Stochastic Energy Management for Microgrids with

Constraints under Uncertainty

Jianzhe Liu Giorgio Rizzoni Benjamin Yurkovich
Department of Electrical Department of Mechanical Center for Automotive Research
and Computer Engineering and Aerospace Engineering the Ohio State University
Center for Automotive Research Center for Automotive Research Columbus, Ohio 43210
the Ohio State University the Ohio State University Email: yurkovich.7@osu.edu
Columbus, Ohio 43210 Columbus, Ohio 43210
Email: liu.2430@osu.edu Email: rizzoni.1@osu.edu

Abstract—This paper studies two energy management prob- with time. Furthermore, robust and stochastic programming
lems for microgrids with constraints involving uncertainty. The approaches have been employed to solve the EMPs [4], [8]–
first energy management problem has constraints on the utiliza- [10].
tion of certain components. It can be used to ensure the pene-
tration level of renewable generation or plug-in electric vehicles. Despite the rich literature about the Microgrid EMPs, the
The second energy management problem aims at refraining the EMPs with constraints under uncertainty have not been ade-
output power of the microgrid. It is required by applications quately studied. This paper solves two microgrid EMPs with
such as peak power shaving and instrument protection. With uncertain constraints on output power. The first EMP attempts
the integration of components having random output power to ensure a level of utilization of certain components. The
such as photovoltaic panels and plug-in electric vehicles in the
microgrid, the two problems have uncertainty in the constraints. utilization of a component is represented by a percentage
A unified chance constrained programming formulation of the share of the output power of the component in the overall
two problems is proposed in the paper. This formulation provides microgrid output power. This constraint is required by the need
probabilistic guarantees on the satisfaction of the constraints and to increase the penetration of RESs [11] or to prevent excessive
transforms the two problems into convex programs that can be usage of expensive components such as battery packs. For
solved efficiently. A practical case study verifies the effectiveness
of the proposed approach. example, if the power consumed by RESs is required to be at
least 50% of the overall microgrid power, then a constraint
I. INTRODUCTION needs to be formulated to ensure this utilization of RESs
The power system is gradually evolving into a smart grid withstanding the random nature of the RESs. The second EMP
that is safer and greener. With advanced cooperative operation endeavors to refrain the magnitude of the microgrid output
techniques, plug-in electric vehicles (PEVs) together with power. This EMP is required by applications such as peak
distributed resources such as energy storage systems (ESSs) power shaving or instrument protection [12]. For example, to
and renewable energy sources (RESs) will play important roles reduce energy cost microgrids typically tend to defer the con-
in the future smart grid. sumption to the time with lower energy prices. This may cause
Microgrid is an emerging distribution infrastructure that large rebounds in power demands, and the rebounds could
coordinates PEVs, ESSs, and RESs with other local energy be dangerous for distribution facilities. Therefore, constraints
components. With proper energy management, a microgrid on the output power magnitude are needed for instrument
is able to act as one controllable energy entity that pro- protection purpose. In conclusion, the two EMPs are originated
vides grid-friendly power outputs. The energy management from practical applications, and they set different constraints
problems (EMPs) of microgrids seek to minimize the energy on the microgrid output power, which is uncertain.
or operating costs with respect to certain time-of-use (TOU) A unified chance constrained programming (CCP) formu-
energy prices [1]. The challenges of the microgrid EMPs are lation is proposed in this paper for the two EMPs. The CCP
contributed by the uncertain output power of the microgrid. approach involves probabilistic constraints that only need to
Different approaches to tackle the uncertainty have been be satisfied with a high probability. Through utilizing the pro-
presented in the literature. In [2], [3], the authors exploit posed method, the two types of constraints are formulated by
numerical prediction techniques to forecast the uncertainty in one generic chance constraint that sets probabilistic guarantees
the microgrid and solve the EMPs in a deterministic way. on the satisfaction of the two constraints. The proposed CCP
However, the credibility of meteorology and load forecasts can be transformed into equivalent convex programs that have
drops substantially with the extent of time [4]. To reduce computationally efficient algorithms to find the solution. This
the accumulation of numerical forecasts errors, in [5]–[7] the unified formulation is an extension of the chance constraint on
EMPs are solved through a rolling horizon strategy, which output framework (CCOF) proposed in the authors’ work [13].
makes planning at each time step to refine the forecasts Notice that the CCP formulation of this paper is more generic

978-1-5090-0403-4/16/$31.00 ©2016 IEEE

such that it covers more applications, like the utilization where xPj,1 (k) and xPj,2 (k) are the minimum charging time
bounding problems, than the CCOF does. and maximum waiting time of the j-th on-site PEV at time k
The rest of the paper is organized as follows. Section II respectively. The initial state is given by xPj,0 = [ej /E¯jP , tj −
describes the model of the microgrid and the two EMPs, ej /E¯jP ], where ej is the required energy and tj is the number
Section III presents the unified CCP formulation, Section IV of time steps from time 0 to the charging deadline. The two-
uses a practical microgrid application to verify the proposed dimensional control inputs uPj take values in set U defined as
method, and Section V concludes the paper and discusses U = {(a1 , a2 ) ∈ {0, 1}2 |a1 + a2 = 1}. The constraints on
future research directions. uPj (k) are given by,
II. S YSTEM M ODEL AND P ROBLEM D ESCRIPTION uPj,1 (k) + uPj,2 (k) = 1, (6)
A. System Model uPj,1 (k) ≤ min{xPj,1 (k), 1}, (7)
We consider a microgrid consisting of n batteries, m on- uPj,1 (k) ≥ max{1 − xPj,2 (k), 0}, (8)
site plug-in electric vehicles (PEVs), p types of renewable
energy sources (RESs), and 1 composite uncontrollable load uPj,1 (k) = 0, k > tj (9)
in this paper. This configuration represents many microgrids Constraint (6) is set due to the fact that an on-site PEV is
in practice [14]–[18]. either waiting or charging at one time step. Equations (7) and
Batteries can improve the reliability of and provide more (8) set constraints on the energy consumption of the PEVs and
flexibility for the microgrid operations. In this paper, we ensure the required energy to be charged before the deadline.
assume that a battery management system is installed for each Constraint (9) ensures that the PEV charging event do not
battery, so we can observe the battery state of charge (SOC) happen after the specified deadline.
estimation and control the battery output power within its RESs, like solar and wind generation, can provide clean
physical limits. In addition, we assume that the batteries have energy for the microgrid to reduce electricity cost and CO2
no power loss during operation. Under these conditions, the emission, but their volatile and uncertain output power makes
dynamics of the i-th battery is given as follows, it hard for their integration to the grid. The microgrid co-
 B
xi (k + 1) = xBi (k) + uBi (k), ordinates the RESs with other controllable components to
(1)
xBi (0) = Ei0
B
/Capi , i = 1, 2, · · · , n alleviate their volatility. With adequate historic data and proper
estimation technique, the probability distributions for RESs
where xBi (k) is the i-th battery’s SOC at time k, uBi (k) is the generation can be obtained over a certain time span [20]. In
control variable that indicates the change of SOC at k with this paper, we assume that for the l-th RES its output energy
positive sign indicating increase of the SOC, Ei0B
is the initial at time k, wlR (k) is given by
energy stored in the battery, and Capi is the rated capacity.
Notice that Capi · uBi (k) is the energy output at k. Define xBi wlR (k) ∼ WlR (k), l = 1, 2, · · · , p (10)
and uBi be the upper bounds on the state and control, xBi and
where WlR (k) is the distribution of the l-th renewable resource
uBi be the lower bounds on the state and control, and fiB be
at time k.
the designated final SOC of the battery. The constraints on the
An uncontrollable load cannot be deferred. We assume that
control uBi (k) are given by,
there is a composite uncontrollable load in the microgrid which
uBi (k) ≤ min{xBi − xBi (k), uBi }, (2) is composed of lighting load, computer load, etc. Like a RES,
the probability distribution of the composite load can also be
uBi (k) ≥ max{xBi − xBi (k), uBi }, (3)
estimated. Define wU (k) as the energy consumption of the
xBi (N ) = fiB , (4) composite load at k, it is given by
where (2) and (3) implement physical constraints for output wU (k) ∼ WU (k), (11)
power, and (4) ensures the final SOC reaches the desired
values. where W (k) is the distribution of the composite load at k.
U

When one PEV plugs into the microgrid it becomes on-site. Define x ∈ Rn+2m , u ∈ Rn+2m , y ∈ R, w ∈ Rp+1 , where
i We assume that agreements are signed between the PEV  T
owners and the microgrid so that an on-site PEV provides x(k) = xB1 (k) · · · xBn (k) (xP1 (k))T · · · (xPm (k))T ,
the microgrid with its required energy and charging deadlines  T
upon plugging in, and the microgrid is allowed to control each u(k) = uB1 (k) · · · uBn (k) (uP1 (k))T · · · (uPm (k))T ,
on-site PEV’s charging power as long as the required energy  T
and charging deadlines are respected. For the j-th on-site PEV, w(k) = w1R (k) · · · wpR (k) wU (k) .
its dynamics is given as follows [19]: Using (1), (5), (10), and (11), an aggregated discrete time
 P
xj,1 (k + 1) = xPj,1 (k) − uPj,1 (k) linear model of the microgrid is given by,
xPj (k + 1) = , 
xPj,2 (k + 1) = xPj,2 (k) − uPj,2 (k) (5) x(k + 1) = Ax(k) + Bu(k),
xPj (0) = xPj,0 , j = 1, 2, · · · , m y(k) = Du(k) + Ew(k),
(12)
where y is the overall power output of the microgrid, which The constraint function gk (u(k), w(k)) is a linear com-
equals the sum of all the energy generation and consumption bination of the energy output of all the controllable and
within the microgrid, matrices uncontrollable components at time k. The coefficients ηk , θk ,
  and γk are chosen according to different applications.
In 0
A = In+2m , B = , For utilization of component bounding problem, the co-
0 −I2m
efficients are chosen to be the weight of each component’s
where In is an n-dimensional identity matrix. D ∈ R1×(n+2m) energy output, and the linear combination in the constraint
and E ∈ R1×(p+q) are two row vectors, where D = function is a weighted sum of the energy outputs of all
[Cap1 , · · · , Capn , E1P , 0, · · · , Em
P , 0] and E = [1, 1, · · · , 1]. the components. Therefore, by varying the coefficients the
penetration of certain components’ energy output in the overall
B. Problem Description energy output can alter. If we want the penetration of PEV to
In this paper, we solve two microgrid EMPs with constraints be less than pk ∈ (0, 1) to reduce the stress on power supply,
under uncertainty. The first EMP requires a probabilistic the coefficients are given by,
guarantee on the utilization of certain components. It is rooted  T
ηk= −pk 11×n (1 − pk )11×2m , θk=−pk 1(p+1)×1 , γk=0,
in applications with special demands on the usage of some
components. For example, when bidding in the energy markets where 1i×j is an i × j matrix with all its elements being 1.
the user may want to make full use of the renewable generation For example, we have 1 PEV and 1 composite load in the
to save operation costs. Therefore, a constraint to guarantee microgrid, and pk = 0.1. For simplicity, define PEV energy
the minimum utilization of renewable generation is needed. output as Ep and that of load as El . Then the chance constraint
In addition, to have a desired penetration level of PEV a is given by,
constraint can be set to refrain the utilization of PEVs. These P({El : 0.9Ep −0.1El ≤ 0})≥1 − 
constraints involve the uncertain output power in microgrids.
Ep
The second EMP incorporates a probabilistic bound on the → P({El : ≤0.1})≥1 − .
Ep + El
output power of microgrids. It can be applied to refrain de-
mand rebounds and to protect power instruments. For example, To bound the utilization of other components, the coefficients
some switching power instruments such as power transmission can be chosen similarly.
lines have capacity limits. Without considering these limits In addition, if at time k we want to constrain the microgrid
may cause impractical or harmful planning. Due to the random energy output with an upper (resp., lower) bound bk , we can
nature of the output power, the constraints are also under choose ηk = 1(n+2m)×1 (resp., ηk = −1(n+2m)×1 ) and
uncertainty. θk = 1(p+1)×1 (resp., θk = −1(p+1)×1 ), and γk = −bk (resp.,
γk = bk ). For example, for the microgrid with 1 PEV and a
III. C HANCE C ONSTRAINED P ROGRAMMING composite load, we want the energy output at k to be less than
F ORMULATION bk , then the chance constraint is given by,
In this section, a unified chance constrained programming P ({El : Ep + El ≤ bk }) ≥ 1 − .
(CCP) formulation is proposed for the two EMPs described in
Section II-B. The formulation is given as follows, Therefore, the unified chance constraint (13b) can be used to
N represent the constraints under uncertainty of the two EMPs.
 Chance constraint (13b) can be transformed into a linear
min E c(k)y(k) , subj. to (13a) constraint in decision variable u(k). From (13b) and (14),
u(1),··· ,u(N )
k=1

(2), (3), (4), (6), (7), (8), (9), (12), P {w(k):ηkT diag(D)u(k)+θkT diag(E)w(k)+γk ≤ 0} ≥1 − ,


P({w(k): gk (u(k), w(k)) ≤ 0}) ≥ 1 − , k = 1, · · · , N (13b) →P {w(k):θkT diag(E)w(k)≤−ηkT diag(D)u(k)−γk } ≥1 − ,


→ F −ηkT diag(D)u(k) − γk ≥ 1 − ,
where N is the optimization horizon, c(k) is the energy price
of the k-th time step. We aim to minimize the expectation of → F −1 (1 − ) + ηkT diag(D)u(k) + γk ≤ 0, (15)
the energy cost over a time horizon N . where F is the cumulative distribution function of random
The chance constraint (13b) requires that the probability variable θkT diag(E)w(k). F −1 is the inverse function of F .
measure of the set {w(k) : gk (u(k), w(k)) ≤ 0} to exceed Notice that given WRl , l = 1, · · · , p, and WU , F −1 can be
a confidence level 1 − . The constraint function gk (·, ·) : obtained. In (15), F −1 (1 − ) and γk are constants, thus (15)
Rn+2m × Rp+1 → R is given by is a linear constraint in u(k).
gk (u, w) = ηkT diag(D)u + θkT diag(E)w + γk , (14) The cost function (13a) is also a linear term in the decision
variable. Substituting y(k) = Du(k) + Ew(k) in the cost
n+2m p+1
where ηk ∈ R , θk ∈ R are two coefficient vectors, function, then (13a) is given by,
and diag : Rn → Rn×n is a function which yields a diagonal N

matrix with its diagonal entries being the elements of the E (c(k)Du(k) + c(k)Ew(k)) ,
argument, and γk ∈ R is an additive coefficient. k=1
 N
 N

  We have level 2 PEV chargers in the microgrid whose
→ c(k)Du(k) + E c(k)Ew(k) ,
charging power is uni-directional and ranges from 0 to 6kW .
k=1 k=1
We suppose that there are 80 charging jobs for these PEV
where the second term cannot be optimized by the decision chargers in one day. Half of the charging events take place in
variable, therefore it can be removed from the cost function. the morning, and the other half occur in the afternoon. The
The cost function (13a) is transformed into distributions for the arriving and leaving events are summa-
N
rized in TABLE II, where U (i, j) is the uniform distribution

c(k)Du(k), (16) between the hours of i and j. In all the events, each charging
k=1
job would need 24kW h of energy to charge. From the table,
all the charging events need to finish in one day.
which is a linear term in the decision variable.
Notice that cost function (16), constraints (2), (3), (4), (6), TABLE II: Distributions of PEV charging job
(7), (8), (9), and (15) are all linear in decision variables, and
the system (12) is a discrete time linear model, CCP (13a) Events Distribution
can be transformed equivalently into a static linear program. Morning arriving event U (7, 9)
In [13], there is a similar transformation that has a detailed Morning leaving event U (12, 15)
Afternoon arriving event U (13, 15)
derivation, thus we do not report the reasoning here. Afternoon leaving event U (18, 20)

IV. CASE STUDY

The PV generation and building consumption are assumed
A. Microgrid Testbed
to be uncontrollable. The joint probability distribution of these
In this section we verify the proposed CCP formulation two components can be obtained from historical data. For
using a case study. We use a microgrid testbed motivated by simplicity, here we assume that the joint energy distribution
a microgrid constructed at Center for Automotive Research at time k is N (μ(k), 5). The values for μ(k) are listed in
(CAR), the Ohio State University (OSU). The configuration TABLE III. Notice that the power rises in the morning since
of the microgrid testbed is shown in Fig. 1. The PV panels people come to work. The PV generation reaches its peak
and batteries are connected to a DC bus through DC/DC in the noon, compensating some building consumption. When
converters. The DC bus links to the AC side which supplies people leave for home in the afternoon the overall power drops.
PEV chargers and a building.
TABLE III: Average Hourly PV and Building Load
DC Bus Grid
Grid
PV DC/DC
Hour(k) μ(k) (kW )
DC/AC 1-5 30
6-9 80
10 - 15 50
Battery DC/DC 16 - 20 80
21 - 24 30
AC Bus
Building

There is a grid-coupling controller in the microgrid. It

provides local coordination of all the components. At each
C trolllle
Con
Controller
PEV Chargers time step, it collects information of all the components and
then implements the CCP formulation in Section III for one
Fig. 1: Microgrid configuration designed and being building time step. In this paper, we assume that N = 6, and one time
at CAR, OSU step is one hour. After solving the CCP at one time step, the
optimal planning for this time step is assigned to each of the
We connect 50 20kW h lithium-ion batteries to the mi- controllable component.
crogrid, and we assume that they are identical. TABLE I B. Power Line Protection
concludes the specifications of each battery.
We demonstrate the effectiveness of the proposed method
in this subsection. In this case study, the power line that
TABLE I: Specifications of one battery
connects the grid and the microgrid has a power capacity.
Capacity 20 kW h Since excessive usage of the power line may cause dangerous
Initial and final SOC 0.5 over heating issue, to protect the power line the absolute
Upper bound of SOC 0.9 switching power of the microgrid to the power grid should
Lower bound of SOC 0.1 be refrained with an upper bound. In addition, we minimize
Upper bound of power 4 kW
Lower bound of power -4 kW the energy costs of the microgrid based on a time of use
(TOU) price. The TOU pricing scheme is meant to use time
varying prices to leverage load imbalance by giving incentives peak power is refrained under 250kW . This shows that using
to customers to adjust their consumption profiles. the proposed CCP formulation the 250kW constraint can be
fulfilled. When  ∈ [0.2, 0.9] though the absolute peak power is
TABLE IV: Three-tier Energy Price higher than 250kW it is still much smaller than 300kW . This
demonstrates that the risk level of the chance constraints plays
Peak Hours (12-18) $0.18/kWh an important role for their satisfaction. The smaller the risk
Part-peak Hours (8-12 & 18-22) $0.09/kWh
Off-peak Hours (1-8 & 22-24) $0.04/kWh level is the better the constraint can be met. Nevertheless, as
long as the chance constraints are incorporated the microgrid
can still operate under safe condition. When the chance
constraints are removed from the planning, the absolute peak
TABLE V: Absolute peak power of the microgrid with differ- power is soaring to over 452.21kW , which is much higher than
ent risk levels the line capacity 300kW . This shows that using the proposed
CCP formulation for this problem is crucial for instrument
Risk Level () Absolute Peak Power (kW )
protection and safe operation.
 =0.01 244.27
 =0.10 249.51
 =0.20 251.72
 =0.30 253.32 500
 =0.40 254.67
 =0.50 255.95 450

Absolute Peak Power (kW)

 =0.60 257.25
 =0.70 258.59
400
 =0.80 260.18
 =0.90 262.38
Greedy planning 452.21 350

300
The simulation is set up as follows. The capacity of the
power line is 300kW . For the instrument protection purpose, 250
we use a conservative upper bound for the absolute power out-
put of the microgrid as 250kW . The TOU price is concluded 200
0.01 0.2 0.4 0.6 0.8 Greedy
in TABLE IV. The TOU price has three tiers according to
Risk 
the different stress of the power grid to supply the load. The
time span for each tier and its energy price are set according Fig. 2: Peak power at different risk levels
to realistic data [21]. We simulate the electrical profile of the
microgrid testbed in Section IV-A for one day. In this case study, for each planning the time horizon is
We use the CCP formulation described in Section III to composed of 6 time steps, and each time step equals to 1
solve this problem. Two chance constraints are formulated hour. On a laptop with Intel Core i7, one planning takes 8.31
respectively to refrain the microgrid power to lie in the range seconds in average using toolbox CVX [22], [23].
[−250kW, 250kW ]. Define gk1 and gk2 : Rn+2m × Rp+1 → R
to be constraint functions in the chance constraints. From (13b) V. CONCLUSION
and (14), the chance constraints are given as follows, This paper solves two EMPs for microgrids integrated with
uncertain components such as PEVs and distributed resources.
P({w(k): gk1 (u(k), w(k)) ≤ 0}) ≥ 1 − , k = 1, · · · , 6 The first EMP requires constraints on the penetration level
P({w(k): gk2 (u(k), w(k)) ≤ 0}) ≥ 1 − , k = 1, · · · , 6 of certain components in the microgrid output power. It
gk1 = 11×(n+2m) diag(D)u(k)+11×(p+1) diag(E)w(k)−250, can help boost the utilization of renewable generation and
the penetration of PEVs. The second EMP seeks to bound
gk2 = −11×(n+2m) diag(D)u(k)−11×(p+1) diag(E)w(k)−250.
the output power of the microgrid for applications such as
With the two chance constraints this power line protection peak power shaving and instrument protection. The difficulty
problem can be formulated by the proposed CCP method. of the problems lies in the uncertain constraints concerning
We vary the risk level  in the chance constraints and com- the microgrid output power. A unified CCP formulation is
pare the simulation results with that of the simulation scenario proposed for the two problems in which the two types of
in absence of chance constraints to showcase the effectiveness constraints under uncertainty are formulated by one generic
of the chance constraints for instrument protection. chance constraint. By doing so, probabilistic guarantees on
The absolute peak power of each simulation scenario is the satisfaction of the constraints are obtained. Since the
collected and shown in TABLE V and Fig. 2. For the scenario proposed CCP formulation can be transformed into convex
with the chance constraints,  varies in [0.01, 0.90]. As for programs with numerically effective solutions, the proposed
the greedy planning scenario, it stands for the one without the CCP formulation has good applicability for practical usage.
chance constraints. From Fig. 2, when  ≤ 0.10 the absolute A case study is presented to verify the proposed method. For
future research, we will investigate optimization problems for
large-scale power systems.
R EFERENCES
[1] Y. Xiang, J. Liu, and Y. Liu, “Robust energy management of microgrid
with uncertain renewable generation and load,” IEEE Trans. Smart Grid,
vol. PP, no. 99, pp. 1–1, 2015.
[2] A. Chaouachi et al., “Multiobjective intelligent energy management for
a microgrid,” IEEE Trans. Ind. Electron, vol. 60, no. 4, pp. 1688–1699,
2013.
[3] H. Beltran et al., “Daily solar energy estimation for minimizing energy
storage requirements in pv power plants,” IEEE Trans. Sustain. Energy,
vol. 4, no. 2, pp. 474–481, 2013.
[4] E. Dall’Anese and G. Giannakis, “Risk-constrained microgrid reconfig-
uration using group sparsity,” IEEE Trans. Sustain. Energy, vol. 5, no. 4,
pp. 1415–1425, 2014.
[5] R. Palma-Behnke et al., “A microgrid energy management system based
on the rolling horizon strategy,” IEEE Trans. Smart Grid, vol. 4, no. 2,
pp. 996–1006, 2013.
[6] P. O. Kriett and M. Salani, “Optimal control of a residential microgrid,”
Energy, vol. 42, no. 1, pp. 321–330, 2012.
[7] A. Bemporad, F. Borrelli, and M. Morari, “Model predictive control
based on linear programming - the explicit solution,” IEEE Trans. Autom.
Control, vol. 47, no. 12, pp. 1974–1985, 2002.
[8] Y. Zhang, N. Gatsis, and G. B. Giannakis, “Robust energy management
for microgrids with high-penetration renewables,” IEEE Trans. Sustain.
Energy, vol. 4, no. 4, pp. 944–953, 2013.
[9] Z. Wu et al., “Economic optimal schedule of CHP microgrid system us-
ing chance constrained programming and particle swarm optimization,”
in proc. IEEE Power Energy Soc. General Meeting, 2011, pp. 1–11.
[10] D. T. Nguyen and L. B. Le, “Risk-constrained profit maximization for
microgrid aggregators with demand response,” IEEE Trans. Smart Grid,
vol. 6, no. 1, pp. 135–146, 2015.
[11] Q. Wang, J. Wang, and Y. Guan, “Price-based unit commitment with
wind power utilization constraints,” IEEE Trans. Power Syst, vol. 28,
no. 3, pp. 2718–2726, 2013.
[12] W. Zhang et al., “Aggregated modeling and control of air conditioning
loads for demand response,” IEEE Trans. Power Syst., vol. 28, no. 4,
pp. 4655–4664, 2013.
[13] J. Liu et al., “Energy management problems under uncertainties for grid-
connected microgrids: a chance constrained programming approach,”
IEEE Trans. Smart Grid, vol. PP, no. 99, pp. 1–1, 2016.
[14] D. Boroyevich, I. Cvetkovic, R. Burgos, and D. Dong, “Intergrid: A
future electronic energy network?” Emerging and Selected Topics in
Power Electronics, IEEE Journal of, vol. 1, no. 3, pp. 127–138, 2013.
[15] X. She et al., “On integration of solid-state transformer with zonal dc
microgrid,” IEEE Trans. Smart Grid, vol. 3, no. 2, pp. 975–985, 2012.
[16] D. Dong et al., “Grid-interface bidirectional converter for residential dc
distribution systems-part one: High-density two-stage topology,” IEEE
Trans. Power Electron., vol. 28, no. 4, pp. 1655–1666, 2013.
[17] X. Tang, W. Deng, and Z. Qi, “Investigation of the dynamic stability of
microgrid,” IEEE Trans. Power Syst, vol. 29, no. 2, pp. 698–706, 2014.
[18] H. Kanchev et al., “Energy management and operational planning of a
microgrid with a pv-based active generator for smart grid applications,”
IEEE Trans. Ind. Electron., vol. 58, no. 10, pp. 4583–4592, 2011.
[19] J. Liu, S. Li et al., “Planning and control of electric vehicles using
dynamic energy capacity models,” in proc. IEEE 53rd Annu. CDC, 2013,
pp. 379–384.
[20] Y. Atwa et al., “Optimal renewable resources mix for distribution system
energy loss minimization,” IEEE Trans. Power Syst, vol. 25, no. 1, pp.
360–370, 2010.
[21] N. Zhou et al., “Assessment of distributed energy adoption in commer-
cial buildings. part 1: An analysis of policy, building loads, tariff design,
and technology development,” Bulletin of The Faculty of Engineering
Kyushu Sangyo University, vol. 42, p. 8, 2005.
[22] M. Grant and S. Boyd, “CVX: Matlab software for disciplined convex
programming, version 2.1,” http://cvxr.com/cvx, Mar. 2014.
[23] ——, “Graph implementations for nonsmooth convex programs,”
in Recent Advances in Learning and Control, ser. Lecture Notes
in Control and Information Sciences, V. Blondel, S. Boyd, and
H. Kimura, Eds. Springer-Verlag Limited, 2008, pp. 95–110,
http://stanford.edu/∼boyd/graph dcp.html.