State information based identification methods towards low order modeling

In this paper we propose a model reduction framework for obtaining low order linear and non-linear models for large scale non-linear, reactive fluid flow systems. Our approach is based on the combination of the method of Proper Orthogonal Decomposition (POD), and System Identification techniques. The proposed methods involve two steps. In the first step POD is used to separate the spatial and temporal patterns and in the second step different model structures of linear and of non-linear types are proposed to approximate the temporal patterns and corresponding model parameters are identified. In particular, model structures of LTI, LPV and of tensorial or multi-variable polynomial type in lower dimensional subspace are identified. It is shown here that the POD modal coefficients can be viewed as the states of the reduced model that is to be identified. This has allowed us to propose different reduced model structures. The resulting lower dimensional models need significantly low computation time. The methods are of generic nature and are promising to different large scale applications characterized by existence of coherent patterns. Moreover, to accommodate the existing knowledge in the form of plant output measurements in the reduced order modeling framework, a new approach is proposed. The efficiency of proposed methods are illustrated on a large scale benchmark problem depicting an Industrial Glass Manufacturing Process. The results show good performance of the proposed methods.