Heuristic Information Based Multiobjective Evolutionary Algorithm And Its Applications

As common and fundamental problems in the industrial domain and daily life,the multiobjective optimization problems(MOP)attract more and more attention in recent years.Since the production efficiency can be greatly improved by solving relevant multiobjective optimization problems,the multiobjective evolutionary algorithms become a popular research topic due to its capability in efficiently solving the MOPs.When dealing with specific MOP,the MOEAs are provided with various heuristic information,and there are some common properties of this heuristic information.At the same time,the mapping features,structural features of the population,the dependency features between problems,the quantitative features,and the control and dependency properties of the decision variables are closely related to the performance of the MOEAs.In order to investigate the effect of the heuristic information contained in these features,in this paper,we proposed several heuristic informationbased MOEAs to tackle some MOPs of different features.These proposed algorithms are verified on the multiobjective deployment optimization of the near space communication system(NSCS).This paper is summarized as follows:(1)Regarding the continuous mapping features in the decision variable space of the obtained approximate Pareto optimal solutions,we considered the relationship among the neighboring subproblems and proposed a heuristic multiobjective evolutionary algorithm based on the local search of the Pareto front under the framework of the multiobjective evolutionary algorithm based on decomposition(MOEA/D).The proposed algorithm adaptively searches the difference regions of the neighboring individuals to further improve the obtained Pareto front.By optimizing the deployment of the NSCS,the proposed algorithm proves its effectiveness.(2)As independent heuristic information of specific MOPs,the structural features of the population during the iterations have a positive impact on improving the performance of the MOEAs.From the perspective of the decision variables,we proposed a similarity-based multiobjective evolutionary algorithm which is able to obtain the evaluations of the neighboring subproblems on the decision variable space by calculating their similarity.By analyzing the similarity and difference of their evaluations,the proposed algorithm is able to utilize the structural information among these individuals to effectively search the decision variable space globally and locally.The proposed algorithm is applied in solving a modified NSCS deployment optimization problem and proves its superiority.(3)In many real-world applications,some relevant MOPs require to be solved in a synergetic manner.The characteristic of these problems is that they consist of several similar but not the same MOPs.These MOPs are closely related by constraints and cannot be solved independently.To solve these problems,we proposed a heuristic cooperative multiobjective evolutionary algorithm based on memetic computing.The heuristic-based crossover,mutation and local search utilize the problem related information to exchange information between related problems and optimize these problems in a cooperative way.The proposed algorithm successfully proves its effectiveness by cooperatively solving a proposed periodic multi-phased NSCS deployment optimization problem.(4)From the perspective of individuals,there are similarity and difference between the decision variables of two individuals of the neighboring subproblems when solving MOPs.As for the entire population,the decision variable distributions differ from different subproblems.The distribution information of the decision variables not only can provide information of the regions of high fitness but also help generate robust individuals.In order to investigate the quantitative features of the decision variable distributions of various subproblems,we proposed a local incremental estimation of distribution algorithm with asymmetrical domination under the MOEA/D framework.The proposed algorithm estimates the decision variable distribution of the neighboring subproblems within several latest iterations and obtains a local incremental distribution model.By sampling this model,robust offspring can be generated.The proposed asymmetrical domination relationship is able to evaluate these solutions in the dynamic environment more accurately as well.By optimizing the deployment optimization problem of the dynamic NSCS,the proposed algorithm can obtain robust solutions in the dynamic environment and thus proves its effectiveness.(5)The MOEAs are usually faced with enormous decision variable space when dealing with MOPs with too many decision variables.The optimization of the MOPs with a large number of decision variables requires much more computation resource.A common strategy to deal with numerous decision variables is to analyze their control and dependency properties.However,this analysis task also consumes too much computation resource.To address this problem,we proposed a decision variable analysis algorithm based on distributed computing.The proposed distributed computing-based method divides the original analysis task into a set of smaller independent subtasks,and these subtasks are solved simultaneously on a distributed computing platform.In the experiment,the CPU time of the proposed distributed computing-based algorithm is compared with that of its sequential version,the proposed method is able to obtain results within a bearable time while the sequential algorithm cannot obtain these results.