Data driven-based health prognostics and charge estimation for lithium-ion batteries under varying discharging patterns

With the widespread adoption of lithium-ion batteries in energy storage and power systems, ensuring their stability and safety has become a critical concern. Herein, this study presents an efficient framework for accurately estimating the state of health (SOH) and state of charge (SOC) of lithium-ion batteries under complex conditions such as varying external circumstances and indiscernible internal response. Initially, to achieve high-precision SOH estimation, health features (HFs) are extracted from sensor data during the constant current charging process based on incremental capacity analysis (ICA) and are further identified through correlation analysis. Additionally, this study proposes a novel data-driven model for battery state estimation, namely Multi-scale Channel Attention Network with Adaptive Denoising filter (ADMCAN), which is built upon two components: Adaptive Spectral Block (ASB) to reduce high-frequency noise while highlighting crucial information in the frequency domain; and Multi-Scale Channel Attention (MSCA) block tailored to capture temporal dependencies at different scales. Two battery datasets under different chemistries and working conditions are utilized to verify the effectiveness of the proposed method. Experimental results show that the proposed method achieves superior predictive accuracy, enhanced robustness, and greater efficiency in both SOH and SOC estimation tasks compared to existing methods.