IEEE JOURNAL OF EMERGING AND SELECTED TOPICS IN INDUSTRIAL ELECTRONICS, VOL. 4, NO.

2, APRIL 2023 549

Real-Time AI-Based Anomaly Detection and

Classification in Power Electronics Dominated Grids
Matthew Baker, Student Member, IEEE, Amin Y. Fard , Student Member, IEEE,
Hassan Althuwaini, Student Member, IEEE, and Mohammad B. Shadmand , Senior Member, IEEE

Abstract—Real-time anomaly detection system (ADS) and These challenges include control, privacy, stability, planning,
anomaly classification system (ACS) techniques are becoming a resilience, and cyber-physical security aspects [2].
crucial need for future power electronic dominated grid (PEDG). The PEDG resilient operation is constrained with predefined
Artificial intelligence techniques such as recurrent neural net-
works, specifically long short-term memory (LSTM) provide a boundaries for variables such as frequency, voltage, and power
promising solution to detect anomalies in power grids. The main quality indices. Any anomaly could cause destructive conse-
challenge is the implementation of these methods for real-time quences on the operation of the PEDG. These anomalies could
detection and classification for preventing catastrophic failure in have different characteristics such as component failure, faults,
PEDG. This article is addressing the challenge for detection and or malicious intrusive attacks. By considering the recent intru-
classification of anomalies in real-time in PEDG. The proposed
approach is based on integration of model predictive control (MPC) sive attempts on the power systems, cyber-physical security of
and LSTM for realizing real-time ADS and ACS. The LSTM the system gains extensive attention from governmental agencies
detection network can utilize the same time-series input data as and decision-making organizations, since it affects the national
the MPC, allowing for anomaly classification and correction. The security. Thus, prevention, detection [4], mitigation [5], and, if
proposed integrated LSTM-MPC approach has features of power needed, isolation of these cyber-physical intrusions are trending
electronics internal failure detection and corrective actions, which
is an important aspect in future PEDG to differentiate inverters research topics for modern PEDG. High penetration of DERs in
internal failures versus anomalies. Such internal failures include the grid requires various agents for proper functionality. These
open circuit fault that needs to be detected and classified from a include smart meters, PMUs, observers, smart loads, and the hi-
potential cyber-attack, allowing resilient operation of PEDG. The erarchical control architecture of the PEDG. Thus, it is necessary
proposed integrated LSTM-MPC scheme for real-time ADS and
to incorporate multilayer anomaly detection and classification
ACS scheme is tested on a realistic 14-bus system dominated with
inverters forming PEDG. (MADC) systems performing in multiple timescales. This is
inevitable to ensure the secure and seamless operation of the
Index Terms—Anomaly classification, anomaly detection, entire system.
cyberattack, fault-tolerance, inverter fault detection.
One of the most common cyber-attacks on power systems is
false data injection (FDI). FDIs target the data integrity of the
I. INTRODUCTION system which would push the control system to adopt inadequate
HE power system is experiencing a massive change to decisions [3], [4]. Attack consequences could vary, depending on
T be able to house ever-increasing distributed energy re-
sources (DERs) across the grid, to decrease the dependency
the number of compromised assets, attacker’s level of knowledge
on the system, attack propagation, and its persistence [5], [6].
on nonrenewable-based sources, and shift it toward renew- Even attacking one node of the system could impact the other
able resources, i.e., photovoltaics and wind. This new energy nodes across the PEDG, since the compromised DER may im-
paradigm is called power electronics dominated grid (PEDG) pact the neighboring nodes’ power quality and optimal operation
[1]. With all the benefits the PEDG introduces, it brings up of other DERs. The consequence of this attack is cascading
some challenges that need to be properly addressed prior to full failures across the grid. Thus, it is crucial for resilient operation
implementation of such a complex system in the real world. of PEDG to isolate the anomalous nodes to avoid cascading
failure across the PEDG that could results in catastrophic failure
Manuscript received 27 June 2022; revised 21 September 2022 and 26 October
of the power grid.
2022; accepted 17 November 2022. Date of publication 5 December 2022; date The supervisory layer of the PEDG must be able to dif-
of current version 23 March 2023. This work was supported by the Qatar ferentiate between these cyber intrusions and internal inverter
National Research Fund (QNRF is a member of Qatar Foundation) under
Grant NPRP12S-0226-190158. This paper was presented in part at the IEEE
failures to be able to make the most optimal decision for the
International Conference on Smart Grid and Renewable Energy, Doha, Qatar, PEDG. Without this knowledge, the supervisory layer will not
2022. (Corresponding author: Mohammad B. Shadmand.) be able to adopt the best decisions for a system with fleets
The authors are with the Department of Electrical and Computer Engi-
neering, University of Illinois Chicago, Chicago, IL 60607 USA (e-mail:
of smart assets across it. In the literature, numerous inverter
[email protected]; [email protected]; [email protected]; shadmand@uic. fault-tolerant control schemes are proposed [7], [8], [9], [10].
edu). The existing solutions are mainly focusing on detection, location
Color versions of one or more figures in this article are available at
https://doi.org/10.1109/JESTIE.2022.3227005.
of the faulty switch, and updating the switching sequences
Digital Object Identifier 10.1109/JESTIE.2022.3227005 accordingly. The missing piece of the puzzle in improving the

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
550 IEEE JOURNAL OF EMERGING AND SELECTED TOPICS IN INDUSTRIAL ELECTRONICS, VOL. 4, NO. 2, APRIL 2023

resiliency of the PEDG is differentiating these internal faults strong contender for scenarios in which the NN-based solutions
from malicious intrusions. Differentiating between these allows do not significantly increase computational time compared to
for optimal corrective action. A critically important missing conventional MPC and where real-time detection of anomalies
piece in exiting solutions is ensuring this process occurs in is of upmost importance. The work in [27] demonstrates an
real-time. If the MADC cannot detect and classify the anomaly ability for MPC-based controllers to detect open circuit faults
in a real-time-basis, the existing window for making proper using the MPC cost function. This control scheme utilizes the
decision might be missed. main core of MPC to generate modified switching sequences
The existing solutions for anomaly detection can be classi- constraint by the faulty switches without the need of major com-
fied into two main categories, which are system model-based putation. The model used for generating the switching sequences
techniques [11], [12], and data-driven schemes [13], [14], [15], is proven to be sufficient for open-circuit faults. However, for
[16], [17]. Generally, in model-based approaches, a model of more comprehensive anomaly detection and classification, this
the system must be developed, and the system parameters article proposes an NN module integrated within the MPC
must be estimated. Since there is not a training mechanism in framework. Therefore, a NN-based-detection scheme must pro-
model-based approaches, data mining is not needed. However, vide additional utility which the conventional MPC itself cannot.
the main drawbacks of model-based techniques is scalability of The main technical challenge for NN integration with MPC for
the detection mechanism, meaning that if the system changes, MADC is sufficient data collection for accurate classification in
a new model needs to be developed [18]. In model-based real-time.
schemes, for detection purposes, usually an observer is designed Existing solutions mostly focus on detecting FDI attacks and
to oversee the dynamic behavior of the system. The observers they neglect the internal failure of the DERs [2]. Also, the
for these model-based techniques could use Kalman filter [19], existing literature, mostly focuses on detection only and they
principle component analysis [20], and weighted least-square are not proposing any solutions after the detection is executed in
[21]. These model-based techniques could be implemented in real-time [2]. Additional utility is presented in classifying FDI
real-time applications. However, since they are highly depen- attacks and circuit faults with a single NN. The main contribution
dent on the mathematical model of the system, they are prone of this article is the realization of a framework, which is able to
to uncertainties, unforeseen disturbances, and computational detect and classify the anomalies across the PEDG in real-time.
burden as the system becomes more complex. Additionally, The proposed control scheme could be used at the primary layer
the model must be modified to account for any change in the of the control hierarchy of the PEDG to ensure detection and
PEDG. On the other hand, data-driven-based schemes can be classification of anomalies of any nature, i.e., inverter internal
used to perform effective detection and classification of wide failure or cyber-attack, in a real-time manner. Thus, integra-
range of anomalies such as FDIs in complex systems. As an tion of the proposed MADC in the primary control level at
example, artificial-neural-network-based approaches illustrated microseconds time scale results in smart self-learning inverters
proper performance for nonlinear systems. These tools are an operating at the grid-edge toward resilient and secure PEDG. The
appealing candidate for anomaly detection system (ADS) and considered attack model includes noise injection over the voltage
anomaly classification system (ACS) in PEDG. However, these and current measurements. The noise-based attacks are among
schemes require huge datasets for training purposes in order to the difficult-to-detect malicious activities, and their destructive
provide accurate performance. Other machine learning-based impacts are considerable, varying from service interruptions to
schemes such as k-nearest neighbor (k-NN), support vector ma- cascading failures across the grid. Specifically, the proposed
chine (SVM) [22], [23], and deep learning [24] suffer from the control framework employs recurrent neutral networks (RNN).
same drawbacks and real-time implementation. As illustrated in RNNs are utilized as feedforward NN techniques classification
[14] and [25], for power system applications, SVM illustrates proved unable to determine fault classification [28]. Thus, for
better performance than k-NN, however, the performance of anomaly classification a deep NN topology is preferred. This
SVM highly depends on kernel type selection. In [26], a deep- category of neural networks are among the top candidates to per-
learning-based method, which is modified version of WaveNet, form classifications on time-dependent variables while taking
was proposed and applied to IEEE 14-bus system while high into the account the previous inputs of the system [29]. Among
penetration of renewable resources are considered. With all these different variants of RNNs, the long short-term memory (LSTM)
methodologies, real-time implementation and scalability of the employs an internal memory to create the predictions for next
ADC and ACS are the remailing challenges. steps of the system [29], a perfect match to be integrated with
Incorporating neural network (NN) based scheme, thus, has MPC with long-time horizon prediction and optimization for the
an opportunity to cooperate with control methods to ensure application in hand and real-time implementation. This internal
an impactful contribution. In this article, the inherent charac- memory with better predictive capabilities makes the LSTM the
teristics of model predictive control (MPC) is leveraged for best option for fault classification within a system as complex as
effective real-time integration of NN-based network for realizing the PEDG. The initial part of the proposed framework employs
MADC in highly nonlinear PEDG. Conventional finite-set MPC LSTM in conjunction with MPC to perform anomaly detection
determines optimal switching sequences based on the model and classification for inverters at the primary layer controller.
and a cost function for optimization. The main focus of this The considered inverter topology in this article is cascaded
article is to develop a framework for integration of MPC and multilevel inverted (CMI) due to its fault-tolerant capability
NN for real-time MADC. The proposed work seeks to be a [27], [30]. It is worth mentioning that the same approach could
BAKER et al.: REAL-TIME AI-BASED ANOMALY DETECTION AND CLASSIFICATION IN POWER 551

be employed for any other inverter architecture to reach to a TABLE I

CMI OPERATIONAL PARAMETERS
similar performance. After detection and classification, the pro-
posed LSTM-MPC framework adopts corrective actions to per-
forms self-healing model predictive control to ensure fault-ride
through capability in the case of internal failure of the inverter.
The integrated LSTM-based MADC with self-healing MPC
enhances the resiliency and cybersecurity of the PEDG in the
case of inverter internal failure and noise-injection-based cyber
intrusions. Thus, to put it in a nutshell, the major contributions
of the article are as follows.
1) An integrated MPC-LSTM technique for detecting and
to prevent impact on nearby nodes and other healthy DERs.
classifying anomalies in real-time followed by a self-
Therefore, corrective action is taken to prevent further attacks
healing solution and/or isolating the intruded DER in the
on the grid and ensure resilient operation.
network to ensure the resilient operation of the system by
The utility of the presented NN scheme is twofold. First, the
preventing cascading failure scenarios.
NN scheme can detect potential anomalies in real-time. The goal
2) Comprehensive approach for data collection and training
of the NN detection scheme is to, in less than one line cycle,
of LSTM for anomaly classification and effective mitiga-
determine whether anomalous behavior exists and classify it as
tion in real-time, which is crucial for the application in
either a fault or cyber-attack; the unique contribution of this
hand.
article. This quick detection allows for the supervisory layer
3) Analysis and case studies to demonstrate the scalability of
to address anomalies before the impact is visible on the entire
proposed LSTM-MPC.
PEDG. The second feature of the NN approach is efficient use
4) Discussion on the hyperparameter selection process as a
of computational resources. When the NNs are trained to reduce
guideline for design of LSTM.
the computational burden, such as by having a single NN system
The rest of this article is organized as follows. Section II
detect numerous anomalies with the same input data, the local
is a description of the PEDG system, LSTM, and MPC used.
LSTM-MPC is operating efficiently. It also allows for further
Section III details the collection of training data and the opera-
anomaly detection in future works where the single network
tion of the proposed LSTM-MPC. Section IV demonstrates the
approach can be used to classify more anomalies without ex-
case studies. Finally, Section V concludes this article.
treme increase in the number of neurons. Thus, this two-level
approach both on the local and supervisory level allows for
II. SYSTEM DESCRIPTION discrete implementation. The proposed LSTM-MPC MADC
framework is implemented on each phase of a CMI with the
The proposed LSTM-MPC detects anomalies and, with the
properties shown in Table I. A complete overview of the system
guidance of the supervisory controller, determines whether the
implemented is seen in Fig. 1.
anomaly would allow for continued operation of the affected
DER. If the anomaly presents a larger threat, it may ultimately
A. Communication Infrastructure of PEDG
lead to isolation from rest of the grid to ensure harmonious
operation of other DERs and nearby nodes. While the proposed The proposed control framework consists of MPC and NN
LSTM-MPC can be applied to any power system, the proposed parts working collaboratively to ensure short-term and long-term
scheme is tested primarily using a 14-Bus system with high voltage and frequency stabilities of the PEDG. The MPC-based
penetration of DERs as the PEDG, and a seven-level cascaded controller is utilized in the local controller. The local controller
multilevel inverter as power electronic interface of each DER. has the responsibility to receive the power set-points from the
To detect an anomaly, the NN system receives input data at each upper control layers, convert them to switching signals, and
sampling instance Ts . This data is fed into the LSTM system, turn power electronic switches ON and OFF in the power stage
which is detecting for one of two potential anomalies considered accordingly. The reasons for opting MPC for the local controller
in this article. Fig. 1 illustrate the schematic of the 14-Bus system include: easy to leverage into constrained multiobjective control
with proposed LSTM-MPC. problems, eliminating the hurdle of PID controller tuning for
Two classifications of anomaly are presented. The first detects different operating points, better power quality, and proper for
whether a physical converter failure has occurred. In this article, inner control loop with high compatibility with outer control
it is an open circuit switch failure, a leading cause of semi- loops. Considering these features, MPC is the best candidate for
conductor failure in power electronic systems [31]. This type the local layer of the proposed controller. However, the local
of anomaly, once detected, allowing for continued operation. controller does not have proper knowledge of the entire PEDG,
Thus, the supervisory controller can provide a corrective action thus, the communication layer is required between the local layer
to the MPC and continue operation. The second type of anomaly of the controller and the supervisory layer.
represents an FDI. This represents a potential cyber-attack on the The supervisory layer is responsible for overseeing the en-
current measurement sensors. When such a detection occurs, tire grid including generations, consumptions, reserved power,
a DER can no longer be treated as a trusted source to the and power loss. Using all the gathered information, this layer
supervisory layer and, thus, must be isolated from the PEDG assigned the power set-points for each individual inverter.
552 IEEE JOURNAL OF EMERGING AND SELECTED TOPICS IN INDUSTRIAL ELECTRONICS, VOL. 4, NO. 2, APRIL 2023

Fig. 1. Complete system under study for this work. The proposed neural network scheme exists as a per phase detection of the cascaded multilevel inverter at
DER 4. This anomaly detection network is connected to the higher level 14-bus system at Bus 3. Anomalies are detected initially at the local controller, then verified
at the supervisory level before corrective action is taken.

Various communication protocols are developed for smart grid loop or communication link of each individual inverter. In that
applications for secure communications with minimal latency case, although the supervisory layer has assigned specific set
[32], [33], [34]. These protocols include, IEEE 802.15.4 (Zig- points to the MPC-based layer of the framework, the noise-based
Bee), IEEE 802.11 (Wireless LAN (WLAN) or Wi-Fi), IEEE- FDI attack prevents proper power injection when the measured
802.16 (WiMAX), GSM and GPRS, and DASH7. But even with current is incorrect. This means the balance between generation
these improvements in communication protocols, the commu- and consumption will deteriorate, and voltage or frequency
nication layer is the most vulnerable layer of the PEDG. stabilities of the system will be endangered. On the other hand,
FDI attacks are possible attacks on the communication layer. the supervisory layer of the PEDG needs an accurate model of
In this article, the attacks considered are noise-based FDI at- the entire system to perform a load flow analysis to determine the
tacks on the sensors measuring the current injected by the DERs. voltage of each node and consequently determine the power set-
When there is communication between a supervisory controller points for each individual inverter. Within the concept of PEDG,
and local controller, an attacker’s plan can be to interfere with where the distributed resources and consumers are changing
the communication layer and inject noise into the feedback dynamically according to the time of the day and season, having
BAKER et al.: REAL-TIME AI-BASED ANOMALY DETECTION AND CLASSIFICATION IN POWER 553

grid-tied inverter applications as highlighted in [35] and [36].

The proposed LSTM-MPC is focused on leveraging the inherent
characteristics of state-of-the-art MPC with additional features.
The MPC operates with a second order generalized integrator
phase lock loop, which generates the grid angle needed for utiliz-
ing active and reactive power references, which are synchronized
with the grid. The reference current i∗k in (3) is determined from
Fig. 2. Fundamental LSTM Cell.
the active and reactive power references, and predicted current
ik+1 in (4) is determined from using the forward Euler method
a precise model of the system is almost impossible. This will
i∗k = i∗d,k sin(θk ) + i∗q,k cos(θk ) (3)
paralyze the supervisory layer of the PEDG in the case of FDI  
attacks. For this reason, in the proposed control framework the R TS
ik+1 [s] = 1 − Ts ik + [vinv,k+1 [s] − vk ]. (4)
supervisory layer of the controller is equipped with an NN-based L L
algorithm. It performs the responsibilities of the supervisory
The cost function (5) is determined using i∗k and ik+1
layer without an accurate model of the entire integrated system.
The proposed NN-based supervisory layer not only assigns J[s] = |i∗k − ik+1 [s]| ∀s ∈ N ≤ 64 s.t. state s1x64 [s] = 0
the set points for each individual inverter, but also detects
sk+1 = arg minimize(J[s]) (5)
noise-based cyber intrusion. The proposed LSTM-MPC-based
control framework enables the PEDG to differentiate between where s is a specific switching state, and the array states lists,
inverter internal failures and FDI attacks. This will enable the which of the switching sequences are permissible. Under normal
supervisory layer to take corrective actions using an MPC-based operating conditions, states is a 1 × 64 array representing
controller in the case of power stage failure of the inverter, or all valid switching states for a seven-level CMI. Whichever
label the inverter as compromised in the case of FDI detection. switching state minimizes the cost of (5) is applied to the
inverter.
B. LSTM Description and Formulation
III. NEURAL NETWORK DETECTION SCHEME: LSTM-MPC
RNNs are a category of NNs, which analyze sequential data.
Whereas more traditional methods, such as the feed forward A. Data Collection
neural network, produce outputs solely based on the current
The LSTM-MPC is created by training data collected from
inputs to the NN. RNNs are trained to also incorporate past
the MPC system described in Section II-B. The diagram of the
states of the network. These networks are strong contenders
LSTM-MPC operational principle is shown in Fig. 3. The aim in
for anomaly classification, as instantaneous data may or may
data collection for this system is to ensure real-time detection is
not be anomalous depending on where it occurs in a sequence.
possible, and no additional measurements or sensors are neces-
In this article, the LSTM RNN topology is used to enable the
sary beyond those needed by conventional MPC. Additionally,
implementation of the LSTM cell, as shown in Fig. 2. The two
proper simulation of the grid voltage during the collection of
activation functions used in each LSTM cell are the sigmoid and
training data will allow for scalability of the LSTM-MPC into a
hyperbolic tangent functions, represented in
wider range of power system topologies. If the system needs to
1 be retrained for changing system properties such as the number
σ(x) = (1)
1 + e−x of switches, power level, and voltage level the same process as
ex − e−x described here can still be utilized. This eases the requirements
tanh(x) = (2) for data collection and allows for more general usage of this
ex + e−x
system and training process. With this design principle in mind
for collecting data, the data considered as input to the NN are
Fig. 2 demonstrates each LSTM cell has 12 parameters deter-
1 × 7 array INP given by
mined through the training process. In this article, the structure  
of the LSTM network is constant; there is no variance in the i[k] , i[k−1] , Vinv[k] , Vinv[k−1]
IN P = (6)
number of features, LSTM layers, dropout rate, or classification OS[k] , OS[k−1] , Vgrid[k]
layer. The major investigation for the training process is the
where at sampling instant k, i[k] and i[k-1] are the inductor
number of hidden units in each LSTM layer. The training process
currents, Vinv[k] and Vinv[k-1] are the inverter voltages, OS[k] and
accounting for changes in hidden states is detailed in Section III.
OS[k-1] are the optimal switching states applied by the MPC, and
Vgrid[k] is the grid voltage at the point of common coupling.
C. Model Predictive Control Formulation The voltage and current measurements are the same as the
The MPC applied for the CMI is shown in Fig. 1. This system measurements used in the MPC cost function formulation. The
is an adaptation of the finite-set MPC, which can eliminate optimal states are the output of the MPC formulation and, thus,
switching sequences made impossible by an open circuit fault. the switching states applied. Therefore, the variables chosen
This MPC scheme, as summarized in this section, is similar to are impactful to detecting anomalies involving FDI and fault
the state-of-the-art approaches, with wide range of benefits for failures as it will impact the data of INP adversely and noticeably.
554 IEEE JOURNAL OF EMERGING AND SELECTED TOPICS IN INDUSTRIAL ELECTRONICS, VOL. 4, NO. 2, APRIL 2023

Fig. 3. Operational diagram of the LSTM-MPC. The input data of INP is collected in step one. Then, is given to the 2-layer LSTM-MPC to create a probability
of anomaly in step two. The number of hidden units in each layer of step two is variable and based on the results of Table II. The final step triggers corrective action
when the probability passes the trigger threshold, as explained in Section III.

The aim of the NN training process is to ensure proper network 2) False Data Injection Data: Data collection for the FDI
structure and training procedure, which can detect these anoma- attack follows a similar setup as the open circuit fault data. To
lies in real-time and accurately. model an FDI attack, white noise is injected into the current sen-
The DER as described in Section II and Table I is created sor, which outputs normally distributed random noise. The noise
in MATLAB/Simulink version 2021b to collect training data. has a power of 0.1. The attack occurs at 0.5 s, halfway through the
The Simscape Specialized Power systems toolbox is utilized for data collection time of 1 s. To accumulate data for a wide range
solving power flow equations at each discrete step instance. The of situations, the amplitude of the noise and the power reference
data creation is executed in this environment as it would also be are varied. The output of the noise generator is multiplied by a
used to verify the network operation in the case studies section. constant NFDI , which varies from 0 to 2 in increments of 0.05.
Thus, a comprehensive dataset is collected from a realistic PEDG A total of 32 200 000 data points from this collection process
bus as a true replica of DER in real world. In fact, this model can are used for the FDI anomaly detection NN scheme. The data
be considered as the digital twin of an actual physical system are grouped into blocks the same as the fault data.
for collecting comprehensive data set for training purpose of
NN network. Matching the training data accumulation to the
method in which the system is tested and operated eliminates B. Training
additional noise or unrealistic data, which might potentially 1) Neural Network Topology: With the training data col-
skew the training process. lected, the next step is to describe the operation LSTM system.
1) Open Circuit Fault Data: To collect data for an internal The fully trained MADC is able to determine when an anomaly is
fault anomaly, the DER operates as a three-phase system with probable, which of the two classes the anomaly is in, and signals
a seven-level CMI. An open circuit fault is simulated through to the supervisory controller that corrective action is needed. The
the opening of a breaker in series with the faulted switch. LSTM-MPC has three classes to represent the training data,
Thus, for open circuit faults, the potential number of faulty “normal,” “fault,” and “cyberattack.” Should different data be
switches in each inverter is 36. To ensure sufficient training collected to train another LSTM-MPC, the number of classes
data is collected, the simulation is run 36 times with a different will change to match the additional data and classifications. To
switch causing an open circuit fault in each instance. For the data create more balanced training data, half of the “normal” class
collection scheme, the simulation in each iteration is run for 1 s of data from each data collection process is excluded from the
of simulation time where the open circuit fault occurs at 0.5 s. As training data set. The final data set classes are approximately
the system operates with a line frequency of 60 Hz, 30 complete 29.2% “fault,” 37.4% “cyberattack,” and 33.3% “normal.” The
line cycles occur for each open circuit fault. This amount of data goal of the LSTM-MPC is to have an accuracy above 90%. This
is sufficient for NN training and not so large as to unnecessarily data is sufficiently balanced to meet this goal since 90% accuracy
increase training time. The seven inputs of INP are collected at a requires high accuracy of each class and no individual class can
sampling rate of 100 kHz, as well as the 1 × 36 array denoting the dominate training. After detection, the supervisory controller
status of each fault. Should a controller need the NN to operate at can confirm the anomaly with a mirrored NN, should extra
a slower frequency, the entire dataset can be under sampled to the validation be required. Finally, corrective action is implemented.
required frequency. Combining the sample rate simulation time 2) Hyperparameters Selection: The LSTM network is
and number of iterations along with the input and class data for trained in MATLAB 2021b. Determining the exact hyperparam-
the network, a total of 25 200 000 data points from this collection eters of the system, specifically the number of LSTM units in
process are used for training the fault anomaly detection NN each of the two layers, is cause for further investigation to the
scheme. These data are grouped into blocks 50 sample long specific control system. In this article, various hyperparameters
to minimize the sample size needed for classification, leading are tested and examined to pick the most proper NN system
to improved response time. Finally, data are randomly split into for anomaly detection. Top priorities of operation in the 14-bus
training, validation, and testing subsets, which are divided as system are accuracy, detection speed, computational effort, and
70%, 15%, and 15% of the total data acquired, respectively. scalability. Accuracy is crucial for resilient operation of PEDG,
BAKER et al.: REAL-TIME AI-BASED ANOMALY DETECTION AND CLASSIFICATION IN POWER 555

TABLE II TABLE III

LSTM HYPERPARAMETERS LSTM TRAINING PARAMETERS

generalization allows for consistent data comparisons across

machines of differing computational speeds.
From the results generated in Table II, an optimal LSTM
network is selected for implementation into the final system. The
network selected is the least computationally expensive network
with an accuracy of at least 90%. Thus, the 10 × 10 network
is used; its accuracy is 96.66% with a computational burden of
only 59.39%.
fast detection speeds allow the MADC to take corrective action
fast enough to prevent a violation of grid standards, low com- C. Corrective Action
putational effort enables the anomaly detection network to run
After the LSTM-MPC has been trained, it is incorporated
in parallel with the MPC, and scalability allows for deployment
into the system as the “inverter neural network” shown in
even in changing grid topology, perhaps due to more DERs com-
Fig. 1. During operation, the inputs to the LSTM-MPC are the
ing online. Finally, a scalable system Therefore, the operation of
data of INP at each sampling instance. The trained 10 × 10
the MPC will be used as a benchmark for the NN operation. The
LSTM-MPC produces a probability an anomaly has occurred.
lower the execution rate, the better the NN system. While other
The three trained classification for the network are “normal,”
hyperparameters such as batch size, learning rate, and activation
“fault,” or “cyberattack.” The LSTM-MPC produces a number
function, may be considered necessary for other applications,
from 0-1 for each class, where the higher the number is, the
the concerns of this article are primarily on execution rate of
higher the probability the data is of the specified class. To
the scheme once the NN has been developed. Therefore, the
prevent false positives from affecting an operational DER, the
hyperparameters adjusted are the number of neurons in each of
outputs of the fault detection network are fed into a moving
the two layers of the LSTM.
average filter to filter out potential false positives and ensure a
Hyperparameter tuning is performed by adjusting the number
fault is only triggered when enough successive terms indicate
of LSTM units in each layer. It is determined two layers are
high probability of a fault. Then, if the moving average filter
sufficient to correctly identify faults. In this article, the number
exceeds a threshold Vt , the anomaly fault flag is tripped, and
of units will vary from 1 through 100 for the first layer, and
the supervisory controller is notified. Before Vt is exceeded
the second layer will consist of either an equal number of units,
for either “fault” or “cyberattack,” the system is assumed to
or half (rounding up) the units of the first layer. The results
be “normal”. For this article, Vt = 0.16 for the fault detection
of the hyperparameter tuning are seen in Table II. This table
and Vt = 0.45 for the cyberattack detection. This threshold is
provides the hyperparameters of various LSTM networks along
manually set low enough to ensure fast response time, but high
with the accuracy and computational effort required to execute
enough to prevent false positives. The threshold can be adjusted
each NN. Both the number of neurons and number of layers are
depending on the accuracy of the NN, as well as the desired
seen in the leftmost column. For example, “10 × 5” denotes
response time of the system operator. After Vt is exceeded, the
a two-layer network of 10 and 5 neurons, respectfully. The
supervisory controller confirms the anomaly and takes corrective
“anomaly detection accuracy” is the accuracy of the network
action. Two corrective actions are implemented here: when an
using the testing data after five epochs of training. All other
internal fault occurs, the corrective action detailed in [27] is
hyperparameters of the system remain constant, as described in
executed, where states is updated to remove faulty switching
Table III.
states. If an FDI attack is determined, the bus is isolated and
The computational rate is determined using the Performance
power references are adjusted, as explained in Section IV and (7).
Advisor Simulink Profiler in MATLAB/ Simulink. To account
for variances in computational time due to hardware variance
on the testing computers, the NN processing time is shown as IV. DISCUSSION AND VALIDATION
a percentage of the processing time to enact the computations To verify the effectiveness of the proposed LSTM-MPC as an
necessary for the MPC. For example, if the MPC required effective MADC, anomalies are intentionally created in Phase A
100 s of total execution time in the profiler, and the NN system of DER 4 of the 14-bus system. These can be either open circuit
needed 75 s, the computational burden is considered 75%. This faults, or FDI attacks.
556 IEEE JOURNAL OF EMERGING AND SELECTED TOPICS IN INDUSTRIAL ELECTRONICS, VOL. 4, NO. 2, APRIL 2023

Fig. 5. Results of the second case study, which examines the impact of the
Fig. 4. Results of the first case study, verifying the neural network’s ability
fault correction network across a 14-bus system. In both simulations, an open
to detect and correct an open circuit fault in Phase A of DER 4 of the 14-bus
circuit fault occurs in S1 at t = 2 s. (a) Detection network is present. It detects and
system. In both simulations, an open circuit fault occurs at t = 2 s. When the
applies corrective action to mitigate the fault in under one line cycle, allowing
fault detection network has high enough confidence to determine a fault, the
for continued operation. (b) No mitigation technique is present. The low power
Fault Status flag goes HIGH and the states array is updated to exclude switching
quality after the fault forces DER 4 to isolate from the 14-bus after five line
states requiring the faulty switch to be closed. (a) Fault occurs in S1. (b) Fault
cycles. Since the synchronous generator is unable to compensate, the system
occurs in S2.
frequency collapses.

A. Fault Detection Verification

The first case study is to verify the LSTM-MPC can correctly
identify a fault in both the positive and negative voltage levels of
Phase A. In this article, Pref is 16 kW and Qref is 0 kW. A fault
occurs at t = 2 s for each test. In test one, the fault occurs in switch
S1 and in the second test the fault occurs in switch S2 . When
the output of the moving average filter exceeds Vt the updated
states matrix is applied preventing any faulty switch from being
utilized in the “closed” state. The results of each test are seen
in Fig. 4. When S1 fails, the fault is detected and corrected in
14.32 ms; when S2 fails, the fault is detected and corrected in Fig. 6. Results of the third case study. Here, the robustness of the NN during
4.08 ms. As seen in each example, the proposed LSTM-MPC FDI is shown, as changes in the power reference do not cause. When FDI does
occur, the system responds promptly by alerting the supervisory controller,
scheme can quickly identify faults in less than one line cycle. which is then able to isolate the attacked DER from the system to minimize
This allows quick identification for each fault and prevents the the effects on the rest of the system.
need to isolate the faulty DER due to inferior current quality for
grid standards, thus providing real-time solution at the grid-edge.
the DER becomes zero. With the synchronous generator unable
B. Impact of Compromised DER Isolation to provide the necessary power to regulate frequency, it rapidly
denigrates and the frequency of the entire system collapses.
To demonstrate the added benefit of the open circuit fault Therefore, the LSTM-MPC scheme ensures the DER does not
detection network, the second case study compares the proposed cause system collapse even when an open circuit fault occurs in
network to a system without any fault detection. In this case a critical DER.
study, DER 4 is considered a major contributor to power gener-
ation in the 14-bus system. The Pref is increased to 20 kW, the
C. Changing Power Reference in Response to Anomalies
14-bus system is operated during low loading conditions, and the
synchronous generator has a maximum capacity of 100 kW. For The case study here demonstrates the LSTM-MPC scheme’s
simplicity, all other DER generators remain constant. Therefore, ability to detect an FDI attack, as well as the system robustness
failure of DER 4 will lead to an inability to supply power across to changing Pref values. A change in Pref can represent an
the system as the synchronous generator is unable to provide anticipated load increase, change in solar irradiance if the DER is
the needed power and frequency support. In this case study the a PV system, or any other ramp increase in power generation. In
open circuit fault occurs in switch S1 at 2 s. Fig. 5(a) shows this article, the system operates under the same initial conditions
the LSTM-MPC system operating to take corrective action to presented in Section VI-B. At t = 2 s, the Pref for DER 4 ramps
ensure proper operation despite the fault. Fig. 5(b) shows the up at a rate of 5000 kW/s. At t = 3 s, Pref ceases to increase. Once
alternative, where the open circuit fault occurs at 2 s. After this state is reached, an FDI attack occurs on the current sensor
five-line cycles, DER 4 must isolate from the 14-bus network when t = 3.25 s. The results of this case study are shown in
due to unacceptable power quality. Thus, the power injected by Fig. 6. As depicted, the proposed LSTM-MPC scheme operates
BAKER et al.: REAL-TIME AI-BASED ANOMALY DETECTION AND CLASSIFICATION IN POWER 557

TABLE IV
BUS VOLTAGES DURING FDI ATTACK

Fig. 7. Fourth case study demonstrates the impact of corrective action after an
FDI attack is determined. The bus voltage and current injected by Phase A are
as designed. No false positive occurs during the change in Pref , shown at the bus with the FDI. (a) Without the anomaly corrective action. (b)
With anomaly detection and corrective action. After isolation from the system,
and the proposed scheme determines a FDI in 13.4 ms after it the system with the NN scheme has a smaller impact on the bus voltage due to
occurs, which is less than one line cycle. the corrective action taken by the supervisory controller after the fault.

D. Impact of the LSTM-MPC on the System Resiliency

The fourth case study investigates the effect of isolating a DER
under attack. In this scenario, a FDI attack occurs at DER 4 at
t = 4.5 s, injecting noise into the current sensor. The increased
harmonics caused by the noise injection and the reduced power
quality of the DER causes breakers to trip, isolating the DER
from the grid. The scenario is tested under two conditions, with
and without the LSTM-MPC scheme. Without the LSTM-MPC,
the supervisory layer is not alerted to the anomaly and mismatch
occurs between the power generation and demand. With the
proposed LSTM-MPC, the anomaly is properly identified and
verified by the supervisory controller 12.0 ms after the attack.
The corrective action in this scenario accounts for the removal of
DER 4 and its power contributions are evenly divided amongst
the remaining DERs using
∗
∗
Pref,n
Pref,n = Pref,n + Pref,4 (7)
Pavalible Fig. 8. Results of the fifth case study. The topology of the power system is
∗ expanded into a total of 18 buses by introducing additional loads and distribution
where Pref,n is the power reference for DER n, Pref,nis the lines. (a) Open circuit fault occurs at t = 3.5 s. (b) FDI attack occurs at t = 3.5 s.
power reference of DER n before the attack, and Pavalible is the Despite the change in system architecture, the LSTM-MPC remains capable of
power generated by the safe DERs before the attack. Another detecting both anomalies and triggering corrective action.
approach can replace (7) should it be desired by the supervisory
controller. The results of the system under both scenarios are
Table I. These parameters accurately describe the low voltage
recorded in Table IV and Fig. 7. Table IV displays the voltage at
operation of the 14-bus system. This match allows for easy
each bus before the attack, as well at the voltage at 5 s. With the
implementation of the LSTM-MPC in DER 4 at Bus 3. This
LSTM-MPC and corrective action taken, the average voltage
implies the LSTM-MPC is properly trained to operate in any
deviation is 0.7% the p.u. voltage before the fault, compared
power system topology; with the important assumption that the
to an average deviation of 2.2% without the corrective action.
bus the LSTM-MPC is connected to is accurately reflected in
Additionally, with the corrective action, all buses are >0.9 V
the training data. Changing the topology of the system should
p.u., whereas buses 1 and 2 do not meet this criterion without
not prevent the LSTM-MPC from operating properly.
the LSTM-MPC network.
To verify this assumption, the 14-bus system is expanded to 18
buses in this case study. The LSTM-MPC is tested in conditions
E. Scalability Analysis
where the DER matches Case Study 1. Two anomaly scenarios
The previous case studies incorporate the LSTM-MPC to a are tested. First, a switch failure in S1 occurs at 3.5 s, and second,
DER connected to an IEEE 14-bus system. The DER is rated an FDI attack occurs in the current sensor of Phase A at 3.5 s.
for a grid voltage of 220 V and 60 Hz, as per the parameters in The results of these tests are seen in Fig. 8. Similar to the case
558 IEEE JOURNAL OF EMERGING AND SELECTED TOPICS IN INDUSTRIAL ELECTRONICS, VOL. 4, NO. 2, APRIL 2023

studies involving the original 14 bus system, the LSTM-MPC is [10] R. Peuget, S. Courtine, and J.-P. Rognon, “Fault detection and isolation
capable of detecting both anomalies in less than one line cycle. on a PWM inverter by knowledge-based model,” IEEE Trans. Ind. Appl.,
vol. 34, no. 6, pp. 1318–1326, Nov./Dec. 1998.
The fault is mitigated after 11.10 ms and the FDI after 11.26 ms. [11] K. Manandhar, X. Cao, F. Hu, and Y. Liu, “Detection of faults and
These results support the premise that the topology of the power attacks including false data injection attack in smart grid using Kalman
system is of little impact to the well trained LSTM-MPC. If the filter,” IEEE Trans. Control Netw. Syst., vol. 1, no. 4, pp. 370–379,
Dec. 2014.
DER is connected to a bus where training data is collected with [12] W. Ao, Y. Song, and C. Wen, “Adaptive cyber-physical system attack
appropriately, the system can be scaled easily to detect these detection and reconstruction with application to power systems,” IET
anomalies locally. Control Theory Appl., vol. 10, no. 12, pp. 1458–1468, 2016.
[13] A. Abdullah, “Ultrafast transmission line fault detection using a DWT-
based ANN,” IEEE Trans. Ind. Appl., vol. 54, no. 2, pp. 1182–1193,
V. CONCLUSION Mar./Apr. 2018.
This article presents an anomaly detection and correction [14] M. Ozay, I. Esnaola, F. T. Yarman Vural, S. R. Kulkarni, and H. V. Poor,
“Machine learning methods for attack detection in the smart grid,” IEEE
scheme for PEDG. The proposed MADC and corrective action Trans. Neural Netw. Learn. Syst., vol. 27, no. 8, pp. 1773–1786, Aug. 2016.
is based on an integration of LSTM and MPC, which features [15] J. Yan, H. He, X. Zhong, and Y. Tang, “Q-learning-based vulnerability
real-time solutions for enhancing the resiliency of PEDG when analysis of smart grid against sequential topology attacks,” IEEE Trans.
Inf. Forensics Secur., vol. 12, no. 1, pp. 200–210, Jan. 2017.
anomalies occurs. In the proposed scheme, when an anomaly [16] S. A. Foroutan and F. R. Salmasi, “Detection of false data injection
is detected, the supervisory layer controller can implement attacks against state estimation in smart grids based on a mixture Gaussian
corrective action to ensure the power system continues proper distribution learning method,” IET Cyber-Phys. Syst., Theory Appl., vol. 2,
no. 4, pp. 161–171, 2017.
operation. The data collection technique and training process of [17] F. Li et al., “Detection and identification of cyber and physical attacks on
the long short-term memory network is described as well as the distribution power grids with PVs: An online high-dimensional data-driven
methodology of selecting proper hyperparameters. The training approach,” IEEE J. Emerg. Sel. Topics Power Electron., vol. 10, no. 1,
pp. 1282–1291, Feb. 2022.
data is selected to increase scalability of the LSTM-MPC should [18] O. Boyaci et al., “Graph neural networks based detection of stealth false
the network topology expand as more DERs are integrated into data injection attacks in smart grids,” IEEE Syst. J., vol. 16, no. 2,
the PEDG. An optimal LSTM network is used to test and validate pp. 2946–2957, Jun. 2022.
[19] H. M. Khalid and J. C.-H. Peng, “Immunity toward data-injection attacks
the operation of the network. The case studies verify the system using multisensor track fusion-based model prediction,” IEEE Trans.
ability to allow robust deployment after anomalies occur, and the Smart Grid, vol. 8, no. 2, pp. 697–707, Mar. 2017.
impact of the corrective actions by the supervisory controller is [20] Z.-H. Yu and W.-L. Chin, “Blind false data injection attack using PCA
approximation method in smart grid,” IEEE Trans. Smart Grid, vol. 6,
demonstrated. no. 3, pp. 1219–1226, May 2015.
[21] R. Deng, G. Xiao, and R. Lu, “Defending against false data injection
ACKNOWLEDGMENT attacks on power system state estimation,” IEEE Trans. Ind. Inform.,
vol. 13, no. 1, pp. 198–207, Feb. 2017.
The statements made herein are solely the responsibility of [22] S. Zhang, Y. Wang, M. Liu, and Z. Bao, “Data-based line trip fault
the authors. prediction in power systems using LSTM networks and SVM,” IEEE
Access, vol. 6, pp. 7675–7686, 2018.
REFERENCES [23] H. Goyel and K. S. Swarup, “Data integrity attack detection using ensemble
based learning for cyber physical power systems,” IEEE Trans. Smart Grid,
[1] A. Khan, M. Hosseinzadehtaher, M. B. Shadmand, S. Bayhan, and to be published, Aug. 2022.
H. Abu-Rub, “On the stability of the power electronics-dominated grid: A [24] G. Abdelmoumin, D. B. Rawat, and A. Rahman, “On the performance of
new energy paradigm,” IEEE Ind. Electron. Mag., vol. 14, no. 4, pp. 65–78, machine learning models for anomaly-based intelligent intrusion detection
Dec. 2020. systems for the Internet of Things,” IEEE Internet Things J., vol. 9, no. 6,
[2] O. H. Abu-Rub, A. Y. Fard, M. F. Umar, M. Hosseinzadehtaher, and pp. 4280–4290, Mar. 2022.
M. B. Shadmand, “Towards intelligent power electronics-dominated grid [25] J. Yan, B. Tang, and H. He, “Detection of false data attacks in smart grid
via machine learning techniques,” IEEE Power Electron. Mag., vol. 8, with supervised learning,” in Proc. Int. Joint Conf. Neural Netw., Jul. 2016,
no. 1, pp. 28–38, Mar. 2021. pp. 1395–1402.
[3] A. S. Musleh, G. Chen, and Z. Y. Dong, “A survey on the detection [26] F. Almutairy, L. Scekic, R. Elmoudi, and S. Wshah, “Accurate detection
algorithms for false data injection attacks in smart grids,” IEEE Trans. of false data injection attacks in renewable power systems using deep
Smart Grid, vol. 11, no. 3, pp. 2218–2234, May 2020. learning,” IEEE Access, vol. 9, pp. 135774–135789, 2021.
[4] R. Das, G. Karmakar, J. Kamruzzaman, and A. Chowdhury, “Measuring [27] M. Easley, M. Baker, A. Khan, M. B. Shadmand, and H. Abu-Rub, “Self-
trustworthiness of smart meters leveraging household energy consump- healing model predictive controlled cascaded multilevel inverter,” in Proc.
tion profile,” IEEE J. Emerg. Sel. Topics Ind. Electron., vol. 3, no. 2, IEEE Energy Convers. Congr. Expo., 2019, pp. 239–244.
pp. 289–297, Apr. 2022. [28] M. W. Baker, H. Althuwaini, and M. B. Shadmand, “Artificial intelligence
[5] A. Sargolzaei, A. Abbaspour, M. A. Al Faruque, A. Salah Eddin, and based anomaly detection and classification for grid-interactive cascaded
K. Yen, “Security challenges of networked control systems,” in Sustainable multilevel inverters,” in Proc. 3rd Int. Conf. Smart Grid Renewable Energy,
Interdependent Networks. Berlin, Germany: Springer, 2018, pp. 77–95. Mar. 2022, pp. 1–6.
[6] K. Xiahou, Y. Liu, and Q. H. Wu, “Decentralized detection and mitigation [29] V. Sze, Y.-H. Chen, T.-J. Yang, and J. S. Emer, “Efficient processing of
of multiple false data injection attacks in multiarea power systems,” IEEE deep neural networks: A tutorial and survey,” Proc. IEEE, vol. 105, no. 12,
J. Emerg. Sel. Topics Ind. Electron., vol. 3, no. 1, pp. 101–112, Jan. 2022. pp. 2295–2329, Dec. 2017.
[7] B. Lu and S. K. Sharma, “A literature review of IGBT fault diagnostic and [30] N. A. Rahim, M. F. M. Elias, and W. P. Hew, “Transistor-clamped H-bridge
protection methods for power inverters,” IEEE Trans. Ind. Appl., vol. 45, based cascaded multilevel inverter with new method of capacitor voltage
no. 5, pp. 1770–1777, Sep./Oct. 2009. balancing,” IEEE Trans. Ind. Electron., vol. 60, no. 8, pp. 2943–2956,
[8] A. M. S. Mendes and A. J. Marques Cardoso, “Voltage source inverter Aug. 2013.
fault diagnosis in variable speed AC drives, by the average current park’s [31] S. Yang, D. Xiang, A. Bryant, P. Mawby, L. Ran, and P. Tavner, “Con-
vector approach,” in Proc. IEEE Int. Electric Mach. Drives Conf., 1999, dition monitoring for device reliability in power electronic converters: A
pp. 704–706. review,” IEEE Trans. Power Electron., vol. 25, no. 11, pp. 2734–2752,
[9] K. Rothenhagen and F. W. Fuchs, “Performance of diagnosis methods Nov. 2010.
for IGBT open circuit faults in three phase voltage source inverters for [32] Y. Yan, Y. Qian, H. Sharif, and D. Tipper, “A survey on smart grid commu-
AC variable speed drives,” in Proc. Eur. Conf. Power Electron. Appl., nication infrastructures: Motivations, requirements and challenges,” IEEE
Sep. 2005, Paper 10. Commun. Surv. Tut., vol. 15, no. 1, pp. 5–20, First Quarter 2013.
BAKER et al.: REAL-TIME AI-BASED ANOMALY DETECTION AND CLASSIFICATION IN POWER 559

[33] US Department of Energy, “Communications requirements of smart Hassan Althuwaini (Student Member, IEEE) re-
grid technology,” 2010. [Online]. Available: https://www.energy.gov/ ceived the B.S. degree in electrical engineering from
sites/default/files/gcprod/documents/Smart_Grid_Communications_ Kansas State University, Manhattan, KS, USA, in
Requirements_Report_10-05-2010.pdf 2019.
[34] M. Ghorbanian, S. H. Dolatabadi, M. Masjedi, and P. Siano, “Communi- He was a Research Assistant with JKLab.org from
cation in smart grids: A comprehensive review on the existing and future 2018 to 2019. He was with the Department of Elec-
communication and information infrastructures,” IEEE Syst. J., vol. 13, trical and Computer Engineering, The University of
no. 4, pp. 4001–4014, Dec. 2019. Illinois Chicago in Jan. 2021 to continue his M.S
[35] S. Vazquez et al., “Model predictive control: A review of its applications studies under the supervision of Prof. M. Shadmand.
in power electronics,” IEEE Ind. Electron. Mag., vol. 8, no. 1, pp. 16–31, His research interests include smart microgrids, en-
Mar. 2014. ergy storage system, characterization of SiC, and GaN
[36] C. R. D. Osório et al., “Modulated model predictive control applied to power devices.
LCL-filtered grid-tied inverters: A convex optimization approach,” IEEE
Open J. Ind. Appl., vol. 2, pp. 366–377, 2021.

Mohammad B. Shadmand (Senior Member, IEEE)

Matthew Baker (Student Member, IEEE) received
received the Ph.D. degree in electrical engineering
the B.S. degree in electrical engineering from Rock-
from Texas A&M University, College Station, TX,
hurst University, Kansas City, MO, USA, in 2018. USA, in 2015.
Since 2020, he has been working toward the Ph.D.
From 2015 to 2016, he was an Instructor with the
degree in power electronics and controls with the
Department of Electrical and Computer Engineering,
Department of Electrical and Computer Engineering,
Texas A&M University. From 2016 to 2017, he was a
University of Illinois Chicago, Chicago, IL, USA. Research Engineer with the Renewable Energy and
From 2018 to 2020, he was a Ph.D. student with
Advanced Power Electronics Research Laboratory,
the Department of Electrical and Computer Engineer-
College Station, TX, USA. From 2017 to 2020, he
ing, Kansas State University, Manhattan, KS, USA.
was an Assistant Professor with the Department of
He has internship experiences at the Electric Power
Electrical and Computer Engineering, Kansas State University, Manhattan, KS,
Research Institute in 2022, and Sandia National Laboratories in 2021 and 2022.
USA. Since 2020, he has been an Assistant Professor with the University of
Illinois Chicago, Chicago, IL, USA. He has authored and coauthored more
than 100 journal and conference papers. His current research interests include
distributed self-learning control schemes, advanced model predictive control,
Amin Y. Fard (Student Member, IEEE) received the grid-following and grid-forming inverters, and intrusion detection system for
B.Sc. degree from Azarbaijan Shahid Madani Uni- power electronics dominated grids.
versity, Tabriz, Iran, in 2011, and the M.Sc. degree Dr. Shadmand was the recipient of Michelle Munson Serban Simu Keystone
with “second rank honor” from the University of Research Scholar, Kansas State University in 2017, the 2019 IEEE Myron
Tabriz, Tabriz, Iran, in 2014, both in electrical power Zucker Faculty-Student Research Grant, and has awarded multiple best paper
engineering. Since 2020, he has been working toward awards at different IEEE conferences. He was a Technical Program Co-Chair of
the Ph.D. degree in power electronics and controls the 2019 and 2022 IEEE Smart Grid and Renewable Energy Conference. He was
with the Department of Electrical and Computer En- an Associate Editor for IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS,
gineering, University of Illinois Chicago, Chicago, IEEE TRANSACTIONS ON INDUSRTY APPLICATION, and IET Renewable Power
IL, USA. Generation.
From 2014 to 2018, he was with Roshdiyeh Higher
Education Institute as a Lecturer, the supervisor with Electrical Machinery and
High Voltage Laboratories, and the Educational Manager with the Engineering
Faculty. He was a Ph.D. student with K-State University from 2018–2020. His
research interests include renewable energy systems like photovoltaic systems
and wind turbines, power electronics, distributed generation, and power quality.