Dynamic Multi-objective Optimization via Sliding Time Window and Parallel Computing.

Tracking changing Pareto front (PF) in the objective space and Pareto set (PS) in the decision space is an important task in dynamic multi-objective optimization (DMO). Similarly, maintaining population diversity and reusing previous evolutionary information are useful to explore promising regions and to find high-quality solutions quickly in time-varying environments. To this end, a sliding time window based on parallel computing (STW-PC) is introduced in the present study. In the STW-PC, obtained time-sequence solution sets aim to preserve the diversity and facilitate a fast convergence since problems in successive time/environments are usually related. The parallel computing method is also employed to reduce the computational time. The STW-PC is incorporated into a multi-objective evolutionary algorithm and is compared with two competitors on 12 dynamic multi-objective optimization problems. The results show that the STW-PC can both improve the tracking performance of the selected algorithm in different degrees of changes, and significantly reduce the calculation time compared with transfer learning.