Hyperlocal Forecasting with Sensor Feedback for Reduced Uncertainty in Dynamic Line Rating

Wind speed is the most important weather variable, so rating forecasting is largely a problem of wind speed forecasting. However, existing wind speed forecasts from traditional weather models suffer from multiple problems that make them inappropriate for use in forecasting ratings. Most notably, they assume a default height of 10 m and are too coarse in resolution to model the individual terrain (including topography, vegetation, and buildings) which can slow down winds passing over the lines. The rating of a full line is the rating of the most limiting section within that line and, in most cases, the most limiting section is the type of wind sheltered area where traditional weather models perform the worst. We have developed a computational fluid dynamics (CFD) based downscaling methodology which simulates the effects of terrain under a variety of atmospheric conditions, including prevailing wind direction. By utilizing our sensor network installed in utility right of ways we have been able to tune our CFD model and calculate error metrics. Across a sample of 178 weather sensors, our CFD model reduced the mean absolute error (MAE) of wind speed by 50.4%. By making some assumptions about typical conductors, we can convert wind speed errors into rating errors. In that space we see a similar 40.6% improvement in MAE. These results were statistically significant using nonparametric tests. This accuracy improvement is even greater at the type of sheltered sites which most strongly influence line-level ratings.