Preventing wide-area blackouts through adaptive islanding of transmission networks

最近的停电和骚乱表明，两个驱动因素：a)更好地利用输电和配电网络的商业压力，以及b)分布式发电(DG)渗透率的增加，可能会降低安全裕度，并导致更高的停电概率。这个跨学科的项目涉及电力工程、图论和运筹学，将研究在连锁故障迫在眉睫时，通过预防性地将大型网络分裂为岛屿来限制停电发生和成本的方法。形成的孤岛应保持良好的需求/发电平衡，而不违反任何传输约束，避免任何发电机的机电不稳定。该项目涉及的挑战包括确定何时可以安全地启动预防性孤岛，建立形成岛屿和/或隔离网络的患病部分的技术，以及展示控制DG渗透率高的岛屿的创新方法。该建议适用于复杂科学呼叫，因为我们努力了解电网元素的小范围局部行为如何影响整个电网的弹性，并有助于防止灾难性后果。建议方法的一个主要创新之处在于，它是基于测量的，而不是基于模型的。当电力系统处于高度扰动状态，且其实际模型与正常模型有明显不同时，通常会做出孤立电力系统的决定。因此，我们建议基于从广域测量系统(WAMS)收集的实际测量数据进行分析，这些系统预计将在未来广泛使用，作为实现基于低碳可再生能源流动愿景所需的下一代监测和控制技术的一部分。该项目的两个方面特别具有冒险精神：它将数学、运筹学和电力工程联系在一起，具有独特的跨学科性质；它面向未来，应用于具有高DG渗透率的网络，即在不牺牲安全和供应质量的情况下实现2020/2050年的碳减排目标。从数学的角度来看，孤岛问题以一种令人兴奋的方式与现代公制空间分析联系在一起。考虑到网格电抗可以在网格上产生一种新的度量结构，我们将在大量研究的等周问题的数学背景下分析孤岛问题。我们的主要工具将是离散拉普拉斯算子的频谱分析，它将被加权以适应物理信息，并用于确定平衡岛屿的可能候选者。我们将使用离散版本的微积分和离散莫尔斯理论来获取有关潮流的信息，开发确定平衡孤岛的替代技术，并评估断开孤岛对网络其余部分的影响。与通常的图分区问题不同，在图分区问题中，分区标准是先验已知的(并且保持不变)，孤岛电力传输的困难来自于需要将分区调整到最近观察到的网络状态。因此，已被证明在稀疏矩阵重新排序或超大规模集成设计的上下文中如此成功的已知图划分技术不能直接适用。我们提出的解决方案是扩展现有的技术以考虑基于测量的(因此动态变化的)图的属性，并设计/实现新的图划分启发式算法。新技术将结合图形解剖启发式和随机化，并将由拉普拉斯算子的谱分析理论工具指导。

Recent blackouts and disturbances have shown that the twin drivers of: a) commercial pressures for better utilisation of transmission and distribution networks and b) increased penetration of Distributed Generation (DG) are likely to reduce security margins and lead to a higher probability of blackouts. This interdisciplinary project, involving power engineering, graph theory and operational research, will investigate methodologies to limit the occurrence and cost of blackouts through preventive splitting of large networks into islands when a cascade fault is imminent. The formed islands should preserve a good demand/generation balance, without violating any transmission constraint and avoiding electromechanical instability of any generator. The challenges addressed in this project include identification of conditions when preventive islanding can safely be activated, establishing techniques for forming islands and/or isolating a sick part of the network, and demonstrating innovative methods for control of islands with a high penetration of DG. The proposal applies to Complexity Science call, as we strive to understand how small-scale local behaviour of elements of an electrical grid influences the resilience of the grid as a whole, and helps to prevent catastrophic consequences.A major novelty of the proposed approach is that it is measurement, rather than model based. The decision to island a power system is usually taken when the system is in a highly disturbed state and its actual model is significantly different from the normal one. Hence, we propose to base the analysis on actual measurements gathered from Wide Area Measurement Systems (WAMS), which are expected to become widely available in the future, as a part of the next-generation monitoring and control technologies necessary for enabling the vision of low-carbon renewable-based energy flows. Two aspects of this project are particularly adventurous: its unique interdisciplinarity linking mathematics, operational science and power engineering, and its future-oriented application to networks with a high DG penetration, i.e. achieving 2020/2050 carbon reduction targets without sacrificing security and quality of supply.From a mathematical point of view, problem of islanding connects in an exciting way with modern analysis of metric spaces. Taking the view that grid reactances can give rise to a new metric structure on the grid, we will analyse the problem of islanding in the mathematical context of much studied isoperimetric problems. Our main tool will be spectral analysis of the discrete Laplace operator, which will be weighted to accommodate the physical information and used to identify possible candidates for balanced islands. We will employ discrete version of calculus and discrete Morse theory to capture the information about the power flows, to develop alternative techniques for identifying balanced islands, and to assess the effect of disconnecting an island on the rest of the network.In practical terms, the islanding of transmission network is a challenging graph partitioning problem. Unlike the usual graph partitioning problems, in which the partitioning criteria are known a priori (and remain constant), the difficulty in islanding electricity transmission comes from the need to adjust the partitioning to the very recently observed state of the network. As a result, the known graph partitioning techniques, which have proved so successful in the context of sparse matrix reordering or very large scale integration design, are not directly applicable. The solution we propose is to extend the existing techniques to take into account measurement-based (hence dynamically changing) graph properties and design/implement novel graph partitioning heuristics. The new techniques will combine graph dissection heuristics with randomisation and will be guided by the theoretical tool of spectral analysis of the Laplace operator.