The Study On Dynamic Multi-objective Evolutionary Algorithms Based On Prediction

In the real world,there are a class of problems with multiple objectives which change over time.These problems are called dynamic multi-objective optimization problems(DMOPs),such as traffic scheduling and wireless network design.Because many of these problems are highly complex and nonlinear,they are very difficult to be solved by conventional mathematical methods.However,the evolutionary algorithms can behave well at this time.Therefore,these evolutionary algorithms for solving DMOPs are called dynamic multiobjective evolutionary algorithms.Evolutionary algorithms are a kind of algorithms that simulate the biological evolution with the character of survival of the fittest in the nature.The research to dynamic multi-objective evolutionary algorithms is still in the preliminary stage.Among them,the response strategy for dynamic process is an important aspect of the research.Among them,the prediction is a hot research direction.The method based on prediction is to find a certain rule by observing the optimal solutions in some environments,by using some learning methods,such as the autoregressive learning method,to predict the optimal solutions of the next environmental change.The purpose is to make the population converge rapidly.Increasing the accuracy of prediction is an important aspect of today's research.Based on the research of dynamic multi-objective optimization at home and abroad,this paper proposes a prediction strategy based on the center points and knee points(CKPS).In addition,a diversity preserving mechanism which can adaptively introduce random individuals according to the difficulty of the problem is proposed.The mechanism can solve DMOPs well,especially for the more complex problems.In addition,this paper proposes a prediction strategy based on special points,which is on the basis of CKPS,and introduces some special points of other characteristics,such as ideal points and boundary points.These special points are actually key points,and they are also the embodiment of some preference information.In addition,the adaptive diversity maintenance strategy is improved,and the number of random individuals is still generated according to the original strategy.But here,we predict the boundaries of random individuals,which are generated in a way similar to difference evolution within this boundary range.This makes diversity maintenance strategy more intelligent.Compared with the other four algorithms on some classical test problems,the experimental results show that SPPS is effective in solving DMOPs.