Improving Applicability of DE Algorithm by Designing Multi-Operator Variant of DE
Surendra Tripathi, K.K. Mishra and Shailesh Tiwari
Recent research on Differential Evolution (D.E.) algorithm involves creating a new variant of D.E. that improves the algorithm’s convergence rate and provides necessary diversity among solutions for performing an effective search in the search space. These variants are developed either by updating the equation of mutation or crossover strategies of classical D.E. algorithms or by performing parameter tuning. Availability of multiple mutations and crossover strategies and parameter tuning techniques promoted scientists for creating multi-operator versions of D.E. Recently a multi-operator approach and self-adaptation have been used in the IMODE algorithm. Although IMODE is faster than other D.E. variants, its performance can be improved further by using a new mutation and crossover strategy that can provide an effective search in the search space. We created a new multi-operator algorithm named MIMODE by defining new strategies for implementing crossover and mutation in D.E. These new strategies are then combined with existing mutation strategies for creating a multi-operator version. The proposed mutation strategy is providing more bandwidth covering more search space.
Additionally, the suggested crossover operator is designed to improve the exploiting capability of the existing crossover operator. These changes are made to improve the convergence rate of the MIMODE algorithm. To check the proposed algorithm’s performance, it is compared with state of art algorithms on the latest benchmark function taken from CEC 2021 on the PLATEMO framework. The latest multi-operator versions were also included in the comparisons. The result verifies that the proposed approach is good as compared to other latest D.E. variants.
Keywords: Differential evolution (D.E.), evolutionary algorithms (E.A.s), multi-operator version, mutation strategy