Learning and knowledge-guided evolutionary algorithm for the large-scale buffer allocation problem in production lines

Abstract

The large-scale buffer allocation problem (LBAP) in production lines represents a significant optimization challenge, centered on the efficient allocation of limited temporary storage areas. Prior research has predominantly addressed the LBAP through dynamic programming, search algorithms, and metaheuristics. However, these methodologies are often problem-specific and inefficient when applied to large-scale scenarios. Consequently, there is a pressing need to investigate innovative algorithms beyond existing approaches. This paper presents a novel learning and knowledge-guided evolutionary algorithm designed for the LBAP in production lines. The proposed algorithm develops an adaptive genetic algorithm and a variable neighborhood search algorithm, incorporating a simulated annealing-based strategy. An online Q-learning algorithm is employed to dynamically select the more effective of the two preceding algorithms for solution updates, while the simulated annealing-based strategy regulates the acceptance of these updated solutions. Furthermore, The proposed algorithm dynamically adjusts crossover, mutation, and shaking rates to adapt to the neighborhood structure. It also leverages conflict knowledge obtained from prior update experiences to inform the search process, thereby enhancing solution quality and computational efficiency. Numerical results indicate that the proposed algorithm surpasses state-of-the-art methods in addressing the LBAP. Additionally, empirical ablation studies demonstrate that the knowledge-guided approach efficiently explores promising solution regions by eliminating low-value solutions, while the learning-guided approach effectively generates improved solutions by selecting optimal strategies. This proposed algorithm significantly advances dynamic production resource allocation in large-scale systems.