An adaptive approach to improve the accuracy of a rolling load prediction model for a plate rolling process

We present a method that integrates off-line rule identification and an on-line adaptive approach to improve the accuracy of a rolling load prediction model for a plate rolling process. Based on the physical model of a plate rolling process, this work presents an empirical and adaptive approach to improve the accuracy of a rolling load prediction model. Our method consists of an off-line rule identification method and an on-line adaptive method. Using a hierarchical clustering method, our rule identification method finds a set of optimal rules that determine appropriate model parameters depending on an operational environment. In contrast to traditional approaches where such rules are determined in an ad-hoc manner, our method provides a systematic method to find optimal rules under the specification on model accuracy. Then, using a recursive least-square error method, our on-line adaptive method tunes model parameters by feeding back the observed model errors. Our off-line approach is effective to deal with nonlinear characteristics of the process, and our adaptive approach guarantees to maximize and to maintain the accuracy even if time passes. A successful application of the proposed approach to the plate rolling process is also shown.