Retired battery sorting method via feature visualization and machine learning under partial charging voltage segments

With the rapid development of the electric vehicles, the echelon utilization of retired power batteries has become a current research hotspot. A retired battery sorting method based on Gramian angle difference fields (GADF) and Swin Transformer models is proposed to address the issues of poor sorting accuracy, low efficiency, and complex feature engineering in the echelon utilization of retired batteries. Part of the charging voltage curve is converted into images using the GADF, and the images are classified using the Swin Transformer model. Firstly, we extract voltage segments from the constant current charging stage to obtain voltages above 3.9V. Then, we perform Piecewise Aggregate Approximation (PAA) data dimensionality reduction on some voltage segments and encode the voltage information into the image using GADF method. Finally, Swin Transformer is used to classify the transformed image. Due to the existence of a shifted window, this algorithm can integrate information between different patches, making the resulting image contain more comprehensive and authentic information. Finally, we validated and compared the effectiveness of different retired battery sorting methods on the modified NASA dataset, and the results showed that the proposed method achieved an accuracy of 97.67% for retired battery sorting, achieving high precision and efficiency.