Predicting the electric power consumption of office buildings based on dynamic and static hybrid data analysis.

Currently, building energy consumption accounts for a considerable proportion of the world's energy consumption, e.g., 35 % in China, which has become a key concern with its rising proportion. Accurate prediction of building hourly energy consumption is key to realizing green, low-carbon, and energy-saving modern buildings. However, most of the current prediction methods are only based on dynamic data from IoT (Internet of Things) systems that may contain a large number of outliers, resulting in lower accuracy and reliability. To solve this problem, with the help of rich data provided by BIM (Building Information Modelling), we propose an approach named dynamic and static hybrid data analysis (D&S-HDA) that can provide a solution with better prediction accuracy. Technically, the novelties of D&S-HDA are fourfold. Firstly, we establish the D&S-HDA framework to obtain more accurate prediction outcomes, where the electricity consumption prediction results based on dynamic data analysis are weighted with the estimation values based on static data analysis. Secondly, in the dimension of dynamic data analysis, we design an improved temporal convolutional networks (TCNs) for parallelly outputting dynamic data samples to make the prediction results more intuitive. Thirdly, in the dimension of static data analysis, a novel concept of building hourly power consumption coefficient(BHPCC) is proposed, and its calculation method using static data analysis is designed to estimate hourly electricity consumption. Finally, in order to validate the effectiveness of our D&S-HDA approach, we design multidimensional evaluation metrics and conduct multi-group comparative experiments under different conditions, including climate models, electricity usage behaviour and time scales. The experimental results reveal that the proposed D&S-HDA approach outperforms current mainstream works in terms of root mean square error (RMSE), mean absolute error (MAE) and mean absolute percentage error (MAPE), with values of 1.5248, 1.0693 and 2.9505, respectively, which shows the efficiency and feasibility of the proposed D&S-HDA. [ABSTRACT FROM AUTHOR]