To Charge or to Sell? EV Pack Useful Life Estimation via LSTMs, CNNs, and Autoencoders

TL;DR

This work tackles the practical problem of estimating the remaining useful life of Li-ion battery packs in EVs by introducing an ampere-hour–based $RUL$ definition ($ah$-$RUL$) and evaluating deep learning pipelines that rely only on measurable inputs. It compares an autoencoder-based feature extractor followed by CNN or LSTM predictors (NASA Randomized data) with a plain LSTM predictor (UNIBO Powertools), validating across two diverse datasets to ensure generalization. The approach achieves robust performance, with autoencoder reconstructions yielding an RMSE of about 0.0356 and ah-$RUL$ predictors reaching RMSEs in the 0.07–0.08 range on NASA data, and about 0.021 on UNIBO data, demonstrating practical applicability for BMS and second-life battery reuse. The findings support more accurate residual value estimation and safer, more sustainable battery management, while outlining future directions such as transformer-based models and integrated policy considerations for recycling.

Abstract

Electric vehicles (EVs) are spreading fast as they promise to provide better performance and comfort, but above all, to help face climate change. Despite their success, their cost is still a challenge. Lithium-ion batteries are one of the most expensive EV components, and have become the standard for energy storage in various applications. Precisely estimating the remaining useful life (RUL) of battery packs can encourage their reuse and thus help to reduce the cost of EVs and improve sustainability. A correct RUL estimation can be used to quantify the residual market value of the battery pack. The customer can then decide to sell the battery when it still has a value, i.e., before it exceeds the end of life of the target application, so it can still be reused in a second domain without compromising safety and reliability. This paper proposes and compares two deep learning approaches to estimate the RUL of Li-ion batteries: LSTM and autoencoders vs. CNN and autoencoders. The autoencoders are used to extract useful features, while the subsequent network is then used to estimate the RUL. Compared to what has been proposed so far in the literature, we employ measures to ensure the method's applicability in the actual deployed application. Such measures include (1) avoiding using non-measurable variables as input, (2) employing appropriate datasets with wide variability and different conditions, and (3) predicting the remaining ampere-hours instead of the number of cycles. The results show that the proposed methods can generalize on datasets consisting of numerous batteries with high variance.

Figures (5)

• Figure S1: Classification of battery SOH estimation methods.
• Figure S2: The structure of the autoencoder used to compress the cycles. The skip link allows us to retain both local and global information
• Figure S3: Example of the results of the autoencoder reconstruction capabilities on the testing set. The voltage, current, and temperature curves of a cycle in the battery's middle life were reconstructed from the extracted features.
• Figure S4: A side-by-side example of the results of the CNN and the LSTM in ah-RUL prediction on the NASA Randomized dataset. The two examples show better stability and curve fitting of the CNN but a lower offset from the LSTM.
• Figure S5: An example of test results of the LSTM ah-RUL prediction on the UNIBO Powertools dataset.