Use of a Stochastic Regression Model GARCH for Wind Fluctuation Study

Abstract

In contrast to mechanical ventilation, natural ventilation’s performance is usually impacted by uncertain factors, including wind fluctuations, outside air temperature variations, and occupants’ behaviours. These factors require a stochastic evaluation process for natural ventilation at the design stage such as quantifying the effective ventilation rates and optimising control strategy. A stochastic regression method based on GARCH (generalized autoregressive conditional heteroskedasticity) model is used in this paper to capture natural wind’s dynamic fluctuating characteristics and verify the existence of heteroskedasticity of wind fluctuation. The stochastic regression result is thus applied to forecast wind speed under natural ventilation condition. By comparing the power spectrum and phase space reconstruction against a simple Gaussian model generated fluctuating wind speed, we find wind speed forecasted from the GARCH model with a small modification maintains more dynamic characteristics of natural wind.