An online modelling method is developed for nonlinear and time-varying processes based on the application of the structure-time-varying nonlinear ARX model (NARX) in a pH waste water neutralisation process. In the present study an online model structure and parameter updating algorithm is applied for identification of the polynomial NARX model. Both the model structure and the related parameters are updated online using an exponential window. The selection of regression variables is based on the contribution these candidate variables make to the output at each selection step, so that the sum of squares errors is minimised. The time-varying model consisted of a series of sub-models can accommodate variations in both the system structure and the parameters. These sub-models may provide a basis for the design of an adaptive controller for the system. The new method was applied to a laboratory scale plant.
展开▼