Recognizing occupant presence status in residential buildings from environment sensing data by data mining approach

Abstract

Actual information of occupant presence is an essential input for the simulation of building energy use and home automation system. Previous studies installed motion sensors, cameras, environment sensor networks or smart meters to detect the occupant presence status directly or indirectly in offices. These approaches would raise the household cost and privacy concerns when were applied to residential buildings. Recently, home indoor environment monitors are becoming a part of daily life with the population of smart home products, and massive indoor environment data in the residential buildings could be collected remotely. Therefore, we conducted an experiment in three bedrooms that were already installed with home indoor environment monitors, based on the hypothesis that information of occupant presence could be extracted from indoor environment data by appropriate data mining approach. Two algorithms were designed in the experiment for the data mining, where one recognized the presence status directly from the environment data, and the other indirectly from the usage information of light and AC that were recognized from illuminance, air temperature and relative humidity. Results showed that it was difficult to develop an algorithm that performed well both on the training and testing datasets when we expected to directly mine the occupant presence from indoor environment parameters, but the performance could be greatly improved when we added the bridge information of light and AC operations into the algorithm, and then the recognition accuracy could be improved higher than 92% in all the three rooms.