Abstract
Due to increasing and variable power flows, many overhead lines approach their traditionally static or seasonal rating limits, which were established decades ago based on very conservative weather conditions for more stable flows. Since expanding infrastructure often entails lengthy delays and high costs, Dynamic Line Rating (DLR) of overhead lines (OHLs) is an alternative solution to leverage unexploited line capacity. In DLR, wind is the primary cooling factor. However, wind speed and direction vary greatly over short distances due to local environmental features like trees, buildings, terrain roughness, conductor level, turbulence, and the atmospheric boundary layer physics. Despite continued advancements, meteorological models still struggle to account for local disturbances, which are very complex in practice to be explicitly incorporated into digital models. Typically, the approach at Ampacimon involves adjusting weather provider wind values and forecasts according to local measurements from sensors installed on the conductor at critical spans. This paper examines the constraints and opportunities of a hybrid approach that employs Artificial Intelligence (AI) / Machine Learning (ML) techniques to extrapolate wind data from critical span locations to adjacent spans with comparable line and environmental conditions. Three conditions must be met: low distance and angle between spans (such as parallel circuits) and little wind obstruction around the span. This new approach improves the scalability of DLR systems with a more optimized number of sensors, reducing installation costs, yet still ensuring high accuracy. Despite the high variability of wind, satisfactory results, validated in the field against sensor measurements over one year of data at multiple locations, are presented. This study also highlights the critical need to include local sensor-based wind measurements in the process to avoid overestimating the real available capacity.