Order Determination and Input Selection with Local Model Networks

Abstract An automatic order determination and input selection method for nonlinear dynamic models following the external dynamics approach is proposed. Based on a wrapperlike input selection approach using local model networks (LMN), the proposed method is able to automatically identify important operating point (OP) variables solely based on measured data. If local linear models are used, the OP variables describe regions in which the process can sufficiently be described by linear, dynamic models. The necessary order of these linear, dynamic models is determined altogether with the OP variables in one framework by the proposed method. We show that this method is able to improve the model accuracy and lead to a more concise model for a real-world heating, ventilating and air conditioning (HVAC) system.