A prediction approach with mode decomposition-recombination technique for short-term load forecasting. (October 2022)

• Record Type: Journal Article
• Authors: Yue, Weimin; Liu, Qingrong; Ruan, Yingjun; Qian, Fanyue; Meng, Hua
• Is Part Of: Sustainable cities and society. Volume 85(2022)
• Publication Date: 2022-10
• Subjects: Short-term load forecasting -- Mode decomposition-recombination technique -- Feature selection -- Long short-term memory networks -- Bayesian optimization algorithm; Sustainable urban development -- Periodicals; Sustainable buildings -- Periodicals; Urban ecology (Sociology) -- Periodicals; 307.76
• DOI: 10.1016/j.scs.2022.104034
• Languages: English
• ISSNs: 2210-6707

Abstract

Highlights: A mode decomposition-recombination method is used to pre-process original data. The optimal input variables are selected by Spearman rank correlation coefficient. LSTM network with Bayesian optimization is used to build the prediction model. The influence of each sub-sequence on the predicted result is given. Abstract: Short-term load forecasting (STLF) is critical for ensuring smooth and efficient functioning of power systems. In this study, a prediction approach, combining ensemble empirical mode decomposition (EEMD), permutation entropy(PE), feature selection(FS), long short-term memory(LSTM) network, and Bayesian optimization algorithm(BOA), is proposed to enhance the accuracy of load forecasting. Firstly, EEMD is used to preprocess the original electricity load series and obtain the different frequency components. Then, PE is introduced to distinguish the complexity of each component and reconstruct components with similar entropy values to obtain a new set of sub-sequences. The Spearman rank correlation coefficient is then employed to determine optimal feature sets for new sub-sequences. Subsequently, LSTM networks are used to establish prediction models for the new sub-sequences, and the BOA is applied to identify the hyperparameter in the LSTM network. Finally, the predicted results of each component are superimposed to obtain the total prediction result. Through the analysis and study of different cases, the results show that the prediction performance of the proposed model is better than the comparable models. Furthermore, this study discusses the influence of each sub-sequence after mode decomposition-recomposition on the prediction results, and reasonable explanations are given for the physical meaning of each component.