An AI Approach to Forecasting Net Loads and Distributed Solar Production for Utilities

Accurate forecasts of a utility's net electrical load—the power it must supply to meet its customers' demand—provide valuable decision support information for the utility's energy production, distribution and trading operations. Electrical loads are determined by complex interactions between customer energy usage patterns and weather conditions; for instance, the power required for cooling on a hot summer day depends on the day of week, the temperatures of recent days, and the presence of clouds and precipitation. The situation is complicated further by the rapid growth of distributed solar power production (e.g., rooftop solar) “behind the meter” -- by supplying some of its customers' power needs, distributed solar reduces a utility's net demand during sunny periods, but it also increases net load variability under rapidly changeable weather conditions. To address this complex forecasting challenge, the National Center for Atmospheric Research has developed an artificial intelligence system that provides short-term distributed solar and electrical net load forecasts. The system utilizes forecasts of key meteorological variables from the Dynamic Integrated foreCast (DICast®) system, developed at NCAR and run operationally by the Global Weather Corporation, to produce hourly 0-168 hour ahead forecasts. The AI forecast models are trained using a historical database of observed loads, solar percent capacity production, characteristics of distributed solar installations, and weather variables, and are automatically updated at regular intervals. The load forecasts are accompanied by the identification of several historical “analog days” having similar forecasted weather characteristics to the target day; providing the observed load profiles for these days aids users in interpreting the deterministic load forecast and mitigating risk. It is anticipated that the NCAR load and distributed solar forecasts will provide improved information to utility decision makers, and ultimately will facilitate the efficient integration of an increasing base of distributed solar installations.