Research On Evolutionary Algorithms For Multi-Solution Optimization Problems

There are a large number of optimization problems in the real world whose goal is to find the acceptable solution for the decision maker.Generally speaking,in some optimization problems the decision maker only needs a single acceptable solution to the problem,however,in many scenarios the decision maker wants to obtain multiple acceptable solutions,which provides more options for the decision maker to choose the final solution according to the preference and environment.Such problems for finding multiple acceptable solutions are called multi-solution optimization problems and are manifested as multimodal optimization problems with multiple global optima and local optima,and multiobjective optimization problems with multiple conflicting objectives.Evolutionary algorithms are commonly used to solve optimization problems,but the classical evolutionary algorithms are difficult to cope with complex multi-solution optimization problems.This dissertation conducts research on multi-solution optimization problems and intend to explore the design of algorithms on complex multi-solution optimization problems in terms of improving the performance for solving multi-solution optimization problems and constructing benchmarks for describing complex multisolution environments.The contributions and innovations of this dissertation are described as follows.(1)To address the difficulty of finding all acceptable solutions in solving multimodal optimization problems,the formulation,balance and keypoint-based differential evolution(FBK-DE)is proposed.FBK-DE focuses on the formulation,balance and keypoint of species,to find as many global optimal solutions as possible while ensuring the population convergence.Experimental results show that compared with other multimodal optimization algorithms,FBK-DE is able to find more global optimal solutions and performs competitive.(2)To address the difficulty of finding multiple acceptable solutions in dynamic environments when solving dynamic multimodal optimization problems,the dynamic multimodal clonal selection algorithm(DMMCSA)based on clonal selection is proposed.To cope with the dynamic changes of the problem,DMMCSA uses a memory set strategy to preserve multiple optimal solutions under different environments.The experimental results verify the superiority of DMMCSA.Meanwhile,the corresponding solution framework PopDMMO is proposed to explore the impact of different components in the optimization algorithm on the solution performance.(3)To address that the algorithms have not yet considered the dynamic and constrained environment in solving dynamic constrained multimodal optimization problems,a scalable algorithm platform called EvoDCMMO is proposed.The platform constructs a general framework for solving these problems and employs various methods in four classes of components:optimizers,nichingg methods,constraint handling techniques,and dynamic response strategies.Experimental results on these components show that the dynamic multimodal clone selection operators,nearest-better clustering,superiority of feasible solution and memory-based immigrants strategy perform the best in the corresponding components,respectively.(4)To address the difficulty of the optimization algorithm in coping with preference changes when solving multiobjective optimization problems,a dynamic preference change model is proposed and two optimization algorithms(g-dominance and archivebased nondominated sorting clonal selection algorithm,ga-NSCSA,and g-dominance and archive-based nondominated sorting genetic algortihm Ⅱ,ga-NSGA-Ⅱ)are designed.The experimental results show that ga-NSCSA performs better compared with the existing algorithms.In summary,this dissertation conducts research on the difficulty that algorithms are unable to obtain multiple high-quality acceptable solutions on complex multi-solution optimization problems,mainly by proposing new algorithms to improve the performance of searching for multiple acceptable solutions,comparing multiple methods to analyze performance differences,and proposing multi-solution optimization benchmarks with other features to simulate complex multi-solution environments.This work has achieved some research results,which provides reference value for further research of evolutionary algorithms in the field of multi-solution optimization.