A non-intrusive data-driven model for detailed occupants’ activities classification in residential buildings using environmental and energy usage data

Recently, in many existing buildings, detailed energy consumption data and various indoor/outdoor environmental data could be collected by BEMS (Building energy management system) or monitoring systems. Since the characteristics of the occupants (e.g., behavior and status) affect the indoor environment, it is possible to predict occupant information by developing an inverse model that utilizes environmental data. In this study, we proposed a method for detecting detailed occupants activities based on environmental data (e.g., temperature, relative humidity, CO2, concentrations, and dust concentrations) and energy consumption data (e.g., related to heating and cooling, lighting, and electronic appliances) which can be monitored in buildings. To this end, the occupants’ characteristics (e.g., what the occupants do for 24 h and how often they do it) were investigated through experiments in which subjects live in a test-bed similar to a residential environment. In this study, we identified that seven specific activities of the occupants (sleeping, resting, working, cooking, eating, exercising, and away) accounted for 98% of the total behavior. Furthermore, the major environmental factors affecting the classification model were identified. The occupants' activities detection accuracy of the three classification algorithms including Random forest (RF), K-nearest neighbor (KNN), and Support vector machine (SVM) was compared and analyzed. The best performance of the model for the occupants' activities was achieved by RF. And the combination of environmental variables and energy use variables of the equipment led to improved accuracy of the average f1-score of 0.94. Detection for detailed activities of occupants was made possible by the development of a classification model using machine learning algorithms, which is expected to be further used for automatic control systems based on the prediction of occupants’ activities and status.