Landscape-based adaptive operator selection mechanism for differential evolution

Evolutionary Framework With Reinforcement Learning-Based Mutation Adaptation

“…Despite the numerous DE variants that exist in the literature, multiple research papers have shown that a single operator may not work well for a wide range of optimization problems. Moreover, given the weakness of single-operator DE algorithms (i.e., the variants that used a single mutation operator) to solve all forms of test problems, research on multi-operator DE variants have recently gained much more interest [4]. Fan et al [37] introduced an auto-selection mechanism (ASM) to choose an appropriate DE variant to solve combinatorial optimization problems.…”

“…They also integrated a learning strategy and selection probability to reinforce the probability of a particular DE variant being chosen and their results for selective data sets demonstrated the efficacy of their ASM method. In a similar analysis conducted in [4], an adaptive operator selection (AOS) technique was proposed. This AOS used the information of the problem's landscape and performance histories to automatically select the most appropriate strategy from a set of many.…”

“…As with all other EAs, DE uses three main operators (mutation, crossover and selection) to guide a set of solutions to obtain acceptable results. DE has attracted the attention of many researchers and practitioners because of its ease of use, its simple structure, and its robustness [4], [5]. Covariance matrix adaptation evolution strategy (CMA-ES) [6] has also attracted the attention of researchers [4], [7].…”

“…DE has attracted the attention of many researchers and practitioners because of its ease of use, its simple structure, and its robustness [4], [5]. Covariance matrix adaptation evolution strategy (CMA-ES) [6] has also attracted the attention of researchers [4], [7]. Both of these algorithms have shown very good performance in solving the bound-constrained and real-application optimization problems [1], [2], [8].…”

“…Both of these algorithms have shown very good performance in solving the bound-constrained and real-application optimization problems [1], [2], [8]. However, there is no single operator (or parameter) and EA that is considered to be the best for all kinds of optimization test functions [2], [4], [9], this is in line with the no free lunch theory [10]. This has motivated and encouraged practitioners and researchers to propose algorithmic frameworks that utilize the benefits and strengths of various parameters, operators and algorithms.…”