Short-term load forecasting based on CEEMDAN and Transformer

Abstract

• Mode decomposition, sample entropy and TR model are combined. • The TR-based hybrid models are used for load prediction ranging from 4h-24h. • The impact of different mode decomposition algorithms is verified. • Analysis is provided to validate the proposed model. Short-term load forecasting (STLF) is an essential part of energy plan, and it is very meaningful for energy management. Recently, some deep learning models have been popular in load forecasting. This study focuses on the Transformer (TR) model which can solve the long memory loss problem by introducing the attention mechanism. A hybrid model incorporating Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN), sample entropy (SE) and TR is proposed which is named CEEMDAN-SE-TR. To demonstrate the superiority of the proposed CEEMDAN-SE-TR model, a variety of machine learning models are used as comparison models, the one with the best performance is used to compare with the TR model and the proposed CEEMDAN-SE-TR model. Meanwhile, the Empirical Mode Decomposition (EMD) technique is used to compare with CEEMDAN and validate the superiority of CEEMDAN-SE-TR. In this study, STLF (4h, 8h, 12h, and 24h, respectively) is performed for New York city. The results show that the CEEMDAN-SE-TR gets the best forecasting results of all the comparison models, taking the forecasting result of 24h as an example, mean absolute percentage error (MAPE) is 4.80%, root mean square error (RMSE) is 1.26, R2 is 0.80, and mean absolute error (MAE) is 0.95.