Multi-stage Battery Degradation Modelling and Prediction with Uncertainty Estimation

Accurate estimation of battery capacity degradation can benefit a range of stakeholders, including Battery Energy Storage Systems (BESS) developers, investors, and asset operators. However, modelling the degradation behaviours of lithium-ion batteries is challenging, particularly when the batteries show a two-phase distinct degradation trend under certain operating conditions. This behaviour makes it even harder for forward forecasting within available test points, as a small difference with the model prediction for existing points can result in a large discrepancy in future predictions. In this paper, we propose a multi-stage modelling method that combines artificial intelligence (AI) and semi-empirical approaches to model this challenging degradation trend in a piecewise manner. Furthermore, we analyse the uncertainty both from curve fitting and model prediction to give the confidence interval. Results on experimental dataset show that the proposed method can successfully model the battery degradation trend with accelerated trajectories.