Intelligent control strategy for coal to ethylene glycol wastewater emission reduction based on dynamic simulation and reinforcement learning

Key Words:

Abstract:Coal-to-Ethylene Glycol (CtEG) process discharges large amounts of wastewater due to the lack of control strategy to cope with uncertain disturbances, which increases the treatment cost, environmental pressure, and safety hazards. In this paper, a novel intelligent control strategy is developed to adjust the operating status of CtEG in real time to minimize wastewater emission. Firstly, a rigorous mechanistic modelling of CtEG process is conducted to obtain data set under dynamic disturbances. Based on the data set, the causal relationship between the control variables and wastewater is analyzed using Granger Causality Test (GCT) to establish the global causality map. Subsequently, the proxy model based on extreme gradient boosting (XGBoost) is constructed to calculate the chemical oxygen demand (COD) and product quality by control parameters in global causality map. Moreover, Shapley Additive Explanations (SHAP) method is used to select the vital control variables as actions of reinforcement learning (RL). Thirdly, the intelligent control model based on RL is constructed to determine the optimal decision by considering wastewater discharge, product quality, and operating cost. Finally, the intelligent control model built by coupling multiple models is verified to reduce wastewater COD by 22.8% and 22.98% under two different feed disturbances.

Volume:194

Issue:

Translation or Not:no