Attention-based LSTM Block Model Framework based on static and dynamic variables for modeling fuel ethanol fermentation process

Abstract

In the industrial fuel ethanol process, the initial feed conditions, which are static variables, and each control operation, which is dynamic variable and is changing during the producing process, have an impact on the concentration of ethanol out of the tank. Developing the accurate concentration model of ethanol out of the tank with the static variables and as early as possible dynamic variables is a challenging work and is useful in analyzing and optimizing the production process. Given that the ethanol fermentation process is multiphase and dynamic, a block concentration model of ethanol out of the tank based on multilayer perceptron (MLP) and long short-term memory (LSTM) is proposed to deal with the coexistence state of the static variables and the dynamic variables. A Channel-Squeeze-and-Excitation (CSE) module is constructed to solve the problem of high-dimensional input and low-dimensional output caused by the block model. The proposed model is applied for the industrial ethanol fermentation process. The results show that CSE-MLSTM is significantly different from other prediction models and improves the prediction accuracy. Temperature control experiments and ablation experiments effectively verify the reliability and effectiveness of CSE-MLSTM.