Step-by-step improvement of JADE and SHADE-based algorithms: Success or failure?

Abstract

Abstract For about two decades Differential Evolution (DE) algorithms belong to the most successful optimization metaheuristics. Among plentiful DE versions proposed so far those that are based on mutation strategies and control parameter adaptation methods introduced within JADE and SHADE variants show especially encouraging performance. However, many modifications of JADE or SHADE are developed simultaneously by various researchers, and are rarely compared or discussed against each other. As most of JADE or SHADE-based variants are tested on recently introduced sets of artificial benchmark functions with a single, pre-specified maximum number of function calls, it remains unclear how they would perform on real-world problems, or when the number of allowed function calls differs from the standard values. In this study in-deep insight into the performance of twenty two JADE/SHADE-based variants on a large sets of artificial benchmarks and real-world problems is presented. The impact of the pre-assumed maximum number of function calls and the algorithm population size on the results is verified and discussed. Finally, overall comparison of algorithms that originate from JADE or SHADE against other kinds of metaheuristics is presented. The main aim of the study is to point out these among recently introduced JADE or SHADE-based operators that turn out to be especially successful, and to determine conditions under which they either achieve desired results or fail.