CAREER: Learning Smart Meter Data to Enhance Distribution Grid Modeling and Observability

This NSF CAREER project aims to provide the theoretical and computational foundation that will allow unlocking the untapped potential of smart meters and radically enhance electric power distribution grid observability in both normal and outage conditions. The project will transform existing distribution grid modeling and monitoring that relies on costly sensors to a scalable and robust method using widely-deployed smart meters.

The intellectual merits of the project include developing new optimization and probabilistic graph learning methods to enable data-driven real-time network modeling, rapid detection of large-scale outages, and robust distribution system state estimation. The broader impacts of the project include integrating power engineering education with data science through training professional workforce for data challenges, developing open-source datasets, and providing outreach to high-school students for interactive learning of smart grids.

If successful, this project will provide U.S. utilities with better situational awareness at minimum sensor investment cost, thus saving millions of dollars, while promoting seamless integration of renewable energy, and enhancing grid resilience.

The increasing deployment of smart meters extends monitoring capability to grid edges and provides unprecedented amounts of data. However, most utilities use smart meters for billing purposes only, without exploring insights or gaining actionable information from them because these data are limited to low-resolution and unsynchronized measurements. The proposed project will open a new venue to enable utilities to extract useful intelligence from smart meters through three major technical innovations: (1) Real-time topology identification, where the approach is to design a Laplacian-like matrix that can capture the physical network feature and leverage its inherent sparse structure to discover nodal connectivity even from low-quality measurements.

For online parameter identification, a novel bottom-up optimization algorithm using only smart meter data is proposed. (2) A new graph learning approach that takes advantages of intrinsic conditional independencies among smart meters and other outage information sources to serve as a data fusion framework for fast, scalable, and accurate outage detection. (3) A multi-objective robust data recovery technique to minimize smart meter asynchrony error. A hierarchical reinforcement learning-aided method is proposed to overcome the scalability issue, and to enable joint primary-secondary distribution system state estimation.

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.