Distribution Network Resilience Enhancement with Topological Neural Networks

气候变化导致的恶劣天气造成的停电数量从21世纪初全国每年约50次增加到过去5年平均每年100多次。随着气候变化导致的网络物理威胁和极端天气事件的增加，电网的恢复能力已经成为一个非常重要的社会问题。通过促进跨学科的思想交流，该项目将开发一种新的几何深度学习模型，通过利用清洁能源资源，显著提高现代电网的安全性、可靠性和效率，从而在广泛的领域影响我们社会的安全和经济福祉。该项目将为研究生提供培训机会。特别是，pi将开发一种新的几何深度学习模型，支持拓扑数据分析工具，用于具有时间主导特征的中断的稳健决策和有效知识转移，特别是专注于预测配电网络的演变和运营决策（通过利用分布式能源和清洁能源），以便在破坏性事件（如自然灾害和对抗性攻击）下恢复能力。通过研究电网的响应和恢复运行问题，将展示通过知识转移（跨事件和运行时间尺度）在通用性、稳健性和学习效率方面的预期收益。所开发的模型将应用于增强配电网络在破坏性事件下的恢复能力，从准备到恢复任务。采取的先发制人措施包括：(1)中断预测和防御性孤岛，(2)网络重构作为恢复性行动。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 number of outages caused by severe weather as a result of climate change rose from about 50 annually nationwide in the early 2000s to over 100 annually on average over the past five years. With the increasing trend of cyber-physical threats and extreme weather events due to climate change, resilience of the power network has emerged as a problem of utmost societal importance. By facilitating the cross-disciplinary exchange of ideas, this project will develop a novel geometric deep learning model to significantly improve the security, reliability, and efficiency of the modern power grid by leveraging clean energy resources – thereby, impacting the safety and economic well-being of our society across a broad front. The project will provide training opportunities to graduate students. In particular, the PIs will develop a novel geometric deep learning model supported with topological data analysis tools for robust decision-making and efficient knowledge transfer for disruptions with time-dominant characteristics, particularly focusing on predicting the distribution network evolution and operational decisions (by leveraging distributed energy resources and clean energy resources) for resilience under disruptive events such as natural disasters and adversarial attacks. Envisioned gains in generalizability, robustness and learning efficiency via knowledge-transfer (across events and operation timescales) will be demonstrated by studying response and recovery operation problems in power grids. The developed model will be applied to enhance the resilience of the power distribution network under disruptive events, ranging from preparatory to restorative tasks. The pre-emptive measures adopted include (i) outage prediction and defensive islanding, while (ii) network reconfiguration is used as a restorative action.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.