Location and manifold based dual prediction method for dynamic multiobjective optimization with stochastic changes

Abstract Most methods for solving dynamic multiobjective optimization problems (DMOPs) are suitable primarily for addressing problems with deterministic changes. However, in practical scenarios, environmental changes in DMOPs are often characterized by stochasticity. Therefore, a location- and manifold-based dual prediction method (LMDP) is proposed for DMOPs with stochastic changes. In this method, one or two historical PSs closest to the PS in the new environment are identified to predict the location of the new PS. Based on the predicted location, submanifolds of the historical PSs similar to the manifold of the PS in the new environment are subsequently identified to predict the manifold of the new PS. This approach improves the ability of the prediction model to capture the location and manifold of the PS in the new environment, thereby increasing its adaptability to stochastic changes. The proposed method is tested on DMOPs designed with stochastic changes. The results demonstrate that our method outperforms several comparative methods, thus validating its effectiveness in solving DMOPs with stochastic changes.