Universities power energy management : A novel hybrid model based on iCEEMDAN and Bayesian optimized LSTM

Research output: Journal Publications and Reviews (RGC: 21, 22, 62)21_Publication in refereed journalpeer-review

16 Scopus Citations
View graph of relations

Related Research Unit(s)


Original languageEnglish
Pages (from-to)6473-6488
Journal / PublicationEnergy Reports
Online published8 Oct 2021
Publication statusPublished - Nov 2021



Rapid growth and development around the world will lead to a gradual increase in electricity consumption. At present, colleges and universities have become the primary unit of daily electricity consumption. Therefore, accurately predicting the power consumption of colleges and universities is of great significance to the energy conservation and emission reduction of colleges and universities. Taking the actual power consumption of colleges and universities as an example, this article first analyzes its power consumption data characteristics. Based on the analysis and ``decomposition and integration'' concept, this paper proposes a hybrid network based on the improved complete ensemble empirical mode decomposition with adaptive noise (iCEEMDAN) and long short-term memory (LSTM) to achieve accurate colleges and universities Short-term load forecasting. First, the original power consumption data is decomposed into a series of patterns with noticeable differences by iCEEMDAN. Then use Bayesian-optimized LSTM to predict each mode individually. Finally, the prediction results of each mode are superimposed and reconstructed to form an overall prediction result. In each training, the Bayesian optimization algorithm is used to select the most suitable LSTM hyperparameter values to match the data characteristics of each model. At the same time, the structure of the LSTM prediction large data set is discussed. The results show that, compared with the prediction errors of other models, the proposed hybrid model can accurately predict university power consumption and provide the highest prediction ability among all survey models. (C) 2021 The Authors. Published by Elsevier Ltd.

Research Area(s)

  • iCEEMDAN, Long short-term memory (LSTM), Bayesian optimizer, Short-term load fore-casting, University power consumption, Deep learning, SHORT-TERM, NEURAL-NETWORK, TIME-SERIES, ELECTRICITY CONSUMPTION, FORECASTING-MODEL, LOAD, DECOMPOSITION, PREDICTION, BUILDINGS

Download Statistics

No data available