A framework for short-term energy consumption prediction based on room air conditioner group characteristics. (15th July 2022)

• Record Type: Journal Article
• Authors: Xue, Kai; Liu, Meng; Ma, Mingjun; Hu, Mengqiang; Yan, Lu; Chen, Xiaoyi; Zeng, Wenmao
• Is Part Of: Journal of building engineering. Volume 52(2022)
• Publication Date: 2022-07-15
• Subjects: Building -- Periodicals; 690.05
• DOI: 10.1016/j.jobe.2022.104400
• Languages: English
• ISSNs: 2352-7102

Abstract

Abstract: Owing to the convenient installation, low energy consumption, wide applicability, and flexible operation of room air conditioner (RAC), high application potential can be achieved in terms of load dispatch or demand response by aggregating numerous single RACs' loads. The diverse behaviors of single occupants using RACs caused by different individual demands is one of the main factors leading to deviations in short-term (e.g., hourly) energy consumption prediction. In addition, owing to the lack of suitable quantitative indicators and subjective data, previous studies are rarely focused on the group behaviors of RACs, which is more appropriate for obtaining specific regional energy consumption. Aiming at optimizing the regional power grid load fluctuations in residential building, this paper proposes a framework for predicting short-term energy consumption based on RAC group characteristics by randomly selecting approximately 2000 samples from a cloud data platform. First, this study constructs multi-dimensional group features and based on the clustering method identifies three typical groups of RACs: i) Cluster a: continuous operation in summer and high energy consumption behavior; ii) Cluster b: annual intermittent operation and high energy consumption behavior; iii) Cluster c: mainly continuous operation in summer and energy-saving behavior. Second, the study generates five different aggregation sizes of RACs and integrates them with feature selection and random forest algorithms, to predict the hourly energy consumption of a residential RAC group. High precision achievement is shown by indicator indexes, such as the coefficient of determination (R 2 ) reaches 90% and the relative mean absolute percentage error (MAPE) is basically within 15%. Finally, by adjusting the group behavior of the load aggregator scenarios of the above three RAC groups in this method, peak shaving capacity of approximately 35% is realized on a typical summer day. Highlights: A framework based on group characteristics of room air conditioner (RAC) is proposed for predicting energy consumption. Three typical RAC groups are identified based on behavior of occupants using K-means clustering. High precision of energy consumption prediction of five RAC groups is realized using random forest. Using this method for application to demand response can achieve 35% peak shaving capacity. Keywords : Clustering; Random Forest (RF); Short-term energy consumption; Demand response (DR).