Prediction of remaining useful life for stochastic distribution systems based on hybrid residual correction method

Abstract Remaining useful life (RUL) prediction serves as a central component of predictive maintenance. This paper proposes a hybrid approach for fault diagnosis and RUL prediction, integrating the Auxiliary Particle Filter (APF), Autoregressive Integrated Moving Average (ARIMA) model, and Long Short‐Term Memory (LSTM) networks. First, a mapping between the system output and the fault is established. Second, system state estimation is performed using the APF, and the ARIMA model provides linear predictions of the second‐order output differences. Then, LSTM is utilized to predict the residual to correct the ARIMA prediction as the nonlinear part. Finally, the multistep ahead prediction of RUL is obtained by means of the fault diagnosis algorithm and the APF framework. The wet‐end control system of a paper machine is used as a case study to verify the effectiveness of the proposed algorithm.