Mixed-Layout Multi-Type Factory Remanufacturing System Optimization via LLM-TD3

Abstract

This work presents a mixed-layout, multi-type factory remanufacturing system optimization problem that considers both linear and U-shaped disassembly lines, with the goal of maximizing profit, and formulates its corresponding mathematical model. Its solution has four stages: product allocation, disassembly line selection, task allocation, and component transportation. Based on the characteristics of each stage, Large Language Model (LLM) is responsible for product allocation and disassembly line selection, while Twin Delayed Deep Deterministic Policy Gradient optimizes task allocation and component transportation according to the LLM’s results. By providing the estimated profits of products under different factory and disassembly line configurations and designing tailored action-state space, the proposed method interacts with the environment to solve the problem. By using various experimental cases, we compare it with CPLEX, Deep Deterministic Policy Gradient, Soft Actor-Critic, and Advantage Actor-Critic to verify its feasibility and effectiveness, demonstrating its potential as a novel solution method.