A multi-step predictor with a variable input pattern for system state forecasting

A reliable predictor is very useful to a wide array of industries to forecast the behaviour of dynamic systems. In this paper, an adaptive multi-step predictor is developed based on a novel weighted recurrent neuro-fuzzy paradigm for system state forecasting. A variable input pattern is proposed to improve the forecasting performance. A hybrid training algorithm, based on the recursive Levenberg–Marquardt algorithm and recursive least square estimate, is suggested to enhance forecasting convergence and to accommodate time-varying system conditions. The viability of the developed predictor is evaluated by simulations on both benchmark data sets and experimental data sets corresponding to machinery condition monitoring. The investigation results show that the developed adaptive predictor is a reliable and robust multi-step forecasting tool. It can capture and track system's response quickly and accurately. It outperforms other related classical forecasting schemes.