Multi-Model Fusion Based NOx Emission Prediction for Power Plant Boilers

Abstract

In view of the complex characteristics of nonlinear and multi operating conditions in the boiler combustion process of power station, a NOx prediction method of boiler combustion process based on MiniBatchKMeans clustering and Stacking model fusion is proposed. This method optimizes the clustering and division of training sets by applying MiniBatchKMeans clustering algorithm, and establishes the stacking fusion framework prediction model (Stacking-XRLL) based on the XGBoost, Random Forest, LightGBM and the Linear Regression, so as to realize the accurate prediction of NOx emissions under the variable operating conditions of power plant boilers. Taking the NOx emission data of a power plant boiler in Guangdong as an example, the modeling simulation and experiment results shows that compared with the single modeling methods such as the BP, the LSTM and the GRU neural network, the Stacking-XRLL modeling method has higher generalization ability and prediction accuracy, and the average accuracy reaches 99%.