Exploring life warning solution of lithium-ion batteries in real-world scenarios: TCN-transformer fusion model for battery pack SOH estimation

Accurate estimation of the state of health (SOH) of lithium-ion batteries (LIBs) is critical for ensuring the operational safety of electric vehicles. While transformer-based models have shown promising results in SOH estimation, their performance on real-world datasets remains limited, particularly in capturing local temporal dependencies. To address this limitation, we propose a transfer learning framework that integrates a temporal convolutional network (TCN) with a transformer model for robust SOH estimation under real-world conditions. First, the actual battery capacity is calculated using an ampere-hour integration method enhanced by Monte Carlo simulation, which reduces estimation errors caused by outliers in the state of charge (SOC) data. Second, key features closely associated with battery SOH are extracted and utilized as inputs to the proposed TCN-Transformer model. This hybrid architecture combines the TCN's efficient parallel processing and long-sequence dependency modeling with the transformer's global feature extraction capabilities, enabling the discovery of deeper and more informative features. Finally, accurate SOH estimation is achieved through transfer learning between the National Aeronautics and Space Administration (NASA) dataset and the real-world vehicle dataset. Experimental results on dataset B1 demonstrate the effectiveness of the proposed approach, yielding a mean absolute error (MAE) of 0.8476 %. After transfer learning, the MAE value is reduced by 28.29 %, demonstrating enhanced estimation accuracy and practical applicability in real-world scenarios.