Building Activity Profiling: Explainable and Predictive Modeling for Building Automation

We propose a technique to model the activities in a building, which we call Building Activity Profiling (BAP). BAP is modeled using a hybrid system composed of continuous time-series data and discrete series of states. The time-series data are provided by various sensors of a Building Automation System, and the states are derived from the characteristics of the time-series data. Unlike a simple list of data, BAP enables us to provide a better understanding of the activities in a building and a precise prediction of what will happen next. The experimental results with real office buildings show that BAP successfully perceives seasonal periodicity and state transitions of activities such as on weekdays and weekends, and predicts electrical loads more accurately than a naive method of the time-series analysis.