Improving Differential Evolution on the Generalized Numerical Benchmark Generator Problem Set by an Ensemble Heuristic Algorithm

Robust optimization strategies (effective and feasible viewpoints) are critical in solving complex, high-dimensional problems in science, where traditional algorithms frequently converge prematurely to suboptimal solutions, and sometimes so far to the global optimum. This paper proposes an Ensemble Heuristic Algorithm to improve the performance of Differential Evolution, specifically tested on the Generalized Numerical Benchmark Generator (GNBG) problem test bed. In this research, the effectiveness of the Differential Evolution approach is examined in several stages of the heuristic optimization process. That is to say, it is an ensemble of the Salp Swarm Algorithm (SSA), Multi-Verse Optimizer (MVO), and a hybrid Differential Evolution, into a ensemble framework in a unified way. The proposed algorithm is tested on 24 problem instances from the GNBG set. This method is superior to standard DE, scoring 12 out of 24 on the success metric, compared to 16.19 out of 24 on the success rate metric. The proposed method shows better convergence and solution quality (non-suboptimal solutions) across the whole benchmark set. An initial exploration phase that uses several metaheuristics, like SSA and MVO, is one of the most important parts of this method. These algorithms generate a diverse quantity of candidate solutions. Following this, a stage called “solution combination stage” uses a hybrid Differential Evolution algorithm and improves the overall search capability (exploration). Finally, we continue with a phase called “intensive local search phase”, this part has adaptive step sizes, which it is tuned in order to get the best solutions and improve convergence to optimal results. Consequently, it is emphasized that the algorithms' limitations in the benchmarking problems, and we proposed a flexible scheme in order to reach the optimal point in the complex optimization challenges.