The early detection of degradation in lithium-ion batteries (LIBs) is crucial for effective predictive maintenance and recycling. However, accurately predicting the future degradation of LIBs in early stage is challenging due to the barely noticeable performance changes at initial charging cycles and the long-term nonlinear degradation pattern. In this work, we propose a two-stage early-stage degradation prediction method, BatteryGPT, which employs a Generative Pre-trained Transformer (GPT) to autoregressively predict the charging data of entire lifecycle and a state-of-health (SOH) estimator to correlates the predicted charging data with ageing features in LIBs. The validation demonstrates that BatteryGPT can predict the future LIB degradation with high accuracy using early charging data, before any capacity degradation is evident. Predicting with the first 30% of the battery lifetime, BatteryGPT significantly outperforms baselines, achieving a root mean square error (RMSE) of 0.213% for SOH variation prediction, and mean absolute percent errors (MAPE) of 2.30% and 1.18% for knee point and EOL predictions. Even predicting with the first 5% of lifetime charging data, BatteryGPT demonstrates strong early-stage prediction performance.

The mechanical coupling of multiple powertrain components makes the energy management of 4-wheel-drive (4WD) plug-in fuel cell electric vehicles (PFCEVs) relatively complex. Optimizing energy management strategies (EMSs) for this complex system is essential, aiming at improving the vehicle economy and the adaptability of operating conditions. Accordingly, a novel adaptive equivalent consumption minimization strategy (A-ECMS) based on the dragonfly algorithm (DA) is proposed to achieve coordinated control of the powertrain components, front and rear motors, as well as the fuel cell system and the battery. To begin with, the equivalent consumption minimization strategy (ECMS) with extraordinary instantaneous optimization ability is used to distribute the vehicle demand power into the front and rear motor power, considering the different motor characteristics. Subsequently, under the proposed novel hierarchical energy management framework, the well-designed A-ECMS based on DA empowers PFCEVs with significant energy-saving advantages and adaptability to operating conditions, which are achieved by precise power distribution considering the operating characteristics of the fuel cell system and battery. These provide state-of-the-art energy-saving abilities for the multi-degree-of-freedom systems of PFCEVs. Lastly, a series of detailed evaluations are performed through simulations to validate the improved performance of A-ECMS. The corresponding results highlight the optimal control performance in the energy-saving performance of A-ECMS.

The escalating environmental concerns and energy crisis caused by internal combustion engines (ICE) have become unacceptable under environmental regulations and the energy crisis. As a promising alternative solution, multi-power source electric vehicles (MPS-EVs) integrate various clean energy systems to enhance the powertrain efficiency. The energy management strategy (EMS) is plays a pivotal role for MPS-EVs to maximize efficiency, fuel economy, and range. Reinforcement Learning (RL) has emerged as an effective methodology for EMS development, attracting continuous attention and research. However, a systematic analysis of the design elements of RL-based EMS is currently lacking. This paper addresses this gap by presenting a comprehensive analysis of current research on RL-based EMS (RL-EMS) and summarizing its design elements. This paper first summarizes the previous applications of RL in EMS from five aspects: algorithm, perception scheme, decision scheme, reward function, and innovative training method. It highlights the contributions of advanced algorithms to training effectiveness, provides a detailed analysis of perception and control schemes, classifies different reward function settings, and elucidates the roles of innovative training methods. Finally, by comparing the development routes of RL and RL-EMS, this paper identifies the gap between advanced RL solutions and existing RL-EMS. Potential development directions are suggested for implementing advanced artificial intelligence (AI) solutions in EMS.

The advancement of electromechanical coupling technologies has fostered the development of four-wheel-drive plug-in hybrid electric vehicles (4WD PHEVs) with multiple power sources, potentially enhancing travel efficiency. Moreover, sophisticated energy management strategies (EMSs) not only augment the efficacy of energy-saving control but also ensure adaptability under varied driving conditions. In this paper, a learning-and-tube-based robust model predictive control (LTRMPC) strategy is proposed for a 4-wheel-drive PHEV (4WD PHEV). The suggested strategy enhances the economic efficiency of the intricate powertrain while preserving control robustness across diverse driving scenarios. Firstly, instead of utilizing a mathematical state predictive model to mirror state changes, a novel state observer is proposed, which uses a deep learning technique named Gated Recurrent Unit (GRU). The novel state observer boasts an enhanced feature fusion ability, thereby reflecting state changes accurately within the predictive horizon. Secondly, to mitigate the negative effects of state observation errors on control outcomes, a tube-based cost function is integrated into the learning-MPC framework to restrain the state changes into a certain range to further reinforce the control robustness. Finally, a simulation evaluation and hardware-in-the-loop (HIL) test validate that the proposed method can improve economic performance across various lengths of the predictive horizon and the energy-saving capability is stable compared with other baselines, showcasing its promising performance.

Existing data-driven technology for prediction of state of health (SOH) has insufficient feature extraction capability and limited application scope. To deal with this challenge, this paper proposes a battery SOH prediction model based on multi-feature fusion. The model is based on a convolutional neural network (CNN) and a long short-term memory network (LSTM). The CNN can learn the cycle features in the battery data, the LSTM can learn the aging features of the battery over time, and regression prediction can be made through the full-connection layer (FC). In addition, for the aging differences caused by different battery operating conditions, this paper introduces transfer learning (TL) to improve the prediction effect. Across cycle data of the same battery under 12 different charging conditions, the fusion model in this paper shows higher prediction accuracy than with either LSTM and CNN in isolation, reducing RMSPE by 0.21% and 0.19%, respectively.