Machine Learning Methods for Forecasting Grid States
Publication at PSCC 2026
With the growing deployment of photovoltaic systems, electric vehicles, and other distributed energy resources, operating distribution grids is becoming increasingly challenging. In this context, our paper “Probabilistic Spatio-Temporal Machine Learning Methods for Distribution System State Forecasting” investigates how future voltage levels and line loadings can be reliably forecast using data-driven approaches.
The work was first presented at the Power Systems Computation Conference (PSCC) 2026 and was subsequently published in the journal Electric Power Systems Research (EPSR). Our results demonstrate that machine learning models combining temporal dynamics with grid topology information provide robust and transferable forecasts across different distribution networks.
A key benefit of state forecasting is the early identification of critical operating conditions, such as voltage violations and line overloads. This creates an important foundation for applications including grid-oriented operating envelopes, dynamic network tariffs, and the proactive control of flexible assets to support grid-friendly operation. In addition, probabilistic forecasts explicitly account for uncertainty, providing valuable decision support for future distribution system operation.
