A Lithium-ion Battery Remaining Useful Life Estimation Method Based on Nonlinear Autoregressive Fusion Model with Exogenous Variables

Remaining useful life (RUL) estimation is an important factor in lithium-ion battery system health management (PHM). After the lithium-ion battery reaches the end-of-life (EOL), the capacity drops rapidly, which can lead to battery failure and even safety incidents. Therefore, accurate RUL estimation is a critical and challenging problem. Additionally, this paper proposes a new RUL estimation method: NAR-MOGPR-PF. Firstly, a nonlinear autoregressive neural network (NARNN) is used to establish a time series model of battery capacity degradation as the state equation. Secondly, the three exogenous variables extracted from the battery data are used as observation parameters, and the multi-output Gaussian process regression (MOGPR) is constructed as an observation equation. Finally, a particle filter (PF) algorithm is constructed for capacity correction to realize the fusion between exogenous variables and the NARNN algorithm. The NAR-MOGPR-PF method improves the long-term prediction accuracy of capacity degradation by taking exogenous variables and nonlinear factors into account. Furthermore, the NAR-MOGPR-PF model is compared with autoregressive (AR) methods and the nonlinear autoregressive with exogenous inputs neural network (NARXNN). The proposed RUL estimation approach has higher accuracy and robustness, according to experimental data, and the RUL estimation error is about 1.