Abstract:There are many dynamic multi-objective optimization problems (DMOPs) in real life. Once the environment changes, evolutionary algorithms are required to quickly track the moving Pareto front (PF) or Pareto set (PS) of optimization problems over time. In this paper, we propose a classification-based multi-strategy prediction method (CMSP). Firstly, the approximate optimal solution obtained by optimization is used to detect the type of the PS change: invariance, translation and others. Secondly, different coping strategies are adopted for different types of change: if type is invariance, the elite individual is retained and diversity is ensured; if translation, a time series is established for the center point of the optimal solution set, and the population is updated by predictive gradient strategy, then, the predicted individuals are compared with the individuals retained from the old population to ensure the accuracy of the prediction; if others, several time series are established for multiple special points to predict the location of individuals in the new environment. Finally, introducing a population preservation strategy and a memory retrieval strategy is beneficial to more fully utilizing historical information. The experimental results show that CMSP can perform dynamic multi-objective optimization well.