Distribution and correlation analysis of typical features of electricity use profiles in non-residential buildings

Building electrical system will play a critical role in the demand response of future distributed energy systems. Understanding the distributive characteristics of the physical features of building electricity use profiles provides a useful reference for planning and designing building electrical systems. This study proposed a methodology and framework for analyzing the distribution and correlations of physical electricity use features. Five physical features were extracted from the electricity use profiles of the buildings. For value-based features, including the weekend ratio and variation factor, distribution fitting and feature regression were conducted to reveal correlations with attributes such as building types and sizes. For curve-based features, including the powertemperature curve, weekday and weekend schedules, clustering analysis was first used to identify several typical curves, and CART classification models were developed to establish correlations between typical curve types and building attributes. The proportions and distributions of all features were analyzed based on a dataset of 4168 non-residential buildings in Jiangsu Province, China. The K-S test p-values of the fitted distributions of weekend ratios and variation factors were 0.074 and 0.189, respectively. The overall accuracies of the CART classification model for power-temperature types, weekday schedule types and weekend schedule types were 62.7 %, 66.3 % and 56.7 % respectively. Potential application procedures and tools were provided for practical use. Variations in weekday and weekend schedules are discussed, and future research perspectives are outlined. The proposed methodology and results provide useful support for the planning and design of distributed building electrical systems.