Intra-hour solar power forecasts using a real-time irradiance monitoring network
Our real-time irradiance network consists of approximately 50 sensors in and around Tucson, AZ. Some sensors are custom-made irradiance sensors that are powered via solar panel and communicate via a cellular data connection, which means they can be placed anywhere with cell reception. We also receive monitoring data from rooftop solar power systems, which we previously used to make retrospective forecasts. We now use our real-time network to make operational forecasts.
To make forecasts, we first gather the data from our central database. We then use pre-generated clear-sky profiles to compute the clearness index for each sensor. Next, we make an interpolated clearness index map and we take a weighted average of this new map and previous maps that are appropriately translated based on cloud motion vectors. To estimate the cloud motion vectors, we can use output from the Univ. of Arizona high-resolution WRF model, upper-air soundings, or correlation between sensor pair time-series. Finally, we use the cloud motion vector to translate the map to make forecasts of clearness index for any location in the domain. We can multiply this clearness index by a power plant's clear-sky profile to obtain a power forecast. We've shown that our forecasts can outperform persistence forecasts at short time horizons. Error statistics and comparison to other forecasting methods will be discussed.