Validation and Verification of Remaining Useful Life Prediction for Lithium-Ion Batteries based on AM-CNN Approach

An empirical model decomposition algorithm, Elman neural networks, and long-term memory are employed to predict the remaining battery life. The method for deconstructing empirical models is given the data on battery capacity vs cycle number that has been acquired. After that, recurrent long short-term memory and Elman neural networks are used to make predictions for the high-frequency and low-frequency sub-layers, respectively. Data gathered at an unprecedented scale during battery testing has aided in model parameterization and performance evaluation. The first three steps of the proposed process are processing the data, choosing the features to use, and training the model. The proposed approach uses a Denoising Transformer (De-Transformer) network to clean up the raw data. The extended Kalman Particle Filter (EKPF) is a crucial sample density function that the proposed system employ in feature selection. AM-CNN is then used to train the models after the features have been chosen. The proposed method is superior to AM and CNN, two common substitutes.