Key result
Deep learning models improve EV battery multi-state estimation, achieving RMSE as low as ~0.03%.
Why the study?
The increasing demand for improved EV battery performance, longevity, and safety necessitates the development of sophisticated multi-battery state estimation methods.
Deep learning models, particularly LSTM-based architectures, outperform traditional methods in accurately estimating electric vehicle battery states.
Supports DL models for EV battery estimation; leaves open prospective validation before real-time deployment.
The increasing requirement for lithium-ion batteries (LIBs) owing to their use in electric vehicles shows how crucial it is to have a reliable battery management system (BMS). This paper presents a methodological review of estimating the multi-battery states of EV batteries. The increasing demand for improved EV battery performance, longevity, and safety necessitates the development of sophisticated estimation methods. This study delves into multiparameter estimation methods, including state-of-charge, state-of-health, and state-of-energy. This review critically compares traditional, machine learning, and deep learning approaches, revealing the superiority of deep learning models in capturing the nonlinear battery dynamics and long-term temporal dependencies. The challenges and prospects of this rapidly evolving field are also discussed, emphasizing the critical role of estimation techniques in optimizing the overall performance and reliability of EV battery systems. This study further investigates the effectiveness of various algorithms, including Support Vector Machines (SVM), Neural Networks (NNs) such as Feedforward Neural Networks (FNN) and Recurrent Neural Networks (RNNs) such as Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU), neural networks with convolution (CNNs), and ensemble methods. The findings indicate that sophisticated deep learning platforms, including LSTM, CNN-LSTM, and cluster-based LSTM, consistently exhibit enhanced performance, characterized by low RMSE and MAE values. Methods such as PSO-LSTM, RS-LSTM, and LSTM-RFR2 significantly improved accuracy, attaining RMSE values as low as 0.0262%. Although these methods demonstrate encouraging outcomes, obstacles persist, including the need for appropriate tuning of hyperparameters and the requirement for extensive, superior datasets. Future research avenues include investigating hybrid models that combine data-driven strategies using physics-based mathematical models, creating reliable techniques for managing noise in addition to uncertainty, and addressing the difficulties of real-time application.
No takes yet. Share an insight, caveat, or question.
Kondru et al. (2026) conducted a review in Electric vehicle batteries. Machine learning and deep learning approaches vs. Traditional approaches was evaluated on Multi-battery states estimation (state-of-charge, state-of-health, state-of-energy). Deep learning models, including LSTM and CNN-LSTM, consistently exhibited enhanced performance for multi-state estimation of EV batteries, attaining RMSE values as low as 0.0262%.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: