Estimation of Lithium Battery Health Status and Prediction of Remaining Useful Life Based on Improved PINN

With the rapid development of new energy vehicles, lithium batteries have gradually become the mainstream choice for their power systems due to their excellent performance. However, during long-term usage, lithium-ion batteries are bound to age and malfunction inevitably. To ensure their performance and safety, accurately estimating the state of health (SOH) and remaining useful life (RUL) of batteries has become an important research topic. This paper proposes a method based on an improved physics-informed neural network (PINN) for SOH estimation and RUL prediction, which can integrate physical models and data-driven models. By increasing the depth of the PINN expression model and using residual connections to improve it, the accuracy is further enhanced while avoiding the impact of gradient vanishing on the model. For the SOH estimation task, experiments are conducted using both explicit physical models and deep hidden physical models (DeepHPM). For RUL prediction tasks that lack explicit physical models, experiments are conducted using DeepHPM to mine potential physical models from the data. Experimental verification is carried out using the MIT-Stanford dataset, and the results show that the proposed method exhibits superior performance in both health state estimation and remaining useful life prediction.