IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 54, NO.

2, MARCH/APRIL 2018 1793

A Quadratic Programming Based Optimal Power and

Battery Dispatch for Grid-Connected Microgrid
Tim George Paul, Student Member, IEEE, Sheikh J. Hossain, Student Member, IEEE, Sudipta Ghosh, Member, IEEE,
Paras Mandal, Senior Member, IEEE, and Sukumar Kamalasadan , Senior Member, IEEE

max
Abstract—In this paper, the concept of end-user-driven mi- Pbatt Maximum power rating of the battery.
crogrid is introduced, and an economic dispatch framework is Pload (t) Total load at instant t.
proposed for optimal power and battery scheduling of such a Ppv (t) Photovoltaic generation at instant t.
microgrid. The main advantage is that unlike other centralized
approaches reported in the literature, the proposed architecture Pgrid (t) Power demand from the electric grid
min
provides end-user flexibility, optimally manages the battery mini- Pgrid Minimum power demand from the electric grid
max
mizing the power required from the grid, and at the same time en- Pgrid Maximum power demand from the electric grid
sures grid-level reliability. The proposed optimal scheduling is first P2 (t) Total power from photovoltaic (PV) plus the battery.
validated on a grid-connected microgrid, including a photovoltaic
SOC State of charge at the battery.
(PV) farm and a battery, considering the grid as a dispatchable
power source. Then, the architecture is compared with one of the MG Microgrid.
existing centralized optimization approaches. To demonstrate the Aeq Coefficient matrix of equality constraints.
applicability and scalability, the architecture is further evaluated A Coefficient matrix of inequality constraints.
on a modified IEEE 33 bus distribution feeder, including multiple beq Vector for constant coefficients.
microgrids suitable for real-time implementation. The test results
b Vector for constant coefficients of inequality con-
indicate that the approach is optimal, scalable, feasible, improves
feeder reliability, and provides maximum utilization of the PV straints.
arrays. H Symmetric matrix describing the coefficients of the
quadratic terms.
Index Terms—Battery power dispatch, economic dispatch,
grid-connected end-user-driven microgrid, optimal scheduling, C Vector of linear terms.
quadratic programming. l Vector of lower bound of variables.
u Vector of upper bound of variables.
x Solution vector.
NOMENCLATURE
I. INTRODUCTION
N Total number of samples.
N ORDER to satisfy the increased electricity demand and
t
E(t)
dt
Time instant.
Instantaneous energy of the battery.
Time interval.
I global environmental regulations, recently there is a move to-
ward realizing a nonfossil fuel based power grid. Grid-connected
Pbatt (t) Instantaneous battery dispatch value. renewable power systems have gained considerable interest in
min
Pbatt Minimum power rating of the battery. this respect. One of the main renewable energy based energy re-
sources is PV systems. As opposed to wind farms that are mainly
controlled and regulated by the utility, PV farms are distributed
Manuscript received May 12, 2017; revised August 15, 2017 and November across the regulated and nonregulated business. A major chal-
6, 2017; accepted November 9, 2017. Date of publication December 11, 2017; lenge for PV generation is in managing the intermittent energy
date of current version March 19, 2018. Paper 2017-IACC-0515.R2, presented
at the 2016 IEEE Industry Applications Society Annual Meeting, Portland, OR, production with varying power demand. On the source side,
USA, Oct. 2–6, and approved for publication in the IEEE TRANSACTIONS ON one way of mitigating this intermittency is to provide additional
INDUSTRY APPLICATIONS by the Industrial Automation and Control Committee energy storage system (ESS) to the PV farm. This is mainly
of the IEEE Industry Applications Society. This work was supported in part
by the Energy Production Infrastructure Center and in part by National Sci- provided using battery ESS (BESS). Especially in the power
ence Foundation under Grant ECS-1309911. (Corresponding author: Sukumar distribution setup, such a discrete energy system consisting of
Kamalasadan.) distributed energy resources, including storage, generation, and
T. G. Paul, S. J. Hossain, S. Ghosh, and S. Kamalasadan are with the Energy
Production Infrastructure Center, and the Department of Electrical and Com- load management, can be developed, namely microgrid. A mi-
puter Engineering, University of North Carolina at Charlotte, Charlotte, NC crogrid can operate in parallel with another microgrid as well as
28223 USA (e-mail: tpaul1@uncc.edu; shossai1@uncc.edu; sghosh9@uncc. independently from the main grid or connected to the grid. Sev-
edu; skamalas@uncc.edu).
P. Mandal is with the Department of Electrical and Computer Engineering, eral approaches have been proposed for energy management
University of Texas at El Paso, El Paso, TX 79968 USA (e-mail: pmandal@ and control of microgrid with renewable sources and storage
utep.edu). especially with PV [1], [2]. Most of such previous approaches
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org. discussed regulated (controlled by the utility) management of
Digital Object Identifier 10.1109/TIA.2017.2782671 microgrids.

0093-9994 © 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications standards/publications/rights/index.html for more information.

Authorized licensed use limited to: UNIVERSIDADE FEDERAL DA PARAIBA. Downloaded on February 04,2023 at 14:06:30 UTC from IEEE Xplore. Restrictions apply.
 1794 IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 54, NO. 2, MARCH/APRIL 2018

As opposed to utility-driven microgrids, end-user-driven mi- problem dynamic, i.e., it is time dependent. Different optimiza-
crogrids are managed by the end-users. Such microgrids nor- tion techniques have been used to formulate this type of dynamic
mally exist at the low-voltage end of the power distribution scheduling problem [8], [11]–[19]. In [12], [13]–[16], [20], and
feeder (either at 120 V or at 480 V level) and connected to [21], dynamic programming is used to solve the dispatch prob-
the utility through inverters. Normally, these systems are seen lem. In [22] and [23], a model predictive control based controller
as rooftops- or community-level installations. Such microgrids is proposed to control the output of BESS that is used in micro-
can provide higher reliability, optimal active power balance (a grid management. In [24], a distributed EMS is implemented
mechanism in which the local power is utilized to the maximum based on alternating direction method of multipliers where dis-
and at the same time grid constraints are satisfied), and enhance patch information is passed between generators and BESS with
customer participation. However, the main challenges of such a goal to meet the total demand. In [25], a distributed formu-
installations are the visibility issue related to the power usage, lation of the economic dispatch is shown where the goal is to
as these installations are mostly behind the meter, and the issues maximize the profit for each energy resource. The main issues
related to the usage of the electricity based on the need and with distributed framework are that maximum benefits for the
comfort of the end-user versus economic and reliable operation utility are not always satisfied and it is also difficult to maintain
of the utility. grid-level stability.
Thus, a coordinating energy management architecture that In our previous work [26], we proposed a dynamic optimiza-
considers the end-user need and economic and reliable opera- tion approach that reduces the grid operational cost, and at the
tion of the grid is essential [3]–[7]. The coordination feature is same time, maximizes the PV output power based on optimal
mainly to achieve an integrated decentralized and centralized scheduling of the battery. This paper is an expanded version
dispatch in spite of the respective modes of operation of the of the work presented in [26], where the proposed quadratic
microgrid. Such microgrid energy management systems (EMS) programming based optimal power scheduling approach is en-
should minimize the operation costs considering the reliability hanced to provide optimal dispatch of the battery that enables
of the grid, and at the same time, allow end-user flexibilities, grid reliability, economic dispatch, and end-user flexibilites. The
such as using the microgrid energy resources based on the end- evaluation is done on utility-scale real-life grid models.
user needs. An economic dispatch approach can perform optimal The main advantages of the proposed approach are follows.
power and battery scheduling considering utility-side and end- 1) The methodology provides economic dispatch by only us-
user-side objectives. Utility-side applications could focus on ing the microgrid tertiary control for calculating the active
optimizing properties of the microgrid output for distribution power set points such that the microgrid uses minimum
upgrade deferral, which means delaying the utility investments power from the grid and the battery dispatch is optimized
in distribution system upgrades, transmission support, and oth- based on short-term PV generation data.
ers, whereas the end-user side’s objective is to maintain the 2) The method works effectively with multiple microgrids,
demand flexibility. and the grid reliability is maintained.
Most of the earlier work related to microgrid management 3) The approach is scalable and can be implemented in real
considered battery constraints to complete a centralized grid systems interconnected with power distribution grid.
power flow optimization problem [8]–[10] such that it bene- 4) The approach provides end-user flexibility and at the same
fits the utility. Such centralized model collects all the necessary time ensures grid requirements.
information for microgrid scheduling and performs centralized This paper is organized as follows. In Section II, theoretical
operation and control. The main drawbacks of the centralized framework and problem formulation is discussed. This includes
scheme are reduced flexibility for the end-user and in adding new the fixed and predictive horizon methodologies. In Section III,
components, and extensive computational requirements. For ex- simulation results for the proof-of-concept and integrated multi-
ample, if a new energy resource (device) is added to the grid or ple microgrids connected to a distribution feeder are presented.
is out for maintenance then the EMS operation is interrupted to Section IV concludes this paper.
add/remove device specific constraints from the algorithm that
implements the EMS. Furthermore, any change of a particular
unit dispatch would require a new optimal power flow to be II. ECONOMIC SCHEDULING OF END-USER
run for the system, because they are all integrated in a single -DRIVEN MICROGRID
optimization problem. In addition, the dynamic nature of en- A conventional framework for a grid-connected microgrid
ergy storage dispatch creates time-domain couplings among all economic dispatch problem can be found in [27]–[32]. In such
decision variables as well as stochastic variables like loads and framework, a centralized economic dispatch optimization is
renewable power. This poses challenges for the computational performed integrating microgrid devices, such as BESS and
complexity of centralized formulation. renewable energy resources. In this paper, the concept of end-
The issues with centralized formulation can be avoided with user-driven microgrid is introduced for the first time. Then, an
decentralized or distributed formulations. In the distributed optimization framework is proposed to satisfy user-defined load
scheme, each component of the microgrid is considered separate (flexibility), maximizing the renewable energy based power ex-
and makes its own decisions. The optimal schedule is obtained traction, and at the same time ensuring minimum power import
by iterative data transfers among the components. Since the from the grid while honoring grid reliability. In this section,
microgrid is integrated with storage it makes the optimization we will first define the end-user-driven microgrid, and provide

Authorized licensed use limited to: UNIVERSIDADE FEDERAL DA PARAIBA. Downloaded on February 04,2023 at 14:06:30 UTC from IEEE Xplore. Restrictions apply.
PAUL et al.: QUADRATIC PROGRAMMING BASED OPTIMAL POWER AND BATTERY DISPATCH FOR GRID-CONNECTED MICROGRID 1795

the proposed dynamic economic dispatch optimization frame-

work including the formulation of the objective function and
constraints. Furthermore, we will discuss the adaptivity of the
method to fulfill end-user and utility requirements and show the
closed-loop implementation flowchart.

A. Customer-Driven Microgrid and End-User-Driven

Microgrid
Electric power microgrids have been broadly classified as
utility microgrids, industrial/commercial microgrids, and re-
mote/residential microgrids [5]. In recent years, a new paradigm
of microgrids is being introduced to bring near real-time infor-
mation and control at the customer end. This type of microgrids Fig. 1. Grid architecture for dynamic dispatch.
is termed customer-driven microgrid [5]. A customer-driven mi-
crogrid is a type of microgrid that can import or export real and
reactive power when operated in a grid-connected mode and
can also operate in an islanded mode [4]. The customer-driven also the past states of the system. In static economic dispatch,
microgrid provides a complete framework for the design, opera- the dispatch solutions in a particular time stage only depend
tion, and economics of the microgrids from the customer stand- on the criteria of that time stage and are independent from the
point. In this paper, the concept of end-user-driven microgrid dispatch solutions at other time stages.
is introduced. The characteristics of end-user-driven microgrid
are as follows. C. Proposed Dynamic Economic Dispatch
1) End-user should be able to drive, control, and manage the For the proposed dynamic economic dispatch, a quadratic
microgrid. programming based approach is used to optimize grid cost,
2) End-user-driven microgrid should be able to sell power to which is considered quadratic with linear constraints. Details of
the utility. a quadratic programming approach is shown in Fig. 2
3) It should be able to share active or reactive power with the
utility grid. 1 T
· x · H · x + CT · x (1)
4) It should be able to work in the islanded mode. 2
The main difference between customer-driven and end-user- A·x≤b (2)
driven microgrids is that in the former type, customer can con-
sume or sell power to another entity apart from the utility grid; Aeq · x ≤ beq (3)
on the contrary, in the latter type, end-user can consume and l ≤ x ≤ u. (4)
share power only with the utility grid. As more distributed en-
ergy resources are connected to the end-user, the necessity for Section II-D discusses the formulation of the objective func-
optimal management of distributed resources and a framework tion and constraints, Section II-E discusses the modification of
for reliable operation of these microgrids with the utility grid is the formulation to enable a predictive horizon approach, and
becoming increasingly more important. Section II-F illustrates the implementation procedure.

B. Overview of the Microgrid Economic Dispatch D. Formulation of Objective Function and Constraints for the
Consider an end-user-driven microgrid shown in Fig. 1, which Proposed Approach
includes a PV farm, local load, and a battery energy storage con- Based on (1)–(4), an objective function, as shown in (5), is
nected to the grid. Let P2 represents the total power that is the proposed for dynamic optimization. It includes the cost function
algebraic sum of the PV plus the battery. The objective of the for the grid and battery and other sources available in the micro-
economic dispatch is to minimize the cost of the power from grid. Considering a quadratic and convex function, the overall
the grid at the point of common coupling in this case P2 , and formulation can be represented as
at the same time, dispatch the battery for optimal microgrid
scheduling. The optimization problem is formulated for a full- 
T

day operation of the microgrid. The microgrid economic dis- [F (Pgrid (t))] (5)
patch has a dynamic formulation due to the presence of energy t=1

storage, which is a time-dependent constraint with power and Pgrid (t) + P2 (t) = Pload (t) (6)
energy limits. This dynamic optimization problem thus pro-
P2 (t) + Pbatt (t) = Ppv (t) (7)
vides the dispatch solution at every time stage dependent on
the solution of all other time stages. For example, in a dynamic Pgrid min
(t) ≤ Pgrid (t) ≤ Pgrid max
(t) (8)
economic dispatch for a daily operation, the dispatch solution
at noon will depend on the state of the system at that instant and P2 m in (t) ≤ P2 (t) ≤ P2 m ax (t) (9)

Authorized licensed use limited to: UNIVERSIDADE FEDERAL DA PARAIBA. Downloaded on February 04,2023 at 14:06:30 UTC from IEEE Xplore. Restrictions apply.
 1796 IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 54, NO. 2, MARCH/APRIL 2018

load, respectively, at any given time. Equation (6) implies that

the total power supplied by the PV and battery plus the grid must
be equal to the load. The constraint in (7) implies that the sum
of the battery charging or discharging power and the power sup-
plied directly to the load from the PV is equal to the PV output at
that time. During the period of low PV power generation, based
on (6), P2 (t) is increased to match the bus load and Pgrid (t) is
decreased. However, as the PV generation is low, the Pbatt (t)
is decreased inheriting a negative sign to satisfy (7). As the PV
generation increases and if excess generation is available after
satisfying the load, then Pbatt (t) is increased to satisfy (7), and
hence, the battery charges. All the other dispatch variables are
constrained by minimum and maximum values, as specified by
(8)–(11). The algorithmic flow chart is represented in Fig. 2.
1) Formation of the Bounded Constraints: Let us consider
the bounded constraints as shown in (12)–(14). The lower bound
and upper bound vectors are represented as shown and then
concatenated to form the vectors as required
     
Pgrid min ≤ Pgrid (t) ≤ Pgrid max (12)
N ×1 N ×1 N ×1
     
P2 m in
≤ P2 (t) ≤ P2 m ax (13)
N ×1 N ×1 N ×1
     
Pbatt min ≤ Pbatt (t) ≤ Pbatt max . (14)
N ×1 N ×1 N ×1

Here, N denotes the total number of samples or time scale

considered. Since the solution vector consists of three variables
the size becomes 3N .
2) Formation of the Equality Constraints: There are two
equality constraints in the above-mentioned problem, as men-
tioned in (6) and (7). A diagonal identity matrix is considered to
implement the above-mentioned constraint. Such an Aeq matrix
is formed as
⎡⎡ ⎤⎡ ⎤⎡ ⎤⎤
1 ... 0 1 ... 0 0 ... 0
⎢ ⎢ .. . . . ⎥⎢ . . . . ⎥⎢ . . . . ⎥⎥
Fig. 2. Flow chart of the dynamic optimization algorithm. ⎢⎣ . . .. ⎦⎣ .. . .. ⎦⎣ .. . .. ⎦ ⎥
⎢ ⎥
⎢ 0 ... 1 ⎥
⎢⎡ ⎤⎡ 0 . . . 1 ⎤⎡ 0 . . . 0 ⎤ ⎥ (15)
⎢ 0 ... 0 1 . . . 0 1 . . . 0 ⎥
⎢ ⎥
⎢⎢ . . . ⎥⎢ . . . ⎥⎢ . . . ⎥ ⎥
Pbatt min (t) ≤ Pbatt (t) ≤ Pbatt max (t) (10) ⎣ ⎣ .. . . .. ⎦⎣ .. . . .. ⎦⎣ .. . . .. ⎦ ⎦
0 ... 0 0 ... 1 0 ... 1

N
  
E m in (t) ≤ E0 + Pbatt (t) · dt ≤ E m ax (t) (11) 2N ×3N
t=1
where the structure of the solution vector x is given by
where T is the time horizon of the optimization problem, and ⎡ ⎤
N is the total number of samples. Pgrid (t)
⎢  N ×1  ⎥
It is worth noting that the PV and battery operating costs are ⎢ P (t) ⎥
⎢ 2 ⎥
not incorporated here. This is due to the fact that as a rate pay ⎢  N ×1  ⎥ . (16)
⎣ ⎦
incentive case, the end-user battery cost is considered always Pbatt (t)
lower than the grid power, and thus, the cost is insignificant. This N ×1
  
is proven in Section III. The battery is modeled with minimum 3N ×1
and maximum charging and discharging limits. The constraints Similarly, the beq matrix is formed by
are formulated such that the load demand is to be met by the ⎡ ⎤
PV farm output. The economic dispatch problem is to deter- Pload (t)
mine the optimum dispatch of the battery at any given time that ⎢  N ×1  ⎥
⎣ P (t) ⎦ . (17)
minimizes the cost while satisfying the demand and operating pv
N ×1
limits. Pgrid (t), P2 (t), and Pbatt (t) are the control variables   
representing power flows from the grid, PV, and battery to the 2N ×1

Authorized licensed use limited to: UNIVERSIDADE FEDERAL DA PARAIBA. Downloaded on February 04,2023 at 14:06:30 UTC from IEEE Xplore. Restrictions apply.
PAUL et al.: QUADRATIC PROGRAMMING BASED OPTIMAL POWER AND BATTERY DISPATCH FOR GRID-CONNECTED MICROGRID 1797

3) Formation of the Inequality Constraints: The inequality

constraint in (11) is separated into lower and upper limits. The
energy stored in the battery at each instant of time is dependent
on the charge stored in the battery at the previous instant of
time and the (charge/discharge) Pbatt (t) solution, so we need
to sum the Pbatt (t) solution of previous instants of time. The
above-mentioned constraint can be divided into two inequality
constraints. Upper limit of inequality can be written as


N
(Pbatt (t) ∗ dt) ≤ E m ax − E0 (18)
t=1

where the A matrix as shown in (2) is formed by concatenating

matrices (19) and (24)
⎡⎡ ⎤⎡ ⎤⎡ ⎤⎤
0 ... 0 0 ... 0 1 ∗ dt . . . 0
⎢ ⎢ .. . . . ⎥⎢ . . . . ⎥⎢ . .. .. ⎥ ⎥
⎢⎣ . . .. ⎦⎣ .. . .. ⎦⎣ .. . . ⎦⎥
⎣ ⎦.
0 ... 0 0 ... 0 1 ∗ dt . . . 1 ∗ dt
N ×N N ×N N ×N
  
N ×3N

(19)
Fig. 3. Flow chart for the predictive horizon based optimization.
The b1 matrix is formed as
 
Emax − E0 . (20) E. Predictive Horizon Based Optimization
N ×1
To illustrate the effect of moving time horizon, a predictive
Lower limit of inequality can be written as horizon based optimization is designed that modifies the opti-
mization discussed in the previous section. The concept is based

N
on receding horizon control [27], [33], [34]. Fixed horizon op-
E m in
≤ E0 + (Pbatt · dt). (21)
timization leads to a dispatch schedule, which begins at the
t=1
current time and ends at some future time. Thus, it does not
By converting the above-mentioned equation into a standard have information regarding the future PV generation, and thus,
form, we get the optimal value is not dynamically proven to be optimized.
In the predictive horizon approach, PV generation is used for

N
E m in − E0 ≤ (Pbatt · dt) (22) a moving horizon ensuring cost reduction of power from the
t=1
grid over that horizon. Fig. 3 shows the algorithm for predictive
horizon based optimization for a day (24 h). In this method,

N
dynamic optimization is performed at each time step and only
−E m in + E0 ≤ − (Pbatt · dt). (23)
the first value of the optimization is used in the current step. For
t=1
next time step, the window is moved in time and optimization is
The A2 and b2 matrices are formed as performed again. In other words, this method constantly looks
⎡⎡ ⎤⎡ ⎤⎡ ⎤⎤ for a time horizon in near future and can adjust according to the
0 ... 0 0 ... 0 −1 ∗ dt ... 0 forecast errors due to passing of clouds or other uncertainties. It
⎢ ⎢ .. .. .. ⎥⎢ .. . . .. ⎥⎢ .. .. .. ⎥⎥
⎢⎣ . . . ⎦⎣ . . . ⎦⎣ . . . ⎦⎥ can be seen that the proposed approach can provide load sched-
⎣ ⎦
0 ... 0 0 ... 0 −1 ∗ dt ... −1 ∗ dt ule set by the end-user (flexibility) and the approach maximizes
N ×N N ×N N ×N the local energy resource usage.
  
N ×3N
(24) F. Framework for Operation With Utility Power Grid
 
E0 − Emin . (25) In order to apply the proposed method to a grid-connected
N ×1 end-user-driven microgrid, security constraints of the intercon-
Concatenating the b1 and b2 matrices to form the b matrix as nected grid needs to be considered. In the previous section,
shown in (2) we discuss how the local optimization is performed consider-
    ing end-user flexibility and local economic benefits. To enable
A1 b1 grid-level reliability and stability, it has to be made sure that
A= ,b = . (26)
A2 b2 the active power imported or exported to the utility from the
2N ×3N 2N ×1 microgrid satisfies the voltage security constraints of the grid.

Authorized licensed use limited to: UNIVERSIDADE FEDERAL DA PARAIBA. Downloaded on February 04,2023 at 14:06:30 UTC from IEEE Xplore. Restrictions apply.
 1798 IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 54, NO. 2, MARCH/APRIL 2018

Fig. 5. Proof of concept system representation.

system (DMS) function every 15–30 min. This closed-loop inte-

gration thus allows grid reliability and at the same time provide
scheduling capability for the grid based on grid-level applica-
tions
    
ΔP JPθ Jpv Δθ
= (27)
ΔQ JQ θ JQ V ΔV
ΔV = JR −1 · ΔP (28)
JR = [Jpv − JP θ J−1
Q θ JQ V ] (29)
ΔVweak
SVweakj = ,j = 1 : m (30)
ΔPj
ΔVstrong
SVstrongj = ,j = 1 : m (31)
ΔPj
1
P minj = · (Vmax − vj ) − Pj , j = 1 : m (32)
SVstrongj
Fig. 4. Proposed overall architecture for operation of end-users-driven micro- 1
grids with utility power grid. P maxj = Pj + · (Vj − vmin ), j = 1 : m (33)
SVweakj
where ΔP is the vector active power deviation for the microgrid-
To ensure that this condition is satisfied, a framework is pro- connected buses, ΔQ is the vector reactive power deviation
posed that demonstrates the reliable operation of the utility grid for the microgrid-connected buses, m is the total number of
with the local optimization of end-user-driven microgrids. Fig. 4 buses where microgrid is connected, SVweakj is the sensitivity
illustrates the proposed framework for integrated grid operation. of weak bus voltage with respect to the active power of jth
In this approach, first, a steady-state analysis (e.g., power microgrid-connected bus, SVstrongj is the sensitivity of strong
flow) is performed at the utility grid to find out the weak and bus voltage with respect to the active power of jth microgrid-
strong voltage buses in the network. Then, using the Jacobian connected bus, P maxj is the maximum power allowed at the
matrix, voltage sensitivities of the weak and strong voltage bus jth microgrid-connected bus, and P minj is the minimum power
are calculated with respect to the active power of microgrid- allowed at the jth microgrid-connected bus.
connected buses. Equations (27)–(31) calculate the sensitivi-
ties of weak and strong buses. Then, using (32) and (33), the III. SIMULATION RESULTS AND DISCUSSION
maximum and minimum active power range for each of the The illustration of the dynamic optimization and the predic-
microgrid-connected bus is calculated. If the active power from tive optimization is first evaluated on a small test system for one
or to the microgrids exceeds these threshold values then the microgrid. Then, the proposed framework in Fig. 4 is imple-
voltage constraints in the utility grid is violated. The maximum mented and evaluated on a real-time practical IEEE test system
and minimum active power flow obtained after this analysis is considering multiple end-user-driven microgrids
then passed to the microgrids as set points. This is then used as
a constraint in the local optimization formulation of microgrid, 3
N · dt = 24(h), N· = 24(h), N = 480. (34)
as explained in 3. This ensures that although the microgrids are 60
economically dispatched to benefit the end users, it also keeps
A. Proof of Concept
the active power flow within the designated range for reliable
operation of the utility grid. Note that power flow is generally A small test system as shown in Fig. 5 is used first to
performed by all utilities as a part of distributed management demonstrate the effectiveness of the proposed method. For the

Authorized licensed use limited to: UNIVERSIDADE FEDERAL DA PARAIBA. Downloaded on February 04,2023 at 14:06:30 UTC from IEEE Xplore. Restrictions apply.
PAUL et al.: QUADRATIC PROGRAMMING BASED OPTIMAL POWER AND BATTERY DISPATCH FOR GRID-CONNECTED MICROGRID 1799

TABLE I
SIMULATION PARAMETERS AND VALUES

T 24 h

dt 30 min
BESS 1 MWh/250 kW
Load 500 kW

TABLE II
QUADRATIC COST COEFFICIENTS

Cost Function x2 x Constant Fig. 7. SOC during optimal dispatch.

Coefficient 0.1 12.6 8

TABLE III
COST OF RHC WITH VARIOUS HORIZON SIZES

Samples N = 48 N = 24 N = 12 N =5

Cost ( $ ) 309.99 308.95 306.46 278.95

Fig. 6. Simulation results based on dynamic optimization.

simulation purpose, it is assumed that the microgrid is always

connected to the grid. The microgrid devices and its ratings are Fig. 8. Predictive horizon battery dispatch for various horizons.
presented in Table I. The assumed cost function is presented in
Table II. The day-ahead forecasted PV power generation utiliz- power profile into consideration. As shown in Table III, the cost
ing the data acquired from a Southern U.S. location is used for for N = 48 is higher as compared to the cost calculated in fixed
the analysis. Half an hour interval data are read from the dataset, horizon optimization, which is $297.68.
and the PV generation profile is extracted, as shown in Fig. 6. From Fig. 8 and Table III, we can observe that as the window
The number of samples obtained from the initial data is based size is reduced, the overall cost for predictive horizon based
on (34). optimization is found to be reduced. For a lower value of win-
1) Results and Discussions—Fixed Horizon Optimization dow size, the battery discharges more, and hence, the power
Considering One Full Day: Fig. 6. shows the dispatch schedule demanded from the grid is reduced. However, when the window
for a full day. During the early morning and night, the load is size increases, as it sees the full next-day PV generation sched-
met by the battery and the grid. The battery dispatch is 106 kW ule, it will give a lower dispatch set point. Even though the cost
during the morning time and 108 kW during evening. As the PV is higher, the advantage is that the SOC of the battery will be
generation increases it produces more power than required by higher at the end of the day. So, for the next day, the battery
the load. This allows the battery to be charged and also satisfy power can be used to satisfy the loads, and hence, grid power
the load requirement. However, the charging power is optimized required will be reduced. If an opportunity cost is added for the
such that there is a uniform value of 23 kW power flowing back conserved SOC then the overall cost is seen to be lower.
to the grid so that the overall cost function is minimized. Fig. 7 3) Comparison Between Fixed and Predictive Horizon Op-
shows that the SOC has hit the lower limit of 0.2 and also the timizations: To demonstrate the advantage of using the predic-
upper limit of 1; hence, the battery is fully utilized. The cost for tive horizon modification, both the algorithms were run for three
the above-mentioned dispatch schedule was $297.68. Note that consecutive days considering sunny and cloudy weather days.
when the SOC is 20% or 100%, the battery has been considered In the fixed horizon case, at the end of each day the battery SOC
fully utilized. was updated to 100%. If SOC was below a particular value then
2) Results and Discussions—Predictive Horizon: In this ap- the cost for the next day was increased by 10% assuming that the
proach, the battery schedule is determined considering a one-day grid demand will be higher for the next day. This is done to bring
horizon optimization. For the full-day horizon, the dispatch set an economic significance to the reduced battery discharge at the
points obtained for the present day also takes the next-day PV end of the day for the predictive horizon optimization dispatch.

Authorized licensed use limited to: UNIVERSIDADE FEDERAL DA PARAIBA. Downloaded on February 04,2023 at 14:06:30 UTC from IEEE Xplore. Restrictions apply.
 1800 IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 54, NO. 2, MARCH/APRIL 2018

TABLE IV
COST OF THREE DAYS PREDICTIVE OPTIMIZATION

Days Day 1 Day 2 Day 3 Total

Cost ( $ ) 309.99 514.70 408.69 1233.38

TABLE V
COST OF THREE DAYS FIXED OPTIMIZATION

Days Day 1 Day 2 Day 3 Total

Fig. 9. Predictive horizon three-day dispatch.
Cost ( $ ) 297.68 566.26 450.29 1314.37

Fig. 10. Fixed horizon three-day dispatch.

Fig. 12. IEEE 33 bus distribution network.

the cost for fixed horizon optimization is higher. Hence, con-

sidering three consecutive days, the total cost is $81 higher
for fixed horizon optimization compared to predictive horizon
Fig. 11. SOC comparison for predictive and fixed horizon dispatch. optimization.

B. Application of the Proposed Method on a Distribution

Figs. 9 and 10 show the comparison of battery dispatch values
and the power consumed from grid for predictive horizon and Feeder
fixed horizon based dispatch. Fig. 11 shows the SOC for both A modified IEEE 33 bus distribution feeder evolved from [35]
of these dispatch methods. is used to show the optimized operation of the feeder if there are
From Fig. 11, it can be seen that the SOC in the case of multiple end-user-driven microgrids. A modified schematic dia-
fixed horizon dispatch has hit the lower limit by midnight of gram of the system including the configuration of the microgrid
the first day, however, the SOC is about 0.6 (i.e., 60%) in the is shown in Fig. 12. Microgrids can be connected at different
predictive optimization case. Hence, the battery can take part in buses for different case studies. In this setup, the batteries are
the next-day dispatch schedule. This illustrates the fact that the considered dispatched according to the economically dispatch
predictive horizon dispatch manages the SOC better than the values obtained from the optimization methods mentioned in
fixed horizon dispatch, which is helpful to tackle future uncer- the previous sections. The voltage profile of the feeder along
tainties in PV power. The net cost was calculated and shown in with the SOC management of the energy storage devices are
Tables IV and V. It can be observed from the PV power profile compared for receding horizon control (RHC) and fixed hori-
of Figs. 9 and 10 that Day 1 is sunny, Day 2 is cloudy and zon methods. Voltage profiles of the sensitive buses are marked
Day 3 is partly cloudy. This is also evident from the results of in red in Fig. 12. Four case studies are performed. Case 1 shows
Tables IV and V. Day 1 consumes less grid power and Day 2 the comparison of the proposed approach with the centralized
consumes more grid power. Moreover, for the first day, the fixed optimization approach. Case 2 and 3 show the scalability of the
horizon optimization gave a lower cost compared to predictive proposed approach for multiple microgrids connected in a dis-
optimization. But for Day-2 since the battery is fully discharged, tribution feeder. In Case 2, two microgrids are located nearby,

Authorized licensed use limited to: UNIVERSIDADE FEDERAL DA PARAIBA. Downloaded on February 04,2023 at 14:06:30 UTC from IEEE Xplore. Restrictions apply.
PAUL et al.: QUADRATIC PROGRAMMING BASED OPTIMAL POWER AND BATTERY DISPATCH FOR GRID-CONNECTED MICROGRID 1801

Fig. 13. Experimental setup.

TABLE VI
PARAMETERS FOR VOLTAGE SENSITIVITY ANALYSIS
Fig. 14. Case 1: Comparison of the proposed approach and centralized ap-
Cases MG SVweakj SVstrongj P maxj P minj proach for fixed load: (a) PV power and load power; (b) grid power; (c) battery
power; and (d) state of charge of battery.
Case 1 MG1 0.515 0.1 713 kW −400 kW
Case 1 MG2 0.505 0.094 717 kW −294 kW
Case 2 MG1 0.495 0.15 722 kW −433 kW optimization algorithm, the power transaction between the grid
Case 2 MG2 0.505 0.094 717 kW −294 kW and microgrid is determined by the centralized dispatch. Then,
the Pgrid obtained is used as a net load for the bus where micro-
grid is connected and an optimal power flow is solved. Fig. 14
and in Case 3, two microgrids are located at different parts shows the comparison of grid power, battery power, and the state
of the feeder (see Fig. 12). Case 4 demonstrates the necessity of charge of battery for both the approaches for fixed load, and
of coordination between local optimization and DMS. Fig. 13 Fig. 15 shows the comparison for varying load. It can be seen
shows the experimental setup for implementing the developed that the centralized optimization approach requires more power
method. A model of the distribution system is illustrated in real- to be consumed from the grid compared to the proposed opti-
time platform. The algorithm for optimization is implemented mization algorithm. Also, large battery idle period can be seen
in MATLAB 2016b software and run on a separate work station in the case of centralized optimization. Tables VII and VIII fur-
with a CPU clock time of 2.5 GHz. MATLAB function “quad- ther quantify the cost of grid power and shows that the proposed
prog” is used to solve the formulated dynamic optimization approach saves more compared to the conventional centralized
problem. It takes on an average of 80 s to run the optimization approach for both fixed and varying load. For the varying load
for a time horizon of four days. Power flow of the system un- condition, total kilowatthour of the battery with the proposed
der study is run in a real-time digital simulator (RTDS). After approach is 780 kWh for three days and with the centralized
obtaining the state variables from the RTDS a static Jacobian- approach the battery output power is 550 kWh. This shows that
based analysis is performed to find the voltage sensitivity of the the proposed approach can use the battery more efficiently.
weak and strong buses. In the system under study, bus 18 is the 2) Case 2: In this case, as mentioned, two microgrids MG1
weak bus and bus 2 is the strong bus. Using the sensitivities and MG2 are connected to two adjacent buses of the feeder, i.e.,
values, the maximum and minimum powers allowed for each of bus 24 and bus 25, respectively. The load at bus 24 and bus 25 is
the microgrid buses are calculated. This is then passed to the changed from 420 to 500 kW and from 420 to 400 kW, respec-
local optimization of each of the microgrids, which generates tively. The PV farm rating and the battery configurations are the
the dispatch values for the microgrid to follow. Table VI sum- same for both microgrids. PV power profiles of both PV farms
marizes the maximum and minimum dispatch values calculated are almost the same as these microgrid buses are considered
from the steady-state sensitivity analysis. to be geographically close. One day from each season is ran-
1) Case 1: In this case, the results from the proposed ap- domly chosen for the PV power profile (i.e., total four different
proach are compared with a centralized dispatch [16], [20]. seasonal days) to demonstrate the application of the proposed
We name this as centralized optimization. In the centralized method in different seasons.

Authorized licensed use limited to: UNIVERSIDADE FEDERAL DA PARAIBA. Downloaded on February 04,2023 at 14:06:30 UTC from IEEE Xplore. Restrictions apply.
 1802 IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 54, NO. 2, MARCH/APRIL 2018

Fig. 16. Case 2: Predictive horizon three-day dispatch for MG1 and MG2.

Fig. 15. Case 1: Comparison of proposed approach and centralized approach

for varying load: (a) PV power and load power; (b) grid power; (c) battery
power; and (d) state of charge of battery.

TABLE VII
COST COMPARISON BETWEEN PROPOSED AND CENTRALIZED APPROACH WITH
FIXED LOAD

Days Day 1 Day 2 Day 3 Total

RHC ( $ ) 534.25 632.7258 536.7758 1703.8

Centralized ( $ ) 546.4786 631.01 539.0671 1716.6

TABLE VIII
COST COMPARISON BETWEEN PROPOSED AND CENTRALIZED APPROACH WITH
VARYING LOAD

Days Day 1 Day 2 Day 3 Total

RHC ( $ ) 518.4685 623.22 532.2858 1668.7

Centralized ( $ ) 518.4685 622.83 533.6113 1674.9
Fig. 17. Case 2: Fixed horizon three-day dispatch for MG1 and MG2.

Figs. 16 and 17 show battery dispatch values, grid power, load

power, and PV power for predictive horizon and fixed horizon
based dispatch, respectively. Fig. 18 also illustrates the compar-
ison between the SOC for predictive horizon and fixed horizon
based optimization method. It shows that predictive horizon
method manages the SOC well and SOC is higher compared to
the fixed horizon method, which is consistent with the results
in previous sections. It has been observed that for both predic-
tive horizon and fixed horizon method voltages of all the buses
are well within the limit which is 0.9–1.1 p.u. for this feeder.
The total loss in the active power is also affected because of Fig. 18. SOC comparison for predictive and fixed horizon dispatch.

Authorized licensed use limited to: UNIVERSIDADE FEDERAL DA PARAIBA. Downloaded on February 04,2023 at 14:06:30 UTC from IEEE Xplore. Restrictions apply.
PAUL et al.: QUADRATIC PROGRAMMING BASED OPTIMAL POWER AND BATTERY DISPATCH FOR GRID-CONNECTED MICROGRID 1803

Fig. 19. Case 2: total active power loss comparison for three days.

TABLE IX
VOLTAGE PROFILE IMPROVEMENT FOR CASE 2

Bus Number Bus 2 Bus 22 Bus 25 Bus 18

Predictive horizon (%) 0.0185 0.3950 0.4879 0.1386

Fixed horizon (%) 0.0221 0.4710 0.5817 0.1653
Fig. 20. Case 3: Predictive horizon three-day dispatch for MG1 and MG2.

minimizing the power consumed from the grid. From Fig. 19,
it can be seen that the active power loss in the feeder is less if
battery is dispatched based upon optimization considering cost.
For fixed horizon, feeder loss is improved by 6.03%, and for
predictive horizon, loss is improved by 5.18% compared to the
base case where no microgrids are connected. Improvements in
voltages are summarized in Table IX.
3) Case 3: In this case, MG1 is placed at Bus 22 and MG2 is
placed at Bus 25. This case is presented to show the effectiveness
of the proposed method if the microgrids are not geographically
at close proximity. Since two microgrids are located in two dif-
ferent places, the PV power profile is considered different for
both the PV farms. For analysis purpose, the load at Bus 22 is
modified from 90 to 100 kW, and at Bus 25, the load is modi-
fied from 420to 500 kW. Figs. 20 and 21 show the optimization
results for predictive horizon and fixed horizon dispatch, respec-
tively. Fig. 22 illustrates the active power profile comparison. It
can be seen that the improvement in losses with fixed horizon
and predictive optimization are 4.11% and 4.26%, respectively.
The voltage improvements are quantified in Table X.
4) Case 4: This case demonstrates the usefulness of defin- Fig. 21. Case 3: Fixed horizon three-day dispatch for MG1 and MG2.
ing maximum and minimum powers for each of microgrid-
connected bus from the DMS. The experimental setup for this
case is similar to the one in Case 2. In this case, no maximum and
minimum power limit is enforced on the grid power. It can be
seen from Fig. 23 that the active power consumed from the grid
for MG1 is greater than the maximum allowed power, which is
from Table IV is 713 kW. Fig. 24 illustrates the voltage level of
the weakest bus in the feeder, which is 18. It can be seen that
the voltage goes below the low voltage limit (0.9 p.u) as the
power absorbed from the grid by MG1 is more than the maxi-
mum power allowed to be consumed from the grid for MG1. So,
based upon the steady-state analysis in the DMS, it is important
to always update the maximum and minimum powers for each
Fig. 22. Case 3: Total active power loss comparison for three days.
microgrid location.

Authorized licensed use limited to: UNIVERSIDADE FEDERAL DA PARAIBA. Downloaded on February 04,2023 at 14:06:30 UTC from IEEE Xplore. Restrictions apply.
 1804 IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 54, NO. 2, MARCH/APRIL 2018

6) Overall Analysis: The proposed architecture proves to

have the capability to include grid economics, reliability and
stability criteria, and maximum utilization of the PV and bat-
tery system and can be easily integrated with the existing power
distribution infrastructure. The method can also be easily modi-
fied to accommodate more constraints and scenarios based upon
real-time operations. For example, an uncertainty matrix that
considers the forecasting errors or change in weather pattern
can be included. Also, ramp rate constraints for the batteries
can be imposed. The approach can also be used to size the
Fig. 23. Case 4: Predictive horizon three-day dispatch for MG1. battery energy storage more effectively.

TABLE X
VOLTAGE PROFILE IMPROVEMENT FOR CASE 3 IV. CONCLUSION
In this paper, a new concept of end-user-driven microgrid is
Bus Number Bus 2 Bus 22 Bus 25 Bus 18
introduced, and an optimal power scheduling approach for the
Predictive horizon (%) 0.0164 0.2677 0.2130 0.0704 battery dispatch in the microgrid is illustrated. The main contri-
Fixed horizon (%) 0.0172 0.2808 0.2267 0.0740 bution of this paper is that unlike other centralized approaches
reported in the literature, the proposed architecture provides
end-user flexibility, optimally manages the battery minimizing
the power required from the grid and, at the same time, en-
sures grid-level reliability. After evaluating the proof of con-
cept, the proposed optimal scheduling is compared with one of
the existing centralized optimization approaches. Furthermore,
the scalability of the algorithm on a power distribution feeder
is illustrated with two microgrids connected at different points.
Test results demonstrated that the proposed approach perform
better than the centralized approach in terms of optimal bat-
tery scheduling, end-user flexibility, and minimizing the overall
Fig. 24. Case 4: Predictive horizon three-day dispatch for MG1. cost. Moreover, the real-time implementation showed that the
proposed architecture is feasible and scalable, and it also ensure
the grid-level reliability.

REFERENCES
[1] J. C. Vasquez, J. M. Guerrero, J. Miret, M. Castilla, and L. G. de Vicuna,
“Hierarchical control of intelligent microgrids,” IEEE Ind. Electron. Mag.,
vol. 4, no. 4, pp. 23–29, Dec. 2010.
[2] A. G. Tsikalakis and N. D. Hatziargyriou, “Centralized control for op-
Fig. 25. Comparison of battery dispatch for MG1 with and without battery timizing microgrids operation,” in Proc. IEEE Power Energy Soc. Gen.
cost. Meeting, Jul. 2011, pp. 1–8.
[3] M. T. Burr et al., “Emerging models for microgrid finance: Driven by the
need to deliver value to end users,” IEEE Electrific. Mag., vol. 2, no. 1,
5) Effect of Battery Cost on the Optimization Result: In this pp. 30–39, Mar. 2014.
section, the cost function of the battery is added in the objective [4] S. Suryanarayanan, R. K. Rietz, and J. Mitra, “A framework for energy
management in customer-driven microgrids,” in Proc. IEEE Power Energy
function of the proposed optimization formulation in (5)–(11). Soc. Gen. Meeting, Jul. 2010, pp. 1–4.
For the cost function of the battery, a simple function like a [5] S. Suryanarayanan and J. Mitra, “Enabling technologies for the customer-
thermal generator is used. The coefficients of the cost function driven microgrid,” in Proc. IEEE Power Energy Soc. Gen. Meeting, Jul.
2009, pp. 1–3.
are a = 0.01, b = 3.2, and c = 2. The objective function in (5) is [6] P. Jain, S. J. Ranade, S. Gupta, and E. Pontelli, “Optimum operation of
replaced by (35). Then, Case 1 is run again to see the difference a customer-driven microgrid: A comprehensive approach,” in Proc. IEEE
in the grid and battery power profile for discarding the battery Int. Conf. Power Electron. Drives Energy Syst., Dec. 2012, pp. 1–6.
[7] Y. Xiang, J. Liu, and Y. Liu, “Robust energy management of microgrid
cost. From Fig. 25, it is observed that there is no noticeable with uncertain renewable generation and load,” IEEE Trans. Smart Grid,
change in the grid and battery power for MG1. Similar affect vol. 7, no. 2, pp. 1034–1043, Mar. 2016.
is seen for MG 2 as well. Similar observations are made for [8] H. L. G. Chang and H. Su, “Short-term distributed energy resource
scheduling for a DC microgrid,” Energy Power Eng., vol. 5, no. 4B,
detailed battery cost model as well pp. 15–21, 2013.
[9] A. D. Giorgio, F. Liberati, A. Lanna, A. Pietrabissa, and F. D. Priscoli,

T
“Model predictive control of energy storage systems for power tracking
[F (Pgrid (t)) + F (Pbatt (t))]. (35) and shaving in distribution grids,” IEEE Trans. Sustain. Energy, vol. 8,
t=1 no. 2, pp. 496–504, Apr. 2017.

Authorized licensed use limited to: UNIVERSIDADE FEDERAL DA PARAIBA. Downloaded on February 04,2023 at 14:06:30 UTC from IEEE Xplore. Restrictions apply.
PAUL et al.: QUADRATIC PROGRAMMING BASED OPTIMAL POWER AND BATTERY DISPATCH FOR GRID-CONNECTED MICROGRID 1805

[10] A. D. Giorgio, F. Liberati, and A. Lanna, “Real time optimal power flow [34] J. Mattingley, Y. Wang, and S. Boyd, “Code generation for receding
integrating large scale storage devices and wind generation,” in Proc. 23rd horizon control,” in Proc. IEEE Int. Symp. Comput.-Aided Control Syst.
Mediterranean Conf. Control Autom., Jun. 2015, pp. 480–486. Des., Sep. 2010, pp. 985–992.
[11] A. Nottrott, J. Kleissl, and B. Washom, “Storage dispatch optimization for [35] M. E. Baran and F. F. Wu, “Network reconfiguration in distribution systems
grid-connected combined photovoltaic-battery storage systems,” in Proc. for loss reduction and load balancing,” IEEE Trans. Power Del., vol. 4,
IEEE Power Energy Soc. Gen. Meeting, Jul. 2012, pp. 1–7. no. 2, pp. 1401–1407, Apr. 1989.
[12] MATLAB. Interior-point-convex quadprog algorithm, 2016. [On-
line]. Available: http://www.mathworks.com/help/optim/ug/quadratic- Tim George Paul (S’08) received the B.E. de-
programming-algorithms.html#bsqspm gree in electrical engineering from Anna University,
[13] P. M. Corrigan and G. T. Heydt, “Optimized dispatch of a residential Chennai, India, in 2010, and the M.S. degree in
solar energy system,” in Proc. 39th North Amer. Power Symp., Sep. 2007, electrical engineering from the University of North
pp. 183–188. Carolina at Charlotte, Charlotte, NC, USA, in 2015.
[14] M. D. Hopkins, A. Pahwa, and T. Easton, “Intelligent dispatch for dis- He joined the Engineering Services Division,
tributed renewable resources,” IEEE Trans. Smart Grid, vol. 3, no. 2, Schweitzer Engineering Laboratories, Pullman, WA,
pp. 1047–1054, Jun. 2012. USA, in 2016, as a Power System Engineer focusing
[15] P. Mahat et al., “A micro-grid battery storage management,” in Proc. IEEE on power system modeling using a real-time digital
Power Energy Soc. Gen. Meeting, Jul. 2013, pp. 1–5. simulator for testing power management schemes. He
[16] S. Teleke, M. E. Baran, S. Bhattacharya, and A. Q. Huang, “Rule-based has experience in the field of power system protection
control of battery energy storage for dispatching intermittent renewable and automation.
sources,” IEEE Trans. Sustain. Energy, vol. 1, no. 3, pp. 117–124, Oct.
2010.
[17] S. Moghadasi and S. Kamalasadan, “Real-time optimal scheduling of
smart power distribution systems using integrated receding horizon control Sheikh J. Hossain (S’16) received the B.Sc. de-
and convex conic programming,” in Proc. IEEE Ind. Appl. Soc. Annu. gree in electrical and electronic engineering from the
Meeting, Oct. 2014, pp. 1–7. Bangladesh University of Engineering and Technol-
[18] S. Abdelrazek and S. Kamalasadan, “Integrated control of battery energy ogy, Dhaka, Bangladesh, in 2013. He is currently
storage management system considering PV capacity firming and energy working toward the Ph.D. degree in electrical en-
time shift applications,” in Proc. IEEE Ind. Appl. Soc. Annu. Meeting, Oct. gineering with the University of North Carolina at
2014, pp. 1–7. Charlotte, Charlotte, NC, USA.
[19] S. A. Abdelrazek and S. Kamalasadan, “A weather-based optimal storage He was a Software Engineer at the Samsung Re-
management algorithm for PV capacity firming,” IEEE Trans. Ind. Appl., search and Development Institute, Bangladesh, from
vol. 52, no. 6, pp. 5175–5184, Nov. 2016. 2013 to 2015. His research interests include dis-
[20] Y. Riffonneau, S. Bacha, F. Barruel, and S. Ploix, “Optimal power flow tributed energy systems integration, modeling and
management for grid connected PV systems with batteries,” IEEE Trans. control, and wide area monitoring, optimization, and control of power system.
Sustain. Energy, vol. 2, no. 3, pp. 309–320, Jul. 2011.
[21] J. Z. H. Tazvinga and X. Xia, “Minimum cost solution of photovoltaic- Sudipta Ghosh (M’12) received the B.Tech. de-
diesel-battery hybrid power systems for remote consumers,” Solar Energy, gree from National Institute of Technology (NIT)
vol. 96, pp. 292–299, 2013. Jamshedpur, Jamshedpur, India, in 2002, the M.Tech.
[22] K. Baker, J. Guo, G. Hug, and X. Li, “Distributed MPC for efficient co- degree in electrical engineering from NIT Durgapur,
ordination of storage and renewable energy sources across control areas,” Durgapur, India, in 2009, and the Ph.D. degree in
IEEE Trans. Smart Grid, vol. 7, no. 2, pp. 992–1001, Mar. 2016. electrical engineering from the Indian Institute of
[23] A. Ortega, P. M. Namara, and F. Milano, “Design of MPC-based controller Technology, Delhi, India, in 2013.
for a generalized energy storage system model,” in Proc. IEEE Power He is currently a Faculty Member with the Uni-
Energy Soc. Gen. Meeting, Jul. 2016, pp. 1–5. versity of North Carolina at Charlotte, Charlotte, NC,
[24] C. Eksin, A. Hooshmand, and R. Sharma, “A decentralized energy man- USA. He was an Assistant Professor in the Depart-
agement system,” in Proc. Eur. Control Conf., Jul. 2015, pp. 2260–2267. ment of Electrical Engineering, Indian School of
[25] M. Mahmoodi, P. Shamsi, and B. Fahimi, “Economic dispatch of a hybrid Mines, Dhanbad, India. His research interests include power system small-
microgrid with distributed energy storage,” IEEE Trans. Smart Grid, vol. 6, signal stability, power system modeling, and dynamic analysis.
no. 6, pp. 2607–2614, Nov. 2015.
[26] T. Paul, S. Ghosh, S. Kamalasadan, and P. Mandal, “A quadratic pro-
gramming based optimal power and battery dispatch for grid connected Paras Mandal (S’05–M’06–SM’12) received the
microgrid,” in Proc. IEEE Ind. Appl. Soc. Annu. Meeting, Oct. 2016, B.E. degree from Kuvempu University, India, in
pp. 1–8. 1998, the M.E. degree from the Asian Institute of
[27] B. Lu and M. Shahidehpour, “Short-term scheduling of battery in a grid- Technology, Thailand, in 2002, and the Ph.D. degree
connected PV/battery system,” IEEE Trans. Power Syst., vol. 20, no. 2, from the University of the Ryukyus, Japan, in 2005,
pp. 1053–1061, May 2005. all in electrical engineering.
[28] S. J. Chiang, K. T. Chang, and C. Y. Yen, “Residential photovoltaic energy He is currently an Associate Professor of electri-
storage system,” IEEE Trans. Ind. Electron., vol. 45, no. 3, pp. 385–394, cal and computer engineering with the University of
Jun. 1998. Texas at El Paso, El Paso, TX, USA. His research
[29] C. Wang and M. H. Nehrir, “Power management of a stand-alone interests include power systems operations and mar-
wind/photovoltaic/fuel cell energy system,” IEEE Trans. Energy Convers., kets, power system optimization, and renewable en-
vol. 23, no. 3, pp. 957–967, Sep. 2008. ergy integration and forecasting.
[30] S. Jain and V. Agarwal, “An integrated hybrid power supply for distributed
generation applications fed by nonconventional energy sources,” IEEE
Trans. Energy Convers., vol. 23, no. 2, pp. 622–631, Jun. 2008.
[31] N. H. El-Farra, X. Wang, and A. Palazoglu, “Operational optimization Sukumar Kamalasadan (SM’01–M’05–SM’17) re-
and demand response of hybrid renewable energy systems,” Appl. Energy, ceived the B.Tech. degree in electrical and electronics
vol. 143, pp. 324–335, 2015. from the University of Calicut, Malappuram, India, in
[32] H. Kanchev, D. Lu, F. Colas, V. Lazarov, and B. Francois, “Energy man- 1991, the M.Eng. degree in electrical power systems
agement and operational planning of a microgrid with a PV-based active management from the Asian Institute of Technology,
generator for smart grid applications,” IEEE Trans. Ind. Electron., vol. 58, Bangkok, Thailand, in 1999, and the Ph.D. degree in
no. 10, pp. 4583–4592, Oct. 2011. electrical engineering from the University of Toledo,
[33] L. Xiaoping, D. Ming, H. Jianghong, H. Pingping, and P. Yali, “Dy- Toledo, OH, USA, in 2004.
namic economic dispatch for microgrids including battery energy stor- He is currently a Professor with the Department
age,” in Proc. 2nd Int. Symp. Power Electron. Distrib. Gen. Syst., Jun. of Electrical and Computer Engineering, University
2010, pp. 914–917. of North Carolina at Charlotte, Charlotte, NC, USA.

Authorized licensed use limited to: UNIVERSIDADE FEDERAL DA PARAIBA. Downloaded on February 04,2023 at 14:06:30 UTC from IEEE Xplore. Restrictions apply.