Short-Term Wind Power Forecasting Based on Wavelet Transform and Machine Learning Approach

Abstract

Due to high uncertainty and fast fluctuations of wind speed in the power system with high penetration of wind power generation spinning reserve scheduling and power dispatching are two major problems encountered To achieve the purposes wind power forecasting is therefore a prerequisite for the integration of a large-scale wind generation farm in the electric power system This thesis aims to present a short-term wind power forecasting method with numerical weather prediction (NWP) data integrated The forecasts are up to 4-hr ahead with a 15-min interval making the wind power forecasts suitable for addressing the spinning reserve scheduling and power dispatching problems The proposed forecasting method integrates probability distribution analysis finite-mixture model multi-resolution analysis (MRA) radial basis neural networks (RBFNNs) fuzzy inference (FI) and near real-time forecasting approaches Tested on the historical one-year wind-speed data the numerical results obtained show that the forecasting accuracy of the wind power in terms of monthly average mean relative error (MRE) and relative error root mean square error (RMSE) for a 2MW wind-generation system is 8 52% and 287 28 kW respectively The performance obtained is obviously better than the compared conventional neural networks (NNs) and MRA-NNs methods in the thesis

Date of Award	2014 Aug 8
Original language	English
Supervisor	Hong-Tzer Yang (Supervisor)