Relational deep learning for energy analytics

At UiT The Arctic University of Norway, the RELAY project is pioneering the future of energy management using relational deep learning, a family of artificial intelligence models for processing data represented as time series and graphs. This innovative approach is designed to address the complexities of energy systems, which traditional models often struggle with. RELAY focuses on three pivotal tasks. 1. Enhanced Load Forecasting: RELAY will integrate grid topology into forecasting models, enabling precise energy distribution and planning predictions, crucial for grid efficiency. 2. Dynamic Power Flow Optimization: The project will develop flexible power solvers, crucial for real-time grid adjustments, ensuring balance and optimal resource allocation. 3. Effective Outage Localization: By representing power grids as dynamic graphs, RELAY will identify risk areas, enabling faster response to outages, a vital aspect in energy-dependent regions. RELAY is not just about grid efficiency; it's about sustainable practices. By tackling the Optimal Power Flow problem, the project aims to minimize environmental impacts, especially CO2 emissions. In addition, the methodological approaches that will be developed to handle spatio-temporal data could revolutionize not only energy but other sectors like epidemiology and traffic management. A key feature of RELAY is its collaborative nature. Working with Norwegian energy companies like Ishavskraft, Lofotkraft, and Finnmark Kraft, the project aligns its research with real-world applications. RELAY is more than a project; it's a step towards a more efficient, sustainable, and resilient future. By leveraging relational deep learning, it promises to reshape our understanding of energy grids and pave the way for advanced, sustainable energy management solutions.

Machine learning models currently used in energy analytics are unable to jointly capture spatial relationships and temporal dependencies that determine, respectively, how energy flows are routed on the power grid and how the load of the electrical nodes changes over time. This, diminish the effectiveness and hinders the adoption of machine learning in the energy sector. The existing limitations can be overcome by using dynamic graphs to represent the energy grid and by processing the data with spatio-temporal models. However, such models are still in their infancy and important key challenges must be solved before deploying them in real-world applications. RELAY will develop the first large-scale deep learning framework capable of handling spatio-temporal data in energy analytics applications. This will be accomplished by advancing the theory and design of spatio-temporal models to let them: i) handle large-scale data; ii) be interpretable in what they learn; iii) quantify the uncertainty in their predictions; iv) learn hierarchical representations that allow analyzing the data at different levels of resolution. While the spatio-temporal framework that will be developed is general purpose and suitable for many different applications, RELAY will focus on three important tasks in energy analytics: forecasting of energy production and demand, power flow optimization, and the detection and localization of outages in the power grid. The project will be carried out in collaboration with the three main energy suppliers in Northern Norway, Ishavskraft, Lofotkraft, and Finnmark Kraft, which will provide data and validate the results.