A deep learning-based particle contribution evaluation mechanism for meta-heuristic optimization algorithms

Meta-heuristic algorithms have been a popular research field nowadays. However, they are prone to falling into local optima, especially when applied to the Problem with Weak Influence of Global Optimal Solutions (PWIGOS), where the global optimal solution has a very small influence area in the search space. In this paper, based on the analysis of the influence of PWIGOS on meta-heuristic optimization algorithms, a novel Particle Contribution Evaluation Mechanism (PCEM) is proposed. Different from the current mechanisms in this field, PCEM is innovative in that it uses deep learning models to infer whether a particle is a high contribution particle within the influence region of the global optimum according to the feature information. This provides meta-heuristics with this additional critical information from outside the optimization process to guide the correct evolution of particle population. Additionally, a dynamic threshold setting method and a particle evolution adjustment method are designed, and three different types of classic and representative meta-heuristic algorithms, differential evolution (DE), Particle Swarm Optimization (PSO) and Gravitational Search Algorithm (GSA) are selected as application examples of PCEM. Experiments are conducted on 27 benchmark functions, CEC2017 benchmark suite and four real-word problems. According to the statistical results, PCEM not only excels in particle contribution assessment but also significantly enhances algorithm performance, especially when addressing challenging PWIGOS.