LSHADE-PS-QS: LSHADE with Population-enhancing Strategy and Quality-enhancing Strategy

Differential Evolution (DE) is one of the most effective metaheuristics for continuous optimization. Among its advanced variants, LSHADE has shown strong performance owing to success-history based parameter adaptation and linear population size reduction (LPSR). This paper proposes LSHADE-PS-QS, an enhanced LSHADE variant with a dual-module mechanism for stagnation handling. Specifically, a Quality-enhancing Strategy (QS) is introduced at the individual level to improve mutation quality through Q-learning-based action selection, elite guidance, and plateau perturbation, while a Population-enhancing Strategy (PS) is designed at the population level to maintain diversity through population expansion, partial restart, and population reduction. To further cope with the non-stationarity caused by LPSR, an LPSR-aware Q-value decay mechanism is incorporated. Experimental results on the CEC 2014 and CEC 2017 benchmark suites demonstrate that LSHADE-PS-QS is highly competitive and often superior to representative baseline algorithms over different dimensions and evaluation budgets.