Multi‐hour ahead forecasting of building energy through a new integrated model

• Published in: Environmental progress & sustainable energy Vol. 41; no. 5
• Main Authors: Cherier, Mohamed Kamal, Hamdani, Maamar, Guermoui, Mawloud, Bouchouicha, Kada, Bekkouche, Sidi Mohammed El Amine
• Format: Journal Article
• Language: English
• Published: Hoboken, USA John Wiley & Sons, Inc 01.09.2022
• Subjects: building energy simulations; forecasting; machine learning; signal decomposition
• ISSN: 1944-7442; 1944-7450
• DOI: 10.1002/ep.13823

The increase in electricity demand requires improved energy planning programs, which involve better energy distribution. Therefore, precise energy demand forecasting is of great importance for optimizing energy distribution. In this respect, a new hybrid forecasting model was proposed in this study for multi‐hour forecasting of total energy requirement (cooling and heating load of buildings) in three different regions in Algeria with different climate conditions. The proposed models are based on two main steps: In the first step, time‐series data were decomposed using Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) into a different intrinsic function (IMFs), and then extreme learning machine (ELM) is employed as essence predictors. The proposed CEEMDAN‐ELM model uses the decomposed IMFs as input data, and the total energy concern as a desired output. The integrated CEEMDAN‐ELM model was evaluated and validated on three different databases each of which had 2 years of measurement on an hourly scale. Experimental results show that the hybridization mechanism CEEMDAN‐ELM outperform the stand‐alone model (ELM) in terms of forecasting errors over the entire forecasting horizon. Forecasting results of CEEMDAN‐ELM led to a normalized root mean square error (nRMSE) in the range of [0.71–4.66] for all studied regions and horizons whereas conventional ELM provides an nRMSE in the range of [0.93–16.89].