Modeling Urban Scale Occupancy Behavior Profile Using Mobile Position Data and Analysing Energy Consumption for DOE Reference Building Types

The building sector consumes approximately 40% of all the energy in the United States. Studies find that a building’s occupancy behavior pattern is one of the primary driving forces behind energy consumption in the building. Hence, the accuracy and credibility of the occupant behavior model directly affect the energy simulation model; therefore, inaccurate results from occupant behavior models cause discrepancies between simulated and measured energy consumption. Thus, the accurate modeling of occupant behavior pattern is a significant challenge in the building energy and urban planning. In this thesis, urban scale occupant behavior profiles are modeled using mobile position data. Real world GPS logged mobile position data is used for extracting the building occupancy behavior profiles of 456 buildings in San Antonio, Texas. The occupancy profiles, which are modeled based on the U.S. Department of Energy (DOE) commercial reference building types, are obtained separately for weekdays, Saturdays and other days. The occupancy behavior profile of each building being dynamic, the average occupancy patterns for each building category must be developed. Therefore, a computational neural network model – Long Short-Term Memory Recurrent Neural Network (LSTM RNN) is proposed in this thesis for modeling occupancy profiles for each building category. The input to the network is the normalized occupancy data extracted from each building of the same category and the output of the network provides the average occupancy profile for each building category. It is benchmarked against a single-layer Feed Forward Neural Network (FFNN). The accuracy of the predicted occupancy behavior profiles, measured as the distance of predicted to the average measured profile, is evaluated using the mean absolute percentage error (MAPE). The results show that the proposed prediction model achieves an average of 20% higher accuracy than the benchmarked model. Finally, to demonstrate the effectiveness of the proposed model, an energy performance model of DOE reference building types are simulated for a full year using whole building energy simulation software – EnergyPlus. Based on the proposed occupancy behavior profile, the schedules of HVAC, lighting, and equipment operations are revised. The results indicate that the energy savings achieved in each building category are up to 12%; the retrofit is implemented without adding any cost to the existing building model. Hence, the simulation results show that the occupancy profile using GPS logged mobile position data help to improve the building model accuracy.