Received October 4, 2021, accepted October 28, 2021, date of publication November 2, 2021, date of current version November

11, 2021.
Digital Object Identifier 10.1109/ACCESS.2021.3125102

Applications of Artificial Intelligence in

Distribution Power System Operation
SIMON STOCK , (Student Member, IEEE), DAVOOD BABAZADEH , (Senior Member, IEEE),
AND CHRISTIAN BECKER, (Member, IEEE)
Institute of Electrical Power and Energy Technology, Hamburg University of Technology, 21079 Hamburg, Germany
Corresponding author: Simon Stock ([email protected])
This work was supported by the Hamburg University of Technology (TUHH) (by paying the Open Access fees in the funding program
Open Access Publishing).

ABSTRACT Due to the energy transition and the distribution of electricity generation, distribution power
systems gain a lot of attention as their importance increases and new challenges in operation emerge. The
integration of renewables and electric vehicles for instance leads to manifold changes in the system, e.g.
participation in provision of ancillary services. To solve these challenges artificial intelligence provides a
variety of solutions based on the increase in sensor data and computational capability. This paper provides a
systematic overview of some of the most recent studies applying artificial intelligence methods to distribution
power system operation published during the last 10 years. Based on that, a general guideline is developed
to support the reader in finding a suitable AI technique for a specific operation task. Therefore, four general
metrics are proposed to give an orientation of the requirements of each application. Thus, a conclusion can
be drawn presenting suitable algorithms for each operation task.

INDEX TERMS Artificial intelligence, distribution power systems, fuzzy logic, machine learning, meta-
heuristics, power systems, power system operation.

I. INTRODUCTION intelligence are getting more attention as a result of the

The supply of electrical energy is a necessity in modern soci- increasing amount of data provided by a growing number of
ety and its reliability has to be guaranteed by the power sys- installed measurement systems. Additionally, the growth of
tem operators. Nevertheless, thanks to the energy transition computational power is enabling the application of powerful
many new challenges regarding the stability and operation of techniques in real-time, which also leads to an enlarged inter-
the energy grid emerge. That being the case, system operators est in data-driven systems [4].
as well as other stakeholders have to find solutions to keep the Given these circumstances, the research output in this field
grid stable and the energy supply reliable, while transitioning also increases. To showcase, the online database of publish-
the energy system to carbon-neutral operation. ers Elsevier, IEEE and Wileys (IET) were searched and the
Hence, governments of multiple countries are developing growth of the number of publications per month compared to
strategies in cooperation with grid operators to increase the the previous year was plotted in percent in figure 1. Thus,
integration of renewables, namely the European commis- a clear trend can be identified. According to this analysis,
sion’s strategy to harness the potential of offshore renew- it can be seen that the number of publications on AI in power
able energy [1], [2]. The German government for instance systems increased by 40 % for the last few years. Recently,
published a double tracked strategy relying on the optimized multiple studies have been done reviewing possible applica-
operation of existing grid structures to reduce reserves as well tions of AI in power systems focusing on different aspects.
as the speed up of building new power lines [3]. Some will be briefly revisited here. Zhao et al. [5] give a broad
Besides the conventional techniques in power system overview about the three lifecycle phases of power electronics
operation, data-driven methods and especially artificial being design, control and maintenance as well as possible
AI applications therein. Another broad review was proposed
The associate editor coordinating the review of this manuscript and by Monti et al. [6] focusing on different types of distributed
approving it for publication was Manoj Datta . intelligence control in smart grids. Also, Omitaomu [7]

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
150098 VOLUME 9, 2021
S. Stock et al.: Applications of Artificial Intelligence in Distribution Power System Operation

performed analysis. In this way, a helpful tool for

designing AI solutions for the proposed applications is
provided.
The structure of this article can be outlined as follows.
Section II presents a short overview of the basic concepts
of the most commonly used AI methods, being machine
learning, fuzzy logic respectively control and metaheuristic.
In section III the review methodology is described and the
metrics further used in this study are proposed. The applica-
tions of AI methods in power system operation are reviewed
in section IV, divided into decision-support and closed-loop
systems. This chapter closes with a guideline for finding
suitable methods for each application. Section V gives a brief
outlook about AI in power systems together with a short
FIGURE 1. Development of AI publications.
overview of emerging concepts. Finally, in section VI this
article is concluded with a summary of the contributions.
This article does not claim to be exhaustive, however,
presented a survey on operation and security concerns focus- it aims to provide a systematic overview and guideline for
ing on AI in smart grids. Some much more specific collec- the selection of suitable AI algorithms in distribution power
tions of applications were presented in the following studies. system operation.
In [8], Cao et al. provide an overview of applications using
deep reinforcement learning to solve problems in modern II. ARTIFICIAL INTELLIGENCE METHODS
power systems. Another study about reinforcement learning The term AI has been discussed in many studies, start-
was proposed by Glavic et al. [9], presenting decision and ing with definitions provided by Turing in the middle of
control applications of reinforcement learning in power sys- the 1940s [17], [18]. Also in recent discussions, there is no
tems. Sun et al. [10] focus on voltage control and give an consensus about one definition. However, most agreed on an
overview of the challenges and opportunities of this con- information-processing system influenced by an environment
trol type. An extensive review about load flow control and that is able to learn and adapt [19], [20]. In this chapter, the
its challenges and opportunities including the integration of basic methods of AI techniques used in the different applica-
AI was proposed by Alhelou et al. [11]. Chai et al. [12] and tions throughout this paper are briefly presented according to
Darab et al. [13] investigate different approaches and their figure 2, this division has been adapted from [5] and [21].
applications of AI in fault detection and diagnosis in power It is noteworthy that in many studies, standard algorithms
systems. Kumar et al. [14] propose a collection of possible are modified to a certain extent, however, they are classified
applications of AI and other emerging techniques used for in one category in this paper. Further information on the
the integration of distributed energy resources into the smart different variations of each standard algorithm can be found
grid. A similar review was presented by Ali and Choi [15]. in the publications. Again, the article does not present a com-
Cai and Lu [16] propose a survey about different metaheuris- prehensive overview but the most frequently used techniques.
tic algorithms and possible applications in power systems.
It can be concluded that a systematic review of applications
of AI techniques in distribution system operation would be a
valuable addition to the current literature. On that account,
different aspects of the applications provided here will be
revisited from a control perspective. Doing so, the aim of
this article is to give a systematic overview of applications of
AI already available and possible concerns. The contributions
of this article include the following.
• Systematic review of AI applications in distribution
system operation including the analysis of individual
requirements as well as their essential functions derived FIGURE 2. Categories of AI.
from the reviewed papers. Because of this, four different
metrics are introduced being runtime, dataset, adaptabil- A. MACHINE LEARNING TECHNIQUES
ity and dynamic. The severity of each metric is rated for Machine learning is one group of techniques that gets used
every application from 0 to 5 based on the analysis of a lot in recent studies. Some basics will be revisited here,
the reviewed studies. divided into three subgroups supervised, unsupervised and
• A guideline is derived in a compact format to present reinforcement learning according to figure 3. In supervised
suitable techniques for every application, based on the learning techniques a dataset consisting of the input and

VOLUME 9, 2021 150099

 S. Stock et al.: Applications of Artificial Intelligence in Distribution Power System Operation

FIGURE 3. Categories of machine learning. FIGURE 4. Categories of metaheuristic methods.

the output/target data of the mapping strategy, such as neu- algorithm, compared to PSO the flies implicitly collaborate to
ral networks, is used for training and validation. The train- build the solution and the algorithms only build a geometrical
ing is performed using an optimizer, minimizing an error representation [26]. The Ant colony optimization is based
function consisting of some kind of distance measurement on the foraging behavior of a real ant colony and was first
between the actual output value and the target value of the introduced in the early 1990s [27]. This algorithm is known
data. Supervised learning is used for regular neural networks to be able to solve complex problems in a short amount of
as well as convolutional neural networks, which also use time. The Genetic algorithm is inspired by natural evolution,
an additional filter layer at the inputs [22]. In contrast to consequently, only the fittest individuals are selected for
that, unsupervised learning does not have the target values reproduction by crossing the parent’s genes [28]. In the dif-
included in the dataset, which leads to a training procedure ferential evolutionary optimization a similar approach is used,
where the learning algorithm has to find the individual target while additionally utilizing the survival of the fittest princi-
itself. Typical methods of unsupervised learning are k-means ple [29]. The immune algorithm was also developed from the
clustering algorithm and Support Vector Machine, applica- genetic algorithm based on the construction of the immune
tions are often found in the field of image classification operator through vaccination and immune selection [30].
and anomaly detection [23]. The last learning technique that Tabu search method is another metaheuristic algorithm that
should be mentioned here is reinforcement learning, which is guides a local heuristic procedure to search the global solu-
an agent-based method to learn a certain action strategy. Here, tion space, based on the incorporation of adaptive memory
the agent has to decide for an action in a specific situation and responsive exploration [31]. The simulated annealing
and earns a reward for this. That way, a utility function is combines the physical behavior of the cool-down phase of
approximated describing the value of a specific action [24]. a solid material after annealing with solving large combina-
torial problems of optimization [32].
B. METAHEURISTIC METHODS
Metaheuristic methods describe a group of algorithms that C. RULE-BASED SYSTEMS
solve a given optimization problem, they are often used Rule-based systems are a group of AI techniques that allow
for finding the hyperparameters of models and controllers. the direct integration of human knowledge. By developing a
The algorithms can be divided into two subgroups being set of if-then rules, the system is able to decide based on the
trajectory-based and population-based methods also called rules given by an expert. Hence, a definition can be derived,
swarm intelligence according to figure 4. describing the rule-based system as a modularized know-how
The Particle swarm optimization (PSO) is probably the system [33]. In multiple studies, rule-based systems are also
most famous of the population-based methods. First invented referred to as expert systems. Besides the Boolean logic,
in 1995 by Eberhart and Kennedy, there are multiple fuzzy logic and control have been used a lot in rule-based
improved versions available, as mentioned in many stud- systems, as it can be seen in figure 5.
ies below. The basic version of the PSO uses a swarm of The main advantage of using fuzzy theory and logic is
particles with an initial position and velocity in a search the description of variables and relations in human lin-
field to find a global optimum, while each particle knows its guistics. Because of this, a fuzzy system normally consists
individual best and the global best position [25]. The fruit of three basic parts. Starting with fuzzification, where the
fly algorithm is another popular metaheuristic optimization input signals are mapped onto a fuzzy membership function

150100 VOLUME 9, 2021

S. Stock et al.: Applications of Artificial Intelligence in Distribution Power System Operation

TABLE 2. Review methodology.

After classifying the studies into groups of distribution

system applications, some general metrics are defined, which
are critical for either distribution grid operation or artificial
intelligence design. In this way, each study can be reviewed
FIGURE 5. Categories of rule-based methods.
with a focus on the defined aspects, to build a base for the
concluding guideline. These metrics are then used to show
the severity of requirements of the individual applications for
using a membership degree. These functions can be of dif-
possible approaches. Every metric allows a rating from 0 to 5,
ferent shapes namely triangular, trapezoidal or Gaussian.
meaning low to very high severity of requirements in this
In the following inference module, the calculated degrees
category. After reviewing the studies for each application,
of membership are integrated into IF-THEN fuzzy rules,
a rating for every metric is chosen, concluding the require-
which were prior derived from expert knowledge about the
ments mentioned in the approaches.
process. As the last step the defuzzification is performed,
While concluding the reviewed studies, a general guide-
which creates an output signal that the physical system is able
line was derived, showing the applicability of all methods
to handle [34]. It is noteworthy, that many combinations of the
described in chapter II to each application. The outcome is
different categories are possible, namely, [35]–[37]. In addi-
shown using a table with a general rating summarizing the
tion, some techniques do not belong to only one category
findings of the review and metrics. In the following, the
and might be classified in others as well. To conclude this
metrics are further described.
introductory chapter to AI techniques, basic advantages and
Dataset: As most of the approaches presented here are
limitations for each group of techniques is listed in table 1
data-based, the database used in each study is presented here
together with applications.
if accessible as well as the required set of measurements.
Runtime: The operating timescale and runtime of the pro-
TABLE 1. AI algorithms.
posed approaches are also mentioned if possible. Doing so,
the practical applicability of each study can be reviewed as
well as real-time operation possibilities.
Dynamic: The consideration of system dynamics is manda-
tory for some applications.
Adaptability: The required effort to adapt the reviewed
approaches to new situations is investigated, such as training
time [44]. This metric is highly correlated with the dataset as
large datasets often lead to long training times, so in table 3
duration of times needed for adaptation are given.
In Table 3 the ratings of the different metrics are specified.
The defined ranges for dataset, runtime and adaptability are
based on the reviewed papers as a quantitative measurement.
Therefore, the values provided in the studies are sorted in five
groups defining the outer limits. The highest rating achieved
in one study is then shown in the corresponding figure in
III. REVIEW METHODOLOGY each subchapter. Same holds for the dynamics, but when no
In this chapter the methodology of the proposed review is dedicated timescales are mentioned, the definitions provided
described in detail. Therefore, table 2 lists all databases that in [45] are utilized as additional information.
were searched as well as the search string, search period and Although certain characteristics (or non-functional
the screening procedure. requirements) of the algorithms like convergence speed,

VOLUME 9, 2021 150101

 S. Stock et al.: Applications of Artificial Intelligence in Distribution Power System Operation

TABLE 3. Definition of metrics.

accuracy as well as exact training and testing time are relevant

to system dynamics, but cannot be considered directly as an
individual metric to compare the approaches. It is thanks to
differences in the test scenarios of reviewed studies, in simu-
lation software and hardware set-ups and the modification of
FIGURE 6. Distribution system operation.
original algorithms.
For example, the provision of frequency ancillary services
IV. APPLICATIONS OF AI IN DISTRIBUTION POWER by distributed generation connected to the distribution sys-
SYSTEMS tem has been researched recently [46]. On that account the
In figure 6 the structural division used in this paper is shown. influence of fast generation changes on the distribution grid
Distribution grid operation is divided into two different types. has to be investigated. So the assessment of frequency will
On the left side, the decision support systems are shown, be of interest for future grids. Moreover, when distributed
which use the measurements to visualize different situations and especially inverter-based generation also has to work in
inside the grid, so the operator is able to take manual control grid-forming mode, the dynamics of the distribution system
actions. In this paper, it is referred to as decision support will also change [45]. For that reason, there are also tasks
systems. Here the decision support systems are human-in- considered in this study that might be useful for future distri-
the-loop or open-loop control, as it is not working fully auto- bution system operation, but are mainly used in transmission
matic. This category includes in particular state estimation, systems right now.
fault diagnosis systems and stability assessment methods. For The maintenance of power system components is also part
every system, there are multiple examples presented using of the system operation and a field where AI is applied
the three different types of artificial intelligence described regularly especially in predictive maintenance. But as it is
in section II. very specific for every component it will not be considered
The right side of figure 6 describes the closed-loop or here. For further information on this topic the authors refer
automatic side of grid control. Here, only fully self-reliant to [47], [48].
systems without the need for human interaction to influence To categorize the applications in line with distribution
the grid are considered. This is shown by the two arrows management system terminology, the wording used by EPRI
under the box representing the closed-loop of the grid and for the description of the advanced distribution automa-
the control. Nevertheless, a visualization of the automatic tion (ADA) functions in [49] is used. Therein, high-level
control actions to the distribution system operator is often functions like real-time Distribution Operation Model and
necessary to check if the control is running as intended. The Analysis (DOMA) and Fault Location, Isolation and Service
applications using artificial intelligence found by the authors Restoration (FLIR) are described in detail. In figure 7, the
contain for example voltage control. These days, there exist four main ADAs investigated in this paper are shown as well
different degrees of automation in distribution grids, but they as the included functions. For example, the modeling of loads
are often run by manual operator decisions. In the following and the analysis of power quality are part of the DOMA.
section A, some of the most recent applications of AI in Figure 8 shows a rough timescale ranging from ms to days,
decision support systems are presented and different aspects which is used here to arrange the applications reviewed in the
are discussed. Section B investigates the applications for the following chapter. The tasks occurring in the distribution grid
closed-loop-control. It is noteworthy that not all reviewed are listed above the timeline. The distribution grid manage-
approaches were originally developed for distribution sys- ment functions to solve these problems can be found under
tems, but they are at least applicable to them. Due to the the timeline. The Distribution Operation Model and Analysis
transition of power systems, e.g. distributed generation, oper- function (DOMA) includes the modeling of distribution nodal
ational tasks that were exclusively for transmission systems loads as well as the analysis of economic efficiency and power
might also be of interest for distribution system operators. quality.

150102 VOLUME 9, 2021

S. Stock et al.: Applications of Artificial Intelligence in Distribution Power System Operation

It is worth mentioning that the power quality is a

sub-function describing all techniques that assess the stability
of the system as well as the exploration of the given oper-
ational limits. The state estimation in distribution systems
ranges from ms to minutes depending on the application
and the measurement update rate, but for future systems the
interest in real-time applications in the range of ms will
most probably increase [50]. The aim of the fault location,
isolation and service restoration is the accurate detection
of faults and anomalies in a minimum amount of time and
their isolation and restoration. In dynamic optimization of
the voltage, reactive or active power of the distribution grid
multiple objectives can be considered, discussed in part B of
FIGURE 7. Applications of distribution system operation. this chapter. The distribution grid operator is also in charge
of the coordination of emergency actions like load shedding
as well as the following restorative actions for instance feeder
re-connection.

A. AI IN DISTRIBUTION SYSTEM DECISION SUPPORT

1) MODELING DISTRIBUTION NODAL LOADS,
DISTRIBUTION MODE CIRCUIT CONNECTIVITY
In the modeling of distribution nodal loads on a short
timescale, the digital twin concept is a technique applied in
very recent studies. The concept of the digital twin is gaining
lots of attention with the rise of industry 4.0 in different
disciplines [51]. It describes the digital representation of a
physical system in which behavior and state are changed
based on parameter information and measurements. This con-
cept is already well established in the manufacturing sector
and gets used for a variety of applications therein. Over the
last few years, some applications were also found in the power
systems sector, such as maintenance and power plant design
and very recently monitoring and control [52]. In figure 9
some basic requirements are defined for the implementa-
tion of nodal load and circuit connectivity modeling in a
power system operation environment. The processing speed

FIGURE 8. Timescale of distribution system operation.

It should be highlighted here, that the fault location, iso-

lation and service restoration’s main task is to identify the
faulted section and location and recommend an optimal isola-
tion strategy of the faulted part of the system. In the following,
the emergency actions have to be coordinated, in particular
load shedding, to keep the system stable. The coordination of
restorative actions is necessary after the emergency appeared FIGURE 9. Severity of basic requirements for modeling distribution nodal
and all emergency actions have been taken. loads, distribution mode circuit connectivity.

VOLUME 9, 2021 150103

 S. Stock et al.: Applications of Artificial Intelligence in Distribution Power System Operation

is essential as well as the accuracy to show the system behav- set for the model is found using Monte Carlo simulations.
ior in every possible state. Hence, a large dataset including For training and testing purposes, the TSAT in DSATools by
multiple situations is necessary. Moreover, the model needs Powertech Labs Inc. was used for creating training exam-
to be able to extract the system behavior from the data and ples utilizing the IEEE 39-bus grid. In contrast to that,
adapt to all operation scenarios, consequently the adaptability Cui et al. [56] propose an LSTM based method for parameter
is rated high. Owing to the ability of the digital twin to estimation of a composite load model using the ZIP model.
change its behavior online, an estimation of the dynamic To extract the temporal relationship between measurements at
parameters is often necessary, which can be performed by the target bus, namely, P, Q, V and the load model parameters,
utilizing AI techniques. This seems to be a very common a stack of LSTMs is used for the parameters as well as for the
technique to build a digital twin, so some recent applications measurements. Afterwards, both are temporarily pooled for
are presented in the following. Zhou et al. [53] propose a the extraction of the average temporal latent, finally they are
digital twin-based framework for online grid analysis. For used to estimate the new set of load parameters through linear
this purpose, a virtual model of the power system containing a regression. For the experimental investigation, the Siemens
bus/breaker, node/breaker, and a bus/branch model is updated PSS/E 23-bus system with a Gaussian variation on every
in real-time by SCADA and state estimation data. When a parameter and a ground fault event simulation on every bus
change in the model is detected, a complex event-processing with a sampling time of 0.4 s in a 32 s timeframe is used.
engine performs a situation awareness analysis and feeds the Additionally, a 68-bus New England and New York intercon-
results into a machine learning framework. Therein, an online nected bus system are considered with a similar parameter
security assessment prediction is performed using a neural variation, but transmission line outages as test cases. The data
network, which was previously trained offline. The comput- has a resolution of 0.1 s and a simulation timeframe of 20 s
ing time for the whole process was less than 300 ms in field in this example.
tests. A digital twin approach for load dynamics identification
He et al. [54] propose a digital twin-based power flow is proposed by Baboli et al. [57], by combining system
calculation using an Artificial neural network. To this end, identification methods with neural networks. Hence, optimal
a mapping of the grid inputs P, Q to the outputs being com- utilization of EVs and DERs is possible. In addition to the
plex voltage through a neural network is developed. A set parameters of the nodal load modeling, the overall system
of 9600 samples of Gaussian power fluctuation for the IEEE parameter and the structure of the system are a necessity for
9-bus system was created in MATPOWER for the training and most calculations. In most distribution systems the topology
testing of the system. Thus, the operator is able to monitor and state of the breaker are not known to full extend. Conse-
the power flow all through the power system in real-time quently, some kind of topology identification system can be
only using operational data. To perform a conventional power helpful.
flow calculation, a model of the whole system is mandatory, Zhao et al. [58] propose a neural network architecture with
including load models in particular. Jereminov et al. [55] binary classifiers for online identification of the line status.
propose a linear first-order load model which can be utilized The developed network is trained using a set of inputs from
for power flow calculations and an algorithm for parameter measurements, e.g., PMUs. Across the hidden layers, a set of
fitting called PowerFit. Relying on linear models has the features is created followed by a line status approximation in
advantage of better convergence of the power flow algorithm. the output layer. Thus, the problem is formulated as binary,
By utilizing load data from Carnegie Mellon University cam- because the output of the neural network is either one or
pus and µPMU data from Lawrence Berkeley National Labo- zero, meaning the line is connected or not connected. For
ratories, the developed algorithm as well as the load model are simulation, the IEEE 30-bus system including 41 lines is
tested. During operation, the algorithm searches for cut points utilized. A set of 300,000 training and test samples is created
in the data, which can be detected by drastic changes in the together with adequate power data, setting the line status as
load data. In case of a detected cut point, the load parameters a Bernoulli random variable with a probability of 0.6 of the
are adapted to the new situation. lines being connected. A different approach is followed by
The first dataset consists of 575 samples in 5 minute steps Jafarian et al. [59] using a deep neural network for topol-
of real voltage and current, as well as imaginary current. ogy identification only utilizing measurements available
For the second dataset 12 days were used including complex to DER management systems. For testing purposes, the
voltage and current with a 120 Hz measurement frequency, IEEE 123 node test feeder with 24 different topologies
which was averaged to 500 samples. and different switching positions, which should be classi-
Wang et al. [43] propose a two-stage approach for load fied by the DNN is used. A training and testing set of
modeling using the Western Electricity Coordinating Council 6,000 load and generator settings is created for every topol-
Composite Load Model (WECC) to capture the dynamic load ogy. Chao et al. [42] propose an approach for checking
response. In this model, each load component aggregates a the topology of a LV distribution grid using fuzzy c-means
different type. That being the case, during the first stage, the clustering. For this purpose, the Smart meter data provided
composition of the load at each bus is investigated using a by the individual household is collected over time and the
DDQN learning agent. In the following stage, a parameter correlation between different users is checked. The data is

150104 VOLUME 9, 2021

S. Stock et al.: Applications of Artificial Intelligence in Distribution Power System Operation

finally compared through a GIS system. Doing so, it can 35,000 samples from NYISO load profiles in 5 min time
be shown if a user is listed in the right transformer area by resolution.
utilizing the fuzzy c-means algorithm. In this study, the data A different approach to integrating physical structures
of 48220 users in 500 transformer areas were used and the into a neural network was proposed by Zamzam and
connection relationship was verified. Sidiropoulos [62]. Herein, the graph structure of the electrical
grid is utilized and copied as the structure of the neural net-
2) STATE ESTIMATION work leading to a graph neural network. Doing so, the com-
As a result of the transition towards distributed generation, plexity and trainable parameters of the network are reduced.
the measurability and controllability of the distribution grid The approach was tested using a large dataset and the IEEE
are getting important. Because the operation practices of the 37- feeder power system. Mestav et al. [63] developed a deep
distribution grid changes, an estimation of the actual states is learning-based framework for real-time distribution system
mandatory for every grid model and control. On account of state estimation only relying on machine learning methods.
the missing topology information and under-determined mea- The system consists of an offline part for training the DNN
surement sets that often occur in distribution systems, con- and an online part, which is a copy of the offline DNN.
ventional approaches are hard to implement. Consequently, When new data arrives, the offline system is trained repeat-
the utilization of AI techniques seems like a consequential edly, followed by the adaptation of the online DNN. The
step, because of their ability to extract information purely offline learning procedure starts with some sets of his-
data-based. In the following, some approaches utilizing torical smart meter data, which are used to estimate the
AI techniques are presented, for further information on dis- injection distribution using Gaussian, Gaussian mixture and
tribution system state estimation, please take a look at Prima- Weibull models. In the following, a Monte Carlo sampling is
dianto [60]. As presented in figure 10, state estimation mostly performed using the estimated injection distributions to gen-
requires a large dataset and high adaptability to new system erate some sets of injection samples. These are fed into
states. Additionally, the runtime is fast in most cases due to a power flow calculation, which then creates the training
the AI approaches. samples for the offline DNN training. With this framework,
the creation of a full training set is possible without full
observability. A bad-data detection is also performed by
investigating the difference between the measurement and
the learned distribution parameters. That being the case, bad
data can be detected pre-estimation. For the simulation a
dataset from Pecan Street collection [64] is used, containing
four months of training data and four months of testing data.
Zhang et al. [65] proposed a real-time state estimation with
an additional forecasting system. This approach focuses on
the nonlinear dynamics of the power system. This is done
by utilizing two different types of DNNs, the first one for
estimation and the second one for prediction. For estimation,
a prox-linear net consisting of a plain-vanilla FNN and a
prox-linear solver is presented. This system is trained offline
using a dataset from the 2012 Global Energy Forecasting
Competition containing real load data. When the whole sys-
tem runs in real-time, three basic steps are performed. First,
FIGURE 10. Severity of basic requirements for state estimation.
the estimation of the states through the prox-linear net, which
is then fed to the Deep RNN for predicting the upcom-
ing states. The results are feedback to the prox linear net,
Wang et al. [61] propose a physics-guided model combin- to improve the estimation accuracy.
ing machine learning methods with established physics-based
methods in a hybrid model to enhance the explainability of 3) POWER QUALITY ANALYSIS
the data-driven model. The basic idea is to include temporal The analysis of power system stability is an important part
correlation of the states to get a better state estimation, which of the power system operation, so in this chapter some of
also takes into account the dynamics of the system. That being the most recent studies using AI techniques for stability
the case, a Deep Neural Network model containing LSTMs assessment tasks in distribution system operation are pre-
is used with the measurement of the current and previous sented. Thanks to their ability to efficiently extract nonlinear
time steps as inputs. In this way, the state of the system is dynamic system behavior and their short runtime, multiple
estimated and fed into an AC power flow model containing AI techniques have been applied here. As this paper is an
the physical parameters of the system. For simulation, the overview over various topics of AI applications in power
IEEE 14-bus and 118-bus systems are trained and tested with systems, it is not as detailed as others. For further information,

VOLUME 9, 2021 150105

 S. Stock et al.: Applications of Artificial Intelligence in Distribution Power System Operation

the reader might take a look at Alimi et al. [66]. Some of the regression problem, both are solved using an extreme learning
tasks and studies mentioned here are traditionally considered machine ensemble. Even though the aggregation of the ELMs
for transmission system operators, in particular frequency is done separately for each stage, the performance validation
stability. However, as generation moves to the distribution is aggregated after the training. Doing so, a multi-objective
system, frequency and non-frequency ancillary services have optimization problem is formulated and solved to find the bal-
to be provided by generators connected to the distribution ance between the earliness and the accuracy of the proposed
system [46]. On that account, the assessment of frequency approach. For database generation of pre-fault condition the
and frequency stability might also be relevant for distribution New England 39-bus system is used, running 10,000 Monte
system operators in the near future. Additionally, a distri- Carlo simulations for an added 700 MW Wind power plant
bution grid with a high share of renewables reacts dynam- and loads. The fault simulations are done using the Transient
ically to system disturbances and might affect the overall Stability Assessment Tool (TSAT) at a 0.01 s simulation
power system stability. It is also worth mentioning that there step size and the RELIEFF algorithm was used for feature
are additional stability classifications like resonance stability selection purposes. For further information on this specific
and converter-driven stability, defined by an IEEE PES task algorithm, the authors refer to [68].
force [45]. A similar approach was proposed by Xu et al. [69] and
To detect dynamic system stability, an accurate model of Zhu et al. [70], which also uses a two-stage system for
system dynamics leading to a large dataset and a fast runtime voltage stability assessment. As a first stage, a stability detec-
on account of rapid changes in stability are necessary as tion is performed here, followed by a trajectory prediction.
shown in figure 11, same holds for the adaptability. Neverthe- Mohammadi et al. [71] propose a SVM for assessing the
less, this does not hold for every application in this chapter, power system voltage stability using PMU measurements.
such as long-term voltage stability. Voltage stability is hard The measurement data is processed using two optimization
to assess, considering the time behavior mentioned at the goals, first the misclassification rate of the SVM. In the
beginning of this chapter. Because of this, Zhang et al. [67] second step the number of input features of the SVM is
propose a hierarchical and self-adaptive data analytic method reduced systematically, thanks to the highly nonlinear rela-
for real-time short-term voltage stability assessment. Based tions between the measurements and the voltage stability.
on PMU measurements a voltage instability detection is per- Thus, the authors try to reduce the processing time and
formed in the first place, meaning the voltage is checked for increase the prediction accuracy. For the selection of the
undergoing stable or unstable propagation. The detection of a subset of features, containing the highest amount of infor-
stable status is followed by the prediction of the fault-induced mation, mutual information is used, describing the mutual
voltage delayed recovery (FIDVR) severity. Therefore, the dependence between two random variables. In the following,
root-mean squared voltage dip severity index (RVSI) is used, the dataset is processed by a biogeography-based optimiza-
which is proposed in this paper and evaluates the voltage tion algorithm (BBO), which is an evolutionary optimization
recovery performance of every single bus. That way, a hierar- algorithm. For the first simulation, a 39-bus test system is
chical assessment system is developed, which leads to faster utilized to create a database of 506 pre-fault operation condi-
execution of the process, because the second hierarchy is tions from load patterns, for which the stability of power flow
only activated if the first hierarchy detects a stable point. convergence is checked. Each set contains reactive power
This makes the first stage classification and the second one a flow, line currents, squared voltages and voltage phase angles
calculated from PMUs, as well as the fault location. After-
wards, a 66-bus real power grid in Iran is used for further
testing and 26 operation conditions are created from 15 days
of load data. To this end, 24 PMUs are placed throughout the
grid.
Another approach is proposed by Malbasa et al. [72]
detecting the operating points which are different in the
developed machine learning predictions and the actual system
state. After detection, a training set around the identified oper-
ating points is created, so the machine learning method can be
adapted. That being the case, the incoming data is divided into
three different classes by their voltage stability margin. The
first one contains all operating points with voltage stability
margin (VSM) larger than the mean (stable) VSM, the second
one operating points with a VSM in the second quantile (alert)
and the third one in the smallest quantile (critical). When
operating online the incoming data from PMUs as well as
SCADA is collected into an unlabeled pool and fed into
FIGURE 11. Severity of basic requirements for power quality analysis. the machine learning system for prediction. In this way, the

150106 VOLUME 9, 2021

S. Stock et al.: Applications of Artificial Intelligence in Distribution Power System Operation

most inaccurate predictions can be found and the unlabeled are kept. For testing purposes, the IEEE 39-bus system is used
datasets are handed over to an offline PSSE simulation, which to create datasets containing 165 measurements (bus voltage/
creates accurate labels for these operating points. This leads generation/load, branch power flows, etc.) by performing
to a labeled data pool for further training. All through this Monte Carlo simulations on active power generation and bus
study, ANN, RF and SVM are compared as possible predic- voltage.
tion techniques. For simulation purposes, a version of the The frequency stability is traditionally a transmission sys-
WECC system is used with a dataset of 10,000 operating tem task, but as described above this might also become
points created through PSSE simulation environment. interesting for distribution system operators. Therefore,
In power system operation, transient stability also has to be Xu et al. [76] propose an online predictor of frequency sta-
considered. Owing to the fast appearance, the detection algo- bility utilizing an Extreme Learning Machine. The frequency
rithms have to operate on a very short timescale, as mentioned stability margin is described as a combination of the distance
at the beginning of the chapter. On that account, Tan et al. [73] between the actual frequency, the minimal frequency allowed
propose an approach for transient stability assessment based and the duration of the undergoing. For the training of the
on PMU data considering different signal-to-noise ratios. ELM, a database is constructed by utilizing the New England
Stacked autoencoders (SAE) are used for feature extraction, 39-bus system consisting of generation and load at each bus as
followed by a convolutional neural network (CNN) to per- well as the total load and generation serving as inputs for the
form representational learning for noise filtering. The learn- system. A 30 s simulation is performed for a tripping genera-
ing process is performed offline based on historical data, tor under 360 different system conditions. After offline train-
utilizing unsupervised learning for the features and super- ing, the ELM-predictor can be applied in an online scenario.
vised learning for classification by the CNN. In online oper- A partially similar approach is followed by Mestav et al. [63]
ation, the real-time data provided by PMUs is used for the proposing a two-stage framework for online usage based on a
transient stability analysis. A simulation database is created DNN for the estimation and a stacked ELM for the correction
using the 39-bus New England grid and a PSD-BPA software phase. During the first stage a DNN is used to estimate the
to perform power flow calculations at different load levels, frequency stability metrics being frequency nadir and the
three-phase short-circuits are applied to create an unstable time to reach it, Rate of Change of Frequency (RoCoF) and
system. That way, 4,000 samples were obtained including quasi-steady-state frequency. The results are then handed to
different levels of SNR. Another two-stage approach the second stage and the frequency metrics are corrected
for online transient stability prediction was proposed by using a stacked ELM. For Simulation purposes a modified
Zhu et al. [74] by utilizing a hierarchical convolutional neural IEEE RTS-79 bus system is utilized to generate 30,000 sam-
network. PMU data is used to build the fault-on trajectories ples with 261 inputs for the estimation stage, including pri-
of voltage magnitude, rotor angle, frequency deviation, active mary reserves, inertia constants and load damping coefficient.
and reactive power of each generator. From this transient Yurdakul et al. [77] propose a methodology for the prediction
profile a spectral representation is extracted using discrete of system frequency based on LSTMs. Multiple variables are
Fourier transform and a 2D- graphical representation is cre- used as system inputs in a certain time, including frequency
ated, called the transient image. These images are fed into the measurements, loads, day of the week and hour of the day.
first stage of the proposed system, consisting of a CNN as a These variables are fed into a multilayer LSTM network,
regression model for the stability margin. For the second stage which is followed by a neural network with one neuron to
of the model, all incoming data is divided into subsets by the finally provide the frequency forecast to the operator. For
estimated stability margin. A CNN is trained for every subset testing, a dataset from NGESO containing two months of
to get a more precise estimation and a binary stability signal. secondly measured frequency is utilized and sampled down
For simulation, the IEEE 39-bus system and the Guangdong to a resolution of one minute. All through the study, multiple
Power Grid system in south China are utilized, performing tests were performed including different look-back windows
simulations in the PSD-BPA simulation package released by for the inputs from 1 to 30 minutes. A much more holistic
China-EPRI. 7200 transient cases were created by varying approach is proposed by You et al. [78] utilizing an arti-
load and topology. ficial intelligence model for assessing transient, small sig-
In contrast to that, Chen et al. [75] propose an indirect nal and frequency stability at the same time based on the
PCA approach to reduce the dimensionality of inputs for same input parameters. For this purpose, dispatch data from
stability assessment using ML systems. Because of this, only the scheduling model is obtained by simulation to calculate
the relevant data points are kept. In direct PCA, the reduc- the stability margins for different scenarios in the first step.
tion is performed by cutting off the smallest eigenvalues, By utilizing the generator dispatch levels and network data as
which are not necessarily the most irrelevant for stability input features the artificial intelligence system is able to pre-
assessment in power systems. For this purpose, an indirect dict the stability margin indices for frequency, transient and
PCA approach is presented, which calculates the difference small-signal stability after training. Throughout the study,
between stable and unstable projections for every single a neural network as well as random forests are trained using
dimension after acquiring the necessary values. Doing so, an 18-bus test system with 288 stability scenarios, calculated
the most important, meaning the most different dimensions every 5 minutes for 24 hours. A similar approach is followed

VOLUME 9, 2021 150107

 S. Stock et al.: Applications of Artificial Intelligence in Distribution Power System Operation

by Hotz and Becker [41], by utilizing an ANN for online 10-bus systems are used to run multiple tests. Whereas the
detection of small signal stability. first two grids remain static and the latter one dynamic
In [79] a framework for power quality disturbance analysis in its load and generation behavior. Chen et al. [82]
is proposed combining compressive sensing and machine applied an improved particle swarm optimizer using
learning algorithms. Therein, a two-stage reduction is used, biogeography-based learning to the economic dispatch prob-
first random projection is utilized to reduce the fault signals lem. By integrating a comprehensive learning strategy and
dimension, second a k-nearest-neighbor algorithm is applied biogeography-based optimization, the PSO particles are able
to find the best k nearest neighbor training samples from the to learn from each other, which leads to an efficient bal-
whole dataset and create a reduced set for training. Finally, ance between exploration, exploitation and unintentional
the fault is classified by solving an objective formulation convergence. Across the study, five test systems with vary-
consisting of a combination of L1-norm and L2-norm. For ing numbers of generators and loads were implemented in
testing purposes, sixteen different scenarios including flicker MATLAB and 50 individual simulations were performed for
and harmonics were simulated 200 times each. the generation of statistical information about the perfor-
mance of the optimization algorithm.
4) ANALYSIS OF ECONOMIC EFFICIENCY Besides metaheuristic methods, reinforcement learning
In the following, the economically optimal generation of is also applied to economic optimization problems a lot
a distribution system is investigated. This topic might be recently. Lin et al. [83] present an approach based on deep
most relevant for the balancing group manager since the reinforcement learning for real-time economic dispatch in
liberalization of energy markets. Nevertheless, the evaluation a virtual power plant. By integrating edge computing, the
of economic efficiency is mentioned in [49] as part of the computational and communicational load is reduced. More-
DOMA and the optimization might also be extended to other over, a 3-layer system is implemented with the virtual power
optimization goals. The economic dispatch problem was orig- plant (VPP) operator on the highest stage, followed by an
inally intended to minimize the generation cost, today some agent for every region of generation and load, which is the
approaches also consider the reduction of carbon as the main lowest stage. To solve the economic dispatch problem, a DNN
goal or at least as secondary. Consequently, a cost func- is trained offline at the VPP stage using historical data on an
tion is formulated, integrating all the different optimization hourly timescale. The results are handed over to the agents as
goals [80]. Due to slow change of the problem variables, the set points. Doing so, the agents are able to solve the economic
optimization does not require a fast runtime or adaptability as dispatch for their own region online. For testing purposes,
shown in figure 12. Nevertheless, a certain amount of data is a three-area system with multiple loads and generators was
required for proper optimization. designed and 45,000 samples with 24 hours of data were used
to train the network. Dai et al. [84] also propose a distributed
reinforcement learning algorithm for solving the economic
dispatch, additionally unknown generation cost functions are
taken as a premise. That being the case, the state-action
value function approximation is utilized to solve this problem.
A simple 4-generator system and the IEEE 39-bus system
with 10 generators are used for testing. All through the study,
twelve different load situations are implemented on both
systems to test the algorithm.
Across this chapter, multiple approaches for modeling and
analysis of power systems utilizing AI have been reviewed,
so some major concerns in practical application should be
briefly discussed here.
• The integration of large measurement systems, which
are crucial for most modeling and analysis approaches,
increases the vulnerability to cyberattacks. As this topic
FIGURE 12. Severity of basic requirements for analysis of economic
is not an essential part of this study but a major con-
efficiency. cern when integrating AI approaches, it should be men-
tioned here. It is also worth noting that some approaches
To solve this optimization task a lot of studies were pre- already integrate trust metrics for measurement values
sented utilizing metaheuristics methods, a few are reviewed in modeling, e.g., [85].
in the following. Liang et al. [81] propose an improved fruit • The online adaptability of models and analysis tools
fly optimization algorithm for solving the economic dispatch is crucial for long-term application to power systems,
problem. To this end, multiple modifications are implemented as they change permanently. Consequently, new training
like penalty functions for the integration of operation con- and adaptation of the model have to be performed on a
straints of the system. For testing the IEEE 6-, 40- and regular basis, which also requires new datasets including

150108 VOLUME 9, 2021

S. Stock et al.: Applications of Artificial Intelligence in Distribution Power System Operation

TABLE 4. Application of AI to decision-support systems.

updated system data. Hence, such models and analysis

tools have a high maintenance demand.
Based on the basic requirements proposed for every applica-
tion as well as the usage statistics of the applied AI meth-
ods, a rough guideline for selecting a suitable algorithm for
every application is presented in table 4. Using a quantitative
approach, every algorithm’s applicability to a certain appli-
cation is rated. A bad rating in this table does not mean that
the technique can not be applied to a problem in either case,
it just provides an orientation.

B. APPLICATIONS IN DISTRIBUTION SYSTEM

CLOSED-LOOP CONTROL
In the following chapter, some of the most recent studies
working on closed-loop controls in distribution systems are FIGURE 13. Severity of basic requirements for power system stabilizer.
presented. For the most part, closed-loop controls are used to
optimize voltage, active or reactive power. PSS as well as a coordinated multi-power system stabilizer
for power system stabilization and reduction of low fre-
1) VOLT/VAR/WATT OPTIMIZATION quency oscillations. The NFC combines a fuzzy controller
In the first part of this chapter, the low voltage oscillations and an ANN, so the advantages are the integration of expert
resulting from multiple voltage regulators in the system knowledge into fuzzy logic, no need for a plant model as
are investigated. For the most part, this was a problem of well as the ANNs ability to learn. In this study a 6-layer
the transmission system operator, but when operating an NFC with 2 inputs, the error and the change of the error,
inverter-based generation connected to the distribution grid is developed. For the coordinated multi-power system PSS
in voltage-controlled mode, this might also be interesting for a metaheuristic farmland fertility algorithm is utilized, which
distribution grid operators. The integration of continuously divides the problem into different sections and optimizes each
acting voltage regulators in most conventional generating one separately. For testing purposes, an eigenvalue simulation
units has a significant impact on the steady-state stability of analysis is performed based on SMIB, IEEE 3-machine,
the power system. The reason being the low frequency and 9-bus and the 10-machine 39-bus New England grid
small magnitude oscillations, which can get dangerous for using MATLAB. Three different scenarios are simulated
the system without additional control. Because of this, a sup- in this study, first a symmetrical three-phase fault, fol-
plementary excitation control known as the power system lowed by a drop and a sudden rise of generation at a cer-
stabilizer was developed for synchronous generation [86]. tain number of generators. A similar method is applied by
The runtime of the implemented controller has to be fast Douidi et al. [88], using a cascaded controller consisting of
to react to the dynamic behavior of the system, nevertheless several PD fuzzy control blocks to act as a nonlinear lead-lag
the optimization of controller parameters can be significantly for low frequency damping and a krill herd algorithm for
slower as it does not have to happen in real-time as shown in parameter optimization. Throughout the paper, disturbance
figure 13, same holds for the adaptability. To improve the tests are performed using a 3-machine 9-bus IEEE grid for
controller performance, Sabo et al. [87] propose a simulation as well as a larger 16-machine 68-bus system.
Neuro-Fuzzy controller (NFC) to replace the conventional Masrob et al. [89] proposed an ANN to adjust the controller

VOLUME 9, 2021 150109

 S. Stock et al.: Applications of Artificial Intelligence in Distribution Power System Operation

parameters in real-time for a responsive control behavior.

To mimic the behavior of a PD controller a neural network
with one layer and two nodes is trained based on the rotor
speed aberration and its derivative. On that account, the
utilized grid model is reduced in the first place by keeping
the dominant eigenvalues. A one-machine infinite-bus system
is used for simulation and small changes in the reference
voltage are studied. A similar approach is followed by
Rana et al. [90], except the ANN is used to estimate the
optimal parameters for a conventional PSS online. There-
fore, a set of 1,000 samples in different conditions is gen-
erated and optimized offline for training and testing of
the ANN. After training, a one-machine infinite-bus system
is used for testing combined with different load conditions.
Chitara et al. [91] proposed a metaheuristic approach and
applied the cuckoo search optimization algorithm as a power
FIGURE 14. Severity of basic requirements for voltage and reactive power
system stabilizer to reduce low frequency oscillations. To this control.
end, the algorithm is used to optimize a cost function that
consists of the damping ratio and damping factor of the for each agent in its zone. During offline training, a power
eigenvalues and operating points. By selecting the damping flow calculation is performed in the first step and the results
factor and ratio, the unstable eigenvalues will be placed in a are handed over to the individual agents to detect voltage
D-shape region in the left half of the s-plane. For simulation band violations. To clear the violation, each agent identifies
purposes, the New England 39-bus system with 10 generators control actions and a second power flow is calculated as well
is utilized and three operating conditions are tested varying as the rewards. By utilizing the Illinois- 200-bus system, four
from low to high loading. Additionally, three three-phase different test cases with 5,000 samples each are investigated
fault scenarios are tested on different buses. It is worth in this study. A broad spectrum of possible failures is covered,
mentioning that the computation times for all implemented ranging from load change to line tripping and communication
algorithms are over 15 min. The application of metaheuristic failures. Diao et al. [97] and Duan et al. [98] also propose
algorithms to PSS is a logical consequence, as the problem a deep reinforcement learning framework named GridMind,
can be easily formulated as a figure of merit. That being which is able to take online control actions. When a set of
the case, similar approaches as the ones described above are real-time measurements arrives, a power flow is calculated for
proposed by Dasu et al. [92], Syahputra and Soesanti [93] voltage band violation detection. The obtained states are then
and Ekinci and Hekimo ǧlu [94]. processed by the deep reinforcement learning agent together
A different approach is followed by Zhu and Jin [95]. with the calculated reward, to find the control actions which
Here, a reinforcement learning framework is applied to the lead to the highest future reward. In the following, a second
optimization problem of the power system stabilizer. Thus, power flow is solved also considering the suggested control
the Q learning algorithm is used to optimize the PSS parame- actions to check for voltage violations again. The results are
ters based on the reward received for specific control action. then feedback into the process to update the reward. Multiple
Using Kundur’s four machine two-area system, multiple systems are utilized for testing purposes, namely, the IEEE
tests were performed in MATLAB/SIMULINK. Two dif- 14-bus system to create 10,000 samples with PSAT software.
ferent operation modes were utilized, the first one contain- A hybrid system combining a metaheuristic approach and
ing pulse inference and the second one a three-phase short reinforcement learning is presented in [36]. Here, a two-stage
circuit. system is proposed consisting of a real-time automatic
In the next part of this chapter, the control of voltage and voltage regulation (AVR) with secondary voltage control.
reactive power in distribution systems is briefly investigated. Therein, an artificial emotional reinforcement learning algo-
An in-depth analysis of challenges of voltage control in smart rithm is implemented for each generator’s AVR, followed by
grids can be found in Sun et al. [10]. In this study, only some an improved dragonfly algorithm. The algorithm individuals
recent approaches will be reviewed to give a broad overview. consist of a real and imaginary part to extend the search
On account of multiple different load and generation situa- space for performing a coordinated secondary voltage con-
tions in a power system, the voltage control requires a short trol. As an additional optimization goal, the carbon emission
runtime, high adaptability and therefore a comprehensive ratios for each generator are taken into consideration by the
dataset as shown in figure 14. dragonfly algorithm. Doing so, the voltage control is moved
A multi-agent framework for voltage control using deep from the conventional centralized three-stage approach to a
reinforcement learning was proposed by Wang et al. [96]. For decentralized two-stage version. The proposed framework is
this purpose, the voltage control problem is formulated as a tested on the IEEE 57-bus, 118-bus and 300-bus systems with
Markov game and only the local measurements are available multiple simulations. Relying on metaheuristic optimization,

150110 VOLUME 9, 2021

S. Stock et al.: Applications of Artificial Intelligence in Distribution Power System Operation

Yoshida and Fukuyama [99] and Iwata and Fukuyama [100]

propose a parallel multi-population differential evolutionary
particle swarm optimization for voltage and reactive power
control. On that account, the problem is formulated as a mixed
integer nonlinear optimization problem with AVR operating
values, OLTC tap positions and the number of reactive power
compensation equipment as state variables. Constrained by
the min, max voltage and power flow, the formulation is
optimized by the PSO. In this approach, multiple sub-swarms
are built with agents migrating between the sub-swarms to
exchange information. In total 100 trials simulations are per-
formed on the IEEE 118-bus system. A hybrid particle swarm
optimization was proposed by Chen [101], utilizing a fuzzy
adaptive inference to control the reactive power and voltage in
a distribution system. Thus, a hybrid PSO is used, consisting
of three PSO variants, searching for the optimal solution of
the formulated mixed integer non-linear programming prob- FIGURE 15. Severity of basic requirements for fault identification,
lem. Fuzzy adaptive inference is used to improve the search isolation and service restoration.

process of the proposed PSO as it tends to converge to local

minima. For testing, the IEEE 33-bus system is adapted and the diagnosis results as a membership and non-membership
simulations are performed based on real-world data from a function. Based on this approach, a fault diagnosis model is
Chinese grid operator and multiple scenarios created by load built, which collects information from each device in the out-
and generation variation. Guliyev [40] developed a fuzzy con- age area, calculates the fault confidence levels and finds the
troller for reactive power control of capacitor banks. Based faulty component. The proposed approach is further inves-
on the reactive power, the derivative of reactive power, the tigated using two different test grids in different sizes and
voltage and the number of commutations of the capacitor voltage levels. The first one is 69 kV with 10 sections and the
bank, a control action is obtained using 96 calculated fuzzy second one 348 kV with 18 system sections. For both grids,
rules. Furthermore, the functionality of the proposed con- three fault scenarios are tested ranging from single faults
troller is tested using a capacitor bank model for simulation without failure devices to multiple faults with the rejection
with multiple variations of load and generation. of circuit breakers.
A different approach by Lin et al. [39] utilizes a hybrid
2) FAULT IDENTIFICATION, ISOLATION AND SERVICE system combining the advantages of the Genetic Algorithm
RESTORATION and the Tabu Search for fault diagnosis. Therefore, the objec-
Another important part of power system operation is the tive function containing the state of the system, the breakers
detection and diagnosis of faults, losses and anomalies, in par- in the system and the protection are enhanced by integrat-
ticular different kinds of short circuits, communication out- ing the influence between the main and backup protection.
ages, and cyberattacks to avoid power outages. For more This was done to improve the problem of non-uniqueness,
detailed information on this specific topic, the reader might which was investigated in prior studies. For simulation pur-
check Gururajapathy et al. [102]. In figure 15 it is shown, poses, a typical power system consisting of 4 substations,
that the detection of faults and anomalies requires an accurate 28 components, 84 protections, and 40 breakers is utilized
modeling of system dynamics and a fast runtime, due to and possible failure scenarios are tested throughout the paper.
fast changing system states. The systems adaptability is also Jamil et al. [104] propose a two-stage approach for fault clas-
important, owing to changing topologies, e.g, the addition sification consisting of a wavelet transformation and a genetic
of a feeder. Some techniques focus on a specific error type, algorithm. In the first stage, the incoming current signal is
so a comprehensive dataset is not mandatory in any case. decomposed into a high and low frequency part by a filter
It is noteworthy, that the system level restoration and recovery and then separated into detail and aggregated components
could be open-loop in some cases, but in future scenarios we using multi resolution analysis. Here, the detail coefficients
assume that it will be automatic. are unique for every type of fault and are used to construct
First, the detection and classification of faults, mainly short the input datasets for the following genetic algorithm. Ten
circuits, will be reviewed in this chapter. Peng et al. [103] different types of faults are simulated on a transmission
propose an intuitionistic fuzzy spiking neural p system for line in MATLAB with different values of fault inception
fault diagnosis. According to the author’s analysis, this angle and fault resistance. In [105] Wang et al. combine the
approach has three main advantages, the first one being advantages of metaheuristics and machine learning by using a
the use of an intuitionistic fuzzy number, which shows the SVM for the classification of anomalies in generation control,
amount of alarm information and the impreciseness. The optimized by an improved PSO algorithm. Thus, the PSO
fuzzy reasoning mechanism followed by the representation of is extended by an adaptive speed weighting and population

VOLUME 9, 2021 150111

 S. Stock et al.: Applications of Artificial Intelligence in Distribution Power System Operation

splitting to overcome convergence speed and local minima load factor and entropy were selected from a range of pos-
problems. For experimental validation, a dataset provided sible features using the neighborhood component analysis.
by the Electric Consumption & Occupancy is utilized, con- In [108] Goswami et al. studied three different machine learn-
sisting of aggregated consumption data of six households ing techniques for fault analysis and classification, focus-
over 8 months in one second resolution. By adding an addi- ing on the identification. For this classification task a set
tional analysis stage, Deng et al. [38] proposed a hybrid of 11300 samples is created, 1,000 for each fault type, using
three-stage approach combining techniques from different MATLAB. Each sample consists of one voltage and current
fields to detect the faults of a motor bearing. In the first stage, value per phase, which makes six features. The time span
the original vibration signal is decomposed into different set for the fault data captured during the simulation is 10 ms
intrinsic mode functions using the empirical mode decom- to 280 ms, this enables the trained classifiers to identify the
position followed by a fuzzy information entropy, to obtain fault types based on their dynamic behavior. Three classifica-
the features used in the following stages. All through the tion algorithms are trained, namely, a K-Nearest-Neighbor,
second stage, an improved PSO algorithm is proposed using Support Vector Machine and a Decision Tree utilizing the
different methods to tailor it to the existing problem and prior described dataset.
the optimization of the parameters of a Least-squares SVM, In [79] the framework also applied in [109] is applied to
which is trained using the improved PSO algorithm. Finally, fault classification. A real-world test grid with 13.2 kV is built
the trained Least-squares SVM is applied to the actual classi- in MATLAB and 10 different faults are simulated 100 times
fication task. For testing of the developed algorithm vibration each with multiple locations and fault parameters to generate
data from Bearing Data Center of Case Western University a training dataset. Wang et al. [110] present an approach for
was used, measured at a frequency of 12,000 Hz for 10 s. online anomaly detection in a data attack situation with auto-
Another three-stage approach targeting the identification and matic generation control using a multi-class classification
localization of anomalies based on PMUs was proposed by based on k-nearest neighbors. Therefore, k-means clustering
Li et al. [106]. To avoid costly labeling work, unsupervised is performed offline to form the classes, followed by an online
learning was used, so there is no need for historical labels. The classification based on three conformity metrics that rely
developed framework consists of three main parts, first event on the received Area Control Error. The developed system
detection based on the change-point method, which detects is tested using the IEEE 39-bus grid with synthetic data
abrupt changes in the data matrix, generated from PMU and six developed test scenarios, for instance flip and ramp
signals. In the second part, an identification approach based attack. An ensemble system for the detection of anomalies
on two stages is proposed. The first one is a PCA, which finds in PMU data is proposed by Zhou et al. [37]. Therein, a set
the most important features to cluster the events, followed of base detectors is trained offline in the first place. When
by a compactness evaluation stage. Here, the compactness of detecting an anomaly in an online operation, the anomaly
the normal and event data distribution is evaluated. The final scores are calculated and aggregated as a decision base. For
step is the localization of the event that occurred, which is testing purposes, a stream of synthetic PMU data created by
done by the change-point method. That way, the location of a real-time digital simulator as well as real-world PMU data
the event is estimated by finding the most significant change is used. For the latter one, three different types of anomalies
in neighboring PMUs. Across the experiments performed in are detected, voltage is zero, data during events and data
the study different events as well as different PMU pene- beyond 5 % of the mean value and the previous and fol-
tration levels are considered. Blazakis et al. [107] propose lowing point. Ren et al. [111] also focused on online
an adaptive neuro fuzzy inference system (ANFIS) for the anomaly detection and proposed a machine learning approach
detection of nontechnical losses such as illegal electricity integrating HPC.
power consumption, e.g., meter tampering, or grid manip- The detection is performed after the anomaly appeared,
ulation. The ANFIS system is the combination of an ANN so a prediction algorithm seems valuable for the sys-
using backpropagation with a Sugeno fuzzy inference system tem operator to take preventive actions. To this end,
consisting of five layers. The first one being a fuzzy layer, Zhang et al. [112] developed a two-step system for fault
followed by a product layer combining the results from the prediction based on historical data. In the first stage, three
first layer. In the third layer, all values are normalized fol- LSTM subnetworks extract the temporal information from
lowed by a defuzzification layer before all nodes are aggre- current, voltage and active power measurements. The result-
gated in the output layer. As testing scenarios, three base ing features are fed into a SVM classifier for fault estimation.
cases are identified, being partial theft, when the consump- In this study, a dataset from the China Southern Power Grid in
tion is constantly lower, overload, when the consumption is Wanjiang from the years 2012-2014 was used, with
constantly higher and periodic theft, reduced consumption 2500 samples for training and testing, consisting of 500 mea-
during specific hours of the day. By varying the percentage surement points each. These points are recorded either before
of the overall consumption, thirteen different scenarios are a line trip or during normal operation in a 15 min resolution
created. The dataset used in this study contains 5,000 house- and labeled with the event that finally appeared. The practical
hold data from Ireland collected over 6 months in a 30 min applicability of the proposed approach is highlighted by the
resolution. As an input for the ANFIS, the mean, median, authors, as a result of the performed experiments and the

150112 VOLUME 9, 2021

S. Stock et al.: Applications of Artificial Intelligence in Distribution Power System Operation

ability of the system parameters to constantly update to new For simulation purposes, the CIGRE 23-bus system is uti-
states of the power system online. Ashok et al. [113] propose lized. Multiple different situations are extensively tested
an approach to detect cyber-attacks in measurement systems including different turbine controllers, frequency-dependent
and their influence on the state estimation by forecasting the loads and different numbers of operating points.
state behavior and comparing the prediction with the actual Usman et al. [116] propose an approach to solve the
measurement. In [85] the anomaly detection is also integrated optimal load shedding coordination when undergoing volt-
into the state estimation by the addition of a trust metric for age limits. By using a multi-objective minimization prob-
every measurement. lem formulation, power loss, voltage deviation and cost of
the load shedding are taken into account. An evolutionary
3) COORDINATION OF EMERGENCY ACTIONS PSO algorithm is used to solve this optimization problem.
When the grid is in a critical mode, the coordination of Additionally, the computational efficiency is increased by
emergency actions is an important part of distribution system integrating an evolutionary competition between the current
operation. A possible reaction to a frequency drop besides and previous positions of particles. Throughout the study, the
control actions is the shedding of loads. That being the case, approach is tested on the IEEE 33-bus distribution grid using
under-frequency load shedding relays get installed that dis- a daily demand profile. In [117], Hasanat et al. propose an
connect the load when the threshold is reached. Thanks to ant colony optimization algorithm to minimize the amount
the complexity of power systems and their dynamic behavior, of load shedding. On that account, the algorithm is extended
the optimal load shedding strategy is hard to find. Because of using a local search to improve the solution. As this approach
this, the proposed algorithms have to be adaptable but do not is purely developed on the graph structure of the electrical
require large datasets as shown in figure 16. grid, a graph generator is used in combination with data from
the national grid data of Bangladesh to create benchmark
datasets. Dreidy et al. [118] propose another study on the
optimization of the load shedding amount comparing PSO,
binary evolutionary programming and binary genetic algo-
rithm. Because of this, a part of the Malaysian distribution
system with high penetration of PVs is modelled and ten loads
are flexibly prioritized while two remaining a fixed priority.

4) COORDINATION OF RESTORATIVE ACTIONS

The importance of the distribution system operator in restora-
tive actions after a blackout is increasing as more gener-
ation happens to be connected to the distribution system.
That being the case, multiple studies were recently presented
working on the integration of distributed generation into the
restoration strategy [119], [120]. Moreover, the integration
of energy storage in restoration is also proposed in many
studies, including the utilization of EVs [121], [122]. Applied
FIGURE 16. Severity of basic requirements for coordination of emergency
actions. approaches have to be adaptable, as a result of the differ-
ent situations that appear after a blackout, as can be seen
In [114] an approach for optimal coordination of in figure 17.
under-frequency load shedding is proposed. An analytical Zhou et al. [123] propose a multi-agent system to restore a
hierarchy process algorithm is used to rate the importance of distribution grid. Therein, two classes of agents are defined,
each load for the creation of a ranking as well as different being load agents and distribution substation agents. The first
load shedding strategies. K-means clustering is used to divide one’s goal is to restore the own load and offer energizing
the appearing instability mode into different clusters based on actions to the neighboring loads in the following, whereas
the detected measurements. Moreover, an ANN is trained to the second one monitors the substation power flow and holds
choose the best load shedding strategy for every cluster. For a list of each load agent. For testing purposes, a 16-bus
training purposes, 667 datasets are created through offline 24 branch system is used considering load and substation
simulation of system faults causing instability. The trained faults. Another multi-agent based approach is provided by
algorithm is then tested on the IEEE 39-bus grid using three Sampaio et al. [124]. A system based on four different types
different fault scenarios. Another approach was proposed by of agents is proposed, being substation agent, feeder agent,
Malkowski and Nieznanski [115] using fuzzy logic to create branch agent and equipment agent. Every agent communi-
an adaptive load shedding algorithm. Therefore, a member- cates its name, status and equipment as well as loading and
ship function for frequency as well as the derivative of the priority loads if accessible. For testing, a simulator was devel-
frequency is created. The output of the inference block shows oped to represent a real MV system with four substations.
the number of load groups that have to be disconnected. In this study, two fault situations were tested. In contrast

VOLUME 9, 2021 150113

 S. Stock et al.: Applications of Artificial Intelligence in Distribution Power System Operation

TABLE 5. Applications of AI to closed-loop control systems.

• In control theory, the robustness of controllers has been

studied extensively, but in AI there have only been
few studies and for some AI approaches the robustness
is hard to investigate due to their black-box structure.
Nevertheless, for the safe operation of power systems,
robustness is crucial, as the provision of energy to the
consumers has to be guaranteed by the system opera-
tor. Hence, future research should consider this prob-
lem, which is also closely related to the explainability
that will be discussed in more detail in the next
chapter.
• The dynamics in power systems change as a result of the
integration of inverter-coupled participants [45], which
also leads to interactions between multiple controllers
with different runtime and optimization goals. As a con-
FIGURE 17. Severity of basic requirements for coordination of restorative
sequence, the influence of AI based controllers has to be
actions. investigated in depth.
To conclude the analysis of closed-loop systems, a table
to that, a hybrid approach using machine learning and is presented similar to the one proposed at the end of the
metaheuristic methods is proposed in [35]. A total of four previous subchapter. In this way, table 5 provides a guideline
metaheuristic algorithms being a modified PSO, frog leaping for the selection of an AI algorithm for closed-loop control
algorithm, genetic algorithm and ant colony optimization applications. As prior described, this only serves as a guide-
algorithm are used in combination with a multi-class support line and does not exclude possible applications of low-rated
vector machine. To create a database, 320 faults were sim- methods for certain problems.
ulated using the IEEE 69-bus system and for every scenario Throughout this chapter, some of the most recent studies
each metaheuristic algorithm finds its best restoration solu- for distribution grid operation utilizing AI techniques were
tion. The best solution of the four algorithms serves as a target systematically presented and reviewed. Doing so, the current
value for the SVM and features that are extracted utilizing state of the art is presented in a compact format and research
a discrete wavelet transform are used as inputs. In [125] directions for the individual problems can be identified.
another approach using metaheuristic and machine learn- In table 6 the distribution of the application in publications
ing techniques is proposed for optimal restoration strategies. is shown to visualize possible research gaps or underrep-
For this purpose, a PSO is used to optimize the switch posi- resented research fields. It can be seen from the table that
tions for power loss minimization. Additionally, an ANN is the field of economic dispatch and power system stabilizers
fed with switch positions, system level loads and line power to has not been extensively researched for the past few years.
find a load balancing index and the voltage profile. To create Regarding the underrepresentation of research in the field of
a database for training the IEEE 33-bus by NRM is utilized economic dispatch, a reason might be that this topic is very
and load variations are performed. complex thanks to market mechanisms and government rules.
In the following, some concluding remarks regarding the Power system stabilizer has been researched for some time,
practical implementation of closed-loop AI algorithms in but they seem to lose relevance during the last few years
power systems will be discussed similar to the previous according to the analysis. One reason for that might be the rise
subchapter. of inverter-based generation also connected to the distribution

150114 VOLUME 9, 2021

S. Stock et al.: Applications of Artificial Intelligence in Distribution Power System Operation

TABLE 6. Summary of usage statistics of applications in distribution has been paid to this topic during the last few years [134].
power systems operation.
As a result of the restrictions, a central collection and usage of
data for individual loads is sometimes not possible. Whereas
being necessary to protect the data, the restrictions lead to
suboptimal datasets and slow down further development to a
certain point.
C. REDUCTION OF COMPUTATIONAL LOAD
Despite the fact that computational power has increased mas-
sively during the last decade, it still requires a lot of time
for most AI techniques to learn complex behaviors. So, the
application to real-time tasks in power systems is limited,
especially when the learning process is performed online,
e.g., online adaptation of Deep Neural Networks. This is also
grid, which needs at least an adaption of controls, as they are a critical point in metaheuristic approaches when applied to
not coupled synchronously. This also leads to an enhanced optimization tasks, they perform an extensive trial-and-error
description of stability, which has to be considered when process, which takes a long time to converge. Moreover,
designing controllers. For further information, the authors online adaptability of the developed models and approaches is
refer to [126]–[128]. necessary due to long time changes in power systems, namely
V. OUTLOOK: AI IN POWER SYSTEMS aging of components.
It can be concluded that AI is already used a lot in power The following conclusion regarding the application of
systems research these days but there is still room for revisited AI approaches to future power system operation is
improvement and further research. In this chapter, a broad not comprehensive but highlights three points that should be
outlook will be given together with some major concerns in considered:
AI practical implementation, which have to be addressed in • Enhancement of explainability to increase the plausibil-
future studies after the potential of AI application has been ity and traceability of AI systems
shown in the last few years. • Enhancement of robustness in every state of the system,
so safe operation is possible
A. EXPLAINABILITY OF AI • Development of comprehensive datasets for training and
Regarding the implementation in real-world systems a con- testing of developed approaches
cern is the explainability of the AI system. This is especially • Reduction of computational demand to allow real-time
relevant in closed-loop control, where the system and not application and online adaptability
the operator take control actions. Owing to the black-box In the last few years, a lot of new concepts and techniques
structure of most AI approaches, it is not possible to check if emerged in power systems that have lots of potential also for
the developed system is behaving as intended in all situations. AI applications. Some of them will be mentioned here briefly.
An approach that has been researched a lot recently is the
D. SECTOR COUPLING
integration of physics into AI techniques and explainable AI.
There are multiple possibilities to explain the behavior of For the transition of the power system to carbon neutrality, the
an AI system ranging from understanding what the model coupling of energy sectors is a research field that gained a lot
has learned [129] to the explanation of individual predic- of attention during the last decade [135], [136]. Concepts like
tions [130]. Nevertheless, during this review, only a few stud- energy quarters were developed for optimizing whole sector
ies on explainable AI in power systems were found, so this coupled buildings and neighborhoods, which are placed in the
topic has lots of potential for future research activities. distribution grid. They can also be utilized as flexibility in the
grid [137], depending on the storage capability e.g., in bat-
B. DATABASE tery storage and EVs. Therefore, various control concepts
As already mentioned across this review, the database is and algorithms have to be developed and implemented, also
essential for most AI applications. The complexity of power AI seems to be a helpful tool here [138], [139].
systems is high, which leads to a need for extensive data
collection and sorting for training and testing of the devel- E. PROVISION OF ANCILLARY SERVICES
oped models and algorithms. There are already some open As a result of the volatility of renewable energy genera-
source data collections available [64], [131], but these are tion and new load characteristics, the provision of ancillary
developed on specific benchmark grids or recorded in a spe- services is an upcoming research topic. On that account,
cific real-world situation. Therefore, the general applicability flexible loads, namely, electric vehicles (EVs) can be uti-
might not hold for every approach. Nevertheless, there are lized besides DER [140], [141] through advanced load-
already collections and surveys available concluding multiple ing concepts [142]–[145]. A problem that occurs in the
databases [132], [133]. Additionally, access to data is also participation of EVs in ancillary services is the missing
limited because of data privacy regulations. A lot of attention infrastructure [146]. Moreover, thanks to the increase in

VOLUME 9, 2021 150115

 S. Stock et al.: Applications of Artificial Intelligence in Distribution Power System Operation

asynchronous generation the acquisition and provision of sys- [14] N. M. Kumar, A. A. Chand, M. Malvoni, K. A. Prasad, K. A. Mamun,
tem inertia has to be considered as it is no longer inherently F. R. Islam, and S. S. Chopra, ‘‘Distributed energy resources and the
application of AI, IoT, and blockchain in smart grids,’’ Energies, vol. 13,
provided by synchronous generators. This leads to multiple no. 21, p. 5739, Nov. 2020.
questions on the behavior of low inertia systems [147]. There- [15] S. S. Ali and B. J. Choi, ‘‘State-of-the-art artificial intelligence techniques
fore, also AI techniques can be applied [148]. for distributed smart grids: A review,’’ Electronics, vol. 9, no. 6, pp. 1–28,
2020.
[16] H. Cai and X. Lu, ‘‘A survey of artificial intelligence algorithm in
VI. CONCLUSION power system applications,’’ in Proc. IEEE 3rd Int. Elect. Energy Conf.
In this article, some of the most recent applications of AI (CIEEC), Sep. 2019, pp. 1902–1906.
[17] A. Hodges, ‘‘Alan Turing and the Turing test,’’ in Parsing the Turing Test,
in distribution power system operation are reviewed. Con-
R. Epstein, G. Roberts, and G. Beber, Eds. Dordrecht, The Netherlands:
sequently, the basic functionality of the main AI methods Springer, 2007.
of rule-based systems, metaheuristic methods and machine [18] A. Turing, ‘‘Computing machinery and intelligence,’’ in Parsing the
Turing Test, R. Epstein, G. Roberts, and G. Beber, Eds. Dordrecht, The
learning are introduced and the exertion on power system Netherlands: Springer, 2007.
specific problems is shown. Throughout this study, the appli- [19] P. Wang, ‘‘On defining artificial intelligence,’’ J. Artif. Gen. Intell.,
cations are divided into decision-support and closed-loop vol. 10, no. 2, pp. 1–37, Jan. 2019.
[20] D. Dobrev, ‘‘A definition of artificial intelligence,’’ 2012, pp. 1–7,
control systems. A guideline for selecting a suitable algorithm arXiv:1210.1568.
for an application is developed in this review. In doing so, four [21] J. Fulcher, Computational Intelligence: An Introduction, vol. 115. Berlin,
general metrics are proposed, the severity of requirements Germany: Springer, May 2008, pp. 3–78.
[22] D. P. Kingma and J. L. Ba, ‘‘Adam: A method for stochastic optimiza-
on the database, runtime, dynamics and adaptability. The tion,’’ in Proc. 3rd Int. Conf. Learn. Represent. (ICLR), 2015, pp. 1–15.
metrics are quantitatively assigned to each application. Based [23] B. Kiran, D. Thomas, and R. Parakkal, ‘‘An overview of deep learning
based methods for unsupervised and semi-supervised anomaly detection
on the revised studies and the provided metrics, a conclusion in videos,’’ J. Imag., vol. 4, no. 2, p. 36, Feb. 2018.
is provided rating the suitability of each technique to the [24] K. Arulkumaran, M. P. Deisenroth, M. Brundage, and A. A. Bharath,
application. ‘‘Deep reinforcement learning: A brief survey,’’ IEEE Signal Process.
Mag., vol. 34, no. 6, pp. 26–38, Nov. 2017.
[25] L. Abualigah and A. Diabat, ‘‘Advances in sine cosine algorithm: A com-
REFERENCES prehensive survey,’’ Artif. Intell. Rev., vol. 54, pp. 2567–2608, Jan. 2021.
[1] An EU Strategy to Harness the Potential of Offshore Renewable Energy [26] S. S. Mohar, S. Goyal, and R. Kaur, ‘‘Fruit fly optimization algorithm for
for a Climate Neutral Future, Eur. Commission, Brussels, Belgium, 2020, intelligent IoT applications,’’ in Fog, Edge, and Pervasive Computing in
p. 27. Intelligent IoT Driven Applications, D. Gupta and A. Khamparia, Eds.
[2] E-Energy Smart Grid Standardization Roadmap Strategy Group. (2016). IEEE, 2020.
The German Roadmap E-Energy/Smart Grid. [Online]. Available: [27] T. Nurcahyadi and C. Blum, ‘‘Adding negative learning to ant colony
https://www.dke.de/resource/blob/2018916/3fac8b29cb6deb28527ed9681 optimization: A comprehensive study,’’ Mathematics, vol. 9, no. 4, p. 361,
0859ca3/dke-normungsroadmap-1-en-data.pdf 2021. [Online]. Available: https://www.mdpi.com/2227-7390/9/4/361
[3] Aktionsplan Stromnetz, BMWi, Federal Ministry Econ. Affairs Energy, [28] S. Katoch, S. S. Chauhan, and V. Kumar, ‘‘A review on genetic algo-
Berlin, Germany, 2018, pp. 1–4. rithm: Past, present, and future,’’ Multimedia Tools Appl., vol. 80,
[4] I. Goodfellow, Y. Bengio, and A. Courville, Deep Learning. Cambridge, pp. 8091–8126, Oct. 2020.
MA, USA: MIT Press, 2016. [Online]. Available: http://www. [29] V. Yadav, A. K. Yadav, M. Kaur, and D. Singh, ‘‘Trigonometric mutation
deeplearningbook.org and successful-parent-selection based adaptive asynchronous differential
[5] S. Zhao, F. Blaabjerg, and H. Wang, ‘‘An overview of artificial intelli- evolution,’’ J. Ambient Intell. Humanized Comput., Jul. 2021.
gence applications for power electronics,’’ IEEE Trans. Power Electron., [30] J. K. Pattanaik, M. Basu, and D. P. Dash, ‘‘Optimal power flow with
vol. 36, no. 4, pp. 4633–4658, Apr. 2021. FACTS devices using artificial immune systems,’’ in Proc. Int. Conf.
[6] A. Monti, F. Ponci, A. Benigni, and J. Liu, ‘‘Distributed intelligence Technol. Adv. Power Energy (TAP Energy), 2017, pp. 1–6.
for smart grid control,’’ in Proc. Int. School Nonsinusoidal Currents [31] R. Toorajipour, V. Sohrabpour, A. Nazarpour, P. Oghazi, and
Compensation (ISNCC), Jun. 2010, pp. 46–58. M. Fischl, ‘‘Artificial intelligence in supply chain management:
[7] O. A. Omitaomu and H. Niu, ‘‘Artificial intelligence techniques in smart A systematic literature review,’’ J. Bus. Res., vol. 122, pp. 502–517,
grid: A survey,’’ Smart Cities, vol. 4, no. 2, pp. 548–568, Apr. 2021. Jan. 2021. [Online]. Available: https://www.sciencedirect.com/science/
[8] D. Cao, W. Hu, J. Zhao, G. Zhang, B. Zhang, Z. Liu, Z. Chen, and article/pii/S014829632030583X
F. Blaabjerg, ‘‘Reinforcement learning and its applications in modern [32] E. K. Burke and G. Kendall, Search Methodologies: Introductory Tutori-
power and energy systems: A review,’’ J. Mod. Power Syst. Clean Energy, als in Optimization and Decision Support Techniques. Boston, MA, USA:
vol. 8, no. 6, pp. 1029–1042, 2020. Springer, 2014.
[9] M. Glavic, R. Fonteneau, and D. Ernst, ‘‘Reinforcement learning for [33] K. R. Chowdhary, Fundamentals of Artificial Intelligence. New Delhi,
electric power system decision and control: Past considerations and per- India: Springer, 2020.
spectives,’’ IFAC-PapersOnLine, vol. 50, no. 1, pp. 6918–6927, Jul. 2017. [34] L. Caponetti and G. Castellano, Fuzzy Logic for Image Processing a
[10] H. Sun, Q. Guo, J. Qi, V. Ajjarapu, R. Bravo, J. Chow, Z. Li, R. Moghe, Gentle Introduction Using Java. Cham, Switzerland: Springer, 2017.
E. Nasr-Azadani, U. Tamrakar, G. N. Taranto, R. Tonkoski, G. Valverde, [35] I. Srivastava, S. Bhat, V. S. G. Thadikemalla, and A. R. Singh, ‘‘A hybrid
Q. Wu, and G. Yang, ‘‘Review of challenges and research opportunities machine learning and meta-heuristic algorithm based service restoration
for voltage control in smart grids,’’ IEEE Trans. Power Syst., vol. 34, no. 4, scheme for radial power distribution system,’’ Int. Trans. Electr. Energy
pp. 2790–2801, Jul. 2019. Syst., vol. 31, pp. 1–25, Sep. 2021.
[11] H. H. Alhelou, M. E. Hamedani-Golshan, R. Zamani, [36] L. Yin, S. Luo, Y. Wang, F. Gao, and J. Yu, ‘‘Coordinated complex-
E. Heydarian-Forushani, and P. Siano, ‘‘Challenges and opportunities of valued encoding dragonfly algorithm and artificial emotional reinforce-
load frequency control in conventional, modern and future smart power ment learning for coordinated secondary voltage control and automatic
systems: A comprehensive review,’’ Energies, vol. 11, no. 10, p. 2497, voltage regulation in multi-generator power systems,’’ IEEE Access,
2018. vol. 8, pp. 180520–180533, 2020.
[12] E. Chai, P. Zeng, S. Ma, H. Xing, and B. Zhao, ‘‘Artificial intelligence [37] M. Zhou, Y. Wang, A. K. Srivastava, Y. Wu, and P. Banerjee, ‘‘Ensemble-
approaches to fault diagnosis in power grids: A review,’’ in Proc. Chin. based algorithm for synchrophasor data anomaly detection,’’ IEEE Trans.
Control Conf. (CCC), Jul. 2019, pp. 7346–7353. Smart Grid, vol. 10, no. 3, pp. 2979–2988, Mar. 2019.
[13] C. Darab, R. Tarnovan, A. Turcu, and C. Martineac, ‘‘Artificial intelli- [38] W. Deng, R. Yao, H. Zhao, X. Yang, and G. Li, ‘‘A novel intelligent
gence techniques for fault location and detection in distributed genera- diagnosis method using optimal LS-SVM with improved PSO algo-
tion power systems,’’ in Proc. 8th Int. Conf. Mod. Power Syst. (MPS), rithm,’’ Soft Comput., vol. 23, no. 7, pp. 2445–2462, 2019, doi: 10.1007/
May 2019, pp. 2019–2022. s00500-017-2940-9.

150116 VOLUME 9, 2021

S. Stock et al.: Applications of Artificial Intelligence in Distribution Power System Operation

[39] X. Lin, S. Ke, Z. Li, H. Weng, and X. Han, ‘‘A fault diagnosis [59] M. Jafarian, A. Soroudi, and A. Keane, ‘‘Distribution system topology
method of power systems based on improved objective function and identification for DER management systems using deep neural net-
genetic algorithm-tabu search,’’ IEEE Trans. Power Del., vol. 25, no. 3, works,’’ in Proc. IEEE Power Energy Soc. Gen. Meeting (PESGM),
pp. 1268–1274, Jul. 2010. Aug. 2020, p. 4.
[40] H. B. Guliyev, ‘‘Reactive power adaptive control system in networks with [60] A. Primadianto and C.-N. Lu, ‘‘A review on distribution system state
distributed generation based on fuzzy set theory,’’ in Proc. Int. Artif. Intell. estimation,’’ IEEE Trans. Power Syst., vol. 32, no. 5, pp. 3875–3883,
Data Process. Symp. (IDAP), Sep. 2019, pp. 1–5. Sep. 2017.
[41] C. Hotz and C. Becker, ‘‘Online monitoring of power system [61] L. Wang, Q. Zhou, and S. Jin, ‘‘Physics-guided deep learning for power
small signal stability using artificial neural networks,’’ in Proc. system state estimation,’’ J. Mod. Power Syst. Clean Energy, vol. 8, no. 4,
Conf. Sustain. Energy Supply Energy Storage Syst. (NEIS), 2019, pp. 607–615, 2020.
pp. 44–49. [62] A. S. Zamzam and N. D. Sidiropoulos, ‘‘Physics-aware neural net-
[42] L. Chao, Z. Lei, and L. Yuhang, ‘‘Topology checking method for low volt- works for distribution system state estimation,’’ IEEE Trans. Power Syst.,
age distribution network based on fuzzy C-means clustering algorithm,’’ vol. 35, no. 6, pp. 4347–4356, Nov. 2020.
in Proc. IEEE Int. Conf. Artif. Intell. Comput. Appl. (ICAICA), Jun. 2020, [63] K. R. Mestav, J. Luengo-Rozas, and L. Tong, ‘‘Bayesian state estimation
pp. 1077–1080. for unobservable distribution systems via deep learning,’’ IEEE Trans.
[43] X. Wang, Y. Wang, D. Shi, J. Wang, and Z. Wang, ‘‘Two-stage WECC Power Syst., vol. 34, no. 6, pp. 4910–4920, Nov. 2019.
composite load modeling: A double deep Q-learning networks approach,’’ [64] Pecan Street Inc. Pecan Street Dataport. Accessed: Feb. 1, 2021. [Online].
2019, pp. 4331–4344, vol. 11, no. 5, arXiv:1911.04894. Available: https://www.pecanstreet.org
[44] K. Andresen, Design and Use Patterns of Adaptability in Enterprise [65] L. Zhang, G. Wang, and G. B. Giannakis, ‘‘Real-time power sys-
Systems (Reihe Wirtschaftsinformatik). Berlin, Germany: GITO, 2006. tem state estimation and forecasting via deep neural networks,’’ 2018,
[Online]. Available: https://books.Google.de/books?id=Zs-_u1EcIiAC pp. 4069–4077, vol. 67, no. 15, arXiv:1811.06146.
[45] N. Hatziargyriou, J. Milanvic, C. Rahmann, V. Ajjarapu, C. Canizares, [66] O. A. Alimi, K. Ouahada, and A. M. Abu-Mahfouz, ‘‘A review of machine
I. Erlich, D. Hill, I. Hiskens, I. Kamwa, B. Pal, P. Pourbeik, learning approaches to power system security and stability,’’ IEEE Access,
J. Sanchez-Gasca, A. Stankovic, T. Van Cutsem, V. Vittal, and C. Vournas, vol. 8, pp. 113512–113531, 2020.
‘‘Definition and classification of power system stability—Revisited &
[67] Y. Zhang, Y. Xu, Z. Y. Dong, and R. Zhang, ‘‘A hierarchical self-adaptive
extended,’’ IEEE Trans. Power Syst., vol. 36, no. 4, pp. 3271–3281,
data-analytics method for real-time power system short-term voltage sta-
Jul. 2021.
bility assessment,’’ IEEE Trans. Ind. Informat., vol. 15, no. 1, pp. 74–84,
[46] S. Möws, V. Scheffer, and C. Becker, ‘‘Probabilistic power forecast Jan. 2019.
of renewable distributed generation for provision of control reserve
[68] C. Eiras-Franco, B. Guijarro-Berdiñas, A. Alonso-Betanzos, and
using vine copulas,’’ IET Gener., Transmiss. Distrib., vol. 14, no. 25,
A. Bahamonde, ‘‘Scalable feature selection using ReliefF aided by
pp. 6312–6318, Dec. 2020.
locality-sensitive hashin,’’ Int. J. Intell. Syst., vol. 36, no. 11, pp. 1–19,
[47] P. T. Baboli, D. Babazadeh, A. Raeiszadeh, S. Horodyvskyy, and
2021.
I. Koprek, ‘‘Optimal temperature-based condition monitoring system for
wind turbines,’’ Infrastructures, vol. 6, no. 4, p. 50, Mar. 2021. [69] Y. Xu, R. Zhang, J. Zhao, Z. Y. Dong, D. Wang, H. Yang, and K. P. Wong,
‘‘Assessing short-term voltage stability of electric power systems by a
[48] P. T. Baboli, A. Raeiszadeh, D. Babazadeh, and J. Meiners, ‘‘Two-stage
hierarchical intelligent system,’’ IEEE Trans. Neural Netw. Learn. Syst.,
condition-based maintenance model of wind turbine: From diagnosis
vol. 27, no. 8, pp. 1686–1696, Aug. 2016.
to prognosis,’’ in Proc. IEEE Int. Smart Cities Conf. (ISC), Sep. 2020,
pp. 1–6. [70] L. Zhu, C. Lu, I. Kamwa, and H. Zeng, ‘‘Spatial–temporal feature learning
in smart grids: A case study on short-term voltage stability assessment,’’
[49] N. Markushevich, M. Lachman, and F. Cleveland. (2010).
IEEE Trans. Ind. Informat., vol. 16, no. 3, pp. 1470–1482, Mar. 2020.
Distribution Grid Management (Advanced Distribution Automation)
Functions. [Online]. Available: https://smartgrid.epri.com/UseCases/ [71] H. Mohammadi, G. Khademi, M. Dehghani, and D. Simon, ‘‘Voltage
Distribution_Grid_ManagementSG_UC_nm3.pdf stability assessment using multi-objective biogeography-based subset
[50] J. Zhao, A. Gómez-Expósito, M. Netto, L. Mili, A. Abur, V. Terzija, selection,’’ Int. J. Electr. Power Energy Syst., vol. 103, pp. 525–536,
I. Kamwa, B. Pal, A. K. Singh, J. Qi, Z. Huang, and A. P. Meliopoulos, Dec. 2018.
‘‘Power system dynamic state estimation: Motivations, definitions, [72] V. Malbasa, C. Zheng, P.-C. Chen, T. Popovic, and M. Kezunovic,
methodologies, and future work,’’ IEEE Trans. Power Syst., vol. 34, no. 4, ‘‘Voltage stability prediction using active machine learning,’’ IEEE Trans.
pp. 3188–3198, Jul. 2019. Smart Grid, vol. 8, no. 6, pp. 3117–3124, Nov. 2017.
[51] F. Tao, H. Zhang, A. Liu, and A. Y. C. Nee, ‘‘Digital twin in [73] B. Tan, J. Yang, X. Pan, J. Li, P. Xie, and C. Zeng, ‘‘Representa-
industry: State-of-the-art,’’ IEEE Trans. Ind. Informat., vol. 15, no. 4, tional learning approach for power system transient stability assessment
pp. 2405–2415, Apr. 2019. based on convolutional neural network,’’ J. Eng., vol. 2017, no. 13,
[52] C. Brosinsky, D. Westermann, and R. Krebs, ‘‘System control cen- pp. 1847–1850, Jan. 2017.
ters,’’ in Proc. IEEE Int. Energy Conf. (ENERGYCON), Jun. 2018, [74] L. Zhu, D. J. Hill, and C. Lu, ‘‘Hierarchical deep learning machine for
pp. 1–6. power system online transient stability prediction,’’ IEEE Trans. Power
[53] M. Zhou, J. Yan, and D. Feng, ‘‘Digital twin and its application to Syst., vol. 35, no. 3, pp. 2399–2411, May 2020.
power grid online analysis,’’ CSEE J. Power Energy Syst., vol. 5, no. 3, [75] K. Chen, ‘‘Indirect PCA dimensionality reduction based machine learn-
pp. 391–398, 2019. ing algorithms for power system transient stability assessment,’’ in
[54] X. He, Q. Ai, R. C. Qiu, and D. Zhang, ‘‘Preliminary exploration on digital Proc. IEEE Innov. Smart Grid Technol.-Asia (ISGT Asia), May 2019,
twin for power systems: Challenges, framework, and applications,’’ 2019, pp. 4175–4179.
pp. 1–8, arXiv:1909.06977. [76] Y. Xu, Y. Dai, Z. Y. Dong, R. Zhang, and K. Meng, ‘‘Extreme learning
[55] M. Jereminov, A. Pandey, H. A. Song, B. Hooi, C. Faloutsos, and machine-based predictor for real-time frequency stability assessment
L. Pileggi, ‘‘Linear load model for robust power system analysis,’’ in of electric power systems,’’ Neural Comput. Appl., vol. 22, nos. 3–4,
Proc. IEEE PES Innov. Smart Grid Technol. Conf. Eur. (ISGT-Eur.), no. 1, pp. 501–508, Mar. 2013.
Sep. 2017, pp. 1–6. [77] O. Yurdakul, F. Eser, F. Sivrikaya, and S. Albayrak, ‘‘Very short-
[56] M. Cui, M. Khodayar, C. Chen, X. Wang, Y. Zhang, and M. Khodayar, term power system frequency forecasting,’’ IEEE Access, vol. 8,
‘‘Deep learning-based time-varying parameter identification for system- pp. 141234–141245, 2020.
wide load modeling,’’ IEEE Trans. Smart Grid, vol. 10, no. 6, [78] S. You, Y. Zhao, M. Mandich, Y. Cui, H. Li, H. Xiao, S. Fabus, Y. Su,
pp. 6102–6114, 2019. Y. Liu, H. Yuan, H. Jiang, J. Tan, and Y. Zhang, ‘‘A review on artifi-
[57] P. T. Baboli, D. Babazadeh, and D. R. K. Bowatte, ‘‘Measurement-based cial intelligence for grid stability assessment,’’ in Proc. IEEE Int. Conf.
modeling of smart grid dynamics: A digital twin approach,’’ in Proc. 10th Commun., Control, Comput. Technol. Smart Grids (SmartGridComm),
Smart Grid Conf. (SGC), Dec. 2020, pp. 6–11. Nov. 2020, pp. 1–6.
[58] Y. Zhao, J. Chen, and H. V. Poor, ‘‘Efficient neural network architecture [79] L. Cheng, Z. Wu, R. Duan, and K. Dong, ‘‘Adaptive compressive sensing
for topology identification in smart grid,’’ in Proc. IEEE Global Conf. and machine learning for power system fault classification,’’ in Proc.
Signal Inf. Process. (GlobalSIP), Dec. 2016, pp. 811–815. SoutheastCon, Mar. 2020, pp. 1–7.

VOLUME 9, 2021 150117

 S. Stock et al.: Applications of Artificial Intelligence in Distribution Power System Operation

[80] A. Mahor, V. Prasad, and S. Rangnekar, ‘‘Economic dispatch using parti- [100] S. Iwata and Y. Fukuyama, ‘‘Dependability evaluations of parallel differ-
cle swarm optimization: A review,’’ Renew. Sustain. Energy Rev., vol. 13, ential evolutionary particle swarm optimization for voltage and reactive
no. 8, pp. 2134–2141, Oct. 2009. power control,’’ IEEJ Trans. Power Energy, vol. 137, no. 1, pp. 52–58,
[81] J. Liang, H. Zhang, K. Wang, and R. Jia, ‘‘Economic dispatch of power 2017.
system based on improved fruit fly optimization algorithm,’’ in Proc. 14th [101] S. Chen, W. Hu, C. Su, X. Zhang, and Z. Chen, ‘‘Optimal reactive power
IEEE Conf. Ind. Electron. Appl. (ICIEA), Jun. 2019, pp. 1360–1366. and voltage control in distribution networks with distributed generators by
[82] X. Chen, B. Xu, and W. Du, ‘‘An improved particle swarm optimization fuzzy adaptive hybrid particle swarm optimisation method,’’ IET Gener.,
with biogeography-based learning strategy for economic dispatch prob- Transmiss. Distrib., vol. 9, no. 11, pp. 1096–1103, Aug. 2015.
lems,’’ Complexity, vol. 2018, pp. 1–15, Jul. 2018. [102] S. S. Gururajapathy, H. Mokhlis, and H. A. Illias, ‘‘Fault location and
[83] L. Lin, X. Guan, Y. Peng, N. Wang, S. Maharjan, and T. Ohtsuki, ‘‘Deep detection techniques in power distribution systems with distributed gen-
reinforcement learning for economic dispatch of virtual power plant in eration: A review,’’ Renew. Sustain. Energy Rev., vol. 74, pp. 949–958,
internet of energy,’’ IEEE Internet Things J., vol. 7, no. 7, pp. 6288–6301, Jul. 2017, doi: 10.1016/j.rser.2017.03.021.
Jul. 2020. [103] H. Peng, J. Wang, J. Ming, P. Shi, M. J. Perez-Jimenez, W. Yu, and C. Tao,
[84] P. Dai, W. Yu, G. Wen, and S. Baldi, ‘‘Distributed reinforcement learning ‘‘Fault diagnosis of power systems using intuitionistic fuzzy spiking
algorithm for dynamic economic dispatch with unknown generation cost neural P systems,’’ IEEE Trans. Smart Grid, vol. 9, no. 5, pp. 4777–4784,
functions,’’ IEEE Trans. Ind. Informat., vol. 16, no. 4, pp. 2258–2267, Sep. 2018.
Apr. 2020. [104] M. Jamil, S. K. Sharma, and D. K. Chaturvedi, ‘‘Fault classification of
[85] M. Brand, S. Ansari, F. Castro, R. Chakra, B. H. Hassan, C. Krüger, three-phase transmission network using genetic algorithm,’’ Int. J. Eng.
D. Babazadeh, and S. Lehnhof, ‘‘A framework for the integration of Appl. Sci., vol. 5, no. 7, pp. 84–88, 2015.
ICT-relevant data in power system applications,’’ in Proc. IEEE Milan [105] Z. Wang, Y. Fu, C. Song, P. Zeng, and L. Qiao, ‘‘Power system anomaly
PowerTech, Jun. 2019, pp. 1–6. detection based on OCSVM optimized by improved particle swarm opti-
[86] A. A. Ba-Muqabel and M. A. Abido, ‘‘Review of conventional power mization,’’ IEEE Access, vol. 7, pp. 181580–181588, 2019.
system stabilizer design methods,’’ in Proc. IEEE GCC Conf. (GCC), [106] H. Li, Y. Weng, E. Farantatos, and M. Patel, ‘‘An unsupervised learning
Mar. 2006, pp. 1–7. framework for event detection, type identification and localization using
[87] A. Sabo, N. I. A. Wahab, M. L. Othman, M. Z. A. M. Jaffar, H. Acikgoz, PMUs without any historical labels,’’ in Proc. IEEE Power Energy Soc.
and H. Beiranvand, ‘‘Application of neuro-fuzzy controller to replace Gen. Meeting (PESGM), Aug. 2019, pp. 1–5.
smib and interconnected multi-machine power system stabilizers,’’ Sus- [107] K. V. Blazakis, T. N. Kapetanakis, and G. S. Stavrakakis, ‘‘Effective elec-
tainability, vol. 12, no. 22, pp. 1–42, 2020. tricity theft detection in power distribution grids using an adaptive neuro
[88] B. Douidi, L. Mokrani, and M. Machmoum, ‘‘A new cascade fuzzy fuzzy inference system,’’ Energies, vol. 13, no. 12, p. 3110, Jun. 2020.
power system stabilizer for multi-machine system stability enhance- [108] T. Goswami and U. B. Roy, ‘‘Predictive model for classification of power
ment,’’ J. Control, Autom. Electr. Syst., vol. 30, no. 5, pp. 765–779, system faults using machine learning,’’ in Proc. IEEE Region Conf.
Oct. 2019, doi: 10.1007/s40313-019-00486-7. (TENCON), Oct. 2019, pp. 1881–1885.
[89] M. A. Masrob, M. A. Rahman, and G. H. George, ‘‘Design of a neural
[109] L. Cheng, Z. Wu, S. Xuanyuan, and H. Chang, ‘‘Power quality distur-
network based power system stabilizer in reduced order power system,’’
bance classification based on adaptive compressed sensing and machine
in Proc. IEEE 30th Can. Conf. Electr. Comput. Eng. (CCECE), no. 1,
learning,’’ in Proc. IEEE Green Technol. Conf. (GreenTech), Apr. 2020,
Apr. 2017, pp. 3–8.
pp. 65–70.
[90] M. J. Rana, M. S. Shahriar, and M. Shafiullah, ‘‘Levenberg–Marquardt
[110] P. Wang, M. Govindarasu, A. Ashok, S. Sridhar, and D. McKinnon,
neural network to estimate UPFC-coordinated PSS parameters to
‘‘Data-driven anomaly detection for power system generation control,’’
enhance power system stability,’’ Neural Comput. Appl., vol. 31, no. 4,
in Proc. IEEE Int. Conf. Data Mining Workshops (ICDMW), Nov. 2017,
pp. 1237–1248, Apr. 2019.
pp. 1082–1089.
[91] D. Chitara, K. R. Niazi, A. Swarnkar, and N. Gupta, ‘‘Cuckoo search
[111] H. Ren, Z. Hou, and P. Etingov, ‘‘Online anomaly detection using machine
optimization algorithm for designing of a multimachine power sys-
learning and HPC for power system synchrophasor measurements,’’ in
tem stabilizer,’’ IEEE Trans. Ind. Appl., vol. 54, no. 4, pp. 3056–3065,
Proc. IEEE Int. Conf. Probabilistic Methods Appl. Power Syst. (PMAPS),
Jul./Aug. 2018.
Jun. 2018, pp. 1–5.
[92] B. Dasu, M. Sivakumar, and R. Srinivasarao, ‘‘Interconnected multi-
machine power system stabilizer design using whale optimization algo- [112] S. Zhang, Y. Wang, M. Liu, and Z. Bao, ‘‘Data-based line trip fault
rithm,’’ Protection Control Mod. Power Syst., vol. 4, no. 1, pp. 1–11, prediction in power systems using LSTM networks and SVM,’’ IEEE
Dec. 2019. Access, vol. 6, pp. 7675–7686, 2018.
[93] R. Syahputra and I. Soesanti, ‘‘Power system stabilizer model based on [113] A. Ashok, M. Govindarasu, and V. Ajjarapu, ‘‘Online detection of stealthy
fuzzy-PSO for improving power system stability,’’ in Proc. Int. Conf. false data injection attacks in power system state estimation,’’ IEEE Trans.
Adv. Mechatronics, Intell. Manuf., Ind. Automat. (ICAMIMIA), Oct. 2015, Smart Grid, vol. 9, no. 3, pp. 1636–1646, May 2018.
pp. 121–126. [114] T. N. Le, N. Au Nguyen, and H. A. Quyen, ‘‘Emergency control of load
[94] S. Ekinci and B. Hekimoǧlu, ‘‘Parameter optimization of power system shedding based on coordination of artificial neural network and analytic
stabilizer via salp swarm algorithm,’’ in Proc. 5th Int. Conf. Electr. hierarchy process algorithm,’’ in Proc. Int. Conf. Syst. Sci. Eng. (ICSSE),
Electron. Eng. (ICEEE), May 2018, pp. 143–147. Jul. 2017, pp. 57–60.
[95] X. Zhu and T. Jin, ‘‘Research of control strategy of power system sta- [115] R. Małkowski and J. Nieznański, ‘‘Underfrequency load shedding:
bilizer based on reinforcement learning,’’ in Proc. IEEE 2nd Int. Conf. An innovative algorithm based on fuzzy logic,’’ Energies, vol. 13, no. 6,
Circuits Syst. (ICCS), Dec. 2020, pp. 81–85. p. 1456, Mar. 2020.
[96] S. Wang, J. Duan, D. Shi, C. Xu, H. Li, R. Diao, and Z. Wang, [116] M. Usman, A. Amin, M. M. Azam, and H. Mokhlis, ‘‘Optimal under
‘‘A data-driven multi-agent autonomous voltage control framework using voltage load shedding scheme for a distribution network using EPSO
deep reinforcement learning,’’ IEEE Trans. Power Syst., vol. 35, no. 6, algorithm,’’ in Proc. 1st Int. Conf. Power, Energy Smart Grid (ICPESG),
pp. 4644–4654, Nov. 2020. Apr. 2018, pp. 1–6.
[97] R. Diao, Z. Wang, D. Shi, Q. Chang, J. Duan, and X. Zhang, [117] M. Hasanat, M. Hasan, I. Ahmed, M. I. Chowdhury, J. Ferdous, and
‘‘Autonomous voltage control for grid operation using deep reinforcement S. Shatabda, ‘‘An ant colony optimization algorithm for load shedding
learning,’’ 2019, arXiv:1904.10597. minimization in smart grids,’’ in Proc. 5th Int. Conf. Informat., Electron.
[98] J. Duan, D. Shi, R. Diao, H. Li, Z. Wang, B. Zhang, D. Bian, and Vis. (ICIEV), May 2016, pp. 176–181.
Z. Yi, ‘‘Deep-reinforcement-learning-based autonomous voltage control [118] M. Dreidy, H. Mokhlis, and S. Mekhilef, ‘‘Application of meta-heuristic
for power grid operations,’’ IEEE Trans. Power Syst., vol. 35, no. 1, techniques for optimal load shedding in islanded distribution network
pp. 814–817, Jan. 2020. with high penetration of solar PV generation,’’ Energies, vol. 10, no. 2,
[99] H. Yoshida and Y. Fukuyama, ‘‘Parallel multipopulation differen- p. 150, Jan. 2017.
tial evolutionary particle swarm optimization for voltage and reac- [119] J. Xu, M. Li, Q. Chen, and M. Wu, ‘‘Evaluation method of DG black
tive power control,’’ Electr. Eng. Jpn., vol. 204, no. 3, pp. 31–40, start of distribution network,’’ in Proc. 5th Asia Conf. Power Electr. Eng.
Aug. 2018. (ACPEE), Jun. 2020, pp. 2141–2147.

150118 VOLUME 9, 2021

S. Stock et al.: Applications of Artificial Intelligence in Distribution Power System Operation

[120] J. Zhao, Q. Zhang, Z. Liu, and X. Wu, ‘‘A distributed black-start opti- [141] A. Fernández-Guillamón, G. Martínez-Lucas, Á. Molina-García, and
mization method for global transmission and distribution network,’’ IEEE J. I. Sarasua, ‘‘An adaptive control scheme for variable speed wind tur-
Trans. Power Syst., vol. 36, no. 5, pp. 4471–4481, Sep. 2021. bines providing frequency regulation in isolated power systems with
[121] L. Sun, X. Wang, W. Liu, Z. Lin, F. Wen, S. P. Ang, and M. A. Salam, thermal generation,’’ Energies, vol. 13, no. 13, p. 3369, Jul. 2020.
‘‘Optimisation model for power system restoration with support from [142] S.-A. Amamra and J. Marco, ‘‘Vehicle-to-grid aggregator to support
electric vehicles employing battery swapping,’’ IET Gener., Transmiss. power grid and reduce electric vehicle charging cost,’’ IEEE Access,
Distrib., vol. 10, no. 3, pp. 771–779, Feb. 2016. vol. 7, pp. 178528–178538, 2019.
[122] A. Di Giorgio, A. Giuseppi, F. Liberati, and A. Pietrabissa, ‘‘Controlled [143] M. S. H. Nizami, M. J. Hossain, S. Rafique, K. Mahmud, U. B. Irshad, and
electricity distribution network black start with energy storage system G. Town, ‘‘A multi-agent system based residential electric vehicle man-
support,’’ in Proc. 25th Medit. Conf. Control Automat. (MED), Jul. 2017, agement system for grid-support service,’’ in Proc. IEEE Int. Conf. Envi-
pp. 781–786. ron. Electr. Eng. IEEE Ind. Commercial Power Syst. Eur. (EEEIC/I&CPS
[123] X. Zhou, H. Wang, Y. Ma, and Z. Gao, ‘‘Research on power system Europe), Jun. 2019, pp. 1–6.
restoration based on multi-agent systems,’’ in Proc. Chin. Control Decis. [144] G. Wenzel, M. Negrete-Pincetic, D. E. Olivares, J. MacDonald, and
Conf. (CCDC), Jun. 2018, pp. 4955–4960. D. S. Callaway, ‘‘Real-time charging strategies for an electric vehicle
[124] R. F. Sampaio, L. S. Melo, R. P. S. Le ao, G. C. Barroso, and J. R. Bezerra, aggregator to provide ancillary services,’’ IEEE Trans. Smart Grid, vol. 9,
‘‘Automatic restoration system for power distribution networks based on no. 5, pp. 5141–5151, Sep. 2018.
multi-agent systems,’’ IET Gener., Transmiss. Distrib., vol. 11, no. 2, [145] W. Khan, F. Ahmad, and M. S. Alam, ‘‘Fast EV charging
pp. 475–484, 2017. station integration with grid ensuring optimal and quality power
[125] D. M. Soomro, M. Y. Al-Izzi, H. A. Soodi, N. M. Elasager, and exchange,’’ Eng. Sci. Technol., Int. J., vol. 22, no. 1, pp. 143–152,
S. C. Chong, ‘‘Optimal restoration of distribution system by using PSO Feb. 2019. [Online]. Available: https://www.sciencedirect.com/
and ANN,’’ J. Telecommun., Electron. Comput. Eng., vol. 9, nos. 3–7, science/article/pii/S2215098617315057
pp. 1–6, 2017. [146] A. Aldik, A. T. Al-Awami, E. Sortomme, A. M. Muqbel, and
[126] R. H. Lasseter, Z. Chen, and D. PattabiRaman, ‘‘Grid-forming inverters: M. Shahidehpour, ‘‘A planning model for electric vehicle aggregators
A critical asset for the power grid,’’ IEEE J. Emerg. Sel. Topics Power providing ancillary services,’’ IEEE Access, vol. 6, pp. 70685–70697,
Electron., vol. 8, no. 2, pp. 925–935, Jun. 2020. 2018.
[127] P. Unruh, M. Nuschke, P. Strauß, and F. Welck, ‘‘Overview on grid- [147] F. Milan, F. Dorfler, G. Hug, D. J. Hill, and G. Verbic, ‘‘Foundations and
forming inverter control methods,’’ Energies, vol. 13, no. 10, p. 2589, challenges of low-inertia systems (invited paper),’’ in Proc. Power Syst.
May 2020. Comput. Conf. (PSCC), Jun. 2018, pp. 1–25.
[128] T. Dragicevic, S. Vazquez, and P. Wheeler, ‘‘Advanced control methods [148] Y. Cui, S. You, and Y. Liu, ‘‘Ambient synchrophasor measurement based
for power converters in DG systems and microgrids,’’ IEEE Trans. Ind. system inertia estimation,’’ in Proc. IEEE Power Energy Soc. Gen. Meet-
Electron., vol. 68, no. 7, pp. 5847–5862, Jul. 2021. ing (PESGM), Aug. 2020, pp. 1–5.
[129] A. Mahendran and A. Vedaldi, ‘‘Visualizing deep convolutional neural
networks using natural pre-images,’’ Int. J. Comput. Vis., vol. 120, no. 3, SIMON STOCK (Student Member, IEEE) rec-
pp. 233–255, 2016. eived the B.Sc. and M.Sc. degrees from the Uni-
[130] W. Samek, T. Wiegand, and K. R. Müller, ‘‘Explainable artificial intel- versity of Bremen, Germany, in 2017 and 2018,
ligence: Understanding, visualizing and interpreting deep learning mod- respectively. He has been a Research Assistant
els,’’ 2017, arXiv:1708.08296. with the Institute of Electrical Power Engineer-
[131] S. Meinecke, D. Sarajlić, S. R. Drauz, A. Klettke, L.-P. Lauven, ing (IEET), Hamburg University of Technology,
C. Rehtanz, A. Moser, and M. Braun, ‘‘SimBench—A benchmark dataset since 2020. From 2018 to 2020, he worked with
of electric power systems to compare innovative solutions based on power the project team of ArcelorMittal Bremen before
flow analysis,’’ Energies, vol. 13, no. 12, p. 3290, Jun. 2020. joining the IEET. His research interests include
[132] E. Proedrou, ‘‘A comprehensive review of residential electricity load the optimal operation of distribution grids and the
profile models,’’ IEEE Access, vol. 9, pp. 12114–12133, 2021. application of AI in power systems.
[133] F. Wiese, I. Schlecht, W.-D. Bunke, C. Gerbaulet, L. Hirth, M. Jahn,
F. Kunz, C. Lorenz, J. Mühlenpfordt, J. Reimann, and W.-P. Schill, ‘‘Open DAVOOD BABAZADEH (Senior Member, IEEE)
power system data—Frictionless data for electricity system modelling,’’ received the master’s degree in electrical engineer-
Appl. Energy, vol. 236, pp. 401–409, Feb. 2019. ing, with a focus on electrical power systems, from
[134] European Union. General Data Protection Regulation. Accessed: Apr. 5, the KTH Royal Institute of Technology (KTH),
2021. [Online]. Available: https://gdpr.eu// Sweden. He continued working as a Research
[135] C. Heckel and J.-P. Becker, ‘‘Investigation of the voltage stability Assistant at KTH and received the Ph.D. degree,
in the integrated energy system of northern Germany,’’ in Proc. with a focus on operation and control of HVDC
Conf. Sustain. Energy Supply Energy Storage Syst. (NEIS), grids, in 2017. He has been a Lecturer and a Senior
Hamburg, Germany, Sep. 2020, pp. 1–6. [Online]. Available: Scientist with the Institute of Electrical Power
https://ieeexplore.ieee.org/document/9273415
Engineering (IEET), Hamburg University of Tech-
[136] M. Z. Oskouei, P. T. Baboli, and D. Babazadeh, ‘‘Robust stochastic
nology, since 2021. After that, he joined the OFFIS Research Institute,
optimization for energy sharing between multi-carrier microgrids using
transactive energy management system,’’ in Proc. ETG Congr., 2021, Germany, and led a research team focusing on automation and control of
pp. 1–6. power systems. He has been involved in many national and European projects
[137] J. F. Heinekamp, M. Voß, S. Krutzsch, F. Sick, K. Strunz, and S. Albayrak, dealing with modeling, simulation, and validation of smart grid solutions.
‘‘Integration of building inertia thermal energy storage into smart grid CHRISTIAN BECKER (Member, IEEE) was
control,’’ in Proc. Int. Conf. Smart Energy Syst. Technol. (SEST), born in Germany, in 1972. He received the
Sep. 2020, pp. 1–6. Dipl.-Ing. and Dr.-Ing. degrees from TU Dort-
[138] R. Jing, M. Wang, H. Liang, X. Wang, N. Li, N. Shah, and Y. Zhao,
mund University, in 1996 and 2001, respectively.
‘‘Multi-objective optimization of a neighborhood-level urban energy
From 2002 to 2015, he worked with the Research
network: Considering game-theory inspired multi-benefit allocation
constraints,’’ Appl. Energy, vol. 231, pp. 534–548, May 2018, doi: and Development Division of Airbus. He has
10.1016/j.apenergy.2018.09.151. been a Full Professor and the Head of the Insti-
[139] A. Kathirgamanathan, M. De Rosa, E. Mangina, and D. P. Finn, tute of Electrical Power and Energy Technology,
‘‘Data-driven predictive control for unlocking building energy flexi- Hamburg University of Technology, Germany,
bility: A review,’’ Renew. Sustain. Energy Rev., vol. 135, Jan. 2021, since 2015. His research activities and professional
Art. no. 110120, doi: 10.1016/j.rser.2020.110120. experiences are focused on power system stability and control engineering
[140] I. Ranaweera and O.-M. Midtgård, ‘‘Optimization of operational cost for a for terrestrial as well as on-board electrical power systems with dedicated
grid-supporting PV system with battery storage,’’ Renew. Energy, vol. 88, focus on grid integration of power electronics equipment and FACTS.
pp. 262–272, Apr. 2016, doi: 10.1016/j.renene.2015.11.044.

VOLUME 9, 2021 150119