On-line identification of thermal process using a modified ts-type neuro-fuzzy system

In this paper, a modified TS-type neuro-fuzzy system (MTSNFS) for on-line identification is proposed, which possesses six layers of neurons to perform the fuzzy inference. A modified self-organizing competitive learning algorithm with capabilities of dynamical rules recruitment and cancellation is proposed for structure identification. A hybrid learning algorithm combining recursive least squares (RLS) estimation and ordered derivative learning is used for parameter estimation. Both the structure and parameters could be automatically determined online without a priori knowledge. Comparisons with other related works are made via identification of Box-Jenkins furnance. Identification of bed temperature of a circulating fluidized bed boiler using the MTSNFS is also presented in this paper. The results demonstrate that the proposed identification approach is of high accuracy and compactness, and suitable for on-line modeling and prediction.