Data-driven methods to improve resource utilization, fraud detection, and cyber-resilience in smart grids

Data;Machine Learning;Security;Smart Grids;Artificial Intelligence;Neural Network;Power;Energy;Wind;Solar;Networks;Attack;Detection;Generation;Consumption;Anomaly;Fraud;Resilience;Outage;Frequency;Demand Response;Signal Processing

Badrinath Krishna, Varun

关键词: Data;    Machine Learning;    Security;    Smart Grids;    Artificial Intelligence;    Neural Network;    Power;    Energy;    Wind;    Solar;    Networks;    Attack;    Detection;    Generation;    Consumption;    Anomaly;    Fraud;    Resilience;    Outage;    Frequency;    Demand Response;    Signal Processing;   

Others  :  https://www.ideals.illinois.edu/bitstream/handle/2142/102493/BADRINATHKRISHNA-DISSERTATION-2018.pdf?sequence=1&isAllowed=y

美国|英语

来源: The Illinois Digital Environment for Access to Learning and Scholarship

【 摘 要 】

This dissertation demonstrates that empirical models of generation and consumption, constructed using machine learning and statistical methods, improve resource utilization, fraud detection, and cyber-resilience in smart grids.The modern power grid, known as the smart grid, uses computer communication networks to improve efficiency by transporting control and monitoring messages between devices. At a high level, those messages aid in ensuring that power generation meets the constantly changing power demand in a manner that minimizes costs to the stakeholders. In buildings, or nanogrids, communications between loads and centralized controls allow for more efficient electricity use. Ultimately, all efficiency improvements are enabled by data, and it is vital to protect the integrity of the data because compromised data could undermine those improvements. Furthermore, such compromise could have both economic consequences, such as power theft, and safety-critical consequences, such as blackouts.This dissertation addresses three concerns related to the smart grid: resource utilization, fraud detection, and cyber-resilience. We describe energy resource utilization benefits that can be achieved by using machine learning for renewable energy integration and also for energy management of building loads. In the context of fraud detection, we present a framework for identifying attacks that aim to make fraudulent monetary gains by compromising consumption and generation readings taken by meters. We then present machine learning, signal processing, and information-theoretic approaches for mitigating those attacks. Finally, we explore attacks that seek to undermine the resilience of the grid to faults by compromising generators' ability to compensate for lost generation elsewhere in the grid. Redundant sources of measurements are used to detect such attacks by identifying mismatches between expected and measured behavior.