CAREER: Intelligent Battery Management with Safe, Efficient, Fast-Adaption Reinforcement Learning and Physics-Inspired Machine Learning: From Cells to Packs

电动汽车(EVS)对于运输部门的脱碳至关重要。然而，仍然存在限制电动汽车更广泛采用的主要瓶颈，包括电池充电速度慢导致的里程焦虑以及与电池退化相关的安全问题。为了应对这些挑战，需要快速充电策略来减少充电时间，并且必须开发可靠的监测机制来提供电池健康状况的早期预测。该提案旨在通过开发具有新型强化学习(RL)和机器学习方法的智能电池管理系统来应对这些挑战。建议的方法将实现安全、高效和自适应的快速充电策略，以及可靠和准确的电池健康预测。该计划还将通过考虑电池组之间的相互作用和不一致性来促进电池组管理方面的知识。研究生和本科生的研究都将在这个项目中得到支持。该项目的研究成果将紧密结合到现有课程中，并为化学工程专业的学生创建一门新的机器学习课程。通过综合研究、教育和推广活动，该项目还将培训学生使用数据科学知识和编程技能来应对未来的工程挑战。该项目旨在应对与开发下一代智能电池管理系统相关的挑战。对于快速充电协议的优化，现有的方法往往依赖过于复杂的电池电化学模型，其中许多模型不能适应电池的运行条件。对于电池健康预测，如容量估计和使用寿命预测，目前的方法通常需要昂贵的人工特征提取。此外，对于电池组的快速充电和健康预测，必须考虑电池之间的相互作用和不一致性。这项研究将通过研究：(1)基于深度RL的方法，基于安全、分层和元RL，以实现安全和自适应的快速充电协议；(2)高效和基于物理信息的基于变压器的健康预测，包括用于容量估计和寿命预测的电池老化物理学；以及(3)将建议的方法扩展到电池组，同时考虑到单个电池之间的不一致性。将利用开源电池仿真平台和电池实验试验台来验证所提出的方法。为实现这些算法而开发的软件将向更广泛的受众公开，以进一步推动该领域的研究并培养下一代电池工程师。研究生和本科生的研究都将在这个项目中得到支持。推广活动，如将该项目的研究成果纳入化学工程课程，教育K-12学生决策，培训大学生编程，将广泛培养整个STEM管道的“数据思维”思维。这一奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

Electric vehicles (EVs) are critically important to the decarbonization of the transportation sector. However, major bottlenecks still exist that limit the wider adoption of EVs, including mileage anxiety due to the slow charging speed of batteries and safety concerns associated with battery degradation. To address these challenges, fast charging strategies are needed to reduce charging time and reliable monitoring mechanisms must be developed to provide early prognostics of battery health. This proposal aims to address these challenges by developing intelligent battery management systems with novel reinforcement learning (RL) and machine learning methods. The proposed methodologies will enable safe, efficient, and adaptive fast-charging strategies, as well as reliable and accurate health prognostics for batteries. This program also will advance knowledge on the management of battery packs by considering cell-to-cell interactions and inconsistencies. Both graduate and undergraduate student research will be supported in this project. Research results from this project will be tightly integrated into the existing curriculum and in the creation of a new machine learning course for chemical engineering students. Through integrated research, education, and outreach activities, this project also will train students to use data science knowledge and programming skills to meet future engineering challenges. This project aims to address challenges related to the development of the next-generation intelligent battery management systems. For the optimization of fast-charging protocols, existing methods often depend on overly complex battery electrochemical models, many of which fail to adapt to battery operating conditions. For battery health prognostics, such as capacity estimation and useful-life prediction, current methods often require expensive manual feature extraction. Moreover, cell-to-cell interactions and inconsistencies must be considered for fast-charging and health prognostics of battery packs. This research will address these knowledge gaps by studying: (1) deep RL-based methods, based on safe, hierarchical, and meta RL, to enable safe and adaptive fast-charging protocols; (2) efficient and physics-informed transformer-based health prognostics that incorporate battery aging physics for capacity estimation and lifetime prediction; and (3) the extension of proposed methods to battery packs while considering the inconsistencies among individual cells. Both open-source battery simulation platforms and battery experimental testbeds will be employed to validate the proposed methods. Software developed in implementing these algorithms will be made publicly available for a broader audience to further advance research in this field and to educate next-generation battery engineers. Both graduate and undergraduate student research will be supported in this program. Outreach activities, such as incorporating research findings of this project into the chemical engineering curriculum, educating K-12 students about decision-making, and training college students about programming, will widely cultivate a “data thinking” mindset across the entire STEM pipeline.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.