A Novel Soft Sensor Model Based on Stacking Ensemble Learning Framework

Abstract

In this paper, an auxiliary variable selection method based on the coefficient of variation and maximum mutual information coefficient (CV-MIC) is proposed to select the most favorable variables for the model as auxiliary variables. On this basis, a new soft sensing model based on stacking ensemble learning framework is established. The model uses LSTM-GPR, SDAE-SVR and XGBoost soft sensing methods as the base learner for ensemble learning, and DNN as the meta-learner for ensemble learning to combine the base learners. To verify the effectiveness of the proposed method, the model is applied to the prediction of the rotor deformation of the boiler air preheater in thermal power plants. The experimental results show that the soft sensor model based on ensemble learning has higher prediction accuracy than the traditional single-model soft sensor method. Key Words: auxiliary variable selection, ensemble learning, soft sensor model, rotor deformation