Data Driven Multi-Agent Deep Reinforcement Learning Based Operational Optimization of Industrial Aerodynamic System

Industrial aerodynamic system (IAS) involves a series of compressors to provide the compressed air for consumer of steel industrial park, which consumes a substantial amount of electrical energy and attracts a considerable attention for energy conservation and efficiency enhancement. The usage pattern of these compressors, influenced by the cyclic and intermittent nature of production schedules, leads to varying operational profiles. Traditional manual adjustments are often inadequate due to the differ in expert knowledge, resulting in increased energy consumption. To tackle this challenge, we propose an innovative approach termed as data-driven multi-agent deep reinforcement learning (DMA-DRL) to achieve the operational optimization of IAS. Initially, we construct a multi-operational model for the air compressor group by leveraging historical data from industrial sites and understanding equipment operation principles; Considering the constraints in terms of production process conditions and equipment safety, we establish a data-driven optimization scheduling method for air compressor units based on multi-agent deep reinforcement learning, which ensures the secure and stable operation of the system under complex conditions. Finally, the proposed method is applied to the performance testing of an industrial park air compressor simulation system, which outperforms the various optimization strategies in aspect of operational cost and energy efficiency.