FF-LSTM: A Novel Modeling Method for pH Neutralization Process

摘要

It is a great challenge to model chemical industrial processes because of their complex nonlinear characteristics and large time delay features. Traditional modeling technologies require establishing explicit model equations and more prior knowledge about reaction process. In this paper, a deep learning (DL) based approach: a feed-forward neural network embedded with long short-term memory (FF-LSTM) network, is introduced for modeling chemical industrial process, the pH neutralization process is studied as a case. A dropout technique is used to prevent over-fitting. Two types of experiments are implemented, both experiments are compared with the recurrent neural network (RNN), long short-term memory network and feed-forward neural network methods. Two evaluation indicators are selected to verify the quality of the model, namely root mean square error (RMSE) and the determination coefficient (R^{2). Experimental results show that the proposed method can achieve better results than the other methods in modeling the process of pH neutralization.}