Capacity Prediction Model of Lithium-ion Battery Based on 1/f Noise and Electrochemical Impedance Spectroscopy

Gaining a deep understanding of the degradation mechanisms and characteristics of lithium-ion batteries, as well as accurately predicting their remaining capacity, is crucial for ensuring the reliability of circuit system equipment and achieving real-time monitoring for lithium-ion battery safety.A lithium-ion battery remaining capacity prediction model based on 1/f noise and electrochemical impedance spectroscopy (EIS) was proposed. The analysis of EIS and 1/f noise can sensitively reflect internal degradation and defects. In terms of experiments, this project investigates an economical and simple self-built impedance spectroscopy measurement device for lithium-ion batteries. A battery cycling charge-discharge testing platform was constructed, and 65 cycles of charge-discharge aging tests were conducted on 8 14500 batteries, obtaining datasets of noise and impedance spectra during degradation. Regarding model construction, deep learning methods were employed to build convolutional neural networks (CNNs) based on single-type data and combined neural networks based on two types of data, utilizing collected EIS and ECN data to train the neural networks to obtain regression models. Using a random search algorithm, the hyperparameter combinations of the models were optimized, resulting in three types of lithium-ion battery remaining capacity prediction models based on battery failure process data. Through validation, it was found that the prediction models obtained through deep learning can accurately predict the remaining capacity of lithium-ion batteries. Additionally, prediction results obtained through deep learning using two types of data exhibit higher accuracy.