Prediction of Coke Oven Gas Production with Multivariable Input Based on LSTM Neural Network

Coke-oven is not a closed, stable and continuous reaction device in the actual production process. It has periodic operation procedures and the transformation of production rhythm. These factors lead to the strong nonlinear and uncertain characteristics of coke-oven production process. It is difficult to accurately model the gas generation process from the reaction mechanism. In this paper, a digital twin model for forecast of coke-oven gas generationcoke-oven in metallurgical industry based on long short term memory network (LSTM) is proposed. The forecasting models of coke-oven gas based on neural network does not need to study the mechanism of complex chemical reaction. It only needs to determine the correlation characteristics of parameters, and then conduct model training through sufficient historical data, so as to obtain accurate predictive results, which can provide data support for advanced applications such as abnormal working condition early warning and integrated energy management system.