Electric vehicle battery prognostics using an attention-based deep learning framework with Harris Hawks optimization

Electric vehicles are increasingly central to sustainable transportation systems, relying on lithium-ion battery (LiB) technology as the dominant electrochemical energy storage solution for effective energy storage and utilization. Reliable assessment of battery state-of-health (SOH) and remaining useful life (RUL) is therefore essential for battery management systems to ensure safe operation, optimal utilization, and effective lifecycle management of lithium-ion batteries (LiBs). In this paper, a hybrid deep learning framework is introduced to enhance the efficiency of LiB prognostics. This framework incorporates convolutional neural networks (CNN), a multi-head attention (MHA) module, and bidirectional long short-term memory (BiLSTM) units, collectively referred to as Convolutional attention-based BiLSTM with Harris Hawks Optimization (CABL-HHO). Furthermore, the Harris Hawks Optimization (HHO) method is adopted to jointly optimize the architectural configuration and hyperparameter set of the CABL-HHO framework. Extensive experimental evaluations are performed using the publicly available NASA LiB dataset, which captures long-term degradation behavior. The results demonstrate that the CABL-HHO achieves improved prediction accuracy compared to benchmark models, leading to more reliable operation of LiB management systems. • A hybrid deep learning model is developed for battery prognostics. • The framework integrates CNN, multi-head attention, and BiLSTM modules. • Harris Hawks Optimization (HHO) jointly tunes network architecture and hyperparameters. • The attention mechanism enhances discriminative temporal–spatial feature extraction. • Experimental results on NASA's dataset outperform existing benchmark models.