Auto-Enhanced Population Diversity with Two Options

This paper proposes an auto-enhanced population diversity with two options (TO-AEPD) for addressing the premature convergence of CMA-ES. TO-AEPD is founded on a modification of AEPD. By quantifying the population distribution in each dimension, TO-AEPD allows the identification of population convergence or stagnation instances. When convergence or stagnation is identified, the population must be diversified to acceptable levels. Two population diversification enhancement methods are used to balance exploration and exploitation. TO-AEPD selects the population diversification strategy with exploration when the current round of optimization does not produce a new optimal solution or when convergence accuracy is achieved. Otherwise, population diversification with an exploitation strategy is implemented. TO-AEPD-CMAES describes the CMAES that incorporates TO-AEPD. Experimentation demonstrated that TO-AEPD-CMAES outperforms several competing algorithms on benchmarks, significantly enhancing the algorithm’s performance and validating the model’s viability and efficacy.