A method for battery state-of-charge estimation without using dynamic models

Accurate state of charge (SOC) estimation is essential for reliable battery management systems. Traditional model-based approaches often suffer from both structural limitations and parameter mismatch under complex operating conditions and dynamic environments, compromising estimation accuracy. This study proposes a data-driven approach that reformulates SOC estimation as a classification problem rather than a regression task. A convolutional neural network (CNN) is trained to compare the similarity of two groups of given data in the sense of SOC similarity. The well-trained CNN is then used in the testing phase to compare the unlabeled data with data labeled with known SOC. In this way, battery SOC can be obtained without using traditional dynamic models that describe the input–output relationship of a battery. Experimental validation shows that the proposed approach achieves SOC estimation errors below 3.0% for different battery types in a wide range of aging (100%–70%) and temperature (5 °C–45 °C). This strategy significantly reduces dependency on model accuracy while improving robustness and generalization in real-world operating scenarios.