Research On Dynamic Multi-objective Evolutionary Algorithms Based On Hybrid Strategy And Adaptive Strategy

Dynamic multi-objective optimization problems(DMOPs)are widely existed in industrial applications and scientific research.This kind of problems not only has multiple conflicting objectives,but also its number of objectives,objective functions or constraints may change over time.In DMOPs,different types of dynamic changes might cause the Pareto optimal front(PF)and/or Pareto optimal set(PS)of DMOPs to change over time,which makes the problem much more difficult.Therefore,dynamic multi-objective evolutionary algorithms(DMOEAs)should be able to track the movement of PF and PS efficiently over time.In recent years,evolutionary algorithms(EA)have been widely used to solve DMOPs,and the research of DMOEAs has made great progress.In general,the current development of dynamic multi-objective optimization field is still in its infancy,and there is a lot of work that remains to be done in dynamic multi-objective optimization field,including proposing practical dynamic multi-objective test problems,robust DMOEAs and standard performance indicators.This thesis is devoted to the research of DMOEAs.By combining with the changing characteristics of existing DMOPs,we analyzed the advantages and disadvantages of existing DMOEAs in solving DMOPs.We explored some effective dynamic handling strategies and proposed two improved DMOEAs.The specific researches are as follows:1)A hybrid of memory and prediction strategies(HMPS)is proposed.When a new change is detected,HMPS firstly identifies the similarity of the new change to the historical changes,based on which two different response strategies are applied.If the detected change is dissimilar to any historical changes,a differential prediction based on the previous two consecutive population centers is utilized to relocate the population individuals in the new environment;otherwise,a memory-based technique devised to predict the new locations of the population members is applied.HMPS makes full use of the advantages of memory-based technique and prediction-based technique to solve DMOPs.HMPS is compared with three state of the art DMOEAs on fourteen DMOPs with different characteristics.The experimental results show that HMPS outperforms the other compared algorithms on most test problems.2)An adaptive exploration and feedback strategy(AEFS)is proposed.When a new change is detected,based on some individuals in population at t-th time step,AEFS firstly adopts dynamic handling strategies to build an exploring population and reevaluates the individual of the exploring population in the new environment.Then,AEFS identifies the non-dominated individuals in the exploring population,whose positions represent the potential locations of the optimal solution set of the new environment,and counting the proportion of the non-dominated individuals in the exploring population.If the proportion is more than 50%,the prediction of the remaining population individuals is guided by these non-dominated individuals;Otherwise,the remaining population individuals not only need non-dominated individuals to guide the prediction,but also some randomly generated individuals should be introduced to increase the population diversity,which helps to improve the global search ability of population.Based on the environmental changes,AEFS can adaptively adjust the dynamic response strategy,which helps to improve the quality of change reaction.AEFS is compared with three state of the art DMOEAs on fourteen DMOPs with different characteristics.The experimental results show that AEFS outperforms the other compared algorithms on most test problems.