An Energy-Based Method for Predicting Battery Aging Using Newly Constructed Features in Fractal-Gradient-Enhanced LSTM Networks

As an effective way, the incremental capacity analysis (ICA) method has been widely adopted in the battery aging prediction field. Using the aging features from the ICA method, the success of long short-term memory (LSTM) networks in predicting battery aging has been widely reported. However, the extraction of aging features via the ICA method poses a challenge if filtering methods are not applied, which leads to the loss of the original information of the features. Meanwhile, it becomes extremely difficult to obtain aging features if the lithium-ion batteries (LIBs) are severely aged. To address these issues, this article proposes an energy-based feature method for predicting battery aging by training fractal-gradient-based LSTM networks. An energy-based feature method is proposed, and a new aging feature is proposed by incorporating the energy into the ICA method. A fractal-gradient-enhanced LSTM network is presented to efficiently use the proposed feature. Based on the definition of Hausdorff derivative, a fractal derivative with error information is proposed, and the convergence of the parameter in the proposed network is proved. Meanwhile, an adaptive order tuning method is proposed to enhance the performance of the proposed scheme. The experiments on different LIBs validate the effectiveness of the proposed scheme, with a satisfactory result.