1

A Distributed Solution to Real-Time Economic

Dispatch Problem Under Power Flow Congestion
Yun Liu, Student Member, IEEE, Huanhai Xin, Member, IEEE, Zhihua Qu, Fellow, IEEE,
and Deqiang Gan, Senior Member, IEEE

Abstract—The shortcoming of conventional power system outputs of the GUs deviate from the most ED schedule to
operation is manifest: economic dispatch (ED) only optimizes maintain system stability. For this reason, the concept of
generation schedule periodically, and any power change of dis- real-time ED (RTED) is proposed almost two decades ago,
tributed generators or system loads can make the power outputs
of the generating units deviate from the optimality. To this which is required to execute very frequently (e.g., from a
end, a distributed real-time ED approach is introduced, which few automatic generation control (AGC) cycles to a few
integrates ED and secondary frequency control, and considers minutes) and communicate with the AGC [3]. Even so, the
power balance, power flow limit and generation limit constraints execution time cannot be made arbitrarily small since the long-
at the same time. With distributed measure of system frequency distance communication and data processing at the energy
and line flows, the proposed approach can achieve ED in the real-
time and the system frequency can be stabilized to the nominal management system still take time. Moreover, conventional
value. Simulation results in IEEE 9-bus test system validate the ED and secondary frequency control, which are implemented
effectiveness and efficiency of the proposed approach. in a centralized manner, are susceptible to single-point failures
Index Terms—Consensus, economic dispatch, power flow con- and cannot meet the requirement of plug-n-play [4].
gestion, distributed control, multi-agent system. The concept of smart grid, which refers to a modernized
grid controlled by computer-based automation and two-way
communication structures [5], provides a new tool in power
I. I NTRODUCTION system design. Recently, distributed algorithms based on the
Power systems have always been operated in a hierarchical cyber-physical system (CPS) have been extensively applied to
manner [1]. Economic dispatch (ED) minimizes the total power systems. In [6], the authors initiatively apply consensus
generation cost under a set of constraints (e.g., power balance, algorithm to ED problem by driving the incremental cost of
generation limit, and power flow limit constraints, etc.) based each GU to a common value. However, this method is not
on load and generation estimation [2]. It determines the power completely distributed since the aggregated power output of
generation setpoints of the generating units (GUs) in the next the GUs is assumed to be known to the leader GU. Reference
period of time (e.g., 15 min or 30 min). However, due to the [7] solves the same problem by distributively and collectively
unavoidable estimation error as well as the power fluctuation estimating the mismatch between demand and aggregated
of distributed generators (DGs) and system loads, the sole power output of the GUs, which is used as a feedback mech-
effort of ED cannot guarantee stable system operation. anism to adjust the power output of each GU. Reference [8]
Frequency control is essential to maintain frequency stabili- considers the transmission loss in the power balance constraint
ty. In a short time-scale, primary frequency control adjusts the for the first time. Different from the above literature, reference
power output of each GU in response to frequency deviation [4] proposes the concept of online optimal generation control,
according to the power/frequency droop curve, which leads which seamlessly integrates ED and frequency control, and
to a steady-state frequency deviation. In a long time-scale, can restore system frequency and minimize the total genera-
system frequency is restored to its nominal value by adjusting tion cost simultaneously. A distributed gradient algorithm is
the power generation setpoints through secondary frequency introduced to solve this problem. Since the speed governor
control. is redesigned to provide signal to the distributed controller,
However, certain shortcomings exist in the above scheme. rather than the GU directly, this method is not compatible
Due to the power change of DGs and loads, the power with existing synchronous generator control structure.
All of the aforementioned works have a common limita-
Y. Liu, H. Xin, and D. Gan are with the department of the Electri- tion: they only consider power balance and generation limit
cal Engineering, Zhejiang University, Hangzhou 310027, China (e-mails: constraints. In practice, power flows can never exceed their
[email protected]; [email protected]; [email protected]). Their work is
jointly supported by National High Technology Research and Development thermal limits, otherwise, it will result in overheat of the
Program of China (2015AA0500695) and National Nature Science Foundation power lines and contingencies may well happen to the grid.
of China (No.51177146). Y. Liu is currently studying at University of Central Therefore, power flow limit constraints must be considered in
Florida, while H. Xin was a postdoctor at University of Central Florida.
Z. Qu is with Electrical and Computer Engineering, University of Central power system dispatch/control [2], [9].
Florida, Orlando, FL 32815 USA (e-mail: [email protected]). His work is supported Based on the distributed cooperative control theory [10], a
in part by US National Science Foundation under grants ECCS-1308928 RTED approach, which integrates ED and secondary frequen-
and CCF-0956501, by US Department of Energys award DE-EE0006340,
by US Department of Transportations award DTRT13-G-UTC51, by L-3 cy control/AGC, is introduced in this paper. This approach
Communications contract 11013I2034, and by Leidos contract P010161530. is applicable to synchronous generators, DGs, as well as

978-1-4673-8040-9/15/$31.00 ©2015 IEEE

 2

virtual power plants [11]. Different from conventional ED, the Distributed RTED
Communication Link
proposed one runs in the real-time in a close-loop fashion, Communication
GU Control Signal
Network
utilizing distributively measured signals (i.e., the frequency Power Line
l ( l ) , m (jkl )
deviation and power flows) to dynamically adjust the power l (i ) , m (jki ) Speed Governor
(l Î Ni )

Power System
generation setpoints of the primary frequency control of GUs. Pi 0
f
(P , f )
0
Distributed i 0 Pi
As a result, the frequency deviation converges to zero while Controller #i GUi
the generation cost is minimized under power balance, power F jk - F jk
Pi

flow limit, and generation limit constraints. Df

Distributed
The proposed approach has several distinct features: Measure to GUi
• Power flow limit constraints are handled by using dis-
tributed measure of power flows of the CPS, which avoids Fig. 1. Architecture of the distributed control
the estimation of global information, leads to a faster
convergence, and is also in line with the trend of smart
grid. distributively measuring power system quantities (as shown in
• By the integration of ED and secondary frequency con- Fig. 1), so that the following control objectives are achieved.
trol, the frequency deviation is always restored to zero, Firstly, the frequency deviation converges to zero at the
while the GUs are always kept at the most ED schedule, equilibrium, i.e.,
regardless of loads/DGs changes. lim ∆f = 0. (3)
• It is an upper-level control, which only adjusts power t→∞

generation setpoints of the speed governors and thus is Secondly, the generation cost is minimized at the equilib-
compatible with the existing GU controllers. rium, i.e., when (3) is attained, Pi = Pi0 (for all i ∈ Sgen )
The rest of the paper is organized as follows. Section II converge to the optimal solution of the ED problem as
formulates the problem of RTED. Section III presents the P
min f (P ) = i∈Sgen fi (Pi ) (4)
distributed RTED approach. Simulation results are included
in Section IV, and conclusions are drawn in Section V. s.t.
P
Pload + Ploss − i∈Sgen Pi = 0 (5)
II. P ROBLEM F ORMULATION
Pimin ≤ Pi ≤ Pimax , ∀i ∈ Sgen (6)
In a power system with synchronous generators, the dynam-
ics of system frequency is governed by the swing equation [1], Fij ≤ F̄ij , ∀(i, j) ∈ Sline (7)
which can be expressed as
The total generation cost is to be minimized in (4). P =
d∆f 1 X
= ( Pi − Pload − Ploss ), (1) [P1 , · · · , PNgen ]T is the decision vector, where Ngen is the
dt 2H cardinality of Sgen , and Pi is the power output of the ith
i∈Sgen
GU. The cost of the ith GU fi (Pi ) is a convex function, and
where ∆f = f − f0 , f , and f0 = 1 p.u. are the frequency without loss of generality, it is approximated as a quadratic
deviation, system frequency, and nominal frequency, respec- function as [9]
tively; H is the combined inertia constant of the synchronous
generators and turbines (p.u.); Sgen is the set of all GUs; fi (Pi ) = ai Pi2 + bi Pi + ci , (8)
Pi , Pload , and Ploss are the power output of the ith GU, the
aggregated load consumption, and power loss across the power where ai , bi , and ci are cost coefficients.
lines, respectively. Eq. (5) is the power balance constraint. In this paper, we
The speed governor provides primary frequency control to actually use ∆f as a feedback signal to reach the power
each GU (when a GU refers to a DG or a virtual power balance in the grid, which avoids the estimate of Pload and
plant, a virtual governor can be designed to have the same Ploss as is done in [8]. The attainment of (3) actually indicates
function [12]). The power output Pi is determined by the (5), but it is still included here for the integrity of the ED
power/frequency droop curve described as problem and the ease of illustrating the proposed approach in
the next section.
Pi = Pi0 − kDi ∆f, (2) Eq. (6) is the generation limit constraint of the ith (∀i ∈
Sgen ) GU, where Pimin and Pimax are the lower and upper
where Pi0 is the power generation setpoint of the ith DG, and bounds of its power output, respectively.
kDi is a predefined frequency droop coefficient. Eq. (7) is the power flow limit constraint of Line (i, j)
Conventionally, the value of Pi0 is determined in an open- (∀(i, j) ∈ Sline , where Sline is the set of power lines in the
loop manner by the ED periodically, and during each period, grid). Fij and F̄ij are the actual power flow from bus i to j
Pi0 of the GUs participating in secondary frequency control and its corresponding thermal limit, respectively. There is no
is adjusted in a close-loop manner to bring the frequency lower limit on Fij , such as Fij > −F̄ij , since Fij and Fji are
deviation back to zero with the help of AGC. considered separately. Fij can be calculated as
In this paper, the distributed RTED approach determines Pi0
of each GU by exchanging information with its neighbours and Fij = Gij Vi [Vi − Vj cos(δi − δj )] − Vi Vj Bij sin(δi − δj ), (9)
 3

where Gij and Bij are the conductance and susceptance of distributed controller of each GU. Once the structure of the
Line (i, j), Vi and δi are the voltage magnitude and phase power grid changes, B can be updated in a peer-to-peer fashion
angle of the ith bus, respectively. These values can be obtained immediately.
by iteratively solving the AC power flow with, e.g., Newton- In what follows, λ and µjk (∀(j, k) ∈ Sline ) are calculated
Raphson method [1]. distributively, so that Pi0 can be calculated from (12).
Define λ = [λ(1) , · · · , λ(Ngen ) ]T and xλ =
(1) (Ngen ) T (i)
III. T HE D ISTRIBUTED RTED A PPROACH [xλ , · · · , xλ ] , where λ(i) and xλ are the estimate of
Define the Lagrangian function of the ED problem (4)-(7) λ and internal state at the ith GU, respectively. Frequency
as deviation ∆f , which can be measured locally, is used as a
P feedback signal to drive the system to stability, so that (11f)
L = f (P ) + λ(Pload + Ploss − i∈Sgen Pi ) can be satisfied.
+ i∈Sgen νi1 (Pi − Pimax )
P
(10) For the ith GU, its λ(i) evolves as
+ i∈Sgen νi2 (Pimin − Pi )
P
(i)
P
+ (i,j)∈Sline µij (Fij − F̄ij ), λ(i) = xλ − kP ∆f, (14a)
(i)
X
where λ, νi1 , νi2 (∀i ∈ Sgen ), and µij (∀(i, j) ∈ Sline ) are ẋλ =α aij (λ(j) − λ(i) ) − kI ∆f, (14b)
the Karush–Kuhn–Tucker (KKT) multipliers. j∈Ni
Lemma 1: In the steady state (∆f → 0), where α, kP , kI > 0 are the control gains. Ni is the
[P10 , · · · , PN0 gen ]T = P is the optimal solution of the neighbour set of the ith GU, which means that if j ∈ Ni ,
ED problem if it satisfies KKT conditions in (11) [13]. the jth GU can send its information to the ith GU. Note that

λ − bi − νi1 + νi2 − (j,k)∈Sline µjk djk
P i ∈ Ni for all i ∈ Sgen , i.e., there is a self loop in every agent.
 0 i
 Pi =

 (11a) For any choice of ξij > 0, aij can be initialized as

 2a i
ξij dij

νi1 (Pi0 − Pimax ) = 0, νi2 (Pimin − Pi0 ) = 0 (11b)



 aij = P ,
l∈Sgen ξil dil

 P min ≤ P 0 ≤ P max , ∀i ∈ S

(11c)
i i i gen

 µjk (Fjk − F̄jk ) = 0 (11d) where dij = 1 if the jth GU is the neighbor of the ith GU,
and dij = 0 if otherwise.




 Fjk ≤ F̄jk , ∀(j, k) ∈ Sline (11e)

P 0 The sufficient condition for the convergence of (14) is that

 i∈Sgen i
 P − P load − P loss = 0 (11f)

 the communication network should be strongly connected,
νi1 , νi2 , µjk ≥ 0, ∀i ∈ Sgen , ∀(j, k) ∈ Sline (11g) which means that it is possible to reach any agent starting
from any other agent by traversing edges in the direction(s)
where (11a) follows from ∂L/∂Pi |Pi =Pi0 = 0, and djk
i is equal
in which they point [10]. It can be verified that in the steady
to ∂Fjk /∂Pi .
state, ∆f = 0 and λ(i) = λ(j) for all i and j.
It follows from KKT conditions (11a), (11b), and (11c) that
According to the KKT conditions (11d) and (11e), when
λ − bi − (j,k)∈Sline µjk djk
P
Fjk < F̄jk , its corresponding multiplier µjk = 0. Otherwise,
 
0 i min max
Pi = SAT , Pi , Pi , µjk > 0 and its value needs to be distributively calculated.
2ai
(12) This is achieved by making use of the abundant measuring
where the saturation function is defined as and communication devices in the CPS. Measuring devices

 min if x < min are installed at each bus (except the generator buses) of the
SAT(x, min, max) = max if x > max . power system, measuring the power flows starting from the

x otherwise bus. The thermal limit of each line from that bus is also stored
at the corresponding measuring device. To avoid long-distance
In high/medium voltage networks, djki can be accurately communication, its information only needs to be transmitted
approximated as to at least one geographical neighbour GU, as depicted in Fig.
 2.
 1 ([B̃ −1 ] − [B̃ −1 ] ) if i 6= N
ji ki bus Assume that power flow Fjk is to reach its thermal limit
djk = X , (13)
i
 0 jk otherwise
(1) (N ) (i)
F̄jk . Define µjk = [µjk , · · · , µjk gen ]T , where µjk is the
estimate of µjk at the ith GU. For the ith GU, define
where Nbus is the total number of buses, Xjk is the reactance
(i) P (l) (i) (i)
of Line (j, k), and matrix B̃ equals the susceptance matrix vjk = γ l∈Ni ail (µjk − µjk ) + gjk β(Fjk − F̄jk ),
B excluding the Nbus th column and row. The entry in the (i)
susceptance matrix B is defined as its corresponding µjk should evolve according to
(
(i) (i)

−1/X ij if i 6= j (i) 0 if µjk = 0 and vjk < 0
Bij = P . µ̇jk = , (15)
l6=i 1/Xil if i = j (i)
vjk otherwise
The derivation of (13) is explained in detail in the Appendix. (i)
It should be noted that, to estimate djk locally at each where β and γ are constant gains. gjk = 1 if Fjk is known
i
(i)
GU, the information of matrix B needs to be stored in the to the ith GU and gjk = 0 if otherwise.
 4

TABLE II
6 DM6
2 DM8 8 Load C 3 DC3 ACTIVE /R EACTIVE P OWER C ONSUMPTION OF THE L OADS
DC2

DM7 Active Power Reactive Power

7 Load# Bus#
GU2 T2 T3 GU3
(MW) (MVar)
DM9 DM5 A 9 125 50
5
9 B 5 90 30
Load A Load B C 7 100 35
4 DM4
TABLE III
T1 PARAMETERS OF THE P OWER L INES
Measurement Flow
1
Communication Link GU1 To From Resistance Reactance Thermal Limit
Line#
DMi Measuring Device at Bus i
Bus# Bus# (p.u.) (p.u.) (MW)
1 1 4 0 0.0576 250
DC1
DCi Distributed Controller at GU i 2 4 5 0.017 0.092 250
3 5 6 0.039 0.17 150
4 3 6 0 0.0586 150
Fig. 2. IEEE 9-bus test system
5 6 7 0.0119 0.1008 150
6 7 8 0.0085 0.072 70
7 8 2 0 0.0625 250
(i) 8 8 9 0.032 0.161 250
In the steady state, all the µjk s converge to a consensus
9 9 4 0.01 0.085 250
value, which guarantees that Fjk = F̄jk .
Remark 1: In a power system, only a small number of power
lines might reach the power flow limits (otherwise, the power
gird needs upgrade). Since most of the µjk s equal zero, it is In case 1, the GUs without generation limits are considered.
trivial to transmit and calculate all of them, which increases the Initially, the power system is at its equilibrium. At 10 s, Load
pressure on the communication system. µjk is distributively C increases from 100 M W to 150 M W all of a sudden.
calculated only when its corresponding power flow limit is Fig. 3 shows the responses of the system. Fig. 3(a) shows
active. To make the other GUs identify the overloaded line, that system frequency decreases to 0.986 p.u. before restoring
its indices should also be broadcast to them in a peer-to-peer to 1 p.u. since power generation of the GUs cannot react to
manner. the load change at once. Fig. 3(b) shows the responses of the
The property of the proposed approach can be summarized GUs power outputs. The aggregated power output increases
as the following theorem. to reach the power balance again, but the power output of
Theorem 1: The power generation setpoint of each GU in (2) GU2 decreases due to power flow congestion in Line (8, 7).
determined by the distributed RTED approach (12)-(15) can Table IV compares the results of the proposed approach with
restore the frequency deviation ∆f to zero, while the power AC optimal power flow (OPF) calculated by MATPOWER
outputs of the GUs converge to the optimal solution of the ED at the new equilibrium, which indicates that the proposed
problem (3)-(7) in the steady state. method is accurate enough to converge to the optimal solution.
The proof of Theorem 1 is not included due to the limit of The variations of power flows are shown in Fig. 3(c). All
space. the power flows are within their thermal limits except Line
(8, 7). F87 exceeds 70 M W right after the load change. As
its corresponding µ87 increases (as indicated by Fig. 3(d)), F87
IV. S IMULATION R ESULTS decreases until it is bounded by its thermal limit. It should be
Simulation results with the 345 kV IEEE 9-bus test system noted that this overload will not damage the power line for
are presented in this section to validate the effectiveness of the two reasons: 1) It is only transient, e.g., it lasts 5 s in this
proposed approach. The single-line diagram and configuration case; 2) The thermal limit on the nameplate always includes
of the communication and measuring devices are shown in Fig. a margin.
2. The cost parameters and generation limits of the three GUs
TABLE IV
are shown in Table I. The active/reactive power consumption C OMPARISON OF THE R ESULT (W ITHOUT G ENERATION L IMITS )
of the loads is shown in Table II. The parameters of the
power lines are shown in Table III, where the thermal limit AC Proposed Relative
of line (7, 8) is deliberately chosen a small value to make its OPF (MW) Approach (MW) Error (%)
GU1 108.21 106.37 -1.70
corresponding thermal limit constraint active. GU2 124.44 126.24 1.44
GU3 135.86 136.62 0.56
TABLE I
PARAMETERS OF THE GU S
In case 2, the same simulations are performed except that
Cost Parameter Pimin Pimax kDi
GU# Bus#
ai bi ci (MW) (MW) (MW/Hz) generation limit constraints are considered. The result after
1 1 0.11 5 150 30 250 6 convergence is also compared to that of AC OPF (Table V),
2 2 0.085 1.2 600 30 300 5 which shows that the power output of GU3 reaches its upper
3 3 0.1225 1 355 30 120 6 bound and the proposed approach can almost converge to the
optimal solution.
 5

System Frequency (p.u.) Power Outputs of the GUs (MW)

1.01 160
GU1
Remark 2: According to [14], Assumptions 1 and 2 gen-
1.00 140
GU2
GU3 erally hold in high/medium voltage networks. Assumption
0.99 120
3 can also be strengthened with distributed reactive power
0.98 100
optimization methods [15].
With slight abuse of notation, we define P =
0.97 80
0 20
Time (Sec)
40 60 0 20
Time(Sec)
40 60 [P1 , · · · , PNbus ]T and δ = [δ1 , · · · , δNbus ]T , where Pi and
(a) (b) δi are the net power injection and voltage phase angle at the
Load Flow of the Power Lines (MW) Calculation of the Multipliers at the GUs
90 27 240 ith bus, respectively. According to DC power flow [2],
l m87

26 l
(1)

(2)
160 P = Bδ, (16)
l
Line 4-9
60 (3)
1
B

Line 8-9 l
Line 6-7 25
(1)
m87 80 Fjk = (δj − δk ), ∀(j, k) ∈ Sline (17)
Line 6-5
Line 4-5
m87
(2)

(3)
Xjk
Line 8-7 m87
30
0 20 40 60
24
0 20 40 60
0
Suppose that the Nbus th bus is the reference bus with
Time (Sec) Time(Sec)
(c) (d) δNbus = 0. Eq. (16) can be simplified as

Fig. 3. Responses of the system to a load change (without generation limit [P1 , · · · , PNbus −1 ]T = B̃[δ1 , · · · , δNbus −1 ]T , (18)
constraints) (Nbus −1)×(Nbus −1)
where B̃ ∈ R is a full-rank matrix. Pre-
TABLE V multiplying the two sides of (18) by B̃ −1 leads to
C OMPARISON OF THE R ESULT (W ITH G ENERATION L IMITS )
[δ1 , · · · , δNbus −1 ]T = B̃ −1 [P1 , · · · , PNbus −1 ]T , (19)
AC Proposed Relative Therefore,
OPF (MW) Approach (MW) Error (%)
[B̃ −1 ]ij

GU1 140.22 139.93 -0.21 ∂δi if j 6= Nbus
GU2 107.80 108.64 0.78
= (20)
∂Pj 0 otherwise
GU3 120.00 120.00 0.00
Eq. (13) directly follows from (17) and (20).
R EFERENCES
V. C ONCLUSION
[1] P. Kundur, Power system stability and control. McGraw-hill New York,
A distributed RTED approach, which integrates ED and 2006.
[2] D. Gan, D. Feng, and J. Xie, Electricity Markets and Power System
secondary frequency control, is proposed in this paper. The Economics. CRC Press, 2013.
proposed approach considers power balance, generation limit, [3] J.-Y. Fan and L. Zhang, “Real-time economic dispatch with line flow
and power flow limit constraints together, which makes it and emission constraints using quadratic programming,” IEEE Trans.
Power Syst., vol. 13, no. 2, pp. 320–325, May, 1998.
more practical and competitive to its centralized counterpart. [4] W. Zhang, W. Liu, X. Wang, L. Liu, and F. Ferrese, “Online optimal
Despite the continual change of system loads and DGs power, generation control based on constrained distributed gradient algorithm,”
the frequency deviation always converges to zero, while the IEEE Trans. Power Syst., to appear.
[5] X. Fang, S. Misra, G. Xue, and D. Yang, “Managing smart grid
generation cost is minimized in the real-time with only a trivial information in the cloud: opportunities, model, and applications,” IEEE
error. This is more economically efficient than the conventional Netw., vol. 26, no. 4, pp. 32–38, Jul./Aug., 2012.
scheme, where frequency control makes the power generation [6] Z. Zhang and M.-Y. Chow, “Convergence analysis of the incremen-
tal cost consensus algorithm under different communication network
of GUs deviate from the most ED schedule. Simulation results topologies in a smart grid,” IEEE Trans. Power Syst., vol. 27, no. 4, pp.
from the IEEE 9-bus test system demonstrate the effectiveness 1761–1768, Nov., 2012.
of the approach. [7] S. Yang, S. Tan, and J.-X. Xu, “Consensus based approach for economic
dispatch problem in a smart grid,” IEEE Trans. Power Syst., vol. 28,
no. 4, pp. 4416 – 4426, Nov., 2013.
[8] G. Binetti, A. Davoudi, F. L. Lewis, D. Naso, and B. Turchiano, “Dis-
A PPENDIX tributed consensus-based economic dispatch with transmission losses,”
For any congested Line (j, k), djk i = ∂Fjk /∂Pi should be
IEEE Trans. Power Syst., vol. 29, no. 4, pp. 1711 – 1720, Jul., 2014.
[9] A. J. Wood and B. F. Wollenberg, Power generation, operation, and
known to all GUs, so that their Pi0 can be calculated from control. John Wiley & Sons, 2012.
(12). It is difficult to precisely calculate djk
i since the classic
[10] Z. Qu, Cooperative control of dynamical systems. Springer, 2009.
[11] D. Pudjianto, C. Ramsay, and G. Strbac, “Virtual power plant and system
AC power flow, which consists of both active and reactive integration of distributed energy resources,” IET Renew. Power Gener.,
power flows, is a nonlinear problem and can only be solved vol. 1, no. 1, pp. 10–16, Mar. 2007.
iteratively. To this end, DC power flow, which simplifies AC [12] H. Xin, Y. Liu, Z. Wang, D. Gan, and T. Yang, “A new frequency
regulation strategy for photovoltaic systems without energy storage,”
power flow and makes the problem linear, is considered in IEEE Trans. Sustainable Energy, vol. 4, no. 4, pp. 985 – 993, Oct.,
estimating djki . DC power flow is accurate enough under three 2013.
assumptions: [13] S. Boyd and L. Vandenberghe, Convex optimization. Cambridge
university press, 2009.
Assumption 1: The resistance of each line can be neglected, [14] K. Purchala, L. Meeus, D. Van Dommelen, and R. Belmans, “Usefulness
i.e., R  X, thus Ploss can be ignored. of dc power flow for active power flow analysis,” in proc. Power
Assumption 2: Voltage angle difference is small across each Engineering Society General Meeting, Jun., 2005, pp. 454–459.
[15] A. Maknouninejad and Z. Qu, “Realizing unified microgrid voltage
line, so that sin(δjk ) ≈ δjk and cos(δjk ) ≈ 1. profile and loss minimization: A cooperative distributed optimization
Assumption 3: Voltage profile is unified, i.e., the voltage and control approach,” IEEE Trans. Smart Grid, vol. 5, no. 4, pp. 1621
magnitude of each bus is 1 p.u. – 1630, Jul., 2014.