Collaborative Optimization of Multiple Operating Parameters for Process Industries Based on Multi-Agent Reinforcement Learning

Process industries are often confronted with strong multi-operational parameter couplings, intricate process topologies, and difficulties in multi-stage coordination, which render conventional localized optimization methods inadequate for achieving global optimality. To address these challenges, this paper proposes a graph spectral theory-based process topology-aware multi-agent reinforcement learning collaborative optimization method for multiple operating parameter collaborative optimization in complex topological process industries. Specifically, a topology analysis framework based on Laplacian spectral analysis is developed to characterize structural coupling relationships among multiple operating parameters, thereby supporting agent task allocation and coordinated decision-making. Subsequently, a temporal perception module integrating long short-term memory networks with a multi-head attention mechanism is designed to extract key temporal dependencies from historical state trajectories. Furthermore, a hierarchical spatial attention mechanism is introduced to enable dynamic and adaptive regulation of optimization attention across organizational, variable, and continuous control domains. On this basis, a hierarchical reinforcement learning architecture is constructed to coordinate local and global policy optimization, facilitating cooperative control and strategy optimization among multiple agents. Simulation experiments using industrial data from a continuous stirred tank reactor system and a representative salt-lake chemical process validate the effectiveness of the proposed method. Experimental results show that the proposed method achieves up to a 41.2% performance improvement over conventional approaches, exhibiting superior convergence behavior and policy stability, and providing a viable technical pathway for multiple operating parameter collaborative optimization in process industries.