关键词: 多模态多目标优化, 白鲸优化, 密度聚类, 图神经网络

Abstract:

In order to solve the problems of unequal distribution and convergence of the equivalent solutions obtained by the multi-modal multi-objective optimization algorithm， a multi-modal multi-objective optimization algorithm based on density clustering （MMO_DBSCAN_BWO） is proposed. Firstly， the chaotic mapping initializes the population， makes the individuals evenly distributed in the decision space， and enhances its global search ability. Secondly， density clustering algorithm is used to generate niches based on non-dominated solution sets of different levels， identify stray individuals， and propose hierarchical evolutionary strategies and hierarchical ring topological evolutionary strategies to enhance search ability. Then， the archival mechanism is introduced to preserve the optimal solution of individual history. Finally， the optimal Pareto solution with uniform distribution is selected based on special congestion distance non-dominated sorting and Euclidean distance truncation mechanism. The overall performance of MMO_DBSCAN_BWO is better than that of the comparison algorithm by testing on the 2020 CEC multimodal multiobjective test function and comparing with the three algorithms. At the same time， the algorithm is applied to the hyperparameter optimization problem of JavaScript malicious code detection graph model， and the accuracy of the model is further improved， reaching 99.3%， which proves the effectiveness of the algorithm.

Key words: multimodal multi-objective optimization, beluga optimization, density clustering, graph neural networks