Iterative identification algorithms for parametric separable nonlinear network modelling

Mathematical models are fundamental for describing the dynamic behaviour of systems. System identification aims to estimate model parameters based on observation data. Selecting an appropriate model along with designing a high-precision parameter estimation algorithm can improve the modelling accuracy. This paper studies the parameter estimation problem for a class of parametric separable nonlinear network models, i.e. the RBF-ARX model. By exploiting the inherent parameter separability property, two gradient-based iterative sub-algorithms are derived to estimate the linear and nonlinear parameters separately through iterative search. A decomposition coordination-based iterative parameter estimation algorithm is then proposed by integrating the sub-algorithms. To incorporate real-time observation data into the modelling process, a dynamic window-based data utilisation scheme is introduced, leading to the development of an enhanced iterative parameter estimation algorithm. The proposed algorithms are validated through a numerical simulation example and applied to electric load forecasting of a real-world power system.