Research On Multi-objective Evolutionary Algorithms Handling Complicated Constraints

Constrained multi-objective optimization problems widely exist in scientific research and engineering practice,which often contains multiple conflicting objective functions and complex constraints.Such as vehicle routing problems require minimizing the total travel distance and the number of vehicles simultaneously,and need to satisfy the constraints of the vehicle loading tolerance,time window,etc;the optimal allocation of water resources requires minimizing the total water shortage in the receiving area and maximizing the economic utility of the resource utilization,and satisfying the constraints of water supply capacity and water demand capacity,etc.However,how to efficiently and accurately obtain the optimal solution of this kind of problem is a key problem that needs to be solved urgently in the field of scientific research.Constrained multi-objective evolutionary algorithm is a kind of method to solve constrained optimization problem by using the characteristics of high search efficiency,strong versatility and good robustness of evolutionary algorithm,which has attracted the interest of many scholars and proposed a large number of constraint handling techniques.The existing constrained multi-objective evolutionary algorithms can obtain a set of uniformly distributed Pareto optimal solutions when solving simple constrained multiobjective optimization problems,with excellent convergence and diversity.However,in the face of complicated constrained multi-objective optimization problems,the existing constrained multi-objective algorithms have encountered some difficulties such as effectively dealing with constraints,and balancing constraints and objectives.In order to overcome these problems,this paper analyzes the characteristics of the existing constraint handling techniques based on the existing evolution algorithms,and designs new constrained multi-objective evolutionary algorithms.The specific contents are as follows:(1)For complex constraints,which are difficult to deal with effectively.This paper a new adaptive ε-constraint-handling technology and detection-and-escape mechanism are embedded in the MOEA/D algorithm,called MOEA/D-εDAE.The algorithm adaptively adjusts the setting value of ε based on the feasible solution information in the objective space to avoid the unreasonable setting value of ε.In addition,the algorithm can make full use of the infeasible solution information in the objective space to search for the complete feasible regions.(2)For the problem that is difficult to balance objective optimization and constraint satisfaction.This paper researches the search bias of different constraint handling techniques,and proposes a new adaptive framework,called ACMO,which adaptively selects constraint handling techniques according to the evolution information of the population to adjust the search bias,balance objective optimization and constraint satisfaction,and search for global optimal solutions.(3)By solving a large number of complicated benchmark problems,the proposed algorithm is compared with several popular constrained multi-objective evolutionary algorithms.The results show that the proposed algorithm has obtained better performance and has certain competitiveness.