A fitness landscape ruggedness multiobjective differential evolution algorithm with a reinforcement learning strategy

Abstract

Optimization is the process of finding and comparing feasible solutions and adopting the best one until no better solution can be found. Because solving real-world problems often involves simulations and multiobjective optimization, the results and solutions of these problems are conceptually different from those of single-objective problems. In single-objective optimization problems, the global optimal solution is the solution that yields the optimal value of the objective function. However, for multiobjective optimization problems, the optimal solutions are Pareto-optimal solutions produced by balancing multiple objective functions. The strategic variables calculated in multiobjective problems produce different effects on the mapping imbalance and the search redundancy in the search space. Therefore, this paper proposes a fitness landscape ruggedness multiobjective differential evolution (LRMODE) algorithm with a reinforcement learning strategy. The proposed algorithm analyses the ruggedness of landscapes using information entropy to estimate whether the local landscape has a unimodal or multimodal topology and then combines the outcome with a reinforcement learning strategy to determine the optimal probability distribution of the algorithm’s search strategy set. The experimental results show that this novel algorithm can ameliorate the problem of search redundancy and search-space mapping imbalances, effectively improving the convergence of the search algorithm during the optimization process.