Temperature Prediction Model for Roller Kiln by ALD-Based Double Locally Weighted Kernel Principal Component Regression

Abstract

In the roller kiln of cathode materials for lithium batteries, the timely measurement of temperature is very important for effective process control. However, it is sometimes difficult or costly to measure the temperature timely. To handle this problem, a kind of soft sensor modeling framework with double locally weighted kernel principal component regression based on approximate linearity dependence (ALD) is proposed, which simultaneously carries out sample and variable weighted learning in the high-dimensional and nonlinear space to solve the process time-varying and strong nonlinearity problems. Moreover, a kind of just-in-time learning framework based on ALD is adopted for selectively updating the online local models. By setting a reasonable threshold of ALD, the prediction time can be effectively reduced, and the prediction accuracy can be maintained. The effectiveness of the proposed method is demonstrated on an industrial roller kiln. The results show that the proposed method can meet the requirements of prediction accuracy and time efficiency of the roller kiln of cathode materials for lithium batteries.