Aalborg Universitet

Optimal Power Flow in Microgrids with Energy Storage

Levron, Yoash; Guerrero, Josep M.; Beck, Yuval

Published in:
I E E E Transactions on Power Systems

DOI (link to publication from Publisher):

10.1109/TPWRS.2013.2245925

Publication date:
2013

Document Version
Early version, also known as pre-print

Link to publication from Aalborg University

Citation for published version (APA):

Levron, Y., Guerrero, J. M., & Beck, Y. (2013). Optimal Power Flow in Microgrids with Energy Storage. I E E E
Transactions on Power Systems, 28(3), 3226-3234 . DOI: 10.1109/TPWRS.2013.2245925

General rights
Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright owners
and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights.

? Users may download and print one copy of any publication from the public portal for the purpose of private study or research.
? You may not further distribute the material or use it for any profit-making activity or commercial gain
? You may freely distribute the URL identifying the publication in the public portal ?
Take down policy
If you believe that this document breaches copyright please contact us at [email protected] providing details, and we will remove access to
the work immediately and investigate your claim.

Downloaded from vbn.aau.dk on: september 04, 2018

 > REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 1
Levron, Y.; Guerrero, J. M.; Beck, Y.; , "Optimal Power Flow in Microgrids With Energy Storage," Power Systems, IEEE Transactions on , vol.PP, no.99, pp.1-9.
doi: 10.1109/TPWRS.2013.2245925
URL: http://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=6469193&isnumber=4374138

Optimal Power Flow in Microgrids with Energy

Storage
Yoash Levron, Josep M. Guerrero, Member, IEEE, Yuval Beck


Abstract— Energy storage may improve power management in
microgrids that include renewable energy sources. The storage
devices match energy generation to consumption, facilitating a
smooth and robust energy balance within the microgrid. This
paper addresses the optimal control of the microgrid's energy
storage devices. Stored energy is controlled to balance power
generation of renewable sources to optimize overall power
consumption at the microgrid point-of-common-coupling (PCC).
Recent works emphasize constraints imposed by the storage
device itself, such as limited capacity and internal losses.
However, these works assume flat, highly simplified network Fig. 1. A typical microgrid.
models, which overlook the physical connectivity. This work
proposes an optimal power flow solution that considers the entire
system: the storage device limits, voltages limits, currents limits
In islanded mode, the main goal of power management is to
and power limits. The power network may be arbitrarily stabilize the system, in terms of frequency and voltage. In
complex, and the proposed solver obtains a globally optimal grid-connected mode, typical objectives are to minimize the
solution. price of energy import at the PCC, to improve power factor at
the PCC, and to optimize the voltage profile within the
Index Terms— microgrid, smart grid, distributed generation microgrid [4], [5]. This work addresses grid-connected
(DG), optimal power flow (OPF), energy storage.
networks.
Energy management in microgrids is usually thought of as a
I. INTRODUCTION
three level hierarchical control system [6]. The first control
M ICROGRIDS have received increasing attention as a
means of integrating distributed generation into the
electricity grid [1], [2]. Usually described as confined clusters
level, often called ‘primary’ or ‘autonomous’ control, consists
of a number of local, autonomous controllers. Each controller
governs a power electronics converter, and is responsible to
of loads, storage devices and small generators, these
interface generators, storage devices and loads with the
autonomous networks connect as single entities to the public
microgrid [7] These controllers are the fastest, as they operate
distribution grid, through a point of common coupling (PCC).
in the millisecond range, employing a droop control in
Fig. 1 illustrates a typical microgrid network. Microgrids
comprise a variety of technologies: Renewable sources, such islanded mode [2]. A secondary control level employs a low
as photovoltaic and wind generators are operated alongside bandwidth communication to fix the frequency and amplitude
traditional high-inertia synchronous generators, batteries and of the microgrid's units, restoring their nominal values.
fuel-cells [3]. Thus, energy is generated near the loads, Finally, the tertiary control level [1] ,[6] is related to the
enabling the utilization of small-scale generators that increase control of active and reactive power flow. This level of control
reliability, and reduce losses over long power lines. is related to Energy Management Systems (EMS, see [2]) and
The locality of the microgrid network enables an improved to the optimization of the microgrid resources, and is the main
management of energy. Generators (and possibly loads) may subject of this present work.
be controlled by a local energy management system (EMS) to The Tertiary control level coordinates power flow within
optimize power flow within the network. The objectives of the microgrid, and therefore often utilizes an optimal power
energy management depend on the mode of operation: flow (OPF) solver. Such solvers have been extensively studied
Islanded, or grid-connected. by many. Surveys may be found at [8], [9]. However, classical
power flow solutions are not tailored for microgrid analysis,
particularly due to the lacking representation of distributed
Manuscript submitted September 2012. energy sources, storage devices, and pricing methods.
- Y. Levron is with the department of Electrical Engineering, Technion Recently, several studies have shown optimal power flow
institute of technology, Israel. E-mail: [email protected].
- d. J. M. Guerrero is with the Institute of Energy Technology, Aalborg
models that highlight the unique aspects of microgrids. These
University, 9220 Aalborg East, Denmark. E-mail: [email protected] studies can be categorized by focus. A first group of studies
- Y Beck is with the Electrical engineering faculty at Holon Institute of concern the allocation and optimal power sharing of
Technology, 52 Golumb st. Holon, Israel. E-mail: [email protected]
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 2

distributed generators, most often solar or wind [10] ,[11]. A TABLE I. Units of the microgrid
second group highlights the economic revenue. Their Unit Symbol Constraints
model
objective is usually to minimize the overall price of energy, or refs
to maximize the profit from energy generation [6] ,[12]. power line Constraints:
bus i bus j Ii,j,x(t) < Ii,j,max
A third group examines the optimal dispatch of energy load Constraints:
storage devices [13]-[16]. Energy may be stored when Pi,x(t) = -PL,ix(t) (fixed)
renewable power is available, or when energy import is Qi,x(t) = -QL,ix(t) (fixed)
Vix,min ≤ Vix(t) ≤ Vix,max
inexpensive. This stored energy may be consumed later when
Free variables: Vix(t), δix(t)
demand is high, or when renewable power is unavailable. The Renewable Constraints:
objective here is to optimize price, efficiency, and stability, generator Pix(t) = +Pg,ix(t) (fixed)
considering the constraints imposed by the storage devices, G Qix(t) = +Qg,ix(t) (fixed)
Vix,min ≤ Vi(t) ≤ Vix,max
such as limited capacity and internal losses. In Free variables: Vix(t), δix(t)
studies [13], [14] the storage device operates as a mediator of storage Eix(t) = stored energy or state [13]-
power generation. Overall power generation is optimized to be device of charge (SOC). [15]
Typical constraints:
as constant as possible, reducing fuel costs, while taking into Vix(t) = VS,i (fixed)
account the limited storage capacity. Study [15] employs 0 ≤ Ei(t) ≤ Ei,max
storage to time-shift the generation of renewables, matching -Pi,rated ≤ Pi(t) ≤ +Pi,rated
State equation (one-phase):
generation to consumption. Study [16] addresses a wind farm, ݀
compensated by a battery energy storage. Their goal is to ‫ ܧ‬ൌ ݂௜ ሺܲ௜ ǡ ‫ܧ‬௜ ሻ
݀‫ ݐ‬௜
control the storage device for improving the predictability of Free variables:
power generation. Pi(t), Qi(t), δi(t), Ei(t)
point of The PCC is always indexed as [8],
All the above studies assume trivial network topologies. common bus 1, i=1. [10]
None of them inspect storage devices integrated in a general coupling PCC Constraints:
power network. An optimal solution to a generally meshed (PCC) δ1,x(t) = 0
V1,x(t) = Vin,x (fixed)
network with storage devices has not been shown. The reason utility
Q1x,min ≤ Q1,x(t) ≤ Q1x,max
grid
for this is the tremendous numerical complexity of the P1x,min ≤P1,x(t) ≤ P1x,max
problem, which includes both the network domain, and the Free variables: P1(t) , Q1(t)
time domain, related with storage. Traditional gradient based
solvers (such as Newton-Raphson), while extremely useful in The PCC corresponds to the ‘slack’ bus. It is always
the network domain, are inadequate in time domain, and indexed as bus 1 (i=1), and is described as a V-δ bus, with
cannot be applied to the combined network-storage problem V1(t)=Vin(t), a uncontrollable voltage signal. Loads and
(See section III). renewable generators are uncontrollable and are therefore
To cover this gap, this work introduces a new solution represented by fixed power signals (power vs. time). The
method to this problem: an optimal power flow (OPF) solver network power-flow expressions are given in (1), for a
that integrates storage devices. The suggested method balanced three-phase system. These may be found in many
computes the globally optimal power flow, in both network classical textbooks, such as [8].
N
domain and time domain. It considers both the limitations of
the storage device, and the limitations of the network
Pi Vi ˜ ¦ Yij ˜ V j ˜ cos G i  G j  Tij 
j 1
regarding voltages, currents and powers. The method
Vi ˜ ¦ Yij ˜ V j ˜ sin  G i  G j  T ij 
N

combines a power flow solver with a dynamic programming Qi (1)

j 1
recursive search, achieving a numerically efficient solution.
iG i iG j
I ij Vi ˜ e  Vj ˜ e
II. NETWORK TOPOLOGY AND POWER FLOW EQUATIONS Yij and θij are the admittances’ magnitude and phase. Yii are the
This work utilizes the usual terminology of a power-flow self-admittances, and Yij (i≠j) are the cross admittances. Iij are
analysis [8]. Buses are denoted with the running index i, the line currents (magnitude).
where i=1..N. N is the number of buses. Each bus is described The system may also be an unbalanced three-phase system.
by four independent signals: The power flow equations for this case are too complex to be
o Pi (t) – the active power, injected from the bus into described in the present scope, and are fully detailed in [17].
the grid (positive for generators, negative for loads). For a simplified unbalanced system, in which leakage currents
o Qi (t) – the reactive power, injected into the grid. are neglected, and the neutral line impedances are taken as
o Vi (t) – the voltage magnitude of the bus. zero, the power-flow equations are given by (2), where the
o δi (t) – the phase angle of the voltage Vi. sub-index ‘x’ represents the phase: A,B or C. It is later
Basic units of the microgrid are defined in table I. It describes explained how these various system representations are
single phase units, balanced three-phase units, or unbalanced integrated into the proposed time-domain solver.
three-phase units, with per phase representation (‘x’ denote the
phase, A,B, or C, and is omitted for balanced systems).
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 3

N C(t) is a price signal ($/MW), usually a time-dependent

Pi , x 
Vi , x ˜ ¦ Yij , x ˜ V j , x ˜ cos G i , x  G j , x  Tij , x  function. For an unbalanced three phase system (Fig. 2) the
j 1
objective is:
N
Qi , x 
Vi , x ˜ ¦ Yij , x ˜ V j , x ˜ sin G i , x  G j , x  Tij , x  (2)
³  PA  t   PB  t   PC  t   ˜ C  t  dt o min
T

(6)
j 1
0
iG i , x iG j , x
I ij , x Vi , x ˜ e  Vj,x ˜ e An alternative objective for unbalanced three phase systems
may be balancing the power among the three phases:

³  PA  t   PB  t     PB  t   PC  t  
T
Storage devices are modeled by an inner state variable – 2 2
dt o min (7)
the stored energy Ei(t). They are defined by a general state 0

equation, f(·), given by:

One phase storage device: III. GRADIENT BASED SOLVERS IN TIME DOMAIN
d
Ei fi  Pi , Ei  (3) An optimal solution computes powers (P,Q), voltages (V),
dt currents (I), phase angles (δ) and stored energy (E), see table I.
Three phase storage device: All quantities are time dependent. While Gradient based
fi  Pi , A , Pi , B , Pi ,C , Ei 
d solvers (like Newton-Raphson) have been proved to solve
Ei (4)
dt networks efficiently, they are inadequate to storage problems.
Where Pi,A, Pi,B, Pi,C, are the three phase powers, at the PCC. The following section explains this claim.
(See Fig. 2b). The desired solution is a function of stored energy vs. time,
The objective is to minimize the overall cost of energy E(t). To reach a solution, the solver must determine E(t) at
import from the public grid, determined by incoming power at every time point, thus each time point is a free variable.
the PCC. For a single phase system the objective is: Apparently, many time-energy functions are local minima.
T
Consider, for instance, the straight line in Fig. 3b. Assume a
³ P1  t  ˜ C  t  dt o min (5) small perturbation over this possible solution. Due to
0
inefficiencies of the storage device, charging and discharging
are wasteful, therefore, in respect to the objective, the
Pi , Qi
perturbation is less efficient, and is worse than the straight line
solution. The straight line is therefore "surrounded" by
P1 Ei  t  solutions which are worse, and is in-fact a local minimum.
single phase
PCC storage
utility
grid E t 
Emax
Pj , Q j
G globally
E j t  optimal
solution
single phase
(a) storage time
(a)
G
E t 
feeder
A
PA Emax a small variation over
the local minimum

three phase Pi , A
transformer Local minimum
feeder PB Pi ,B “trap function”
B
PCC
time
utility Pi ,C Ei  t  (b)
grid G
Fig. 3. Global and local solutions in time domain. (a) global solution. (b) A
three phase local solution (bold) with a small variation (thin). The variation is
storage energetically worse than the local solution, due to charge and discharge losses.
feeder
C
PC
While this explanation is by no means a mathematical
(b) proof, it highlights the numerical difficulties of gradient based
Fig. 2. Notations of one-phase or three-phase networks and storage devices (a)
solvers, applied to time domain problems. If a gradient based
Single phase system with two storage devices. (b) Three phase system with a solver reaches a local solution, such as the straight line in Fig.
three-phase device. 3b, it will conclude that it is optimal. However, this local
solution does not resemble the global one, and have no desired
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 4

properties, such as charging when energy is available and so E(t) and its first derivative, ௗ௧ௗ E(t), revealing Pi(t).
forth. This local solution is unsuitable, and cannot be used in a 3. The storage device is replaced with an auxiliary P-V
real power system. source, with P=Pi(t), V=VS,i. A network power flow
analysis is computed using Gauss-Seidel, Newton-
IV. THE DYNAMIC PROGRAMMING APPROACH Raphson, or any other method.
Unlike gradient based methods, dynamic programming 4. If the power flow solution complies with all network
algorithms (see [18]) scan all feasible solutions, to locate the constraints, The differential, ΔV, is assigned a value
global optimum. A direct scan of the entire solution space is according to the power at the PCC, P1(t). Otherwise, it is
numerically impossible, so the optimal solution is designed assigned a value of infinity:
recursively, combining dynamic allocation in time domain, ­ P  t  ˜ C  t  ˜ dt in ½
° constraints °¾
with a traditional power flow solver on the network domain. 'V ® 1 (11)
°¯ f otherwise °¿
A. A single storage device – one dimensional solution
With a single, one phase storage device, the stored energy Having computed V(t,E) over all times and energies, the
function, Ei(t), governs the power flow of the network. For a optimal energy E*(t) may be evaluated. This is done by a
given energy function, the power output of the storage device, forward recursion process. Known values of V(t,E) are
Pi(t) may be computed using the storage state equation (3). substituted in the Bellman equation to recover the optimal
Assuming that the voltage magnitude of the device is solution:
specified, Vi(t) = VS,i, the device may be replaced by an
auxiliary P-V unit, with known power and voltage values. V  t, E  E t  E  t  dt 
Recall that loads and renewable generators are specified, so
7 6
given the storage power output, power flow over the entire dV=2
network may be computed. This is easily achieved using 8
dV=4
standard power flow algorithms, such as Gauss-Seidel, or 4
Newton-Raphson. The problem is therefore one-dimensional, E dV
=∞
with a single controllable state variable, Ei(t).
The challenge is to determine an energy function, Ei(t)= 1
E(t), that minimize the objective, (5) and comply with all
constraints listed in table I. To this end, a value function V(·) t
is defined: t t  dt
T (a)
V t, E  ³ P W  ˜ C W  ˜dW
t
1
(8)
with an initial condition E  t  E
The objective (5) is equivalent to minimizing V(0,0), that is,
Ei
to minimize overall cost over the entire period, starting with
an empty storage, E=0. Calculations are numeric, over a
discrete grid. dt marks the time step, and dE marks the energy Ej
step. The optimal solution is computed recursively by the
Bellman equation:
° 'V  E, E  t  dt  
­ ½
°
V  t , E  min ® ¾ (9)
E  t  dt 
°
¯  V  t  dt , E  t  dt   °
¿ t t
The value-function, V(t,E) is numerically computed by
t  dt
(b)
backward recursion. The process starts at the final time, t=T, Fig. 4. The backward recursion process, using the Bellman equation. The
where the value function is known: V(T,E)=0. Applying (9), value function at each point equals the minimum over all differential paths
the value function may be computed at T-dt, revealing V(T- from t to t+dt. Possible paths appear as dashed lines, and the optimal path is
marked bold. (a) A single storage device, with one state variable E(t). (b)
dt,E) over all the energy values. The process continues until Multi-dimensional - Two storage devices with two variables – Ei(t), Ej(t).
reaching t=0. A backward recursion step is shown at Fig. 4a. White dots mark the numerical grid. Black dots mark feasible solutions.
The differential cost, ΔV, is defined for every two arbitrary
points, {t, E(t) } and {t+dt, E(t+dt)}. It represents the cost of
transition between the two points. ΔV is computed in steps: E*  t  arg min ®

­ 'V E *  t  dt  , E  t   ½
¾ (12)
1. The first derivative of energy is evaluated by: E t  ¯  V t, E t  ¿
d E  t  dt   E  t 
E| (10)
dt dt Energy at t, E*(t), is calculated in relation to a previous
2. Power output of the storage device, Pi(t) is evaluated. The energy value, E*(t-dt). The computation process starts at t=0,
storage state equation (3) is solved using known values of in which optimal energy is known and equals the starting
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 5

condition, usually E*(0)=0. Optimal energy at the next time smoothing total power injected to the grid.
step, E*(dt), is evaluated by (12). The process continues until The following description is duplicated from [16]: Wind
the entire optimal energy path has been discovered, up to the power is represented by Pwind, storage power is PES, total
final time t=T. Knowing the optimal energy path, all powers, power is Ptotal. The battery is modeled by its power capacity
voltages and phase angles may be computed directly. Prated, the storage capacity Jrated, and the battery State of
Charge (SOC), in the range 0...1. This represents energy in
B. Multiple storage devices
this problem. The storage state equations are:
Each storage device adds a dimension to the solution space.
Thus, a network with two storage devices is a two- K ˜ PES
d
dimensional problem, with two free variables: Ei(t) and Ej(t). SOC 
dt J rated
Power flow is now governed by two energy functions, instead
of one. A two-dimensional computation is shown at Fig. 4b. ­Kout PES ! 0 ½
K ® ¾ (16)
Multi-dimensional solutions are essentially equivalent to ¯ Kin PES  0 ¿
single dimension solutions. The major difference is that the
 Prated d PES d Prated
value function, V(·) is now multi-dimensional. Consider, as an
example, a one-phase network with two storage devices. The 0 d SOC d 1
value function is now defined as follows:
T Pwind

V t , Ei , E j  ³ P W  ˜ C W  ˜dW
1
(13) Ptotal G
wind
farm
t utility
initial condition: Ei  t  Ei , E j  t  E j grid PES energy
storage
(a)
It is a function of both Ei(t) and Ej(t). The Bellman equation
now involves minimization over both energy variables: 1

^ `
Pwind
'V  Ei , Ei  t  dt  , E j , E j  t  dt  
power [p.u.]

V  t , Ei , E j  (14)
Ptotal
Ei  t  dt ,  V  t  dt , Ei  t  dt  , E j  t  dt  
min
0.5
E j  t  dt 
The differential cost, ΔV is computed for two-dimensional
points: {t, Ei(t), Ej(t) } and {t+dt, Ei(t+dt), Ej(t+dt)} (See Fig. 0
4b). The computation involves evaluation of two derivatives:
Ei  t  dt   Ei  t 
0.5 Prated
d
power [p.u.]

Ei |
dt dt
(15) 0
E j  t  dt   E j  t 
d
Ej |
PES  Prated
dt dt
-0.5
These are employed for computing the output power of both
storage devices. Using this data, the network power flow is
evaluated normally, at each time point, using Gauss-Seidel, or 1
Newton-Raphson. The differential, ΔV, is assigned a value
SOC

according to the power at the PCC, or assigned a value of 0.5

infinity in case the solution is infeasible.
Three phase storage devices may be one-dimensional, or 0
three dimensional. A balanced three phase device, with equal 201.5 202 202.5 203 203.5 204
Time [days]
powers, Pi,A=Pi,B=Pi,C, is one dimensional. The stored energy,
E(t), determines the three phase powers, in accordance with (b)
the storage state equation (4). Knowing the storage output Fig. 5. Power system of Brekken et. Al. [16]. (a) Wind farm and storage. (b)
Optimal solution. Top – wind power and total power. Middle – storage power.
powers, a three-phase power flow analysis is computed at each Bottom –Battery State of Charge (SOC).
time point. If phase powers are each individually controlled,
using a dedicated power converter, than the problem is three The parameters are chosen as follows: Prated=0.34, Jrated=0.4,
dimensional. The free variables may be Pi,A(t), Pi,B(t), Pi,C(t). ηin=0.85, ηout=1.15. Wind power, Pwind, is sampled from [16].
The proposed dynamic programming analysis is applied to
V. MICROGRID CASE STUDY I this system, optimizing the utilization of storage. A price
To demonstrate the proposed method, we examine a power signal is unavailable, so a minimal price objective cannot be
system proposed by Brekken et. Al. [16]. The system (Fig. 5a) evaluated. Instead, we chose to optimize the power output of
includes a wind farm (renewable source), coupled with a the system by minimizing losses over the mutual power line.
battery energy storage. During high winds, energy is stored in Assuming a resistance of R=0.01, and a bus voltage V1=1.0,
the battery. Stored energy is released when wind is low, the objective is:
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 6

§
T The microgrid contains 2 renewable generators, 6 loads (the
2 ·
³
R
¨ Ptotal  2 ˜ Ptotal ¸ dt o min (17) sixth is a capacitor bank), and 2 storage devices. The
0© V1 ¹ generators are photovoltaic sources having installed power
The value function V(·) is defined over time and the state of peaks of 1 MW and 0.5 MW. They provide only active power.
charge, SOC. It embeds the objective, as follows: Power signals were generated randomly, over a 72 hours
T
§ R · period, as shown in Fig. 7. The storage device capacities are
V  t , SOC  ³ ¨ Ptotal W   2 ˜ Ptotal
2
W  ¸ dW Emax = 0.4 MWhr (Mega Watt – hour) each. The storage state
(18)
t © V1 ¹
with an initial condition SOC  t  SOC
equation is:
d
Ei D ˜ Ei  P  Pi  ˜ Pi
The analysis is computed over a numeric grid, using time dt
­1 / K0 Pi t 0 ½
step of dt=0.1 hr, and dE = d(SOC) = 0.01. V(·) is evaluated P  Pi  ® ¾ (19)
using backward recursion, scanning all possible paths of SOC ¯ K0 Pi  0 ¿
over time. Then, the optimal function SOC(t) is constructed by 0 d Ei  t  d Ei ,max
forward recursion. Optimal power flow results are shown in
where Ei,max is the capacity of the device, α is the rate of self-
Fig. 5b.
discharge, and η0 is the efficiency of the device. The chosen
parameters are: Ei,max= 0.4 MWhr (Mega Watt – hour), for
VI. MICROGRID CASE STUDY II each device, and α=0.02 1/hr. η0 is varying, taking the values
The second system case study combines both a non-trivial of η0=1.0, 0.6, 0.4, 0.1.
network and storage devices. Power flow is optimized to Power flow is optimized using a one dimensional process
satisfy both the storage device constraints, and the physical (Fig. 4). The two storage devices have the same ratings, and
constraints of the network. The objective is to optimize the are treated as a unified device. V(t,E) is numerically computed
cumulative price of energy at the PCC (eq. (5)). by backward recursion (eq. (8) – (11)), with dt=0.5 hr,
The network is shown at Fig. 6. This microgrid is a medium dE=0.05 MWh. At each time step, network power flow is
voltage (MV) network. It is supplied by a central transformer solved using the simple Gauss-Seidel analysis. Optimal stored
at the PCC, which ratings are: V1(t)=Vin(t)= 13.8 kV, Snom=5 energy is rebuilt by forward recursion (eq. (12)). The resulting
MVA. Impedances are specified in per-unit (in percent), using optimal power signals are shown in Fig. 8.
a base equal to the transformer's ratings. Active power at the
PCC is limited by the transformer: -5≤P1(t) ≤+5 MW.

L5
G1: photovoltaic source, installed peak: 1 MW
G2: photovoltaic source, installed peak: 0.5 MW
0.5 MW
S1, S2: energy storage, capacity: 0.4 MWhr
L6: capacitor bank for reactive power compensation (0.3 MVAr) 1.3+ L4 G2
j4.2%

69 kV / 13.8 kV 1.4 +
5 MVA j2.6%
0.6 + j5.0 %

PCC 13.8 kV 0.07+j0.2 % 0.13+j0.45 %

utility
0.8 +
grid 1.1 + j1.7%
j2.6 %
L6
0.3 MVAr 1.5 + 1.1 + 1.1 +
j1.7 % j4.4 % j2.6 % L3
L1
Vnom (base) = 13.8 kV
S (base) = 5 MVA
S1
0.4 MWhr G1
L2
1 MW S2
0.4 MWhr
Fig. 6. Microgrid case-study II.
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 7

The dynamic programming approach presents several

5 important advantages: First, it reveals the globally optimal
Loads
[MW]

solution, since the algorithm scans the entire solution space.

0 Second, the model is extremely general. It does not bind to a
certain model or constraint. Storage device are described by
2 general state equations (eq. (3)-(4)). Different storage devices
[MVAr]
Loads

with entirely different properties may be integrated in the

0
same network. One phase networks may be solved just as well
-2 as three phase networks. Different objective functions may be
considered (eq. (5)-(7),(17)). The solution method is always
1 the same: a recursive scan of all feasible system states.
Generators

A main disadvantage of the propose approach is that

[MW]

0.5
numerical complexity grows in power law with the number of
0 (different) storage devices. This is evident in Fig. 4b: Each
0 12 24 36 48 60 72
storage device contributes an extra dimension to the solution
time [hr]
space. From our experience, a network with a single storage
Fig. 7. Microgrid case-study II. Powers of loads and generators. Top – sum of
active load power. Middle – sum of reactive load power. Bottom – active device is evaluated in seconds, even if it contains hundreds of
power of generators. buses. However, a system with four or five devices may be
evaluated if it contains only a few buses.
VII. DISCUSSION Two system case studies are shown. The first system is a
This work proposes an optimal power flow analysis, two bus topology (Fig. 5). The second system (Fig. 6) is a
integrating storage devices. An exact global solution to this network, with two renewable sources, and two storage
problem has not been shown so far, due to the numerical devices. The systems differ in both topology and design
complexity of the problem, which includes both the network objectives, and employ different storage devices. The
domain, and the time domain, related with storage. Traditional objective in the first system is to minimize loss over the
solvers view the problem as a function to be minimized. This transmission line, whereas in the second system, the objective
way of thinking leads to traditional solution methods, such as is to minimize the price of imported energy. The same
gradient search and linear programming. However, traditional dynamic programming analysis is applied to both systems. the
solvers, such as Newton-Raphson, are inefficient in time cost function V(∙), shown in eq. (8), is defined, and evaluated
domain (See section III). We propose a new way of thinking, by backward-recursion, as shown in eq. (9)-(11). Optimal
which leads to an entirely different solution, capable of stored energy and power flow is revealed by forward
circumventing the numerical complexity: Instead of looking at recursion, described in eq. (12).
the problem as a function to be minimized within constraints, The resulting optimal power management (Fig. 5, Fig. 8)
we regard stored energy as a resource to be allocated. Instead reveals a similar conceptual strategy for both systems: energy
of a minimization problem, we view it as an allocation is stored according to the availability of the primary source:
problem: energy is allocated in time domain, to optimize wind power, in the first example, or low-cost energy at the
power import at the point-of-common coupling. PCC, in the second example. Stored energy is released to the
Allocation problems are easily solved by dynamic network when wind power is not available (first example) or
programing algorithms, so this is the main "engine" of our when the price of energy is high (second example). In both
solver. The solver combines a recursive dynamic systems, stored energy tends to equalize the total power, in
programming scan on time domain, with a traditional solver order to reduce losses as much as possible. The second system
(we used Gauss-Seidel) on the network domain. The solution includes the non-zero admittances of power lines, and the
process is general, and can be used with any network topology power limit of the transformer. This topology of the network
(single phase, balanced or unbalanced three-phase). Likewise, effects the management of stored energy. For example, the
there is no assumption on the power-flow solver, which can be storage device compensates the load, to sustain the 5 MVA
Newton-Raphson, Gauss-Seidel, or a specific per-phase capacity of the PCC's transformer, as shown in Fig. 8c. Fig. 8
solver, such as forward backward sweep or TCIM. At each reflects the effect of storage efficiency (η0) on power
time point, the value function, V(·),is evaluated in respect to a management. In Fig. 8a (η0=1), the storage is charged to full
previous value. The network power flow is solved at each time capacity. In Fig. 8b (η0=0.6), losses are higher, and the storage
point, after replacing the storage devices with dummy P-V is not fully charged. In Fig. 8c (η0=0.4), the storage device is
sources. This power output is evaluated only by the current so lossy that it becomes economically worthless. In Fig. 8d
and previous energy values. Thus, each time point is evaluated (η0=0.1), the storage losses dominates. Here, the storage
once. There is no need to re-compute the entire energy path, so device cannot supply enough power to compensate the load's
the calculation is numerically efficient. power peak, and the network will become unstable. There is
no feasible solution.
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 8

Fig. 8. Microgrid case-study II. Resulting optimal powers for different values of storage efficiency. (a) η0=1.0. (b) η0=0.6. (c) η0=0.4. (c) η0=0.1. The graphs
from top to bottom: C(t) – price. E(t) – stored energy. P1(t) is the power supplied by the utility grid (at the PCC), Pgen(t) - the combined active power generation
of all loads and renewable generators (taken negative). PS(t) – combined active power supplied by the storage devices. QS(t) – combined reactive power supplied
by the storage devices.
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 9

VIII. CONCLUSION [18] D. P. Bertsekas, Dynamic Programming and Optimal Control, Vol. I,
Athena Scientific, 1995.
This work suggests an algorithm to compute the optimal
energy management of storage devices in grid-connected BIOGRAPHIES
microgrids. Stored energy is controlled to balance the power Yoash Levron holds a Ph.D. from Tel-Aviv University, Israel. His research
of loads and renewable sources, over the time domain, interests include power electronics, power systems and distributed power
minimizing the overall cost of energy at the PCC. The management.
algorithm incorporates an arbitrary network topology, which Josep M. Guerrero (S’01–M’04–SM’08) is a Full Professor at the Institute of
can be a general one-phase, balanced or unbalanced three- Energy Technology, Aalborg Universiy, Denmark, and responsible of the
phase system. It employs a power flow solver in network Microgrid research program. His research interests include power electronics
converters for distributed generation and distributed energy storage systems,
domain, within a dynamic programming recursive search in control and management of microgrids and islanded minigrids, and
time domain. This combination is robust and numerically photovoltaic and wind power plants control. He is an associate editor of the
efficient, and reveals the globally optimal stored-energy vs. IEEE Transactions on Industrial Electronics, the IEEE Ind. Electron.
Magazine, and the IEEE Transactions on Power Electronics.
time for each storage device.
Yuval Beck received his B.Sc. in 1997, M.Sc in 2002 and PhD in 2007, at Tel
Aviv University, all in electrical engineering. His fields of interest are: power
REFERENCES systems, power electronics, lightning phenomena, photovoltaic systems and
[1] J. C. Vasquez, J. M. Guerrero, J. Miret, M. Castilla and L. G. de Vicuña, smart grid. Dr. Beck is head of the energy and power systems department in
"Hierarchical Control of Intelligent Microgrids", IEEE Industrial the Holon Institute of technology (HIT).
Electronics Magazine, vol. 4, pp. 23-29, Dec. 2010.
[2] E. Barklund, N. Pogaku, M. Prodanovic, C. Hernandez-Aramburo and T.
C. Green, "Energy Management in Autonomous Microgrid Using
Stability-Constrained Droop Control of Inverters," IEEE Trans. Power
Electronics, vol. 23, pp. 2346-2352, Sept. 2008.
[3] J. Arai, K. Iba, T. Funabashi, Y. Nakanishi, K. Koyanagi and R.
Yokoyama, "Power electronics and its applications to renewable energy
in Japan," IEEE Circuits and Systems Magazine, vol. 8, pp. 52-66, Third
Quarter, 2008.
[4] F. Katiraei, R. Iravani, N. Hatziargyriou and A. Dimeas, "Microgrids
management," IEEE Power and Energy Magazine, vol. 6, pp. 54-65,
May-June, 2008.
[5] R.H. Lasseter, “Microgrids“, IEEE Power Engineering Society Winter
Meeting, Vol. 1, pp. 146-149, Feb. 2001.
[6] A. G. Tsikalakis and N. D. Hatziargyriou, "Centralized Control for
Optimizing Microgrids Operation," IEEE Trans. Energy Conversion,
vol. 23, pp. 241-248, March. 2008.
[7] R. Majumder, A. Ghosh, G. Ledwich and F. Zare, "Power Management
and Power Flow Control With Back-to-Back Converters in a Utility
Connected Microgrid," IEEE Trans. Power Systems, vol. 25, pp. 821-
834, May. 2010.
[8] V. Del Toro, Electric power systems, vol. II. Prentice-Hall, 1992
[9] N. P. Padhy, "Unit commitment-a bibliographical survey," IEEE Trans.
Power Systems, vol. 19, pp. 1196-1205, May. 2004.
[10] Y. Atwa, E. El-Saadany, M. Salama and R. Seethapathy, "Optimal
renewable resources mix for distribution system energy loss
minimization", IEEE Trans. Power Systems, vol. 25, pp. 360-370, Feb.
2010.
[11] Hassan Nikkhajoei and Reza Iravani, "Steady-State Model and Power
Flow Analysis of Electronically-Coupled Distributed Resource Units,"
IEEE Trans. Power Delivery, vol. 22, pp. 721-728, Jan. 2007.
[12] G. Celli, E. Ghiani, S. Mocci and F. Pilo, "A multiobjective evolutionary
algorithm for the sizing and siting of distributed generation", IEEE
Trans. Power Systems, vol. 20, pp. 750-757, May 2005.
[13] Y. Levron and D. Shmilovitz, "Power systems’ optimal peak-shaving
applying secondary storage," Electric Power Systems Research, vol. 89,
pp. 80-84, 8, Aug. 2012.
[14] Y. Levron and D. Shmilovitz, "Optimal Power Management in Fueled
Systems With Finite Storage Capacity," IEEE Trans. Circuits and
Systems I: Regular Papers, vol. 57, pp. 2221-2231, Aug. 2010.
[15] J. P. Barton and D. G. Infield, "Energy storage and its use with
intermittent renewable energy", IEEE Trans. Energy Conversion, vol.
19, pp. 441-448, June 2004.
[16] T.K.A. Brekken, A. Yokochi, A. von Jouanne, Z.Z.Yen, H.M. Hapke, D.
A. Halamay, "Optimal Energy Storage Sizing and Control for Wind
Power Applications", IEEE Trans. Sustainable Energy,vol.2, no.1,
pp.69-77, Jan. 2011.
[17] J. C. M. Vieira, W. Freitas ,A. Morelato, "Phase-decoupled method for
three-phase power-flow analysis of unbalanced distribution systems",
IEE Proceedings Generation, Transmission and Distribution, vol. 151,
pp. 568-574, Sept. 2004.