Soft sensor based on DBN-IPSO-SVR approach for rotor thermal deformation prediction of rotary air-preheater

Abstract The rotor thermal deformation prediction is a challenging task due to the sophisticated processes involved in rotary air-preheater and limited hardware sensors, so it is necessary to employ a soft sensor model to get the accurate deformation value for the existing sealing technology. Additionally, aiming at the problems of low accuracy and underutilization of data of traditional methods, a novel soft sensor for air-preheater based on DBN-IPSO-SVR is proposed. The grey relational analysis (GRA) method is employed to provide reliable input variables for model training. The deep belief network (DBN) and the support vector regression (SVR) with the improved particle swarm optimization (IPSO) as data-driven model are employed to extract the features in the data. The results demonstrate the IPSO can obtain better parameters, the proposed soft sensor model significantly improved the performance of rotor thermal deformation prediction and is therefore a valuable non-contact measure tool for controlling air leakage.