Temporal distribution-based prediction strategy for dynamic multi-objective optimization assisted by GRU neural network

To solve dynamic multi-objective optimization problems, evolutionary algorithms must be capable of quickly and accurately tracking the changing Pareto front such that they can respond in a timely and effective manner when detecting environmental changes. To address this challenge, we propose a dynamic multi-objective prediction strategy based on the temporal distribution characteristics assisted by a gated recurrent unit (GRU) neural network (GTBP). First, a time series is created using the improved historical centroid information, and the time series distribution information is represented by the temporal distribution characteristics. Subsequently, a GRU neural network is used to maximize the distribution characteristics and minimize the losses to train the network model. Finally, the individuals composed of the estimated manifold and the population centroids predicted by the model are combined with some individuals randomly generated for increasing the population diversity to form the initial population at the next moment. To evaluate the GTBP performance, it was compared with four dynamic multi-objective algorithms for 15 test problems. The experimental results demonstrated that GTBP is competitive for solving dynamic multi-objective optimization problems.