Differential evolution with rankings-based fitness function for constrained optimization problems.

When evolutionary algorithms are employed to solve constrained optimization problems (COPs), how to efficiently make use of the information of some promising infeasible solutions is very important in the process of searching for the optimal feasible solution. In this paper, for selecting and making full use of some better infeasible solutions, a rankings-based fitness function method is designed. Specifically, the final fitness function of each individual is obtained by weighting two rankings, which are got after sorting the population based on the ɛ constraint technique and only based on the objective function, respectively. Furthermore, the weight is dynamically adjusted by considering the proportion of feasible solutions and generation information. By doing this, the tradeoff in constraints and objective can be addressed. Moreover, the promising offspring are generated by three differential evolution strategies with distinct characters to balance diversity and convergence. In addition, 116 benchmark problems from three test suites are used to evaluate the performance of the proposed method. Nine commonly used practical problems are selected to test the potential of the algorithm to solve real-world problems. Experimental results indicate that the proposed method shows superior or competitive to other state-of-the-art methods tailored for COPs. Moreover, the effectiveness of each introduced component in the proposed algorithm is investigated by the ablation study. • A new constrained evolutionary optimization algorithm is developed, denoted as DERFF. • Two rankings are got for each individual by two ways, one prefers constraint and the other prefers objective function. • The weight for summing these two rankings is dynamically adjusted by the proportion of feasible solutions and generation information. • Three evolutionary strategies are combined to balance the convergence and diversity of the population. • Experimental results indicate the effectiveness of the DERFF against other advanced algorithms. [ABSTRACT FROM AUTHOR]