Date of Completion

4-12-2017

Embargo Period

4-10-2017

Keywords

weather forecast, wind speed, power outage prediction, numerical weather prediction, planetary boundary layer

Major Advisor

Emmanouil N. Anagnostou

Associate Advisor

Joshua P. Hacker

Associate Advisor

Thomas M. Hopson

Associate Advisor

Marina Astitha

Associate Advisor

Guiling Wang

Field of Study

Civil Engineering

Degree

Doctor of Philosophy

Open Access

Campus Access

Abstract

Skillful wind speed forecasts in the Northeast U.S. allows for improved preparedness when potentially severe storms advance toward the region. Power outages are frequently caused by uprooted and broken trees and branches, which in turn are provoked by gusts and sustained-high wind speeds. Combining wind forecasts with outage prediction models allow utilities to confidently recruit and allocate crews near vulnerable areas, so power is restored fast and efficiently. The Outage Prediction Model at the University of Connecticut (UConn-OPM) guides emergency response of electric utilities in the Northeast providing outage predictions over their service territory with up to three-day lead time. The UConn-OPM operates with inputs from a Numerical Weather Prediction (NWP) model that provides forecasts for upcoming storms throughout the year. The system performance is determined by the weather forecast skill in simulating storms of varying scales, and in resolving small- scale features driving localized gusts. This thesis encompasses an analysis of NWP wind-speed error sources, a novel methodology to estimate forecast probabilities, and an investigation of error propagation from NWP simulations onto the UConn-OPM outage predictions. First, NWP simulations are evaluated along with an investigation of sources of wind-speed errors, such as terrain, season, and flow patterns. Wind-speed biases depend strongly on flow patterns and the NWP performance is influenced by atmospheric stability. Results motivate the development of a new technique to derive probabilistic forecasts from deterministic NWP model outputs. The object-based analog (Obj-An) forecast technique devised in this study uses spatial features to identify similar storms in the past and previous forecast errors to derive an ensemble representation of the deterministic forecasting uncertainty. The forecast probabilities obtained with the Obj-An technique are based on an algorithm capable of distinguishing flow and spatial patterns through image processing techniques. Finally, the effect of NWP wind simulation errors on power outage predictions is assessed through wind gust simulations using different Planetary Boundary Layer parameterizations and an off-line gust parameterization. The forecasts from the various parameterizations are used in the UConn-OPM to predict outages, through which we evaluate the error propagation from wind and gust simulations to outage predictions.

COinS