Risk Control of Energy Performance Fluctuation in Multi-Unit Housing for Weather Uncertainty

• Published in: Buildings (Basel) Vol. 13; no. 7; p. 1616
• Main Authors: Wang, Jiahe, Mae, Masayuki, Taniguchi, Keiichiro
• Format: Journal Article
• Language: English
• Published: Basel MDPI AG 01-07-2023
• Subjects: Air conditioning; Algorithms; Architecture and energy conservation; building performance gap; Canada; Clustering; clustering algorithm; Datasets; deep learning; Economic development; Energy consumption; Energy efficiency; Energy use; Green buildings; Housing; Japan; Mean; multi-unit housing; Multiple dwellings; Neural networks; Normal distribution; Optimization; Population density; Probability distribution; Risk management; robust optimization; Robustness; Sampling; Sampling methods; Simulation; uncertain weather conditions; Uncertainty; Urban development; Weather; Canada; Japan
• ISSN: 2075-5309; 2075-5309
• DOI: 10.3390/buildings13071616

With the acceleration of urban development, the population density of urban cities has increased. As the spatial characteristics of multi-unit housing (MUH) perfectly fit this developmental trend and, simultaneously, have high energy efficiency, the number of MUHs has increased rapidly in recent decades. Although many studies have proposed high energy efficiency strategies, weather uncertainty leads to errors between the operational performance of building energy and simulated values. This study introduces a robust optimization framework that incorporates uncertainty considerations into the optimization process to suppress energy consumption fluctuations and improve the average building energy consumption performance. Neural networks are used to model the uncertainty of multiple weather elements as normal distributions for each hour, and the accuracy of the uncertainty model is validated by calculating the mean absolute percentage error (MAPE) between the mean values of the distribution and the measurement values, which ranges from 3% to 13%. The clustering algorithm is proposed to replace the sampling method to complete the sampling work from the normal distribution space of the weather elements to serve the subsequent optimization process. Compared with the traditional method, the sampling results of the clustering algorithm show better representativeness in the sample space. The robust optimization results show that the average energy consumption of the optimal scheme decreases by 13.4%, and the standard deviation decreases by approximately 17.2%, which means that the optimal scheme, generated by the robust optimization framework proposed in this study, has lower average energy consumption results and a more stable energy consumption performance in the face of weather uncertainty.