Adaptive data-driven secondary control and cyberattack-resilient secondary control for AC microgrids

加拿大有100多个偏远社区没有与国家电网连接。他们主要依靠柴油发电，成本高，污染严重。与此同时，暴风雪和野火造成的损失成为加拿大停电的主要原因。为了增加可再生能源供应，提高电网在极端天气下的弹性，微电网，小型和自给自足的能源系统，可以发挥至关重要的作用。与大容量电力系统相比，微电网具有低能耗、高自愈能力等优点，可同时运行于并网模式和孤岛模式。尽管有许多好处，但微电网也带来了新的挑战。微电网以变流器接口的分布式能源（DER）为主，具有低惯性的特点，这意味着微电网的电压和频率往往会受到实时功率不平衡的影响而出现较大的偏差。这种偏差如果不及时纠正，可能会导致更大的功率不平衡，甚至灾难性的系统崩溃。二次控制被认为是校正电压和频率偏差的有效手段，这通常是基于先验精确的物理模型和完整的传感器数据来执行的。 然而，由于转换器的不同操作模式和不同的网络拓扑结构，准确的物理模型可能并不总是可用的，而由于信息基础设施的脆弱性，传感器数据容易受到网络攻击。从最近的入侵事件中可以明显看出，针对电力部门的网络攻击是不可预测的，例如，2017年，美国堪萨斯伯灵顿附近的一座核电站被渗透。为了应对这些挑战，在本研究项目中，我们打算为交流微电网开发新型自适应二次控制和网络攻击弹性二次控制。利用Koopman算子理论，分布式控制和强化学习，我们将：1）。为微电网开发一种新的在线自适应二次控制框架，以实时校正频率和电压偏差，而无需电网模型和预热训练的任何先验知识; 2）.研究跨层网络攻击对已开发的二级控制方法的影响; 3）使用深度学习技术设计网络攻击弹性二级控制方法。由于极端天气或恶意网络攻击导致的大范围停电不仅代价高昂，而且会对数百万人的日常生活造成严重破坏，并严重破坏基本服务的提供（例如，食品、保健）。拟议的研究将产生新的网络攻击弹性二次控制算法，以确保微电网在不同的电网拓扑结构，工作模式，甚至网络攻击下的稳定运行。研究成果将有助于发展安全稳定的微电网，提高可靠的电力，以及发展安全，清洁和有弹性的社区。

More than 100 remote communities in Canada are not connected to the national electrical grid. They rely mainly on diesel for electricity, suffering from high costs and pollutions. Meanwhile, damage from snowstorms and wildfires becomes a leading cause of power outages in Canada. To increase the renewable energy supply and enhance the resiliency of electric power grids under extreme weather, microgrids, small-scale and self-sufficient energy systems, can play a crucial role. Compared to the bulk power systems, microgrids that can operate in both grid-connected and islanded modes have the advantages of low-carbon consumption and high self-healing capability. Despite many benefits, microgrids also bring new challenges. Microgrids dominated by the converter-interfaced distributed energy resources (DERs) are characterized with low inertia, meaning that microgrid voltage and frequency tend to experience large deviations subject to real-time power imbalance. Such deviations without timely correction may cause a severer power imbalance or even catastrophic system collapse. Secondary control is conceived as an effective means to correct the voltage and frequency deviations, which is conventionally carried out based on an apriori accurate physical model and intact sensor data. Nevertheless, an accurate physical model may not always be available due to diverse operating modes of converters and varying network topologies, while sensor data is susceptible to cyberattacks because of the vulnerability of information infrastructure. The plausibility of cyberattacks against the electric power sector is evident from the recent intrusion events, e.g., the penetration of a U.S. nuclear power plant near Burlington, Kansas, in 2017.To address these challenges, in this research project, we intend to develop novel adaptive secondary control and cyberattack-resilient secondary control for AC microgrids. Leveraging on Koopman operator theory, distributed control, and reinforcement learning, we will: 1). develop a novel online adaptive secondary control framework for microgrids to correct frequency and voltage deviations in real-time, without any prior knowledge of the grid model and warm-up training; 2). study the impacts of cross-layer cyberattacks on the developed secondary control methodologies; 3) design cyberattack-resilient secondary control methods using deep reinforcment learning techniques.Widespread power outages due to extreme weather or malicious cyberattacks are not only costly but also wreak havoc on millions of people's daily lives and profoundly disrupt the delivery of essential services (e.g., food, health care). The proposed research will yield novel cyberattack-resilient secondary control algorithms to ensure the stable operation of microgrids under varying grid topologies, working modes, and even under cyberattacks. The research outcomes will contribute to the advancement of developing secure and stable microgrids, the enhancement of reliable electric power, and the development of secure, clean, and resilient communities.