In order to solve the dynamic multi-objective problems, a predictive dynamic multi-objective particle swarm optimization algorithm based on decomposition is proposed. Firstly, the target problem is divided into several sub-problems by decomposing. When the problem changes, the optimized individuals corresponding to different sub-problems are chosen to detect the severity of changes to improve the ability of the algorithm to adapt and respond to different dynamic problems. Then, a group prediction strategy, which converts individual positions at different moments in the same convergence direction of the target space into time series, is designed to improve the diversity and validity of the prediction population by introducing the time series prediction method and forecasting the position of the next moment, which can effectively improve the algorithm convergence speed after the problem changes. Finally, in order to avoid the mismatch between individuals and subproblems after the environment changes, a rematching strategy is designed to effectively improve the accuracy of the prediction strategy. The experimental results show that compared with the four dynamic multi-objective optimization algorithms, the proposed algorithm has significant advantages in convergence, distribution and stability over six standard dynamic multi-objective problems.