An Improved Two-Step Q-Learning Algorithm for Solving the Disassembly and Assembly Line Balancing Problem

The rapid development of modern industry has rendered traditional disassembly and assembly techniques inadequate. This study focuses on addressing the disassembly and assembly line balancing problem (DALBP). Given the limited exploration in the integration of disassembly and assembly line balancing, a simplified DALBP model is proposed, demonstrating its potential for application in reinforcement learning solutions. This research delves into how different task allocation strategies affect the minimization of disassembly costs and the maximization of profits when robots perform similar tasks, and based on this, a mixed-integer programming mathematical model aimed at maximizing disassembly profits is constructed. To tackle this issue, an innovative improved two-step Q-learning algorithm (ITQLA) is designed, which efficiently selects actions through an action selection table to optimize task allocation and effectively solve the DALBP model considering robot operation times. Comparative experiments with three other reinforcement learning algorithms further validate the outstanding performance and stability of the ITQLA algorithm, fully showcasing its application prospects and high efficiency.