Day-ahead Prediction of Building Occupancy using WiFi Signals

摘要

Advance knowledge of occupancy in commercial buildings facilitates implementation of occupant-centric control schemes that reduce energy use and increase comfort. However, training and validation of occupancy prediction models can be challenging since ground truth data is not always easily obtainable. In fact, not only is the collection of ground truth costly because of the manual labor involved, it might be restricted in time and space for security and privacy reasons. As a result, prediction based on semi-supervised learning techniques using limited ground truth data can be a promising approach with a slight compromise on accuracy. In this paper, an innovative method for day-ahead prediction of total building occupancy is proposed which leverages the opportunistic probing signals from a WiFi network. Using only two days of ground truth occupancy data, a model based on a combination of linear regression and artificial neural networks is able to predict day-ahead occupancy count with 90 percent accuracy.