Directional considerations when tracking time-variant parameters

The need for directional considerations when tracking time-variant parameters became apparent during the development of a predictive model for an industrial blast furnace, and a method capable of such considerations is suggested. The method utilizes the structure of the Kalman filter, but does not require any prior knowledge of the variations in the parameters. Instead, some simplifications motivated in the context of parameter estimation and identification are introduced, which give an approach to determine individual “gains” in the update of the parameters plausible. Along with simulations, the application of the method to a challenging full-scale process, which exhibits an a priori unknown time-variant behavior, clearly illustrate the potential of the proposed method