Elite knowledge transfer within lower-level searches for bilevel optimization

Bilevel optimization problems pose significant challenges for evolutionary algorithms (EAs) due to their nested structure. This paper introduces an efficient evolutionary bilevel algorithm that leverages elite knowledge transfer to tackle these challenges. Firstly, this paper employs a biobjective source selection strategy to balance convergence quality with relevance to the target lower-level problem. Building on this, a multi-source elite knowledge transfer mechanism constructs an elite Gaussian distribution model from source lower-level solutions, facilitating efficient parameterized knowledge transfer to accelerate the optimization of the target lower-level problem. Additionally, an adaptive strategy for reducing the lower-level population size further enhances algorithmic efficiency. Evaluated on benchmark test suites and real-world problems, the proposed algorithm demonstrates superior efficiency and accuracy compared to state-of-the-art bilevel optimization algorithms, underscoring the effectiveness of the elite knowledge transfer and adaptive reduction strategies. The source code for EKTBO has been publicly released at the following link: https://github.com/tg980515/EKTBO