Soft Sensor Method Based on Deep Belief Network for Rotor Thermal Deformation of Rotary Air Preheater

Abstract

In this paper, to detect the rotor thermal deformation of rotary air preheater, a soft sensor method based on Deep Belief Network (DBN) is proposed. The rotor thermal deformation of rotary air preheater can be accurately detected by the proposed method, so the air leakage of rotary air preheater under the harsh working environment can be well controlled. In the study, latent variables which are closely related to rotor thermal deformation are obtained by grey relation analysis method. Then, DBN network is trained by using labeled data and unlabeled data, in which the features in the data set are extracted by DBN module. The features extracted by DBN module are input Support Vector Regression (SVR) as input data. At the same time, Particle Swarm Optimization (PSO) algorithm is used to select the appropriate parameters for SVR. SVR is used as the predictor of the continuous target change value in the soft sensor model. The new soft sensor model of rotor thermal deformation is obtained by the superiority of DBN and SVR algorithms. Simulation result shows that the identification accuracy of this new model is higher, and the prediction of rotor thermal deformation is accurate, so it can predict the output well.