CO Emission Prediction of MSWI Processes Based on Nonlinear Feature Reduction Long Short-Term Memory Neural Network

Carbon monoxide (CO), one of the by-products of the municipal solid waste incineration (MSWI) processes, is a toxic gas that is harm to human health. Moreover, its emission concentration relates to the dioxins (DXN) from the MSWI plant directly. Thus, the CO emission concentration should be predicted in terms of assisting the optimal control of pollutant emission in the MSWI process. In this article, a prediction method of CO emission concentration based on nonlinear feature reduction and long short-term memory neural network (LSTM) is proposed. Firstly, the nonlinear feature selection based on mutual information (MI) is carried out on the preprocessed data to remove the features with weak correlation. Then, the nonlinear features were extracted based on one-dimensional convolution (1DCNN), which is fed into LSTM to construct the prediction model. The convolutional layer and LSTM parameters are updated based on the loss function. Finally, the validity and rationality of the proposed method are verified based on the benchmark dataset and the actual industrial CO dataset.