IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 29, NO.

1, JANUARY 2014 121

Dynamic Optimal Power Flow for Active

Distribution Networks
Simon Gill, Ivana Kockar, Member, IEEE, and Graham W. Ault, Member, IEEE

Abstract—Active Network Management is a philosophy for Intertemporal equality constraints.

the operation of distribution networks with high penetrations of
renewable distributed generation. Technologies such as energy Intertemporal inequality constraints.
storage and flexible demand are now beginning to be included in Grid connection point number.
Active Network Management (ANM) schemes. Optimizing the op-
eration of these schemes requires consideration of inter-temporal Number of grid connection points.
linkages as well as network power flow effects. Network effects
are included in optimal power flow (OPF) solutions but this only Power injection from grid connection
optimizes for a single point in time. Dynamic optimal power flow point and limits on power from grid
(DOPF) is an extension of OPF to cover multiple time periods. connection point.
This paper reviews the generic formulation of DOPF before
developing a framework for modeling energy technologies with
Reactive power injection from grid
inter-temporal characteristics in an ANM context. The framework connection point and limits on reactive
includes the optimization of nonfirm connected generation, princi- power from grid connection point.
ples of access for nonfirm generators, energy storage, and flexible
demand. Two objectives based on maximizing export and revenue
Time-step.
are developed and a case study is used to illustrate the technique. Length of time step.
Results show that DOPF is able to successfully schedule these
energy technologies. DOPF schedules energy storage and flexible Number of time-steps in optimization.
demand to reduce generator curtailment significantly in the case Summation step-time variable.
study. Finally, the role of DOPF in analyzing ANM schemes is
discussed with reference to extending the optimization framework Bus number, number of busses.
to include other technologies and objectives.
Bus voltage and voltage limits.
Index Terms—Active Network Management (ANM), dynamic
optimal power flow (DOPF), energy storage, flexible demand, op- Line number, number of lines.
timal power flow (OPF). Apparent line power flow and apparent
power limit.
I. NOMENCLATURE Nonfirm generator number.
Number of nonfirm generator.
A. General DOPF Indices of nonfirm generator number.
Nonfirm power output, and maximum
Vector of OPF control variables.
nonfirm generator output.
Vector of OPF fixed parameters.
Priority order scaling constant.
Vector of intertemporal variables.
Priority order number.
Vector of OPF derived variables.
Vector of external power prices.
Objective function.
OPF equality constraints. B. Energy Storage System (ESS)
OPF inequality constraints.
Input and output ESS efficiency.
Manuscript received October 30, 2012; revised February 21, 2013; accepted ESS state of charge.
August 08, 2013. Date of publication September 11, 2013; date of current ver-
sion December 16, 2013. This work was supported in part by the Wind Energy
Permissible state of charge limits.
Systems Centre for Doctoral Training at the University of Strathclyde under
Grant EPSRC EP/G037728/1. Paper no. TPWRS-01215-2012.
Power injected into grid by ESS.
S. Gill is with the Wind Energy Systems Centre for Doctoral Training, Univer- Power injected into grid by ESS when
sity of Strathclyde, Glasgow, G1 1XW, U.K. (e-mail: [email protected]).
I. Kockar and G. W. Ault are with the Institute for Energy and En- charging (negative).
vironment, University of Strathclyde, Glasgow, G1 1XW, U.K. (e-mail: Power injected into grid by ESS when
ivana.kockar,[email protected]).
Color versions of one or more of the figures in this paper are available online discharging (positive).
at http://ieeexplore.ieee.org.
ESS rated power.
Digital Object Identifier 10.1109/TPWRS.2013.2279263

0885-8950 © 2013 IEEE

Authorized licensed use limited to: Politecnico di Bari. Downloaded on February 14,2024 at 17:35:34 UTC from IEEE Xplore. Restrictions apply.
122 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 29, NO. 1, JANUARY 2014

Reactive power injection from PSS. storage [11], and the operation of energy storage and wind gen-
eration [12], [13].
Energy capacity of ESS.
An area where inter-temporal technologies are being de-
ployed is the management of distribution networks (DNs).
C. Flexible Demand Active Network Management (ANM) is the philosophy of
planned and real-time management of a DN and connected de-
Power injected into grid by vices [14], [15]. It is being driven by the pressure to increase the
flexible demand units (negative). penetration of renewable generation connecting to the power
Rated power of flexible demand system as distributed generation (DG). Traditional operation
units. of DNs assumes a “fit-and-forget” strategy which significantly
limits DG penetration [16]. ANM allows additional DG to
Energy storage capacity of connect under nonfirm connection agreements and applies
SMFD. curtailment to these generators to maintain network constraints
Demand to extract heat from [17], [18]. Technologies such as energy storage systems (ESS)
SMFD. [19] and flexible demand [20], both of which are inter-temporal
Total energy demand of are now forming important parts of ANM schemes.
flexible demand units across ANM technologies help distribution network operators
optimization. (DNOs) deal with the variability of generation from renewable
DG technologies such as wind and its lack of correlation
State of charge of storage with demand. The optimization of ANM schemes containing
managed flexible demand. inter-temporal technologies and nonfirm DG is an open
Minimum and maximum levels problem, and DOPF can provide an important tool for analysis
of state of charge. of these problems.
Efficiency/coefficient of This paper lays out a full framework for the development
performance for storage managed of a general DOPF for use with ANM schemes. We formu-
flexible demand. late AC DOPF as a nonlinear (NL) programming problem in-
cluding generation curtailment with principles of access (PoA),
as well as ESS and flexible demand in an ANM context. The
II. INTRODUCTION key contributions of this paper are as follows. First, the full
AC DOPF for ANM schemes is the first application DOPF in a

T HE operation of power systems has for a long time been

informed by optimal power flow (OPF) [1]. The process
is used to dispatch available generation plant in a way that min-
way that fully models existing and developing ANM schemes.
Second, we present a method of encoding PoA for nonfirm gen-
erators within a DOPF. Third, we make the first presentation of
imizes a particular objective function. OPF can fully represent a DOPF which fully models both efficiency and the flexibility of
the network equations and nodal power balance. It also main- an ESS. Fourth, we present a model of flexible demand linked
tains limits on bus voltage, branch power flows, and generator to heat-storage. Finally, we discuss the optimization of ANM
outputs. Standard OPF formulations include minimizing oper- schemes over a one-day timescale.
ating costs and minimizing network losses. Other objectives
have been developed to include minimizing emissions [2] and
III. ACTIVE NETWORK MANAGEMENT
maximizing renewable penetration in distribution network [3],
[4]. The algorithms used to solve OPF problems include both Over the past decade, the ANM concept has been deployed in
mathematical programming techniques and heuristic optimiza- a number of R&D and pilot projects [21]. Current projects and
tion [5], [6]. future plans are moving beyond the simple model of real-time
OPF is used to optimize for a particular point in time; to ex- monitoring and control to day-ahead scheduling of technologies
tend the problem in the time domain requires an extension of such as ESS and flexible demand [20].
the basic OPF format. Intertemporal or dynamic optimal power The objective of ANM to date has been to maximize the
flow (DOPF) was initially developed as an optimization in hy- ability of distribution networks to use renewable DG. They at-
drothermal power system problems, where stored energy can tempt to increase the efficiency with which existing assets are
be dispatched at different times, but operation at one time af- used. Here, we describe the key technologies relevant to ANM
fects the ability to operate at other times [7], [8]. DOPF solves with an emphasis on inter-temporal effects.
the general problem of how to optimally dispatch generation
across a network and across a time-horizon to meet demand A. Generation Curtailment
within that time-horizon. It allows the modeling of ‘intertem- Under “fit-and-forget,” the maximum DG capacity is limited
poral’ technologies and effects including energy storage, flex- by power flow constraints, either thermal line limits or bus-
ible demand, and generator ramp rates. A number of investi- voltage limits. The capacity limit is the maximum local genera-
gations using DOPF are presented in the literature. This in- tion that the network can accommodate whilst respecting power
cludes interruptible demand investigated as part of an electricity flow constraints during all nonfault operating scenarios. This
market [9], the combination of markets for power and ancillary “firm generation” can be allowed to operate at all times at any
services [10], active and reactive power dispatch from energy level of output.
Authorized licensed use limited to: Politecnico di Bari. Downloaded on February 14,2024 at 17:35:34 UTC from IEEE Xplore. Restrictions apply.
GILL et al.: DYNAMIC OPTIMAL POWER FLOW FOR ACTIVE DISTRIBUTION NETWORKS 123

An ANM scheme provides the necessary communications Direct load control mechanisms have been discussed in a
and control infrastructure to allow additional “nonfirm” connec- number of papers (of which [26] gives an overview) and in-
tions for DG under the requirement that the output of these new volves dictating scheduling of power delivery to an electricity
generators will be curtailed under instruction from the DNO. load. In this paper, a directly managed flexible demand (DMFD)
Curtailment will be applied to avoid breaching power flow con- models this form of demand flexibility by allowing the ANM
straints and is likely when both demand is low and firm genera- scheme to take direct control of a particular demand. For ex-
tion is high. Under these circumstances, network branches will ample, it may be a commercial process where the timing of when
be close to full capacity, and there is no network capacity for the process occurs is not critical.
additional generation [16].
ANM schemes allow the network operators to monitor and E. Voltage Control Techniques
provide control signals to DGs in real time. In the U.K., the
There can be significant variability in voltage levels across
Orkney ANM scheme monitors and controls real-time gener-
a distribution network and problems can be accentuated by the
ation curtailment to ensure thermal and voltage limits are main-
addition of DG. ANM communications can be used to coordi-
tained across the network and on undersea cables linking the
nate the use of on load-tap changing transformers and combine
distribution and transmission networks [17], [18].
this with reactive power dispatch from DG and other devices.
B. Nonfirm PoA In general, these techniques are “time-independent.”

When multiple nonfirm generators connect, the remaining F. Intertemporal ANM Management and Objectives
network capacity is shared out according to a particular PoA
[22]. These are sets of rules defining which generator can ac- The aim of ANM schemes are closely linked to the growth in
cess the limited network capacity. An example is a priority PoA, renewable DG. ANM objectives are based around maximizing
where each generator has a fixed priority number and network the use of distributed renewable generation. With intertemporal
access is always given to high priority generators first. This prin- devices, DOPF allows objectives to be maximized across an op-
ciple of access has been deployed as “Last-in-first-out: (LIFO) timization time-horizon rather than at each time-step individu-
with priority decided by the order of connection. ally. This can lead to a reduction in DG output during some
Other examples of PoA include “shared percentage” where time-steps if this maximizes the objective overall.
all generators are curtailed by the same fraction of their output
and “technical-best” where the generators which allow the net- IV. GENERAL DOPF FORMULATION
work to best meet its objectives are used.
The basis of a DOPF is the standard OPF formulation for
C. Energy Storage Systems the network. The generic formulation of an OPF and DOPF are
shown in Table I. In OPF, a set of control variables are adjusted
ESSs provide a way of time-shifting energy from times suit- by the optimization process, fixed parameters define limits and
able for generation to times where it is most useful. In an ANM parts of the system that cannot be adjusted, and derived vari-
scheme ESSs can be dispatched to reduce curtailment or manage ables are those which are functions of control variable and fixed
network congestion. A number of technologies are available parameters.
with the ability to time-shift multi-MWh quantities of energy DOPF breaks the time-horizon up into time-steps and ex-
over periods of hours or longer [23]. These include chemical tends all OPF variables and some parameters into time series.
batteries, pumped storage hydro schemes, and compressed air. The network must obey the typical OPF constraints such as
ESS units such as chemical batteries operate with a power flow equations during each time-step independently. Fur-
power-electronic interface linking the device to the elec- thermore, additional inter-temporal variables and constraints are
trical grid. These interfaces have been shown to be able to created. An example of an inter-temporal variable is the state
generate or consume reactive power whilst continuing to of charge (SOC) of an ESS, that is, the fraction of total energy
charge or discharge [24]. So-called “four-quadrant operation” capacity currently used. SOC depends on charge and discharge
requires a constraint on apparent power and providing reactive values in multiple time-steps. The DOPF objective is a function
power support will reduce the capacity for real power charging across all time-steps.
and discharging.

D. Flexible Demand V. DOPF FOR ANM

Demand flexibility can be achieved in several ways, through The development of a DOPF structure for ANM problems is
response to price signals or by managing the delivery of energy. based on a formulation of OPF suitable for use with distribu-
This managed flexible demand model is being implemented in tion networks, and concentrates on the combination of firm and
ANM schemes, for example on the Shetland Islands, U.K. [20], nonfirm DG, energy storage, and managed flexible demand. The
[25], where domestic heating and industrial-scale electrical inclusion of other ANM technologies, here called time-indepen-
heating are being combined with heat storage. The amount of dent, are discussed in Section VII. The variability of voltage
energy to be delivered to a particular load is fixed across a across a DN caused by the high ratio and mainly radial
day, but there is flexibility in when it is delivered. In this case, layout means that a full AC OPF formulation is required.
it is the storage of heat that provides flexibility and will be The following scenario provides the structure for the scenario
described here as storage managed flexible demand (SMFD). modelled within the DOPF.
Authorized licensed use limited to: Politecnico di Bari. Downloaded on February 14,2024 at 17:35:34 UTC from IEEE Xplore. Restrictions apply.
124 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 29, NO. 1, JANUARY 2014

TABLE I
GENERAL FORMULATION FOR OPF AND DOPF

• A distribution network containing both firm and nonfirm This objective uses an exogenous electricity price and is of
generation. The network can have any number of grid con- interest where the load and DG capacity on the distribution net-
nection points (gcps) linking it to the transmission network. work is small compared with the overall electricity market. The
• Generic ESSs connected to specified busses. objective is of particular interest where the DN or part of a DN
• Managed flexible demand connected to specified busses. is operating as an integrated entity such as a micro-grid.
• Objectives related to the utilization of renewable genera-
tion are modeled. B. Standard OPF Formulation
As the DOPF is a network-focused optimization, sign-con- Distribution network problems require a full AC-formulation
ventions reflect this: ESS power injections are positive for dis- of the power flow equations, and as such OPF problems must be
charging as they are modeled as generators. solved using nonlinear programming methods such as gradient
search or interior point methods. The formulation of the static
A. Objectives OPF applied at each time-step as part of the DOPF is as follows.
The objectives of the ANM scheme need to be converted into • The power balance equations
suitable mathematical formulations. Two are described here and
further discussed in Section VI. (3)
1) Minimize Distribution Network Imports/Maximize
Exports: include constraints on nodal power balance and the power
flow equations.
• Voltage levels at each bus

(4)
(1)
The optimization sums the imports to the network across each where it is assumed that the maximum and minimum
gcp and all time-steps in the optimization horizon. This objec- voltage limits remain fixed across the optimization
tive aims to maximize the utilization of DG (rather than simply horizon.
its generation) either through the ability of DG to meet local de- • Thermal line limits constrain the apparent power flow
mand or to export energy from the distribution network. Simply along each line:
maximizing distributed generation raises the possibility that the
optimal solution increases losses as a way of increasing gener-
ation, particularly when ESS losses are included. (5)
2) Minimize Cost of Import/Maximize Revenue From
Export: • Each gcp is modeled as a generator with the ability to
supply positive and negative values of real and reactive
power:

(2) (6)
Authorized licensed use limited to: Politecnico di Bari. Downloaded on February 14,2024 at 17:35:34 UTC from IEEE Xplore. Restrictions apply.
GILL et al.: DYNAMIC OPTIMAL POWER FLOW FOR ACTIVE DISTRIBUTION NETWORKS 125

Busses connected to a gcp operate as PV buses (where real 2) Shared Percentage: A shared percentage scheme is one
power and voltage are fixed parameters) and the primary in which all generators receive the same percentage net overall
gcp bus will act as the reference bus (where the voltage curtailment across the time-horizon of the optimization. This
angle reference is defined). can be modeled through additional constraints on the nonfirm
generation control variables given as
C. Firm Distributed Generation
Firm generation is “must-take” and is treated as negative (9)
fixed demand; the available generation from a firm generator is
subtracted from the fixed bus demand. It is assumed that firm
generators operate at fixed power factor and reactive power is where and refer to different nonfirm generators. Here the
also subtracted from bus demand. fractional energy curtailed at each wind farm must be the same
across the optimization-horizon. Shared percentage provides
D. Nonfirm Distributed Generation flexibility regarding which generators can be curtailed when.
It will be useful when carrying out planning studies over
Nonfirm generation has a maximum output within each time- time-horizons of months rather than for day-ahead scheduling.
step, the value of for each time-step is defined by the available 3) Technical Best: Technical best PoA means dispatching the
wind resource generator that best helps the network meet its objective. In the
DOPF formulation, this will be achieved by using the original
(7)
objective and allowing the DOPF to choose which generator to
The power of nonfirm generators is a control variable and can curtail.
be scheduled anywhere within its range.
F. Energy Storage Systems
E. PoA for Nonfirm Generation The role of power lines in an OPF is to link network nodes
Modeling the PoA for nonfirm generators in a DOPF can be which are spatially separated. In DOPF, ESS can be thought of
achieved by modifying the objective function or adding addi- as carrying out a similar role linking nodes that are separated in
tional constraints as set out below. time rather than space. Both power lines and ESSs act as vectors
1) Priority Order (e.g., LIFO): For the DOPF to apply a pri- transferring energy around the network.
ority order, there must be a distinction in the value of generation ESS systems can be modeled as generators with the ability to
from different generators. This is accomplished by modifying inject positive or negative power onto the network and a SOC
the objective function. variable to keep track of the stored energy. The relationship be-
For the objective of minimize import, (1) can be modified to tween the power injections in each period and the SOC of the
include a second term defining the priority order to yield store includes a discontinuity as follows:
if
if (10)

(8) is positive when discharging and negative when charging

and refers to the flow of power as seen by the grid. The efficiency
factors relate this value to the power injected or removed from
where is an integer between 1 and with high numbers rep- the ESS including conversion losses. The round-trip efficiency
resenting low priority generators. The definition of the constant of the store is the product of and . The discontinuity of
is discussed below. (10) cannot be directly accommodated in a nonlinear program-
The original objective has now been adjusted so that genera- ming solution.
tion from high priority generators have more value than for low This discontinuity can be removed by modeling the ESS with
priority generators in minimizing the objective, but this ‘pri- separate generators for charging and discharging so that
ority order’ value should be small so that the first term in (8)
dominates. This ensures that the optimization continues to min- (11)
imize imports whilst the small adjustment of the second term
provides the priority and chooses which DG to dispatch. This ef- with the charging generator operating with negative values.
fect is achieved by choosing so that and Fig. 1 shows this conception of the ESS model. The real-power
for the lowest priority generator (highest ). It is also impor- generation of the two generators is constrained as follows:
tant that is sufficiently large that the priority term is larger
than the tolerance value used by the DOPF algorithm to define
convergence. (12)
Applying priority orders to objective 2, defined in (2), can
be achieved in a similar way, and the second term from (8) can
be added to (2) to create the required objective. In this case, the where is the power of the “charging generator” and is
requirement on is that for the lowest priority negative relative to the grid, and is the discharging
generator. generator.
Authorized licensed use limited to: Politecnico di Bari. Downloaded on February 14,2024 at 17:35:34 UTC from IEEE Xplore. Restrictions apply.
126 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 29, NO. 1, JANUARY 2014

of for the optimal . If this

minimum occurs when for charging periods
and for discharging periods. Thus, for a positive
cost of generation and an ESS efficiency less than 1, an optimal
solution will only operate at most one of the two charging/dis-
charging generators during each period.
In addition to the real power constraints described in (12), the
ESS unit is able to provide reactive power support. The combi-
nation of real and reactive power from an ESS is limited by the
Fig. 1. Two-generator model of an ESS. apparent power according to

(17)
The SOC is now related to the power flows of two generators where
and for each period is defined by
(18)

G. Managed Flexible Demand

(13) A flexible demand unit is modeled as a generator with neg-
ative output where the output magnitude during each time-step
is a control variable. The two models proposed in Section II are
where is the initial SOC. The SOC is constrained to
now described.
remain within limits as follows:
1) Directly Managed Flexible Demand (DMFD): The total
(14) energy delivered to the load across the time-horizon must be
equal to the energy requirement. This is simply
These limits may be 0 and 1; however, it is likely that
some operational strategies will want to avoid charging and (19)
discharging an ESS to its theoretical limits. In general, it is
expected that the initial and final SOC are the same: For each time-step, the power delivered is bound by the rated
capacity of the load and a minimum power delivery for that
(15) time-step, the value of may be 0 for all or may be
higher to allow for some consumer preference regarding min-
This formulation provides a significant improvement over
imum power delivery at predefined times:
existing techniques. For example, in [11], charging and dis-
charging time-steps are predefined as inputs to the formulations; (20)
in [12], the formulation does not predefine charging discharging
time-steps but at the expense of a full formation of efficiency. 2) Storage Managed Flexible Demand (SMFD): If a form of
Note that in this formulation there is no constraint to stop storage such as a heat store is used to provide flexibility, (19) and
charging and discharging generators operating during the same (20) should be augmented with equations to define and manage
time-step. It is therefore a mathematically feasible solution to the SOC on that heat store. It is assumed that a fixed demand
charge and discharge during the same time-step although obvi- schedule for energy delivery (e.g., ) is buffered by the
ously not physically realizable in an ESS. Such a solution, whilst heat storage:
feasible in this framework can be seen to be nonoptimal if: 1)
the round-trip efficiency of the ESS is less than 1 and 2) the
“cost” of all generation in the objective is positive. The situa-
tion is therefore avoided in the optimal solution.
To demonstrate this, consider the physical system. During a (21)
particular time-step there is an optimal change in SOC
relating to a charge or discharge of the ESS. In the mathematical
Note that, as is not related to the electrical network, it is
formulation, this can be achieved by a combination of charging
defined as positive for demand; is the efficiency of conversion
and discharging that obeys
between electrical and stored energy (known as the Coefficient
of Performance), and for heat-pump technologies this may be
(16)
greater than 1 [27]. is modeled as a generator with
negative power flows. SOC constraints are the same as those
where represents power that must be generated else- for ESS:
where (either by DG or imported via a gcp). If the cost of
generating this power is positive in the objective function, (22)
then minimizing the objective includes minimizing the value (23)
Authorized licensed use limited to: Politecnico di Bari. Downloaded on February 14,2024 at 17:35:34 UTC from IEEE Xplore. Restrictions apply.
GILL et al.: DYNAMIC OPTIMAL POWER FLOW FOR ACTIVE DISTRIBUTION NETWORKS 127

TABLE II
PARAMETERS OF DISTRIBUTED GENERATION AND ANM COMPONENTS

Fig. 2. Case study distribution network. All generators represent wind farms
and additional ANM devices are shown connected.

and the rate of charge is constrained by

(24)

VI. CASE STUDY

Here, we present a case study of a distribution network op-

erating an ANM scheme with inter-temporal components. The
DOPF formulated using (1)–(24) is illustrated on a radial net-
work with some meshing typical of a rural distribution network.
It is based on the U.K. Generic Distribution System simplified
Fig. 3. Normalized input time-series. All time-series have been normalized
rural Extra High Voltage network, and an outline of the network against their maximum value.
is shown in Fig. 2.1
DG and ANM components are added to the network as shown TABLE III
in Fig. 2, while the parameters used are listed in Table II, and CURTAILMENT AT NF WIND FARMS IN EACH OF THE SCENARIOS MODELLED
time-series for demand and generation are shown in Fig. 3. For
simplicity, all DG is modeled as wind generation with the same
wind profile. The optimization is for a 24-h horizon split up
into 15-min time-steps. Fixed demand at each bus and available
generation at each wind farm is the product of total capacity
and the relevant normalized time series. As DG penetration is
high, the objectives from Section IV part A are referred to as
“maximizing export” and “maximize revenue.”
The following scenarios are used to illustrate the effective-
ness of the DOPF model and summarized in Table II.
1) Firm and nonfirm wind only, maximize export objective,
priority-order PoA with a value of in (8). No
other ANM devices are added.
2) As for scenario .
3) As for scenario and SMFD. Flexible demand Algorithm (MIPS) is used as the solver. On a quad-core 3-GHz
replaces fixed demand so that total energy demand is the desktop the optimization solves in approximately 3.5 s.
same as for scenarios 1 and 2.
4) As for scenario 3 using the maximize revenue objective. B. Results
Results for curtailment, exported energy, and losses are
shown in Table III which also shows revenue from export in
A. Implementation scenarios 3 and 4.
The results show that, when the system is operated as a cur-
The DOPF has been implemented in conjunction with the
tailment scheme only (i.e., without ESS or flexible demand),
MATPOWER suite for power system analysis [28]. The exten-
18% of the available non-firm generation is curtailed across
sible architecture of the software allows for easy customization
the whole day. This curtailment is distributed in line with the
of standard OPF problems. MATPOWER’s own Interior Point
priority order, with the highest priority wind farm receiving
1[Online]. Available: http://www.sedg.ac.uk/ no curtailment and the lowest priority generator receiving the
Authorized licensed use limited to: Politecnico di Bari. Downloaded on February 14,2024 at 17:35:34 UTC from IEEE Xplore. Restrictions apply.
128 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 29, NO. 1, JANUARY 2014

Fig. 6. SOC of the ESS in scenario 2.

Fig. 4. Curtailment of WF4 and 5 in scenario 1.

Fig. 7. Scheduling of SMFD at bus 8 compared with curtailment at WF4 with

and without flexible demand.

(between busses 1 and 2) rather than due to its local conges-

tion, the ESS device is able to reduce the curtailment of WF4 as
well. In addition, the ESS is able to use its reactive capabilities
to manage the reactive power and voltage levels. For example,
during time-steps 52–54, Fig. 5. shows that the increase in wind
Fig. 5. ESS schedule for scenario 2. generation is greater than the power demand by the ESS. Due
to the high penetration of DG at bus 16, the maximum voltage
level of 1.06 p.u. is a constraining factor, leading to curtailment
most, at 55% of output. The time-series of curtailment for WF4 of WF5. By regulating reactive power, the ESS is able to allow
and WF5 is shown in Fig. 4. WF5 is fully curtailed during 16 additional generation at that bus.
time-steps and at these times WF4 is partially curtailed. There The replacement of fixed demand with flexible demand in
are 25 time-steps (out of 96) when WF4 is not curtailed and scenario 3 leads to a further reduction in curtailment and a rise
WF5 is partially curtailed. There are also 24 time-steps where in export. The flexible demand is scheduled by DOPF for time-
both WF4 and WF5 are partially curtailed. At these times, local steps where it reduces curtailment. Fig. 7 shows the schedule
thermal constraints on the line between bus 4 and bus 6 limit the for the SMFD at bus 7 and the curtailment experience by WF4
output from WF4 without affecting that from WF5. with and without flexible demand: the demand has been moved
In scenario 2, the addition of ESS increases the energy ex- to coincide with periods of curtailment and leads to a reduction
ported by 12 MWh (1.3%) and reduces curtailment by 16 MWh in that curtailment. The DMFD at bus 2 is not local to either
(1.5%). The operating schedule for the ESS device is shown of the wind farms but by managing congestion on the export
in Fig. 5. and the SOC schedule in Fig. 6. The curtailment is constraint it further reduces curtailment at both WF4 and WF5.
decreased by more than the export rises as overall losses rise The network setup for scenario 4 is the same as that for sce-
by 4 MWh. Whilst the export increase is relatively small com- nario 3, with the objective changed to reflect the external market
pared with total export, a more useful comparison is with the price for energy. The change in objective leads to a number of
maximum energy that such an ESS can time-shift in a 24-h pe- changes in the optimal solution. First, the total energy export is
riod. A 3-MW, 18-MWh unit can in total time-shift 36 MWh in very slightly smaller, but there is a change in the timing of export
a day, so a value of 12 MWh equates to 33% of this value. It to coincide with periods of high market price. When compared
is seen that the SOC reaches its minimum level, therefore max- with the revenue that would be raised by the results for the pre-
imizing the utilization within a single charge-cycle. Given the vious scenario there is a rise of 1.3% in revenue. Fig. 8 shows
distribution of curtailment throughout the day, greater utiliza- the change in export between the two scenarios and the market
tion could be achieved if the initial and final SOC were greater price. When the export is not constrained the “maximize rev-
than 0.5. The ESS device is located at the same bus as WF5 and enue” objective leads to reduced export when price is low and
is able to directly manage congestion on the line linking it to increased export when price is high. Time-steps with no change
the rest of the network. However, during periods where WF4 between the scenarios are those where the export link is at full
is constrained due to congestion in exporting from the network capacity.
Authorized licensed use limited to: Politecnico di Bari. Downloaded on February 14,2024 at 17:35:34 UTC from IEEE Xplore. Restrictions apply.
GILL et al.: DYNAMIC OPTIMAL POWER FLOW FOR ACTIVE DISTRIBUTION NETWORKS 129

example is the load recovery period where demand spikes

above the average at the end of a period where the delivery
of energy to a load has been restricted [29]. DOPF is able
to model these type of effects, for example, by defining the
energy to be delivered during each three hour block of the day
or limiting load controllability to particular parts of the day.
The results presented here show that within the DOPF model
simple optimal load-scheduling can be combined with ESS
Fig. 8. Change in export between scenario 3 (maximize export) and scenario and curtailment schemes in a format that is flexible enough to
4 (maximize revenue), positive value represent greater export under maximize consider more complex schemes in future work.
revenue. The usefulness of this model in the operational context is that
it can be run using forecasts of generation and demand to pro-
vide the optimal solution for the forecast conditions; creating
TABLE IV
EXPORT INCREASE RELATIVE TO SCENARIO 1 FOR effective schedules based on day-ahead forecasts is a require-
VARIATIONS IN SIZE AND LOCATION OF ESS ment of the Shetland ANM [20]. Retrospectively it can be used
with historic measurements of demand and generation to bench-
mark the success of the actual schedules used. The short time
required to solve the DOPF (3.5 s) also allows the method to
be combined with probabilistic techniques such as Monte Carlo
simulation. For example,the case study used can be solved for
approximately 1200 demand and generation cases within 1 h on
a standard desktop.
The methodology presented here can be applied to larger and
more complex networks, with operational speed being the lim-
VII. DISCUSSION iting factor. When a larger network consisting of 68 busses and
The case study illustrates that the DOPF structure presented representing an existing distribution network, the solution time
can be successful in producing optimal dispatches for ESS and is approximately 7 s.
flexible demand in an ANM context. It is able to use the flexi- Whilst this case study concentrated on the inter-temporal as-
bility to raise export by 2.9% compared with simple curtailment. pects, other ANM technologies that are time-independent can
The case study presented models ANM examples which are easily be incorporated. There are OPF formulations in the lit-
now being rolled out to real-world distribution networks. This erature to deal with coordinated voltage control, power-factor
includes the extension of the Orkney smart grid to include en- control from DG, and ANM power flow management [30]–[32].
ergy storage [18], and the use of SMFD on the Shetland Dis- Including these in the DOPF formulation simply involves ex-
tribution network [20]. As such schemes develop and the pene- tending the variables and constraints used in the original work to
tration of these enabling technologies increases, so too will the time-series variables and applying them independently at each
ability to manage curtailment. DOPF provides a tool for investi- time-step.
gating the effect of the sizes and location of ESS units. Table IV There are a range of objectives that the DOPF formulation
gives examples of the total export increase (compared with sce- can be used for. The two illustrated here concentrate on the
nario 1) if the location and size of the ESS in scenario 2 is varied. effect of DG and ANM on the overall aim rather than on the
For a similar sized ESS unit, locating it at bus 2 rather than bus DG itself: maximizing exported energy or revenue is the ob-
16 leads to greater increases in exported energy as it is able to jective rather than simply maximizing the output by DG. It is
more effectively utilize curtailment from more NF wind farms. likely that a feasible dispatch of ESS and FD exists with curtail-
But in both locations, the continued increase in ESS capacity ment lower than the 189MWh achieved in Scenarios 3 and 4 but
leads to diminishing returns in terms of increased export. This with higher losses; it is important that the use of DG is consid-
is because on the margin there is less curtailment to remove and ered in the objective. The importance of losses in systems with
the network capacity is more fully utilized. The optimal levels of high penetration of renewable generation needs careful consid-
ESS at each bus will be a function of economic factors as well as eration when developing objectives for DOPF. OPF analysis of
technical factors. A key area of future research utilizing DOPF networks with mainly conventional generation often uses the
based tools will be to investigate the trade-offs between size, objective of minimizing losses. When renewable curtailment is
costs, location and the ownership models used for distributed involved, both losses (network or ESS) and curtailment are sim-
ESS. ilar in that they represent the non-use of renewable energy. It is
Similar analysis can be carried out with flexible demand, important that the objective function represents the real aims of
however these technologies raise a number of other issues re- the ANM scheme.
lated to the behavior of consumers. The models presented here The use of priority-order PoA is one that has so far been fa-
assume that consumer demand can be predicted accurately in vored due to its simplicity and transparency to wind farm devel-
advance, and that those consumers are willing to give up their opers. But others may be more effective at encouraging the de-
ability to control the load. The likelihood of these assumptions velopment of DG within ANM schemes. In scenario 3, the level
being true depends on the type of load being controlled. One of curtailment varies significantly across the non-firm portfolio
Authorized licensed use limited to: Politecnico di Bari. Downloaded on February 14,2024 at 17:35:34 UTC from IEEE Xplore. Restrictions apply.
130 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 29, NO. 1, JANUARY 2014

with WF3 receiving zero curtailment and WF5 almost 50%. It [11] A. Gabash and P. Li, “2012, active-reactive optimal power flow in
is very likely that if this is indicative of a normal day that WF5 distribution networks with embedded generation and battery storage,”
IEEE Trans. Power Syst, vol. 27, no. 4, pp. 2026–2035, Nov. 2012.
will be uneconomical. That value of 50% should be contrasted [12] A. J. Lamadrid, T. D. Mount, and R. J. Thomas, “Scheduling of energy
with that of 15% curtailment across the entire non-firm port- storage systems with geographically distributed renewables,” in Proc.
folio. In the U.K., the average capacity factor of wind farms was 9th Int. Symp. Parallel Distrib. Process. Applicat. Workshops, 2011,
27% during 2011 [33]. Using this as an estimate for “viable” pp. 85–90.
[13] V. Stanojevic, V. Silva, D. Pudjianto, G. Strbac, P. Lang, and D.
development, a wind farm that is curtailed for 50% of the time McLehman, “Application of storage and demand side management to
would require an available capacity factor of 54%. If curtail- support the integration of intermittent distributed generation,” in Proc.
ment were 15% the uncurtailed wind farm capacity factor would CIRED, Prague, Czech Republic, 2009, pp. 1–4.
[14] DTI A Technical Review and Assessment of Active Network Manage-
need to be 32% for it to be viable after curtailment. Finding sites ment Infrastructures and Practices. London, U.K.: Dept. Trade and
with the wind resource required for a wind farm with 32% ca- Industry, 2006.
pacity factor is feasible, especially in areas of high resource, [15] S. K. Merz, “Active network management (ANM) technology. Current
such as those where current ANM development is happening. issues and identification of technical opportunities for active network
management,” Dept. Business and Regulatory Reform, London, U.K.,
The DOPF structure can be used to inform PoA development. 2008.
[16] N. Jenkins, R. Allan, P. Crossley, D. Kirschen, and G. Strbac, Em-
VIII. CONCLUSION bedded Generation. London, U.K.: Inst. Electr. Eng., 2000.
[17] “Facilitate generation connections on Orkney by automatic distribution
This paper presents a general DOPF framework for use with network management,” SSE, Dept. Trade and Industry, 2004.
ANM schemes. ANM technologies are presented with a focus [18] “Orkney Smart Grid,” SSE, 2012. [Online]. Available: http://www.
ssepd.co.uk/OrkneySmartGrid/
on inter-temporal effects. The DOPF is developed specifically [19] A. Nourai, “Installation of the first distributed energy storage system
to model curtailment of renewable distributed generation, en- (DESS) at American Electric Power (AEP),” Dept. Energy, Albu-
ergy storage systems and flexible demand. A case-study net- querque, NM, USA, 2007.
[20] M. J. Dolan, G. W. Ault, D. F. Frame, S. Gill, I. Kockar, O. Anaya-
work and ANM scheme is solved to illustrate the effectiveness Lara, S. Galloway, B. O’Neill, C. Foote, A. Svalovs, and K. Jackson,
of the framework in scheduling for realistic time-horizons such “Northern isles new energy solutions: Active network management sta-
as one day. Under an objective of maximizing total energy ex- bility limits,” in Proc. ISGT, Berlin, Germany, 2012, pp. 1–9.
ported the framework optimizes energy storage and flexible de- [21] R. MacDonald, R. Currie, G. Ault, and J. Douglas, “Register of active
management pilots, trials, research, development and demonstration
mand to raise export by 2.9%, whilst in the process reducing cur- activities,” [Online]. Available: http://www.ensg.gov.uk/index.php?ar-
tailment of non-firm generation by 14%. Finally, the extension ticle=952008.
of DOPF to include other ANM relevant technologies including [22] R. Currie, B. Neill, C. Foote, A. Gooding, R. Ferris, and J. Douglas,
“Commercial arrangements to facilitate active network management,”
existing OPF formulation for time-independent technologies is in Proc. CIRED, Frankfurt, Germany, 2011, pp. 1–4.
discussed. [23] M. Beaudin, H. Zareipour, A. Schellenberglabe, and W. Rosehart,
“Energy storage for mitigating the variability of renewable electricity
sources: An updated review,” Energy for Sustainable Development,
REFERENCES vol. 14, pp. 302–314, 2010.
[1] H. W. Dommel and W. F. Tinney, “Optimal power flow solutions,” [24] L. H. Walker, “10-MW GTO converter for battery peaking service,”
IEEE Trans. Power App. Syst., vol. PAS-87, no. 10, pp. 1866–1876, IEEE Trans. Ind. Applicat., vol. 26, no. 1, pp. 63–72, Jan./Feb. 1990.
Oct. 1968. [25] S. Gill, M. J. Dolan, D. Frame, and G. Ault, “The role of the elec-
[2] J. H. Talaq, F. El-Hawary, and M. E. El-Hawary, “Minimum emissions tric heating and district heating networks in the integration of wind en-
power flow,” IEEE Trans. Power Syst., vol. 9, no. 1, pp. 429–435, Feb. ergy to island networks,” Int. J. Distrib. Energy Resources, vol. 7, pp.
1994. 245–262, Jul. 2011.
[3] G. P. Harrison and A. R. Wallace, “Maximising distributed generation [26] D. S. Callaway and I. A. Hiskens, “Achieving controllability of electric
capacity in deregulated markets,” in Proc. IEEE PES Transm. Distrib. loads,” Proc. IEEE, vol. 99, no. 1, pp. 184–199, Jan. 2011.
Conf. Expo., 2003, vol. 2, pp. 527–530. [27] K. J. Chua, S. K. Chou, and W. M. Yang, “Advances in heat pump
[4] G. P. Harrison and A. R. Wallace, “Optimal power flow evaluation of systems: A review,” Appl. Energy, vol. 87, pp. 3611–3624, Dec. 2010.
distribution network capacity for the connection of distributed gener- [28] R. D. Zimmerman, C. E. Murill-Sanchez, and R. J. Thomas, “Mat-
ation,” Proc. Inst. Electr. Eng.—Gen., Transm. Distrib.-, vol. 152, pp. power: Steady-state operations, planning, and analysis tools for power
115–122, 2005. systems research and education,” IEEE Trans. Power Syst., vol. 26, no.
[5] F. Milano, Power System Modelling and Scripting, 1st ed. London, 1, pp. 12–19, Feb. 2011.
U.K.: Springer, 2010. [29] V. Silva, V. Stanojevic, M. Aunedi, D. Pudjianto, and G. Strbac, Smart
[6] K. Lee and M. El-Sharkawi, Modern Heuristic Optimization Tech- Domestic Appliances as Enabling Technology for Demand Side In-
niques Theory and Applications to Power Systems, 1st ed. Hokoken, tegration: Modeling, Value and Drivers, M. Pollit, Ed. Cambridge,
NJ, USA: Wiley, 2008. U.K.: Cambridge Univ., 2011.
[7] M. E. El-Hawary and D. H. Tsang, “The hydrothermal optimal load [30] S. N. Liew and G. Strbac, “Maximising penetration of wind genera-
flow, a practical formulation and solution techniques using newton’s tion in existing distribution networks,” Proc. Inst. Electr. Eng.—Gen.,
approach,” IEEE Trans. Power Syst., vol. PS-1, no. 3, pp. 157–166, Transm. Distrib., vol. 149, pp. 256–262, 2002.
Aug. 1986. [31] L. F. Ochoa, C. J. Dent, and G. P. Harrison, “Distribution network
[8] W. Hau, H. Sasaki, J. Kubokawa, and R. Yokoyama, “Large scale hy- capacity assessment: Variable DG and active networks,” IEEE Trans.
drothermal optimal power flow problems based on interior point non- Power Syst., vol. 25, no. 1, pp. 87–95, Feb. 2010.
linear programming,” IEEE Trans. Power Syst., vol. 15, no. 1, pp. [32] M. J. Dolan, E. M. Davidson, I. Kockar, G. W. Ault, and S. D. J.
396–403, Feb. 2000. McArthur, “Distribution power flow management utilizing an online
[9] W. Uturbey and A. S. Costa, “Interruptible load management assess- optimal power flow technique,” IEEE Trans. Power Syst., vol. 27, no.
ment via dynamic optimal power flow,” in Proc. IEEE Power Tech 2, pp. 790–799, May 2012.
Conf., Bologna, Italy, 2003, vol. 4, p. 7. [33] “Digest of UK Energy Statistics (DUKES): Load factors for re-
[10] A. L. Costa and A. S. Costa, “Energy and ancillary service dispatch newable electricity generation,” DECC, Sep. 2012. [Online]. Avail-
through dynamic optimal power flow,” Electr. Power Syst. Res., vol. able: http://www.decc.gov.uk/en/content/cms/statistics/energy_
77, pp. 1047–1055, 2007. stats/source/renewables/renewables.aspx

Authorized licensed use limited to: Politecnico di Bari. Downloaded on February 14,2024 at 17:35:34 UTC from IEEE Xplore. Restrictions apply.
GILL et al.: DYNAMIC OPTIMAL POWER FLOW FOR ACTIVE DISTRIBUTION NETWORKS 131

Simon Gill received the M.S. degree in astrophysics Graham W. Ault (M’98) received the B.Eng. degree
from the University of Edinburgh, Edinburgh, Scot- in electrical and mechanical engineering and Ph.D.
land, in 2003. He is currently working toward the degree in electrical power systems from the Univer-
Ph.D. degree at the Wind Energy Centre for Doctoral sity of Strathclyde, Glasgow, U.K., in 1993 and 2000,
Training, University of Strathclyde, Glasgow, U.K. respectively.
After completing his graduate work, he spent four Since 1996, has been researching power system
years teaching physics. His research interests include planning and operations issues relating to distributed
energy storage, active management of distribution energy resources in distribution systems. He is cur-
networks, and the integration of renewable energy rently a Professor with the Institute for Energy and
into power systems. Environment, University of Strathclyde, Glasgow,
U.K.

Ivana Kockar (S’99–M’04) received the Dipl.Ing.

(M.Sc.) degree from the University of Belgrade, Bel-
grade, Yugoslavia, and the M.Eng. and Ph.D. degrees
from McGill University, Montreal, QC, Canada, in
2004.
She is a Lecturer within the Institute for Energy
and Environment at the University of Strathclyde,
Glasgow, U.K. Her research interests include power
system operation planning and economics of energy
systems.
Dr. Kockar chairs the IEEE PES Computing and
Analytical Methods Subcommittee (CAMS) and co-chairs the IEEE PES Task
Force on Agent-Based Modeling of Smart Grid Market Operation.

Authorized licensed use limited to: Politecnico di Bari. Downloaded on February 14,2024 at 17:35:34 UTC from IEEE Xplore. Restrictions apply.