TFDDA: a time-frequency deep domain adaptation method for cross-domain state of health estimation of lithium-ion batteries

Abstract Lithium-ion batteries have become an important part of the energy storage field. Accurately estimating their state of health (SOH) is a key precondition for ensuring the safe and efficient operation of the system. The performance of most SOH estimation methods based on data-driven generally relies on the assumption that the training data and the test data have the same distribution. However, in practical applications, differences in manufacturing processes, operating conditions, and battery types lead to varying degradation patterns and inconsistent data distributions among different batteries. This affects the estimation accuracy of such methods in cross-domain scenarios. To address this issue, a time-frequency deep domain adaptation (TFDDA) method is proposed. Specifically, the method constructs a dual-branch convolutional neural network that extracts and integrates the time-domain signal of the charging curve and the frequency-domain features transformed by the Fast Fourier transform (FFT) separately. It utilizes frequency-domain components with stronger domain invariance to compensate for the limitations of relying solely on time-domain features. This enables the model to automatically learn aging features with more discriminability and robustness. At the same time, maximum mean discrepancy (MMD) is employed to align feature distributions between source and target domains, thereby achieving cross-domain knowledge transfer. To verify the effectiveness of the proposed method, three experimental scenarios are designed: transfer within the same battery type under identical and different operating conditions, and transfer between different battery types. The experimental results demonstrate that the proposed method maintains good performance even as the data distribution differences gradually increase.