Dynamic Constrained Multi-Objective Optimization with Combination Response Mechanism

In the dynamic multi-objective optimization problem DMOP, several objective functions are optimized simultaneously. On the other hand, in a DMOP problem, the variables or objective functions may change over time, called changing the environment. The DMOP optimization algorithm must find the optimal solutions for the current environment before changing the environment. In recent years, the use of DMOP algorithms in solving various real-world problems, including control, planning, resource management, routing, and mechanical design problems, is increasing. However, due to the specific challenges of real-world problems, most DMOP algorithms do not apply to these problems or are not efficient enough. Most real-world issues have several constraints that change over time. Changing constraints often complicates the solution area. Many solutions suitable for the previous environment may be impossible for the new environment. Most DMOP algorithms lose their efficiency in solving these problems. In this paper a hybrid response mechanism called RM-RPM was proposed. In this mechanism, three proposed methods produce the initial population for the new environment. In the first method, random solutions are generated with the help of the DE\rand \1 operator and Cauchy mutation, which preserves the population diversity and the ability to search globally. In the second method, a combination of transfer learning and particular points is used to predict the location of appropriate solutions in the new environment. In the third method, a crowding-based replication mechanism was proposed that helps to create new solutions with acceptable variability around the best solutions. Experiments on four benchmark functions in the DMOP domain and five formulations of real-world problems demonstrate the proper performance of the proposed framework