Learning-based distributed load forecasting in energy grids

Khalid Kalbat, Ali Tajer

Abstract: This paper proposes a generic distributed framework for load forecasting, as a pivotal function in energy grids. Energy grids are rapidly evolving towards complex interconnected subnetworks that are potentially operated by different entities with distinct physical constraints, generation capacitance, and load demands. Such subnetworks, nevertheless, are also interconnected through shared sensing, actuating, and controlling modules. The goal of this paper is to exploit such underlying interconnectivity structures among the subnetworks and formalize a mechanism for 1) forming local predictions in the subnetworks based on local stochastic dynamics and historic data, and 2) aggregating the local predictions for reaching network-wide optimal load predictions. The advantages of the proposed framework is supported by simulations in the standard IEEE-14 bus power system.