Accepted Manuscript

Title: Swarm Robotics Search and Rescue; a Novel Artificial

Intelligence-Inspired Optimization Approach

Authors: M. Bakhshipour, M. Jabbari Ghadi, F. Namdari

PII: S1568-4946(17)30107-2
DOI: http://dx.doi.org/doi:10.1016/j.asoc.2017.02.028
Reference: ASOC 4078

To appear in: Applied Soft Computing

Received date: 1-4-2016

Revised date: 24-2-2017
Accepted date: 26-2-2017

Please cite this article as: M.Bakhshipour, M.Jabbari Ghadi, F.Namdari,

Swarm Robotics Search and Rescue; a Novel Artificial Intelligence-
Inspired Optimization Approach, Applied Soft Computing Journal
http://dx.doi.org/10.1016/j.asoc.2017.02.028

This is a PDF file of an unedited manuscript that has been accepted for publication.
As a service to our customers we are providing this early version of the manuscript.
The manuscript will undergo copyediting, typesetting, and review of the resulting proof
before it is published in its final form. Please note that during the production process
errors may be discovered which could affect the content, and all legal disclaimers that
apply to the journal pertain.
Swarm Robotics Search & Rescue; a Novel Artificial

Intelligence-Inspired Optimization Approach

M. Bakhshipour *, M. Jabbari Ghadi †, F. Namdari ‡
*‡
Dept. of Electrical Engineering, Faculty of Engineering, University of Lorestan, Khorramabad, Iran
†
Dept. of Electrical Engineering, Faculty of Engineering, University of Guilan, Rasht, Iran
*
Email: [email protected]
†
Email: [email protected]
‡
Email: [email protected]

†
Corresponding Author: M. Jabbari Ghadi

Postal address: Dep. of Electrical Engineering, Faculty of Engineering, University of Guilan, P. O. Box 3756,

Rasht, Iran, Tel:+98-131-6690274, Fax:+98-131-6690271, E-mail: [email protected].

Graphical abstract

Flow chart of the proposed SRSR optimization algorithm

1
Abstract

In this paper, a novel heuristic algorithm is proposed to solve continuous non-linear optimization problems. The

presented algorithm is a collective global search inspired by the swarm artificial intelligent of coordinated robots.

Cooperative recognition and sensing by a swarm of mobile robots have been fundamental inspirations for

development of Swarm Robotics Search & Rescue (SRSR). Swarm robotics is an approach with the aim of

coordinating multi-robot systems which consist of numbers of mostly uniform simple physical robots. The ultimate

aim is to emerge an eligible cooperative behavior either from interactions of autonomous robots with the

environment or their mutual interactions between each other. In this algorithm, robots which represent initial

solutions in SRSR terminology have a sense of environment to detect victim in a search & rescue mission at a

disaster site. In fact, victim location refers global best solution in SRSR algorithm. The individual with the highest

rank in the swarm is called master and remaining robots will play role of slaves. However, this leadership and

master position can be transitioned from one robot to another one during mission. Having the supervision of master

robot accompanied with abilities of slave robots for sensing the environment, this collaborative search assists the

swarm to rapidly find the location of victim and subsequently a successful mission. In order to validate effectiveness

and optimality of proposed algorithm, it has been applied on several standard benchmark functions and a practical

electric power system problem in several real size cases. Finally, simulation results have been compared with those

of some well-known algorithms. Comparison of results demonstrates superiority of presented algorithm in terms

of quality solutions and convergence speed.

Index Terms

Swarm Robotics, Evolutionary algorithms, Nonlinear optimization

1. INTRODUCTION

With the advent of technological advances in field of computer science during recent years, researches in case

of computational intelligence have attained significant attentions of scholars. In this regard, a number of

optimization algorithms have been introduced based on swarm intelligence (SI). In fact, nature has been the ultimate

source of inspiration for engineers in this field of science. Due to incapability of analytical-based methods such as

linear programming, dynamic programming, or Lagrangian approaches for finding exact solution of large-scale or

2
non-differentiable engineering and science problems, these heuristic based approaches which are puissant to search

solution space for an approximate or a near-optimum solution in a reasonable time frame have attracted

considerable attentions. They mostly have inspirations from evolution or swarm intelligence of creatures on planet

earth [1].

Recently, numerous researches have been conducted in case of nature-inspired meta-heuristic optimization

algorithms with the aim of reaching a methodology that has superiority over previous algorithms in terms of

optimality and convergence speed by proposing a novel algorithm or by metamorphosing/ hybridization of existing

optimization algorithms. The common principal of all heuristic optimization algorithms is that they begin with a

number of initial solutions, iteratively produce new solutions by some generation rules and then to evaluate these

new solutions, and eventually report the best solution found during the search process. Some of them are

summarized in Table 1

.In this paper, a new meta-heuristic optimization algorithm called Swarm Robotics Search & Rescue (SRSR)

inspired by artificial intelligence of robots in the swarm robotics is proposed. In addition to several standard test

functions, proposed algorithm is applied on unit commitment problem (UC) in electric power systems. UC is an

non-convex optimization problem to determine the operation schedule of the electrical generating units at every

hour interval with varying loads under different operational constraints.

The main contributions of this paper compared to the previous studies can be detailed as follows:

 Proposing an artificial intelligence inspired optimization algorithm: As opposed to previous heuristic

algorithms that were inspired by evolutionary or natural phenomena, presented algorithm is based on

an artificial intelligence concept. Considering the fact that personal movements of robots and their

internal interactions are programed by human brain, a wide range of movements and characteristics can

be obtained which are mostly rare or impossible to observe simultaneously in a specific creature or

phenomenon in the nature. Therefore, high flexibility of implementation and programing is achievable.

 Proposing new operators: Three new operators have been presented in proposed optimization algorithm,

which are different than those of presented by previous algorithms. a) Accumulation: the first operator

moves solutions to their new positions by utilizing normal distribution function. To this end, a new

method for calculating mean and variance factors for each of solutions are presented. b) Exploration:

the second operator may displace solutions toward/backward the best solution by a twofold master-

slave concept. None of previous algorithms can lead some of solutions on counter direction of best

3
 solution during a controlled procedure to explore the entire of search space better. Besides, as opposed

to previous algorithms like PSO, GA and etc. in which positions of low quality solutions were basis of

their new respective positions, some positions around these solutions are basis of movement for low

quality solutions in proposed operator. c) Local search: well-known operators of exponent and round

effecting on float part of variables are employed for the first time in this paper as the third operator.

However, local search policy effects on limited number of solutions during each of iterations, these

operators considerably accelerate convergence of algorithm when solutions are gathered near the most

optimal position.

 Enjoying high robustness: Contrary to previous algorithms that their overall convergence qualities were

highly dependent to several controlling parameters, SRSR has only one controlling parameter limited

within a short interval that subsequently minimizes familiar challenges related to setting of controlling

parameters.

 During section “progress assessment” of proposed method, position of solutions are compared with that

of elite solutions using a modified approach which helps to maintain diversity of solutions; While, some

optimization algorithms omitted this methodology, or this the problem about other remaining

algorithms like PSO and GA results in a pre-mature convergence of algorithm and plighting in local

optima.

 Enjoying high convergence in term of solution quality: In case of number of cost function evaluation in

each of iterations of algorithm, this value is about 2 for proposed algorithm. However, this value for

some algorithms like PSO, GA and etc. are less than SRSR, as results will confirm, quality of proposed

operators suitably compensate this deficiency.

Structure of the paper is as follows: After presenting a brief review about the emergence of robotics in section

2, Section 3 provides some explanations for swarm robotics and its practical applications in real world. Details of

the proposed optimization algorithm inspired by swarm robotics based algorithm of search & rescue are described

in section 4. The performance of the proposed algorithm is illustrated in Section 5 using case studies. Concluding

remarks are presented in Section 6.

2. EMERGENCE OF ROBOTICS

The development and progress of robotics in human life can be categorized into two main phases. During the

first phase, electric machines that could be programmed to carry out some specific tasks have been emerged [30].
4
They didn't really have any obvious interactions with the real world, such as the highlighted reciprocal performance

can be seen in automotive manufacturing in high-tech companies over the last century. Japanese companies were

pioneer in case of employing industrial robots to different markets including pharmaceutical packaging, auto

manufacturing, foundries, distribution centers and many others. As opposed to the first stage of robotics industry,

novel robots are programmed not only to basically perform repetitive tasks, but also to recognize objects based on

information absorbed and correspondingly respond to objects in their environment utilizing gathered information

with higher accuracy. In recent years, a large number of different models have been proposed to constitute multi-

robots system or swarm robotics form of collections of robots which are principle inspiration for proposed

algorithm.

3. SWARM ROBOTICS, DEFINITION AND APPLICATION

3.1. Definition of swarm robotics

Swarm robotics is a newfangled algorithm to coordinate a large number of robots in form of multi-robot or

swarm systems [31]. It is supposed that interactions of robots either with the environment or between each other

result in a desired collective behavior [32]. Swarm robotics was an anonymous field of research until the model

was proposed by Reynolds in 1986 [30]; A flock of birds in flight has been simulated based on local interactions

and only few simple rules[30]. Independence of swarm robotics to a central wireless system and their abilities to

utilize own wireless communications assist a swarm of robots to be efficient in industrial environments where the

quality of receiving wireless signals is highly variable and communication failures are to be expected. Some of the

highlighted characteristics of members in these swarms can be summarized as follows [33]:

i. They should be completely autonomous mobile robots with ability to analyze data perceived by sensation

of the real environment.

ii. Based on complexities of manufacturing and control systems, number of robots in a swarm can vary from

several to thousands.

iii. Although, homogeneity of robots is another key feature of individuals existing in a swarm, it is seen there

are a limited number of robots with specific superiorities in order to accomplish particular obligations in a

mission.

5
 iv. Inability of sole individuals of swarm to reach ultimate goal desired for the entire system.

v. Due to local and limited sensing and communication capabilities of robots in a swarm, coordination of

individuals will be distributed and therefore, scalability becomes a property of the system.

As opposed to robots can be witnessed during routine days which mostly enjoy significant computer processing

power with complex components, robots utilized in a swarm are often relatively inexpensive and simple [33].

Cooperative operation of these simple machines eliminates the necessity to more powerful and expensive robot. In

order to satisfy such the necessity, oblivion should be an inseparable feature of robots in such the style which means

they cannot remember observations or computations performed. In the best state of construction, they can

remember only a limited number of previous steps. Based on the environment in which the swarm will be utilized,

different communication systems like radio frequency, infrared or any other types of wireless transmission systems

are employed. With respect to the fact, the sensing range of robots may vary from zero to visibility domain. Unlike

the distributed computing in which many computers work on a single large task, each individual in the swarm-style

robotics deals with unique stimuli. For instance, each robot might be in a different area at a given time.

3.2. Search & Rescue application of swarm robotics

Research for applications of swarm robotics in the field of target search has grown significantly during recent

years [32]. Problems related to applications of swarm robotics can be categorized into two classes. First class

includes pattern based problems (e.g. self-organizing grids, migration, cartography, aggregation and area coverage)

while destination in the second class is object existence (e.g. searching for ore veins in wild region, locating targets

in an unknown environment, foraging, detecting the odor sources and search & rescue (S&R) mission

accomplishment for a victim in a post-disaster area) [34].

In this regard, the focus of majority of contributions has been on groups of machines which have relatively small

number of members. However, a swarm having 1024 single robots was demonstrated by Harvard in 2014 and that

has been the largest so far [35]. Swarm robotics system is mostly common for the tasks requiring a large area of

space (e.g. the tasks that cover large areas). A simple example is searching and collecting multiple targets in an

open area [42]. The area might be very large and the swarm can take advantage of the parallel searching with

several small groups within the sensing ranges of the robots.

Besides, research for applications of swarm robotics in the field of target search has grown significantly during

recent years for situations in which search mission is associated with high level of danger or in inaccessible working

places. To this end, applications of swarm robotics in forms of autonomous armies in military and S&R task
6
accomplishment in endangered semi-ruined environments are two most highlighted. In fact, one of the most

promising applications of multi-robots and swarm robotics concerns about S&R process; in which an unsafe semi-

destroyed structure after a disaster should be explored to rescue a victim [34]-[35]. This procedure was the

inspiration of proposed optimization algorithm. In case of places that are not reachable for rescue workers to safely

detect the presence of life using passive infrared sensor (PIR sensor), a swarm of robots with pre-specified sizes

can be highly effective.

It is notable that there are mainly two types for interactions among individuals in a swarm of robots. In first and

most common type, no individual has superiority over others; while in second type which is known as master-slave

swarm robotics [36]-[37], however all individuals have same characteristics, each member of swarm has ability to

play role of leader (master robot) in some periods of mission accomplishment. This type of swarm robotics as basis

of proposed algorithm has been detailed in section 3.3.

3.3. Master-slave swarm robotics

The communication system among robots is one of primary issues in design of a swarm. However, several

communication systems like Bluetooth, IR and ZigBee have been proposed and currently employed for

communication among individuals of swarms, no strong evidence of superiority can be found in case of one of

these systems. In this research, master/slave technique as the communication system is used [36]-[37]. In this type

of communication system, the swarm consists of a single master robot and slave robots. The master robot sends its

current position to slave robots to track its position in addition to their local searches during each interval of S&R

mission. Therefore, only the communication between the each slave robot and present master robot is necessary.

The master robot controls the entire swarm. In fact, the master robot should communicate with the other robots to

help them identify their relative location to the master robot.

In this system, all robots should be programmed to act both the roles of master and slave; and the role of master

robot can be change among all the robots existing in a swarm iteratively based on quality robots position in search

space. The slave robots will depend on master robot to avoid getting stuck at local optima. Low cost of the final

system compared to the other options such as Bluetooth, simple structure of the antenna and possibility to have

flexible antenna, short range of communication which brings security and no interfering signals and possibility to

setup a leader-less swarm structure with satisfactory efficiency are main advantages of slave-master based

communication system. Detailed procedure and interactions between members of a slave-master based swarm of

robots based S&R mission is provided in section 4.

7
4. SWARM ROBOTICS SEARCH & RESCUE (SRSR) ALGORITHM

In this section, the structure of proposed algorithm is described based on cooperative behavior of a swarm of

robots to search a victim in a post-disaster location. Flowchart depicted in Fig. 1 shows the procedure of algorithm

implementation. As like as other evolutionary optimization algorithms, it starts with creation of a population of

robots. In next step, robots should be distributed in catastrophe areas. Then, they start to search and appraise area

near their position and by utilization of high-tech sensors mounted on robots for attaining specified vital signs of

the victim, mostly human. Based on data gathered by robots, the superior individual among the swarm plays role

of the master robot and should conduct group to promoted situations; while it is mandatory for other robots to

perform as the slaves. In such the condition, quality of a robot position is completely related to its sense of

environment and, subsequently its ability to track vital signs of victim. It is noteworthy that all robots in a swarm

have similar construction specifications, and therefore their abilities to response to environmental signals and to

analyze the received information are identical as well. During the search and rescue (S&R) mission, with leadership

of the robot named master due to highest peripheral perception from victim, three policies including 1)

Accumulation: gathering of slave robots in a position near the position of master robot,

2) Exploration: search of distance between slave and master robots by slave robots and, 3) Local search: search of

specified number of robots surrounding the position of master robot should be accomplished.

As opposed to individuals in natural swarms, semi-advanced robots in a homogeneous group have a limited

ability to record previous steps. This characteristic assists members to return to previous positions if they don’t

reach a favorable position during the period of mission accomplishment. However, due to the mentioned hardware

limitations of robots in a swarm with large number of members which are designed for a coherent behavior in a

rescue mission, the chip design of robots will not permit them to record more than one or two previous steps.

Besides, the robustness of swarm robotic systems comes from the implicit redundancy in the swarm. This

redundancy allows the swarm robotic system to degrade peace-fully making the system less prone to catastrophic

failures. For instance, swarm robotic systems can create dynamic communication networks in the battlefield. The

first hypothesis for these swarms assumes that each individual owns a unique ID for communication and

cooperation. Such networks can enjoy the robustness achieved through re-configuration of the communication

nodes when some of the nodes are hit by enemy fire. However, there are some limitations for swarm like central

automatic control by which slave robots are restricted to communicate solely with master robot. While a slave robot

reaches a position better than that of master robot, it should guide swarm for rest of mission as the master robot;

8
while prior master will play role of a slave robot. In fact, they should exchange their duties and responsibilities of

S&R mission for remaining minutes ahead. The rest of paper illustrates how S&R mission is modeled based on

swarm robotics for optimization task.

4.1. Generating initial swarm of robots

In order to obtain a global solution of an optimization problem, the first step is to form decision variables of

problem as an array. In different codifications of optimization algorithms, this array has been called with different

names. For instance, it is named as chromosome, social-political position of a country and habitat in GA, ICA and

COA, respectively. In the proposed algorithm, vector of decision variables is called robot position. In an

optimization problem with N var decision variables, each initial solution is composed of an array with dimension

1 N var shows the position of a robot in search area. This position is defined by Eq. (1):

i
Positionrobot   x1 , x2 ,...., xNvar  , i  1,2,,...., Nrobots (1)

Variables x1 , x2 ,...., xNvar can be floating point numbers. Value of victim’s vital signs (VVS) monitored by each

robot is defined by Eq. (2):


VVS  f Positionrobot
i
 
 f  x1 , x2 ,...., xNvar  ,  i  1, 2,,...., N robots (2)

Where, f is referred to optimization function. Considering Nrobots as number of robots in initial swarm, a

matrix with dimension of Nrobots  N var is generated as initial population of robots. In a solution search space with

maximum and minimum limits of Varmax and Varmin , respectively; initial position of a robot can be obtained by Eq.

(3):


Positionrobot  Urnd Varmin ,Varmax ,Varsize  (3)

Function Urnd generates an array of random floating-point numbers from continuous uniform distribution with

lower and upper endpoints specified by Varmin and Varmax , respectively; and Varsize  1 Nvar .

As mentioned, after generating initial solutions and evaluating of initial population, master and slave robots are

determined based on fitness of VVSs. Then, three main policies of algorithm including accumulation, exploration,

and local search are implemented as follows:

4.2. Accumulation

During the first policy of algorithm, all slave robots should move to a position near the position of the master

9
robot aiming at ameliorating its current sense of VVS. This accumulation nearby the position of best robot is

implemented with random movements and by utilizing normal probability distribution function (PDF). The

probability density of normal distribution is given by Eq. (4):

 x   
2
1
y  f  x |  ,   e 2 2 (4)
 2

Here, parameter  is mean or expectation of distribution and parameter  is its standard deviation with its variance

then  2 . A random variable with a Gaussian distribution is said to be normally distributed and is called a normal

deviate. If   0 and   1 , the distribution is called the standard normal distribution denoted by N  0,1 . As it

can be seen in Fig. 2, with variations of two parameters  and  , different ranges of PDF based random floating

numbers can be obtained.

The master robot transmits its current position to slave robots employing predesigned communication systems

like Bluetooth technology, global positioning system (GPS) and etc. as an accumulation command. Subsequently,

slave robots start to change their position toward the position of the master robot.

Comparing normal PDF to some other well-known PDFs like Weibull, Gamma or Pareto, it should be mentioned

that the most important aspect about these functions is that these are skewed and asymmetric. These distribution

functions are mostly used to model phenomena in which generated points should be distributed with high

probabilities at first/ending stages of phenomenon. On the contrary, the great power of the normal distribution is

that the concept of Master-Slave in a swarm of robots can be transformed into a normal distribution via Central

Limit Theorem [38]. Moreover, it also serves a physical manifestation. Due to fact that there is no certain

justification about superiority of movement direction for an individual in a swarm of robots with limited level of

intelligence, especially during first iterations, robots should explore both directions with same probability to

maintain diversity and also to prevent plighting in local minimums. Using Weibull, Gamma or Pareto distributions,

probability to search on one direction toward the master robot will be dominant and logically, because of high

convergence toward local minimums during first iterations of algorithm execution, an extensive exploration of

search space for global optimum cannot be achieved.

Values  and  for master and slave robots can be obtained as follows:

4.2.1. Calculation of parameters  and  for master robot

10
 As aforementioned, master robot calls slave robots up to its current position with aim of directing population to

location with the maximum likelihood of victim presence. As long as slave robots are accumulating in position of

master robot, it starts a local search nearby the stated position. In fact, by execution of such the search, master robot

tries to improve its situation during period other robots are moving toward position in which maximum VVS has

been found. Considering C1 as first control parameter, values  and  for master robot are defined by Eq. (5) and

Eq. (6), respectively:

 robot M   r (5)


robot M   1   1
Eit 1
 
  robot M  / max 0,  1
Eit 1
  1  C   1  Position
1 robot  M  (6)

r is a random number within [0,1], Eit denotes elapsed time of S&R mission in SRSR terminology, refers to

number of iterations in algorithm codification and Positionrobot M  is position of master robot.

4.2.2. Calculation of parameters  and  for slave robots

During implementation procedure of accumulation policy, all slave robots should be relocated by their own

parameters  and  . Parameters for slave robots are defined by Eq. (7) and Eq. (8):

  
 irobot S  SCF i  Positionrobot M   Positionrobot
i
S  
(7)
 r 2  Positionrobot  S   j  | i  1,...., N slaves ; j  1,...., N var
i

,
 Positionrobot  M   j   Positionrobot  S   j  0.05
i
 

robot  S   C1  Positionrobot M   1  C1   Positionrobot S  , i  1,...., N slaves

i i
(8)

SCF i is a correction factor for parameter  corresponding to slave robot i. Calculation method of SCF value has

been detailed in section 4.2.5. The schematic representation for calculation procedure of new position of a slave

robot by executing accumulation policy in relation to master robot is shown in Fig. 3. It should be mentioned that

 
the location  p1 ,  p 2 ,  p3 enjoys maximum likelihood to be chosen as the next position of slave robot. However,

based on the variance obtained for each of the axes which refer to decision variables, the locations near the location

of master robot might be proportionally chosen with lower probabilities.

4.2.3. Generating new positions

In this section, new position of master and slave robots by utilizing parameters  and  corresponding to each

robot should be determined.

11
  
NewPositionroboti   Nrnd roboti  , robot i  , i  Master , Slaves (9)

Function Nrnd generates an array of random floating point number from normal PDF with mean parameter 

and standard deviation parameter  obtained from two previous sections. It is noteworthy; this equation is valid to

determine new position for both master and slave robots. In order to bound new obtained positions in range of

search space, following equations are utilized:


NewPositionroboti   j   max Varmin  j  , NewPositionroboti   j  ,  i  Master , Slaves; j  1,...., Nvar (10)


NewPositionroboti   j   min Varmax  j  , NewPositionroboti   j  ,  i  Master , Slaves; j  1,...., Nvar (11)


NewVVSroboti   f NewPositionrobot i  ,  i  Master , Slaves (12)

4.2.4. Progress assessment

In nature, there are a large number of creatures with some elements of cooperative behaviors. In animal

communities, many empirical evidences suggest that performance of a cooperative behavior may not be desirable

in a period of time and it was impossible to return to previous situation. Lack of a temporary memory to record

accurate previous situation and time variant features of environmental elements are two key factors for occurrence

of these types of events. However, such the happenings are not mostly probable in artificial intelligence territory.

As an illustration, a simple robot in a S&R swarm with moderate functionality which has ability to record one

previous step can change its current position to previous one if desired improvements are not achieved in new

position. This capability has been considered for rescuer robots in swarm robotics. Mathematic formulation of this

characteristic is defined by Eq. (13):

i   VVSrobot i  , i  Master , Slaves; j  1,...., N var

j j
if NewVVSrobot

i   NewVVSroboti 
j j
VVSrobot
(13)

Positionrobot[i]  NewPositionroboti 

end

However, it should be mentioned that by implementation of this policy during every minute of the S&R mission

for all robots, they rapidly gather at a local optimum location and algorithm will lose its diversity factor gradually

in search process. Therefore, after relocation of robots to their new position and sorting them based on their new

VVS quality, one of values 50% or 100% of population should be selected randomly by algorithm. If 50% is

12
selected, robot population should be divided to two groups. Then, robots in the first group which enjoy higher VVS

values are permitted to reside in new positions, if NewVVSrobot  VVSrobot ; otherwise, they are subject to return to

previous positions. Whereas, members of the second group should reside in new positions without any

consideration of improvement in NewVVS to maintain diversity of whole the population. If value 100% is selected,

method employed for the first group should be utilized for all members of population.

4.2.5. Calculation of SCF

After settling of robots at their new position at the end of each time interval of S&R mission, SCF value should

be modified for coming one minute (refers to next iteration in algorithm implementation) based on VVS change of

robots in current iteration. That is the robot which enjoys highest VVS change compared to previous one minute

should be considered as basis for modification of robots movement in coming one minute of S&R mission.

Calculation procedure of SCF is detailed in the following steps:

Step 1: Let SIF  6 as an initial value for the first iteration (algorithm is not sensitive to this value, because

value of this temporary variable will be modified iteratively).

Step 2: Calculate SCFi  SIF  ri , i  1,...., Nrobots ; ri is random number within [0,1] corresponding to robot i.

Step 3: Calculate
i
VVSMovement  VVSrobot
i
 NewVVSrobot
i
, i  1,...., Nrobots

Step 4: Select index i which refers to robot enjoys maximum VVSMovement

  
Step 5: Calculate SIF  1  max Varmax  Varmin ,  r   SCFmax ; SCFmax is value of SCF corresponding to
 

robot which is selected in step 4.

 
Step 6: Limit SIF value obtained in previous step within 0,max Varmax  , derived from six sigma theory [39].
 

Step 7: Go step 2 for new iteration.

i
It is noteworthy variable VVSMovement in step 3 should be determined before implementation of “section 4.2.4

progress assessment” in which variables Position and VVS are updated based on improvements of robots;

otherwise, it is probable about some of robots to return to their respective position if new positions are not more

justified in term of VVSrobot value.

4.3. Exploration

13
 Exploration policy is another main operator of proposed algorithm by which slave robots are allowed to search

spaces around themselves in a direct path toward/backward master robot. New position of slave robots by execution

of exploration policy can be obtained by Eq. (14):

 S   r  Positionrobot S 
i i
NewPositionrobot
(14)
+GB   Positionrobot M   Positionrobot
i
S    MF , i  1,...., N slaves

Variable GB  1 is selected randomly and vector of movement factor  MF  is calculated by

 
max 1, Positionrobot M   PositionrobotWS  at the end each iteration; here index WS refers to slave robot with the

worst VVS . Parameter MF can control convergence speed and accuracy of search at different intervals of algorithm

execution. This movement is shown in Fig. 4. As it can be seen, a random percent of current position of a slave

robot is added to/subtracted from a factor of direct path from position of slave robot to position of master robot. In

fact, the first term of exploration equation changes distance and initial angle of position of a slave robot with respect

to master robot (first step), randomly; while the second term of exploration equation moves position of slave robot

toward/backward master robot (second step). After creating new position of slave robot, all steps proposed in

section “4.2.4 progress assessment” should be implemented to determine optimal position of slave robots in current

one minute of S&R mission.

4.4. Local search

In this policy, a specified number of robots which have lowest qualities perform as worker robots. These robots

are assigned to a local search around the position of master robot. In fact, this policy is an extra accurate search by

which worker robots, that refer to low quality solutions in optimization algorithm, are permitted to improve their

position. However, residing of worker robots at new positions is possible only when they reach a position better

than position of robot enjoying second rank. Otherwise, they should return to the previous position after an

unsuccessful search. In order to implement such an expression, some arithmetic operators with elementary

operations have been employed to determine the new position of worker robots as follows:

1. Round operator
14
  
NewPosition1robotW   sign round  Positionrobot  M   robot [M]
(15)

2
NewPositionrobot  

W   sign round Positionrobot  M  
robot [M]
(16)

Operators round

 x  / round   x  round elements of x to nearest integers greater than/ less than or equal to x
, respectively; and operator sign  reflects signs of  to  element by element.

2. Exponential operator

 
  
1
3
NewPositionrobot W 


 sign  Int Positionrobot M   Fra PositionrobotM  e1 

robot[M]
(17)


     
e2
W   sign  Int Positionrobot M   Fra Positionrobot  M 
4
NewPositionrobot  (18)
 robot[M]

Functions Int    and Fra   return integer and fractional parts of elements of  , respectively. Parameters

e1 and e2 are two random integers within [2,4].

3. Combined operator

This operator is retrieved from crossover policy in GA. In fact, new position of worker robot 5 is a combination

3 4
of NewPosition robot W  and NewPositionrobotW  with random shares of P1 and P2, respectively; while,

P1  P2  100% .

5
NewPositionrobot 
 3

 
 
W     P1 % NewPositionrobotW    P2 % NewPositionrobotW   
4
 (19)

It is noteworthy such implementation is accomplished about a certain number of decision variables, not all

variables of position array to prevent sudden chaotic changes in position of obtained solutions.

4.5. Control parameters

Control parameters of algorithm are as follows:

 Mu factor [C1]: controls mean values for master and slave robots. This parameter controls dominance

degree of the master robot.

 Movement factor [C2]: determines movement pace of robots during exploration policy

 Sigma factor [C3]: determines level of divergence of the slave robots from master robot.

15
  Sigma limit [C4]: limits value of sigma factor

In this algorithm, control parameters C2, C3 and C4 are automatically tuned; while C1 should be adjusted by

user within [0.5, 0.85].

5. CASE STUDY AND RESULTS

In this paper, the effectiveness of the presented method has been assessed in two scenarios. For first scenario, a

number of standard test functions have been employed; while, in second scenario, the proposed method has been

applied to obtain optimum scheduling table for medium-scale (10 unit system) to large-scale (100 unit system)

thermal electric power generating systems. The proposed heuristic algorithm has been implemented on an Intel®

Core™ i5-460M Processor (3M Cache, 2.53 GHz).

5.1. Scenario 1: standard benchmark functions

For this scenario, the proposed method has been applied to a set of unimodal (F1-F10; for scales of small (D1),

medium (D2) and large (D3)) and multimodal (F11-F14) standard benchmark functions. These functions have been

detailed in Table 2 and. Table 3The proposed algorithm has been run for 50 times. In order to guarantee a fair

comparison among different algorithms and to avoid consideration of execution time as the comparison index

(because different qualities of codification for a same problem formulation may result in different execution times),

maximum number of function evaluations (NFE) is considered as stop criterion of algorithm implementation. NFE

parameter has been set to 5000 in case of unimodal and 10000 in case of multimodal benchmark functions. For all

other algorithms, the maximum NFE has been set to that of the proposed algorithm. Multimodal functions include

two functions with a few local minima (F11 and F12 depicted by Fig. 5) and two functions with many local minima

(F13 and F14 depicted by Fig. 6).

In implementation procedure of algorithm, stop criterion is defined by Eq. (20):

fend .it (robot M  )  global min  StC (20)

fend .it (robot M  ) is cost of the best obtained solution at last iteration of algorithm and StC is stopping criterion

of the algorithm, when it converges into 10-6 and 10-3 tolerance in cases of unimodal and multimodal functions,

respectively. Mean and standard deviation for obtained results are calculated by Eq. (21):

16
 1
1 n 2 2
Std . dev     xi  x   (21)
n 
 i 1 

xi is vector of solutions resulted by runs of program and x is mean value of solutions:

1 n
x  xi
n i 1
(22)

Obtained results including best (min), arithmetic average (mean), worst (max) and standard deviation (Std. dev.)

after 50 independent executions of proposed SRSR optimization algorithm have been compared with those of

TLBO, ABC, ICA, BBO, GSA, PSO and DE in Table 4- Table 6 for dimensions D1-D3, respectively. As it can

be deduced from Table 4, proposed SRSR methodology shows the best performances among presented methods;

while the worst performance is related GSA. ABC and ICA enjoy relatively equal performances and also weaker

performances compared to SRSR and TLBO. Obviously, proposed algorithm demonstrates its superiority and

efficiency in cases of reaching global solution and lower standard deviations. Moreover, in case of max (worst)

solution, proposed algorithm has shown its excellence in comparison task for all benchmark functions. Although,

BBO presents a better performance than GSA, solutions obtained by executions of these algorithms do not enjoy

expected quality to be comparable with other algorithms. Moreover, considering little changes in ranking, PSO and

DE provide solutions relatively as similar as TLBO in most of the cases.

Referring to Table 5 and Table 6, superiority of SRSR method can be highly demonstrated by comparison of

obtained results with those of literature review for medium and large scale unimodal problems.

Convergence curves of the proposed algorithm versus NFE for different benchmark functions are compared

with those of ABC, TLBO, ICA, BBO and GSM in Fig. 7. In the case of single objective functions, superiority of

SRSR over other algorithms in term of the solution quality is evident; however such the superiority cannot be

observed for all benchmark functions in term of convergence pace. It should be mentioned that convergence quality

of algorithms is completely related to features of benchmark functions. Needless to say, level of adaptability of an

optimization algorithm cannot be determined by applying to a limited number of problems. As an illustration, GSM

algorithm does not provide quality solutions in most of cases, while its convergence speed surpasses that of other

algorithms for function F1.

17
 In case of multimodal functions (F11-F14), results of comparison of SRSR with other algorithms based on

success times (number of runs in which all global peaks have been found), mean (average final value of all runs of

algorithm) and average number of optima found (the average number of peaks found over 50 runs with StC=1×10-
3
) are detailed in Table 7.As it can be seen, maximum number of success times has been obtained by SRSR.

Considering more challengeable feature of multimodal benchmark functions compared to unimodal functions,

success rates for multimodal functions are not as high as unimodal functions. In case of multimodal functions F11-

F14, TLBO and ABC have better performances than ICA in most of cases. It is noteworthy, with the increment of

complexity of benchmark functions from F11 to F14, quality of solutions are reduced. As an illustration, average

optima found by SRSR for the function F11 is 2 where there are two global peaks (100 % success rate); However,

there are 18 global peaks for function F14 and the best average optima found is 16.18 by SRSR (89.9 % success

rate). Such the statement is true about number of success times and mean values as well. Results obtained by BBO

and GSA implementations show none of them are capable of comparing with other algorithms. However, TLBO

has approximately equal performances to SRSR about functions F11 and F12, superiorities of proposed algorithm

are obvious in cases of benchmark functions F7 and F8 where complexities of test functions are increased.

5.2. Scenario 2: practical power system problem

In order to evaluate the quality of the proposed technique for a practical problem, unit commitment as one the

most important problems in operation of electric power system is assessed. This problem is a large-scale mixed

integer problem (MIP) in which day-ahead power production scheduling of thermal generating units should be

determined aiming at minimizing operating costs including fuel cost, start-up and shut-down costs.

This problem is subjected to some system (e.g. demand and reserve) and units (e.g. minimum on/off time, ramp

up/down rates) constrains. It is noteworthy due to large scale and non-convex features of this problem; it is a costly

problem from computational view point which has been investigated in many researches [44]-[51].

5.2.1. UC problem formulation

A. Objective Function

The objective function of UC problem (UCOF) for a generating system consisting N units and for T hours of

scheduling horizon can be formulated as:

18
 UCOF  minimize TC  (23)

Here:

 
T N
TC   Ci P i, t  .I i, t   ST i, t  .I  i, t  .[1  I i, t 1 ] (24)
t 1 i 1
 

I i, t  commitment state of unit i at time t, and fuel cost for generator i with fuel cost coefficients ai , bi and ci

is defined by Eq. (25):

 
Ci P i , t  .I i , t   ai  bi  P i , t   ci  P i , t 2 (25)

Start-up costs can be expressed as an exponential or linear function of the time [39]. In this paper, start-up costs

are modeled as a two-valued (hot start/cold start) staircase function [44]. For unit i with hot start-up cost ( HSCi

) and cold start-up ( CSCi ), time-dependent cost of start-up is defined by Eq. (26) [44]:

 HSCi , if MDTi  DTi  MDTi  CSH i

ST i, t    (26)
CSCi , if DTi  MDTi  CSH i

MDTi is minimum down-time limit of unit i, CSH i is cold start hour of unit i, and DTi is the unit’s down time.

Similarly, Shutdown cost can be considered as a fix, staircase or exponential cost for each unit per shut down.

However, in this paper, the shut-down cost of units is not considered during simulation procedure.

B. Constraints

The PBUC problem is formulated subject to following system and unit constraints [46].

Unit Constraints

1. Unit generation limits

Pgmin
i
 P i , t  I i , t   Pgmax
i (27)

Power production of units is constrained to minimum and maximum limits of Pgmin

i
and Pgmax
i
, respectively.

2. Ramp up/down rates

By considering ramp up/down rates constrains, difference of values in production levels is limited to a maximum

value in consecutive periods of scheduling horizon. For generator i with the ramp-up RU i and ramp-down RDi ,

this constrain is formulated by Eq. (28) and Eq. (29):

19
 When generator ramps up: 
P i, t   min Pgmax
i
, Pi, t 1    RUi  (28)

When generator ramps down: 

P i , t   max Pgmin
i
, P i , t 1    RDi  (29)

 is equal to 60 min and it is the UC time step.

3. Unit minimum ON/OFF durations

This constrain limits minimum duration of ON status of generator operation before shutting the unit down as

well as minimum hours that should be spent on OFF status before next start up. This constrain can be formulated

by Eq. (30):

 X ic  MUTi if X ic  0
 c (30)
 X i  MDTi if X ic  0

c
MUTi is minimum up-time limit of unit i, and X i is a signed integer which represents the continuous ON/OFF

c
status duration of the ith cycle of the ith unit. The sum of the absolute values of X i for all units must be equal to

the scheduling horizon.

System Constraints

1. Demand constraint

N
 Pi, t  .Ii, t   PDt  t  1,..., T (31)
i 1

PD t  is system load demand at hour t.

2. Spinning reserve of the system

N
 Pi, t  .Ii, t   PDt  +PRt  t  1,...,T (32)
i 1

PR t  is system reserve at hour t. P i , t  should be obtained by (28)-(29) with   10min . This is 10-min maximum

response-rate-constrained active power of generating unit i.

Obviously, economic load dispatch (ELD) should be executed in order to reach optimal power of each

generating unit after determining commitment status of units. Traditional ELD has been employed to specify power

dispatched during scheduling horizon.

20
5.2.2. Case study

The proposed SRSR algorithm has been applied to systems with 10, 20, 40, 60, 80, and 100 generating units for

a 24-h scheduling time horizon with one period per hour. Data for the generation system consisting 10 units is

detailed inTable 8.. The data for forecasted load is given in Table 9[45]. Single line diagram of the IEEE 39 bus

system with ten generation units is shown in Fig. 8. Method of MIP codification of solutions is based on [45].

For the 20-generating unit system, the data of the ten generating unit system duplicated and the load data

multiplied by 2. Similarly, data for 10 unit system and respective load demand scaled appropriately in order to

provide data for larger-scale test systems. 10% of hourly load demand of system is considered as spinning reserve

of generation in all cases. Optimum power dispatched corresponding to the best solution obtained by SRSR for

the given 10-unit system and commitment status of generating units for 20, 40, 60, 80 and 100 unit system are

shown in Table 10 to Table 15 Operation costs corresponding to the obtained solutions for the given test systems

are detailed in Table 16.The results obtained by SRSR have been compared with other methods in Table 17 for 6

given generating systems. It can be seen, with the increment of scale of test systems, hourly reserve constrains are

satisfied marginally. In other words, changes in size of case studies donate algorithm freedom to search marginal

points more precisely. Optimal parameters of the proposed algorithm have been obtained through several runs of

program. Obviously, number of initial solutions should be proportionally increased along with increment of

generating system scale. Needless to say, a trade-off should be placed between the number of initial solutions and

optimality of results as a further increase in the number of solutions may cause algorithm to incur higher

computational costs; while, convergence may not improve as expected. Effectiveness of proposed SRSR algorithm

in term of optimality of obtained solutions (lower operation cost of generating systems) has been validated by

comparison of the results as detailed in Table 17.

6. CONCLUSION

In this paper, a novel algorithm called SRSR has been proposed based on collective search approach of a swarm

of robots to find a victim in post-disaster locations. In contrast to previously nature-inspired optimization

algorithms, collective intelligence in artificial life is the main inspiration of the presented method. A set of problems

including several standard benchmark functions and a practical electric power system problem have been utilized

to assess the performance of the presented optimization algorithm in terms of quality solutions as well as

convergence pace. However, quality of the solutions obtained by SRSR surpasses those of other algorithms in

21
almost all the cases; obviously, it is not possible to conclude that the proposed algorithm is universally superior to

all other previously proposed population-based methods. In fact, the acceptable performance of SRSR in cases of

benchmark functions and UC problem provides some evidences that this algorithm can be successfully applied to

other challengeable optimization problems in different branches of science. Finally, authors proposed some

research directions for future works as follows:

 Considering eye-catching role of mean and variance factors of distribution function on convergence of

algorithm, a novel method for updating these factors can ameliorate algorithm performance.

 However normal PDF has shown an acceptable performance about presented problems, utilizing of other

PDFs like Gamma and Pareto during ending iterations of algorithm can improve convergence pace.

 However, authors defined movement factor as distance between best and worst solutions, a more targeted

calculation method of this factor can advance quality of second operator and subsequently overall

performance of optimization algorithm.

 The functions round and exponent can be categorized as basic operators that have never been utilized

separately to improve and also accelerate performance of any optimization algorithm. Employing similar

operators my assist algorithm to reach a better quality.

22
 References

[1] Bertsekas, D. P., & Scientific, A. (2015). Convex Optimization Algorithms. Athena Scientific.

[2] Dorigo, M. (1992). Optimization, Learning and Natural Algorithms. Ph.D. thesis, Dipartimento di Elettronica,
Politecnico di Milano, Italy.

[3] Karaboga, D. (2005). An idea based on honey bee swarm for numerical optimization. Technical report-tr06,
Erciyes university, engineering faculty, computer engineering department.

[4] Farmer, J. D., Packard, N., Perelson, A. (1986). The immune system, adaptation and machine learning. Physica
D: Nonlinear Phenomena, 2, 187–204.

[5] Passino, K. M. (2002). Biomimicry of bacterial foraging for distributed optimization and control. Control
Systems, IEEE, 22(3), 52-67.

[6] Yang, X. S. (2010). A new meta-heuristic bat-inspired algorithm. In Nature inspired cooperative strategies for
optimization (NICSO 2010) (pp. 65-74). Springer Berlin Heidelberg.

[7] Erol, O. K., & Eksin, I. (2006). A new optimization method: big bang–big crunch. Advances in Engineering
Software, 37(2), 106-111.

[8] Simon, D. (2008). Biogeography-based optimization. Evolutionary Computation, IEEE Transactions on, 12(6),
702-713.

[9] Kaveh, A., & Talatahari, S. (2010). A novel heuristic optimization method: charged system search. Acta
Mechanica, 213(3-4), 267-289.

[10] Lam, A., & Li, V. O. (2010). Chemical-reaction-inspired meta-heuristic for optimization. Evolutionary
Computation, IEEE Transactions on, 14(3), 381-399.

[11] Yang, X. S., & Deb, S. (2009, December). Cuckoo search via Lévy flights. In Nature & Biologically Inspired
Computing, 2009. NaBIC 2009. World Congress on (pp. 210-214). IEEE.

[12] Rajabioun, R. (2011). Cuckoo optimization algorithm. Applied soft computing, 11(8), 5508-5518.

[13] Storn, R., & Price, K. (1997). Differential evolution–a simple and efficient heuristic for global optimization over
continuous spaces. Journal of global optimization, 11(4), 341-359.

[14] Yang, X. S. (2010). Nature-inspired metaheuristic algorithms. Luniver press.

[15] Yang, X. S. (2012). Flower pollination algorithm for global optimization. In Unconventional computation and
natural computation (pp. 240-249). Springer Berlin Heidelberg.

[16] John Henry Holland. (1992). Adaptation in natural and artificial systems: an introductory analysis with
23
 applications to biology, control, and artificial intelligence. MIT press.

[17] Rashedi, E., Nezamabadi-Pour, H., & Saryazdi, S. (2009). GSA: a gravitational search algorithm. Information
sciences, 179(13), 2232-2248.

[18] Mirjalili, S., Mirjalili, S. M., & Lewis, A. (2014). Grey wolf optimizer. Advances in Engineering Software, 69, 46-
61.

[19] He, S., Wu, Q. H., & Saunders, J. R. (2009). Group search optimizer: an optimization algorithm inspired by
animal searching behavior. Evolutionary Computation, IEEE Transactions on, 13(5), 973-990.

[20] Geem, Z. W., Kim, J. H., & Loganathan, G. V. (2001). A new heuristic optimization algorithm: harmony search.
Simulation, 76(2), 60-68.

[21] Shah-Hosseini, H. (2009). The intelligent water drops algorithm: a nature-inspired swarm-based optimization
algorithm. International Journal of Bio-Inspired Computation, 1(1-2), 71-79.

[22] Atashpaz-Gargari, E., & Lucas, C. (2007, September). Imperialist competitive algorithm: an algorithm for
optimization inspired by imperialistic competition. In Evolutionary Computation, 2007. CEC 2007. IEEE
Congress on (pp. 4661-4667). IEEE.

[23] Kashan, A. H. (2011). An efficient algorithm for constrained global optimization and application to mechanical
engineering design: League championship algorithm (LCA). Computer-Aided Design, 43(12), 1769-1792.

[24] Eberhart, R. C., & Kennedy, J. (1995, October). A new optimizer using particle swarm theory. In Proceedings
of the sixth international symposium on micro machine and human science (Vol. 1, pp. 39-43).

[25] Kirkpatrick, S., Gelatt, C. D., & Vecchi, M. P. (1983). Optimization by simulated annealing. science, 220(4598),
671-680.

[26] Pattnaik, S. S., Bakwad, K. M., Sohi, B. S., Ratho, R. K., & Devi, S. (2013). Swine influenza models based
optimization (SIMBO). Applied Soft Computing, 13(1), 628-653.

[27] Rao, R. V., Savsani, V. J., & Vakharia, D. P. (2012). Teaching–learning-based optimization: an optimization
method for continuous non-linear large scale problems. Information Sciences, 183(1), 1-15.

[28] Carrizosa, E., Dražić, M., Dražić, Z., & Mladenović, N. (2012). Gaussian variable neighborhood search for
continuous optimization. Computers & Operations Research, 39(9), 2206-2213.

[29] Bayraktar, Z., Komurcu, M., & Werner, D. H. (2010, July). Wind Driven Optimization (WDO): A novel nature-
inspired optimization algorithm and its application to electromagnetics. In Antennas and Propagation Society
International Symposium (APSURSI), 2010 IEEE (pp. 1-4). IEEE.

[30] Reynolds, C. W. (1987, August). Flocks, herds and schools: A distributed behavioral model. In ACM Sig-graph
24
 Computer Graphics (Vol. 21, No. 4, pp. 25-34). ACM.

[31] Trianni, V. (2008). Evolutionary swarm robotics: evolving self-organising behaviours in groups of autonomous
robots (Vol. 108). Springer.

[32] Tan, Y., & Zheng, Z. Y. (2013). Research advance in swarm robotics. Defence Technology, 9(1), 18-39.

[33] Trianni, V., Nolfi, S., & Dorigo, M. (2008). Evolution, self-organization and swarm robotics. In Swarm
Intelligence (pp. 163-191). Springer Berlin Heidelberg.

[34] Murphy, R. R., Tadokoro, S., Nardi, D., Jacoff, A., Fiorini, P., Choset, H., & Erkmen, A. M. (2008). Search and
rescue robotics. In Springer Handbook of Robotics (pp. 1151-1173). Springer Berlin Heidelberg.

[35] Mondada, F., Floreano, D., Guignard, A., Deneubourg, J. L., Gambardella, L., Nolfi, S., & Dorigo, M. (2002).
Search for rescue: an application for the SWARM-BOT self-assembling robot concept. Technical Report, LSA2-
I2S–STI. Swiss Federal Institute of Technology: Lausanne, Switzerland.

[36] Witkowski, U., & Zandian, R. (2013). Novel Method of Communication in Swarm Robotics Based on the NFC
Technology. In Towards Autonomous Robotic Systems (pp. 377-389). Springer Berlin Heidelberg.

[37] Pilania, V. K., Panda, S., Mishra, S., & Mishra, A. (2009, February). A Novel Approach to Swarm Bot
Architecture. In Informatics in Control, Automation and Robotics, 2009. CAR'09. International Asia Conference
on (pp. 418-422). IEEE.

[38] Khaluf, Y., Birattari, M., & Rammig, F. (2013, December). Probabilistic analysis of long-term swarm
performance under spatial interferences. In International Conference on Theory and Practice of Natural
Computing (pp. 121-132). Springer Berlin Heidelberg.

[39] Hahn, G. J., Doganaksoy, N., & Hoerl, R. (2000). The evolution of six sigma. Quality Engineering, 12(3), 317-
326.

[40] Rubenstein, M., Ahler, C., & Nagpal, R. (2012, May). Kilobot: A low cost scalable robot system for collective
behaviors. In Robotics and Automation (ICRA), 2012 IEEE International Conference on (pp. 3293-3298). IEEE.

[41] Murphy, R., & Stover, S. (2006). Gaps analysis for rescue robots. ANS 2006: Sharing Solutions for Emergencies
and Hazardous Environments.

[42] Tan, Y. (2013). Swarm Robotics: Collective Behavior Inspired by Nature. J Comput Sci Syst Biol, 6, e106.

[43] Schlenoff, C., & Messina, E. (2005, November). A robot ontology for urban search and rescue. In Proceedings
of the 2005 ACM workshop on Research in knowledge representation for autonomous systems (pp. 27-34).
ACM.

[44] Damousis, I. G., Bakirtzis, A. G., & Dokopoulos, P. S. (2004). A solution to the unit-commitment problem using
25
 integer-coded genetic algorithm. Power Systems, IEEE Transactions on, 19(2), 1165-1172.

[45] Damousis, I. G., Bakirtzis, A. G., & Dokopoulos, P. S. (2004). A solution to the unit-commitment problem using
integer-coded genetic algorithm. Power Systems, IEEE Transactions on, 19(2), 1165-1172.

[46] Hadji, M. M., & Vahidi, B. (2012). A solution to the unit commitment problem using imperialistic competition
algorithm. Power Systems, IEEE Transactions on, 27(1), 117-124.

[47] Virmani, S., Adrian, E. C., Imhof, K., & Mukherjee, S. (1989). Implementation of a Lagrangian relaxation based
unit commitment problem. Power Systems, IEEE Transactions on, 4(4), 1373-1380.

[48] Uyar, A. Ş., Türkay, B., & Keleş, A. (2011). A novel differential evolution application to short-term electrical
power generation scheduling. International Journal of Electrical Power & Energy Systems, 33(6), 1236-1242.

[49] Eslamian, M., Hosseinian, S. H., & Vahidi, B. (2009). Bacterial foraging-based solution to the unit-commitment
problem. Power Systems, IEEE Transactions on, 24(3), 1478-1488.

[50] Ebrahimi, J., Hosseinian, S. H., & Gharehpetian, G. B. (2011). Unit commitment problem solution using shuffled
frog leaping algorithm. Power Systems, IEEE Transactions on, 26(2), 573-581.

[51] Azizipanah-Abarghooee, R., Niknam, T., Bavafa, F., & Zare, M. (2014). Short-term scheduling of thermal power
systems using hybrid gradient based modified teaching–learning optimizer with black hole algorithm. Electric
Power Systems Research, 108, 16-34.

26
Fig. 1. Flow chart of the proposed SRSR algorithm

27
Fig. 2. Normal distribution probability density function

28
Fig. 3. Movement of slave robot toward master robot by the accomplishment of accumulation policy

29
Fig. 4. Two steps of exploration procedure

30
Fig. 5. Multimodal functions with a few local minima (f11: Six Hump Camel Back and f12: Himmelblau)

31
Fig. 6. Multimodal functions with many local minima (f13: Hansen and f14: Shubert)

32
(a) Function F1 (b) Function F2

(c) Function F3 (d) Function F4

(e) Function F5 (f) Function F6

(g) Function F7 (h) Function F8

33
(i) Function F9 (j) Function F10

Fig. 7. Algorithms’ convergence versus NFE for functions F1-F10

34
Fig. 8. Single line diagram of IEEE 39 bus system

35
 Table 1. A literature overview of heuristic algorithms
Algorithm Abbr. Year Inspiration

1 Ant colony optimization ACO [2] 1992 Pheromone trail laying behavior of real ant colonies
2 Artificial Bee Colony ABC [3] 2005 Natural foraging behavior of honey-bees

3 Artificial immune system algorithm AIS [4] 1986 Principles and processes of the vertebrate immune system

4 Bacterial foraging optimization BFA [5] 2002 Behaviors of E. coli bacteria during their whole lifecycle,
including chemo-taxis, communication, elimination,
reproduction, and migration
5 Bat Algorithm BA [6] 2010 Echolocation characteristics of micro-bats
6 Big Bang–Big Crunch BB–BC [7] 2006 Evolutionary theory of the universe based on the big bang
and big crunch
7 Biogeography-Based Optimization BBO [8] 2008 Geographical distribution of biological organisms (the
migration behavior of island species)

8 Charged system search CSS [9] 2010 Coulomb’s law and laws of motion

9 Chemical Reaction Optimization CRO [10] 2010 Nature of chemical reactions process

10 Cuckoo search CS [11] 2009 Obligate brood parasitic behavior of some cuckoo species
in combination with the Levy flight behavior of some birds
and fruit flies
11 Cuckoo Optimization Algorithm COA [12] 2011 Special lifestyle of Cuckoo birds and their characteristics in
egg laying and breeding
12 Differential evolution DE [13] 1997 It uses multi agents or search vectors to carry out search
13 Firefly Algorithm FA [14] 2008 Flashing behavior of fireflies.
14 Flower Pollination Algorithm FPA [15] 2012 Pollination process of flowers

15 Genetic Algorithm GA [16] 1975 Natural evolution, such as inheritance, mutation, selection,
and crossover
16 Gravitational Search Algorithm GSA [17] 2009 Newton's law of universal gravitation and mass interactions

17 Grey wolf optimizer GWO [18] 2014 Leadership hierarchy and hunting mechanism of grey
wolves in nature
18 Group search optimizer GSO [19] 2009 animal searching behavior and group living theory

19 Harmony Search Algorithm HSA [20] 2001 Musical process of searching for a perfect state of harmony
(improvisation of music players)
20 Intelligent Water Drops IWD [21] 2009 Actions and reactions that occur between river's bed and the
water drops that flow within
21 Imperialist Competitive Algorithm ICA [22] 2007 Imperialistic competition among empires
22 League Championship Algorithm LCA [23] 2014 Competition of sport teams in a sport league

33 Particle Swarm Optimization PSO [24] 1995 Animal collective behavior, the movement of the swarm and
the intelligence of the swarm
24 Simulated Annealing SA [25] 1983 Metallurgic process of heating and controlled cooling of a
material
25 Swine Influenza Model based SIMBO [26] 2013 Susceptible–Infectious–Recovered (SIR) models of swine
Optimization flu
26 Teaching–learning-based TLBO [27] 2012 Philosophy of teaching and learning process in a classroom
optimization
27 Variable Neighborhood Search VNS [28] 2012 Perturbing the incumbent solution by adding some
Gaussian distribution generated noise
36
28 Wind Driven Optimization WDO [29] 2010 Wind flow from high pressure points to low pressure points
(Newton's second law of motion)

Table 2. Standard unimodal test functions

Global Dimension
Func Definition Interval
minimum D1 D2 D3
N

F1 f ( x)   xi2 [-100,100] 0 10 50 200

i 1

D
F2 f   [100( xi 2 - xi 1 )2  (1- xi ) 2 ] [-10,10] 0 2 50 200
i 1

f ( x, y)  y sin(4 x)  1.1x sin(2 y)

[−10, 10]
F3 -19.8623 2 - -
n
F4 f  10n   ( xi 2  10cos(2 xi )) [-5.12,5.12] 0 3 50 200
i 1

1 D 2 1 D
F5 f (x )  -20exp(-0.2 
30 i 1
x i ) - exp(  cos 2 x i )
30 i 1
[−30, 30] 8.8818e-16 10 50 200

N
F6 f ( x)   ( xi2  xi ) cos( xi ) [-10,10] -100.22375×D 2 50 200
i 1

F7 f ( x, y)  x sin(4 x)  1.1y sin(2 y) [-10,10] -18.5547 2 - -

N
F8 f ( x)   ( xi  10 cos( xi ) [-10,10] -10×D 10 50 200
i 1

sin 2 ( ( x12  x22 ))  0.5

F9 f ( x)  0.5  [-10,10] 0 2 - -
1  0.001( x12  x22 )
N
xi2 N
x
F10 f ( x)  1     cos( i ) [-10,10] 0 10 50 200
i 1 4000 i 1 i

37
Table 3. Standard multimodal test functions
Global Global
Func Definition Interval
minimum peaks
 x4 
F11  3 

f (x)   4  2.1x12  1  x12  x1 x2  4  4 x22 x22  x1 ∈ [−1.9, 1.9]
x2 ∈ [−1.1, 1.1]
−1.03162 2

x1 ∈ [−6, 6]
  x 
2 2
F12 f (x)  x12  x2  11  x22  7 0 4
1
x2 ∈ [−6, 6]
 5   5  x1 ∈ [−10, 10]
F13 f (x)    i cos   i  1 x1  i      j cos   j  1 x2  j   −176.541 18*
 i 1   j 1 

x2 ∈ [−10, 10]

 5   5  x1 ∈ [−10, 10]
F14 f (x)    i cos   i  1 x1  i      j cos   j  1 x2  j   −186.730 18*
 i 1   j 1 

x2 ∈ [−10, 10]

* indicates minimum number of peaks for 2D function

38
Table 4. Comparison of best obtained solutions from different algorithms (unimodal function); dimension: D1
ABC TLBO ICA BBO GSA PSO DE SRSR
-06 -14 -09
Min (best) 0.0368 3.24×10 51.6464 4.4543 5.8254 4.16×10 41.82×10 0
Mean 0.4282 9.34×10 -05
196.9432 17.3496 15.2565 8.99×10-12 4.57×10-08 0
F1
Max (worst) 1.4496 0.0011 606.8374 52.3666 24.0362 1.83×10-10 2.73×10-07 0
Std. dev. 0.3499 1.82×10 -04
117.6103 10.7962 4.0361 2.70×10-11 5.95×10-08 0
Min (best) 5.61×10-07 6.83×10-05 1.06×10-05 0.0306 0.0086 3.13×10-16 1.82×10-18 0
Mean 0.0017 2.42×10-02 0.0272 1.1487 0.349 374×10-09 1.96×10 -09
0
F2
Max (worst) 0.0177 0.1654 0.5372 5.5865 1.1296 4.95×10-08 7.14×10-08 0
Std. dev. 0.0035 0.0328 0.0787 1.3068 0.2913 1.03×10-08 1.07×10-08 0
Min (best) -19.8623 -19.8623 -19.8623 -19.7337 -19.7744 -19.8623 -19.862 -19.8623
Mean -19.8623 -19.8621 -19.786 -18.3463 -18.5486 -17.0653 -19.337 -19.8623
F3
Max (worst)-19.8623 -19.8582 -18.5916 -16.8589 -17.4261 -11.6434 -17.494 -19.8622
Std. dev. 4.48×10-10 5.71×10-04 0.3048 0.6638 0.5188 1.9560 0.96656 1.69×10-06
Min (best) 1.99×10-06 0 2.95×10-05 0.011 1.0318 0 0 0
Mean 0.3137 2.54×10-13 0.2013 0.6115 4.2586 1.3730 0 0
F4
Max (worst) 1.6941 5.01×10-12 1.0074 1.7147 7.7237 4.9748 0 0
Std. dev. 0.4339 8.80×10-13 0.3833 0.4815 1.5559 1.3164 0 0
Min (best) 0.1132 0.0027 3.7334 1.4843 4.8953 5.32×10-08 1.38×10-05 8.88×10-16
Mean 0.4015 0.0451 6.5576 2.8951 6.1779 0.1715 1.28×10-04 8.88×10-16
F5
Max (worst) 1.3313 0.955 10.5058 4.7718 7.3042 1.6462 1.73×10-03 8.88×10-16
Std. dev. 0.2694 0.1332 1.2746 0.599 0.5294 0.4344 2.49×10-04 0
Min (best) -200.4475 -200.4475 -200.4475 -200.4471 -200.4395 -200.4475 -200.4475 -200.4475
Mean -200.0689 -200.4475 -198.5618 -199.9672 -199.0868 -150.7990 -197.6926 -200.4474
F6
Max (worst) -181.5918 -200.4475 -181.5918 -198.2014 -193.2686 -89.8412 -181.5918 -200.4470
Std. dev. 2.6664 1.56×10-12 5.71 0.5861 1.3386 42.3979 4.6158 1.08×10-04
Min (best) -18.5547 -18.5547 -18.5547 -18.5544 -18.5516 -18.5547 -18.5547 -18.5547
Mean -18.5521 -18.5547 -18.5547 -18.4316 -18.3571 -16.6552 -18.5547 -18.5547
F7
Max (worst) -18.4758 -18.5547 -18.5547 -16.9746 -17.8192 -10.3993 -18.5547 -18.5545
Std. dev. 0.0126 1.73×10-07 1.31×10-07 0.2269 0.1867 2.0012 1.79×10-14 4.73×10-05
Min (best) -76.2009 -80.2417 -70.9271 -87.5371 -45.7483 -76.7780 -80.574 -100
Mean -65.3395 -73.1429 -59.3421 -77.647 -39.1211 -65.9410 -67.848 -100
F8
Max (worst) -59.7662 -67.8236 -35.7792 -71.1918 -32.0331 -61.2966 -60.995 -100
Std. dev. 3.4832 2.8901 8.1086 3.8414 3.5161 4.2108 -60.995 0
Min (best) 0 1.24×10-04 3.98×10-12 2.90×10-05 3.60×10-05 0 0 0
Mean 9.46×10-10 0.0163 8.30×10-03 0.0402 0.0073 0.0161 2.02×10-03 0
F9
Max (worst) 4.71×10-08 0.0406 0.0447 0.0445 0.0265 0.0447 4.47×10-02 0
Std. dev. 6.66×10-09 0.0106 1.73×10-02 0.0125 0.0066 0.0217 8.86×10-03 0
Min (best) 4.40×10-04 6.34×10-05 0.9997 0.0168 0.3207 0 0.0309 0
Mean 0.1279 0.0334 1.0004 0.0579 0.5894 0.0399 0.1502 0
F10
Max (worst) 0.4574 0.0456 1.0019 0.1129 0.7471 0.1229 0.3316 0
Std. dev. 0.1069 0.0166 5.69×10-04 0.0222 0.1001 0.0313 0.0574 0

39
Table 5. Comparison of best obtained solutions from different algorithms (unimodal function); dimension: D2
ABC TLBO ICA BBO GSA PSO DE SRSR
3 86.7748 338.26
Min (best) 20.9616 0.0770 90.4533 2.65×10 202.0650 0
Mean 259.4144 1.2408 331.8823 4.32×103 300.4854 8.3405 685.04 0
F1
Max (worst) 1013.7 5.9629 869.7397 6.20×103 628.7836 493.2527 1248.9 0
Std. dev. 248.9055 1.1920 168.0125 767.6956 65.5196 96.6024 229.03 0
Min (best) 431.4416 49.0658 799.9275 1.21×104 7.15×104 156.6446 3075.6 48.5066
Mean 8.48×103 49.9723 1.93×103 2.60×104 1.61×105 330.3673 9507.8 48.7696
F2
Max (worst)1.16×105 52.8635 4.03×103 5.21×104 2.71×105 883.0405 26352 48.9163
Std. dev. 1.77×104 0.8572 857.8237 8.86×103 4.91×104 151.5299 5542 0.1051
Min (best) 0 24.9020 130.1062 97.1867 478.2112 65.4286 327.76 0
Mean 61.4583 101.6300 231.9110 117.6285 520.0191 99.5531 417.87 0
F4
Max (worst) 252.7431 195.3031 342.9199 152.6470 553.5434 151.2542 480.95 0
Std. dev. 85.4016 42.9534 48.3793 11.8371 17.0121 22.4923 30.229 0
Min (best) 5.9047 0.0621 6.0671 8.7395 8.2637 2.7354 4.7953 8.88×10-16
Mean 9.6925 0.2186 9.3658 10.1661 9.1511 4.5972 7.6771 8.88×10-16
F5
Max (worst) 14.6906 0.4458 17.2838 11.6550 9.7827 7.6744 20.126 8.88×10-16
Std. dev. 2.1077 0.0935 2.5389 0.5837 0.3446 1.0973 3.0628 0
Min (best) -4236.7 -4695.8 -3018.2 -4621.7 -1181.9 -2881.8
-4702.4 -4813.2
Mean -3845.1 -4545.1 -2529.5 -4445.5 -806.6711 -2219.7
-4473.6 -4681.6
F6
Max (worst) -3490.0 4344.4 -1936.4 -4222.2 -539.5142 -1602.1
-4086.1 -4480.3
Std. dev. 183.5911 81.2469 218.2588 79.4057 134.3862 285.28
115.3663 62.73
Min (best) -316.7334 -299.1715 -336.5846 -318.0996 -115.5267 -350.1117 -184.01 -500
Mean -289.7390 -276.2601 -295.9097 -298.1003 -90.9463 -331.5742 -129.29 -500
F8
Max (worst) -249.4120 -249.5225 -184.8459 -281.6284 -64.5994 -317.5358 -82.62 -500
Std. dev. 15.6372 11.4651 36.8068 7.5289 11.6338 8.5092 25.41 0
-05 0.0031 0.1020
Min (best) 0.0288 3.24×10 0.0278 0.5816 0.9832 0
-04 0.0304 0.2985
Mean 0.1971 3.14×10 0.870 0.7703 1.0399 0
F10
Max (worst) 0.7102 0.0012 0.1915 0.9054 1.0546 0.1073 0.8113 0
-04 0.0218 0.1604
Std. dev. 0.1194 2.34×10 0.3888 0.0694 0.0117 0

40
Table 6. Comparison of best obtained solutions from different algorithms (unimodal function); dimension: D3
ABC TLBO ICA BBO GSA PSO DE SRSR
5 4 5 3 10103 89361
Min (best) 1.37×10 0.6033 4.18×10 1.50×10 4.88×10 0
5
Mean 1.83×105 2.7282 5.78×104 1.74×105 8.91×103 15292 1.22×10 0
F1
Max (worst) 2.32×105 7.5115 7.91×104 1.93×105 1.39×104 22349 1.63×105 0
Std. dev. 2.42×104 1.4664 8.66×103 9.53×103 1.98×103 2397.4 17172 0
Min (best) 3.49×106 200.1324 5.16×105 4.16×106 4.23×105 41095 2.68×106 198.5736
Mean 7.09×106 203.7160 9.23×105 5.13×106 6.40×105 74358 4.07×106 198.8343
F2
Max (worst) 1.06×107 212.1350 1.82×106 6.49×106 1.27×106 134610 5.79×106 198.8995
Std. dev. 1.56×106 2.8224 2.71×105 4.67×105 1.38×105 18740 8.33×105 0.0603
Min (best) 1.46×103 18.5455 1.34×103 1.48×103 1.90×103 777.6756 2345.9 0
Mean 1.66×103 66.9449 1.57×103 1.66×103 2.06×103 896.1709 2472.8 0
F4
Max (worst) 1.88×103 143.7546 1.81×103 1.83×103 2.18×103 10624 2657.4 0
Std. dev. 96.1158 29.4746 115.0177 55.7392 63.1357 65.0518 66.821 0
Min (best) 18.4168 0.0860 17.2205 17.5759 10.2026 9.7014 17.137 8.88×10-16
Mean 18.9445 0.1843 18.4381 18.2013 10.9454 11.4041 18.598 8.88×10-16
F5
Max (worst) 19.4393 0.3355 19.0665 18.6056 11.8548 12.9827 20.464 8.88×10-16
Std. dev. 0.2471 0.0639 0.3963 0.1885 0.3725 0.7696 1.1491 0
Min (best) -11743 -17852 -8871.9 -11514 -3205.6 -7705.4 -16931 -18013
Mean -10152 -17378 -7355.8 -10670 -2258.1 -4058.1 -15858 -17333
F6
Max (worst) -9073.1 -16613 -5964.5 -10146 -1736.1 -1778.5 -14444 -16182
Std. dev. 505.3396 226.7118 678.2347 305.3386 273.3307 1637.9 578.42 407.7910
Min (best) -957.4032 -636.7343 -668.7113 -527.6198 -489.0815 -789.6414 -112.03 -2000
Mean -665.9442 -472.1799 -506.2347 -423.7153 -360.1752 -874.9599 134.1 -2000
F8
Max (worst) -384.3000 -337.2687 -272.9559 -345.7854 -273.9183 -10054.21 386.44 -2000
Std. dev. 133.2749 68.1014 100.6667 44.3224 51.9544 50.7416 87.317 0
Min (best) 1.3415 4.04×10-05 1.0771 1.3743 1.1068 0.5447 1.2346 0
Mean 1.5124 2.30×10-04 1.1493 1.4342 1.1689 0.7379 1.3241 0
F10
Max (worst) 1.5947 6.02×10-04 1.2056 1.4971 1.2161 0.9790 1.4589 0
Std. dev. 0.522 1.29×10-04 0.0232 0.0287 0.0173 0.0935 0.0509 0

41
Table 7. Optimization results obtained from different algorithms for multimodal functions
ABC TLBO ICA BBO GSA SRSR
Succ. times 48 49 46 40 32 50
F11 Mean −1.02801 −1.03147 −1.0176 −1.0026 −0.9783 −1.03162
Ave. optima 1.94 1.98 1.88 1.76 1.68 2
Succ. times 40 43 37 35 30 47
F12 Mean 3.2546×10-6 4.2860×10-7 5.7585×10-6 5.9763×10-5 4.5213×10-3 4.6294×10-8
Ave. optima 3.55 3.63 3.49 3.22 2.67 3.79
Succ. times 32 34 29 21 18 40
F13 Mean −170.8438 −173.8538 −169.5672 −165.1547 −152.2435 −175.7126
Ave. optima 12.65 13.37 11.89 10.95 8.02 17.18
Succ. times 28 29 24 21 13 34
F14 Mean −178.1639 −179.2478 −175.9807 −171.2474 −163.8247 −181.5655
Ave. optima 13.58 13.62 12.89 11.43 6.07 16.18

42
Table 8. The data for 10 unit case study
Unit 1 Unit 2 Unit 3 Unit 4 Unit 5 Unit 6 Unit 7 Unit 8 Unit 9 Unit 10

Pigmax 455 455 130 130 162 80 85 55 55 55

Pigmin 150 150 20 20 25 20 25 10 10 10

ai 1000 970 700 680 450 370 480 660 665 670
bi 16.19 17.26 16.6 16.50 19.70 22.26 27.74 25.92 27.27 27.79
2
ci  10 0.048 0.031 0.200 0.211 0.398 0.712 0.079 0. 413 0.222 0.173
MDTi 8 8 5 5 6 3 3 1 1 1
MUTi 8 8 5 5 6 3 3 1 1 1
HSCi 4500 5000 550 560 900 170 260 30 30 30
CSCi 9000 10000 1100 1120 1800 340 520 60 60 60
CSH i 5 5 4 4 4 2 2 0 0 0
Ini. Si 8 8 -5 -5 -6 -3 -3 -2 -1 -1

Table 9. Hourly forecasted load and power price in power market

Hour [h] 1 2 3 4 5 6 7 8 9 10 11 12
Load [MW] 700 750 850 950 1000 1100 1150 1200 1300 1400 1450 1500
Hour [h] 13 14 15 16 17 18 19 20 21 22 23 24
Load [MW] 1400 1300 1200 1050 1000 1100 1200 1400 1300 1100 900 800

43
Table 10. Power dispatch for 10 unit system using SRSR
Time Load U1 U2 U3 U4 U5 U6 U7 U8 U9 U10 Reserve
[h] [MW] [MW] [MW] [MW] [MW] [MW] [MW] [MW] [MW] [MW] [MW] [% Load]
1 700 455 245 0 0 0 0 0 0 0 0 30.00
2 750 455 295 0 0 0 0 0 0 0 0 21.33
3 850 455 370 0 0 25 0 0 0 0 0 26.12
4 950 455 455 0 0 40 0 0 0 0 0 12.84
5 1000 455 390 0 130 25 0 0 0 0 0 20.20
6 1100 455 360 130 130 25 0 0 0 0 0 21.09
7 1150 455 410 130 130 25 0 0 0 0 0 15.83
8 1200 455 455 130 130 30 0 0 0 0 0 11.00
9 1300 455 455 130 130 85 20 25 0 0 0 15.15
10 1400 455 455 130 130 162 32 26 10 0 0 10.86
11 1450 455 455 130 130 162 44 54 10 10 0 10.83
12 1500 455 455 130 130 162 53 85 10 10 10 10.80
13 1400 455 455 130 130 162 32 26 10 0 0 10.86
14 1300 455 455 130 130 85 20 25 0 0 0 15.15
15 1200 455 455 130 130 30 0 0 0 0 0 11.00
16 1050 455 310 130 130 25 0 0 0 0 0 26.86
17 1000 455 260 130 130 25 0 0 0 0 0 33.20
18 1100 455 360 130 130 25 0 0 0 0 0 21.09
19 1200 455 455 130 130 30 0 0 0 0 0 11.00
20 1400 455 455 130 130 162 32 26 10 0 0 10.86
21 1300 455 455 130 130 85 20 25 0 0 0 15.15
22 1100 455 455 0 0 145 20 25 0 0 0 12.45
23 900 455 425 0 0 0 20 0 0 0 0 10.00
24 700 455 345 0 0 0 0 0 0 0 0 13.75

44
 Table 11. Commitment status for 20 unit system using SRSR
Time Reserve Time Reserve
ON Units ON Units
[h] [% Load] [h] [% Load]
1 1-4 30.00 13 1-16 10.86
2 1-4 21.33 14 1-13 11.88
3 1-4, 10 16.59 15 1-10 11.00
4 1-4, 9, 10 12.84 16 1-4, 6-10 20.67
5 1-4, 7, 9, 10 13.70 17 1-4, 6-10 26.70
6 1-5, 7-10 15.18 18 1-4, 6-10 15.18
7 1-5, 7-10 10.17 19 1-4, 6-10, 12, 14 12.46
8 1-10 11.00 20 1-4, 6-18 10.14
9 1-13 11.88 21 1-4, 6-14 10.15
10 1-16 10.86 22 1-4, 8-11, 13 10.86
11 1-18 10.83 23 1-4, 10 10.11
12 1-20 10.80 24 1-4 13.75

45
 Table 12. Commitment status for 40 unit system using SRSR
Time Reserve Time Reserve
ON Units ON Units
[h] [% Load] [h] [% Load]
1 1-8 30.00 13 1-32 10.86
2 1-8 21.33 14 1-24, 28 10.25
3 1-8, 20 11.82 15 1-8, 11, 13-24, 28 11.31
4 1-8, 15, 17, 18, 20 12.00 16 1-8, 11, 13-20 17.57
5 1-8, 15-20 13.70 17 1-8, 11, 13-20 23.45
6 1-8, 12-20 12.23 18 1-8, 11, 13-20 12.23
7 1-8, 10, 12-20 10.17 19 1-8, 11, 13-24, 27 11.31
8 1-20 11.00 20 1-24, 27, 29-36 10.23
9 1-24, 27 10.25 21 1-24, 27 10.25
10 1-32 10.86 22 1-10, 12, 14, 16, 18-20, 24 10.36
11 1-36 10.83 23 1-10, 12 11.94
12 1-40 10.80 24 1-7, 9, 10, 12 11.72

46
 Table 13. Commitment status for 60 unit system using SRSR
Time Reserve Time Reserve
ON Units ON Units
[h] [% Load] [h] [% Load]
1 1-12 30.00 13 1-43, 45-48 10.20
2 1-12 21.33 14 1-16, 18-36, 38, 39, 41 10.22
3 1-12, 29 10.24 15 1-12, 15, 16, 19-36 10.44
4 1-12, 25-29 10.00 16 1-12, 15, 16, 19-30 18.60
5 1-12, 21, 23-30 13.70 17 1-12, 15, 16, 19-30 24.53
6 1-12, 17, 19-30 11.24 18 1-12, 15, 16, 19-30 13.21
7 1-12, 14, 17-30 10.17 19 1-12, 15-17, 19-31, 35, 39, 41 10.17
8 1-30 11.00 20 1-36, 39, 41, 43-52, 54 10.08
9 1-37, 41 10.79 21 1-36, 39, 41 10.79
10 1-46, 48 10.20 22 1-14, 17, 18, 20, 22-24, 26, 28, 29, 32-34, 36 10.70
11 1-52, 54 10.20 23 1-14, 17, 18 10.74
12 1-59 10.19 24 1-7, 9-14, 18 12.40

47
 Table 14. Commitment status for 80 unit system using SRSR
Time Reserve Time Reserve
ON Units ON Units
[h] [% Load] [h] [% Load]
1 1-16 30.00 13 1-58, 60-64 10.37
2 1-16 21.33 14 1-16, 18, 20-48, 52-56 10.20
3 1-16, 35, 38 11.82 15 1-16, 20, 21, 23, 25-45, 47, 48 10.06
4 1-16, 29, 34-36, 38-40 10.29 16 1-16, 20, 21, 23, 25-40 19.12
5 1-16, 25, 26, 28, 29, 33-40 13.70 17 1-16, 20, 21, 23, 25-40 25.08
6 1-16, 23, 25-40 10.75 18 1-16, 20, 21, 23, 25-40 13.70
7 1-17, 22-40 10.17 19 1-17, 19-21, 23, 25-40, 47, 48, 51, 53 10.38
8 1-40 11.00 20 1-48, 51, 53, 57-72 10.23
9 1-48, 50, 54 10.25 21 1-48, 51, 53 10.25
10 1-63 10.37 22 1-19, 22, 24-26, 30-33, 36, 37, 39, 41-46 10.32
11 1-66, 68-72 10.35 23 1-19, 22, 24 10.14
12 1-74, 76-80 10.34 24 1-9, 11-16, 18, 22, 24 12.73

48
 Table 15. Commitment status for 100 unit system using SRSR
Time Reserve Time Reserve
ON Units ON Units
[h] [% Load] [h] [% Load]
1 1-20 30.00 13 1-70, 72, 74-80 10.07
2 1-20 21.33 14 1-52, 55-60, 62, 66, 68, 70 10.00
3 1-20, 44, 46 10.87 15 1-22, 26-28, 31-52, 55-57, 59, 60 10.25
4 1-20, 35, 37, 41, 44, 46-50 10.46 16 1-22, 26-28, 31-50 20.67
5 1-20, 32, 35-38, 41-50 13.70 17 1-22, 26-28, 31-50 26.70
6 1-21, 31-50 10.45 18 1-22, 26-28, 31-50 15.18
7 1-22, 24, 29-50 10.17 19 1-22, 26-28, 31-50, 52-55, 62, 63, 68 10.37
8 1-50 11.00 20 1-60, 62, 63, 68, 71-86, 88-90 10.14
9 1-53, 55, 57-60, 65, 66, 68, 70 10.00 21 1-60, 62, 63, 68 10.58
10 1-76, 79, 80 10.07 22 1-20, 23-25, 29-33, 36-38, 40-44, 46, 50, 51, 56-60 10.11
11 1-83, 85-89 10.07 23 1-20, 23-25, 29, 30, 41 10.13
12 1-92, 94-98, 100 10.07 24 1-12, 14-18, 20, 23-25, 29, 30 10.50

49
Table 16. Hourly operation cost of 10, 20, 40, 60, 80 and 100 unit systems
Time Number of generating units
[h] 10 20 40 60 80 100
1 13683.13 27366.71 54733.45 82100.23 109467.03 136833.86
2 14554.50 29109.33 58218.79 87328.40 116438.15 145547.50
3 17709.45 34011.32 66615.07 99218.83 133230.42 165834.15
4 18597.67 38096.56 76680.05 114745.19 152316.56 190875.63
5 20580.02 40017.23 80374.47 120951.70 161308.93 201666.17
6 23487.04 46378.28 91066.97 135905.90 180743.83 225582.95
7 23261.98 46009.27 93119.81 140229.34 187339.66 234449.45
8 24150.34 49400.88 98802.56 148204.46 197605.55 247007.43
9 27986.06 54914.53 108771.65 163687.05 217544.78 272019.09
10 29992.61 60505.22 121530.44 181283.95 242309.18 302222.69
11 31842.63 63685.31 127370.62 190346.28 254031.62 317007.28
12 33705.61 67411.34 134823.19 201516.21 268927.89 335621.13
13 29932.61 59865.22 119730.44 178903.95 238769.18 297942.69
14 27126.06 53714.53 106891.65 160926.34 214964.13 267219.09
15 24150.34 48300.88 98012.22 146328.39 194774.02 243043.57
16 21513.66 42407.15 84198.06 126603.94 169010.71 212036.98
17 20641.82 40659.26 80698.79 121357.68 162016.74 203297.19
18 22387.04 44158.28 87726.97 131860.95 176019.24 220791.42
19 24150.34 49415.52 98952.22 147412.07 196253.02 244890.84
20 30422.61 62028.61 123262.92 184218.94 245895.97 307272.21
21 27126.06 54044.05 106891.65 160607.05 213784.78 267499.34
22 22610.52 44456.61 87762.31 132272.91 176276.93 219994.65
23 17645.36 34862.82 70560.61 105531.35 140502.59 175362.51
24 15427.42 30854.93 62637.38 93492.18 124347.33 155510.67
Total 562684.87 1121673.83 2239432.31 3355033.29 4473878.23 5589528.49

50
Table 17. Comparison of operation cost obtained by SRSR and other algorithms for different case studies
Total operation cost of generating units
No. of units
LR [47] ICGA [44] DE [48] BF [49] ICA [46] SFL [50] GMTLBO–BH [51] SRSR
10 565,825 566,404 566,070 564,842 564,405 564,769 563,937 562,684
20 1,130,660 1,127,244 1,126,121 1,124,892 1,124,274 1,123,261 1,123,297 1,121,673
40 2,258,503 2,254,123 2,255,077 2,246,223 2,247,078 2,246,005 2,245,601 2,239,432
60 3,394,066 3,378,108 - 3,369,237 3,379,120 3,368,257 - 3,355,033
80 4,526,022 4,498,943 - 4,491,287 4,508,762 4,503,928 - 4,473,878
100 5,657,277 5,630,838 - 5,611,514 5,617,913 5,624,526 5,611,105 5,589,528

51