Wind Speed Multi-Mode Ensemble Forecasting for Wind Farms Based on Machine Learning

Abstract

[Introduction] With the extensive construction of wind farms, the combination of researches on different machine learning algorithms and meteorological forecasting modes has received widespread attention. [Method] This paper was based on the spatial distribution characteristics of wind energy resources in Hubei Province, and utilized representative stations in combination with experimental data analysis to conduct in-depth discussions on the results. [Result] The wind farms in operation and under construction in Hubei Province are all located in the "Three Zones and One Area", including the north-south wind zone from Jingmen to Jingzhou in the central part of Hubei Province, the east-west wind zone from Zaoyang to Yingshan in the north of Hubei Province, certain lake islands and zones along the lake, as well as some high mountainous areas in the southwest and southeast of Hubei Province. This research uses four different numerical forecasting products, namely CMA-WSP, CMA-GD, WHMM, and EC, to compare with the measured wind speeds and investigated the applicable range of these four numerical modes. [Conclusion] By analyzing the performance of five ensemble forecasting methods based on machine learning and the mean method, we identified suitable algorithm and forecasting model combinations, providing references for improving the accuracy of ensemble forecasting.

Keywords

• wind power day-ahead forecasting
• machine learning algorithms
• random forest
• lightgbm
• adaboost
• gru
• lstm
• ensemble forecasting.