George Young, Thesis Advisor/Co-Advisor George Young, Thesis Advisor/Co-Advisor
Keywords:
AR/ARMA Wind Speed Prediction Neural Networks
Abstract:
Projected construction of nearby wind farms motivates this study of statistical forecasting of wind speed, for which accurate prediction is critically important to the fluid integration of wind power into the electricity grid and energy market. An 18-year record of hourly wind speed data from Williamsport, Pa. is used to develop a series of Autoregressive (AR) or Autoregressive Moving Average (ARMA) models. Performance assessments of these advanced persistence models allow for the quantification of baseline skill in wind speed forecasting. Further investigation reveals marked annual and diurnal patterns in the wind speed record, prompting the creation of scaled variables with mixed success. For each method, a persistent skill wall in modeled wind speed is observed, but this threshold is surpassed using Artificial Neural Networks (ANN).