Short-term power load forecasting based on hybrid decomposition method and attention mechanism-temporal convolutional network

Abstract

Power load forecasting is a key task in power system operation and control. Due to the non-stationarity and complex randomness of power loads, it is challenging to accurately predict power loads. A short-term load forecasting method is proposed based on a hybrid decomposition approach and time convolutional network with an attention mechanism. At the outset, the load data are decomposed into a trend component and a daily fluctuation component by the ensemble patch transform algorithm. Next, the daily fluctuation components are decomposed into different frequency components by complete ensemble empirical mode decomposition with adaptive noise. Then, the trend component and the components of different frequencies are extracted through a temporal convolutional network with attention mechanism to predict the features. The comparison with the experimental results of the existing model shows that the proposed method is accurate in load prediction, which verifies the accuracy of the proposed method.