Neural Representation-Based Wiener Process With Meta-Learning for Battery RUL Prediction Under Time-Varying Degradation Rate

The burgeoning development in industrial technology and the rapid evolution in the realm of new energy have precipitated an increasing need for dependable reliability assessment of lithium-ion batteries. However, the complexity of time-varying degradation rates poses a significant challenge in accurately predicting the remaining useful life (RUL) for lithium-ion batteries. Additionally, the acquisition of high-quality lithium-ion battery degradation data entails substantial time and financial investments. Consequently, a novel degradation model that employs the neural representation-based Wiener process is developed. The integration of the stochastic process with neural representation equips the proposed model with an enhanced capability for nonlinear fitting. Besides, the incorporation of meta-learning for model training facilitates the prediction capability effectively in application scenarios under time-varying degradation rates and limited available data. The efficacy of the proposed model is validated through comprehensive case studies on battery data, where the model is well trained just with the first 10% of it.