Lightweight AI-Based Battery Health Prediction for Electric Vehicles Using Temporal Attention and Multi-Feature Data

View at publisher Summarise with VIRAT
Authors: Chinmay Parikh, K. Rajakumar, Arjun Sreedhar, Shashwat Chaudhary, Sai Bhargav Kasetty
Cited by: 0
References: 14

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

The rapid adoption of EVs has made lithium-ion batteries the cornerstone of eco-friendly transportation. However, the highly accurate estimation of the SOH and RUL is crucial to their lifespan and performance. The use of state-of-the-art conventional methods based on deep learning and physics-based models has been restricted to real-time BMS because they either require intricate calculations or are not interpretable, respectively. This study describes the creation of a lightweight temporal degradation attention network that takes into account multiple features from the NASA battery degradation dataset in order to predict SOH and RUL in an effective and comprehensible manner. It has a Random Forest regression layer for prediction and a temporal attention encoder for identifying critical cycles in degradation. Temperature statistics, voltage range, capacity fade, and discharge and charge times are important features. High predictive performances are shown by$\mathbf{R 2} \boldsymbol{=} \mathbf{0. 9 9 8}$for both SOH and RUL, with MAEs of 0.006 and 7.13 for each, according to experimental evaluation using Group K-Fold cross-validation. The accuracy, interpretability, and computational efficiency of the suggested model are all balanced. It is therefore appropriate for embedded applications in EV BMS. Generally speaking, the TDAN + Random Forest hybrid framework allows for real-time, explicable battery health prediction by bridging the gap between data-intensive deep learning and lightweight machine learning.

Fields of Study & Subjects

Battery Modeling and Management Advancements in Battery Materials Electric and Hybrid Vehicle Technologies

Publication details

Journal / source
International Conference on Intelligent Data Communication Technologies and Internet of Things (IDCIoT)
Pages
204–209
Publication date
4 February 2026
Language
EN