An REAE-LTransformer model for lithium-ion battery RUL prediction based on IFVIM

<title>Abstract</title> <p>Accurate prediction of the remaining useful life (RUL) of lithium-ion batteries is critical for battery safety and reliability. Existing models still suffer from inadequate feature extraction and suboptimal hyperparameter selection, limiting prediction accuracy and generalization. Therefore, a lithium-ion battery RUL prediction model is proposed by integrating a Residual-Enhanced Autoencoder (REAE), a lightweight Transformer (LTransformer), and an Improved Four-Vector Intelligent Metaheuristic (IFVIM) algorithm. Three innovations are featured in this framework: Firstly, a Residual-Enhanced Autoencoder (REAE) is designed to effectively extract stable degradation features from complex noise. Second, LTransformer captures long-term temporal dependencies with reduced computational overhead, improving prediction accuracy and cross-battery generalization. Finally, an IFVIM optimization module is proposed, wherein the global search and adaptive adjustment of key hyperparameters are conducted through parameter space optimization, and prediction errors are thereby effectively minimized. Under all testing scenarios, a Mean Absolute Error of less than 0.024 is achieved, and a coefficient of determination (R²) exceeding 0.99 is attained. Compared with mainstream models, the average Root Mean Square Error on the UNIBO and CALCE datasets is reduced by 20.3% and 47.1%, respectively. Experimental results demonstrate that the proposed model effectively captures battery degradation dynamics and achieves superior prediction performance.</p>