Abstract
Accurate State of Charge (SOC) estimation for retired power batteries remains a critical challenge due to their degraded electrochemical properties and heterogeneous aging mechanisms. Traditional methods relying solely on electrical parameters (e.g., voltage and current) exhibit significant errors, as aged batteries experience altered internal resistance, capacity fade, and uneven heat generation, which distort the relationship between electrical signals and actual SOC. To address these limitations, this study proposes a surface temperature-assisted SOC estimation method, leveraging the distinct thermal characteristics of retired batteries. By employing infrared thermal imaging, key temperature feature regions-the positive/negative tabs and central area-are identified, which exhibit strong correlations with SOC dynamics under varying operational conditions. A Gated Recurrent Unit (GRU) neural network is developed to integrate multi-region temperature data with electrical parameters, capturing spatial-temporal thermal-electrical interactions unique to retired batteries. The model is trained and validated using experimental data collected under constant current discharge conditions, demonstrating superior accuracy compared to conventional methods. Specifically, our method achieves 64.3-68.1% lower RMSE than traditional electrical-parameter-only approaches (V-I inputs) across 0.5 C-2 C discharge rates. Results show that the proposed method reduces SOC estimation errors compared to traditional voltage-based models, achieving RMSE values below 1.04 across all tested rates. This improvement stems from the model's ability to decode localized heating patterns and their hysteresis effects, which are particularly pronounced in aged batteries. The method's robustness under high-rate operations highlights its potential for enhancing the reliability of retired battery management systems in secondary applications such as energy storage.