Prediction of nitrogen oxide emission concentration in cement production process: a method of deep belief network with clustering and time series.

Abstract

The concentration of nitrogen oxide (NOx) emissions is an important environmental index in the cement production process. The purpose of predicting NOx emission concentration during cement production is to optimize the denitration process to reduce NOx emission. However, due to the problems of time delay, nonlinearity, uncertainty, and data continuity in the cement production process, it is difficult to establish an accurate NOx concentration prediction model. In order to solve the above problems, a NOx emission concentration prediction model using a deep belief network with clustering and time series features (CT-DBN) is proposed in this paper. Particularly, to improve data sparsity and enhance data characteristics, a clustering algorithm is introduced into the model to process the original data of each variable; the time series containing delay information are introduced into the input layer, which combines previous and current variable data into time series data to eliminate the influence of the time delay on the prediction of NOx emission concentration. In addition, restricted Boltzmann machine (RBM) is used to extract data features, and a gradient descent algorithm is used to reversely adjust network parameters to establish a deep belief network model (DBN). Experiments prove that the method in this paper has higher accuracy, stronger stability, and better generalization ability in predicting NOx emission concentration in cement production. The CT-DBN model realizes the accurate prediction of NOx emission concentration, provides guidance for denitration control, and reduces NOx emissions.