Knowledge-Guided Data-Driven Model With Transfer Concept for Battery Calendar Ageing Trajectory Prediction

Dear Editor, Lithium-ion (Li-ion) battery has become a promising source to supply and absorb energy/power for many energy-transportation applications. However, Li-ion battery capacity would inevitably degrade over time, making its related ageing prediction necessary. This letter presents effective battery calendar ageing trajectory prediction by deriving a knowledge-guided data-driven model with transfer concept. More specifically, this data-driven model is based on the support vector regression (SVR) technology. To ensure highly-accurate prognostics of battery calendar ageing trajectory under wit-nessed conditions, a knowledge-guided kernel is first developed by coupling the mechanism and empirical knowledge elements of battery storage temperature, state-of-charge (SoC), and time. To im-prove model's generalization ability under unwitnessed conditions, the knowledge-guided data-driven model is then equipped with trans-fer concept by adding a classical Gaussian kernel for all inputs. A well-rounded real battery ageing dataset under eight different storage conditions is collected to evaluate the performance of developed model. Results illustrate that this knowledge-guided battery ageing trajectory prediction model presents satisfactory accuracy for wit-nessed conditions with R2 over 0.98. After using only 20% starting capacity point to tune its transfer part, it can also generalize well for unwitnessed conditions with R2 over 0.97, further heavily reducing the required ageing experimental time and cost.

Automated summary

This letter presents an effective battery calendar ageing trajectory prediction model based on support vector regression (SVR) technology, which combines the mechanism and empirical knowledge elements of battery storage temperature, state-of-charge (SoC), and time. The model achieves highly accurate predictions for witnessed conditions and also demonstrates good generalization ability for unwitnessed conditions, reducing the required experimental time and cost.