Statistical analysis of the challenges to high penetration of wind energy
Date
2014
Authors
O'Connell, Matthew, author
Marchese, Anthony, advisor
Zimmerle, Daniel, advisor
Young, Peter, committee member
Journal Title
Journal ISSN
Volume Title
Abstract
Grid penetration of renewable energy technologies, especially wind power, is higher than ever and continues to increase. The inherent stochastic variability of wind makes predicting wind, and thus power generation difficult. Generating companies usually don't openly share power output predictions or historical generation data which increases the level of complexity when determining new wind plant locations or estimating delivered grid level power. This work focuses on statistical data analysis and advanced data modeling related to wind power forecasting and generation. The first part of this thesis uses power output logs from several wind plants and a well-known forecasting method to determine energy storage requirements for individual wind plant contract firming. Forecasts of varying accuracy are used to characterize storage requirements based on contract period length, forecast lead time, and forecast accuracy. Results show that forecast error distributions are effected more by forecast accuracy and lead time than wind plant size and location. The biggest reductions in produced power deviations can be achieved by increasing forecast accuracy and decreasing forecast lead time. The second part of this work develops a statistical analysis which allows estimation of contract firming requirements for a specific wind plant location without the need for time series wind and forecast data. The developed method requires only a wind speed and forecasting error distribution. Using these distributions, deviations between forecast to produced power and energy can be estimated. Results from comparing to historical time series data show this method is accurate to within 10% of actual amounts. Since distributions are much more easily attained than historical time series data, this analysis is useful for developers when evaluating potential new locations. The third part of this work uses a pattern matching algorithm to recognize wind ramp events and separate the forecasting error due to timing from the forecasting error due to magnitude. Wind ramp detection is achieved by developing a pattern matching algorithm which is also shown to work in identifying start and stop transients in electrical device current draw. The analysis confirms wind ramp events can be detected by calculating a bimodal ranking value from a histogram of power data, and the effects of forecast timing and magnitude can be separated from overall forecasting errors. The results of this analysis show magnitude errors contribute more in large wind ramp events, while timing errors contribute more in small ramp events.