Research On Differential Evolution Algorithm And Application Based On Turning Mutation Strategy

Differential evolution algorithm is an effective random search optimization algorithm.By inheriting and developing the advantages of existing optimization algorithms,it has the strengths of fewer control parameters,simple settings,fast convergence speed,and robust optimization results.However,for the more complex single-objective multi-peak optimization problem,the differential evolution algorithm tends to converge too early and stagnate in the local optimal solution.In response to these problems,a strategy based on turning mutation is proposed to improve three differential evolution algorithms.Moreover,the CEC2020 single objective boundary constrained numerical optimization benchmark set is used as the evaluation standard,and then the improved three differential evolution algorithms are applied to three practical engineering problems.First of all,the research status of differential evolution algorithm at home and abroad is discussed and the common improvement directions in view of the shortcomings of differential evolution algorithm are introduced.Then the thesis gives the basic process of standard DE algorithm and its variant SHADE algorithm,as well as the main characteristics of L-SHADE and j SO algorithm evolved from SHADE algorithm.Then a strategy based on turning mutation is proposed for the defect that the differential evolution algorithm tends to converge too early and stagnate in the local optimal solution in the complex single-objective multi-peak optimization problem,and then it is applied to SHADE,L-SHADE and j SO algorithm.The motivation of this strategy is to change the prospective mutation direction in certain circumstances,try to maintain the diversity of the population,and make the algorithm maintain a long global search stage,in order to avoid premature convergence,jump out of the local optimal solution,and then achieve better optimization results.In the thesis,the improved algorithms(Tb-SHADE,Tb LSHADE and Tb-j SO)are verified on 10 test functions of 5 dimensional,10 dimensional,15 dimensional and 20 dimensional benchmark set of single objective boundary constrained numerical optimization(CEC2020).On the whole,the improved algorithms achieve significantly better optimization results.It is also verified that the method can maintain the population diversity by measuring the population diversity and cluster analysis.Finally,so as to further verify the improved algorithms,they are added constraint processing operation and applied to the heat exchanger network design,multiple disk clutch brake design problem and step-cone pulley problem.And the experimental results indicate that the improved algorithms achieve significantly better optimization results.The experimental results indicate that,the differential evolution algorithms improved by using the turning mutation strategy have achieved significantly better results in the CEC2020 single-object boundary-constrained numerical optimization benchmark set and three practical engineering problems when compared with the original algorithms and some of the current better algorithms.It provides new ideas for solving some complex single-objective multi-peak optimization problems,and has important scientific research value for the theory and application of swarm intelligence algorithms,evolutionary algorithms,and optimization.