Research And Application Of Operator Adaptation Strategy In Evolutionary Algorith

In the field of scientific research and engineering application,multi-objective optimization problems are commonly existing.These problems consist of multiple objectives which conflict to each other.It is difficult to solve these problems by conventional methods in limited time.At present,evolutionary algorithm(EA)is widely used to quickly search a set of approximate optimal solutions.The genic operator is the search engine in an EA,and many researches have shown that it is difficult to deal with various types of problems via single genic operator.Therefore,combining a variety of genic operators to improve the generalization of the algorithm has gradually become the mainstream research direction in this field.In order to improve the ability of MOEAs in solving optimization problems with different features,this thesis explores the impact of different adaptive operator strategies on the overall performance of the algorithm from three perspectives: optimizing the existing adaptive operator EAs,proposing a new adaptive operator selection strategy,and then optimizing the engineering parameters in practice.The specific work and innovations are as follows:(1)For the existing MOEAs with AOS methods,in order to reduce their dependence on a single framework and enhance the cooperation between operators in the pool,a multi-objective hybrid operator evolutionary algorithm based on double credit assignment(DCA-MOEA/D)is proposed.The algorithm additionally introduces an external archive set and maintains with nondominated sorting criterion under the framework of MOEA/D.Two of different credit assignment strategies are adopted to allocate the credit value for each operator,which is used for operator selection hereafter.Based on the conventional DE,the concept of search inertia is introduced,and a novel differential evolution based on search inertia(Si DE)is proposed,which is added to the operator pool to improve the cooperation between different operators.Compared with other algorithms,the effectiveness of the proposed algorithm is verified according to the evaluation results of HV and IGD performance indexes on UF,WFG and DTLZ test suites.(2)In order to solve the fuzziness and contingency of traditional AOS strategy,an adaptive operator selection evolutionary algorithm based on classification tree(MOEA/D-CTAOS)is proposed.The algorithm introduces the classification tree model in the field of machine learning,and comprehensively considers the positional relationship between parents and the historical performance of each genic operator under such relationship to select the best one.The algorithm also adds Si DE to the operator pool to speed up the search efficiency in the decision space.Compared with MOEA/D variants and EAs based on other AOS methods,the performance indexes of IGD and HV on UF and LZ09 verify the effectiveness of the proposed algorithm.(3)In engineering applications,the calculation of objective function may often be very complex,and the usage of genic operator has a significant impact on the results.In order to verify the value of this research in specific engineering problems,the above two evolutionary algorithms are applied to the optimization of vehicle thermoelectric power generation parameters.Firstly,the decision variables and search domain of the problem are sorted out,and then the algorithms proposed in this thesis along with other MOEAs are utilized to solve it respectively.According to the HV index,a round of optimal solution result obtained by MOEA/D-CTAOS is selected,and three parameter setting schemes are chosen from the final population by using the knee point criterion.Compared with the benchmark scheme,within the search range of ± 10% of each decision variable,the output power and thermoelectric conversion efficiency of the three solutions are increased by 156.9% and 69.1% respectively in average,which further verifies the application value of this research in practical engineering problems.