Fuzzy Entropy-Based State of Health Estimation of LiFePO₄Batteries Considering Temperature Variation

Sample entropy (SE) has been used as a feature to estimate the state of health (SOH) of batteries as it can capture the voltage variation during battery degradation. However, because the Heaviside function is used to access similarity in the definition of SE, SE is sensitive to parameter selection. Hence, the SE shows an obvious change when the battery is aged at different conditions (e.g., temperatures), leading to a decrease in the estimation accuracy. By introducing the concept of fuzzy membership, the generalized version of SE, fuzzy entropy (FE) is weak influenced by parameters and test condition. Therefore, the FE is proposed as a feature to estimate the SOH of battery in terms of aging temperature variation. The FE-SOH is used as the input-output data pair of support vector machine, then the single-temperature model, full temperature model, and partial-temperature model are established. Compared with the SE-based method, FE-based method not only has better estimation accuracy, but also decreases the dependence on the size of training data. Finally, the effectiveness of the proposed method is verified using experimental results.