See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/320455698

An Economic Robust Programming Approach for the Design of Energy

Management Systems

Chapter · March 2017

CITATIONS READS

3 445

2 authors, including:

Felipe Valencia
Universidad Austral de Chile
88 PUBLICATIONS 1,160 CITATIONS

SEE PROFILE

All content following this page was uploaded by Felipe Valencia on 17 October 2017.

The user has requested enhancement of the downloaded file.

 Chapter 11
An Economic Robust Programming Approach
for the Design of Energy Management Systems
Felipe Valencia Arroyave1
Alejando Marquez Ruiz2

Energy management systems have a key role in the operation of micro-grid appli-
cations. Due to their importance, several approaches have been reported in the spe-
cialized literature. In this chapter a novel formulation for robust optimization-based
EMS for micro-grid applications. The novelty recast in the introduction of transmis-
sion constraints and in the avoidance of the use of prediction models for the uncer-
tain variables. In this formulation a time-variant uncertainty set is used to include
the uncertainty in the optimization process. Unlike in the approaches reported in the
specialized literature the sets are previously defined based on the empirical statis-
tics of historical data available. Thus, at each sampling time the boundaries for the
uncertain variables are obtained without using a prediction model. This reduces the
complexity of the EMS making possible to compute the solution of the associated
optimization problem in the range of the seconds. A modified version of the IEEE 9-
bus system benchmark was used as test-bench. In the modified version the hydraulic
generation unit (the largest unit in the system) was changed by an equivalent wind
power station. The EMS was in charge of assigning the real power and the voltage
level of the remaining two generation units.

1 Solar Energy Research Center, SERC-Chile

Energy Center
Depertament of Electrical Engineering
University of Chile, Santiago
Chile
e-mail: [email protected]
T: +56 2 2978 4200
2 Control Systems Group

Department of Electrical Engineering

Eindhoven University of Technology
Flux 5.076 (5th floor)
P.O. Box 513, 5600 MB Eindhoven
The Netherlands
e-mail: [email protected]
T: +31 64 247 7310
 Running Head Verso

11.1 Introduction
The development of energy management systems (EMS) has received much atten-
tion in recent years due to its key role in the operation of micro-grids, providing an
optimal use of local energy resources and infrastructure available to supply the de-
mand of a predefined area (very often remote and disconnected of a main grid). An
EMS is a control loop used in micro-grid applications to achieve reliable and cost-
effective operation through the dispatch and commitment of the available generation
units. Often the EMS is implemented using optimization-based control techniques
that additionally provide the required coordination among energy sources for a safe
and secure operation. Indeed, at this level, the most cost-effective combination of
generation units to meet the predicted load and reserve requirements is defined [26].
Since micro-grids have became one of the more appealing alternatives for electrifi-
cation purposes it is required the design of an EMS able to properly accomplish with
the aforementioned requirement (coordination among energy resources considering
a cost-effective, save and secure operation of the micro-grid) [11, 15, 6, 11]. How-
ever, it has been found that an adequate performance of optimization-based EMS its
highly related with the degree of representativity of the model used in the optimiza-
tion to represent the behavior of a micro-grid. One way to enhance the performance
of optimization-based EMS approaches is to incorporate an uncertainty model into
the formulation of the optimization problem, so that the effects of the differences
between the model and the real system (due to parameter mismatches, un-modeled
phenomena, the presence of stochastic variables such as wind and solar generation)
are included within the solution of the EMS problem.
Several optimization methods with application to management and decision mak-
ing under uncertainty have been reported in the specialized literature (see e.g., [25,
2]). In [25, 23, 24, 14], these approaches have been classified as: stochastic and
robust. In the former, several realizations of the uncertain variables are considered.
Each realization has a probability of occurrence. Depending on this probability, a so-
lution is obtained. In the later, it is assumed that the parameters and/or constraints in
the optimization model are uncertain and prescribed into a defined set [3]. Depend-
ing on the construction of the uncertainty set, a solution is obtained. The advantage
of robust over stochastic approaches is that in the robust framework i) the use of
probability density functions is not required, i.e., only the range of the uncertain data
is required, and ii) the solution is robust against all realizations of the uncertain data
within a deterministic uncertain set [2]. The disadvantage is that there are no guide-
lines available for the construction of the uncertainty set. Consequently, a solution
more conservative than necessary could be achieved. Examples of both an stochastic
and a robust EMS has been presented in [28, 29]. In [29] probability density func-
tions were used to generate the scenarios for each uncertain variable, while in [28] a
methodology to generate time-dependent uncertainty sets was proposed.
The use of stochastic and robust optimization has not been only limited to micro-
grid applications. In large-scale power systems some additional approaches has also
been reported. In [2, 30, 27] multi-stage optimization-based methodologies were
proposed for the unit commitment in systems with wind generation and price re-
 Running Head Recto

sponsive demand. In these approaches, the first decision stage consisted on the unit
commitment and the second decision stage consisted on the power dispatch. More-
over, in [30] a third decision stage was added to include the uncertainty in the price-
elastic demand curve. Furthermore, in water resources management the authors of
[25, 23, 24, 14] investigated the use of two-stage stochastic fuzzy programming with
interval-valued membership functions. Specifically, the support of water resources
management was addressed in [23], the flood-diversion planning was analyzedin
[24], and the municipal solid waste management under uncertainty was studied in
[25, 14]. Since interval-valued membership functions provided upper and lower
values for the uncertain variables, in these approaches the first decision was made
considering each boundary of the fuzzy model as a deterministic scenario, then a
feasibility test was performed to eliminate the unfeasible solutions, and finally the
second decision was made over the feasible solutions. At the end of the process, a
set of solutions was obtained. How to select an alternative from the resulting set of
solutions is still a pendent issue.
Notwithstanding the broad use of stochastic and robust optimization approaches
in decision making under uncertainty, particularly in energy management in power
systems of different scales (micro-grid applications and large-scale power systems),
very often their performance depends upon the performance of prediction models
used to generate representative realizations of the uncertain variables (stochastic ap-
proaches) and to define the expected range of variation of the uncertain variables
(robust approaches). Moreover, the use of predictions increase the complexity of the
EMS, which increase the computational burden of the solution accordingly. Indeed,
several EMS has solution times in the range of the minutes (e.g. the EMS proposed
in [16] and in [15]) requiring additional control loops to avoid permanent frequency
and voltage deviations produced by the action of local primary controllers. Further-
more, in the context of micro-grid applications little attention has been paid to the
addition of transmission constraints in the EMS.
The present chapter presents a novel formulation for robust optimization-based
EMS for micro-grid applications. In this formulation a time-variant uncertainty set
is used to include the uncertainty in the optimization process. Unlike in the ap-
proaches reported in the specialized literature the sets are previously defined based
on the empirical statistics of historical data available. Thus, at each sampling time
the boundaries for the uncertain variables are obtained without using a prediction
model. This reduces the complexity of the EMS making possible to compute the
solution of the associated optimization problem in the range of the seconds. In this
EMS transmission constraints are also included in the formulation of the optimiza-
tion problem. Since transmission constraints are non-linear non-convex, the formu-
lation proposed in [8, 9, 10, 7] was used. Such formulation allows transforming
the transmission constraints in linear convex ones through a non-linear bi-directional
variable-transformation. The resulting optimization problem then remains convex
(as in the case of the uni-nodal formulations reported in the literature) and therefore
can be solved in finite steps and has an unique solution. Indeed, the EMS becomes
a second-order cone programming problem (SOCP) which can be efficiently solved
and whose complexity scales-up proportionally to the square-root of the number of
 Running Head Verso

decision variables. The aforementioned features of the proposed EMS in addition to

the reduction of the sample-time from the minutes to the seconds allowed to control
both the frequency and the voltage profile all over the micro-grid, without includ-
ing a secondary control. A modified version of the IEEE 9-bus system benchmark
was used as test-bench. In the modified version the hydraulic generation unit (the
largest unit in the system) was changed by an equivalent wind power station. The
EMS was in charge of assigning the real power and the voltage level of the remaining
two generation units. The remainder of this chapter is organized as follows: section
11.2 introduces some basic mathematical concepts of robust programming; section
11.3 presents the proposed EMS; section 11.4 provides the case study used to test
the proposed EMS and the simulation results obtained; and section 11.5 presents the
concluding remarks.

11.2 Robust Programming Framework

In this section some mathematical concepts related with robust programming based
on scenarios are introduced. These concepts will be used in section 11.3 to formulate
the proposed robust EMS. For this purpose, consider an optimization problem whose
objective is to minimize the difference between a mathematical model and a set of
measured values. Very often an optimization problem is formulated to solve this
problem. Let J(x) denote the objective function of such an optimization problem,
where x is a set of decision variables that determines the fulfilling of the objective
of the optimization model. Then, if J(x) is assumed linear such an objective func-
tion has the form J(x) = kAx − bk. Without loss of generality, assume that A is a
random variable with an empirical probability density function describing their vari-
ations. Then, finding a value os x that minimize J(x) becomes a minimization of
the expected value of J(x), whose solution is closely related with the description of
the uncertainty of A. indeed, depending of the way the uncertainty in A is addressed
the resulting optimization problem could be tractable or not. For this reason uncer-
tainty description by means of polytopes or norm-bounded models are proposed in
the literature (see [4] for details).
A particular case in which minimizing the expected value of J(x) is tractable is
the worst-case problem. In worst-case problem the uncertainty is described by a set
of values for A that constitutes an uncertainty set. Given this set, the uncertainty is
analyzed to find the worst conditions for J(x). Once these conditions are identified
the value of x minimizing J(x) is computed. Thereby, the expected value becomes
a min-max optimization problem. In the specialized literature several approaches
have been reported to solve min-max optimization problems [20, 12, 17, 13, 1]. But
often worst-case problems became nonlinear programming problems. With currently
computation resources it is almost impossible to find the exact solution of this type
of optimization problems in real time applications [5]. Therefore transformations
and/or linear approximations are used to speed up the computation of the solution of
worst-case optimization problems.
One of the most used transformations is the norm bounded error. In this case the
uncertainty of A is modeled as an ellipsoid that represents the expected variations
 Running Head Recto

of A. Let A = Ā + ∆A represents the variations of A, with Ā the mean value of A

and kδ Ak ≤ ϒ the variations of A. Then, the worst-case scenario for J(x) is given
by: sup{kĀx − b + δ Ax | δ A ≤ ϒk}. Let k · k denote the Euclidean norm. Then the
resulting worst-case scenario is given by kĀx − bk2 − akxk2 , with A = auν T . Then
the robust approximation minimization problem is formulated as follows:

min kĀx − bk22 + akx|22 (11.1)

x

which is a regularized norm problem, solvable as the second order cone program-
ming problem (SOCP) [4]:

min t1 + at2
x,t1 ,t2

s.t:
kĀx − bk22 ≤ t1
kx|22 ≤ t2
t1 ,t2 ≥ 0 (11.2)

Other alternative to represent the uncertainty of A is the use of scenarios. In this

case it is assumed that A belongs to the set Ξ = {A1 , . . . , Ah }. In this case, the worst-
case scenario is determined by the values of A that maximize the uncertainty. That is,
the worst-case scenario is given by the value of A ∈ Ξ that maximizes the deviation
of Ax. Mathematically, this optimization problem can e formulated as foollows:

min t
x,t

s.t:
kAi x − bk22 ≤ t, i = 1, . . . , H
t ≥0 (11.3)

where H the number of scenarios. Note that minimization problem Eq. (11.3) only
requieres the definition of the scenarios to represent the uncertain variables. These
scenarios could be generated from interval models or from the empirical statistics
of the available data. In this chapter the later approach has been used. To avoid
the requirement of prediction models, empirical statistics have been used. Empirical
statistics have been selected because they can be easily estimated off-line.

11.3 Energy Management System as a Robust Programming

Problem

The current investigation involved the formulation of a robust optimization-based

EMS for micro-grid applications. The proposed EMS was formulated on the basis of
the mathematical concepts presented in section 11.2. For this purpose, the statistics
of historical data available of the uncertain variables has been used to derive their
expected range of variation. In this sense, the use of prediction models was avoided
 Running Head Verso

reducing the complexity of the EMS. Furthermore, transmission constraints were

included to account for the coupling between real and reactive power that arises in
micro-grid applications, and in low-voltage distribution systems in general. Since
transmission constraints are non-linear and non-convex, the variable transformation
proposed in [8, 9, 10, 7] was used to keep the optimization problem convex. As
a consequence, the EMS became a SOCP problem that can be efficiently solved
and whose complexity scales with the square-root of the number of decision vari-
ables. The relationship between complexity and number of decision variables allows
scaling-up the size of the micro-grid (measured as number of nodes and transmission
lines). This provides some degree of flexibility for the application of the proposed
EMS in real systems (micro-grids and even small and medium size power systems).
In real applications an EMS is included with the objective of providing on-
line real power set-points for each generation unit [16]. In this case, in addition
to real power set-points the EMS also sends voltage set-points to each generation
unit. Then, it is expected a better frequency and voltage control without using ad-
ditional control loops (e.g. secondary frequency and voltage controllers). Both real
power and voltage set-point are calculated so that the operational costs are mini-
mized. These costs are often related with the use of diesel generators and/or the
use of the energy stored in battery banks to supply the demand. Let Cgi denote the
generation cost of the i-th diesel generation unit; Pgi (k) denote the real power pro-
vided by the i-th diesel generation unit; CB j denote the cost of using the j-th battery
bank; PB j (k) denote the real power provided by the j-th battery bank; CNS denote
the cost of the un-supplied power; PNS (k) denote the total amount of un-supplied
power; Psl (k), Pwm (k) and Pdn (k) respectively denote the real power supplied by the
l-th solar power plant, by the m-th wind power plant and demanded by the n-th load
center. Then the single-node formulation of an EMS is as follows :
I J
min
Pgi (k),PB j (k),
∑ Cgi Pgi (k) + ∑ CB j PB j (k) +CNS PNS (k)
i=1 j=1
PLo (k)

s.t:
I J N L M
∑ Pgi (k) + ∑ PB j (k) + PNS (k) = ∑ Pdn (k) − PLo (k) − ∑ Psl (k) − ∑ Pwm (k)
i=1 j=1 n=1 l=1 m=1

Vgimin ≤ Vgi (k) ≤ Vgimax

min max
EBi ≤ EBi (k) ≤ EBi
Pgimin ≤ Pgi (k) ≤ Pgimax
min max
PBi ≤ PBi (k) ≤ PBi
PNS (k) ≥ 0
PLo (k) ≤ 0 (11.4)

where PLo (k) is the unused power; Vgimin and Vgimax are the allowed minimum and
maximum values of diesel volume for the i-th generation unit; EBmin max
j and EB j are the
allowed minimum and maximum values of stored energy in the j-th battery bank;
 Running Head Recto

Pgimin and Pgimax are the allowed minimum and maximum values of real power gen-
eration for the i-th diesel generator; PBi min and Pmax are the allowed minimum and
Bi
maximum values of real power generation of the j-th battery bank; and I, J, L, M
and N the number of diesel generators, battery banks, solar power plants, wind power
plants and load centers. Note that PLo (k) is a slack variable used to relax the power
balance equality constraint. Physically, this variable indicates the energy to be dissi-
pated for maintaining the balance in the micro-grid.
In Eq. (11.4) only constraints over the real power are considered. Moreover,
since a single-node transmission/distribution grid is assumed it is not possible to
guarantee that the scheduled powers Pgi (k) and PB j (k) can be transferred from the
sources (diesel generators and battery banks) towards the load centers. Indeed, since
the transmission capacity of the lines is not included in the formulation, it is possible
to obtain values of Pgi (k) and PB j (k) that could exceed their maximum transmission
capacity. Additionally, the real power set-points computed using Eq. (11.4) could
lead to operating conditions highly sensitive to the variations in Psi (k), Pw j (k) and
Pd j (k). To overcome these issues, transmission constraints were added in the for-
mulation of the EMS problem Eq. (11.4)3 . Let Ybus = G + jB denote the admitance
matrix of a micro-grid. Then, the power-balance equations for a node r of the system
are the following:

Pr (k) = νr 2 (k)Gr,r
S
+ ∑ [νr (k)νs (k)Gr,s cos(δr (k) − δs (k)) + νr (k)νs (k)Br,s sin(δr (k) − δs )(k))]
s=1,s6=r

Qr (k) = −νr 2 (k)Br,r

S
− ∑ [νr (k)νs (k)Gr,s cos(δr (k) − δs (k)) − νr (k)νs (k)Br,s sin(δr (k) − δs (k))]
s=1,s6=r

with S the number of nodes in the grid, Pr (k) and Qr (k) the difference between the
generated and demanded power at the r-th node, νr (k) the voltage at the r-th node,
and δr (k) the voltage angle at the r-th node, r, s = 1, . . . , S. Note that a balance
between generated, demanded and transmitted real and reactive powers could be
achieved with several values of Pr (k), Qr (k), νr (k) and δr (k). This makes including
transmission constraints in an EMS a challenging task.
Let Rr,s (k) = νr (k)νs (k) cos(δr (k)−δs (k)); Tr,s (k) = νr (k)νs (k) sin(δr (k)−δs (k));
2
r (k)
ν√
and ur = 2
. Then, from the properties of sin(x) and cos(x) the following relations
hold: Rr,s (k) = Rs,r (k), Tr,s (k) = −Ts,r (k) and 2ur (k)us (k) = R2r,s (k) + Tr,s
2 (k). Fur-

3 For simplicity in the formulation, in the sequel only electric constraints will be considered in the

EMS. In [22, 21] how to relate non-electric variables, such as the diesel volume, and the energy stored in
a battery bank with the decision variables of Eq. (11.4) is presented
 Running Head Verso

thermore, the power balance equations become:

√ S
Pr (k) = 2Gr,r ur (k) + ∑ [Gr,s Rr,s (k) + Br,s Tr,s (k)]
s=1,s6=r
√ S
Qr (k) = − 2Bi,i ur (k) − ∑ [Br,s Rr,s (k) − Gr,s Tr,s (k)] (11.5)
s=1,s6=r

In Eq. (11.5) the balance between generated, demanded and transmitted real and
reactive powers depends upon the values of Pr (k), Qr (k), ur (k), Rr,s (k) and Tr,s (k).
The values of these variables can be obtained by solving a system of linear equations,
which has a unique solution. Accordingly, transmission constraints can be included
in the formulation of the EMS without significantly increasing its complexity. Let
define Pr (k) and Qr (k) in terms of the diesel generation, battery generation, solar
power, wind power and the demand as: Pr (k) = Pgr (k) + PBr (k) + Psr (k) + Pwr (k) −
Pdr (k) and Qr (k) = Qgr (k) + QBr (k) − Pdr (k). Then, the minimization problem Eq.
(11.4) can be reformulated as follows:

I J
min ∑ Cgi Pgi (k) + ∑ CB j PB j (k) + αb(k) + β bp (k) + γbq (k)
Pgi (k),PB j (k), i=1 j=1
b(k),bp (k),bq (k)
ur (k),Rr,s (k),Tr,s (k)

s.t:
I J N L M
∑ Pgi (k) + ∑ PB j (k) = ∑ Pdn (k) − ∑ Psl (k) − ∑ Pwm (k) + bp (k)
i=1 j=1 n=1 l=1 m=1
I J N
∑ Qgi (k) + ∑ QB j (k) = ∑ Qdn (k) + bq (k)
i=1 j=1 n=1

Pgr (k) + PBr (k) + Psr (k) + Pwr (k) − Pdr (k) =
√ S
2Gr,r ur (k) + ∑ [Gr,s Rr,s (k) + Br,s Tr,s (k)] + b(k)
s=1,s6=r
√ S
Qgr (k) + QBr (k) − Qdr (k) = − 2Bi,i ur (k) − ∑ [Br,s Rr,s (k) − Gr,s Tr,s (k)] + b(k)
s=1,s6=r
max
|(Yr,s (ur (k)) − us (k))| ≤ Ir,s
2
2ur (k)us (k) = R2r,s (k) + Tr,s (k)
Pgimin ≤ Pgi (k) ≤ Pgimax
min max
PBi ≤ PBi (k) ≤ PBi
Qmin max
gi ≤ Qgi (k) ≤ Qgi
Qmin max
Bi ≤ QBi (k) ≤ QBi
b(k), bp (k), bq (k) ≥ 0 (11.6)
 Running Head Recto

max is the maximum allowed current flowing through the line connecting
where Ir,s
the nodes r and s; and b(k), bp (k) and bq (k) are slack variables used to relax the
equality constraints associated with the power balances. These slack variables can
be interpreted as active and reactive loses in each case. Moreover, since currently
solar and wind power plants are not being forced to supply reactive power, then their
contribution to the reactive power balance has not been considered.
Minimization problem Eq. (11.6) represents the more general case of energy
management in a micro-grid. In this case, a node r could have all kind of generation
resources (here diesel generation, battery bank, solar generation and wind gener-
ation) connected at the same time. Since real micro-grid projects attempt to take
advantage of the distributed local generation resources available very often only one
type of generation is connected to a particular node. Assuming that a node only have
one generation resource the minimization problem Eq. (11.6) becomes the following
optimization problem:

I J
min ∑ Cgi Pgi (k) + ∑ CB j PB j (k) + αb(k) + β bp (k) + γbq (k)
b(k),bp (k),bq (k) i=1 j=1
ur (k),Rr,s (k),Tr,s (k)

s.t:
I J N L M
∑ Pgi (k) + ∑ PB j (k) = ∑ Pdn (k) − ∑ Psl (k) − ∑ Pwm (k) + bp (k)
i=1 j=1 n=1 l=1 m=1
I J N
∑ Qgi (k) + ∑ QB j (k) = ∑ Qdn (k) + bq (k)
i=1 j=1 n=1
√ S
Pgi (k) − Pdi (k) = 2Gi,i ui (k) + ∑ [Gi,s Ri,s (k) + Bi,s Ti,s (k)] + b(k), i = 1, . . . , I
s=1,s6=i

PB j (k) − Pd j (k) =
√ S
2G j, j u j (k) + ∑ [G j,s R j,s (k) + B j,s T j,s (k)] + b(k), j = 1, . . . , J
s=1,s6=i

Psm (k) − Pdm (k) =

√ S
2Gm,m um (k) + ∑ [Gm,s Rm,s (k) + Bm,s Tm,s (k)] + b(k), m = 1, . . . , M
s=1,s6=i

Pwn (k) − Pdn (k) =

√ S
2Gn,n um (k) + ∑ [Gn,s Rn,s (k) + Bn,s Tn,s (k)] + b(k), n = 1, . . . , N
s=1,s6=i

− Pdr (k) =
√ S
2Gr,r ur (k) + ∑ [Gr,s Rr,s (k) + Br,s Tr,s (k)] + b(k), r = 1, . . . , S − N − M − J − I
s=1,s6=i
 Running Head Verso

√ S
Qgi (k) − Qdi (k) = − 2Bi,i ui (k) − ∑ [Bi,s Ri,s (k) − Gi,s Ti,s (k)] + b(k), i = 1, . . . , I
s=1,s6=r

QB j (k) − Qd j (k) =
√ S
− 2B j, j u j (k) − ∑ [B j,s R j,s (k) − G j,s T j,s (k)] + b(k), j = 1, . . . , J
s=1,s6=r

− Qdr (k) =
√ S
− 2Br,r ur (k) − ∑ [Br,s Rr,s (k) − Gr,s Tr,s (k)] + b(k), r = 1, . . . , S − N − M − J − I
s=1,s6=r
max
|(Yr,s (ur (k)) − us (k))| ≤ Ir,s , r, s = 1, . . . , S
2ur (k)us (k) = R2r,s (k) + Tr,s
2
(k), r, s = 1, . . . , S
Pgimin ≤ Pgi (k) ≤ Pgimax , i = 1, . . . , I
min max
PBi ≤ PBi (k) ≤ PBi , j = 1, . . . , J
Qmin max
gi ≤ Qgi (k) ≤ Qgi , i = 1, . . . , I
Qmin max
Bi ≤ QBi (k) ≤ QBi , j = 1, . . . , J
b(k), bp (k), bq (k) ≥ 0 (11.7)

It is important to remark that in Eq. (11.7) unlike in Eq. (11.6) the real and reactive
power injected by each generation unit is determined by the values of ur (k), Rr,s (k), Tr,s (k),
r, s = 1, . . . , S. Thus, these are the decision variables in such optimization problem.
Accordingly, the constraints and the cost function in Eq. (11.7) can be expressed
as a linear function of them. Furthermore, both Eq.(11.6) and Eq. (11.7) provide an
additional flexibility to the EMS. The additional flexibility consists on the possibility
of sending voltage set-points to any connected device able to regulate the voltage in
the connection point.
Minimization problem Eq. (11.6) (and as an special case minimization problem
Eq. (11.7)) corresponds to the deterministic formulation of the proposed optimization-
based EMS. The implementation of this EMS in a real micro-grid requires a per-
fect knowledge of Psm (k), Pwm (k),Pdr (k), and Qdr (k). In practice these variables
are highly uncertain and a mismatch in their estimation could lead the operation of
a micro-grid to undesirable points. To include the uncertainty in Eq. (11.6) (and
therefore in in Eq. (11.7)) the mathematical framework introduced in section 11.2
has been used. In this framework a set of possible scenarios is constructed using
the combinations of the extreme values of the uncertain variables. In this research,
unlike in [22, 21] and the references therein, these extreme values are determined
using the statistics of historical data available. Therefore, the use of prediction mod-
els or estimators is avoided. Based on the statistics of the historical data, and given
the desired sample time for the EMS, the mean value and the standard deviation are
computed per sample time. For instance, if the EMS is designed to be solved every
minute, then for each of the 1440 minutes of the day the mean µ(k) and the standard
 Running Head Recto

deviation σ (k) are computed. With the values of µ(k) and σ (k) the extreme values
are computed as: xmin (k) = µ(k) − ξ1 σ (k) and xmax (k) = µ(k) + ξ2 σ (k) with ξ1 and
ξ2 tunable parameters determined by the desired confidence level.
min (k), Pmin (k),Pmin (k), Qmin (k) denote the minimum expected value for
Let Psm wm dr dr
the real power supplied by the solar and wind power plants, and for the real and re-
active power demanded by the load centers. Let Psm max (k), Pmax (k),Pmax (k), Qmax (k)
wm dr dr
denote the maximum expected value for the real power supplied by the solar and
wind power plants, and for the real and reactive power demanded by the load cen-
ters. Note that in the minimization problem Eq. (11.6) both Psm (k) and Pwm (k) are
not being controlled. Then, an equivalent demand could be formulated to integrate
all the sources of uncertainty into a single variable. Let PLr (k) denote the equiva-
lent real power demand at node r. Then the extreme values of PLr (k) are given by:
max (k) = Pmax (k) + Pmax (k) − Pmax (k) and Pmin (k) = Pmin (k) + Pmin (k) − Pmin (k).
PLr sr wr dr Lr sr wr dr
Given these definitions, a robust EMS with transmission constraints can be formu-
lated (from Eq. (11.6)) as follows:

I J
min ∑ Cgi Pgi (k) + ∑ CB j PB j (k) + αb(k) + β bp (k) + γbq (k)
Pgi (k),PB j (k), i=1 j=1
b(k),bp (k),bq (k)
ur (k),Rr,s (k),Tr,s (k)

s.t:
I J S
∑ Pgi (k) + ∑ PB j (k) + ∑ PLrmax (k) − bp (k) = 0
i=1 j=1 r=1
I J S
∑ Pgi (k) + ∑ PB j (k) + ∑ PLrmin (k) − bp (k) = 0
i=1 j=1 r=1
I J N
∑ Qgi (k) + ∑ QB j (k) − ∑ Qmax
dn (k) − bq (k) = 0
i=1 j=1 n=1
I J N
∑ Qgi (k) + ∑ QB j (k) − ∑ Qmin
dn (k) − bq (k) = 0
i=1 j=1 n=1

max
√ S
Pgr (k) + PBr (k) + PLr (k) = 2Gr,r ur (k) + ∑ [Gr,s Rr,s (k) + Br,s Tr,s (k)] + b(k)
s=1,s6=r

min
√ S
Pgr (k) + PBr (k) + PLr (k) = 2Gr,r ur (k) + ∑ [Gr,s Rr,s (k) + Br,s Tr,s (k)] + b(k)
s=1,s6=r
√ S
Qgr (k) + QBr (k) − Qmax
dr (k) = − 2Bi,i ur (k) − ∑ [Br,s Rr,s (k) − Gr,s Tr,s (k)] + b(k)
s=1,s6=r
√ S
Qgr (k) + QBr (k) − Qmin
dr (k) = − 2Bi,i ur (k) − ∑ [Br,s Rr,s (k) − Gr,s Tr,s (k)] + b(k)
s=1,s6=r
 Running Head Verso

max
|(Yr,s (ur (k)) − us (k))| ≤ Ir,s
2ur (k)us (k) = R2r,s (k) + Tr,s
2
(k)
Pgimin ≤ Pgi (k) ≤ Pgimax
min max
PBi ≤ PBi (k) ≤ PBi
Qmin max
gi ≤ Qgi (k) ≤ Qgi
Qmin max
Bi ≤ QBi (k) ≤ QBi
b(k), bp (k), bq (k) ≥ 0 (11.8)

Until here, this section has been focused on the formulation of electrical con-
straints. But, as presented in [16] some other features could be added into the
proposed EMS. For example demand-side management signals and programmable
loads. Since both of them are deterministic and the associated constraints are lin-
ear (or at least convex -see [16] for some details-), adding these elements would
not significantly increase the compexity of the EMS. The main expected change
is that in addition to real and reactive power set-points and voltage set-points, the
EMS should send consumption signals to the customers and ON/OFF signals for the
programmable loads. In this sense, although the EMS problem formulated as the
minimization problem Eq. (11.6), and even as the minimization problem Eq. (11.7),
seems to be more complex than those presented in, e.g., [16, 22, 21, 15], the formu-
lation in Eq. (11.6) provides more flexibility to the grid from the control viewpoint.
Indeed, using Eq. (11.6) it is possible to define real and reactive power set-points but
also voltage set-points. Then a better control of the behavior of the grid is expected.
Furthermore, since transmission constraints are considered in the formulation, it is
possible to guarantee that the power injections computed by the EMS can be trans-
ferred from the sources to the load centers. Next the case study is described and the
simulation results are analyzed.

11.4 Case Study and Simulation Results

Results obtained in previous studies [28, 29, 22, 21] using stochastic and robust
programming frameworks to implement an EMS in micro-grid applications indicated
that the use of EMS approaches in which only predictions to deal with the uncertainty
[16] resulted in a lower grid performance from the security viewpoint. According to
the results reported in [21] robust EMS assigned larger power reserves to mitigate the
negative effects of the intermittency of wind generation. The present study proposed
a novel optimization-based robust EMS with transmission constraints. The novelty
of the proposed EMS recast in both the addition of transmission constraints and
the avoidance prediction models to determine the time-variant uncertainty set used
to represent the uncertain variables. The performance of the proposed EMS was
evaluated using a modified version of the IEEE 9-nodes power system benchmark. In
the modified version of the IEEE 9-nodes power system benchmark, the generation
 Running Head Recto

unit originally connected to the node 1 (which is the slack machine) was changed
by an equivalent wind generator (here equivalent means same terminal voltage and
maximum capacity). Figure 11.1 presents a single line diagram of the power system
used as test-bench.

Figure 11.1: Single-line diagram of the EEE 9-nodes power system benchmark
(taken from: http://www.al-roomi.org/power-flow/9-bus-system)

The system shown in Figure 11.1 was implemented in Matlab/Simulink software.

The parameters used in [18, 19] were considered to implement the simulation model.
To implement the simulation model the following considerations were done: the
generation units were modeled through sub-transient models (7th order model); the
transmission lines were modeled by means of the π lumped-parameters model; the
load centers were modeled as constant impedance with nominal powers of: 125+ j50
MVA for the load center connected to node 5, 90 + j30 MVA for the load center
connected to node 6 and 100 + j35 MVA for the load center connected to node 8;
and the wind power generator was modeled as a voltage source converter, using an
average model to represent the converter dynamics (this model includes the dynamic
response of both the voltage and the current inner control loops). A one month length
wind power signal measured in Atacama desert (north of Chile), with a resolution
of 1 min, was scaled up to match the capacity equivalence between the original
an the wind power station. This signal was used to generate the current reference
signal required by the inner current loop of the converter. The simulation model
was implemented with a fixed integration step of 50 µs (numerical method ODE4 of
Matlab/Simulink) to solve the differential equations. Since the EMS is discrete by
nature, it was executed every 10 s.
Due to the differences between the execution times rate-transition blocks were
added to emulate the communication delay that appears in real applications. Further-
 Running Head Verso

more, since the load centers were modeled as constant impedance only uncertainty in
the wind generation was considered in the EMS. The deterministic EMS Eq. (11.7)
was implemented for comparison purposes. Figure 11.2 presents the wind power
signal used during the simulation and its extreme values derived from the statistics
of the signal. A base-power of 255.47 MW was used for the normalization of the
signal. As can be seen, the power supplied by the wind generator (blue solid line
) is always bounded by the extreme variables derived from the statistics of the sig-
nal, which is consistent with the procedure used to derive the interval. In this case
ξ1 = 0.3 and ξ2 = 0.3. In the deterministic the measured wind power signal was used
to define the power injected by the wind power generator. In the robust EMS, both
extreme values were used to represent the power injected by the wind power genera-
tor. Since these values were obtained off-line prediction models were not required to
implement the robust EMS. Note that the robust EMS never exactly knows the real
value of the wind power injected to the system.

0.9

0.8

0.7
Wind power [p.u]

0.6

0.5

0.4

0.3

0.2

0.1

0
0 500 1000 1500 2000 2500 3000 3500 4000
Time [s]

Figure 11.2: Wind power time-series used in the simulation (blue solid line), and
extreme values derived from the statistics of the available data (black dashed lines)

Figure 11.3 shows the results obtained with both the deterministic and the robust
EMS. In this case equal generation costs were assigned to generation units 2 and
1.5 1
3. These values were: Cq2 = 1.92 and Cq3 = 1.28 , since in a base of 100 MW their
maximum capacities are 1.92 and 1.28 respectively. As it can be seen in the top and
bottom left-hand-side panels, the deterministic EMS makes an under-estimation of
the real power that generation units 2 and 3 should supply. Consequently, a differ-
ence between the scheduled (black dashed line) and the injected (blue solid line) real
power by each generation appears throughout the entire simulation. By contrast, the
robust EMS adequately estimated the real power that generation units 2 and 3 should
supply (top and bottom right-hand side panels). Accordingly, the difference between
scheduled and supplied real power does not appear. Note that in both cases there
 Running Head Recto

exists a delay in the scheduling of the generation units. The delay is about 10 s and
is a consequence of the discrete nature of both EMS. In the literature several alter-
natives to compensate such a delay have been proposed (e.g., the approach reported
in [?] and the references therein). Any of these alternatives could be combined with
the EMS proposed in this chapter to enhance the performance of the micro-grid as a
whole.

1 1

0.8 0.8
Pg 2 [p.u]

Pg 2 [p.u]
0.6 0.6

0.4 0.4

0.2 0.2

0 0
0 1000 2000 3000 4000 0 1000 2000 3000 4000
Time [s] Time [s]

1 1

0.8 0.8
Pg 3 [p.u]

Pg 3 [p.u]

0.6 0.6

0.4 0.4

0.2 0.2

0 0
0 1000 2000 3000 4000 0 1000 2000 3000 4000
Time [s] Time [s]

Figure 11.3: Scheduled (dark dashed lines) and supplied (blue solid line) real power
for generation units 2 and 3 throughout the simulations. The left-hand side panels
correspond to the results obtained with the deterministic EMS. The right-hand- side
panels correspond to the results obtained with the proposed robust EMS.

A clear difference between the behavior of the system with the deterministic and
the proposed EMS is evident from Figure 11.3. Such a difference is reinforced by
the the behavior of the frequency of the system throughout the simulation. Figure
11.4 presents the frequency measured in all generation nodes (top panels). It can
be observed that during the entire simulation the frequency measured in the gener-
ation nodes was the same (for the three generators) despite the changes in the wind
power. Moreover, it also can be observed that, in the case of the deterministic EMS,
the difference between the scheduled and the supplied real power resulted in a fre-
quency deviation in steady-state throughout the simulation (left-hand side top panel).
Indeed, after the transients produced by the fast changes in the wind power the fre-
 Running Head Verso

quency always stabilized over 1 p.u. This phenomena did not appear when the pro-
posed robust EMS was used (see right-hand side top panel). It is important to remark

1.06 1.06

1.04 1.04

1.02 1.02
f [p.u]

f [p.u]
1 1

0.98 0.98

0.96 0.96

0.94 0.94
0 1000 2000 3000 4000 0 1000 2000 3000 4000
Time [s] Time [s]

1.2 1.2
ExTime [s]

ExTime [s]

1 1

0.8 0.8

0.6 0.6

0 1000 2000 3000 4000 0 1000 2000 3000 4000

Time [s] Time [s]

Figure 11.4: System frequency (top panels) and execution time (bottom panels)
throughout the simulation. The frequency was measured in the generation nodes,
namely, in nodes 1, 2, and 3. As it can be seen all frequencies are overlapped show-
ing that the frequency along the system is the same.

that voltage and reactive power results have not been included in this section since
significative differences were not evidenced. Instead of these variables, the execution
time of each EMS was included in Figure 11.4. Throughout this chapter and special
emphasis has been given to the possibility of efficiently solving the minimization
problem Eq. (11.7). The execution time was computed using the TIC-TOC function
of Matlab/Simulink software. As it can be noticed, in both cases the execution time
was always less than the sample time of 10 s, being lower for the deterministic EMS.
Also, in both cases the execution time was about 0.8 s in average. This fact puts
forward the low computational burden of the solution of the EMS analyzed here.
The results presented in Figures 11.3 and 11.4 were obtained considering the
same generation cost for both generation units. However, these results could change
in these costs are assumed different. Indeed, one of the main issues in EMS ap-
proaches is how to select the generation costs of each technology so that the result-
ing optimization model adequately represents the objectives of the EMS. To illustrate
 Running Head Recto

1
this statement, the generation costs were changed for the following values: Cq2 = 1.92
1
and Cq3 = 1.28 . With this selection the cost of using the generation unit 2 to supply
the demand is lower than the cost of using the generation unit 3 for the same purpose.
Figure 11.5 presents the results obtained with equal generation costs for both gener-
ation units (left-hand side panels) and the results obtained with different generation
costs (right-hand side panels). As it can be noticed, in left-hand side panels both
generation units are scheduled so that they evenly share the demand. By contrast,
in the right-hand side panels the EMS tends to use generation unit 2 more than gen-
eration unit 3 to supply the demand. It is important to notice that in the time-range
(1000s, 3300s) the EMS assigned almost 0 MW of real power to generation unit 2.
This range corresponded to the time-range with the maximum wind power supply.
This condition might correspond in real applications to turn-off the diesel generator
and use only the battery bank to supply the demand.

1 1

0.8 0.8
Pg 2 [p.u]

Pg 2 [p.u]

0.6 0.6

0.4 0.4

0.2 0.2

0 0
0 1000 2000 3000 4000 0 1000 2000 3000 4000
Time [s] Time [s]

1 1

0.8 0.8
Pg 3 [p.u]

Pg 3 [p.u]

0.6 0.6

0.4 0.4

0.2 0.2

0 0
0 1000 2000 3000 4000 0 1000 2000 3000 4000
Time [s] Time [s]

Figure 11.5: System frequency (top panels) and execution time (bottom panels)
throughout the simulation. The frequency was measured in the generation nodes,
namely, in nodes 1, 2, and 3. As it can be seen all frequencies are overlapped show-
ing that the frequency along the system is the same.

The results presented throughout this section suggest that the proposed frame-
work for the development of robust EMS is a suitable alternative to include trans-
mission constraints into the formulation of the EMS problem, without significantly
 Running Head Verso

increasing its complexity, without compromising its application in real micro-grids,

and avoiding the use of prediction models. Furthermore, the use of the proposed
robust EMS allows having a better control of the variables of the grid, namely, real
and reactive powers but also voltages and system frequency.

11.5 Concluding Remarks

Prior work has documented the importance of the robustness of EMS in micro-grid
applications in guaranteeing the security in the operation of the grid and in main-
taining an optimal economic operation; the authors in [28, 29], for example, report
the improvements in the operation of a micro-grid achieved when using different
alternatives to deal with the uncertainty in EMS formulations. However, these stud-
ies have either been affected by the performance of prediction models and have not
included transmission constraints. Then it is not possible to guarantee that the sched-
uled powers can be transferred from the sources to the demand centers in a proper
way. In this study it has been proposed a novel formulation for robust optimization-
based EMS for micro-grid applications, in which prediction models are avoided, and
where transmission constraints are included in the formulation. To represent the un-
certainty, the statistics of the available data of the uncertain variables is used to derive
a time-variant uncertainty set. It has been found that in all cases analyzed the pro-
posed robust EMS provides a better scheduling of the generation units. For compari-
son purposes a deterministic version of the proposed EMS has been used. Moreover,
it has also found that including transmission constraints into the EMS improves the
performance of the micro-grid and adds more flexibility to the EMS without signif-
icantly increasing its complexity. These results extend those of [22, 16], confirming
that improving the micro-grid models used in the formulation of EMS approaches
tend to increase the type of elements to be able to be controlled by the EMS and to
extend the variables that can be directly/indirectly regulated by the EMS. In addition,
the formulation presented in this chapter is independent of the topology of the micro-
grid and the type of generation units. This study therefore indicated that the benefits
gained with the implementation of the proposed EMS could be extended to a broad
range type of micro-grids. Future work should therefore include implementing the
proposed EMS in different types of micro-grids.

Acknowledgement

This research was supported by the FONDAP/CONiCYT project Solar Energy Re-
search Center, SERC-Chile, grant number 15110019.
Bibliography

[1] Walid Al-Gherwi, Hector Budman, and Ali Elkamel. A robust distributed
model predictive control algorithm. Journal of Process Control, 21(8):1127–
1137, 2011.

[2] Dimitris Bertsimas, Eugene Litvinov, Xu Andy Sun, Jinye Zhao, and Tongxin
Zheng. Adaptive robust optimization for the security constrained unit commit-
ment problem. Power Systems, IEEE Transactions on, 28(1):52–63, 2013.

[3] S. Boyd, L. Vandenberghe, and M. Grant. Efficient convex optimization for en-
gineering design. IFAC Symposium on Robust Control Design, Rio de Janeiro,
Brazil, 1994.

[4] Stephen Boyd and Lieven Vandenberghe. Convex Optimization. Cambridge

University Press, 2004.

[5] Jairo Espinosa, Joos Vandewalle, and Vincent Wertz. Fuzzy Logic, Identifica-
tion, and Predictive Control. Springer, 2005.

[6] N. Hatziargyriou, G. Contaxis, M. Matos, J.A.P. Lopes, G. Kariniotakis,

D. Mayer, J. Halliday, G. Dutton, P. Dokopoulos, A. Bakirtzis, J. Stefanakis,
A. Gigantidou, P. O’Donnell, D. McCoy, M. J. Fernandes, J. M S Cotrim, and
A.P. Figueira. Energy management and control of island power systems with
increased penetration from renewable sources. In Power Engineering Society
Winter Meeting, 2002. IEEE, volume 1, pages 335–339 vol.1, 2002.

[7] RA Jabr and BC Pal. Conic programming approach for static voltage stabil-
ity analysis in radial networks. IET generation, transmission & distribution,
1(2):203–208, 2007.

[8] Rabih A Jabr. Radial distribution load flow using conic programming. IEEE
transactions on power systems, 21(3):1458–1459, 2006.

[9] Rabih A Jabr. A conic quadratic format for the load flow equations of meshed
networks. IEEE Transactions on Power Systems, 22(4):2285–2286, 2007.

[10] Rabih A Jabr. Optimal power flow using an extended conic quadratic formula-
tion. IEEE transactions on power systems, 23(3):1000–1008, 2008.
 BIBLIOGRAPHY

[11] John D. Kueck, Robert H. Staunton, Solomon D. Labinov, and Brendan J.

Kirby. Microgrid energy management system. 01/2003 2003.

[12] JH a Lee and Zhenghong Yu. Worst-case formulations of model predictive con-
trol for systems with bounded parameters. Automatica, 33(5):763–781, 1997.

[13] D Limon, I Alvarado, T Alamo, and EF Camacho. Robust tube-based mpc for
tracking of constrained linear systems with additive disturbances. Journal of
Process Control, 20(3):248–260, 2010.

[14] X.H. Nie, G.H. Huang, Y.P. Li, and L. Liu. Ifrp: A hybrid interval-parameter
fuzzy robust programming approach for waste management planning under un-
certainty. Journal of Environmental Management, 84(1):1 – 11, 2007.

[15] D.E. Olivares, C.A. Canizares, and M. Kazerani. A centralized optimal energy
management system for microgrids. In Power and Energy Society General
Meeting, 2011 IEEE, pages 1–6, July 2011.

[16] R. Palma-Behnke, C. Benavides, F. Lanas, B. Severino, L. Reyes, J. Llanos, and

D. Saez. A microgrid energy management system based on the rolling horizon
strategy. Smart Grid, IEEE Transactions on, 4(2):996–1006, June 2013.

[17] Gabriele Pannocchia. Robust model predictive control with guaranteed setpoint
tracking. Journal of Process control, 14(8):927–937, 2004.

[18] A Pizano-Martı́nez, C Fuerte-Esquivel, EA Zamora-Cárdenas, and J Se-

gundo Ramı́rez. Conventional optimal power flow analysis using the matlab
optimization toolbox. ROPEC International, 2010.

[19] Peter W Sauer and MA Pai. Power system dynamics and stability. Urbana,
51:61801, 1997.

[20] Pierre OM Scokaert, James B Rawlings, and Edward S Meadows. Discrete-

time stability with perturbations: Application to model predictive control. Au-
tomatica, 33(3):463–470, 1997.

[21] Felipe Valencia, Jorge Collado, Doris Sáez, and Luis G Marı́n. Robust energy
management system for a microgrid based on a fuzzy prediction interval model.
IEEE Transactions on Smart Grid, 7(3):1486–1494, 2016.

[22] Felipe Valencia, Doris Sáez, Jorge Collado, Fernanda Ávila, Alejandro Mar-
quez, and Jairo J Espinosa. Robust energy management system based on
interval fuzzy models. IEEE Transactions on Control Systems Technology,
24(1):140–157, 2016.

[23] S. Wang and G.H. Huang. An interval-parameter two-stage stochastic fuzzy

program with type-2 membership functions: an application to water re-
sources management. Stochastic Environmental Research and Risk Assessment,
27(6):1493–1506, 2013.
 BIBLIOGRAPHY

[24] S. Wang and G.H. Huang. A two-stage mixed-integer fuzzy programming with
interval-valued membership functions approach for flood-diversion planning.
Journal of Environmental Management, 117(0):208 – 218, 2013.
[25] S. Wang, G.H. Huang, and B.T. Yang. An interval-valued fuzzy-stochastic pro-
gramming approach and its application to municipal solid waste management.
Environmental Modelling & Software, 29(1):24 – 36, 2012.
[26] Brittany Wright. A review of unit commitment. 2013.
[27] Zhe Yu, L. McLaughlin, Liyan Jia, M.C. Murphy-Hoye, A. Pratt, and Lang
Tong. Modeling and stochastic control for home energy management. In Power
and Energy Society General Meeting, 2012 IEEE, pages 1–9, July 2012.
[28] Yu Zhang, Nikolaos Gatsis, and Georgios B Giannakis. Robust energy man-
agement for microgrids with high-penetration renewables. Sustainable Energy,
IEEE Transactions on, 4(4):944–953, 2013.
[29] Binyan Zhao, Yi Shi, Xiaodai Dong, Wenpeng Luan, and J. Bornemann. Short-
term operation scheduling in renewable-powered microgrids: A duality-based
approach. Sustainable Energy, IEEE Transactions on, 5(1):209–217, Jan 2014.
[30] Chaoyue Zhao, Jianhui Wang, J.-P. Watson, and Yongpei Guan. Multi-stage
robust unit commitment considering wind and demand response uncertainties.
Power Systems, IEEE Transactions on, 28(3):2708–2717, Aug 2013.

View publication stats