Near Feasibility Driven Adaptive Penalty Functions Embedded MOEA/D
Akhtar Munir Khan, Muhammad Asif Jan, Muhammad Sagheer, Rashida Adeeb Khanum, Muhammad Irfan Uddin, Shafiq Ahmad, Shamsul Huda · IEEE Access · 2023
Adaptive penalty function methods (APFMs) are promising constraints handling techniques. In an APFM, a penalty parameter which balances constrains’ violations and objective function values is adaptively adjusted. This work modifies an APFM that uses a near feasibility threshold (NFT), a portion around the feasible region where infeasible solutions are considered as good ones, for constrained multiobjective optimization. The modified APFM with five different settings of NFT is embedded in a prominent multiobjective evolutionary algorithm based on decomposition, MOEA/D. This brings in five constrained variants of the base algorithm, denoted by CMOEA/D-TAP1 to CMOEA/D-TAP5. These variants are tested on well-known constrained multiobjective benchmark test suits, the CTP series and the CF series. The proposed variants are compared with four best performing algorithms through HV metric (hyper volume metric) statistics for CTP series, and with seven state-of-the-art algorithms through Wilcoxon rank sum test employed to mean values of both HV and IGD (inverted generational distance matric) metrics for CF series. Simulation results reflect that overall performance of the newly introduced variants is better than the competitors for the taken benchmark test suits.