Deep learning-based battery health prediction for enhancing electric vehicle performance

Area of Science:

• Electrical Engineering
• Computer Science
• Materials Science

Background:

• Reliable battery diagnostics are crucial for electric vehicle (EV) advancement and Sustainable Development Goal 7 (SDG 7).
• Current diagnostic methods face challenges in accuracy and efficiency for complex battery degradation patterns.

Purpose of the Study:

• To develop a hybrid deep learning framework for intelligent EV battery health diagnostics.
• To improve State-of-Health (SOH) prediction accuracy and efficiency for Battery Management Systems (BMS).

Main Methods:

• A hybrid deep learning model integrating 1D-CNN, TCN, and LSTM with an attention mechanism was proposed.
• Differential Voltage (dV/dQ), Differential Current (dI/dV), and Incremental Capacity Analysis (ICA, dQ/dV) features were extracted and denoised.
• The model was trained and validated on extensive battery degradation datasets (NASA PCoE, Oxford, CALCE).

Main Results:

• The hybrid model achieved high SOH prediction accuracy ([Formula: see text]) and low RMSE (0.021).
• It outperformed baseline models (CNN, LSTM, XGBoost) by up to 25% in accuracy.
• The model demonstrated low parameter count (0.35 million), fast inference (6.1 ms), and reduced energy consumption (0.63 mJ/sample), outperforming Transformer architectures.

Conclusions:

• The proposed framework offers a robust, scalable, and real-time feasible solution for embedded BMS.
• It enhances diagnostic precision and extends battery lifespan, contributing to sustainable mobility.
• The method supports energy efficiency and circular economy goals under SDG 7 by reducing computational demands.