Using the National Digital Forecast Database for model-based building controls

Abstract

Abstract An accurate short-term weather forecast enhances the performance of model-based control strategies for a building. Most existing short-term forecast methods and models are based on historical archives of weather variables observed about an individual building. This practice leads to ask: 1) Does a short-term forecast model based on a collection of the past data have sufficient potential to predict future behavior, which is inherently random?; 2) If historical data is not available, or if only a limited number of weather variables is available, what are the other alternatives?; and 3) Would it be possible to derive a relevant weather data set from an incomplete historical data-driven model? The National Digital Forecast Database (NDFD) XML may answer these questions. This paper discusses an applicability of the NDFD XML for model-based building control solutions. It also presents a simplified, easily implementable, and reliable method to predict hourly global horizontal solar radiation. The short-term weather forecast method using the NDFD XML shows an outstanding performance of forecasting erratic and sporadic characteristics of weather, particularly when compared to the historical data-driven method. However, its forecast performance is not always accurate due to the inherent irregularity of weather. In order to increase robustness of model-based controls, we recommend that short-term weather forecasts should be considered as possessing scenario uncertainty . Including multiple scenarios into formulation of control problem can effectively describe scenario uncertainty when part of scenarios account for such erratic nature.