Xiaosong Hu’s research while affiliated with Chongqing University and other places

Deep reinforcement learning (DRL) algorithms have demonstrated impressive performance in developing optimal energy management strategies (EMSs) for individual hybrid electric vehicles (HEVs) under predefined driving cycles. However, in this area of research, the impact of thermal loads and thermal management (TM) is often overlooked. Moreover, HEVs may encounter unseen driving patterns that can hinder the overall performance of EMS. Connected HEVs (C-HEVs) show promising solutions; however, there are existing issues such as privacy, security, and communication loads. This paper proposes a novel integrated thermal and energy management (ITEM) approach based on federated reinforcement learning (FRL) for achieving a generalized policy across multiple C-HEVs. This framework broadens learning from multiple environments while preserving local HEV data privacy and security. The proposed FRL algorithm is iteratively executed between multiple HEVs and a cloud-based center to develop global policies for all ITEMs. For each ITEM, two DRL agents (cabin TM and EMS) build their local policies based on recorded driving data. The only local and global models exchanged between the cloud-based center and the ITEMs reduce communication overhead and preserve driving data privacy. Our findings successfully demonstrate that this approach has the advantage of accelerating convergence speed and achieving total rewards similar to the DRL strategy, which has access to driving cycle information in advance. Furthermore, we demonstrate that the proposed approach delivers excellent performance even when additional DRL agents join the FRL network. The implementation capability is also verified by a hardware-in-the-loop (HIL) test setup.

Energy storage and management technologies are key in the deployment and operation of electric vehicles (EVs). To keep up with continuous innovations in energy storage technologies, it is necessary to develop corresponding management strategies. In this Review, we discuss technological advances in energy storage management. Energy storage management strategies, such as lifetime prognostics and fault detection, can reduce EV charging times while enhancing battery safety. Combining advanced sensor data with prediction algorithms can improve the efficiency of EVs, increasing their driving range, and encouraging uptake of the technology. Energy storage management also facilitates clean energy technologies like vehicle-to-grid energy storage, and EV battery recycling for grid storage of renewable electricity. We offer an overview of the technical challenges to solve and trends for better energy storage management of EVs. (This is a read-only link to the full-text PDF: https://rdcu.be/d8xK1)

Safe and efficient charging of lithium-ion batteries is crucial for the economic viability and convenience of electric flying cars in urban air mobility applications. This article proposes a multi-objective charging optimization framework for large-format batteries based on a physics-data enhanced battery electrothermal-degradation model, a charging economic model, and a battery safety model. Specifically, the coupled battery model emulates the battery's electrical, thermal, and degradation behavior during charging. The economic model evaluates the equivalent cost of battery degradation and the charging electricity cost. The safety model quantifies the battery's state of safety and imposes the constraints for safe charging. The multi-objective charging optimization was achieved with an improved dynamic multi-swarm particle swarm optimization algorithm. Four typical charging strategies were investigated in detail, and the potential influencing factors of optimal battery charging were analyzed. Numerous simulation results indicate that the proposed balanced charging strategy, which achieves the expectant charging goal within 10 min, mitigates degradation by 12 %, reduces cost by 5.56 %, and improves safety by 12.37 % compared to the minimum-time charging.

Accurate, practical, and robust evaluation of the battery state of health is crucial to the efficient and reliable operation of electric vehicles. However, the limited availability of large-scale, high-quality field data hinders the development of the battery management system for state of health estimation, lifetime prediction, and fault detection in various applications. In this work, to gain insights into underlying factors limiting battery management system performance in real-world vehicles, we analyze the operational data of 300 diverse electric vehicles over three years to understand the disparities between field data and laboratory battery test data and their effect on state of health estimation. Furthermore, we propose a deep learning-based multi-modal framework to effectively leverage historical vehicle data for efficient, accurate, and cost-effective state of health estimation. The proposed paradigm exhibits considerable potential for numerous applications in state estimation and diagnostics in multi-sensor systems. Furthermore, we make the field data of these electric vehicles publicly available aiming to promote further research on the development of effective and reliable battery management systems for real-world vehicles.

Tracking the battery discharge capacity is significant, yet challenging due to complicated degradation patterns as well as varying or even random usage scenarios. This work proposes a physics‐constrained domain adaptation framework to predict the capacities during random discharge with non‐destructive mechanism diagnosis using early or random discharging information. By imposing the impedance as physical constraints in a domain adaptative layer, the interpretability and generalization of the model can be improved as the physics‐constrained layer provides physical insights into the battery mechanism characteristics, enabling onboard and non‐destructive diagnosis without complex tests. The learned impedances in the physics‐constrained layer are well‐fitted to the real ones, suggesting accurate physical insights and, therefore, good interpretability of the trained model. Furthermore, apart from capacity prediction, the aging mechanisms of the cell can be interpreted through the learned physics from this deep learning framework without impedance measurement. Such interpretation has also been validated experimentally through post‐mortem analysis. This work provides an example of grey‐box modeling of complex dynamic systems where deep learning models can provide certain physical details to increase the model's interpretability.