A COMPARATIVE STUDY OF FEATURE SELECTION METHODS FOR WIND SPEED

Abstract

Forecasting wind speed or capacity of wind power is playing an important role to serve the problem of resource mobilization of the power system. However, forecasting is still a difficult problem because there are many factors affecting wind speed. In this paper, our goals are using some feature selection methods to find the best approach as well as to select the meteorological parameters that have great influence on wind speed, thereby helping to improve our predictive model. Pearson’s Correlation, Random Forest, Boruta were 3 feature selection methods to be used on 2 weather datasets in 2 different locations. Firstly, each data was analyzed with separate autocorrelation and partial autocorrelation analysis. From this, the hysteresis characteristics of the data were obtained then added to the methods. Following that, we carried out and compared the performance of the feature selection methods based on the evaluation criteria of each method. The results show that the wind speed depends heavily on the lags closest to it and in different geographical locations gives different results.