Conference Paper

Proceedings of BSA Conference 2017: Third Conference of IBPSA-Italy

Weather Scenario Generation for Stochastic Model Predictive Control Using Vector Autoregressive Prediction

Samuel R. Currie, Gregor P. Henze

Abstract: Conventional building energy simulation utilizes characteristic locational weather data to illustrate the typical operation of the modeled facility, generally to provide design or capital investment insight. Because of the uncertainty in the weather, the assumptions behind typical meteorological year (TMY) data tend to perform poorly in building energy modeling applications for realtime control such as model predictive control (MPC) of passive building thermal mass. To account for weather uncertainty in such operational context, we present a strategy for creating an arbitrary number of plausible near-future weather scenarios via a vector autoregressive (VAR) time-series prediction framework. This approach allows us to preserve the relationships between several spatiotemporally interrelated weather variables, for example dry-bulb temperature and absolute humidity, by capturing the variance in the joint time-series. Results from several climates are presented for 24-hour predictions of psychrometric and solar weather variables for a range of samples sizes and the application to stochastic MPC is highlighted.
Pages: 113 - 118
Paper: