Identification of nonlinear system with irregularly and slowly sampled output using LPV model

To deal with the identification of nonlinear systems with slowly and irregularly sampled output, a linear parameter varying (LPV) model is used and it consists of weighted local models, which is also called multi-model structure. In this work, the output error (OE) method is utilized to identify the local models and the global LPV model is represented by the combination of the local models weighted by cubic spline function and asymmetric Gaussian function. Gauss-Newton method and stochastic gradient method are respectively applied to estimate the parameters of the local OE models and weighting functions. Meanwhile, the problem of initialization and stability of the numerical search algorithm are also considered and handled in this paper. To identify a more accurate LPV model, a multi-iterative method is developed. The effectiveness of the proposed approach is illustrated by an industry case.