Remaining Useful Life Prediction of Lithium-Ion Batteries Based on Artificial Neural Network-Supported Wiener Process with Random Effects

Among various data-driven approaches for remaining useful life (RUL) prediction, Wiener process-based degradation models are interpretable and quantify uncertainty. However, existing methods that combine artificial neural networks (ANNs) with the Wiener process have not yet fully exploited the modeling power of neural networks or the analytical tractability of stochastic processes. To overcome this limitation, this paper proposes an ANN-supported Wiener process with random effects for RUL prediction. Specifically, a bias-free four-layer neural network is constructed to characterize the latent nonlinear degradation trajectory, while the drift coefficient is modeled as a random variable to capture unit-to-unit variability. An analytical expression for the time derivative of the network output is derived, facilitating the closed-form solution of the probability density function (PDF) of RULs within the first passage time framework. A training strategy based on the Levenberg–Marquardt algorithm is developed for efficient parameter estimation, and the Akaike information criterion is used to select the optimal network architecture. The proposed method is validated using two lithium-ion battery degradation datasets. Experimental results verify that the proposed method exhibits superior performance to traditional nonlinear degradation models in terms of both fitting performance and RUL prediction accuracy, with the advantages becoming more pronounced in long-term prediction scenarios.