Temperature prediction for roller kiln based on hybrid first-principle model and data-driven MW-DLWKPCR model

Abstract

In this paper, a hybrid temperature prediction model is developed for an industrial roller kiln of lithium-ion battery cathode materials, which is based on first-principle model and moving window-double locally weighted kernel principal component regression (DLKWKPCR). First, the mechanism model is built for the roller kiln according to the energy conservation law and heat transfer mechanism. Since the first-principle model is based on some simplified assumptions, it often results in large estimation errors. Thus, a data-driven error compensation model is further constructed with real-time process running data. In order to handle the strongly nonlinear, highly redundant and gradually time-varying characteristics, the error compensation model is built with moving window based DLWKPCR. Finally, a hybrid temperature prediction model is obtained by combining the compensation model and the mechanism model. An industrial roller kiln is utilized to test the effectiveness of the hybrid prediction model, in which the modeling results demonstrate that the developed hybrid prediction model can correctly estimate the roller kiln temperature.