An optimized ensemble learning framework for lithium-ion Battery State of Health estimation in energy storage system

Battery State of Health (SOH) is critical for the reliable operation of the grid-connected battery energy storage systems. During the long-term Lithium-ion (Li-ion) battery degradation, large amounts of data can be recorded. Unfortunately, massive raw data are naturally with different qualities, which makes it difficult to guarantee the superior performance of one unified and powerful data driven estimator. Thus, this paper proposes a novel ensemble learning framework to estimate the battery SOH, which can boost the performance of the data driven SOH estimation through a well-designed integration of the weak learners. Moreover, the short-term current pulses, which are convenient to be obtained from real applications, act as the deterioration feature for SOH estimation. To establish the weak learners with good diversity and accuracy, support vector regression is chosen to utilize the measurement from a specific condition. A Self-adaptive Differential Evolution (SaDE) algorithm is used to effectively integrate the weak learners, which can avoid the trial and error procedure on choosing the trial vector generation strategy and the related parameters in the traditional differential evolution. For the validation of the proposed method, two LiFePO₄/C batteries are cycling under a mission profile providing the primary frequency regulation service to the grid.