Surrogate-Assisted Level-Based Learning Particle Swarm Optimizer Based on Expected Value Evaluation under High Noise Optimization

Surrogate-assisted evolutionary algorithms (SAEAs) have emerged as an effective approach for addressing expensive optimization problems. However, in scenarios where uncertain factors such as evaluation noises exist, the performance and reliability of most SAEAs are compromised due to inaccuracies and uncertainties, such as biases and measurement errors, which may vary. To mitigate these challenges, this paper proposes a particle re-evaluation technique based on a classifier-assisted approach. Specifically, the technique leverages a classifier-assisted level-based learning swarm optimizer to enhance the algorithm’s performance and reliability. Moreover, it explores various reevaluation rules tailored for uncertain conditions. Experimental results demonstrate that re-evaluation significantly enhances the classifier-assisted level-based learning swarm optimizers’ ability to improve performance and reliability, particularly in environments where objective fitness is sensitive to disturbances. Additionally, the experiments show that re-evaluation substantially boosts the optimizers’ capacity to discover superior solutions and enhances particle trustworthiness in uncertain situations.