Analysis of Open-Source Data in Wind Power Prediction Modeling

Abstract

As identified in the 2021 IPCC AR6 WGIII report, wind energy has a high potential to reduce greenhouse gas emissions. The deployment of wind energy, however, has fallen behind its potential in part because of the need for improved wind power predictions. This thesis combines historical power production data, meteorological station data, reanalysis data, and numerical weather prediction output data (WRF model) to determine the optimal combination of data sources and variables for wind power prediction using a random forests model. A study then further evaluates reanalysis data and methods of bias correction for this type of data, to improve power predictions at 52 wind farms across Canada using power curve and machine learning methods. Recommendations are proposed for: the use of data sources and important input variables; the utility of global reanalysis data sources by terrain features; and the utility of bias correction methods for downstream wind power prediction.