MI-HGRU: A Combine Method of Mutual Information and HGRU Neural Network for Boiler NOx Emission Prediction

Abstract

Accurately predicting NOx emissions during boiler combustion is of significance for the operation and control of the boiler combustion systems in coal-fired power plants. According to the characteristics of the boiler combustion process with strong disturbances, highly nonlinear and multivariate coupling, a fusion model based on mutual information variable selection and Hyperopt optimized GRU neural network (MI-HGRU) is proposed. The method can accurately select the parameters of boiler combustion process base on mutual information feature selection algorithm. A GRU neural network prediction model with Hyperopt optimization is established to achieve accurate prediction of NOx emission during boiler combustion. The modeling experiment was using a NOx emission dataset from a power plant boiler in Guangdong. The experimental results show that the MI-HGRU method has higher generalization ability and prediction accuracy than the RBF, LSSVM, RNN and LSTM neural networks, with an average accuracy of 99.4%.