A Method of Predicting Occupancy in Commercial Building Based on Machine Learning

Abstract

As a reference value, the occupancy guides the building Automation System (BAS) operation, which can significantly reduce energy consumption. However, the occupancy counts of commercials fluctuate dynamically with time, and how to gain the occupancy clusters and accurately predict the occupancy counts has yet to be well solved. To solve the above problems, this research proposes a prediction method for the occupancy counts of commercial buildings based on the integration of Wi-Fi connection counts data, categories of weekdays and holidays, outdoor climate data sets, and the combination of K-Means and decision tree algorithm. First, the K-means algorithm was used to cluster to obtain the representative occupancy daily clusters. Subsequently, the decision tree algorithm recognizes the clusters’ generation rules and constructs the prediction model. The validation experiments were conducted in a commercial building in Chongqing, China. The results showed that the prediction model had an accuracy of 95.24%, with better robustness than independent data sources. The prediction results can provide a practical reference for formulating BAS’s operation control and commercial operation scheme in the low carbon emission reduction environment.