Machine learning for anomaly detection in smart grid

点击翻译按钮获取中文摘要

Utilismart Corporation provides meter data management software solutions and data services for utilities, municipalities, industrial, commercial and residential customers. The company collects, validates, securely stores, analyzes, and presents data from smart meters, interval meters, or any other device monitoring electricity. Many issues can affect the quality of these collected/raw data, for example, communication issues, hardware problems and failures, improper setup, and many others. While the Validation, Editing, and Estimation (VEE) process remedies many of these issues, some irregularities still remain and their resolution typically requires significant manual effort. Consequently, the objective of this project is to design and develop a state-of-the-art solution for identifying these anomalies based on machine learning as well as to devise a novel technique for remedying those anomalies. The solution will consist of two main components i) Anomaly detection component will be responsible for identifying anomalous data. While many anomaly detection solutions exist, they typically belong to the supervised learning category and require a large number of known anomalies for training. In contrast, our technique will learn from predominantly normal data and will be able to detect new types of anomalies. ii) Anomaly resolution component will provide support for resolving the identified anomalies through automated or semiautomated processes. With this project, Canada has an opportunity to create a strong foundation for smart meter data analytics and build strenghth in the smart grid domain. Moreover, training in machine learning and artificial intelligence will help drive Canada's abilities in these fast-growing fields.