State-of-health estimation for lithium-ion batteries with hierarchical feature construction and auto-configurable Gaussian process regression

State-of-Health (SOH) estimation is crucial for the safety and reliability of battery-based applications. Data-driven methods have shown their promising potential in battery SOH estimation, yet creating a high-performance model with a compact structure is still a grand challenge. This paper focuses on constructing the elastic feature to formulate auto-configurable Gaussian Process Regression (GPR) to address this issue. To eliminate the impacts of the kernels on GPR, an evolutionary framework is designed to organize the kernel configuration. Meanwhile, a hierarchical feature construction strategy reduces the complexity of the extracted feature according to the geometry of the charging curve. Experiments on three battery datasets demonstrate the effectiveness of the proposed method, demonstrating the practical value of the proposed method for the battery management system (BMS) to construct feature more feasible, and to provide the optimal kernel configuration automatically.