Capacity Prediction Method for Ternary Lithium Batteries Based on Low-Frequency Noise

This study analyzes the aging mechanisms of lithium batteries and characterizes the aging properties through low-frequency noise analysis. By combining these findings with neural network algorithms, it achieves accurate predictions of battery capacity. Initially, a low-frequency noise experimental analysis platform for ternary lithium batteries was established, and a dataset of low-frequency noise under battery cycle aging was created through accelerated aging tests. Seven statistical features were extracted: mean, standard deviation (STD), skewness, kurtosis, autocorrelation, fitting error, and the square of the fitting errors. A predictive model was then developed using a one-dimensional convolutional neural network (1D-CNN) to extract dynamic characteristics from the low-frequency noise time-domain signals. Finally, the model was trained using conventional battery parameters and low-frequency noise features to predict the capacity of ternary lithium batteries and was subsequently validated. The results indicate that the time-domain features of low-frequency noise are correlated with battery capacity, and combining these features can enhance the predictive accuracy of the model.