expensive constrained multi-objective optimization problems (ECMOPs) present a significant challenge to surrogate-assisted evolutionary algorithms (SAEAs) in effectively balancing optimization of the objectives and sa...
详细信息
expensive constrained multi-objective optimization problems (ECMOPs) present a significant challenge to surrogate-assisted evolutionary algorithms (SAEAs) in effectively balancing optimization of the objectives and satisfaction of the constraints with complex landscapes, leading to low feasibility, poor convergence and insufficient diversity. To address these issues, we design a novel algorithm for the automatic selection of two acquisition functions, thereby taking advantage of the benefits of both using and ignoring constraints. Specifically, a multi-objective acquisition function that ignores constraints is proposed to search for problems whose unconstrained Pareto-optimal front (UPF) and constrained Pareto-optimal front (CPF) are similar. In addition, another constrainedmulti-objective acquisition function is introduced to search for problems whose CPF is far from the UPF. Following the optimization of the two acquisition functions, two model management strategies are proposed to select promising solutions for sampling new solutions and updating the surrogates. Any multi-objective evolutionary algorithm (MOEA) for solving non-constrained and constrainedmultiobjectiveoptimization problems can be integrated into our algorithm. The performance of the proposed algorithm is evaluated on five suites of test problems, one benchmark-suite of real-world constrainedmulti-objectiveoptimization problems (RWCMOPs) and a real-world optimization problem. Comparative results show that the proposed algorithm is competitive against state-of-the-art constrained SAEAs.
In this paper, an adaptive surrogate-assisted MOEA/D framework (ASA-MOEA/D) is proposed for solving computationally expensive constrained multi-objective optimization problems, in which three specific search strategie...
详细信息
In this paper, an adaptive surrogate-assisted MOEA/D framework (ASA-MOEA/D) is proposed for solving computationally expensive constrained multi-objective optimization problems, in which three specific search strategies are adaptively implemented based on the optimization states of subproblems to achieve targeted searches for different subproblems. To maintain feasibility, the RBF-based local search models are constructed by comprehensively considering the orthogonal distance difference and constraint satisfaction information for guiding infeasible solutions of the infeasible subproblems into feasible regions. To maintain convergence, the RBF surrogates of the aggregated objective and constraints are employed to construct local search models for locating better feasible solutions. To maintain diversity, the subregions of unexplored subproblems are effectively explored by utilizing the valuable information of their neighboring elite solutions. Moreover, the solution with the maximum overall uncertainty of RBF surrogates is selected for progressively increasing the prediction accuracies of surrogates. Therefore, ASA-MOEA/D strikes an adaptive balance among diversity, feasibility and convergence with the assistance of RBF surrogates as the optimization progresses. Empirical studies on three classical test suites demonstrate that ASA-MOEA/D with tchebycheff approach achieves highly competitive perfor-mance over other four state-of-the-art algorithms.
expensive constrained multi-objective optimization problems involve computationally expensiveobjectives and constraints, which impose stiff challenges on traditional evolutionary algorithms to optimize within limited...
详细信息
ISBN:
(纸本)9798350377859;9798350377842
expensive constrained multi-objective optimization problems involve computationally expensiveobjectives and constraints, which impose stiff challenges on traditional evolutionary algorithms to optimize within limited function evaluations. To address this, some surrogate-assisted constrainedmulti-objective evolutionary algorithms have been proposed, where surrogate models are constructed to replace expensive function evaluations. However, most existing surrogate models are either regression or classification models, which are liable to poor reliability in approximating complicated constraints. In this paper, a relation-and-regression-assisted constrainedmulti-objective evolutionary algorithm, named RCMOEA, is proposed. In RCMOEA, each regression model is constructed to approximate each objective function, and each relation model is built to learn the relation of constraint values between any two solutions. Based on the constructed surrogate models, a relation-and-regression-based constrained Pareto dominance, denoted as RCPD, is proposed to compare solution pairs. By adopting RCPD as the dominance criterion, the RCPD-based selection strategy is proposed for selecting offspring solutions. Also, the distance-based infill sampling strategy is proposed to preserve the diversity of solutions. Experimental results demonstrate the superiority of RCMOEA over the compared algorithms.
暂无评论