AI-driven battery ageing prediction with distributed renewable community and E-mobility energy sharing

Electrochemical battery storages play multiple functions in district energy systems, like peak shaving, load coverage, renewable penetration, load coverage, frequency regulation and short-term fast response. Battery cycling ageing will significantly affect the battery performance and the ignorance or inaccurate estimation on battery cycling ageing will lead to performance overestimation. However, current literature provides limited progress on these topics. In this study, a coastal district energy community is developed with district buildings, transportations, solar-wind systems, electrochemical battery storages, power grid and multi-directional power interactions. The ultimate target is to predict the dynamic state of health of various batteries with high accuracy and efficiency. In respect to frequent charging from intermittent renewables and discharging from stochastic demands and vehicle driving s, a series of cycling ageing models are developed, tested and compared, using mathematical fitting and machine learning approaches. Considering modelling complexity, time-consuming and product-dependent characteristic of mathematical models, machine learning-based regression learners are applied to firstly learn the underlying ageing mechanism, and then make accurate predictions. A top-down approach, following statistical analysis in training, cross-validation and prediction processes, is proposed for algorithm selection to accelerate the cycling ageing prediction. Results indicate that, during training and cross-validation processes, the final algorithm is identified as the exponential Gaussian Process Regression. Demonstration on battery cycling ageing verifies the feasibility and effectiveness in different battery categories and energy paradigms, so as to avoid techno-economic performance overestimation.