Recursive identification of a turbo-generator plant using structurally adaptive neural networks

This paper presents an enhanced version of the "online adaptive k-tree lattice learning" (ONALAL) algorithm to train "rectangular local linear model" (RLLM) networks. It is especially designed for online identification of nonlinear dynamical systems using the NARX structure. Basically, the algorithm performs a recursive adaptation of the complete structure and all parameters of the network. Thus, the significant inputs of the network (regressors of the NARX structure) as well as the number of local linear models are automatically determined. Furthermore, the parameters of each local linear model are optimized using a recursive optimization method (RLS algorithm). This leads to parsimonious models of SISO or MIMO dynamical systems, a primordial aim when solving nonlinear system identification problems. The effectiveness and the performance of the new approach is demonstrated by the real-time identification of a highly nonlinear plant-a turbogenerator.