Short-term metropolitan-scale electric load forecasting based on load decomposition and ensemble algorithms

This paper presents an ensemble algorithm based on a new load decomposition method to forecast short-term metropolitan-scale electric load. In this method, a decision tree for hourly seasonal attributes and a weighted average method for daily seasonal attributes are first applied to divide seasons into a completely different way. Then, the load of transition seasons is chosen as a basic component according to power load characteristics, and the differences between total load and the basic component are extracted as the weather-sensitive component. Finally, a time-series method is selected to forecast the basic component and SVM (Support Vector Machine) to the weather-sensitive component. This paper takes the annual electricity load of Shanghai as a case study to verify this ensemble method. The results show that compared with the traditional model based on overall daily load and other load decomposition methods-EMD (Empirical Mode Decomposition) and WT (Wavelet Transform), this ensemble model reduces the error from 3 to 5% to lower than 2% when forecasting the power load of workdays, and for non-work days, the error is decreased from 4 to 5% to lower than 4%. (C) 2020 Elsevier B.V. All rights reserved.