A modified dual-population approach for solving multi-objective problems

Abstract

Abstract-Maintaining the balance between convergence and diversity plays a vital role in multi-objective evolutionary algorithms (MOEAs). However, most MOEAs cannot reach a satisfying balance, especially when solving problems having complicated pareto optimal sets. In this paper, we present a modified cooperative co-evolution approach for achieving better convergence and diversity simultaneously (namely DPP2). In DPP2, while populations are trying to achieve both criteria, the priority being set for these criteria will be different. One population focuses on achieving better convergence (by using pareto-based ranking scheme), while the other is for ensuring the population diversity (by using the decomposition-based method). After that, we use a cooperation mechanism to integrate the two populations and create a new combined population with hopes of having both characteristics (i.e. converged and diverse). Performance of DPP2 is examined on the well-known benchmarks of multiobjective optimization problems (MOPs) using the hypervolume (HV), the generational distance (GD), the inverted generational distance (IGD) metrics. In comparison with the original version DPP algorithm, experimental results indicated that DPP2 can significantly outperform DPP on the benchmark problems with stable results.

A Competitive Co-Evolutionary Approach for the Multi-Objective Evolutionary Algorithms

“…Similar to WFG instances, DPPCP achieves better metric values in all of the comparisons (except WFG5). These results show that the competitive co-evolution model proposed in this paper outperforms the co-operative co-evolution methods in ( [31], [32]).…”

“…First, we compare the proposed method with some baseline algorithms (i.e. NSGA-II and MOEA/D-DE) and the state-of-the-art algorithms (ED/DPP in [31] and DPP2 in [32]). Via the comparison results, we could see how good the performance of the proposed method when comparing to the others.…”

“…Therefore, DPP can be considered a state of the art algorithm. In this study, we focus on comparing the DPPCP with the original algorithm DPP (named ED/DPP) and the DPP2 algorithm [32], a modified version of this algorithm.…”

“…In such a case, both mating parents are selected from the same population, rendering the co-evolutionary mechanism meaningless. To address these shortcoming, Vu et al [32] improved this model by proposing a new restricted selection mechanism as well as some small improvements in the DPP model to shorten the running time as well as achieving better results. Li et al [33] proposed a dual-population approach for balancing convergence and diversity.…”