Electric Power Systems Research 146 (2017) 25–32

Contents lists available at ScienceDirect

Electric Power Systems Research

journal homepage: www.elsevier.com/locate/epsr

Multi-temporal Optimal Power Flow for voltage control in MV

networks using Distributed Energy Resources
J.L. Meirinhos, D.E. Rua, L.M. Carvalho, A.G. Madureira ∗
Institute for Systems and Computer Engineering, Technology and Science (INESC TEC), FEUP Campus, Rua Dr. Roberto Frias, 4200-465 Porto, Portugal

a r t i c l e i n f o a b s t r a c t

Article history: Large-scale integration of variable Renewable Energy Sources (RES) brings significant challenges to grid
Received 13 June 2016 operation that require new approaches and tools for distribution system management, particularly con-
Received in revised form 5 January 2017 cerning voltage control. Therefore, an innovative approach for voltage control at the MV level is presented.
Accepted 9 January 2017
It is based on a preventive day-ahead analysis that uses data from load/RES forecasting tools to establish
Available online 19 January 2017
a plan for operation of the different Distributed Energy Resources (DER) for the next day. The approach
is formulated as a multi-temporal Optimal Power Flow (OPF) solved by a meta-heuristic, used to tackle
Keywords:
complex multi-dimensional problems. The tuning of the meta-heuristic parameters was performed to
Batteries
Distributed power generation ensure the robustness of the proposed approach and enhance the performance of the algorithm. It was
Microgrids tested through simulation in a large scale test network with good results.
Renewable Energy Sources © 2017 Elsevier B.V. All rights reserved.
Smart grids
Voltage control

1. Introduction optimal solutions but is simple to implement. Coordinated control,

on the other hand, can be extremely complex as it usually requires
The operation of electrical distribution systems has changed due sophisticated communication infrastructures and makes use of dif-
to the increasing integration of DER with voltage control being one ferent resources for voltage control with different characteristics
of the main aspects that have been significantly affected. In partic- and even ownership models. Traditional voltage regulation devices
ular, voltage rise due to the injection of power from DER, especially are property of the Distribution Network Operator (DNO) while
in rural grids, where traditional voltage regulation methods, often other distributed resources such as storage devices or Distributed
based on automatic voltage regulators or On-Load Tap Changing Generation (DG) units may be privately owned but available for
(OLTC) transformers, can become inadequate [1]. supporting grid operation following some type of bilateral agree-
In this context, possible voltage control strategies require dif- ment, for instance.
ferent control structures that have been identified in the scientific A decentralized agent-based operation is proposed in Ref. [3],
literature [2]: where two methods are used: firstly, a Particle Swarm Optimiza-
tion (PSO) method evaluates possible solutions and their effect on
• Local control—Making use of independent and local control of voltage profiles; secondly, synergies between local control strate-
compensation devices; gies complementary to the optimization are sought. In Ref. [4], the
• Centralized control—Often based on a predefined set of rules that authors propose the deployment of a network of meta-heuristic
can involve coordination up to a certain degree between devices optimizer agents aimed at solving the voltage regulation problem
of the same type (such as capacitor banks along a feeder); using cost functions. The authors in Ref. [5] present an efficient
• Model-based control—Distribution system model-based volt/var hybrid approach for volt/var control in distribution systems, with
management that utilizes real-time data, state estimation, and switching limits on taps and shunts, that combines the strengths of
online power-flow calculations. a gradient technique and a metaheuristic technique. In Ref. [6], an
approach for optimal voltage control strategies for the day-ahead in
In Ref. [2], the main advantages and limitations of each of these active distribution networks is presented that makes use of OLTCs,
solutions are highlighted. Uncoordinated control provides sub- active and reactive power control of DG units and switchable var
compensation devices with the objective of loss minimization and
voltage penalty objective functions. Also, in Ref. [7], an approach to
∗ Corresponding author. Fax: +351 22 209 4150. coordinate different DER in MV distribution networks for voltage
E-mail address: [email protected] (A.G. Madureira). control purposes is presented and formulated as an optimization

http://dx.doi.org/10.1016/j.epsr.2017.01.016
0378-7796/© 2017 Elsevier B.V. All rights reserved.
26 J.L. Meirinhos et al. / Electric Power Systems Research 146 (2017) 25–32

problem that is solved by a meta-heuristic. However, the multi- variable RES subject to a set of technical and operational con-
temporal formulation is not addressed as time dependent actions as straints, namely node voltage limits and branch thermal limits.
such the scheduling if storage devices is not considered. An active- The resulting actions will allow defining the operation plan for
reactive OPF for distribution grids with DG and battery storage the next day.
is presented in Ref. [8]. Both active and reactive power dispatch • N-hours ahead (6 h-ahead analysis) → The same multi-temporal
strategies are optimized for maximizing the total yield and simul- OPF developed for the day-ahead analysis will be used n-hours
taneously minimizing the total costs of energy losses. ahead in order to adjust the control actions previously identified
The approach proposed in this paper is based on a multi- feeding from more recent and accurate data regarding load and
temporal OPF formulation solved by a meta-heuristic that produces RES forecast (generated by the DSO forecasting system). The main
a set of control actions – set-points – for the available DER (such as objective will now be to minimize the deviations in a sliding win-
DG units or storage devices) and other voltage support equipment dow of 6-h ahead (with hourly updates) regarding the scheduled
for the day-ahead to keep voltage profiles within admissible limits scenario in D-1 analysis. This will enable correcting deviations
by relying on both load and Renewable Energy Sources (RES) fore- that may occur and solve technical problems that may arise close
casts. The main innovation is the ability to ensure a coordinated to real-time.
control of all the resources available, for voltage control purposes,
considering a multi-period analysis for an operating window of In Ref. [11], the combination of a solar power forecast method
a full day with hourly discretization that takes into account time and the multi-temporal OPF was proposed to analyze the impact of
interdependencies of specific variables. solar power forecast errors in a real Portuguese MV network.

2. Multi-temporal Optimal Power Flow framework 2.2. Mathematical formulation

2.1. Reference architecture In the optimization problem, the definition of the objective-
function to be used by the algorithm is of a great importance. In
A reference architecture for smart distribution systems was this case, the following two objectives are pursued:
developed in the EU project SuSTAINABLE [9], which is based on the
Portuguese smart grid architecture [10]. This architecture, shown • Minimize the curtailment of DG;
in Fig. 1, is assumed to be based on advanced ICT infrastructure that • Minimize the shedding of controllable loads.
uses data from the existing smart meters.
The proposed control architecture is organized into three main In this case, it was assumed that some specific loads can be
layers: shed under some type of interruptibility contract with the DSO
(according to the types of demand response defined [12]) that allow
• A top control level corresponding to the SCADA/DMS, under the for interruptions to power delivery in exchange for some financial
responsibility of the DNO, that is in charge of managing the whole compensation at the time of interruption or for an overall reduction
distribution grid. in the electricity price. Nevertheless, since the formulation is flexi-
• An intermediate control level at the HV/MV primary substation – ble, other models of demand response schemes used for mitigating
the Smart Substation Controller (SSC) – which is in charge of each network constraints may also be considered, for instance following
MV network and incorporates a set of advanced control function- an agreement with an aggregator, or by flexible network tariffs.
alities that allow a coordinated and efficient operation of the MV Active power losses minimization can also be pursued as an
system exploiting available DER. additional goal to be attained with voltage control [13,14]. The
• A bottom control level located at the MV/LV secondary substa- formulation adopted addresses voltage control as constraint sat-
tion – the Distribution Transformer Controller (DTC) – which is isfaction problem, i.e., it aims at guaranteeing acceptable voltage
responsible for managing a LV network, serving as a gateway of levels while trying to minimize the curtailment of DG and the load
data to the upstream systems and also being able to incorporate shedding when required. Similar formulations have been adopted
some basic control functionalities. by other authors, namely, in Refs. [15,16]. Note that the objective
• A field control level located at the customer premises in which -function described by (1) can be easily adjusted to include the
the smart meter serves as a gateway to control its associated minimization of active power losses, provided that representative
resources, namely microgeneration, controllable loads, storage trade-offs between the three objectives are available.
devices and Electric Vehicles. The resulting objective-function combines these two terms
using a type of trade-off approach, as shown in (1).
The proposed multi-temporal OPF framework will operate at
the level of the HV/MV primary substation and will be responsi-

24
 
h h
min ˛ · Pcurtail + (1 − ˛) · Pshed (1)
ble for managing MV network operation. This algorithm, as well
h=1
as the load and RES forecasting systems, are located at the level
of the Supervisory Control and Data Acquisition/Distribution Man- h
where ˛ is a decision parameter, 0 ≤ ˛ ≤ 1; Pcurtail is the total cur-
agement System (SCADA/DMS). To cope with the uncertainty in the tailment of DG active power, i.e. the difference between the total
forecasts, the multi-temporal OPF is expected to work in two differ- available power and the power actually produced (normalized by
ent time-frames that are summarily described next and illustrated h
the maximum value); Pshed is the total shedding of controllable load
in Fig. 2: active power, i.e. the amount of load necessary to be disconnected
following an availability contract (normalized by the maximum
• Day-ahead (D-1 analysis) → Taking as inputs load and genera- value).
tion bids from the market agents as well as results from load The decision parameter ˛ is used to define the relation between
and RES forecasting, the multi-temporal OPF will produce a set two objectives: minimization of DG curtailment (in kW) and the
of control actions for the next day by MV network node (i.e. by minimization of load shedding (also in kW). This “weight” should
DTC). The main goal, in line with the SuSTAINABLE project con- then reflect the preferences of the decision-maker, which in this
cept, will be to maximize the integration of energy coming from case should be the DNO. This means that the DNO can decide which
 J.L. Meirinhos et al. / Electric Power Systems Research 146 (2017) 25–32 27

Fig. 1. Smart distribution grid reference architecture.

Time Horizon

Day D-1 Day D

Solve voltage violation
problems at the MV grid
considering control Adjust control actions to minimize deviations
actions from DER from scheduled for day D

Multi-temporal Operating Period

OPF at MV level [6 hours ahead]

Best prediction
of the operating
scenario for day
Load + RES Forecasts D
for D

Fig. 2. Proposed approach for the voltage control at the MV level.

objective to favored (DG curtailment or shedding of controllable

loads) by defining the appropriate value for ˛. In case no objective
is to be favored, ˛ should be set to 0.5. This approach is only valid 
N
  
when the trade-offs and the value functions for the two objectives QGh + QDG
h h
+ QCB − QLh = Vih Vkh · Gik · sin ih − kh
i i i i
are linear [17]. k=1
The objective-function, in (1), is subject to restrictions related to  
−Bik · cos ih − kh (3)
the AC power flow balance for each of the N nodes of the network
as presented in (2) and (3):
where PGh and QGh are the active and reactive power generated
i i

N
   upstream the distribution network at bus i in hour h, respectively;
PGh h
+ PDG − PLh + PDh − PCh = Vih Vkh · Gik · cos ih − kh h and Q h are the active and reactive power generation from DG
PDG DG
i i i i i i i
k=1 at bus i in hour h, respectively; PLh and QLh is the active and reactive
  i i
+Bik · sin ih − kh (2) power load at bus i in hour h, respectively; PDh is the active power
i
28 J.L. Meirinhos et al. / Electric Power Systems Research 146 (2017) 25–32

provided from discharging the storage unit at bus i in hour h; PCh is 2.3. Optimization algorithm
i
the active power consumed from charging the storage unit at bus i
h is the reactive power generated from capacitor banks
in hour h; QCB The approach was formulated as Mixed-Integer Non-Linear
i
Programming (MINLP) problem according to the mathematical for-
at bus i in hour h; Vih is the voltage magnitude at bus i in hour h; ih
mulation presented in Section 2.2 and considers the following
is the voltage angle at bus i in hour h.
control variables:
It is assumed that MV networks are mostly balanced systems
since there are usually no single-phase connections directly to MV
and the effect of aggregation ensures that the unbalances at LV • Reactive power injection/absorption from DG units (continuous
level (due to the connection of single-phase loads and/or micro- variable);
generation) are not passed to the MV level, i.e. that phases are • Reactive power injection from capacitor banks (discrete vari-
balanced at the distribution transformer level. able);
The variables concerning the AC power flow model and the DER • Active power curtailment from DG units (continuous variable);
operation must remain within their respective limits as presented • Active power injection/absorption from storage devices (contin-
in (4)–(9). uous variable);
min h max
• Active power load shedding from controllable loads (continuous
PDG ≤ PDG ≤ PDG (4)
i i i variable);
min h max • Tap positions of OLTC transformers (discrete variable);
QDG ≤ QDG ≤ QDG (5)
i i i

Vimin ≤ Vih ≤ Vimax (6) It is known that non-linear optimization techniques, of the vari-
Sik ≤ max
Sik (7) ety known as stochastic search algorithms or meta-heuristics, aim
at finding the global optimum or, at least, a good local optimum
PDGi , PGi , PDi , PCi > 0 (8) without requiring significant assumptions on the problem. This
  approach competes with classical methods when it comes down
tOLTCik = tOLTC1ik , tOLTC2ik , · · · (9) to selecting an optimization algorithm. Evidently, both approaches
min and P max are the minimum and maximum active power have their advantages and drawbacks.
where PDG DG i i The MINLP optimization problem has both continuous and dis-
min and Q max are the
generation from DG at bus i, respectively; QDG DG i i crete variables. Many classical methods for optimization are not
minimum and maximum reactive power generation from DG at bus suitable for dealing with discrete and continuous variables simul-
i, respectively (positive if generating reactive power and negative if taneously because of the “combinatorial explosion” of the values of
consuming reactive power); Vi is the voltage at bus i; Vimin and Vimax discrete variables. On the other hand, meta-heuristics are intrinsi-
are the minimum and maximum voltage at bus i, respectively; Sik cally able to handle both continuous and discrete variables.
is the power flow in branch ik; Sik max is the maximum power flow in
In this paper, the optimization problem is solved using a new
branch ik; tOLTCik is the tap for the OLTC transformer in branch ik. meta-heuristic approach—DEEPSO, which is a variant of the Evo-
An additional constraint is considered for OLTC transformers to lutionary Particle Swarm Optimization borrowing concepts from
restrict the number of tap position changes in two consecutive peri- Differential Evolution [18]. Several examples of the application of
ods in order to prevent the mechanism from deteriorating due to meta-heuristics to the volt/var problem can be found in the avail-
intensive use as shown in (10). able scientific literature [3,4,7,19,20].
Following the tradition of evolutionary algorithms, inequality
|tOLTC hik − tOLTC h−1
ik
| ≤ ıik (10)
constraints are modelled as quadratic penalty functions added
where tOLTC hik and tOLTC h−1 is the tap for the OLTC transformer ik in to the main objective-function while the equality constraints
ik
hour h and h − 1, respectively; ıik is the maximum step size allowed derived from the AC power flow equations were verified using a
for OLTC transformer in branch ik; Newton–Raphson algorithm available in the MATPOWER toolbox
Also, the constraints (11)–(14) must be considered for each stor- [21].
age device:
 
PDh 3. Optimization parameter tuning
SOCih = SOCih−1 + Ci · PCh − i
(11)
i Di
Like any other meta-heuristic, the optimal solution provided by
DEEPSO may be slightly different each time the meta-heuristic is
SOCimin ≤ SOCih ≤ SOCimax (12)
run. In order to make sure DEEPSO output is robust enough to be
SOCimax used in practical applications, the set of strategic parameters (e.g.
SOCi1 = SOCi24 = (13) mutation rate, , communication probability, P, etc.) must be opti-
2
mally defined. Successful tuning algorithms are the ones that find
PCh × PDh = 0 (14) good values for the strategic parameters with the least amount
i i
of computational effort, hence, outperforming naive enumeration
where SOCih is the state of charge in hour h of the storage device
methods that rely on testing every possible combination of param-
placed at bus i; SOCih−1 is the state of charge in hour h − 1 of the stor- eters. There are several parameter tuning processes reported in
age device placed at bus i; Ci and Di are the charging efficiency and the literature, with the Response Surface Methodology (RSM) [17]
discharging efficiency, respectively, of the storage device placed at being one of the most popular.
bus i; PCh and PDh are the active power absorbed and the active power This paper adopted the parameter tuning algorithm reported in
i i
injected, respectively, in hour h of the storage device placed at bus i; Ref. [18] to obtain the best set of parameters for all the experiments
SOCimin and SOCimax are the minimum state of charge and maximum conducted. This algorithm is based on repeated 2 × 2 factorial
state of charge of the storage device placed at bus i. designs, in which there are four treatment combinations being
It must be stressed that this methodology is general and can be analyzed in each iteration. Each treatment is composed of 30 ran-
easily adapted according to different needs or preferences domized DEEPSO runs.
 J.L. Meirinhos et al. / Electric Power Systems Research 146 (2017) 25–32 29

Table 1 Table 2
Mean DEEPSO output. Two-way ANOVA results.

2 × 2 Factorial Design Factor F-test statistic p-Value Main effect

/P 0.2 0.8  32.515 9.17 × 10−8 −1.19

P 57.224 1.00 × 10−11 1.58
0.2 1.507 3.840  ×P 12.760 5.168 × 10−4 0.75
0.8 1.061 1.897

Fig. 3. Boxplot of the DEEPSO for the treatments A = {0.2, 0.2}, B = {0.2, 0.8}, C = {0.8, Fig. 4. Boxplot of the DEEPSO for the treatments A = {0.5, 0.2}, B = {0.5, 0.35}, C = {0.8,
0.2} and D = {0.8, 0.8}. 0.2} and D = {0.8, 0.35}.

The iterative process begins by defining the maximum range ered to 0.5 and two new treatments are defined: B = {0.2, 0.5} and
of variation for the parameters being optimized (i.e.,  and P). The D = {0.8, 0.5}. Thirty new independent runs of DEEPSO are carried
30 best fitness values obtained for each treatment are then jointly out for these new treatments and the former analysis is done once
analyzed using the two-way ANOVA to detect not only if the two again. The iterative process is repeated until all the p-values are all
variables have an impact in the DEEPSO output but also if there is less than or equal to 0.05.
interaction among them. The variable with the highest impact is After 4 iterations of the method, it was found that the treat-
then selected and its allowable interval of variation is decreased ments A = {0.5, 0.2}, B = {0.5, 0.35}, C = {0.8, 0.2} and D = {0.5, 0.35}
using the bisection method. The main impacts of the variables are yield statistically similar DEEPSO output for the optimization prob-
used to decide whether its upper or lower limit should be updated. lem being studied. Hence, any combination of these parameters can
The algorithm stops when ANOVA detects no evidence that the be selected since DEEPSO will produce, in average, the same best
parameters have an influence on DEEPSO output [18]. This can eas- solution. Fig. 4 shows the boxplot for these final four treatments.
ily be done by setting a maximum threshold for the p-value, which
typically is 0.05.
Table 1 shows the mean DEEPSO output for the first iteration 4. Simulation results
of the tuning algorithm applied to the scenario to be simulated.
In this case, the algorithm started with four combinations of two The proposed methodology was then tested on a large scale test
extreme values of  and P, being the lower limit 0.2 and the upper network characterized by high DG penetration using as a reference
limit 0.8. Note that each treatment {, P} consists of 30 indepen- a scenario with no control actions. The MV network is based on a
dent DEEPSO runs. In order to avoid excessive computation effort, real Portuguese MV distribution network, and the diagram is pre-
only 100 iterations were allowed. The population is composed of sented in Fig. 5. In this case, DEEPSO was run with the best set of
30 individuals. All individuals in the initial population have been parameters for 1000 iterations.
randomly assigned to a point in the search space. This network is typically rural, with a radial structure of two
At first glance, the results in Table 1 might indicate that the different voltage levels: 30 kV before the OLTC transformer (shown
mean DEEPSO output is very similar for each treatment. However, on the top of Fig. 5) and 15 kV after the OLTC transformer (bottom
the clear influence of parameter P is noticeable since the best mean of Fig. 5). The OLTC transformer is a 30/15 kV transformer with taps
output is achieved for the lowest limit of this parameter. on the secondary side. The network has a total of 210 nodes and
Fig. 3 shows the boxplots for each treatment, which clearly por- 212 branches. The complete network data can be found in Ref. [7].
trays the randomness of DEEPSO output. As expected, the worst This network includes several DER such as DG units (one CHP
performing treatments are the ones with the highest communica- unit and two wind generators—DFIG) and six microgrids directly
tion probability. connected to MV network. From the MV point of view, each micro-
Table 2 contains the two-way ANOVA results for this first itera- grid was considered as a single bus with an equivalent generator
tion. Note that variable P has the lowest p-value, which means that (corresponding to the sum of all PV microgeneration) and equiva-
it is the one with the highest impact in DEEPSO output. Moreover, its lent load (corresponding to the sum of all LV loads). Furthermore,
positive main effect allows concluding that the best DEEPSO output two large storage devices (batteries) of 0.5 MWh each were con-
is achieved at low values of P. Hence, the upper limit of P is low- sidered with 85% of efficiency for charging/discharging, as well
30 J.L. Meirinhos et al. / Electric Power Systems Research 146 (2017) 25–32

Fig. 5. MV test network.

as a 0.2 Mvar capacitor bank and two controllable loads under an P (M W)

interruptibility contract (around 0.25 MW). 6
In order to run the algorithm for the day-ahead analysis, real
5
forecast data is necessary. In this case, daily profiles for both load
and generation were used to build the simulation scenario. The 4
scenario used was obtained by applying the load and generation
3
profiles to the total load installed capacity and total generation
capacity, respectively. The profile for 24 h is provided by the Por- 2
tuguese network operator (REN) for a day in 2014. The resulting 1
load and DG profiles used are shown in Fig. 6.
0
In this case, the decision parameter ˛ was set at 0.5. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Figs. 7 and 8 show the maximum value and minimum value,
respectively, for voltage for each hour of the day considering that no PV CHP Time (h)
control actions are taken (base case—gray bars), and after running Wind Pload
the optimization algorithm (green bars).
Fig. 6. Load and generation for 24 h.
As it can be observed, without the voltage control algorithm,
voltage values are outside the admissible range of ±5%. Since the
DG units have their peak generation at 14:00 (outside the hours is exported to the HV network which causes overvoltages around
of peak demand), there is an excess of generation in the MV that that hour. On the other hand, around 21:00 (corresponding to the
 J.L. Meirinhos et al. / Electric Power Systems Research 146 (2017) 25–32 31

V (p.u.) P (MW)
1.15 0.3

1.1 0.2

1.05 0.1
1 0
0.95 -0.1
0.9
-0.2
0.85
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 -0.3
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Time (h)
Initial Final lim Time (h)
Storage 1 Storage 2

Fig. 7. Maximum voltage values for 24 h. (For interpretation of the references to Fig. 10. Charging/discharging of batteries.
color in the text, the reader is referred to the web version of this article.)

Q (M VAr)
V (p.u.)
0.15
1.1
0.1
1.05 0.05

1 0
-0.05
0.95
-0.1
0.9 -0.15

0.85 -0.2
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Time (h) Time (h)
Initial Final lim
Fig. 11. Total reactive power provided by DG.
Fig. 8. Minimum voltage values for 24 h. (For interpretation of the references to
color in the text, the reader is referred to the web version of this article.)
Q (M VAr)
0.25
P (M W)
0.5 0.2

0.4 0.15

0.3 0.1

0.2 0.05

0.1 0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
0 Time (h)
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Capacitor Bank
Time (h) Fig. 12. Reactive power provided by capacitor bank.
PV Wind

Fig. 9. Active power curtailment from DG.

hours. In this case, the maximum state of charge of the batteries is
reached during the day in order to avoid voltage violations.
peak load hour), the amount of power provided by the DG units is Also, the reactive power provided by the DG units (positive
smaller than the load, which causes undervoltages. After running when injecting and negative when absorbing), presented in Fig. 11,
the optimization algorithm, all voltage violations were corrected. assists in the voltage control.
In this case no load shedding was required in order to control In hours of peak load, the capacitor banks have also contributed
the voltage profiles. Nevertheless, it was necessary to curtail some to increase the voltage in peak load as portrayed in Fig. 12.
renewable generation in order to bring voltage values back inside Fig. 13 shows the tap values at the OLTC transformer. Accord-
admissible limits, as shown in Fig. 9. The worst hour is at 14:00, ing to the modelling of the transformer with taps on the secondary
where there is a greater penetration of generation from RES that side, tap values above 1 raise the voltage on the secondary of the
leads to the most extreme voltage deviations. transformer, whereas, lower tap values are used when voltage pro-
The contribution of other control variables such as the two bat- files are high and higher tap values are used when voltage profiles
teries, shown in Fig. 10, have an important role to voltage control are low. As can be observed, the constraint used for limiting the
by storing energy (negative values) at hours when there is excess number of switching actions of the OLTC transformer was able to
DG and provide the stored energy (positive values) in peak load limit the number of tap changes in consecutive one-hour periods.
32 J.L. Meirinhos et al. / Electric Power Systems Research 146 (2017) 25–32

Tap References

1.01 [1] N. Jenkins, J. Ekanayake, G. Strbac, Distributed Generation, IET, 2010.

[2] J. Barr, R. Majumder, Integration of distributed generation in the Volt/VAR
1 management system for active distribution networks, IEEE Trans. Smart Grid
6 (2015) 576–586.
[3] E. Drayer, F. Meyer, J. Hegemann, M. Braun, Control strategies for a
0.99
decentralized, real-time operation of distribution grids, 2015 IEEE Eindhoven
PowerTech (2015).
0.98 [4] A. Vaccaro, A.F. Zobaa, Voltage regulation in active networks by distributed
and cooperative meta-heuristic optimizers, Electr. Power Syst. Res. 99 (2013)
0.97 9–17.
[5] A. Mohapatra, P.R. Bijwe, B.K. Panigrahi, An efficient hybrid approach for
volt/var control in distribution systems, IEEE Trans. Power Deliv. 29 (2014)
0.96 1780–1788.
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 [6] M.Z. Degefa, M. Lehtonen, R.J. Millar, A. Alahäivälä, E. Saarijärvi, Optimal
voltage control strategies for day-ahead active distribution network
Time (h) operation, Electr. Power Syst. Res. 127 (2015) 41–52.
[7] A.G. Madureira, J.A.P. Lopes, Coordinated voltage support in distribution
Fig. 13. Tap position of the OLTC transformer. networks with distributed generation and microgrids, IET Renew. Power
Gener. 3 (2009) 439–454.
[8] A. Gabash, P. Li, Active-reactive optimal power flow in distribution networks
In fact, only one tap change has occurred from one period to the with embedded generation and battery storage, IEEE Trans. Power Syst. 27
other. (2012) 2026–2035.
[9] A. Madureira, R. Bessa, L. Seca, J. Pereira, A.A. Messias, D.A. Lopes, P.G. Matos,
Advanced system architecture and algorithms for smart distribution grids:
5. Conclusion the SuSTAINABLE approach, 23rd International Conference on Electricity
Distribution (CIRED) (2015).
[10] C. Gouveia, D. Rua, F.J. Soares, C. Moreira, P.G. Matos, J.A.P. Lopes,
An approach for voltage control that performs a day-ahead anal-
Development and implementation of Portuguese smart distribution system,
ysis is proposed, where an optimal dispatch schedule is sought for Electr. Power Syst. Res. 120 (2015) 150–162.
the next day to coordinate the several DER available in the MV [11] A. Madureira, R. Bessa, J. Meirinhos, D. Fayzur, J. Filipe, A.A. Messias, D.A.
network to avoid technical problems in terms of voltage profiles. Lopes, P.G. Matos, The impact of solar power forecast errors on voltage
control in smart distribution grids, 23rd International Conference on
A multi-temporal OPF was developed capable of coordinating Electricity Distribution (CIRED) (2015).
the available DER connected at the MV level such as OLTC trans- [12] M.H. Albadi, E.F. El-Saadany, A summary of demand response in electricity
formers, DG units, controllable loads or storage devices to keep markets, Electr. Power Syst. Res. 78 (2008) 1989–1996.
[13] M. Oshiro, K. Tanaka, A. Uehara, T. Senjyu, Y. Miyazato, A. Yona, T. Funabashi,
voltage profiles within admissible limits. The problem is formu- Optimal voltage control in distribution systems with coordination of
lated as an optimization problem and solved by a meta-heuristic distribution installations, Int. J. Electr. Power Energy Syst. 32 (2010)
(DEEPSO) that is able to tackle large dimension systems such as 1125–1134.
[14] R.S. Rao, K. Ravindra, K. Satish, S.V.L. Narasimham, Power loss minimization in
real distribution grids with multiple control variables, both contin- distribution system using network reconfiguration in the presence of
uous and discrete. In this case, the main objective is to minimize the distributed generation, IEEE Trans. Power Syst. 28 (2013) 317–325.
DG curtailment and controllable load shedding subject to a set of [15] V. Calderaro, G. Conio, V. Galdi, A. Piccolo, Reactive power control for
improving voltage profiles: a comparison between two decentralized
technical/operational constraints including admissible voltage lim-
approaches, Electr. Power Syst. Res. 83 (2012) 247–254.
its and physical limits of devices, according to the preferences of the [16] M. Farina, A. Guagliardi, F. Mariani, C. Sandroni, R. Scattolini, Model predictive
DNO. An efficient parameter tuning algorithm was used, allowing control of voltage profiles in MV networks with distributed generation,
Control Eng. Pract. 34 (2015) 18–29.
DEEPSO to consistently provide the best output for the optimization
[17] R. Keeney, H. Raiffa, Decisions with Multiple Objectives: Preferences and
problem. Value Tradeoffs, 2nd edition, Cambridge University Press, 1993.
The proposed algorithm was tested through simulation using [18] V. Miranda, R. Alves, Differential evolutionary particle swarm optimization
a real MV network with multiple control variables and revealed (DEEPSO): a successful hybrid, 2013 Congress on Computational Intelligence
and 11th Brazilian Congress on Computational Intelligence (BRICS-CCI &
a good performance, even in extreme scenarios with large-scale CBIC) (2013).
integration of RES, which proved the effectiveness of the proposed [19] A. Khorsandi, S.H. Hosseinian, A. Ghazanfari, Modified artificial bee colony
approach for a day-ahead analysis. algorithm based on fuzzy multi-objective technique for optimal power flow
problem, Electr. Power Syst. Res. 95 (2013) 206–213.
[20] Y. Fukuyama, Parallel particle swarm optimization for reactive power and
Acknowledgment voltage control investigating dependability, 18th International Conference on
Intelligent System Application to Power Systems (ISAP) (2015).
[21] R.D. Zimmerman, C.E. Murillo-Sanchez, R.J. Thomas, MATPOWER:
This work was developed within the framework of the SuSTAIN- steady-state operations, planning, and analysis tools for power systems
ABLE project (contract no. 308755), co-financed by European funds research and education, IEEE Trans. Power Syst. 26 (2011) 12–19.
through the 7th RTD Framework Programme (FP7).