From Data Points to Ampacity Forecasting: Gated Recurrent Unit Networks

• Published in: 2018 IEEE Fourth International Conference on Big Data Computing Service and Applications (BigDataService) pp. 200 - 207
• Main Authors: Molinar, Gabriela, Popovic, Nicholas, Stork, Wilhelm
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.03.2018
• Subjects: ampacity forecasting; Conductors; Forecasting; gated recurrent unit; machine learning; overhead line monitoring; Predictive models; recurrent neural network; Weather forecasting; weather prediction; Wind speed
• DOI: 10.1109/BigDataService.2018.00037

Predicting the maximum current capacity of high voltage overhead lines (the so called, ampacity forecasting) is an interesting and still explored subject in the energy industry. Due to its dependency on weather conditions (and the infamously complex mechanisms which affect these), calculating the maximum amount of power a conductor can transmit in the future remains challenging. The project this paper originates from, proposes the installation of several weather stations along an overhead line in order to acquire meteorological data in the locality of the conductor. This data is then used to model the local atmospheric dynamics using machine learning algorithms. In this paper, experiments based on gated recurrent unit networks to forecast the ampacity of simulated overhead lines up to 24 hours ahead are presented and analyzed. The use of geographically distributed weather stations brings with it an improvement in the accuracy of the forecasts, while a higher precision remains as a goal to be accomplished. Moreover, the impact of feature selection and extraction, bearing in mind the underlying meteorological concepts, is also discussed.