Вы находитесь на странице: 1из 11

International Journal of Mechanical Engineering OF MECHANICAL 0976 6340(Print), INTERNATIONAL JOURNAL and Technology (IJMET), ISSN ENGINEERING ISSN

N 0976 6359(Online) Volume 3, Issue 3, Sep- Dec (2012) IAEME AND TECHNOLOGY (IJMET)

ISSN 0976 6340 (Print) ISSN 0976 6359 (Online) Volume 3, Issue 3, September - December (2012), pp. 627-637 IAEME: www.iaeme.com/ijmet.asp Journal Impact Factor (2012): 3.8071 (Calculated by GISI) www.jifactor.com

IJMET
IAEME

AN INVASIVE WEED OPTIMIZATION (IWO) APPROACH FOR MULTI-OBJECTIVE JOB SHOP SCHEDULING PROBLEMS (JSSPs)
Hymavathi Madivada1, C.S.P. Rao2
1

(Research Scholar, Department of Mechanical Engineering, National Institute of Technology Warangal, Warangal 506004, India, hyma.madivada07@gmail.com) 2 (Professor, Department of Mechanical Engineering, National Institute of Technology Warangal, Warangal 506004, India, csp_rao@rediffmail.com, csp_rao63@yahoo.com)

ABSTRACT In this paper, a new meta-heuristic solution approach for Multi-objective Job Shop Scheduling Problems (JSSP) is presented. The proposed algorithm makes use of Mehrabian & Lucass heuristic Invasive Weed Optimization (IWO) in generating optimal schedules. For performance evaluation of solutions in a Multi-objective scenario, a concept called Fuzzy dominance has been employed. The results obtained from our study have shown that the proposed algorithm can be used as a new alternative solution technique for finding good solutions to the complex Multi-objective Job Shop Scheduling problems. Keywords:Invasive Weed Optimization, Job Shop scheduling, Metahueristics, MultiObjective Optimization 1. INTRODUCTION 1.1. Job Shop Scheduling Scheduling may be viewed as an optimization process where limited resources are allocated over time among both parallel and sequential activities.It comes into picture when certain activities are to be carried out with limited resources. In a Job-shop Scheduling Problem (JSSP) activities refer to the operations on jobs and resources refer to machine hours. It is a NP-hard problem. In many cases, the combination of goals and resources exponentially increases the search space, and thus the generation of good schedule is difficult because we have a very large combinatorial search space. Several problems in various industrial environments are combinatorial. This is the case for numerous scheduling and planning problems. Generally, it is extremely difficult to solve this type of problems in their general form, as it comprises several concurrent goals and several resources which must be
allocated to lead to our goals, which are to maximize the utilization of individuals and/or machines and minimize the time required to complete the entire process being scheduled. Therefore, the exact 627

International Journal of Mechanical Engineering and Technology (IJMET), ISSN 0976 6340(Print), ISSN 0976 6359(Online) Volume 3, Issue 3, Sep- Dec (2012) IAEME methods such as the branch and bound method, dynamic programming and constraint logic programming need a lot of time to find an optimal solution. So, we expect to find not necessarily an optimal solution, but a good one to solve the problem. Realistically, we are satisfied by obtaining a good solution near the optimal one. New search techniques such as genetic algorithms, simulated annealing or tabu search[2] are able to lead to our objective i.e. to find near-optimal solutions for a wide range of combinatorial optimization problems. The task of production scheduling consists in the temporal planning of the processing of a given set of orders. The processing of an order corresponds to the production of a particular product. It is accomplished by the execution of a set of operations in a predefined sequence on certain resources, subject to several constraints. The result of scheduling is a schedule showing the temporal assignment of operations of orders to the resources to be used. Each operation can be performed by some machines with different processing times. The difficulty is to find a good assignment of an operation to a machine in order to obtain a schedule which minimizes the total elapsed time (makespan). 1.2. Multi-Objective Job Shop Scheduling Problem Multi Objective Job Shop Scheduling Problem deals with sequencing the operations so that given set of objectives are achieved. One cannot identify a single solution that simultaneously optimizes each objective. While searching for solutions, one reaches points such that, when attempting to improve an objective further, other objectives suffer as a result. A tentative solution is called non-dominated, Pareto optimal, or Pareto efficient if it cannot be eliminated from consideration by replacing it with another solution which improves an objective without worsening another one. Finding such non-dominated solutions, and quantifying the trade-offs in satisfying the different objectives, is the goal when setting up and solving a multi objective optimization problem. In the present work an attempt has been made to find the solution to Multi Objective Job Shop Scheduling the objectives considered being Minimizing Make-span, make-span being the maximum completion time of all jobs or the time taken to complete the last job on the last machine in the schedule. Minimizing Tardiness, tardiness being the lateness of the job if it fails to meet its due-date, and zero otherwise. Minimizing Mean Flow-time which measures the average response of the schedule to individual demands of jobs for service. 2. A REVIEW OF INVASIVE WEED OPTIMIZATION (IWO) Invasive weed optimization (lWO), first designed and developed by Mehrabian and Lucas, is a relatively novel numerical stochastic optimization algorithm inspired from colonization of invasive weeds. The algorithm is simple but has shown to be effective in converging to optimal solution by employing basic properties, e.g. seeding, growth and competition, in a weed colony. A weed is any plant growing where it is not wanted; any tree, vine, shrub or herb may qualify as a weed, in any specified geographical area, depending on the situation. Weeds have shown a very robust and adaptive nature that renders them undesirable plants in agriculture. In D-dimensional search space, a weed which represents a potential solution of the objective function is represented by W = (w1, w2, , wm) . Firstly, M weeds, called a population of plants, are initialized with random growth position, and then each weed produces seeds depending on its fitness and the colony's lowest fitness and highest fitness to simulate the natural survival of the fittest process. The number of seeds each plant produce increases linearly from minimum possible seed production to its maximum. The generated seeds are being distribution randomly in the search area by normal distribution with mean equal to zero and a variance parameter decreasing over the number of iteration. By setting the mean equal to zero, the seeds are distributed randomly such that they locate near to the parent plant and by decreasing the variance over time, the fitter plants are grouped together and inappropriate plants are eliminated over times. The whole process is depicted in Fig 1 shown below [1] . The model and simulation of the colonizing behavior of weeds in order as a novel optimization algorithm, with some basic properties of the colonization process is presented in the following steps given below. 628

International Journal of Mechanical Engineering and Technology (IJMET), ISSN 0976 6340(Print), ISSN 0976 6359(Online) Volume 3, Issue 3, Sep- Dec (2012) IAEME 2.1 Initialization A finite number of seeds are being dispread over the D-dimensional problem space with random positions (initializing a population)

Fig 1: Flowchart for Invasive Weed Optimization Algorithm


629

International Journal of Mechanical Engineering and Technology (IJMET), ISSN 0976 6340(Print), ISSN 0976 6359(Online) Volume 3, Issue 3, Sep- Dec (2012) IAEME

2.2 Reproduction Every seed grows to a flowering plant and produces seeds depending on their fitness (reproduction): a member of the colony of weeds is allowed to produce seeds depending on its own and the lowest and highest fitness values of the colony, where the number of seeds each plant produces increases linearly from a possible minimum to its maximum; in other words, a plant will produce seeds based on its fitness and the lowest and highest fitness values of the colony to ensure that the increase is linear. This step adds a significant property to the search algorithm. In evolutionary algorithms that are adopted to solve optimization problems, intuitively, the infeasible individuals are not allowed to be reproduced, and feasible individuals could be thought to be the ones with better fitness values than infeasible individuals, although it is possible that some of the infeasible individuals carry more useful information than feasible individuals during the evolution process; in the reproduction method, this chance is given to infeasible individuals to survive and reproduce similar to the mechanisms that occur in nature 2.3 Spatial Dispersion The produced seeds are being randomly dispread over the search area and grow to new plants (spatial dispersal): the generated seeds are being randomly distributed over the Ddimensional search space by normally distributed random numbers with a mean equal to zero but with a varying variance of Witer. Thus, seeds will be randomly distributed such that they abide near the parent plant. The SD of the random function is reduced from a previously defined initial value Wini to a final value Wfin in every step (generation). In simulations, a nonlinear alteration has shown satisfactory performance. = + ..(1)

Where itermax is the maximum number of iterations, Witer the SD at the present time step and n the nonlinear modulation index. This alteration ensures that the probability of dropping a seed in a distant area decreases nonlinearly at each time step, which results in grouping fitter plants and the elimination of inappropriate plants 2.4 Competitive-Exclusion This process continues until the maximum number of plants is attained by fast reproduction. At this stage, only the plants with higher fitness can survive and produce seeds, whereas others are eliminated (competitive exclusion). In this process, after the maximum number of weeds in a colony is reached, each weed is allowed to produce seeds, spread them over the search area, and find their position and rank together with their parents. Next, weeds with lower fitness values are eliminated in order to attain the maximum allowable population in a colony. The course continues until the maximum iterations are reached and hopefully the plant with the best fitness is the closest to the optimal solution. . It is worth mentioning that the IWO has some distinctive properties when compared with the traditional GA and other numerical search algorithms, such as reproduction, spatial dispersal and competitive exclusion. In addition, no genetic operators are employed in the proposed algorithm, which makes it more dissimilar from the GA. In this way the algorithm mimics the ecological process of weed colonization and has been proved to be a powerful tool for finding out competitive solutions. It is capable of solving general multi-dimensional, linear and non-linear optimization problems with appreciable efficiency. It has been shown to be statistically significant than some state of art existing evolutionary algorithms.
630

International Journal of Mechanical Engineering and Technology (IJMET), ISSN 0976 6340(Print), ISSN 0976 6359(Online) Volume 3, Issue 3, Sep- Dec (2012) IAEME

3. IMPLEMENTATION OF IWOFOR SOLVING MULTI-OBJECTIVE JSSP 3.1 Weed representation of JSSP solution The solution to JSSP is a schedule of operation for jobs. In the present work IWO is used to find the optimum schedule[2]. A weed represents feasible schedule in this case .This is similar to a chromosome representing feasible schedule in case of genetic algorithm. In the present work a direct approach Operation based representation is employed. Following is the brief description of the representation. The schedule is represented in the form of a string as shown in figure 3.1. The representation encodes a schedule as a sequence of operations, and each character of the string stands for one operation. All operations for a job are represented by the same symbol, the job number and they are interpreted according to the order of their occurrence in the sequence in which they appear in the solution string. In this case the string is called weed and the algorithm IWO is used to evolve these weeds to discover potential schedules. For example, for a three-job-three-machine problem the representation would be as shown below. 3 2 1 2 1 3 3 1 2 3.2 Inputs for the problem For solving JSSP using IWO, the following inputs are required: Number of jobs Number of Machines Machine order for all the jobs Processing Times of all the operations Due dates for all the jobs Number of iterations to be carried out Maximum population allowable Initial and Final standard deviations of seeds from parent weed Nonlinear modulation index for finding out standard deviation in each iteration. 3.3 Step-wise implementation of IWO for JSSP This section gives a detailed explanation of implementation of the algorithm. The actual algorithm has been slightly modified for enhancing the performance. INVASIVE WEED OPTIMIZATION adopts the colonization process of weeds for finding optimum solutions efficiently. In this case weed is a potential solution and the algorithm helps the weeds to evolve and generate better population thus giving rise to fitter weeds which represent competitive schedules. In the present work the algorithm has been programmed using matlab. Several bench mark problems have been tested using make span and tardiness as objectives. 3.3.1 Initialization In the original algorithm initialization is random. n number of weeds are generated randomly. The modification is that initially 10 weeds i.e 10 different schedules are generated using dispatching rules which present a significant optimization capacity. Since the choice of the initial population has a high impact on the speed of evolution and the quality of final results, the solution scenario would be focused on generating initial population using priority dispatching rules. Priority dispatching rules are actually the most widely used for solving JSSP where all the operations available to be scheduled are assigned a priority .The operation
631

International Journal of Mechanical Engineering and Technology (IJMET), ISSN 0976 6340(Print), ISSN 0976 6359(Online) Volume 3, Issue 3, Sep- Dec (2012) IAEME

with the highest priority is chosen to be sequenced. A priority dispatching rule is a simple mathematical formula that, based on some processing parameters , specifies the priority of operations to be executed. 10 initial schedules i.e. weeds are generated using 10 commonly used priority dispatching rules given in the table below. EXPRESSION Shortest Processing Time (SPT) DESCRIPTION The job with shortest time on machines selected. pi pi+1 pi+2 pn Longest Processing Time(LPT) The job with longest time on machines selected.pi pi+1 pi+2pn Minimum Slack Time Per Time remaining until the due date Processing Operation(MINSOP) timeremaining Minimum Due Date(MINDD) The job with earliest due date is processed first Di Di+1 Di+2. Dn Critical Ratio(CR) Most work remaining (MWKR) Remaining due date/Remaining processing time Select the operation associated with the job of the most work remaining to be processed Least work remaining(LWKR) Select the operation associated with the job of the least work remaining to be processed Shortest remaining Minimum Min(processing tine remaining- minimum Processing Time(SRMPT) processing time) Longest remaining Maximum Max(processing tine remaining- maximum Processing Time(LRMPT) processing time) RANDOM(random selection) Select the next job to be processed randomly. Table 1: Priority Dispatching Rules 3.3.2 Performance Evaluation The initial population is made to reproduce depending on the fitness of the individual. The present problem is a Multi- Objective optimization , the objectives being minimizing make-span, minimizing tardiness and minimizing flow-time. So the fitness of the individual has to be evaluated considering all the three aspects.The individuals are ranked based on their fitness values(all the three). This ranking is done based on concept called Fuzzy-ParetoDominance.[3] The ranking scheme assigns dominance degrees to any set of vectors in a scale-independent, non-symmetric and set-dependent manner. Based on such a ranking scheme, the fitness values of a population can be replaced by the computed ranking values representing the dominating strength of an individual against all other individuals in the population. The three scheduling objectives used in this implementation are (1)make-span of the sequence (2) mean-flow time of the jobs, and (3)the mean tardiness of jobs. 3.3.2.1. Make-Span Module (Fitness 1) In Scheduling literature, make-span is defined as the maximum completion time of all jobs, or the time taken to complete the last job on the last machine in the schedule-assuming that the processing of the first job began at time 0. Make-span is denoted by cmax and computed as cmax =max {Fj} , where Fj is the flow time for job j(the total time taken by job j from the instant of its release to the shop to the time its processing by the last machine is over).
632

International Journal of Mechanical Engineering and Technology (IJMET), ISSN 0976 6340(Print), ISSN 0976 6359(Online) Volume 3, Issue 3, Sep- Dec (2012) IAEME

3.3.2.2. Mean Tardiness Module (Fitness 2) The lateness of a job measures the conformity of the schedule to that jobs due date. Lateness is defined as the amount of time by which the completion time of a job exceeds its due-date. Mathematically Lj=Cj-dj Lj: Lateness of job j Cj: Completion time of job j dj: Due date of job j. Tardiness of job is the lateness if it is positive else it is zero. Mean tardiness(T) is defined as the average of tardiness of all jobs. T = (sum of tardiness of all jobs) / number of jobs 3.3.2.3.Mean Flow Time Module(Fitness 3) The mean flow time measures the average response of the schedule to individual demands of jobs for service. Mathematically, mean flow time is the average of the flow times of all jobs. Mean flow time = sum of flow times of all jobs / number of jobs 3.3.2.4. Fuzzification Of Pareto Dominance And Ranking In multiobjective optimization, the optimization goal is given by more than one objective to be extreme. Formally, given a domain as subset of Rn, there are assigned m functions f1(x1, xn) . . . fm(x1. . . xn). Usually, there is not a single optimum but rather the so-called Pareto set of non-dominated solutions. Evolutionary Computation (EC) has been shown to be a powerful technique for multiobjective optimization (EMO - Evolutionary Multi-Objective Optimization). This biologically inspired methodology offers both flexibility in goal specification and good performance in multimodal, nonlinear search spaces. If we want to solve a highly complex multi-objective optimization problem, we might select one of the best ranked evolutionary approaches reviewed in the literature, like NSGA-II and hopefully start reaching good results quickly.[11] However, all these algorithms need dominated individuals in the population, to perform the corresponding genetic operators. For a higher number of objectives, this might become a problem, since the probability of having a dominated individual in the population will rapidly go to zero. The need for a revision of the Pareto dominance relation for also handling a larger number of objectives was already pointed out in a few studies, esp. given by Farina and Amato. There, we also find the suggestion to use fuzzy-membership degrees for the degree of a point belonging to the Pareto set (so called fuzzy optimality). Authors design their revised dominance measure in a way that the approach to the Pareto front can be registered more early in the search. The approach was shown to work successfully in the domain of more than two objectives. It came out that the use of fuzzy concepts is fruitful in this regard The fuzzification of Pareto dominance relation can be written then as follows: It is said that vector a dominates vector b by degree a with: a (a,b) = , It is said that the vector a is dominated by vector b by degree p with: p (a,b) =

633

International Journal of Mechanical Engineering and Technology (IJMET), ISSN 0976 6340(Print), ISSN 0976 6359(Online) Volume 3, Issue 3, Sep- Dec (2012) IAEME

Here i denote the dimension of vector, in the present case i is 3 as we have considered 3 objectives.The definitions differ in denominator and thus are not symmetric. Dominating by degree and being dominated by degree have different fuzzy values. For a Pareto dominating b, a (a,b) = 1 and p (b,a) = 1 but p (a,b) < 1 and a (b,a) < 1

We use these dominance degrees to rank a set M of multi-variate data(vectors).The vectors in this case are that sets of three fitness values of the schedules in the population.The vector here is (fitness1 ,fitness2 ,fitness3). Consider two weeds or schedules a and b p(a,b) = s/p. Where s = min (fitness1a, fitness1b) min (fitness2a, fitness2b) min (fitness3a, fitness3b) p = fitness1bfitness2b fitness3b Each element of M(set of schedules) is assigned the maximum degree of being dominated by any other element in M, and the elements of M are sorted in increasing order of the value of dominance degree. After sorting the population is ranked sequentally. So the individuals having good rank(lower) are fitter members in the population. With the help of ranking scheme performance evaluation is carried out. 3.3.3 Reproduction The number of seeds generated by the parent weed depends upon the parents fitness, the maximum fitness of the population and the minimum fitness. For all the weeds generated in the first phase number of seeds for each weed is found out using the following formula. N=max Max=max no of seeds a weed can have w =worst rank b =best rank, i =fitness of the weed considered The seed has similarities with the parent. The seed differs from the parent weed by a standard deviation the value of which reduces from Sin to Sfin as iterations progress thus ensuring convergence. At each iteration the value of standard deviation(s) is calculated as follows Sf = Sin =1 (assumption) Sfin=0.001 (assumption) f = ( Sin - Sfin)Sf S = f + Sfin iter = maximum number of iterations for each run n = non-linear index (in this case it is assumed to be 3) i = the iteration for which the standard deviation is being calculated Initially a 10o (o is no of operations) matrix of random numbers is generated. Each row of the matrix corresponds to a weed (schedule) generated in the first step. Let x be the matrix of random numbers and p be the matrix of corresponding schedules(each row of the matrix is a schedule coded in operation-based mode) generated using priority dispatching rules. From the initial set of schedules seeds are generated,the number of seeds depending on the rank of the parent weed.
634

International Journal of Mechanical Engineering and Technology (IJMET), ISSN 0976 6340(Print), ISSN 0976 6359(Online) Volume 3, Issue 3, Sep- Dec (2012) IAEME

3.3.4 Competitive Exclusion This process continues until the maximum number of plants is attained by fast reproduction. At this stage, only the plants with higher fitness can survive and produce seeds, whereas others are eliminated (competitive exclusion). In this process, after the maximum number of weeds in a colony is reached, each weed is allowed to produce seeds, spread them over the search area, and find their position and rank together with their parents. Next, weeds with lower fitness values are eliminated in order to attain the maximum allowable population in a colony. The course continues until the maximum iterations are reached and hopefully the plant with the best fitness is the closest to the optimal solution. . It is worth mentioning that the IWO has some distinctive properties when compared with the traditional GA and other numerical search algorithms, such as reproduction, spatial dispersal and competitive exclusion. In addition, no genetic operators are employed in the proposed algorithm, which makes it more dissimilar from the GA. The initial weeds generated and their seeds are together compared for their fitness.All are sorted in descending order of their fitness. Only the first m solutions are considered and are allowed to proceed to the next generation, m being the maximum allowable population which is an assumption depending upon the problem complexity. Maximum allowable population = 20no of operations (assumption in this case) 3.3.5 Stopping Criteria The stopping criteria is one of the following Maximum number of iterations The populations worst and best fitness becomes equal The cycle of steps explained is carried out until one of the stopping criteria is met 4. RESULTS AND DISCUSSIONS In this section, the execution of the program is presented by using an example problem. The problem is a 5 machine and 10 job problems [2]. The input to the program is as follows S.No Job1 Job2 Job3 Job4 Job5 Job6 Job7 Job8 Job9 Job10 Due date 37 74 111 148 185 222 259 296 333 370 Machine Order Matrix 1 4 3 1 1 3 2 1 5 2 5 4 2 3 Processing Time Matrix 13 16 7 18 15 8 15 17 6 13 19 19 13 16 10 19 20 18 9 9 9 7 17 18 13 7 18 5 20 16 6 14 19 19 17 9 10 20 7 12 7

5 2 1 3 19 2 5 4 1 19 4 5 2 3 14 2 5 4 3 8 2 5 4 1 16 4 3 1 5 14 2 4 5 3 8 4 3 2 1 16 1 3 5 4 12 Table 2: Input data for the problem

1-Best Make-span 2-Best Mean Flow Time


635

International Journal of Mechanical Engineering and Technology (IJMET), ISSN 0976 6340(Print), ISSN 0976 6359(Online) Volume 3, Issue 3, Sep- Dec (2012) IAEME

S.NO 1 2

165 143

IWO 184 128

169 136

190 153.2

NSGA-II 198 140.9

193 142.4

Table 3: Results obtained for the above problem using IWO

Comparison of results of IWO and NSGA


Mean Flow Time 160 150 140 130 120 160 170 180 Make Span 190 200

MOIWO NSGA

Fig2: Comparision of results of IWO and NSGA In view of the results obtained by implementing IWO to solve JSSP (MOIWO), it appears that IWO is efficient. Fuzzy dominance applied for decision making in Multiobjective scenario yielded promising results. The algorithm has been improved by changing its solution coding method and hybridizing with priority dispatching rules leading to fast convergence. The algorithms performance has been compared to that of standard Nondominated- Search- Genetic- Algorithm (NSGA-2) with the help of some bench-mark problems and has been found to be superior to the latter. 5. CONCLUSION In the present work the algorithm has been programmed for JSSP using mat-lab. Bench mark problems have been tested using make span and tardiness as objectives using IWO algorithm and the results are compared with the best known ones. It is concluded that the application of IWO to Multi Objective JSSP is a new area to be explored for competitive solutions REFERENCES Journal Papers [1] Siddharth Pal, Anniruddha Basak and Swagatam Das Linear Antenna Array Synthesis with Invasive Weed Optimization,International Conference of Soft Computing and Pattern recognition, 2009 [2] Edurado Fernandez, Edy Lopez, Segio Bernal Evolutionary multi-objective optimization using a Fuzzy-based Dominance Concept [3] Prithwish Chakraborty, Gaurab Ghosh Roy On Population Variance and Explorative Power of Invasive Weed Optimization World congress on nature and biologicaaly inspired computing, 2009
636

International Journal of Mechanical Engineering and Technology (IJMET), ISSN 0976 6340(Print), ISSN 0976 6359(Online) Volume 3, Issue 3, Sep- Dec (2012) IAEME

[4] Ritwik Giri ,Aritra Chowdhury, Arnob Ghosh A Modified Invasive Weed Optimization for training of feed forward neural networks [5] M.Ramezani Ghalenoei, H.Hajmirsadeghi, C.Lucas Discrete Invasive Weed Optimization and its application to Co-operative multi-task assignment of UAVs Comin prac. 48th IEEE conference on Decision and Control, Dec 2009,in Press [6] Rafal Zdonek, Tomasz Ignor UMTS base station location planning with Invasive Weed Optimization [7] Takeshi Yamada and Ryohei Nakano Genetic Algorithn for Job-shop scheduling problemProceedings of Modern Hueristic for Decision support, pp.March 1997, Pages 67-81 [8] Takeshi Yamada and Ryohei Nakano Job shop scheduling Job Shop Scheduling, pp.IEEE Control engineering Services 55,pages 134-160 [9] S.Q. Liu, H.L. Ong, K.M. Ng Applying Tabu search to Job Shop Scheduling Problem Annals of Operations research 41, 1993, Pages 231-252 Books [10] Tapan P. Bagehi Multi-objective Scheduling By genetic Algorithms Chapters in Books [11] Khald Mesghouni, Pierre Borne Evolutionary Algorith For Job shop scheduling Int.j.appl.Math.Comput.Sci.vol.14,2004,Pages 91-103 [12] Hossein Hajimirsadeghi, Amin Ghazanfari Co-operative co-evolutionary Invasive Weed Optimization and its application to Nash Equilibrium search in Electricity Markets

637

Вам также может понравиться