Intelligent spiking GAN-residual framework for accurate SOH, and RUL prediction in EV battery

Accurate estimation of State of Health (SOH) and Remaining Useful Life (RUL) of batteries in Electric Vehicles (EVs) is critical for enhancing energy efficiency, operational safety, and predictive maintenance. Therefore, this research propose an intelligent framework for reliable battery state estimation using advanced hybrid Deep Learning (DL) based Residual-SpikeGANet approach. Battery time series data are denoised using Discrete Wavelet Transform (DWT), which preserves degradation-related time–frequency information, while an AutoEncoder is employed for obtaining concise and insightful latent features from battery data. Piranha Foraging Optimization Algorithm (PFOA) is implemented for optimal feature selection based on computational efficiency and reduced size. Residual-SpikeGANet architecture integrates residual learning, Spiking Neural Networks (SNN), and Generative Adversarial Networks (GAN) for accurately capturing the nonlinear degradation behaviour of batteries and the long-term temporal dependencies associated with the degradation process. The novelty lies in integrating DWT-based denoising, AutoEncoder-based feature extraction, PFOA-based feature selection and ResidualSpikeGANet.-based prediction model. This proposed framework is implemented in Python, and evaluated using the NASA dataset, achieving a high estimation accuracy, as demonstrated by Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and R² values 0.01334, 0.01031, 0.99707 for SOH; and 0.02073, 0.01635, 0.99389 for RUL respectively.