Lithium-Ion Battery State of Health Estimation Using Resampling-Based Data Simplification Deep Learning Techniques

Abstract

Lithium-ion batteries (Li-ion) are widely used in various applications due to their high efficiency and reliability. However, as these batteries are continuously used, their performance gradually degrades over time, making accurate estimation of the State of Health (SOH) of increasing significance. One of the key challenges in SOH estimation lies in the nature of the measurement data collected during charge and discharge processes, which is typically time-series data with large volume and complexity. This results in increased computational load and reduced efficiency of the estimation models. To address this issue, this research proposes a data simplification method using a resampling technique aimed at identifying the optimal sampling level that maintains estimation accuracy while reducing computational cost. Four deep learning (DL) models Long Short-Term Memory (LSTM), Bidirectional Long Short-Term Memory (Bi-LSTM), Attention-based LSTM (ATT-LSTM), and Gated Recurrent Unit (GRU) are employed in this work. These models are trained and evaluated using public battery datasets that contain complete charge-discharge cycles. The proposed methods had been tested with the Oxford Battery Dataset, which the experimental results demonstrate that the proposed approach achieves higher estimation accuracy while significantly reducing computation time.