Data-driven SOC estimation method for power batteries under driving cycle conditions and a wide temperature range

Accurate estimation of the state of charge (SOC) in lithium-ion batteries is essential for evaluating the driving range of electric vehicles and ensuring the reliable operation of battery management systems (BMSs). However, large-scale datasets encompassing diverse operating conditions remain scarce, while publicly available datasets are often fragmented and restricted to specific chemistries or scenarios. To overcome this limitation, we construct a comprehensive multi-condition dataset covering a wide range of temperatures, battery types, and driving cycles. Building on this dataset, we propose a fast and high-precision SOC estimation method tailored to real-world automotive conditions. The method leverages a novel deep learning architecture that integrates a convolutional neural network (CNN), an enhanced parallel residual temporal convolutional network (PRTCN), and a squeeze-and-excitation (SE) module. The CNN extracts short-term local features from time-series data, the PRTCN captures long-term temporal dependencies, and the SE module adaptively enhances feature representation, thereby improving overall model performance. Extensive validation under varying temperatures and ten dynamic load profiles demonstrates that the proposed method achieves a maximum absolute estimation error below 2 %, with inference times on the order of milliseconds. These results highlight the advantages of the method in terms of accuracy, efficiency, and practical applicability, providing strong technical support for SOC estimation in electric vehicle BMSs.