Adaptation of Model Prediction Error of Process Control by Local Regression Modeling

In this paper, a new method to estimate model prediction error using process database for the purpose of improvement in model prediction accuracy for the steel plant was investigated. The basic concept of the proposed method is to compensate prediction error of physical model by using local regression modeling method, such as Model-On-Demand technique or Just-In-Time modeling. In this method, several local models are obtained using several datasets and then the estimation value of prediction error is calculated by applying ensemble learning method. In addition, the effectiveness of proposed method has been confirmed for the temperature prediction model of plate cooling by numerical simulations and actual online tests.