Abstract:

The traditional short-term load forecasting methods are mostly data-driven machine learning methods,and the application scenarios are mostly macroscopic city/county regional total load forecasting.In the face of the distribution transformer load,the prediction effect is obviously insufficient.In this regard,a load forecasting method based on EMD-Stacking-MLR is constructed.Firstly,the distribution transformer load data are decomposed into finite intrinsic mode function components from high to low frequency by empirical mode decomposition method.The high and low frequency components are divided according to the sample entropy value.Subsequently,Stacking multi-model fusion method and multiple linear regression method are used to predict the high and low frequency components respectively.Finally,the final distribution transformer prediction load curve is obtained by superposition of every components prediction results.Through experimental verification,the results show that this method has achieved remarkable results in improving the load forecasting accuracy and model generalization ability.

Key words: empirical mode decomposition(EMD), Stacking ensemble learning, MLR, short-term load forecasting, distribution transformer