Integrating Reinforcement Learning and LSTM for Raw Material Procurement Optimization in Salt Lake Chemical Manufacturing

In the process industry, fluctuations in raw material prices and changes in production demand pose significant challenges to formulating effective raw material procurement plans during production and operation. Existing procurement optimization methods typically assume known values for price sequences and production demand, which may not adequately address market volatility risks. To overcome these challenges, this study introduces a long-short-term memory (LSTM) model for short-term prediction of raw material market prices and production demand. Subsequently, deep reinforcement learning methods are utilized to make sequential decisions on raw material procurement plans to effectively navigate complex procurement environments. To validate the efficacy of our proposed method, the coal procurement process of a Salt Lake chemical company was chosen as an example. The real operating data was used to compare the difference between the proposed method and the traditional methods regarding decision-making performance. The results demonstrate that our proposed method significantly reduces procurement costs while improving overall enterprise production efficiency.