A Multiobjective Hybrid Evolutionary Algorithm For Constrained Optimization Problems

Evolutionary computation is an intelligent computational method that mimics the naturalbiological evolution process. Recently, employing evolutionary computation to solve constrainedoptimization problems (COPs) has been drawn concerns widely. COPs are a kind of commonproblems in practical engineering applications, and due to the diversity and complexity of constraints,solving this kind of problems is difficult. Therefore, the research of evolutionary computation onCOPs comes to necessity for very important theoretical and practical prospect. Therefore, with thelatest works of constraint handling techniques on evolutionary computation, this paper proposes amultiobjective hybrid constrained optimization algorithm for solving COPs. More specifically, themain work of this paper includes the following several parts:First of all, the importance of multiobjective optimization technique and the information sharingamong communities are analyzed for making the foundation to solve COPs in the subsequent research.Thus, in order to achieve the above purpose, this paper proposes an improved multiobjective invasiveweed optimization (IWO) algorithm based on the information sharing among communities. In thisalgorithm, an information sharing operator is presented, and its effectiveness is tested.Secondly, a hybrid constrained optimization algorithm is proposed based on differentialevolution (DE) as a global search process and invasive weed optimization as a local refinementprocess. In this proposed hybrid algorithm, the global search ensures the diversity of the hybridalgorithm for not being trapped into suboptimal solution and avoiding wasting computationalresources. Meanwhile, local refinement guarantees the convergence of the hybrid algorithm, whichcan lead to a refined search for locating the optimal solution. In this paper, the hybrid algorithm hasbeen applied to solve the standard test functions and the practical engineering problems. Theexperimental results show that the algorithm can efficiently solve these functions and find the optimalsolution of these functions. Furthermore, the proposed algorithm is compared with severalstate-of-the-art constrained optimization algorithms, and the comparison shows that the proposedalgorithm is very competitive when compared with these algorithms.Finally, when the proposed algorithm runs the local refinement process, there is a "step-length"parameter in IWO that is needed to set in advance and is applied to determine the degree of localrefinement. However, it is very difficult for adjust this parameter into an appropriate value before thealgorithm runs, and this parameter can also change in the different stage of search process. Hence, to take this situation into consideration and on the basis of the hybrid algorithm, this paper furtherproposes an improved hybrid algorithm relying on ring topology. Using ring topology can effectivelyestimate the information of individual’s adjacent area, and then through the information estimation,the improved hybrid algorithm can adaptively set the parameter "step-length", which can lead to amore effectively local refinement. Meanwhile, with ring topology, this paper proposes an improvedmutation strategy that is applied to DE for improving the performance of global search. Besides, thecomparison between the improved algorithm and several state-of-the-art algorithms demonstrates thatthe improved algorithm presents an excellent performance. Moreover, this paper further discusses therole of each part on the improved hybrid algorithm.