Você está na página 1de 13

Article

Statistical Performances Evaluation of APSO and


Improved APSO for Short Term Hydrothermal
Scheduling Problem
Muhammad Salman Fakhar 1,‡ , Syed Abdul Rahman Kashif1,‡ *, Noor Ul Ain 1,‡ , Hafiz Zaheer
Hussain †,‡ Akhtar Rasool 2,‡ Intisar Ali Sajjad 3,‡
1 University of Engineering and Technology, Lahore; salmanfakhar@uet.edu.pk; abdulrahman@uet.edu.pk;
noor-ul-ain@uet.edu.pk
2 Sharif College of Engineering and Technology, Lahore; akhtar@sabanciuniv.edu
* Correspondence: abdulrahman@uet.edu.pk; Tel.: +92-345-409-2528
† Power Mechanical Division, National Engineering Services Pakistan (NESPAK),
engr.hafizzaheer@gmail.com
‡ These authors contributed equally to this work.
§ University of Engineering and Technology, Taxila, intisar.ali@uettaxila.edu.pk

Version May 15, 2019 submitted to Appl. Sci.

1 Featured Application: Short term hydrothermal scheduling is physically existing problem


2 which deals with the combined operation of hydro and thermal generators. The two types of
3 generations are so dispatched that they do not violate the constraints related to thermal power and
4 water reservoirs related to the hydro power, meanwhile serving the load demand and covering
5 the transmission power losses. It is a type of economic dispatch problem. The research is still
6 in progress to find algorithm/s that help in such a economic dispatch that the fuel cost, related to
7 the thermal generators be minimized and the water reservoirs are also not completely depleted,
8 which are primarily required for irrigation purpose. This research article is a presentation of an
9 effort to find an elegant metaheuristic optimization algorithm that robustly finds the solution
10 of two out of several types of short term hydrothermal thermal scheduling problems, while
11 considering the standard test problems as already discussed in the literature. The accelerated
12 particle swarm optimization and its improved version is thus presented in this research article
13 that is serving the defined purpose.

14 Abstract: Accelerated Particle Swarm Optimization (APSO) algorithm is an efficient and easiest to
15 implement variant of famous Particle Swarm Optimization (PSO) Algorithm. PSO and its variant
16 APSO have been implemented on the famous short-term hydrothermal scheduling (STHTS) problem
17 in recent research and it has shown promising results. APSO algorithm can further be modified
18 to enhance its optimizing capability by deploying dynamic search space squeezing. This paper
19 presents the implementation of the Improved APSO algorithm that is based on dynamic search space
20 squeezing, on short term hydrothermal scheduling problem. To give a quantitative comparison, a
21 true statistical comparison based on comparing means is also presented to draw conclusions.

22 Keywords: Improved APSO; dynamic search space squeezing; independent sample t-test

23 1. Introduction
24 PSO algorithm has gained much popularity among the meta heuristic optimization algorithms in
25 the recent past due to its ease of implementation and promise towards finding good approximates
26 to global optimum solutions of complex optimization problems [1]. APSO is a simpler yet brilliant

Submitted to Appl. Sci., pages 1 – 13 www.mdpi.com/journal/applsci


Version May 15, 2019 submitted to Appl. Sci. 2 of 13

27 variant of PSO and it has proved to find the good approximates of global optimum solutions in
28 lesser time and iterations as compared to PSO algorithms [2]. Dynamic search space squeezing has
29 been applied on PSO to make yet another variant of PSO named as Improved PSO and it gives
30 better performance to some optimization problems as compared to the PSO itself [3]. Short term
31 hydrothermal scheduling is a non-linear and multi modal optimization problem, which has many
32 forms. These can be of non-cascaded form to cascaded form. In non-cascaded form, there is only one
33 reservoir of water whereas in cascaded form, there is a series of downstream reservoirs. It can be single
34 objective or multi-objective. In single objective STHTS problem, the objective is only to minimize the
35 cost of operation of the thermal generating units, whereas, in multi-objective STHTS problem, the
36 other objectives are to reduce the emission of COx, NOx and SOx gases as well. References [3] to [27]
37 discuss all those types of STHTS problem and implementations of conventional and non-conventional
38 meta-heuristic optimization algorithms on STHTS problems and present superiority of one type of
39 algorithm (usually meta heuristic) over other types. Improved PSO, PSO, Cuckoos search optimization,
40 Quantum behaved PSO, Teaching learning based optimization, Multi-objective differential algorithm,
41 Gravitational search algorithm, artificial bee colony algorithm, genetic algorithm, civilized swarm
42 optimization algorithms, and gravitational search algorithms are better choices for those types of
43 STHTS problems as presented in references [3] – [19]. References [20] to [27] discuss specifically the
44 non-cascaded and single objective STHTS problems in which one hydel and one equivalent thermal
45 unit of a number of thermal units, are dispatched to supply the power demand. PSO and APSO have
46 shown best results for such problems. Specially, according to reference [23] and [24], APSO algorithm
47 has outperformed the other previously implemented deterministic and meta heuristic optimization
48 algorithms on non-cascaded and single objective STHTS problems.
49 References [19] to [24] discuss specifically the non-cascaded problems in which one hydel and
50 one thermal unit are dispatched to supply the power demand. PSO and APSO have shown best results
51 for such problems. Specially, according to [23] and [24], APSO algorithm has outperformed the other
52 previously implemented deterministic and meta heuristic optimization algorithms.
53 Due to the stochastic nature, it is required that there must be a true statistical comparison between
54 the implementations of various algorithms on one type of optimization problem which is not presented
55 in previous works of reference [3] – [27]. Reference [28] has also presented this thesis while applying
56 PSO algorithm using digital pheromones. In STHTS problem, it is therefore, also required to present
57 the quantitative analysis by applying statistical hypothesis tests.
58 This paper is an extended work of references [24] and [25], already published by the authors and
59 presents the implementation of Improved APSO algorithm on the non-cascaded STHTS problems
60 (two cases) and gives its statistical comparison with the previous APSO implementations on the same
61 problems, by the authors of this paper, in references [24] and [25]. Independent sample T-Tests will be
62 performed on the data set of the two implementations to check the superiority one of the two APSO
63 variants on the same STHTS problems as discussed in [24] and [25].

64 2. Accelerated Particle Swarm Optimization and its Improved Version


65 Particle swarm optimization has now been a very famous meta heuristic optimization algorithm
66 which has shown promising results when implemented on many types of complex, non-linear and
67 multi modal optimization problems. The canonical version of PSO is given in equation (1) as:


νin+1 = θνin + αe1 [ g − xin ] + βe2 [ xi∗ − xin ]
(1)
xin+1 = xin+1 + νin+1

68 In equation (1), νin+1 is the velocity update of particle i, at iteration n + 1. This velocity is added
69 to the particle position xin of particle i, at iteration n, to get the updated position xin+1 of particle i. xi∗ is
Version May 15, 2019 submitted to Appl. Sci. 3 of 13

70 the local best position of particle i, in iteration history and g∗ is the global best among all the particles,
71 at iteration n. e1 and e2 are random number ranging from 0 to 1.
72 There are almost more than two dozen variants of this canonical version of PSO algorithm as
73 discussed in [29] and [30]. Accelerated particle swarm optimization is its variant, presented in [1] and
74 [2], which is very elegant, i.e. easy to understand, easiest to program and has proven itself to give
75 good approximates to global optimum solutions. Its single step particle update equation is given as
76 equation (2);


xin+1 = (1 − β) xin + βg + α(e − 0.5) (2)

77 As can be seen, this single step update equation has ignored the use of velocity update, and not
78 used the local best position in the updating of particle i. The typical values of α and β are 0.2 and 0.5
79 respectively. e is a random variable and its value is between 0 and 1, as given in reference [1] and
80 [2]. This variant has proved to find global optimum of highly complex multi-modal functions like
81 Michaelwics 2D function as presented in references [1] and [2]. There can be many possible variants of
82 equation (2), as discussed in [1] and [2]. The variant discussed in this paper is related to the dynamic
83 search space squeezing method in which, at every iteration, the search space of the particles, in all
84 dimensions is readjusted by taking influence from the global best particle i.e. of g∗ each iteration. This
85 is how, the chance of each particle to get oscillated in the search space is decreased and there is lesser
86 chance to stick to local optima. By reducing search space, the range of particle at each dimension is
87 reduced i.e. the constraints are readjusted and new constraints are bounded by the original constraints
88 of the optimization problem, by using the dynamically search space squeezing equations (3) – (7),
89 as presented in reference [25]. Reference [25] has applied the concept of dynamic search squeezing
90 technique on canonical form of PSO i.e. on equation (1), on the velocity update equation. These
91 equations help in squeezing the search space dynamically from given constraints to newer constraints
92 from one iteration to another iteration. This paper presents this concept of dynamic search space
93 squeezing on APSO algorithm, i.e. equation (2) to further enhance its performance. Therefore, rather
94 than squeezing the search space of velocity terms, the search space of particles themselves will be
95 directly squeezed. The dynamic search space squeezing phenomenon is shown in Figure 1.

Figure 1. Dynamic search space squeezing under the influence of g∗ at iteration n + 1. The arrows
indicate the squeezing of search space from iteration n to n + 1.

gi∗n − xi,min
n
∆nlow,i = n n (3)
xi,max − xi,min
n
xi,max − gi∗n
∆nhigh,i = n n (4)
xi,max − xi,min

∆nlow,i + ∆nhigh,i = 1 (5)

n +1
xi,min n
= xi,min + ( gi∗n − xi,min
n
)(∆nlow,i ) (6)

n +1
xi,max n
= xi,max n
+ ( xi,max − gi∗n )(∆nhigh,i ) (7)
Version May 15, 2019 submitted to Appl. Sci. 4 of 13

Figure 2. Generation model of STHTS problem [24].

96 3. Short term hydrothermal scheduling problem


97 Short term hydrothermal scheduling is a type of economic dispatch problem in which a parallel
98 operation of hydel power generating units is combined with thermal power generating units on a
99 generation bus. This problem can have several forms varying from single objective to multi objective,
100 cascaded to non-cascaded and pumped storage to non-pumped storage problem. The generation
101 model of STHTS problem is presented in Figure 2, as taken from reference [24]. The complexities
102 of these problems can be further enhanced by considering valve point loading effects of thermal
103 generating units. The detailed analysis of all these problem types are presented in references [3]
104 to [25]. This paper considers two types of these problems as case study, as already considered in
105 references [19] to [24], to implement improved APSO algorithm and its comparison with the previous
106 implementations of APSO algorithm in references [23] and [24], on a formal quantitative hypothesis
107 testing. The two problem cases belong to the following two types of STHTS problem.

108 1. Non-cascaded pumped storage short term hydrothermal scheduling [23]


109 2. Non-cascaded non-pumped storage short term hydrothermal scheduling while considering
110 transmission losses [24]

111 It seems more logical to check the performance of one type of algorithm by implementing it on as
112 many types of problems as possible. However, according to the “No free lunch theorem” as given in
113 reference [29] – [32], if one algorithm performs best on type A problem, it does not necessarily mean
114 that it performs the best for another type B problem. i.e., an algorithm cannot prove to be the best
115 optimization algorithm for all types of problems. Therefore, it is required to find a good performing
116 algorithm for every type of optimization problem. The problems 1 and 2 belong to the same class of
117 single objective and non-cascaded STHTS problems and are different in terms of structure from the
118 remaining types of STHTS problems. Any algorithm, providing best optimum to these problems, will
119 perform well on all similar types of STHTS problems.
120 APSO algorithms are the simplest and most elegant metaheuristic algorithms which, without
121 having a lot of tuning parameters, and having just a single step particles update equations, are able
122 to solve non-linear and multi-modal objective functions like Michaelwicz 2D functions as given in
123 references [1] and [2]. Therefore, the elegacy and robustness of these APSO algorithms make them
124 the first choice out of many metaheuristic algorithms to implement on optimization problems. The
125 other types of STHTS problems, for example the cascaded STHTS of single objective type problem
126 requires meta-heuristic algorithms with more tuning parameters instead of APSO algorithm like
127 PSO and improved PSO. Another type is multi-objective STHTS problem, which requires variants of
128 meta-heuristic algorithms specially designed for multi-objective problems, like multi-objective PSO,
Version May 15, 2019 submitted to Appl. Sci. 5 of 13

129 ant colony and genetic algorithms, Cuckoos search, gravitational search algorithm, teaching-learning
130 based algorithms and water cycle algorithms [3]-[19].
131 The problems taken into consideration in this paper have been solved using APSO algorithm by
132 the authors of this paper, already in references [23] and [24], that outperformed the other algorithms
133 implemented on the same problems given in references [22], [26] and [27]. In this enhanced research
134 Improved APSO algorithm has been implemented on the same problems that used dynamic search
135 space squeezing, and this modified APSO has shown even better results than APSO algorithm, as
136 presented in the result section. Figure 3 presents the basic generation model of the STHTS problem
137 which is also considered in references [23] and [24].

138 4. Hypothesis testing


139 Due to their stochastic nature, meta heuristic algorithms are destined to give different results
140 on different trials on same optimization problem. It is therefore required to test the performance of
141 implementations by doing statistical analysis on a data set of results obtained. To check whether an
142 algorithm A is superior on another algorithm B in terms of performance, there are hypothesis tests
143 that quantitatively establish the superiority. References [3] – [27] have not taken into consideration
144 this point and though, the performance of algorithms presented in those references are the best, yet
145 the quantitative proof has not been established. This proof can be established by doing hypothesis
146 testing to compare means of performances. For this purpose, a null hypothesis is established. Null
147 hypothesis states that there is no difference in performance of the two algorithms. By performing
148 independent sample t-test, the null hypothesis can either be accepted or rejected [28] and [35]. If the
149 null hypothesis is rejected, it means that there is a significant difference between the performances
150 of the two algorithms. If null hypothesis is accepted, it means, that statistically, there is no difference
151 between the performances of the two algorithms [28] and [35].

152 5. Methodology
According to [19]-[27], the non-cascaded hydrothermal scheduling problem can be described by
equations (8) to (14).

N
min( f ) = ∑ n j Fj (cost f unction or objective f unction) (8)
j =1

153 subjected to:

Phyd + PT = PDemand + PLosses ( Power balance equality constraint) (9)

N
∑ n j Dis j = Distotal ( Discharge rate equality constraint) (10)
j =1

Vmin < Vj < Vmax (Water reservoir volume limits) (11)

Dismin < Dis j < Dismax ( Discharge rate limits) (12)

PT,min < PT,j < PT,max ( Thermal power limits) (13)

Phyd,min < Phyd,j < Phyd,max ( Hydro power limits) (14)


Version May 15, 2019 submitted to Appl. Sci. 6 of 13

The reservoir’s volume and the discharges are balanced by the continuity equation (15) given as;

Vj = Vj−1 + n j ( R j − Dis j − S j ) (15)

154

155 To implement APSO or Improved APSO algorithms on non-cascaded STHTS problem, following steps
156 can be implemented;

157 1. Randomly initialize the volume vectors (particles) within the given volume constraints, randomly.
158 In this paper, uniform random number generators have been used.
159 2. Calculate the water discharge rate, hydro power and thermal power using the values in step 1.
160 3. Check the limits in constraints. If the limits are not met, restart from step 1. If the limits are met,
161 proceed to step 4.
162 4. Find the total cost using thermal power values found in step 2, against each particle.
163 5. Take the minimum cost value and its corresponding volume vector. That volume vector will be
164 the global best particle.
165 6. Update all the particles using APSO/Improved APSO updating “(10)”.
166 7. Dynamically squeeze the volume constraints by applying search space squeezing equations for
167 improved APSO algorithm
168 8. Iterate from steps 2 to 6 till the stopping criterion is reached.
169 9. Get the results

170 6. Results and Discussions


171 This paper presents the essence and importance of true statistical quantitative tests to check the
172 superiority of one type of algorithm to the other type of algorithm. This is because, the meta heuristic
173 optimization algorithms are stochastic as well as deterministic in nature. Due to the stochastic nature,
174 there is an in-built randomness and therefore, no meta heuristic optimization algorithm gives same
175 redundant results when implemented on one type of problem. Therefore, to check if one algorithm is
176 performing on same problem every time, the performance is judged by standard deviations among the
177 trials. To check the performance of two different types of algorithms, on same problem, the data taken
178 from equal number of trials are hypothetically tested for significant difference in the means of all the
179 trials. This type of testing is known as independent sample t tests. This t test is performed on software
180 SPSS i.e. Statistical Package for Social Sciences. To test the performance of APSO and Improved APSO,
181 two test cases have been taken as already discussed in reference [23] – [24].

182 6.1. Case 1: Non-cascaded pumped storage hydrothermal scheduling [23], [27]
183 It’s a two-unit hydrothermal scheduling problem as given in reference [23]. The total number of
184 scheduling periods are six out of which three periods are pumping i.e. discharged water is pumped
185 back to the reservoir, while three periods are generating. Thermal unit characteristics are given in
186 equation (16) as;

FT = 3877.5 + 3.9795PT + 0.00204PT2 $/h (16)

200MW ≤ PT ≤ 2500MW (17)

The generating and pumping models are given by equations (18) and (19) respectively.

Dis( Phyd ) = 2Phyd ( acre. f t/h) f or 0 ≤ Phyd ≤ 300 MW ( Generatingcharacteristics) (18)

Dis( PH,pump ) = −600( acre. f t/h) with PH,pump = −300MW ( Dischargingcharacteristics) (19)
Version May 15, 2019 submitted to Appl. Sci. 7 of 13

187 The starting and ending volume of reservoir = 8000 (acre-ft / h). The load demand at every
188 interval is given in Table 1.

Table 1. Load demand for each scheduling interval of four hours each

Interval 1 2 3 4 5 6
Demand MW 1600 1800 1600 500 500 500

Table 2. Comparison of some performance parameters between APSO and Improved APSO on
pumped storage STHTS problem

Performance parameter or algorithm APSO Improved APSO


Minimum Cost 269642.4001 269642.4000
Average Cost 269642.5318 269642.4752
Maximum Cost 269643.2336 269642.402
No of acceptable convergences 04 out of 100 trials 97 out of 100 trials
Standard deviation 6.2557539 0.0088976
Average No of iterations 7 8
Average computation time ∗ 0.462 sec 0.506 sec
∗ Average computation time is according to MATLAB 2015 at Core i5 second generation processor

Table 3. SPSS Results of independent sample t-test showing comparison between simple APSO and
Improved APSO on pumped storage non-cascaded STHTS problem

Independent Samples T-Test


Levene’s Test for equality
T-Test for the equality of means
of variances
95% Confidence Interval
Comparison Sig. Mean Std. Error
F Sig. t df of the difference
_APSO (2-tailed) Difference Difference
Lower Upper
Equal
variances 62.424 0.000 5.333 198 0.000 3.3361210 0.6255760 2.1024741 4.5697679
assumed
Equal
variances 5.333 99.00 0.000 3.3361210 0.6255760 2.0948425 4.5773995
not assumed

189 The problem is implemented using both APSO algorithm and Improved APSO algorithm.
190 Reference [23] has already discussed the implementation of APSO on this problem, however, this
191 paper presents its comparison with improved APSO algorithm on hypothetical testing. Improved
192 APSO algorithm has shown even more promising results as compared to APSO algorithm. Table 2
193 presents the comparison between the two implementations.
194 Both the algorithms have been tested on the case 1 problem for a hundred trials each to have a
195 normally distributed data set. Their means are compared using hypothesis testing independent sample
196 t-test. The results obtained from SPSS software run, are given in Table 3.
197 The null hypothesis was that there is no difference in performance of the two algorithms on the
198 pumped storage STHTS problem. This hypothesis was tested on confidence interval value of 95%.,
199 gives significance value (2-tailed) equal to 0.000 i.e. lesser than 0.05 or 5% [35]. This value suggests
200 that the null hypothesis is rejected, and the statistical test quantitatively suggests that improved APSO
201 algorithm has performed significantly differently from simple APSO algorithm. Also, the Levine’s
202 test of equality of variances suggest that for significance value equal to 0.000 which is less than 0.05,
203 that variances are also statistically different, significantly Table 2 further elaborates the difference in
204 performance by establishing the superiority of improved APSO over simple APSO algorithm. Table
Version May 15, 2019 submitted to Appl. Sci. 8 of 13

Table 4. Power flow and cost optimization with Improved APSO algorithm implementation on
non-cascaded pumped-storage STHTS problem

Intervals Demand (MW) Ps (MW) Ph (MW) Q (ac-ft/hr) V (ac-ft) Cost ($)


1 1600 1449.99 150 500 5999.95
2 1800 1500 300 800 2799.819
3 1600 1450 159.97 500 800
269642.4
4 500 800 -300 -600 3200
5 500 800 -300 -600 5600
6 500 800 -300 -600 8000

Figure 3. Convergence characteristics of simple APSO algorithm on non-cascaded pumped STHTS


problem

205 2 also presents that though the minimum values achieved by the two algorithms are though same,
206 but improved APSO algorithm has converged to the acceptable result i.e. this paper has considered
207 acceptable cost value less than 269642.4001, 97% times as compared to simple APSO algorithm that
208 converged to the acceptable results for 4% of times. The number of iterations and computation time are
209 very low for both these algorithms, however, due to the dynamic search space squeezing, Improved
210 APSO takes a slightly higher average time to converge to the solution as compared to simple APSO
211 algorithm. However, the difference in computation time is not much significant. Although APSO
212 algorithm has given promising results, but improved APSO algorithm has proved itself better for
213 pumped storage non-cascaded STHTS problem, statistically. Table 4 presents the best results of the
214 improved APSO implementation on non-cascaded pumped-storage STHTS problem.
215 The convergence characteristics of simple APSO and improved APSO are given in Figures 3 and
216 4 respectively. These figures are taken for the successful trials for both the algorithms. Number of
217 particles are taken to be equal to 200 for both implementations. The convergence characteristics of both
218 the algorithms are quite similar and are showing the general nature of PSO variants. The algorithms,
219 as their names suggest are accelerated in nature, i.e. they are fast in finding good approximates of
220 global optimum.

221 6.2. Case 2: Non-cascaded non-pumped STHTS considering transmission losses [24]
222 It is a two unit and non-pumped hydrothermal scheduling problem as given in references [24].
223 Total number of scheduling periods are six and the discharging characteristics of the thermal unit is
224 given in equation (20) and (21); where fuel cost is 1.15 $/MBTU.

FT = 500 + 8PT + 0.0016PT2 ( MBTU/h) (20)

150MW ≤ PT ≤ 1500MW (21)


Version May 15, 2019 submitted to Appl. Sci. 9 of 13

Figure 4. Convergence characteristics of Improved APSO algorithm on non-cascaded pumped STHTS


problem

Table 5. Load demand for each scheduling interval (each interval is of 12 hours making a total
scheduling period of 3 days)

Interval 1 2 3 4 5 6
Demand MW 1200 1500 1100 1800 950 1300

225 The water discharging constraints are given in equation (22) to (25);

Dis ( Phyd ) = 330 + 4.97Phyd ( acre − f t/h) (22)

for
0MW ≤ Phyd ≤ 1000MW (23)

Dis ( Phyd ) = 5300 + 12( Phyd − 1000) + 0.05( Phyd − 1000)2 ( acre − f t/h) (24)

226 for

1000 MW ≤ Phyd ≤ 1100 MW (25)

227 The transmission losses are given by equation (26)

2
( Ploss ) = 0.0008Phyd ( MW ) (26)

228 Discharge rate characteristics are given in equation (27)

5300 ( acre − f t/hr ) ≤ Dis ≤ 7000 ( acre − f t/hr ) (27)

229 The load demand in every interval is given in Table 5.


230 The reservoir’s water’s volume’s flow constraints as given in references [25] and [27] are:

231 1. 100,000 acre-feet is the volume of water in the reservoir prior to the first scheduling period.
232 2. 60,000 acre-feet is the volume of water in the reservoir in the last period.
233 3. Volume limit constraints in the first iteration are in acre-ft 60,000 (acre ft) ≤ V ≤ 120,000 (acre-ft)
234 4. Continuous Incoming flow into the reservoir is 2000 acre-feet/hour throughout the scheduling
235 period.
236 5. The continuity equation as given in equation (15) is to be met.

237 The problem is implemented using both APSO algorithm and Improved APSO algorithm.
238 Reference [24] has already discussed the implementation of APSO on this problem, however, this
Version May 15, 2019 submitted to Appl. Sci. 10 of 13

Table 6. Comparison of some performance parameters between APSO and Improved APSO on
non-cascaded non-pumped STHTS problem

Performance parameter or algorithm APSO Improved APSO


Minimum Cost 727870.00 727855.8
Average Cost 728422.7 728539.4
Maximum Cost 730103.3 732386.8
No of acceptable convergences 15 out of 100 trials 15 out of 100 trials
Standard deviation 413.35 539.34
Average No of iterations 6 7
Average computation time ∗ 1.035 sec 1.227 sec
∗ Average computation time is according to MATLAB 2015 at Core i5 second generation processor

Table 7. SPSS Results of independent sample t-test showing comparison between simple APSO and
Improved APSO on non-cascaded non-pumped STHTS problem

Independent Samples T-Test


Levene’s Test for equality
T-Test for the equality of means
of variances
95% Confidence Interval
Comparison Sig. Mean Std. Error
F Sig. t df of the difference
_APSO (2-tailed) Difference Difference
Lower Upper
Equal
variances 4.674 0.032 -1.133 198 0.259 -76.9991850 67.9525842 -211.0028687 57.0044987
assumed
Equal
variances -1.133 185.468 0.259 -76.9991850 67.9525842 -211.0585712 57.0602012
not assumed

239 paper presents its comparison with improved APSO algorithm on hypothetical testing. Improved
240 APSO algorithm has shown even more promising results as compared to APSO algorithm. Table 6
241 presents the comparison between the two implementations.
242 Both the algorithms have been tested on the case 2 problem for a hundred trials each to have a
243 normally distributed data set. Their means are compared using hypothesis testing independent sample
244 t-test. The results are shown with image files of tables obtained from SPSS software run, in Table 7.
245 For this case 2 problem, the value of significance of 2 tailed t-test shows that both the algorithms
246 are not significantly different in terms of performance because its value is greater than 5%. It means
247 that for more than 25% of times, the results are somewhat repeating. Therefore, at 95% confidence
248 level, the null hypothesis cannot be rejected, and the means are not statistically different. However, the
249 improved APSO has been able to find a better approximation to the global best as given in table 6. The
250 number of iterations and computation time are very low for both these algorithms, however, due to
251 the dynamic search space squeezing, Improved APSO takes a slightly higher average time to converge
252 to the solution as compared to simple APSO algorithm. However, the difference in computation
253 time is not much significant. Therefore, if some more tuning parameters be properly adjusted, both
254 the algorithms have capability to find robust approximations to global optimum solutions with low
255 standard deviations, for non-cascaded non-pumped STHTS problem while considering transmission
256 line losses, improved APSO having more ability to find better approximation to global minimum.
257 Table 8 gives the best cost result of improved APSO performance.
258 The convergence characteristics of simple APSO and improved APSO are given in Figures 5 and 6
259 respectively, on non-cascaded non-pumped STHTS problem considering transmission losses. These
260 figures are taken for the successful trials for both the algorithms. Number of particles are taken to
261 be equal to 200 for both implementations. The convergence characteristics of both the algorithms are
262 quite similar and are showing the general nature of PSO variants. The algorithms, as their names
263 suggest are accelerated in nature, i.e. they are fast in finding good approximates of global optimum.
Version May 15, 2019 submitted to Appl. Sci. 11 of 13

Table 8. Power flow and cost optimization with Improved APSO algorithm implementation

Interval PT (MW) Phyd (MW) Ploss (MW) Dis (acre-ft/h) V(acre-ft) Total cost ($)
1 833.91 377.4 11.399 2206 97527.06
2 950.66 575.86 26.52 3192 83222.43
3 813.76 293.10 6.87 1786.7 85781.40
727855.8
4 1091.94 753.47 45.41 4074.7 60884.22
5 733.34 220.54 3.89 1426.1 67770.92
6 851.08 466.313 17.395 2647.57 60000

Figure 5. Convergence characteristics of simple APSO algorithm on non-cascaded non-pumped STHTS


problem taken from reference [24]

Figure 6. Convergence characteristics of Improved APSO algorithm on non-cascaded non-pumped


STHTS problem

264 7. Conclusions
265 To establish the superiority of one type of algorithm over another type of algorithm, while
266 implementing on one type of optimization problem requires a proper statistical hypothesis testing like
267 t-test. This paper has presented another modification in particle swarm optimization as Improved
268 Accelerated particle swarm optimization algorithm, that dynamically squeezes the search space for the
269 particles in every iteration. Two out of many types of STHTS problems were tested using simple APSO
270 and improved APSO. Improved APSO algorithm has shown much promising results as compared to
271 simple APSO algorithm for the non-cascaded pumped storage STHTS case whereas both the algorithms
272 have shown equivalent results in statistical terms for the non-cascaded non-pumped STHTS problem.
273 However, Improved APSO has successfully found a better approximate to global minimum for this
274 problem. Further research can be made on the tuning of these variants of PSO to find even more
275 promising results for these problems and for other types of STHTS problems.
Version May 15, 2019 submitted to Appl. Sci. 12 of 13

276 Abbreviations
277 The following symbols and abbreviations are used in this manuscript:
278
APSO Accelerated Particle Swarm Optimization
xin+1 Particale position at iteration n + 1
g∗ Global best particle
∆nlow,i Lower increment to be added in ith particle’s lower limit for search space squeezing at iteration n
∆nlow,j Higher increment to be added in ith particle’s upper limit for search space squeezing at iteration n
n +1
xi,min Lower limit of ith particle at iteration n + 1
n +1
xi,max upper limit of ith particle at iteration n + 1
279 f Objective function/cost
Fj Cost of jth generating unit
Vj Volume of reservoir at jth interval
Dis j Discharge rate of reservoir at jth interval
PT,j Thermal power at jth interval
Phyd,j Hydro power at jth interval
Rj Constant water inflow into the reservoir at jth interval
Sj Spillage of water from reservoir at jth interval

280 References
281 1. Yang, X.S., Engineering optimization: an introduction with metaheuristic applications; John Wiley Sons;
282 2010, doi: https://doi.org/10.1002/9780470640425
283 2. Yang, X. S., Deb, S., and Fong, S., Accelerated particle swarm optimization and support vector machine for
284 business optimization and applications, Communications in Computer and Information Science, 2011, 136, 53-66,
285 doi:org/10.1007/978-3-642-22185-9-6.
286 3. Hota, P., Barisal, A. and Chakrabarti, R., An improved PSO technique for short-term optimal hydrothermal
287 scheduling. Electrical Power Systems Research, 2009, 79(7), 1047-1053, doi: 10.1016/j.epsr.2009.01.001
288 4. Mandal, K. K., Basu, M., and Chakraborty, N., Particle swarm optimization technique based short-term
289 hydrothermal scheduling, Applied Soft Computing, 2008, 8(4), 1392-1399, doi: 10.1016/j.asoc.2007.10.006.
290 5. Yu, B. Yuan, X. and Wang, J., Short-term hydro-thermal scheduling using particle swarm optimization
291 method, Energy Convertion and Management, 2007, 48(7), 1902-1908, doi: 10.1016/j.enconman.2007.01.034.
292 6. Thang T. N., Dieu N. V., and Anh V. T., Cuckoo search algorithm for short term hydrothermal scheduling,
293 em Applied Energy, 2014, 132(1), 276-287 doi: 10.1007/978-81-322-2119-7-57.
294 7. Ahmadi, A. Masouleh, M. S. Janghorbani, M. Manjili, N. Y. G. Sharaf, A. M. and Nezhad, A. E., Short
295 term multi-objective hydrothermal scheduling, Electrical Power Systems Research, 2015, 121(1), 357-367, doi:
296 10.1016/j.epsr.2014.11.015.
297 8. Chen, J. J. and Zheng, J. H., Discussion on short-term environmental/economic hydrothermal scheduling,
298 Electrical Power Systems Research, 2015, 127(1), 348-350, doi: 10.1016/j.epsr.2015.04.021.
299 9. Sun, C. and Lu, S., Short-term combined economic emission hydrothermal scheduling using improved
300 quantum-behaved particle swarm optimization, Expert Systems Applications, 2010, 37(6), 4232-4241, doi:
301 10.1016/j.eswa.2009.11.079.
302 10. Roy, P. K., Teaching learning based optimization for short-term hydrothermal scheduling problem
303 considering valve point effect and prohibited discharge constraint, International Journal of Electrical Power
304 Energy Systems, 2013, 53(1), 10-19, doi: 10.1016/j.ijepes.2013.03.024.
305 11. Zhang, H. Zhou, J. Zhang, Y. Fang, N. and Zhang, R., Short term hydrothermal scheduling using
306 multi-objective differential evolution with three chaotic sequences, International Journal of Electrical Power
307 Energy Systems, 2013, 47(1), 85-99, doi: 10.1016/j.ijepes.2012.10.014.
308 12. Farhat, I. A. and El-Hawary, M. E., Optimization methods applied for solving the short-term
309 hydrothermal coordination problem, Electrical Power Systems Research, 2009, 79(9) , 1308-1320, doi:
310 10.1016/j.epsr.2009.04.001.
Version May 15, 2019 submitted to Appl. Sci. 13 of 13

311 13. Tian, H., Yuan, X., Ji, B. and Chen, Z., Multi-objective optimization of short-term hydrothermal scheduling
312 using non-dominated sorting gravitational search algorithm with chaotic mutation, Energy Conversions and
313 Management, 2014, 81(1), 504-519, doi: 10.1016/j.enconman.2014.02.053.
314 14. Zhou, J., Liao, X., Ouyang, S., Zhang, R., and Zhang, Y., Multi-objective artificial bee colony algorithm for
315 short-term scheduling of hydrothermal system, International Journal of Electrical Power Energy Systems, 2014,
316 55(1), 542-553, doi: 10.1016/j.ijepes.2013.10.013.
317 15. Mandal, K. K., and Chakraborty, N. Differential evolution technique-based short-term economic generation
318 scheduling of hydrothermal systems, Electrical Power Systems Research, 2008, 78(1), 1972-1979, doi:
319 10.1016/j.epsr.2008.04.006.
320 16. Mandal, K. K., and Chakraborty, N., Daily combined economic emission scheduling of hydrothermal systems
321 with cascaded reservoirs using self organizing hierarchical particle swarm optimization technique, Expert
322 Systems Applications, 2012, 39(3), 3438-3445, doi: 10.1016/j.eswa.2011.09.032.
323 17. Immanuel S. A., Civilized swarm optimization for multi-objective short-term hydrothermal scheduling,
324 International Journal of Electrical Power Energy Systems, 2013, 51(1), 178-189, doi: 10.1016/j.ijepes.2013.03.002.
325 18. Narang, N., Short-term hydrothermal generation scheduling using improved predator influenced civilized
326 swarm optimization technique, Applied Soft Computing, 2017, 58(1), 207-224, doi: 10.1016/j.asoc.2017.04.065.
327 19. Hota, S. K., Barisal, A. K., and Chakrabarti, R., An improved PSO technique for short-term
328 optimal hydrothermal scheduling, Electric Power Systems Research, 2009, 79(7), 1047-1053, doi:
329 https://doi.org/10.1016/j.epsr.2009.01.001.
330 20. Padmini, S., and Rajan, C. C. A., Improved PSO for short term hydrothermal scheduling, SEISCON, 2011,
331 332-334, doi: 10.1049/cp.2011.0384.
332 21. Samudi, C., Das, G. P., Ojha, P. C., Sreeni, T. S., and Cherian, S., Hydro thermal scheduling using particle
333 swarm optimization, Proceedings of Conference on Transmission and Distribution Exposition, PIMS, 2008, doi:
334 10.1109/TDC.2008.4517221
335 22. Fakhar, M. S., Kashif, S. A. R., Saqib, M. A., and Hassan, T. U., Non cascaded short-term hydro-thermal
336 scheduling using fully-informed particle swarm optimization, International Journal of Electrical Power Energy
337 Systems, 2015, 73 983-990, doi: 10.1016/j.ijepes.2015.06.030.
338 23. Khandualo, S. K., Barisal, A. K., and Hota, P. K., Scheduling of pumped storage hydro-thermal system with
339 evolutionary programming, Journal of Clean Energy Technologies, 1(4), 2013 doi: 10.7763/JOCET.2013.V1.70.
340 24. Fakhar, M. S., Kashif, S. A. R., Saqib, M. A., Mehmood, F., and Hussain, H. Z., Non-cascaded short-term
341 pumped-storage hydro-thermal scheduling using accelerated particle swarm optimization, Proceedings of
342 International Conference on Electrical Engineering, ICEE 2018, (Lahore) Pakistan.
343 25. Hussain, H. Z., Haider, A., Fakhar, M. S., Ahmad J., Butt, M. A., and Khokhar, K. S., Short-term scheduling of
344 non-cascaded hydro-thermal system with transmission losses using accelerated particle swarm optimization
345 algorithm, Pakistan Journal of Engineering and Applied Sciences, 2018, 22, 20-29.
346 26. Wood, A. J., Power Generation Operation and Control, 1996.
347 27. Wood, A. J., and Wollenberg, B., Power generation operation and control, 2nd edition, 2002.
348 28. Kalivarapu, V., and Winer, E., A statistical analysis of particle swarm optimization with and without digital
349 pheromones, Proceedings of Conference on Structures, Structural Dynamics, and Materials, 2017, (Honolulu)
350 Hawaii, 23-26 April 2017.
351 29. Chopard, B., and Tomassini, M., Particle swarm optimization, Springer in Natural Computing Series, 2018.
352 30. Kennedy, J., Particle swarm optimization: Tutorial, Encyclopedia of Machine Learning, 2010.
353 31. Wolpert, D. H., and Macready, W.G., No free lunch theorems for optimization, : IEEE Transactions on
354 Evolutionary Computation, 1997, 1(1), 67-82, doi:10.1109/4235.585893.
355 32. Ho, Y. C., and Pepyne, D. L., Simple explanation of the no-free-lunch theorem and its implications, Journal of
356 Optimization Theory and Applications,2002, 115(3), 549-570, doi: 10.1023/A:1021251113462.
357 33. Yu-Chi, H., and Pepyne, D. L., Simple explanation of the no free lunch theorem of optimization, 2003.
358 34. Kifer, D., and Machanavajjhala, A., No free lunch in data privacy, 2011.
359 35. Cairney, P., Doing Statistics using SPSS, Third Edition, 2009.

360
c 2019 by the authors. Submitted to Appl. Sci. for possible open access publication under the terms and conditions
361 of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

Você também pode gostar