An evolving neuro-fuzzy technique for system state forecasting

A reliable predictor is very useful to real-world applications to forecast the future states of dynamic systems especially when their dynamic characteristics are time-varying. In this paper, an evolving neuro-fuzzy (eNF) technique is developed for system state forecasting. A novel clustering paradigm is developed for cluster generation and rule base modification. A new enhanced LSE method is proposed for online training of the eNF parameters, whose Lyapunov stability is verified theoretically. The effectiveness of the developed eNF predictor is evaluated based on simulation using some benchmark data sets. Next the eNF is implemented for the applications of currency exchange rate forecasting and machinery condition prognostics. Test results show that the enhanced LSE is computationally efficient and can improve reasoning convergence; the developed eNF predictor is an accurate forecasting tool. It can capture the system's dynamic behavior quickly and track the system's characteristics accurately. It is also a robust predictor to accommodate different system conditions.