CAREER: Stochastic capacity scheduling and control of distributed energy storage enabling stacked services

PI的长期研究目标是开发方法，通过利用分布式能源的灵活性，确保具有大量波动的可再生能源的电力系统的可靠性。随着电池成本的下降，越来越多的小型分布式电池被连接到网络中，以提供本地服务。可以聚合这些资源，并在本地服务不需要时使用它们来提供网格服务。类似地，可以聚合和协调灵活的负载，使其表现得像存储一样，并同时提供多个服务。堆叠服务提高了储能利用率和经济性，同时也提高了电网容纳更多可再生能源的能力，改善了其对环境和健康的影响以及能源安全。本提案的研究目标是开发一套计算易于处理的优化和控制算法，使聚合器能够最优地协调数千个高度异构和分布式的小规模存储资源和负载，以提供堆叠服务。研究结果将影响能源政策，特别是围绕堆叠服务的法规。PI的总体教育目标是激励学生应对气候变化和能源安全的重大挑战，开发同时教授电力系统和培养学生基本STEM技能的课程，并提高公众对电力系统现代挑战和机遇的认识和理解。为此目的，教育计划包括四项活动：i)利比里亚的国际教育和外联；Ii)通过开发在线储能游戏进行公众教育和推广，突出与分布式储能相关的机遇和挑战；（三）研究生和本科教育；iv)通过短期课程、网络研讨会和讲座与从业人员、研究人员和政策制定者进行接触。这些教育活动将通过密歇根大学和利比里亚大学的课程创新，以及向利比里亚K-12和本科妇女提供服务，促进劳动力发展。由于电力系统具有不确定性、非线性和高维性，因此研究目标具有挑战性；储存协调所需的传感和通信系统不完善；协调工作不能对配电网产生负面影响。该研究将开发i)调度分布式存储聚合的随机优化方法，以提供堆叠服务，同时管理存储的非凸性；Ii)鲁棒预测非线性控制方法来协调分布式存储的聚合以提供堆叠服务，利用捕获退化动态的新型聚合存储模型；Iii)综合负荷模型（由实验数据确定），将负荷灵活性表示为存储，包括与负荷协调行动相关的效率和退化，以及使用近似存储类型模型导致的模型误差；iv)协调高异构存储和负载聚合的优化/控制方法。结果将在高保真仿真环境中得到验证，最有希望的方法将在现实的物理测试平台中实现。在方法上，该研究将开发i)针对非凸模型的新型随机对偶动态规划（SDDP）方法，利用SDDP对整数规划问题的新兴扩展，以及ii)集成滑模控制和模型预测控制的新型鲁棒预测控制方法。虽然过去的工作已经提出了滑动和模型预测控制相结合的方法，但它们需要在线优化，而我们寻求一种快速在线计算的控制策略。总之，这项研究将有利于电力系统、运筹学和控制系统社区。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

The PI's long-term research objective is to develop approaches to ensure the reliability of electric power systems with massive amounts of fluctuating renewable energy resources by leveraging flexibility from distributed energy resources. As battery costs have decreased, more small-scale distributed batteries have been connected to the network to provide local services. These resources could be aggregated and used to provide grid services when not needed for their local service. Similarly, flexible loads can be aggregated and coordinated to behave like storage and provide multiple services simultaneously. Stacking services improves storage utilization and economics in addition to the ability of the grid to accommodate more renewables, improving its environmental and health impacts along with energy security. The research objective of this proposal is to develop a suite of computationally-tractable optimization and control algorithms that will enable aggregators to optimally coordinate thousands of highly-heterogeneous and distributed small-scale storage resources and loads to provide stacked services. The findings will impact energy policy, specifically the regulations surrounding stacked services. The PI's overall educational objectives are to inspire students to tackle the grand challenges of climate change and energy security, develop curricula that simultaneously teaches electric power systems and builds students' fundamental STEM skills, and increase public awareness and understanding of modern challenges and opportunities in power systems. To this end, the education plan includes four activities: i) international education and outreach in Liberia; ii) public education and outreach via the development of an online energy storage game highlighting the opportunities and challenges associated with distributed energy storage; iii) graduate and undergraduate education; and iv) outreach to practitioners, researchers, and policymakers via short courses, webinars, and talks. The educational activities will promote workforce development through curricular innovation at University of Michigan and the University of Liberia, and outreach to Liberian K-12 and undergraduate women. The research objective is challenging because the power system is uncertain, nonlinear, and high dimensional; sensing and communication systems required for storage coordination are imperfect; and coordination must not negatively impact the distribution network. The research will develop i) stochastic optimization approaches to schedule aggregations of distributed storage to provide stacked services, while managing storage nonconvexities; ii) robust predictive nonlinear control approaches to coordinate aggregations of distributed storage to provide stacked services, leveraging novel aggregate storage models capturing degradation dynamics; iii) aggregate load models (identified with experimental data) representing load flexibility as storage, including the efficiency and degradation associated with load coordination actions and the model error resulting from the use of approximate storage-type models; and iv) optimization/control approaches to coordinate highly-heterogeneous storage and load aggregations. The results will be validated in high-fidelity simulation environments and the most promising approaches will be implemented in a realistic physical test bed. Methodologically, the research will develop i) novel stochastic dual dynamic programming (SDDP) methods for nonconvex models, leveraging emerging extensions of SDDP to integer programming problems, and ii) novel robust predictive control methods that integrate sliding mode control and model predictive control. While past work has proposed methods of combining sliding and model predictive control, they require online optimization whereas we seek a control policy for fast online computation. In summary, the research will benefit the power systems, operations research, and control systems communities.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.