Differential Evolution Algorithm Based on Adaptive Bi-Coordinate Systems for Mixed-Variable Optimization Problem
How to design algorithms for solving Mixed-variable Optimization Problems(MVOPs)is an important research direction in the field of computer algorithm design and analy-sis.The difficulty of this problem lies in the need to optimize both continuous and discrete deci-sion variables,and Evolutionary Algorithms(EAs)are an effective means to solve this problem.However,there exists an issue that the correlation between problem variables is ignored,which deteriorates the algorithm performance to some extent.Therefore,in this work,a differential e-volution algorithm based on adaptive bi-coordinate systems is proposed to solve MVOPs from the perspective of variable correlation.Firstly,the covariance matrix information of the population is used to establish the eigen coordinate system,so that the relevant operations of the algorithm can be performed in the eigen coordinate system to relax the correlation between continuous and dis-crete variables;secondly,to avoid the loss of population diversity,the original coordinate system is retained,while an adaptive strategy is designed to switch the eigen coordinate system and the original coordinate system for taking full advantages of the two coordinate systems;finally,to better optimize the discrete variables,a local search strategy based on the correlation between discrete variables is specially designed,which is helpful for the overall performance of the algo-rithm.To verify our approach,extensive experiments are conducted on a wildly used test suite containing 28 test functions,and the comparative results with five well-known EAs of solving MVOPs confirm that the proposed approach owns better performance.Besides,on a real-world MVOP,i.e.,the welded beam design problem,and our approach is able to obtain the ever best-known solution to the problem.Our approach provides a new idea for the related work in the field of computer algorithm design and analysis.
mixed-variable optimizationdifferential evolutionvariable correlationeigen coor-dinate systembi-coordinate systems
万方数据
维普
