A parameter dispersion based global identification method and its application

Abstract

Local identification method provides one point in the parametric space, which locally minimizes the criteria, the deviation of the object output and the model output. However, for process control system modeling, it often exists a set of points that satisfy to minimize the criteria. This paper proposed a global identification method, which is based on the local identification results to determine the upper bound and lower bound of the model parameters by parameter dispersion method. Then gradient projection is applied to search for the optimal projection direction and project the parameters surfaces on this direction, which is named as equidistance-projection. The application of parameters identification of industrial multiple-step batch polyvinyl chloride synthesis reaction control model shows that the proposed method is efficient and practical in dynamic process control system modeling.