| dc.description.abstract | The promising potential of wind energy as a source for carbon-free electricity is still hampered by the uncertainty and limited predictability of the wind resource. The overarching theme of this dissertation is to leverage the advancements in statistical learning for developing a set of physics-informed statistical methods that can enrich our understanding of local wind dynamics, enhance our predictions of the wind resource and associated power, and ultimately assist in making better operational decisions. At the heart of the methods proposed in this dissertation, the wind field is modeled as a stochastic spatio-temporal process. Specifically, two sets of methods are presented. The first set of methods is concerned with the statistical modeling and analysis of the transport effect of wind—a physical property related to the prevailing flow of wind in a certain dominant direction. To unearth the influence of the transport effect, a statistical tool called the spatio-temporal lens is proposed for understanding the complex spatio-temporal correlations and interactions in local wind fields. Motivated by the findings of the spatio-temporal lens, a statistical model is proposed, which takes into account the transport effect in local wind fields by characterizing the spatial and temporal dependence in tandem. Substantial improvements in the accuracy of wind speed and power forecasts are achieved relative to several existing data-driven approaches. The second part of this dissertation comprises the development of an advanced spatio-temporal statistical model, called the calibrated regime-switching model. The proposed model captures the regime-switching dynamics in wind behavior, which are often reflected in sudden power generation ramps. Tested on 11 months of data, double-digit improvements in the accuracy of wind speed and power forecasts are achieved relative to six approaches in the wind forecasting literature. This dissertation contributes to both methodology development and wind energy applications. From a methodological point of view, the contributions are relevant to the literatures on spatiotemporal statistical learning and regime-switching modeling. On the application front, these methodological innovations can minimize the uncertainty associated with the large-scale integration of wind energy in power systems, thus, ultimately boosting the economic outlook of wind energy. | en |
