Research On Multimodal Optimization Algorithm Based On Multi-objective Method

In reality,multimodal optimization problems(MMOPs)with multiple optimal solutions are commonly seen,such as electromagnetic design,protein detection,pedestrian detection and path optimization.However,the most existing multimodal optimization algorithms are sensitive to niching parameters.To avoid the above problem,multi-objective method is applied to solve multimodal optimization problems.In this thesis,we focus on the multimodal optimization algorithm based on multi-objective method.The main contents are summarized as follows:Firstly,a multimodal optimization algorithm based on the decomposition-based multi-objective method(MMO-MOEA/D)is proposed in the thesis.The original multimodal optimization problem is first transferred into the multi-objective optimization problem(MOP)with two conflicting objectives by the information of the first decision variable and fitness value.The decomposition-based multi-objective evolutionary algorithm(MOEA/D)is applied to optimize the transformed MOP.A set of benchmark functions are employed to evaluate the proposed algorithm.Compared with a variety of classical multimodal algorithms,experimental results of the proposed algorithm demonstrate its effectiveness for MMOPs.Secondly,a multimodal optimization algorithm based on improved multi-objective method based on decomposition and dominance(MMO-IMOEA/DD)is proposed for the incomplete use of decision variable information during the transformation so that a biobjective function is extended to several bi-objective functions and the original Pareto dominance relation is updated.In addition,a novel neighbor updating strategy is designed to balance the diversity and convergence of the population during the evolutionary procedure.Moreover,the multi-objective evolutionary algorithm framework based on decomposition and dominance combined with the updated Pareto dominance relation and the novel neighbor updating strategy is adopted to search for the optimal solutions in the whole space.Experimental results demonstrate the proposed algorithm can solve MMOPs.