No abstract

In recent years, many occupancy studies have used environmental sensor data (such as carbon dioxide [CO2], air temperature, and relative humidity) or Wi-Fi data to predict building occupancy information. However, the value of a data fusion approach that uses both environmental sensing and Wi-Fi sensing to predict occupancy remains an open question. To answer this question, this study conducted an on-site experiment in one office room in City University of Hong Kong. Three feature-based occupancy models using machine learning algorithms-k-nearest neighbors (kNN), support vector machine (SVM), and artificial neural network (ANN)-were selected to learn and predict occupancy information. In the model input, the study tested three data groups: environmental parameters only, Wi-Fi data only, and a combination of both. To assess the three occupancy models, the mean average error (MAE), mean average percentage error (MAPE), and root mean squared error (RMSE) indices were utilized. Results showed that when only the environmental parameter data were applied to learn occupancy, the ANN-based occupancy model was more suitable and accurate, and so will be with the combination of environmental parameters and the Wi-Fi data. The SVM model is more suitable and accurate in learning occupancy information with Wi-Fi data. The ANN model is more suitable environment dataset and the combination. On the other hand, the combination of datasets cannot improve accuracy significantly during three days when compared with

No abstract

Accurate occupancy prediction can improve building control and energy efficiency. In recent years, WiFi signals inside buildings have been widely adopted in occupancy and building energy studies. However, WiFi signals are easily disturbed by building components and the connections between users and WiFi signals are unstable. Meanwhile, occupancy information is often characterized stochastically and varies with time. To overcome such limitations, this study utilizes WiFi probe technology to actively scan the WiFi connection request and response between WiFi signal and smart devices in existing network infrastructures. The Markov based feedback recurrent neural network (M-FRNN) algorithm is proposed in modeling and predicting the occupancy profiles. One on-site experiment was conducted to collect ground truth data using camera-based occupancy sensors, which were used to validate the M-FRNN occupancy prediction model over a 9-day measurement period. From the results, the M-FRNN based occupancy model using WiFi probes shows best accuracy with a tolerance of 2, 3, and 4 occupants can reach 80.9%, 89.6%, and 93.9%, respectively. This study demonstrated WiFi data coupled with machine learning methods can provide valuable people count information to building control systems and thus improve building energy efficiency.