Accurate Remaining Useful Life Prediction of Electric Vehicle Lithium-Ion Batteries Using an Enhanced Hybrid Fourier Neural Operator Convolutional Transform Network

Abstract

As a result of their high-energy density, efficiency and decreasing cost, lithium-ion batteries (LIBs) have emerged as the energy storage technology of choice for electric vehicles (EVs). But precisely estimating RUL is still a significant challenge for guaranteeing battery dependability, security and optimal utilization. This work aims to overcome these drawbacks by developing an enhanced hybrid predictive model using a Fourier Neural Operator–tuned Convolutional Transform Network (FNO-CTN) for precise RUL estimation and assessment of battery health. The model proposed takes advantage of the ability of FNO to learn from global degradation patterns in the frequency domain, and the CTN to learn from locally spatial-temporal dependency characteristics through the information of battery operational data. By integrating FNO into CTN, the nonlinear behavior of the battery can be improved modeled, and the long-term degradation trends can be modeled as well. This EV Battery RUL Health Dataset, with 9200 records, supplies the model with key operating parameters such as voltage, current and temperature. A comprehensive preprocessing pipeline involving min-max normalization and outlier removal is employed to improve data quality and robustness, and feature extract is used by principal component analysis (PCA) to reduce dimensionality and extracting significant features. Moreover, efficient hyperparameter tuning is achieved by integrating Wolf Pack Search (WPS) into the algorithm, which supports faster convergence, stability and generalization under various operating conditions. The research employed Python 3.10, and typical performance measures such MAE at 0.001298, RMSE at 0.001897, Parameters at 0.29, and Power at 70.86 are utilized to assess the efficacy of the FNO-CTN. Experimental results demonstrate that the proposed FNO-CTN significantly outperforms conventional deep learning and hybrid models in prediction accuracy and robustness.