Improving the prediction of heating energy consumed at residential buildings using a combination of support vector regression and meta-heuristic algorithms

• Published in: Energy (Oxford) 2023-06, Vol.272, p.127069, Article 127069
• Main Authors: Khajavi, Hamed, Rastgoo, Amir
• Format: Article
• Language: English
• Subjects: Battle royale optimization (BRO); Consumed heating energy; Meta-heuristic algorithms; Residential buildings; Support vector regression (SVR)
• ISSN: 0360-5442
• DOI: 10.1016/j.energy.2023.127069

The growing population has caused to increase in energy demand worldwide. Since significant energy consumption in the residential building sector is assigned to the heating. Thus, forecasting consumed energy is effective for planning and providing future demand. However, the prediction of the consumed energy of heating systems in residential buildings due to various variables is complex. This study presents a model based on machine learning to improve the prediction of heating energy consumption in residential buildings. In this regard, Support Vector Regression (SVR) was selected as the central core of problem-solving. Then, using the hybridization technique, SVR was combined with six meta-heuristic algorithms to tune and optimize the hyper-parameters of the SVR algorithm. Finally, the proposed method's performance and accuracy compared to hybrid models by conducting a case study were investigated. The results show that the proposed method can accurately predict the heating energy consumed in residential buildings. Also, it is shown that the hybrid SVR- Battle Royale Optimization model had the best performance among all investigated hybrid models. For example, the R2 values of this model are equal to 0.999386 and 0.998898 for the train and test datasets, which are the highest values among all models. •Forecasting of heating energy consumption using SVR and meta-heuristic algorithms.•The comparison of the hybrid models by different statistic indexes.•The SVR-BRO has the smallest prediction's error value.•The proposed prediction method has the lowest time cost and overfitting.•The best R2 value is achieved in the SVR-BRO of about 0.998898.