Ampacity forecasting from Numerical Weather Predictions: a fusion of the traditional and machine learning methods

Gabriela Molinar, Johannes Bassler, Nicholas Popovic, Wilhelm Stork

Abstract: The present document analyzes current-carrying capacity forecast models based on online-available Numerical Weather Prediction (NWP) results. For this task, Feedforward and Convolutional Neural Networks have been implemented. The first one interpolates the NWP to the overhead line, optimizing the ampacity forecast accuracy directly. The second approach considers the spatial grid of NWP results, which are treated as pixels in an image. Convolutions play an important role in this solution, because of their capacity to extract relevant spatial and temporal patterns from the data, which are integrated into the ampacity forecast performance. This paper compares the results of these machine-learning-based forecast models to the direct calculation of the ampacity prediction from the nearest NWP grid point. A standard open-source dataset was developed for this case study, in order to build a reference for future studies in this research area <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">1</sup> .