Blending distributed photovoltaic and demand load forecasts

Abstract

Utilities consider having accurate electric load forecasts to be critical to their day-to-day operations. But the growing penetration of distributed photovoltaic (DPV) solar power production “behind the meter” makes it more difficult to predict load because it is hard to distinguish between increased solar power generation and decreased power consumption at a site with installed DPV. This paper describes the development of a “top down” solar power forecasting system and how it can be integrated with a load forecasting system that incorporates weather information. Both systems depend on accurate weather forecasts, information regarding the utility variables, and daily and seasonal factors. The load forecasting system provides day-ahead forecasts having roughly 2% error, on average, and the solar power forecast is within 5% error. In general, we find that if the load forecasting system incorporates the same weather and solar predictors as the solar power forecasting system, it implicitly accounts for DPV generation during periods of stationary solar power deployment, and a separate DPV forecast may not be necessary. However, for higher penetrations of solar power during times of rapid deployment of additional solar capacity, it may become important to explicitly incorporate the solar power forecasts to avoid degradation of the load forecasts. This approach could allow utilities and independent system operators to better deal with rapidly increasing penetration of DPV generation.