Memory-Enhanced Evolutionary Algorithm for Fast Multi- and Many-Objective Optimization

This paper proposes a memory-enhanced evolutionary algorithm that can effectively address multi- and many-objective optimization problems with various Pareto fronts within a small number of generations. Traditional multi-objective evolutionary optimization algorithms face challenges when dealing with real-world problems featuring many objectives and irregular Pareto fronts. Inspired by human memory mechanisms, this paper integrates a memory mechanism into the CA-MOEA algorithm. By constructing and utilizing a memory network, the location of the Pareto front is predicted, accelerating the convergence of the algorithm and enhancing its performance. The memory network employs a gated recurrent unit neural network, which learns the relationship between individuals from previous generations and the Pareto front through training. Additionally, this paper introduces a memory-based environmental selection strategy, adopts the clustering-based environmental selection method from the original algorithm, and presents a dynamic adjustment method for the environmental selection strategy. This enhances the algorithm’s ability to explore and exploit resources in complex problem spaces. Experimental results demonstrate that MCA-MOEA exhibits faster convergence speed and better performance on most test cases, confirming the effectiveness of the memory mechanism.