EAGER: Real-Time: Effective Power System Operation during Hurricanes using Historical and Real-Time Data

飓风经常导致大规模和长期的停电，造成严重的社会和经济后果。该项目旨在开发模型和软件工具，利用现有的天气预报数据，为电力系统的智能和预防性运行提供指导。预计开发的工具将更好地评估飓风对电网的影响，并确定对飓风造成的破坏具有弹性的运营战略。最终，该项目的成果将有助于减少飓风期间停电的规模和持续时间。避免的停电成本每年可以为美国经济节省数亿至数十亿美元。这是一个跨学科的研究项目，电气工程、土木工程和大气科学专业的学生将密切合作，拓宽他们的技能集，从而促进他们的教育，并增强国家训练有素的劳动力的能力。该项目利用天气预报数据的可用性来指导飓风期间的预防性电力系统运行。目前，天气预报数据没有系统地整合到电力系统运行中，因此，预防性运行是不可能的。目前的研究方法采用一个综合的框架，主要基于物理模型和工程模型，利用飓风预报信息来预测电力系统元件的破坏情景。然后，通过对破坏情景的显式建模，利用随机优化来确定预防性操作决策。现有的模式有两个主要缺点：1)它们忽略了天气预报的不确定性；2)它们对计算的要求非常高。天气预报数据和部件损伤模型存在着很大程度的内在不确定性，影响了最终预防作战策略的有效性以及模型的计算效率。而且，电力系统实时运行的计算时间极其有限。该项目的假设是，一种利用数据可用性的混合方法，同时依赖物理和工程原理，可以克服上述两个挑战。首先，使用历史和实时数据，该项目减少了不确定性，以提高解的质量和计算需求。其次，通过对合成、历史和实时数据应用机器学习技术，该项目用有效的代理确定性模型取代了计算要求很高的随机优化，以实现实时操作的计算可处理性。因此，该项目将首次能够将不确定天气预报数据适当地整合到飓风期间的电力系统运行中，以快速确定有效的预防性运行策略。该奖项反映了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。

Hurricanes often lead to large and long-lasting blackouts, with serious social and economic consequences. This project aims at developing models and software tools to provide guidance for the smart and preventive operation of power systems, using available weather forecast data. It is anticipated that the developed tools will better estimate the impacts of hurricanes on the electric power grid and identify operation strategies that are resilient to the damages induced by the hurricane. Ultimately, the outcomes of this project will help reduce the size and duration of power blackouts during hurricanes. The avoided blackout costs can save the U.S. economy hundreds of millions to several billion dollars each year. This is an interdisciplinary research project, where electrical engineering, civil engineering, and atmospheric science students will closely collaborate and broaden their skill sets, thus advancing their education and empowering the nation's trained workforce.This project exploits the availability of weather forecast data to guide preventive power system operation during hurricanes. Currently, weather forecast data is not systematically integrated into the power system operation and, thus, preventive operation is not possible. The current research approaches employ an integrated framework, mainly based on physical and engineering models, which uses the hurricane forecast information to predict the component damage scenarios for the power system. The preventive operation decisions are, then, determined with stochastic optimization, through explicit modeling of the damage scenarios. The existing models have two main shortcomings: 1) they ignore the weather forecast uncertainties; and 2) they are extremely computationally demanding. There are substantial levels of inherent uncertainties in weather forecast data and component damage models, which affect both the effectiveness of the final preventive operation strategies as well as the model?s computational efficiency. Moreover, the computational time in real-time power system operation is extremely limited. The hypothesis governing this project is that a hybrid approach, which exploits the availability of data, while also relying on physical and engineering principles, can overcome the two challenges, mentioned above. First, using historical and real-time data, this project reduces uncertainties in order to improve both the solution quality and the computational needs. Second, by applying machine learning techniques on synthetic, historical, and real-time data, the project replaces computationally-demanding stochastic optimization with effective proxy deterministic models, to achieve computational tractability for real-time operation. Thus, this project will enable, for the first time, appropriate integration of uncertain weather forecast data within power system operation during hurricanes, to quickly identify effective preventive operation strategies.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.