Adaptive task transition framework assisted evolutionary multitasking for constrained multi-objective optimization

In the real world, there are many practical application problems that require simultaneously optimizing multiple conflicting objectives while satisfying certain constraints, which are called constrained multi-objective optimization problems (CMOPs). Accordingly, the key to solving CMOPs is to handle constraint satisfaction and objective optimization. Many scholars use the idea of divide and conquer in evolutionary multitasking to create auxiliary tasks to solve CMOPs, which is a very effective method. However, due to the different characteristics of different auxiliary tasks, it is obvious that they have different roles in solving CMOPs. Based on the above considerations, this paper proposes a constrained multi-objective evolutionary algorithm with an adaptive task transition framework to assist evolutionary multitasking (ATTCMO). In the adaptive task transition framework, the main population always performs a constrained multi-objective optimization task. The auxiliary population first performs an unconstrained multi-objective optimization task using genetic algorithm to help the population converge quickly and cross the infeasible regions. Moreover, it adaptively transitions the task into a multi-objective optimization task with constrained dynamic relaxation based on the stability of main population changes, and adopts differential evolution as the evolutionary operator, which helps to enrich the diversity of the population. ATTCMO is compared with nine most advanced constrained multi-objective evolutionary algorithms on 23 benchmark test problems and 6 real-world practical application problems, which validates the superior performance of ATTCMO.