Predicción de la salud de la batería basada en aprendizaje profundo para mejorar el rendimiento del vehículo...

Reliable and sustainable battery diagnostics are essential for advancing electric vehicle (EV) technologies and fulfilling Sustainable Development Goal 7 (SDG 7): Affordable and Clean Energy. This study proposes a hybrid deep learning framework that integrates one-dimensional Convolutional Neural Networks (1D-CNNs), Temporal Convolutional Networks (TCNs), and Long Short-Term Memory (LSTM) layers, along with an attention mechanism, for intelligent EV battery health diagnostics. Differential Voltage (dV/dQ), Differential Current (dI/dV), and Incremental Capacity Analysis (ICA, dQ/dV) features were extracted and denoised from over 10,000 charge-discharge cycles sourced from the NASA PCoE, Oxford, and CALCE battery degradation datasets. The proposed model achieved a State-of-Health (SOH) prediction accuracy of [Formula: see text] and RMSE = 0.021, outperforming conventional CNN, LSTM, and XGBoost baselines by up to 25% in accuracy. With only 0.35 million parameters, the model demonstrated an average inference latency of 6.1 ms and energy consumption of 0.63 mJ per sample-a 27% reduction compared to Transformer-based architectures. These results confirm the framework's robustness, scalability, and real-time feasibility for embedded Battery Management Systems (BMS). By improving diagnostic precision, extending battery lifespan, and reducing computational energy demand, the proposed method directly contributes to sustainable mobility and the broader goals of energy efficiency and the circular economy under SDG 7.