An Ensemble Feature Selection Method for Short-Term Electrical Load Forecasting

The fluctuation of electrical power load is affected by numerous external factors, such as climate change, economic development, social form and special events etc. The optimal feature subset selection is the key to improving the accuracy of short-term power load forecasting. The traditional feature selection methods have encountered difficulties in balancing performance and computational consumption. The development of ensemble feature selection methods is a feasible way to solve this dilemma. In this paper, an embedded ensemble feature selection method is established and investigated on a short-term electrical load forecasting task. A new weighting approach is proposed to combine the result of individual embedded feature selection methods. The results of the embedded ensemble feature selection method show a significant increase in accuracy compared to the individual feature selection methods.