Comparison of Top-Down and Bottom-up Methods of Modeling Regional Distributed Solar-based Power Production for Short-Term Forecasting Applications

July 28, 2015
Session T2B: Solar Forecasting: Distributed to Centralized — Room 105
9:45 am  -  10:45 am

“The interest in forecasting the variations in distributed (also known as “behind-the-meter”) solar-based power generation (DSG) has been growing rapidly because the amount of distributed solar-based production on some grid systems has also been increasing rapidly. The prediction of DSG is complicated by the fact that many of the attributes (panel types, panel orientation, maintenance status etc.) of the DSG systems are frequently not available to the forecasting process and indeed in many cases measurements of the aggregated production are not available to the forecaster in either a real-time or historical mode. Thus, many approximations and assumptions often must be employed to estimate the production from the information that is available.

This presentation will provide an overview of a comparison of two different approaches to the estimation of “behind the meter” distributed solar power production: (1) a bottom-up method and a (2) top-down approach. The comparison was based on data from the Big Island of Hawaii for the year of 2014. A total of approximately 50 MW of distributed PV capacity was nominally in production on the island by the end of the year. This represents a substantial fraction of the typical daily system load, which varies from a minimum of about 90 MW a peak around 170 MW.

In the so-called “bottom-up” approach all of the solar generation capacity connected to each substation was treated as a virtual solar generation facility. A statistical power output model was constructed by using the measured or estimated solar irradiance at the substation and samples of measured solar-based production from distributed photovoltaic (PV) installations. This model along with data on the installed PV capacity connected to each substation was used to estimate the output from each virtual generator and then these estimates were aggregated to calculate the island-wide (system-wide) production.

In the “top-down” approach, the island-wide solar generation was estimated from system-wide load data, a population-weighted island-wide average solar irradiance derived from visible satellite image data and traditional explanatory variables (day of the week, temperature etc.) used in load models. No information about the attributes or installed capacity of the distributed solar-based power production was used in this approach. All of the information about the distributed solar-based generation was essentially extracted from the load variations explained by the satellite-based irradiance estimates.

The initial results indicated that the two approaches produced fairly similar results for estimates of the peak, average and day-to-day variability in midday distributed solar production for the year. For example, both approaches yielded a maximum estimated production of about 35 MW at approximately noon on a day with the least cloud cover. However, the issues and limitations of each approach were also evident.”