You are on page 1of 6

Priority list and particle swarm optimization based

Unit commitment of thermal units including


renewable uncertainties
1
Md. Sajid Alam, 2Durga Hari Kiran B, 3Matam Sailaja Kumari
Department of Electrical Engineering
National Institute of Technology, Warangal, India-506004
1
msajid1992@gmail.com, 2bdurgaharikiran@yahoo.co.uk, 3sailaja@nitw.ac.in

Abstract—Utilizing thermal generation alone to meet the Si Start-up cost of unit i


energy demand leads to adverse effects on environment. So, to
minimize the environmental pollution, there is a need to S0i , S1i ,t Cost coefficients of Start-up of unit i
enhance the renewable energy contribution in the grid. In this PC Production cost
paper a hybrid Priority List and Particle Swarm Optimization
(PL-PSO) approach to solve Unit Commitment (UC) and AS Ancillary service
Economic Load Dispatch (ELD) of thermal units integrated Avg average
with renewable sources like wind and solar power plant is DP Dynamic programming
presented. The on/off decision of thermal units is handled by
PL technique, while Particle Swarm Optimization (PSO) I. INTRODUCTION
solves ELD. Both programs are run simultaneously, fine
tuning their solutions in search of a better solution. Wind and Unit commitment (UC) involves turning on/off of thermal
units for forecasted load of particular hour, while considering
solar power output is modeled using MATLAB simulink tool
and coding respectively. Wind and Solar uncertainties are system capacity requirements. The decision on UC also
handled using scheduling of operating reserve Ancillary includes few constraints like spinning reserve, minimum up,
down time and ramp limits [1]. After UC decision, ELD load
service. The problem formulation considers minimum up and
down time, start up cost and spinning reserve constraints. The dispatch is performed for committed units. ELD is carried out
objective is to minimize the fuel cost associated with thermal to find out the output power of each committed unit, to find
optimal fuel cost while satisfying all constraints associated
units while satisfying constraints. Case studies are performed
on 10 and 54 thermal unit systems including solar and wind with thermal units [2] [3] [4] [5]. UC is usually performed by
priority list method and can be applied for large power
generation and the results are encouraging.
systems. Among all available methods priority list technique is
Keywords—Unit Commitment(UC); Economic Load simple to understand and computationally very fast [6]. In this
Dispatch(ELD);particle swarm optimization(PSO); Priority List method incremental cost(IC) of each unit is calculated and
method (PL) based on IC value on/off decision of units is considered.
Dynamic programming based on IC is flexible but
Abbreviations and Acronyms computationally very expensive [7]. [8] has proposed optimal
N Number of units generating schedule including unit commitment using
T Scheduling hour (h) Dynamic Programming. This study also includes wind
uncertainties by weibul PDF method. [9], [10] and [11]
Power generated by Committed unit i in jth
Pij proposed UC and ELD using genetic algorithm. MD sajid
hour
alam et.al. have attempted the solution for UC and ELD
PD ( j ) Demand for hour j problem using PL-GA method [12]. UC and ELD are done by
Priority list based genetic algorithm(PL-GA) but genetic
Pw ( j ) Wind power for j th hour algorithm faced convergence problem, and computational time
Ps ( j ) Solar power for j th hour is also more [12], so, this paper attempted UC including
renewable by priority list based particle swarm optimization
Pu Extra power generation due to uncertainties (PL-PSO). In PL-PSO, UC is done based on incremental cost
and economic load dispatch by particle swarm optimization.
ai , bi , ci Fuel cost coefficients Both genetic and PSO algorithm is converging to the same
SR/SC Spinning reserve/Start up cost point and hence giving the same optimal solution. However, in
case of PL-PSO computational time decreased and also good
MUT Minimum up time
convergence is observed. After integration of renewable
MDT Minimum down time energy source, contribution of thermal generation, line
U ij ON (“1”)/OFF(“0”) status of unit i at j hour loading, transmission cost, fossil fuel consumptions, emissions
are drastically reduced.
nc Total number of combinations
978-1-4673-8848-1/16/$31.00 ©2016 IEEE
II. UC PROBLEM FORMULATION which each particle can be modify their position is given by
The objective of the UC and ELD is to minimize the fuel cost following equation (7)
of thermal units while satisfying equality and inequality Vki +1 = K *(W *Vki + c1r1 ( pbest − Ski ) + c2 r2 ( g best − Ski )) (7)
constraints. The objective function includes total running cost
and starting cost. Since UC and ELD is done for 24 hour so Where c1and c2 are constants usually equal to 2.05. They are
overall objective function is given by (1) called the cognitive parameter and the social parameter,
respectively. r1 and r2 are random numbers between 0 and 1.
A. Objective Function (r1, r2) being generated for acceleration toward pbest and gbest
24 N locations and K is called Constriction factor. The current
Min F (U ij , Pij ) = U ij Fi ( Pij ) + U ij (1 − U ij −1 )Si (1) position (search point in the solution space) can be modified
j =1 i =1 using the equation (8)
Running cost of thermal unit generally is a function of output
power of the particular unit and is given by (2) Ski+1 = Ski +Vki+1 (8)
2
Fi (Pij ) = a P +bi Pij + ci
i ij
(2) .
Constriction factor (K) increases the algorithm's ability to
Minimum up time and down time are associated with start-up converge to a good solution and is defined as:
and shut down costs respectively. In this paper the shut-down
cost is considered to be zero and the time dependent start up 2
cost is given by (3) [13] K= ; φ = c1 + c2 ; φ > 1 (9)
2 −φ − φ2 − 4
Si = S0 i + S1i ( 1 − e −( T / t ) ) (3)
Appropriate selection of inertia weight (W) in (7) provides a
balance between global and local explorations, often decreases
B. Other Constraints on UC
linearly during a run given by following equation
1. Power Balance Constraint
W max − W m in
24 N W = W m ax − * iter (10)
 U
j =1 i =1
P + Pu = PD ( j ) − Ps ( j ) − Pw ( j )
ij ij
(4) itermax
2. Spinning Reserve IV. PRIORITY LIST METHOD APPROACH TO UC
In this paper spinning reserve is considered more than 5%. Let In this method first, full load average production cost is
it be 6%, then scheduled unit of particular hour must satisfy calculated and then priority list of the units based on
the equation (5) incremental cost(IC) is developed. This is the simplest unit
24 N
commitment method for on/off decision of thermal units and is
U
j =1 i =1
P
ij i max ≥ 1.06 PD ( j ) (5) very fast and easy to implement. The unit having least IC
value is preferred first for commitment. For de-commitment
3. Thermal unit output power limitation purpose, the unit which has highest IC value is preferred
Pi min ≤ Pij ≤ Pi max (6) among all committed units. After unit commitment decision
economic load dispatch is carried out to find out optimal
Where i=1………N and j=1………24
power output so as to minimize fuel cost, while satisfying all
4. Minimum up time and down time constraints.
The minimum time for which the unit is to be ON, before
which the unit is de-committed. The minimum time for which V. RENEWABLE ENERGY UNCERTAINTIES
the unit is to be OFF, before which the unit is re-committed. Renewable energy sources are clean, green and emission free
5. Crew constraint and have least cost associated with energy production. Main
problem with renewable energy sources is their intermittent
Two units cannot be turned on at the same time because it
nature of supply. This work considered to maximize the use of
requires enough crew members to attend both the units while
renewable energy sources which are available at particular
starting up
hour, while considering renewable uncertainties. The
III. PARTICLE SWARM OPTIMIZATION uncertainties are managed by extra power generation from
thermal units. The extra power generation from thermal units,
Particle Swarm Optimization is one of the evolutionary
act as ancillary service which will take care of any short fall of
computation techniques. The algorithm starts with initializing
power. Wind generator model is created in MATLAB
a population of random solutions (particles) and searches for
simulation and solar power model is developed using
optimum by updating generations. Each particle remembers
MATLAB code. Renewable uncertainties are handled by
its own best position termed as personal best (pbest) found so
considering generation of extra power from committed
far in the exploration. Moreover, among these pbest solutions
thermal units. The amount of extra generation totally depends
each particle knows the best value which is termed as gbest.
Each particle tries to modify its position using the current upon the renewable output power, which is integrated.
velocity and the distance from pbest and gbest. Velocity with Different case studies have been done to analyze the total
production cost. Extra power generation acts as ancillary
service that helps to maintain any power shortfall upto certain   T 
extent, and hence reliability of supply increased. Vt = N s *  k *    (16)
VI. SIMULATION
  q 
  V + (I ( i ,1 ) * R s )    V + (R s * I (1 ,1 ))  (17)
A. Solar Power modelling I = I o *  exp   − 1  +  
Vt * a  R sh
     

B. Wind Power Simulink Model


The ability of a wind turbine to extract power from wind is a
function of wind power availability. Wind power at any hour
is given in (18). Fig.2 presents the wind turbine simulink
model and fig. 3 the output power.
1
Pw ( j ) = ρAV ( j )3 (18)
2

(a)

Fig. 2. Wind power Simulink model

(b)
Fig. 1. Solar PV Pannel Output a) VI characteristics b) PV characteristics

The PV cell electrical characteristic under solar Irradiance (S)


is given in terms of PV cell output current (I) and PV cell Fig. 3. Wind power Output (1.5 MW)
voltage (V) [14]. Electrical characteristics of the PV cell In fig (4) flow chart for PL-GA is presented. First of all
model, can be elaborated through the following set equations incremental cost(IC) of each generator is calculated and
(11-17). priority list is made on the basis of IC value. After that
Tn initialize the vector called state=zeros [N, 1], then switch on
V tn = N s * ( k * q
) (11)
the generator having least IC value. Check whether the
committed generator is capable of satisfying that particular
I on = I scn /((exp( V ocn /( a * V tn ))) − 1 (12) hour load or not. After feasibility test, economic load dispatch
is performed using genetic algorithm. Finally cost of supplied
  T 3   q * Eg   1 1    load of each hour is printed.
I o = I on *   n  * exp    *  −  

(13)
 T 
  a* k   Tn T   

I pvn = I scn (14)

G
( )
I pv = I pvn + K i * (T − Tn ) *   (15)
 Gn 
VII. CASE STUDIES AND RESULTS DISCUSSION
Case1: Only thermal unit is committed
Case2: Thermal unit commitment considering 10%
integration with renewable source and 5% ancillary service
management
Case3: Thermal unit commitment considering 20%
integration with renewable source and 10% ancillary service
management
Case4: Thermal unit commitment considering 30%
integration with renewable source and 15% ancillary service
management

10 thermal units data and load is taken from [9]. Forecasted


renewable power is considered from [11].

Due to page constraint only few of the results are presented.


TABLE I. UC AND ELD OF 10 THERMAL UNITS BY PL-GA/PL-PSO
Hr AS UC ELD COST
(MW) in Rs.
1 70 1100000000 455 315.04 0 0 0 0 0 0 0 0 14904
Fig. 4. Flow chat of Priority list based Genetic algorithm 2 75 1100000000 455 370.06 0 0 0 0 0 0 0 0 15866
3 59.5 1100000000 455 454.52 0 0 0 0 0 0 0 0 17345
4 95 1111000000 455 329.95 130 130 0 0 0 0 0 0 20917
5 100 1111000000 455 384.95 130 130 0 0 0 0 0 0 21878
6 66 1111000000 455 451.02 130 130 0 0 0 0 0 0 23036
7 115 1111100000 455 455 130 130 94.99 0 0 0 0 0 25463
8 120 1111100000 455 455 130 130 150.1 0 0 0 0 0 26602
9 26 1111100000 455 455 130 130 155.99 0 0 0 0 0 26726
10 140 1111111110 455 455 130 130 162 80 25 55 47.98 0 34299
11 101.5 1111111110 455 455 130 130 162 80 29.5 55 55 0 34617
12 105 1111111111 455 455 130 130 162 80 66.84 55 55 16.1 36772
13 98 1111111110 455 455130 130 162 80 0 55 30.99 0 32658
14 104 1111110000 455 455130 130 162 71.98 0 0 0 0 28861
15 120 1111110000 455 455130 130 130.06 20 0 0 0 0 27003
16 105 1111100000 455 414.95 130 130 25 0 0 0 0 0 23349
17 100 1111100000 455 359.99 130 130 25 0 0 0 0 0 22387
18 110 1111100000 455 455 130 130 40.05 0 0 0 0 0 24351
19 120 1111110000 455 455 130 130 129.99 0 0 0 0 0 27002
20 98 1111110110 455 455 130 130 162 80 0 55 30.99 0 32658
21 26 1111100000 455 455 130 130 155.97 0 0 0 0 0 26725
22 99 1111100000 455 455 130 130 28.98 0 0 0 0 0 24130
23 90 1101000000 455 404.97 0 130 0 0 0 0 0 0 19337
24 80 1100000000 455 424.9712 0 0 0 0 0 0 0 0 16827

TABLE II. COMPARSION OF 10 THERMAL UNITS UC AND ELD


COMPUTATIONAL COMPUTATIONAL PROBABILITY OF
ALGORITHM
TIME TAKEN FOR TIME TAKEN FOR CONVERGENCE
USED
10 UNITS (SECS) 54 UNITS (SECS) 10 AND 54 UNITS
PL-GA 372 694.18 0.5
PL-PSO 9.43 12.45 0.9

Tables I present the production cost for using PL-GA and PL-
Fig. 5. Flow chat of Priority list based Particle Swarm Optimization PSO. From table II it is clear that the PL-PSO algorithm
convergence is considerably fast and probability of
Fig. 5 presents the PL-PSO flow chart for UC. In PL-PSO unit convergence is also more for both 10 and 54 thermal units.
commitment is done using priority list method. After that
economic load dispatch is done using particle swarm
optimization. In this technique GA is replaced by PSO to
overcome convergence problem faced in GA. Also
computational time is less in PL-PSO method.
TABLE III. PRODUCTION COST(PC) COMPARSION OF 54 THERMAL UNITS
BY PL-GA/PL-PSO

Hr. LOAD Production Cost in Rs.


(MW) Case 1 Case 2 Case 3 Case 4
1 4200 59341 55795 51460 46336
2 3960 54951 51623 47589 42873
3 3480 46741 43877 40449 36491
4 2400 30672 28843 26579 24160
5 3000 39169 36796 33948 30735
6 3600 48737 45757 42180 38031
7 4200 59341 55795 51459 46336
8 4680 67622 62736 57825 52823 Fig. 8. UC and ELD using PL-GA/PL-PSO for 54 units
9 4920 71315 67030 61798 55637
10 5280 78224 72660 66972 60289
11 5340 79337 73721 67137 61207 Figure 8 shows the same for 54 generator system. The
12 5040 73518 68320 62316 56704 production cost further reduces as a function of Ancillary
13 4800 69005 64144 58595 53254 Services.
14 4560 64615 60107 54984 49931
15 5280 78145 72625 66158 60289 B. Production Cost comparison of 10 and 54 units
16 5400 80455 74782 68124 62132
17 5100 74642 69376 63265 57588 TABLE IV. TOTAL PRODUCTION COST(PC) COMPARSION OF 10 UNITS
18 5340 79269 73687 67137 61207 ALGORITHM CASE TOTAL AVERAGE TOTAL
TOTAL SR/AVG SR
19 5640 85094 79132 72148 64943 USED STUDY PC PC SC
20 5880 89745 83510 76191 67775 THERMAL 606376 25265.67 1211/50.45 5920
THERMAL
21 6000 91964 85702 78224 69596
AND 594220 24759.167 1736.0088/72.3337 5860
22 5400 80454 74782 68124 62132
WIND
23 5220 77022 71563 65959 59380 DP
THERMAL
24 4920 71315 67030 61798 55637 WIND
586168 24423.67 1821/75.875 5800
AND
A. Cost comparsion graph SOLAR
THERMAL 609240 25385 1979/69.9583 5520
THERMAL
AND 597660 24902.5 2138.02/89.0842 5460
PL-GA/ WIND
PL-PSO THERMAL
WIND
589865 24577.71 2168.0096/90.1254 5460
AND
SOLAR

TABLE V. TOTAL PRODUCTION COST(PC) COMPARSION OF 54 THERMAL


UNITS BY PL-GA/PL-PSO
CASE STUDY TOTAL PC AVG PC
CASE 1 1650696 68779
CASE 2 1539384 64141

Fig. 6. UC and ELD using DP for 10 units CASE 3 1410408 58767


CASE 4 1275480 53145

For 54 thermal units, unit commitment is done by PL-PSO and


PL-GA only because dynamic programming will face
convergence problem and it will take more computational
time. For 54 thermal units, IEEE 118-bus system is used.
C. Results Discussion
UC and ELD of thermal units are implemented by using DP,
PL-GA and PL-PSO. From comparison study Table IV, total
PC, total SC is decreasing and total SR is more after
Fig. 7. UC and ELD using PL-GA/PL-PSO for 10 units integration of renewable sources. When the same problem is
attempted by PL-GA and PL-PSO total SR has further
Figures 6 and 7 indicate the production cost of 10 generator increased and total SC has decreased. Table V shows the
system with renewable and AS management. As the production cost of 54 thermal units of different case study. UC
contributions of renewables are increasing, there is a decrease and ELD of 54 thermal units are performed using PL-GA and
in production cost. PL-PSO only. The results of table II clearly shows that
computational time of PL-GA is high as compared to PL-PSO.
The probability of program being converged is high in case of [5] S. Kiranyaz, T. Ince, A. Yildirim and M. Gabbouj,
PL-PSO. "Fractional Particle Swarm Optimization in
Multidimensional Search Space," IEEE Transactions on
Systems, Man, and Cybernetics, Part B (Cybernetics),
VIII. CONCLUSION vol. 40, no. 2, pp. 298-319, April 2010.
This paper presented unit commitment and economic load [6] A.Wood and B.Wollenberg, Power Generation Operation
dispatch solution to schedule thermal units considering and Control, New York: Wiley, 1996.
renewable sources integration by PL-PSO and results are
compared with PL-Genetic Algorithm (PL GA) and Dynamic [7] W. L. Snyder Jr., H. D. Powell Jr., and J. C. Rayburn,
Programming (DP) method. PL-PSO generates better solution "Dynamic programming approach to unit commitment,"
as compared with DP method. The solution obtained using IEEE Trans.on Power App. Syst., vol. 2, no. 2, p. 339–
PL-PSO is same as that using PL-GA but in case of PL-PSO 350, May 1987.
computational time is very less as well as probability of [8] B.D.Hari Kiran, M.S Kumari, "Optimal generation
problem being converged is more. The basic objective of UC scheduling with operating reserves including wind
is to reduce the fuel cost considering all constraints. From uncertainties," in International Conference on Smart
comparison study table the start up cost is reduced and total Electric Grid (ISEG), Vijayawada, India, 2014.
spinning reserve or average Spinning Reserve has increased by [9] K. S. Swarup and S.Yamashiro, "Unit Commitment
using PL-PSO or PL-GA, hence reliability of supply is Solution Methodology Using Genetic Algorithm," IEEE
improved. To reduce the thermal generation, integration of transaction on power systems, vol. 17, no. 1, pp. 87-91,
renewable source has been considered. To tackle with Feb 2002.
renewable uncertainties, extra power is generated from [10] I.G.Damousis, A.G. Bakirtzis and P.S.Dokopoulos, "A
thermal units so that any shortfall of power would be fulfilled. solution to the unit-commitment problem using integer-
Results clearly demonstrate that PL-PSO provides good coded genetic algorithm," IEEE Transactions on power
solution in minimum possible time with a good probability of systems, vol. 19, no. 2, pp. 1165-1172, May 2004.
convergence. Along with renewable sources, storage
technologies can also be implemented. Stored energy can be [11] Ashiwani Kumar, Devendra Kumar and Lokesh Kumar
used when peak load demand occurs. Yadav, "unit commitment of thermal power plant in
integration with wind and solar plant using genetic
IX. BIBLIOGRAPHY algorithm," International Journal of Engineering
Research & Technology (IJERT), vol. 3, no. 7, July –
2014.
[1] K.S. Swarup, S. Yamashiro, "A genetic algorithm
approach to generator unit commitment," International [12] Md. Sajid Alam, Matam Sailaja Kumari, "Unit
Journal of Electrical Power & Energy Systems, vol. 25, commitment of thermal units in integration with wind and
no. 9, pp. 679-687, November 2003. solar energy considering ancillary service management
using Priority list(IC) based Genetic algorithm," in IEEE
[2] T. O. Ting, M. V. C. Rao and C. K. Loo, "A novel Conference, New Delhi, India, Feb 2016(Presented).
approach for unit commitment problem via an effective
hybrid particle swarm optimization," IEEE Transactions [13] K. Liu, J. Yu, H. C. Shu and Y. Chen, "A New Advanced
on Power Systems, vol. 21, no. 1, pp. 411-418, Feb 2006. Genetic Algorithm for Optimal Unit Commitment of
Power System," in 2009 Asia-Pacific Power and Energy
[3] W. Hu and G. G. Yen, "Adaptive Multiobjective Particle Engineering Conference, Wuhan, China, 2009.
Swarm Optimization Based on Parallel Cell Coordinate
System," IEEE Transactions on Evolutionary [14] Geoff Walker, "Evaluating MPPT converter topologies
Computation, vol. 19, no. 1, pp. 1-18, Feb 2015. using a MATLAB PV model," Journal of Electrical &
Electronics Engineering, vol. 21, no. 1, pp. 49-56, March
[4] A. Y. Saber, T. Senjyu, N. Urasaki and T. Funabashi, 2001.
"Unit Commitment Computation - A Novel Fuzzy
Adaptive Particle Swarm Optimization Approach," in
IEEE PES Power Systems Conference and Exposition,
Atlanta, GA, 2006.

You might also like