Economic dispatch : A mixed-integer linear model for

thermal generating units

Masoud Javadi and Turaj Amraee, Senior Member, IEEE

Electrical Engineering Faculty, K.N. Toosi University of Technology
Tehran, Iran
[email protected] [email protected]

Abstract— Economic Dispatch (ED) is a major study in load flow formulation

operation of a power system. Using the ED study, the power
Imaginary part of grid admittance matrix
system operators determine the optimal power production of
generating units to achieve the minimum total generation cost. To Grid shunt susceptance matrix
reach the global optimum solution of ED, this paper develops a Real part of grid admittance matrix
linear ED model using the mixed integer linear programming
(MILP) formulation. The non-linear equations of the full AC ̅ Breakpoints on the voltage angle difference
load flow model are linearized using Taylor series expansions and Breakpoints on the cosine function
piecewise linear (PWL) approximation techniques. The IEEE 14-,
30-, 39-, 57-, 118-, and 300-bus standard test systems are
employed to evaluate the accuracy and efficacy of the presented
Variables
linear AC load flow model. The simulation results of the MILP-
based ED model are compared with those obtained by the / Active and reactive power demand
conventional non-linear programming (NLP) approach. The / Active and reactive power produced by
developed model is optimized using GAMS software. generators
Active power generation of lth linear
Keywords— linear AC power flow; economic dispatch;
segment
optimization;
/ Minimum active and reactive power
NOMENCLATURE generation of generators
Indices / Maximum active and reactive power
generation of generators
, Index of buses Active power flow of lines
Index of piecewise segments in linear fuel cost Maximum capacity of transmission lines
function
Index of breakpoints in linear AC power balance / Bus voltage magnitude and angle
model / Minimum and maximum values of
voltage magnitude
Sets Slope of lth segment in linear fuel cost
function
Ω Set of all nodes
Generation cost of thermal units
Ω Set of load points
Approximated voltage angle
Ω Set of generating units
Approximated cosine function
Ω Set of transmission lines
SOS2 variable for piecewise linear
approximation in the linearized AC load
Parameters flow formulation
Error of approximated variables
Total number of buses
Total number of load points
Total number of generators I. INTRODUCTION
Number of line segments for linear fuel cost curves Economic and secure operation of electric energy systems
is a priority for power system operators and owners. The main
Number of breakpoints used in the linearized AC
cost of operation is dominated by the cost of thermal

978-1-5386-5186-5/18/$31.00 ©2018 IEEE

generating units. Two main operation studies include the uncertainties of steady state operation studies could be found in
economic dispatch (ED) and unit commitment (UC) problems. [13-15].
In a daily UC study the on/off status of generators along with
the optimal levels of power generation and reserve are In the present paper, the non-linear load flow formulation is
determined. In ED study, by assuming the on/off status of linearized at a given operating point in polar coordinates. The
generating units the optimal power generation of generating cosine functions are then linearized using Special Ordered Sets
units is determined with the aim of achieving the minimum (SOS). The proposed linear ED model is solved using
cost of generation. In other words, the aim of the ED study is to commercial MILP solvers.
minimize the total cost of generation especially the fuel cost of The remainder of this paper is arranged as follows. Detailed
thermal generating units by controlling the generation level. formulation of the proposed linear ED is presented in section
The ED problem is optimized with considering different II. The simulation results are given in section III. Finally, the
constraints including the constraints of generating units and the paper is concluded in section IV.
network. The conventional ED is inherently a non-linear
optimization problem. The nonlinearities arise from the II. FORMULATION
quadratic cost function of generating units and the full AC The goal of the ED problem is to minimize the total cost of
power flow equations. active power production of generating units considering
ED study is a special kind of optimal power flow (OPF) operational constraints. The objective function of the ED
problem which aims at the minimizing total production cost of problem is introduced as follows:
thermal generating units. Like other steady state OPF studies,
the AC power flow equations are a major part of the
= ( ( ) + + ) (1)
optimization model [1-3].
ED model is inherently a nonlinear optimization model. where , , and refer to the constant parameters of the fuel
The techniques proposed for optimizing ED study include the cost function. The following equations are linear equivalents
mathematical programming-based techniques and evolutionary of (1) using the PWL method as described in [9].
techniques. For linear optimization problems, the mathematical
optimization techniques give the optimal solutions while for
nonlinear optimization technique they may fail to give the = + (2)
global optimal solution. Evolutionary techniques may optimize
a vast range of linear and nonlinear problems with a population
based search algorithm. However these techniques will not = + (3)
guarantee the globality of the solution and are time consuming
for large scale problems [4]. Mathematical programming-based −
techniques include interior point [5], Benders decomposition 0≤ ≤ (4)
[6], Semidefinite programming [7], linear and non-linear
programming techniques, and etc. Heuristic optimization The group of constraints given by (5)-(8), relates to the
techniques such as genetic algorithm have been used in linearized load flow equations. It should be noted that line
economic dispatch problems [8]. flows are Taylor-expanded at a given operating point (e.g. =
The conventional NLP-based models of ED may converge 1 . . , and = 0°).
to local solutions. Quadratic fuel cost functions of thermal
generators are incremental and convex. Hence, a piecewise
− = ( ( + + cos( ) − 2)
linear (PWL) approximation technique as, for example,
introduced in [9], can be used to simplify the solution process. (5)
Furthermore, various linearization techniques for linearizing + ) ∈Ω
AC power flow equations has been developed in previous
researches.
A linear model of AC power flow model is introduced for
the islanding problem in [10, 11]. This model is based on TABLE I. LINEAR EQUIVALENTS OF NONLINEAR TERMS IN
AC POWER FLOW EQUATIONS
Taylor series expansion of line flows at the normal operating.
Other applications of linear AC power flow model have been
investigated in some power system studies such as an MILP Non-linear term Linear equivalent
model in the under-voltage load shedding problem using
Taylor series expansion and a piecewise linear technique for cos( ) + + −2
cosine terms [12]. The utilization of the MILP techniques
guarantees a global optimal solution of the ED study. Another sin( )
challenge in the ED study is the uncertainties of some input 2 −1
parameters such as load demand. The focus of this paper is the
deterministic formulation of ED. The details of handling the
 TABLE II. ERRORS OF THE DEVELOPED LINEAR LOAD FLOW
MODEL

IEEE (  ) (  ) (  )
(  )
test grid . .

14-bus 0.0014 0.0114 0.0016 0.0036

30-bus 0.0002 0.0015 0.0004 0.0007

39-bus 0.0016 0.0832 0.0097 0.0336

57-bus 0.0057 0.0512 0.0042 0.0118

118-bus 0.0002 0.0166 0.0073 0.0053

300-bus 0.0013 0.0602 0.0141 0.0174

, =1 (12)

Fig. 1. PWL approximation of cosine function using SOS2 variables The final group of constraints given by (13)-(16) are system
operational limits including voltage magnitudes of buses,
reactive power generations of generators, limits of active power
output of generators, and line flows and expressed as follows.

≤ ≤ ∈Ω (13)
− = ( − ( + + cos( )
(6) ≤ ≤ ∈Ω (14)
− 2)) ∈Ω
≤ ≤ ∈Ω (15)
. = −(2 − 1) + − ≤ , ≤ ∈Ω (16)
, ,
+ + cos( ) − 2 + ∈Ω (7)
III. SIMULATION RESULTS
. = −(2 − 1)( − )+
In this section, first the efficacy of the proposed linear AC
− + + cos( )−2 ∈Ω (8) power flow model is verified for different IEEE test grids. The
The equivalent approximation terms are summarized in approximation errors of the system variables provided by the
Table I. According to the above equations (i.e. (5)-(8) ), the proposed MILP model including voltage magnitude of load
model is still non-linear due to the presence of the cosine points, reactive power generation/absorption of thermal
generating units, and active and reactive line flows are
terms. In order to linearize these non-linear terms, is
calculated to be compared with NLP-based model. Considering
substituted for cosine terms. Then, a PWL technique using the results of the NLP-based load flow model as the
special ordered set of type 2 is applied [16, 17]. The proposed benchmark, the errors of the system variables are calculated
piecewise linear function is depicted in Fig. 1. The following using the mean absolute error (MAE) criterion as follows.
group of constraints are linear equivalents of the non-linear
cosine terms.
1
(  ) = | − | ∈Ω (17)
= , ̅ (9)
1
(  ) = − ∈Ω (18)
= cos( ̅ ) (10)

1
= (  ) = . − . ∈ Ω (19)
, (11) . 2
 4
x 10
2.5
NLP
MILP
Total generating power (MW) 2

1.5

0.5

0
IEEE 14 bus IEEE 30 bus IEEE 39 bus IEEE 57 bus IEEE 118 bus IEEE 300 bus

Fig. 2. Total generating power of test cases in the ED study

the quadratic fuel cost function of thermal units and 33

1
(  ) = . − . ∈Ω (20) breakpoints are also used for PWL approximation of cosine
. 2 terms. Higher number of segments may cause to more
computational burden.
The approximation errors of the proposed linear model (in
per unit) have been reported in Table II. Based on Table II, the The obtained results have been reported in Table III.
approximation errors are relatively small and negligible. According to Table III, the proposed method gives reasonable
Different benchmarks including standard IEEE networks, are errors. In all simulated test cases the relative errors are less
used to evaluate the performance of the proposed linear power than 1 %. It is noted for real time applications it is required to
flow model. The data of the utilized test systems could be reduce the linearized segments. Reducing the number of
found in [18]. The proposed linear model of ED is solved using linearized segments deteriorates the overall accuracy of the
the CPLEX solver [19] in GAMS with an OPTCR (relative proposed linear ED. The total generating power is also depicted
gap) of 0%. The CONOPT solver is also utilized to solve the in Fig. 2. It can be seen that the total power generation obtained
conventional non-linear model. It should be noted that different by the proposed MIP-based ED model are very close to the
NLP solvers are also used to verify the globality of the solution total amount of power generation obtained by the NLP model.
obtained by the CONOPT solver. In the NLP-based methods, a In other words assuming the same amount of total load
local optimum may be obtained instead of the global optimum. demand, the proposed linear ED approximate the total active
For instance, MINOS solver results in local optimal solution in power loss with a reasonable accuracy.
IEEE 300-bus system. The cost function in this case is The focus of this paper is the optimization of active power
725560.863$. Here, 20 line segments are used to approximate production. For other applications such as reactive power
management the proposed linearization technique may be
TABLE III. THE RESULTS OF ED USING THE PROPOSED LINEAR revised to give reasonable errors especially at heavy load
MODEL AND THE NONLINEAR MODEL conditions. Another point is the efficacy of the proposed linear
Total cost ($)
ED method under the network architect or topology. Although
IEEE Relative
test grid error all the errors reported in Table III are lower than 1 %, however
MILP NLP
the error of the linear technique in IEEE 14-bus test system is
14-bus 8039.471 8081.204 5.164×10-3 higher than other test grids. Indeed the IEEE 14-bus test system
30-bus 575.293 574.865 7.445×10-4 has a radial topology in which the load and generation have
been concentrated in two remote centers. Therefore the efficacy
39-bus 41957.648 41973.397 3.752×10-4 of the proposed linear technique is a little lower than other test
57-bus 41683.299 41731.682 1.159×10-3 network which have a meshed topology.
118-bus 129560.050 129660.694 7.762×10-4 IV. CONCLUSION
300-bus 719912.947 719725.099 2.610×10 -4
A linear model was introduced for the ED study. The
objective function of the ED was linearized using a PWL
technique. Also the non-linear load flow formulation were [8] C.-L. Chiang, "Improved genetic algorithm for power economic dispatch
of units with valve-point effects and multiple fuels," IEEE transactions
linearized using SOS technique. It was shown that the non- on power systems, vol. 20, pp. 1690-1699, 2005.
linear ED model may fail to give the optimal solution of [9] A. J. Wood and B. F. Wollenberg, Power generation, operation, and
power generations. However, the proposed linear ED model control: John Wiley & Sons, 2012.
gives the optimal solutions with considering different limits of [10] P. A. Trodden, W. A. Bukhsh, A. Grothey, and K. I. McKinnon,
the ED study. The proposed linear ED model may be extended "Optimization-based islanding of power networks using piecewise linear
to cover a vast range of constraints in future works. AC power flow," IEEE Transactions on Power Systems, vol. 29, pp.
1212-1220, 2014.
REFERENCES [11] F. Teymouri, T. Amraee, H. Saberi, and F. Capitanescu, "Towards
Controlled Islanding for Enhancing Power Grid Resilience Considering
[1] T. Amraee, A. Ranjbar, R. Feuillet, and B. Mozafari, "System protection Frequency Stability Constraints," IEEE Transactions on Smart Grid,
scheme for mitigation of cascaded voltage collapses," IET generation, 2017.
transmission & distribution, vol. 3, pp. 242-256, 2009.
[12] M. Javadi and T. Amraee, "Mixed integer linear formulation for
[2] T. Amraee, B. Mozafari, and A. Ranjbar, "An improved model for undervoltage load shedding to provide voltage stability," IET
optimal under voltage load shedding: particle swarm approach," in Generation, Transmission & Distribution, 2018.
Power India Conference, 2006 IEEE, 2006, p. 6 pp.
[13] I. Kaffashan and T. Amraee, "Probabilistic undervoltage load shedding
[3] B. Mozafari, T. Amraee, and A. Ranjbar, "An approach for under using point estimate method," IET Generation, Transmission &
voltage load shedding using particle swarm optimization," in Industrial Distribution, vol. 9, pp. 2234-2244, 2015.
Electronics, 2006 IEEE International Symposium on, 2006, pp. 2019-
2024. [14] A. Soroudi and T. Amraee, "Decision making under uncertainty in
energy systems: State of the art," Renewable and Sustainable Energy
[4] X. Xia and A. Elaiw, "Optimal dynamic economic dispatch of Reviews, vol. 28, pp. 376-384, 2013.
generation: A review," Electric Power Systems Research, vol. 80, pp.
975-986, 2010. [15] S. Sarhadi and T. Amraee, "Robust dynamic network expansion
planning considering load uncertainty," International Journal of
[5] G. Irisarri, L. Kimball, K. Clements, A. Bagchi, and P. Davis, Electrical Power & Energy Systems, vol. 71, pp. 140-150, 2015.
"Economic dispatch with network and ramping constraints via interior
point methods," IEEE Transactions on Power Systems, vol. 13, pp. 236- [16] E. Beale and J. J. Forrest, "Global optimization using special ordered
242, 1998. sets," Mathematical Programming, vol. 10, pp. 52-69, 1976.
[6] Z. Li, W. Wu, B. Zhang, and B. Wang, "Decentralized multi-area [17] G. L. Nemhauser and L. A. Wolsey, "Integer and Combinatorial
dynamic economic dispatch using modified generalized benders Optimization. Interscience Series in Discrete Mathematics and
Optimization," ed: John Wiley & Sons, 1988.
decomposition," IEEE Transactions on Power Systems, vol. 31, pp. 526-
538, 2016. [18] R. Christie, "Power systems test case archive," Electrical Engineering
dept., University of Washington, 2000.
[7] A. M. Jubril, O. A. Komolafe, and K. O. Alawode, "Solving multi-
objective economic dispatch problem via semidefinite programming," [19] I. CPLEX, "Ilog cplex homepage 2009," Available at [online]:
IEEE Transactions on power systems, vol. 28, pp. 2056-2064, 2013. http://www. ilog. com, 2009.