Integrated framework for battery SOH estimation using incremental capacity and image feature transformation
Ping Ding, Taotao Li, Yajun Qiao, Linfeng Zheng, Hui Deng, Weixiong Wu
Abstract
As the core energy source, Li-ion batteries display intricate aging patterns, posing significant challenges for systematic analysis and precise capacity estimation. The incremental capacity (IC) method is a prevalent technique for assessing battery state of health (SOH). This paper proposes a comprehensive SOH estimation algorithm that integrates IC features with image conversion techniques. The estimate model begins by converting one-dimensional raw IC data into two-dimensional images using the gramian angular field (GAF) algorithm. Subsequently, a CNN-LSTM model extracts intricate features from these images, which are then fed into an XGBoost estimator. Comparative experiments explore the impact of three distinct CNN architectures on SOH estimation performance. Additionally, the paper evaluates the reinforcement learning module to assess its accuracy and robustness in SOH estimation. The results highlight a significant reduction in average RMSE by 68.2% with the addition of the XGBoost module, achieving an estimated RMSE of 1.76% on the NASA dataset. Similarly, for the CALCE dataset, the model achieves an RMSE of less than 1.5% under consistent charge rate between training and test sets, and less than 2.7% when charge rate vary. These findings underscore the effectiveness of the proposed algorithm in enhancing the precision of SOH estimation for lithium-ion batteries across varied operational conditions. • A 1D incremental capacity curve is transformed into a 2D time-series image. • A CNN-LSTM-XGBoost model is used for feature extraction and estimation. • Model accuracy and generalization are evaluated using two distinct battery datasets. • The impact of various CNN architectures on model prediction is compared.