A Competitive Learning QUasi Affine TRansformation Evolutionary for Global Optimization and Its Application in CVRP

Abstract

In this paper, we propose a new Competitive Learning QUasi Affine TRansformation Evolutionary (CL-QUATRE) algorithm for Global Optimization and its application in Capacitated Vehicle Routing Problem (CVRP). In the proposed CL-QUATRE, the population is divided into two subpopulations (i.e., winner and loser) with a pair wise competition mechanism. Each subpopulation utilizes different mutation strategy to reserve the population diversity and improve convergence speed. The winner evolves with a mutation strategy “QUATRE/best/1”, whereas the loser evolves with a modified mutation strategy “QUATRE/target-to-best-win ner/1”, which learns from winner subpopulation to make the algorithm more efficient. Meanwhile, a scale factor updating method, called stochastic scale factor, is introduced into the proposed CL-QUATRE algorithm to jump out of the local optima and avoid falling into stagnation. With these modifications, the proposed algorithm can achieve good balance between exploration and exploitation capability. We compare the proposed algorithm with four QUATRE variants, four DE variants, and four PSO variants on CEC2013 test suite, CEC2014 test suite and two CVRP benchmarks. The experimental results demonstrate that the CL-QUATRE algorithm achieves better or competitive performance.