Evolutionary Dynamic Multi-objective Optimization Algorithm Based On Hybrid Prediction Strategy

There are many dynamic multi-objective optimization problems(DMOPs)in industrial production and scientific research.DMOPs is a type of problems that objectives change over time and conflict with each other.This characteristic that change over time brings challenge for researchers,because of the algorithm is required not only to find the optimal solution,but also can react to the changed environment.As a heuristic research algorithm,evolutionary algorithm can optimization many individuals simultaneously,and has been applied to solve dynamic multi-objective optimization problems.With this in mind,in this paper,we review the state of the arts about evolutionary dynamic multi-objective optimization.Aiming at the difficulties in DMOPs,we design the new prediction strategy,and propose three dynamic multi-objective optimization algorithms.The main work of this paper is presented as follows: 1.A hybrid prediction strategy based on special points(SHPS)is proposed.Firstly,PPS doesn't have a valid prediction strategy when the restored information is not enough,we utilize PRE&VAR in early evolution,which is suitable for little history information and just depend previous one or two times information.Secondly,PPS predict the whole population using the history information of population center and manifold,once the history information is inaccurate,the predict population may diverge the possible search range,in this work;we introduce special points prediction by combining PPS.Then the proposed SHPS are applied to MOEA/D-DE to solve DMOPs.Experimental results show that,the proposed algorithm can react to environment change quickly,and the convergence and diversity gotten by the algorithm are better than five existed dynamic multi-objective optimization algorithms.2.A hybrid prediction strategy based on diversity introduction(DHPS)is proposed.Firstly,the algorithm in chapter 3 is based on the special points which are still the points in population,and their impact on whole population diverge is limited,moreover computing and predicting these special points needs taking cost.So in the proposed DHPS,some diversity solutions generated by different ways are combined solutions by using PPS to overcome the shortage in work(1).By combing the proposed DHPS with a steady change detection method,a new diversity based hybrid dynamic multi-objective optimization is established.In order to test the performance of this algorithm,we use 8 test instance to test,and two set of experiment are conducted.The experimental results show that our algorithm has higher accuracy and can react to the environment change quickly.3.A hybrid prediction strategy based on memory mechanism is proposed.For the periodic problems,the solutions in new environment may come back to the position in previous environment,so using the optimal individuals found in previous environment may accelerate convergence.To utilize the optimal individuals in previous environment,a memory pool is used in our algorithm,when change is detected,each weight value find the best individual form the memory pool,and compare the best individual in memory pool and the current individual,if the individual in memory pool is better than current individual,then the current individual is replaced by the individual in memory pool.The experiment result show that our algorithm is better than other algorithms.