Electric Power Systems Research 127 (2015) 4152

Contents lists available at ScienceDirect

Electric Power Systems Research

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

Optimal voltage control strategies for day-ahead active distribution

network operation
M.Z. Degefa a, , M. Lehtonen a , R.J. Millar a , A. Alahivl a , E. Saarijrvi b
a
Department of Electrical Engineering and Automation, Aalto University, Espoo, Finland
b
Trimble, Espoo, Finland

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

Article history: The aim of this study is to develop a coordinated day-ahead voltage control strategy for an active dis-
Received 6 March 2015 tribution network. A framework comprising a synergy of real-time dynamic thermal rating (DTR) and
Received in revised form 21 May 2015 coordinated voltage control (CVC) is proposed for solving the voltage quality and thermal limit problems
Accepted 23 May 2015
associated with a high penetration level of distributed generation (DG) in an active distribution network.
Available online 9 June 2015
The CVC scheme involves solutions such as On-Load-Tap-Changers (OLTCs), active and reactive power
control of DG units, and switchable shunt VAR compensation devices (SVCs). Loss minimization and
Keywords:
voltage penalty objective functions in the CVC optimization problem are compared. A 147 bus test distri-
Active network management
Distributed generation (DG)
bution network planned for an actual geographical location is used to evaluate the proposed DTR-based
Dynamic thermal rating (DTR) day-ahead CVC strategy. In this study, we have showed that the reactive power absorption/injection
Onload tap changer (OLTC) potential of DG units can play an important role in CVC. Moreover, the study demonstrates that real-time
State estimation thermal rating boosts the utilization of reactive power resources in the distribution system. Finally, the
Voltage control study investigates the practicality of day-ahead active distribution network operation planning for CVC.
2015 Elsevier B.V. All rights reserved.

1. Introduction two possible explanations for the lack of a collective agreement

In todays active distribution network there is an ever increas-
ing penetration level of distributed generation (DG) driven by
technical and policy forces. Among the various limiting factors
inhibiting the further installations of DG units are feeder ther-
mal capacity limits and the steady state voltage rise problem [1].
With regard to dealing with the voltage rise problem, reactive
power contribution by DG units is one of the most commonly pro-
posed approaches [2]. Wind turbines, for instance, by virtue of
their power electronic converters, are able to control active and
reactive power independently [3]. The voltage source inverter in
PVs is also an interface that enables the control of reactive power.
These non-dispatchable DG units, such as PV and wind, operate a
signicant fraction of their time much below their rated power,
during which they can provide reactive power service. Neverthe-
less, the present grid code, for example in Finland, does not allow
distributed generation to participate in distribution network volt-
age control in any way. Moreover, the currently used distribution
network planning tools and procedures are not capable of taking
active voltage control into account, as discussed in [4]. There are
Corresponding author. Tel.: +358 44 5654598; fax: +358 9 470 2991.
E-mail address: merkebu.degefa@aalto. (M.Z. Degefa).
on deploying DG units for voltage control in a distribution net-
work. The rst is the insufcient measurements and consequently
the limited state estimation services in distribution networks. The
second reason could be the absence of power ow constraints
such as bottlenecks in many existing under-loaded distribution
networks, which tends to have inhibited the deployment of real-
time line and cable rating programs. Given the rapid development
of active distribution networks, both aforementioned reasons are
becoming obsolete. With the installation of automatic meter read-
ing (AMR) devices, the accessibility of distribution network data
has signicantly increased in terms of resolution as well as clar-
ity. Contradicting the second reason, the increasing installations
of DG units are creating voltage level and capacity limit prob-
lems in todays distribution network. Hence, broad studies, with
dependable control mechanisms for the coordinated operation of
various types of voltage regulating options including DG units,
are required. In addition, the formulation for the multi-objective
optimization with coordinated voltage control involving DG units
needs to be understandable and economical. An effective method-
ology for multi-objective DG operation for distribution system
volt/var control during normal and emergency situation is there-
fore vital.
There have been numerous studies solving optimization prob-
lems for coordinated voltage control in distribution systems. Some

http://dx.doi.org/10.1016/j.epsr.2015.05.018
0378-7796/ 2015 Elsevier B.V. All rights reserved.
42 M.Z. Degefa et al. / Electric Power Systems Research 127 (2015) 4152
studies considered the coordination of only two voltage con-
trol methods, while others investigated all the available methods,
including the reactive power control of DG units. The study in [5]
presents a coordination of OLTC and Static-var-Compensators (SVC)
in an unbalanced distribution system. The proposed approach is a
two-stage decision making procedure, where in stage one, an opti-
mization problem of loss minimization is solved and in stage two,
the minimization of switching due to economic and technical con-
siderations is solved. Nevertheless, in [5] the DG units have been
assumed to operate with unity power factor; no reactive power
supply is considered from the DG units. In [6], coordination among
multiple SVCs, OLTC and DG units is presented for an online voltage
control in distribution systems. The synchronous machine-based
renewable DG units are also involved in the voltage regulation,
which minimized, according to [6], the total tap operation of SVRs.
The method in [6] uses pseudo measurements for load and DG gen-
eration; however, it doesnt include the uncertainties involved with
the measurements or uncertainties in the voltages from distribu-
tion state estimation. The study in [7] proposes a hierarchical rule
based coordinated voltage control strategy involving OLTCs and DG
units. In mitigating voltage problems, predened steps tap chang-
ing steps, Q regulation of DG units and PQ regulation of DG units
are initiated sequentially until the problem is alleviated. In the
studies [4,8], the authors presented a method to enable distribu-
tion system operators to integrate the voltage level management
potential of DG units in their network operation and planning prin-
ciples. These studies are efforts to relate the vast optimization and
rule-based coordinated voltage control theoretical studies to the
practical conditions of the existing network. In [8], a planning pro-
cedure is proposed so that implementation in the currently used
network planning tools is convenient.
The coordination between substation voltage and DG reactive
power is claimed to be the least cost method in [4], which used sta-
tistical distribution planning to select voltage control strategy. In
the statistical planning, one year load and production curves were
used to conduct load ows which can then be used to evaluate the
costs of different control strategies [4]. Nevertheless, the rule-based
procedures in [4,8] do not investigate the costs of the optimal coor-
dination of different voltage control strategies. A comprehensive
voltage control strategy among OLTCs, substation switched capac-
itors and feeder-switched capacitors is presented in [9]. The study
also investigated the impact of DG units on the control strategies,
in which they found that a constant voltage operation of DG units
is benecial for a signicant reduction in OLTC operation. In [9],
however, the DG is set to generate constant active power, which is
not possible in the case of PVs and Wind, ruling out the possibility
of PQ control of DG.
The study in [10] proposes a simple DG local reactive power
control with occasional communication with distribution network
operators (DNO). The proposed approach aims to guarantee that
active power generation does not cause voltage rise. With an objec-
tive function that minimizes DG curtailment and voltage violations,
the study in [11] proposes a comprehensive centralized voltage
constraint management approach. Provided that the VCM problems
are formulated properly, the MINLP solvers generally provide an
acceptably fast solution, as most distribution systems are equipped
with a relatively small number of discrete control means [11]. In
[11], the transfer of DG between feeders using only remotely con-
trolled switches only occurs when the DG curtailment cost exceeds
the cost of switching.
There are only a few studies (such as [12,13]) which tackle the
real-time management of voltage and thermal constraints local
to DG connection. The decentralized approach in [12] aims to
avoid extensive sensing and communications, where the thermal
constraint is managed by setting a constant line capacity thresh-
old which triggers the trimming of wind generation if violated.
A centralized management of thermal constraints, while inhibi-
ting violation of voltage limits, is presented in [13] by employing
remotely controlled switches to reduce the DG curtailment. The
study claims that the additional degree of freedom provided by
remotely controlled network switches leads to less DG curtailment.
However, as in [12], in [13] the thermal limits of all lines are set to
a constant value.
In practice, both the capacity threshold and DG generation
uctuates a lot, following weather variations in real-time. Hence,
dynamic thermal rating is proposed in this study to manage voltage
level and network losses while the real-time line thermal limit is
being respected in the constraint. In addition, in this study OLTCs,
DG units and SVCs are coordinated in addressing the voltage and
thermal constraints in an active distribution network.
The purpose of this study is to provide a CVC strategy for
the day-ahead operation of an active distribution network with
updated network capacity using real-time thermal rating (RTTR).
The CVC involves tap changing transformers, switchable static VAR
compensators and DG units. The day-ahead control strategies use
day-ahead forecasts of load, weather variables and DG generation.
Voltage penalty function and network loss minimization objectives
are compared in solving the CVC problem involving DG units.
The limitations of the method proposed in this study are mostly
associated with uncertainties in load and weather variable fore-
casts. Further, the execution complexity due to inconsistencies
between the planned day-ahead settings of voltage control devices
and their intraday local measurement-based closed loop operation
pose a challenge. Nevertheless, with the development of network
operation planning from day-ahead to hour- and minute-ahead
planning, the inaccuracies are likely to become less signicant.
Section 2 reviews the various voltage regulating mechanisms
and discusses the reactive power potential of DG. In Section 3,
a brief discussion of the real-time thermal rating (RTTR) method
employed in this study is presented. The subsequent section, Sec-
tion 4, presents the CVC formulation, which incorporates the DG
Q/V droop control variables and network component ratings. In
this section, the loss minimization and voltage deviation penalty
function based objectives are also discussed. In Section 5, the test
active distribution network and the load and weather variables uti-
lized in the analysis are presented. Section 5 also discusses the main
observations of the analysis while Section 6 briey discusses prac-
tical concerns related to communicating control set-points. The last
section, Section 7, summarizes the main ndings of this study.
2. Voltage control methods
There are two typical voltage level problems in distribution
systems. The short-term problem, which lasts for not more than
a minute, and the long-term problem, where the voltage level
remains outside the 10% limit for more than 1 min. Over voltage
and under voltage events require proper management that utilizes
the dependencies of voltage and other variables, such as active and
reactive load and generation, as shown in Fig. 1 and (1). The voltage
at busbar 2 in Fig. 1 can be approximated as
R (PG PL ) + (QG QL QC ) X
V2 V1 + (1)
V2
Fig. 1. A simple illustration of voltage dependency in distribution network [14].
 M.Z. Degefa et al. / Electric Power Systems Research 127 (2015) 4152 43

where PG and QG are local generation connected with local load PL ,

and QL and QC are supplied from a reactive compensator.

2.1. Distribution system voltage regulation devices

2.1.1. On-load tap-changers (OLTCs)

An OLTC is a transformer component controlled automati-
cally by a relay to increase or decrease voltage by altering the
tap position of the transformer [14]. Usually, OLTCs operate fol-
lowing the connection point voltage level as a feedback control
loop. In a CVC scheme, however, the operation of OLTCs might
be needed for correcting voltage level problems elsewhere in
the distribution network. The CVC scheme, therefore, needs to
communicate the optimized set-points for the OLTCs. Neverthe-
less, the OLTCs might also need to take corrective actions for
local contingencies without waiting for the centrally running opti- Fig. 2. Inverter reactive power capability curve. (For interpretation of the references
to color in this gure legend, the reader is referred to the web version of this article.)
mization. Hence, an intelligent OLTC operating with two layers
of control regimes, central and local, is the solution. This con-
trol scheme is presented briey in Section 6. Nevertheless, in the
test case, OLTCs receiving only centrally optimized set-points are In a highly resistive network, such as a low voltage distribu-
considered. tion network, the voltage is more dependent on active power than
The OLTCs are mainly installed in the HV/MV primary sub- reactive power. Conversely, when the reactance of a power system
station transformer. Nevertheless, their installation in the MV/LV network is more signicant than the resistance the voltage will be
secondary substation has been recommended to abet the voltage more sensitive to changes in reactive power. Hence, in the former
problem introduced by DG units, as in [15]. In this study, the optimal scenario a P/V droop controller is recommended and in the latter
placement or optimal OLTC installation problems are not discussed. case a Q/V droop controller would be more effective. DG control for
However, with a given OLTC installation level and without further voltage regulation involves either soft or hard curtailment. With
device investment, we attempt to improve voltage quality through hard curtailment, we disconnect the DG altogether. However, with
CVC, involving DG units and an RTTR system. soft curtailment either of the P/V or Q/V droop control methods can
be utilized.

2.1.2. Switchable static VAR compensation devices (SVC) 2.1.3.1. Q/V droop controller. This type of control is also called volt-
An SVC is a shunt-connected static var generator or sink whose age dependent reactive power control. The DG units considered
output is adjusted to exchange capacitive or inductive current. in this study are PVs and wind turbines. The Q/V droop method
Hence, SVCs are capable of either supplying or absorbing reactive does not need a source of real power for generating the neces-
power. The response time of SVCs is also fast enough to respond to sary reactive power for compensation. Since the main objective of
transitional voltage uctuations. In this study, the SVCs comprise PV generators is to produce active power, their reactive power is
either thyristor-switched capacitors or reactors. limited to the maximum apparent capacity of the inverter.

V

  Qmax , V < 0.9 or V > 1.1
2.1.3. DG reactive power control methods
V 

The reactive power capability of DG units (wind turbines and

solar PVs in this study) originates from their inverter circuits con-

0, 1 D V 1 + D
necting to the grid. Except for limiting the maximum reactive power Q = Q
max (3)

(V + 1 + D) , 1 + D < V 1.1
intake or supply, the active power generated does not have an
0.1 D


impact on the reactive power capacity of DG units, as formulated in

(2). This fact enables solar PVs to be utilized in reactive power bal- Qmax V + 1 D , 0.9 V < 1 D
0.1 D
ance, even during night time when no generation is possible. The
inverter apparent power rating, however, sets the absolute possible Qmax = P tan (acos (PFlim )) (4)
reactive power capacity, as shown in (2) and Fig. 2.
where the power factor limit is PFlim and the deadband range set-
 ting value is D. The graphical presentation of (3) and (4) is presented
Qmax (t) = 2
Smax Pact (t)2 (2) in Fig. 3.

where Smax (red line in Fig. 2) is the inverter apparent power rating,
Pact (t) is the actual active power generated by DG at time t (green
line in Fig. 2) and Qmax (t) is the corresponding maximum reactive
power supply or intake capability.
Standards such as IEEE 1547 do not currently allow active
voltage regulation by DG inverters at the point of common cou-
pling (PCC) to access maximum amount of real power and to
avoid unnecessary interaction among voltage controllers. Other-
wise, there is no other technical limitation inhibiting the usage
of reactive power of DG units. Many inverters have the capabil-
ity of providing reactive power to the grid. Inverters can also be
oversized just to increase the capacity of their reactive power
balance. Fig. 3. Q/V control with dead-band.
44 M.Z. Degefa et al. / Electric Power Systems Research 127 (2015) 4152
2.1.3.2. DG P/V droop controller. The P/V droop can be implemented
with either soft curtailment in the case of dispatchable DG or dis-
connection of the DG altogether in the case of non-dispatchable DG.
The P/V droop controllers are effective in resistive networks. Both
the P/V and Q/V -control could be a constant droop or with a dead-
band. In this way, the DG units would only react to voltages that
exceed a certain threshold voltage. In the P/V droop control, the DG
units contribute to the load sharing, depending on both their ratings
(droops) and the line impedances. The P/V droop control strategy
is based on the formulation shown in the following equation:
are used in the RTTR system. The underground cable thermal model
developed by the authors and the implementation procedures of
the RTTR system are detailed in [16,17].
4. Optimal day ahead CVC formulation
The main goal of CVC in distribution networks is to compen-
sate for the load and DG generation variations so that the customer
supply voltages are kept within certain bounds. The utilization
of DG units coordinating with SVCs and OLTCs for voltage con-
 
Pmax,i ki Vi Vcrit,i Vi Vcrit,i
Pi =
Pmax,i Vi < Vcrit,i
where Vcrit,i is the voltage above which the power injected by the DG
is decreased with the droop coefcient ki and Pmax,i is the nominal
power or the maximum power available by the DG. Vi and Pi are
the voltage level and DG output at node i. The P/V droop equation
in (5), hence, executes the active power curtailment required for
keeping the bus voltage within an acceptable range. In this study,
the Q/V droop controller method with dead-band is used to control
the reactive power supply or intake of DG units.
3. Real-time thermal rating
Static distribution network ratings are usually calculated
assuming conservative weather conditions. For example, the rat-
ing of overhead lines is calculated under low wind speed conditions
and high ambient temperature. These ratings are usually provided
seasonally and are xed within the season. The RTTR method, how-
ever, estimates the real-time thermal states and provides hour
ahead ratings, for example, by using the previous loading condi-
tions and local weather measurements or forecasts. The weather
variables relevant to the thermal ratings are wind speed, ambi-
ent temperature, solar irradiation and ground soil temperature.
The emerging active distribution network is facing distributed
generation and demand response induced stochastic loading char-
acteristics. Hence, to implement the real-time monitoring and
operation of an active distribution network, a real-time thermal
rating system is essential.
The RTTR system requires dynamic thermal models of the essen-
tial components such as underground cables, overhead lines and
substation transformers. Thermal models are expected to be capa-
ble of giving the maximum next hour carrying capacities by using
the current thermal state information and forecasts of weather vari-
ables. The equivalent ladder network circuit of underground cables
installed in unlled conduit is presented in Fig. 4. It shows the
thermal resistances and thermal capacitances emulating the tran-
sient thermal responses. IEEE Std. 738 and IEEE Std. C57.91 loading
guidelines for overhead conductors and oil-immersed transformers
trol can impose higher loading on the network. In other words,
(5) to tap the supply and absorption capacity of reactive power in a
distribution system the lines and cables are required to cope with
increased loading. The network loading capacity in a static rating
regime, however, is heavily limited compared to real-time thermal
rating.
The day-ahead CVC framework presented in Fig. 5 comprises
weather variable and load forecasts. The DG output forecast is based
on the weather variables, while the component ratings depend
on both the weather variables and forecasted loading. The errors
in the forecasting will in fact be carried through to the CVC set-
tings. Hence, evaluating the uncertainties in the control settings
is also equally crucial. Besides, while being the general day-ahead
guideline for the voltage regulating devices, the optimal control set-
tings from the proposed framework require updates within each
day. In this study we mainly focus on providing the optimal CVC
set-points for day-ahead distribution network operation. Practical
matters, such as communication, uncertainties and contingencies
are briey discussed in Section 6 from the multi-agent system per-
spective.
In the optimal decision of set-points for CVC, the ratings of
components such as lines and cables must be considered. How-
ever, these capacity limits could either be set based on static rating
principles or real-time thermal rating methods. The incorporation
of an RTTR system with CVC brings two main benets: the rst

Fig. 4. A seven loop thermal model for an underground cable inside an unlled
conduit.
Fig. 5. Three level CVC strategy procedure.
 M.Z. Degefa et al. / Electric Power Systems Research 127 (2015) 4152 45

is the lower cost of operation of a facility involved in CVC and

the second is a higher utilization and integration potential of DG
units.

4.1. Objective functions

Dening an objective function to maintain power quality and

reduce losses is a challenge. For loss minimization, the reactive
power ow along the branch is preferred to be zero (i.e. Qij = 0).
Nevertheless, to maintain the voltage variation between buses i Fig. 6. Voltage penalty function.
and j at a minimum, the reactive power ow is preferred to be
equal to (rij /xij ) Pij . In [18], it is suggested that a control scheme 4.2. The equality constraints
should be adaptable to easily allow for smooth transitions between

  V  G
emphases on power quality and distribution losses. The minimiza- N
   
tions of loss and penalty objective functions formulated in (6) and PDGi + PDi + Vi j ij cos ij + Bij sin ij = 0 i = 1, . . ., N (8)
(7), respectively, are compared in Section 5. j=1
Loss minimization demonstrates the economical aspect while
voltage deviation in the objective represents the technical lim-
  V  G    
N

itations. The utilization of DG reactive power and impact of QCi QDGi + QDi + Vi j ij sin ij Bij cos ij = 0 i = 1, . . ., N
real-time thermal rating for CVC is the focal point of this study. j=1

Ultimately, however, advanced centralized optimal coordinated (9)

voltage controls are expected to be multi-objective, encompassing
the following: Vi

V  PDGi tan (acos (PFlim )) , Vi < 0.9 or Vi > 1.1

i


Maintain voltage within permitted range and atten voltage pro-
D Vi 1 + D
0, 1
le along feeders QDGi = P tan (a cos (PF )) (10)
Minimize the sum of:

DGi lim
(Vi + 1 + D) , 1 + D < Vi 1.1


0.1 D
o Energy losses

o Curtailed DG energy

PDGi tan (acos (PFlim )) Vi + 1 D , 0.9 Vi < 1 D
0.1 D
o Operation (wear) of network components (OLTC, VR, SC etc.)
o Reactive power ow through HV/MV transformer where  ij = i j is the voltage angle difference between connected
o Reactive power injection/absorption by DG buses i and j. QDGi is the reactive power supplied or absorbed by
the DG connected to bus i. (8) and (9) calculate active and reac-
tive power injections at bus i, respectively. The network admittance
4.1.1. Loss minimization objective matrix is given by Yij = Gij + i Bij . (10) sets the DG reactive power
The loss minimization objective is formulated in the following supply following the Q/V droop characteristics,
equation:
4.3. Inequality constraint

N
N
  
Min PL = gi,j Vi 2 + Vj 2 2Vi Vj cos i j (6) Vimin Vi Vimax , i = 1, . . ., N, (11)
i=1 j=1 QDGi min QDGi QDGi max , i NDGq (12)
j=
/ i
QCi min QCi QCi max , i NCq (13)

where N is the total number of buses and gk (gi,j ) is the conductance akmini aki akmaxi i NOLTC (14)
of branch k or the line connecting node i to j. Vi and Vj are the  
Sij  Sij,max (15)
voltage magnitudes and i and j are voltage angles for nodes i and
j, respectively. 0 PDGi PDG gen,i , i = 1, . . ., N, (16)

where (11) limits the upper and lower voltage magnitude at bus i,
4.1.2. Voltage penalty function objective (12) sets the limits for DG reactive power at bus i, (13) limits the
In addition to the loss minimization objective dened in (6), a reactive power supply from the SVCs, (14) gives the range of OLTC
penalty function for the voltage exceeding a certain threshold can set-points and (15) gives the real-time dynamic thermal rating or
be dened as in the following equation: static rating of branch ij. Eq. (16) prevents the utilized DG generated
power PDGi at node i from exceeding the actual DG generated power
PDG gen,I , NDGq , NCq , and NOLTC are the set of all DG units, SVCs and
k(V Vi )2 ; Vi < Vimin
imin OLTCs connected at bus i, respectively.
Wi = 0; Vimin Vi Vimax (7) Sij,max is the capacity of the line or cable which is calculated

using the dynamic thermal models of the overhead lines and under-
k(Vi Vimax )2 ; Vi > Vimax ground cables. The previous hours optimal loading of the branch
is used to attain the initial conductor temperature for calculating
where k is the penalty factor. the coming hours line or cable rating. Hence, the iteration between
The objective function in (7), also drawn in Fig. 6, can be used the OPF with CVC and the DTR is established. In the optimization
when the voltage quality is more signicant than the overall distri- problem dened by (6)(16), the control variables are the reactive
bution network losses. power of the DG units (QDGi ), the reactive power of the SVCs (QCi )
46 M.Z. Degefa et al. / Electric Power Systems Research 127 (2015) 4152
Fig. 7. Greeneld network plan based on actual loading data and geographical location [21]. (For interpretation of the references to color in this gure legend, the reader is
referred to the web version of this article.)
and the tap set-point (ak ). The state variables, on the other hand,
are the bus voltage magnitude and angle, while the line load ows
are output variables.
Due to the nonlinearity of power systems, linear programing
loses accuracy due to linear assumptions [19]. Hence, the OPF is
solved using the nonlinear IPOPT solver of the General Algebraic
Modeling System (GAMS), while the DTR has been implemented
in MATLAB [20]. For numerical efciency within the power ow
or the OPF solution all discrete variables (transformer taps and
shunt steps) are treated as continuous until the optimal solu-
tion is found. Then they are rounded off to their nearest discrete
values.
The day-ahead dynamic thermal ratings of the distribution net-
work components in Fig. 5 are attained from the load and weather
variable forecasts input into the dynamic thermal models of the
components. The day-ahead dynamic thermal rating procedure is
presented in [17].
5. Test case: An active distribution network with DG units
and voltage regulating devices
For the test case, a Greeneld distribution network plan based
on actual loading data and geographical location is used, as shown
in Fig. 7. In the single line network diagram the green lines represent
MV underground cables while the red lines are overhead. The test
distribution network has 146 20/0.4 kV secondary substations and
a 110/20 kV primary substation.
The distribution network has voltage regulating devices, an
OLTC at primary substation and two SVCs installed at the load
center of the network. Besides, wind turbines are installed at four
locations and solar panels are installed at three locations, as shown
in Table A.1. In this work, various formulations of the CVC problem
for day-ahead distribution network operation are investigated. The
most important alternative solutions studied are the involvement
of DG units in supplying or absorbing reactive power through the
CVC scheme and the implementation of a real-time thermal rating
system.
In the test network, the secondary substation one day loading is
selected randomly from a pool of hourly AMR meter measurements
from actual households. The random selection considers the data
pool to be uniformly distributed. The respective reactive power
loads are calculated to correspond to the residential loading power
factor range of 0.955 to 0.98. Although the forecasts of the day ahead
load and weather variables are used to compute the optimal CVC
set-points for day ahead network operation, the forecasting method
is not elaborated in this study. The day ahead dynamic thermal rat-
ings and weather and load forecasting methods employed in this
study are presented in publications [16,17].
By denition, the DG penetration level in a system is the ratio of
gross annual energy generated by the DG to the total annual energy
demand in the system. In this study DG penetration levels of 1.13%,
11.3%, 22.6% and 65% are investigated. The DG capacity presented
in Table A.1 is at 1.13% penetration level, which is also the rst case
scenario. Mainly the scenarios for ten and twenty-fold increases
in the DG penetration levels are investigated, which are 11.3% and
22.6%, respectively. To represent a high DG penetration we also
simulated a 65% penetration level. The installed DG units, however,
are not evenly distributed throughout the network; rather, only the
installed capacities are varied at their respective xed geographical
locations, as shown in Fig. 7.
There are around 30 radial feeders connected to the primary
substation, of which the longest line is 60 km. Details of the test
distribution network are provided in Tables A.1 and A.2. In Fig. 8,
the bus voltages of the feeder starting from the primary substation
are plotted for a network without DG units and a network with 65%
penetration of DG units. There is no CVC applied in the case shown
in Fig. 8 and the voltage levels are the average of the 24 h voltage
levels. The test distribution network without DG integration (see
Fig. 8a) has 2.4% higher losses on the selected day than the DG inte-
grated test network (see Fig. 8b) with a 65% DG penetration level.
In Fig. 9, the substation voltage level on the 30 feeders of the test
 M.Z. Degefa et al. / Electric Power Systems Research 127 (2015) 4152 47

1.001

Voltage level (pu)

0.999

0.998

0.997

0.996
(a) 0 10 20 30 40 50 60
Distance from primary substation (km)

1.15

1.1
Voltage level (pu)

1.05

0.95
(b) 0 10 20 30 40 50 60
Distance from primary substation (km)

Fig. 8. Daily average voltage levels of buses on the 30 feeders radiating from the primary substation for the test distribution network without DG (a) and with 65% DG
penetration (b) (in (b) the blue feeders are connected with wind turbines and the red feeders are connected with solar panels). In both (a) and (b), there is no CVC applied.
(For interpretation of the references to color in this gure legend, the reader is referred to the web version of this article.)

network is plotted before and after the CVC is applied. Fig. 9a shows
the case for a 65% DG penetration level while Fig. 9b shows the case
for 1.11% DG penetration. The CVC corrects both the overvoltages
caused by DG connected at the end of radial feeder and undervolt-
ages due to long distances from the primary substation. From Fig. 9
it is also apparent that the two-third rule for SVC placement on a
radial feeder does not apply in the test network, where the DG units
are scattered at the ends of the feeders.
In this study, the day ahead model predictive CVC strategy is
applied on the test distribution network by varying the CVC objec-
tive functions and the network rating types. The results in Table 1
compare the six types of CVC formulation involving DG units at
1.13% penetration level with DG curtailment, and at 11.3% and
22.6% penetration level, with and without DG curtailment. Table 1
presents the overall network losses after applying the control set-
points for a period of 24 h. Besides, the number of OLTC and SVC
operations per day, and the cumulative reactive power supplied by
DG units and SVCs are presented. The lowest network losses are
attained, as expected, for a CVC formulation where loss minimi-
zation is the objective. Apparently there is no difference between
static and dynamic rating when loss minimization is the only objec-
tive as DTR does not directly reduce losses. Nevertheless, the DG
soft curtailment is too high when voltage loss minimization is the
objective. DTR shows a clear advantage over static rating in terms
of maximizing the utilization of the DG units, while SVCs are effec-
tive for voltage control and reducing the stress on OLTCs. However,
the greater utilization of active power generation from the DG units
came with a slight increment in losses. A voltage penalty function
can minimize the voltage burden on the customer. As shown in
Fig. 10, the static and DTR ratings coupled with both the voltage
penalty function and loss minimization gave the voltage level clos-
est to the nominal. DTR, therefore, enables the proper utilization
of resources for CVC, where a superior benet can be attained by
using DTR coupled with the penalty function and loss minimization
objectives.
The following characteristics of the CVC objective function for-
mulation methods are observed:
Observation no. 1: The CVC formulations experienced a higher
voltage level from the nominal without the involvement of DG reac-
tive power capacity than with (see Fig. 11). In addition, the optimal
set-point is achieved at signicantly higher network loss values for
CVC without DG reactive power involvement.
Observation no. 2: With the voltage deviation penalty function
objective; there is a higher requirement for the operation of the
SVCs and OLTCs, as shown in Table 1. In both Static and DTR rating
methods, the voltage penalty function objective utilizes a higher
reactive power supply from the SVCs and OLTCs while injecting
reactive power into the DG units. This is because the objective
places less emphasis on losses, leading to a greater tendency to
use reactive power absorption and supply to keep the voltage level
in the entire network close to unity. With the voltage penalty func-
tion objective, the voltage level stays closer to the at one per
48 M.Z. Degefa et al. / Electric Power Systems Research 127 (2015) 4152

1.16
65% DG penetration without CVC
1.14 65% DG penetration with CVC+DTR+LOSS+PENALITY

1.12

Voltage level (pu)

1.1

1.08

1.06

1.04

1.02

0.98
0 10 20 30 40 50 60
(a) Distance from primary substation (km)

1.0005
1.13% DG penetration without CVC
1.13% DG penetration with CVC + DTR + Loss

1
Voltage level (pu)

0.9995

0.999

0.9985
0 10 20 30 40 50 60
(b) Distance from primary substation

Fig. 9. An average day-ahead voltage level prole of the secondary substations on the 30 feeders, with and without the implementation of CVC. (a) shows the prole with
65% of DG penetration and (b) shows the result with a 1.13% DG penetration level.

unit compared to the loss minimization objective function. How- also lower the DG active power generation curtailment better than
ever, the network losses increased tremendously with the voltage static ratings, as shown in Table 1.
penalty function objective, as shown in Table 1. With the penalty Observation no. 3: DTR shows no signicant difference from
function objective, DTR can use the reactive power resources and static rating when loss minimization is an objective. In principle,
the DTR utilizes the opportunities provided by the weather for the
cooling of distribution network components. It is evident that with
1.0015
No CVC
CVC+STATIC+Loss
CVC+STATIC+Penality
1.001 CVC+DTR+Loss 1.15
CVC+DTR+Penality No CVC
Bus voltage magnitude (pu)

CVC+STATIC+Loss+Penality STATIC loss minimization CVC without DG

CVC+DTR+Loss+Penality
1.0005 STATIC loss minimization CVC with DG
Voltage level (pu)

1.1

1.05

0.9995

1
0.999 0 10 20 30 40 50 60
0 5 10 15 20 25 30 35
Distance from primary substation (km) Distance from primary substation (km)

Fig. 10. Voltage level at hour 4:00 on feeder 10, where a wind turbine is connected Fig. 11. Voltage level on the longest feeder of the network after implementation
at the furthest node. The DG penetration level is 1.13% and the CVC involves DG of CVC with and without the involvement of DG reactive power supply. (In all the
curtailment. three scenarios, 65% DG penetration.)
 M.Z. Degefa et al. / Electric Power Systems Research 127 (2015) 4152 49

Table 1
Evaluation of the CVC formulation techniques for the period of 24 h ahead implementation.

No. Specication DG penetration DG Losses Max. No. of OLTC No. of SVC DG Qgen. DG Pgen SVC Qgen.
(objective and rating) (%) curtailment (kW h) Vdev* (%) operations operations (MVARh) curtailed (%) (MVARh)

1 Obj: losses & rating: 1.13 Yes 25.6 0.14 0 0 2.439 31.63 0.856
static 1.13 No 34.5 0.14 0 0 1.016 0 0.994
11.3 Yes 23.0 0.14 0 0 1.459 90.39 0.946
11.3 No 2,610.7 0.91 0 12 36.455 0 7.521
22.6 Yes 22.8 0.14 0 0 1.457 94.97 0.946
22.6 No 10,352.5 1.83 0 27 73.620 0 15.146

2 Obj: losses & rating: 1.13 Yes 25.6 0.14 0 0 2.438 31.64 0.858
DTR 1.13 No 34.5 0.14 0 0 1.017 0 0.991
11.3 Yes 23.0 0.14 0 0 1.459 90.39 0.947
11.3 No 2,609.6 0.89 0 10 36.161 0 6.584
22.6 Yes 22.831 0.14 0 0 1.454 94.97 0.954
22.6 No 10,352.5 1.83 0 25 73.620 0.00 15.147

3 Obj: penalty & rating: 1.13 Yes 47.9 0.24 11 1 1.506 2.27 9.866
static 1.13 No 55.8 0.28 9 2 2.185 0 11.915
11.3 Yes 2,981.9 3.56 10 20 48.952 0.69 50.989
11.3 No 3,261.6 3.56 16 31 50.713 0 56.733
22.6 Yes 10,798.6 1.86 16 31 87.857 0.12 63.214
22.6 No 11,502.7 3.88 17 32 92.136 0 78.740

4 Obj: penalty & rating: 1.13 Yes 49.8 0.21 5 0 1.647 2.10 10.292
DTR 1.13 No 57.0 0.29 13 0 2.296 0 12.351
11.3 Yes 2,926.5 1.94 13 19 49.411 1.39 53.125
11.3 No 3,041.3 2.54 12 27 48.449 0 48.768
22.6 Yes 10,866.95 1.92 7 22 89.359 0.14 68.587
22.6 No 11,546.9 3.99 16 36 94.552 0 87.551

5 Obj: losses + penalty & 1.13 Yes 36.0 0.16 6 2 0.929 4.23 1.867
rating: static 1.13 No 34.5 0.14 0 0 1.021 0 0.979
11.3 Yes 2,544.5 0.89 5 0 36.076 0.97 6.775
11.3 No 2,613.9 0.91 2 3 36.641 0 8.051
22.6 Yes 10,302.71 1.83 4 4 73.316 0.31 14.752
22.6 No 10,352.5 1.83 0 3 73.619 0 15.143

6 Obj: losses + penalty & 1.13 Yes 34.6 0.14 8 0 1.026 3.77 1.454
rating: DTR 1.13 No 34.5 0.14 0 0 1.018 0 0.990
11.3 Yes 2,542.9 0.89 4 0 36.113 0.82 6.752
11.3 No 2,610.1 0.90 2 1 36.352 0 7.213
22.6 Yes 10,338.9 1.83 4 3 73.553 0.14 15.142
22.6 No 10,352.5 1.83 0 3 73.619 0 15.144
*
Vdev = max(|1 Vij |), i = 1; 24, j = 1; 147.

DTR network losses could even increase as long as they remain Observation no. 4: The combined loss minimization and voltage
within the economically acceptable level. As shown in Fig. 12 and penalty objective function with both static and DTR rating reduced
Table 1, the voltage penalty minimization objective provides bet- signicantly the operations of OLTCs and SVCs, while utilizing both
ter voltage levels with DTR than with static rating, even though the the active and reactive power resources of the installed DG units to
total losses are the same. the utmost. This has been observed in all the four cases presented
in Table 1.
The simulation results for the network loss minimization
objective and the voltage penalty function objective in the CVC
1.035
DTR voltage penalty minimization formulations are presented in Table 1. When network losses are
STATIC voltage penalty minimization insignicant, such as in the test network in Fig. 7, the voltage
1.03
penalty function objective performs better, by bringing the volt-
1.025 age to a at stable level close to one per unit. In addition, the
incorporation of static rating or DTR with a loss minimization objec-
Voltage leve (pu)

1.02
tive brought down the network losses by about 29%, compared
1.015 to the static or DTR rating with only the voltage penalty func-
tion objective. Furthermore, when the voltage penalty function and
1.01
loss minimization objectives are used together the observed overall
1.005 performance is superior.

1
6. Discussion on practical implementations
0.995
0 20 40 60 80 100 120 140 150
The 147 buses in order of their number Planning CVC for day-ahead distribution network operation
requires reliable load and environmental variable forecasting.
Fig. 12. The voltage level at hour 16:00, after the static and DTR based CVC control
settings are applied with the voltage penalty minimization objective and with 20% Nevertheless, the control setting planned a-day-before faces two
DG penetration. fundamental challenges. The rst is the question of where to
50 M.Z. Degefa et al. / Electric Power Systems Research 127 (2015) 4152

the benets, primarily by reducing DG curtailment as observed in

[13].

7. Conclusion

In this study, the synergy between coordinated voltage con-

trol and real-time dynamic thermal rating is investigated for
an increased utilization of DG reactive power potential. Beyond
its foremost benets, enhancing network capacity and efcient
utilization of network components, dynamic thermal rating has
demonstrated its advantage for a coordinated voltage regulation
involving DG units. The benet of incorporating DTR with CVC
is two-pronged. On one hand, the generated DG active power
can be utilized better, due to an increased distribution network
component carrying capacity. On the other hand, the required
reactive power for voltage regulation can be transferred without
violating network component limits if DTR is applied in a real-
time basis. In the analysis, the CVC formulation with a combined
loss and voltage penalty function objective and real-time dynamic
thermal rating method showed better results in both minimiz-
ing the network losses and keeping the voltage close to one per
unit.
The day-ahead CVC network operation planning framework is
Fig. 13. Integrated local and central voltage control strategy. not tested with very large networks, where computation times
could better be investigated. The complete intended application,
involving the interaction between day-ahead network operation
conduct the central optimization for CVC and how to communi- planning and intraday local network operation control actions, has
cate the optimal set-points to the respective equipment such as not yet been simulated.
OLTCs, SVCs and DG units. The second challenge is how to react With the improved measurement points in todays active dis-
to contingencies during the same day and the requirements for tribution network, there has never been such a high degree
updating. of visibility and forecasting potential for distribution network
Usually, voltage regulators operate by following the voltage component states and customer loads. Hence, multi-objective pre-
level at their connection point. In a CVC strategy, however, the opti- dictive control strategies could utilize DG for voltage regulation,
mal set-points are computed and communicated from elsewhere. and employ real-time thermal rating to increase the integra-
We propose a connection point voltage dependent on central and tion potential of new DG and to utilize available reactive power
local control strategies for the specic voltage regulators, as shown resources.
in Fig. 13. The day ahead optimal set-points will be sent to the
specic voltage regulating device that it follows, unless the con- Acknowledgements
nection point voltage level does not violate a certain threshold.
When the connection point voltage level violates a given preset The authors of this paper would like to acknowledge that this
level, however, the local closed loop control system will be ini- work is jointly funded by the Aalto energy efciency program
tiated (see Fig. 13). This strategy can be implemented naturally through the SAGA and STEEM projects, and would also like to thank
with the Multi-Agent System (MAS) control architecture of active Jussi Niskanen of Loiste Shkoverkko for his cooperation.
distribution networks. With the specied MAS approach, the volt-
age regulating devices need to be agents with decision making and
communication capability. Appendix A.
Another challenge in executing the optimal control settings is
the proper queuing of the control actions. Hence, not only do we Tables A.1 and A.2.
need to communicate the day-ahead control settings, but the asso-
ciated time lags are also needed.
The proposed method is mainly for active distribution net- Table A.1
DG, SVC and OLTC connections.
work operations planning which might range from minutes ahead
to up to a day-ahead. For real-time application, as is discussed Type Node DG rating and number
briey in this section, we propose a cooperative central and DG WIND 102 50 kW 2 Turbines
distributed control strategy. In the proposed scheme, the day- 108 50 kW 3 Turbines
ahead centralized operation planning (which can be referred to 92 50 kW 2 Turbines
as coordinated control agents) will be used as a guideline for 66 50 kW 3 Turbines
PV 28 215 Wp 8 Panels
local controls (unit control agents). We believe that, to accom-
83 215 Wp 32 Panels
modate the distributed potentials such as DG units and demand 117 215 Wp 16 Panels
response, a relaxed centralized cooperation strategy with decen-
SVC (0.5/opp.) Node Max (MVA) Min (MVA)
tralized control functions will be most useful. Such a multi-agent 85 15 0
based hierarchical hybrid control architecture is presented in 121 30 0
[22]. OLTC (0.0125/opp.) Place Max tap Min tap
In future work, the addition of network reconguration capabil- Primary 1.11 0.91
substation
ities in the proposed comprehensive CVC method would enhance
 M.Z. Degefa et al. / Electric Power Systems Research 127 (2015) 4152 51

Table A.2
The 147 bus active distribution test network specications (40 MVA base 20 kV MV system).

No. From To r (pu) x (pu) No. From To r (pu) x (pu)

1u 16 2 0.008 0.006 74 77 75 0.037 0.046

2u 32 3 0.005 0.004 75 97 76 0.017 0.021
3u 20 4 0.010 0.008 76 48 77 0.019 0.023
4u 13 5 0.005 0.004 77 81 78 0.042 0.052
5u 1 6 0.003 0.003 78 78 79 0.037 0.046
6u 38 7 0.004 0.003 79 86 80 0.021 0.026
7u 2 8 0.013 0.010 80 80 81 0.020 0.025
8u 41 9 0.019 0.015 81 46 82 0.036 0.044
9u 2 10 0.006 0.005 82 79 83 0.027 0.034
10u 1 11 0.053 0.042 83 76 84 0.043 0.053
11u 30 12 0.006 0.005 84 47 85 0.028 0.035
12u 14 13 0.006 0.005 85 94 86 0.032 0.039
13u 1 14 0.007 0.006 86 79 87 0.037 0.046
14u 3 15 0.005 0.004 87 89 88 0.024 0.029
15u 22 16 0.010 0.008 88u 90 89 0.020 0.016
16 33 17 0.021 0.026 89 27 90 0.285 0.351
17u 28 18 0.014 0.011 90 90 91 0.036 0.044
18u 21 19 0.005 0.004 91 55 92 0.082 0.102
19u 29 20 0.005 0.004 92 49 93 0.053 0.065
20u 1 21 0.017 0.014 93u 70 94 0.020 0.016
21u 1 22 0.018 0.014 94 74 95 0.047 0.058
22u 4 23 0.017 0.014 95 69 96 0.029 0.036
23u 8 24 0.027 0.021 96 48 97 0.027 0.033
24u 10 25 0.006 0.005 97 73 98 0.050 0.062
25u 35 26 0.004 0.003 98 134 99 0.130 0.161
26 37 27 0.022 0.027 99 99 100 0.026 0.032
27u 9 28 0.013 0.010 100 100 101 0.058 0.071
28u 12 29 0.005 0.004 101 101 102 0.051 0.063
29u 6 30 0.003 0.002 102 113 103 0.047 0.058
30u 1 31 0.022 0.018 103 103 104 0.029 0.036
31u 7 32 0.004 0.003 104 104 105 0.066 0.082
32 18 33 0.033 0.040 105 107 106 0.064 0.078
33u 30 34 0.002 0.002 106 108 107 0.104 0.128
34u 40 35 0.007 0.006 107 56 108 0.256 0.315
35u 41 36 0.005 0.004 108 64 109 0.022 0.028
36 17 37 0.011 0.013 109 111 110 0.077 0.095
37u 1 38 0.002 0.002 110 125 111 0.032 0.039
38u 5 39 0.006 0.005 111 110 112 0.019 0.023
39u 23 40 0.005 0.004 112u 100 113 0.023 0.018
40u 31 41 0.024 0.019 113 104 114 0.086 0.105
41 24 42 0.035 0.043 114 123 115 0.024 0.029
42 49 43 0.035 0.043 115 126 116 0.040 0.050
43u 19 44 0.018 0.015 116 110 117 0.028 0.035
44 78 45 0.018 0.023 117 116 118 0.028 0.035
45u 72 46 0.011 0.008 118 128 119 0.028 0.035
46 82 47 0.013 0.017 119 135 120 0.032 0.039
47 72 48 0.032 0.039 120 140 121 0.028 0.035
48 84 49 0.035 0.043 121 138 122 0.027 0.034
49 68 50 0.046 0.057 122 122 123 0.041 0.051
50 63 51 0.113 0.139 123 115 124 0.055 0.068
51 43 52 0.027 0.033 124u 11 125 0.040 0.032
52 52 53 0.066 0.082 125 119 126 0.015 0.018
53 58 54 0.027 0.033 126 112 127 0.062 0.076
54 62 55 0.047 0.058 127 146 128 0.029 0.035
55 55 56 0.098 0.121 128 143 129 0.019 0.023
56 52 57 0.018 0.023 129 142 130 0.060 0.074
57 71 58 0.024 0.030 130 130 131 0.016 0.020
58 54 59 0.026 0.032 131 127 132 0.031 0.039
59 61 60 0.071 0.087 132 141 133 0.072 0.089
60 59 61 0.074 0.091 133 133 134 0.021 0.026
61 61 62 0.050 0.061 134 129 135 0.097 0.120
62 50 63 0.088 0.109 135 137 136 0.055 0.068
63 45 64 0.020 0.025 136 127 137 0.068 0.084
64u 51 65 0.011 0.008 137 139 138 0.072 0.089
65 65 66 0.013 0.016 138 140 139 0.026 0.032
66 43 67 0.019 0.023 139 136 140 0.039 0.048
67 96 68 0.032 0.040 140 124 141 0.032 0.039
68 84 69 0.049 0.061 141 129 142 0.046 0.056
69u 44 70 0.018 0.014 142 145 143 0.027 0.033
70 98 71 0.048 0.060 143 124 144 0.048 0.059
71 109 72 0.029 0.035 144 119 145 0.021 0.026
72 74 73 0.021 0.026 145 147 146 0.042 0.051
73 88 74 0.138 0.170 146 42 147 0.133 0.165
u
Branches of underground cables. All others are overhead lines.
52 M.Z. Degefa et al. / Electric Power Systems Research 127 (2015) 4152
References
[1] M.E. Elkhatib, R. El-Shatshat, M.M.A. Salama, Novel coordinated voltage control
for smart distribution networks with DG, IEEE Trans. Smart Grid 2 (Dec (4))
(2011) 598605.
[2] A. Samadi, R. Eriksson, L. Soder, B.G. Rawn, J.C. Boemer, Coordinated active
power-dependent voltage regulation in distribution grids with PV systems,
IEEE Trans. Power Delivery 29 (June (3)) (2014) 14541464.
[3] S. Engelhardt, I. Erlich, C. Feltes, J. Kretschmann, F. Shewarega, Reactive power
capability of wind turbines based on doubly fed induction generators, IEEE
Trans. Energy Convers. 26 (March (1)) (2011) 364372.
[4] A. Kulmala, S. Repo, P. Jrventausta, Using statistical distribution network plan-
ning for voltage control method selection, in: IET Conf. Proc. on Renewable
Power Generation, Sept 2011, Edinburgh, UK, 2011.
[5] N. Daratha, B. Das, J. Sharma, Coordination between OLTC and SVC for volt-
age regulation in unbalanced distribution system distributed generation, IEEE
Trans. Power Syst. 29 (1) (2014) 289299.
[6] D. Ranamuka, A.P. Agalgaonkar, K.M. Muttaqi, Online voltage control in
distribution systems with multiple voltage regulating devices, IEEE Trans. Sus-
tainable Energy 5 (Apr (2)) (2014) 617628.
[7] A. Viehweider, B. Bletterie, D.B. De Castro, Advanced coordinated voltage con-
trol strategies for active distribution network operation, in: CIRED 20th Int.
Conf. on Electricity Distribution, 811 Jun, 2009, Prague, 2009.
[8] A. Kulmala, K. Maki, S. Repo, P. Jarventausta, Including active volt-
age level management in planning of distribution networks with dis-
tributed generation, in: Proc. IEEE PowerTech, Jun. 2009, Bucharest, 2009,
pp. 16.
[9] F.A. Viawan, D. Karlsson, Voltage and reactive power control in systems with
synchronous machine-based distributed generation, IEEE Trans. Power Deliv-
ery 23 (Apr (2)) (2008) 10791087.
[10] P.M. Carvalho, P.F. Correia, L.A. Ferreira, Distributed reactive power generation
control for voltage rise mitigation in distribution networks, IEEE Trans. Power
Syst. 23 (May (2)) (2008) 766772.
[11] F. Capitanescu, I. Bilibin, E. Romero Ramos, A comprehensive centralized
approach for voltage constraints management in active distribution grid, IEEE
Trans. Power Syst. 29 (Nov (2)) (2013) 933942.
[12] T. Sansawatt, L.F. Ochoa, G.P. Harrison, Smart decentralized control of DG for
voltage and thermal constraint management, IEEE Trans. Power Syst. 27 (Mar
(3)) (2012) 16371645.
[13] I. Bilibin, F. Capitanescu, Contributions to thermal constraints management
in radial active distribution systems, Electr. Power Syst. Res. 111 (Jun) (2014)
169176.
[14] T.J.T. Hashim, A. Mohamed, H. Shareef, A review on voltage control methods
for active distribution networks, Prz. Elektrotech. 88 (Jun) (2012) 304312.
[15] M. Vandenbergh, V. Helmbrecht, D. Craciun, R. Hermes, H. Loew, Technical
solutions supporting the large scale integration of photovoltaic systems in the
future distribution grids, in: CIRED 22nd Int. Conf. on Electricity Distribution,
Jun 2013, Stockholm, 2013.
[16] M.Z. Degefa, M. Humayun, A. Safdarian, M. Koivisto, R.J. Millar, M. Lehtonen,
Unlocking distribution network capacity through real-time thermal rating for
high penetration of DGs, Electr. Power Syst. Res. 117C (2014) 3646.
[17] M.Z. Degefa, M. Koivisto, R.J. Millar, M. Lehtonen, Dynamic thermal state fore-
casting of distribution network components, in: 13th Int. Conf. on Probabilistic
Methods Applied to Power Systems (PMAPS 2014), 710 Jul., 2014, Durham,
2014.
[18] K. Turitsyn, S. Petr, B. Scott, C. Michael, Options for control of reactive power
by distributed photovoltaic generators, Proc. IEEE 99 (6) (2011) 10631073.
[19] W. Zhang, F. Li, L.M. Tolbert, Review of reactive power planning: objectives, con-
straints, and algorithms, IEEE Trans. Power Syst. 22 (Nov (4)) (2007) 21772186.
[20] E.R. Sichard, GAMS: A Users Guide, Tutorial, GAMS Development Corporation,
Washington, Mar. 2014.
[21] R.J. Millar, S. Kazemi, M. Lehtonen, E. Saarijrvi, Impact of MV connected
microgrids on MV distribution planning, IEEE Trans. Smart Grid 3 (4) (2012)
21002108.
[22] C.X. Dou, B. Liu, Multi-agent based hierarchical hybrid control for smart micro-
grid, IEEE Trans. Smart Grid 4 (Jun (2)) (2013) 771778.