Information Fusion for Real-Time Occupancy Estimation Using CO2 Dynamics and PIR Sensors in Smart Building

Real-time estimation of occupants number in buildings is crucial for managing and reducing the energy consumption of HVAC systems. Currently, estimation is mainly performed using black box models. These models require large datasets with ground truth values, which are often unavailable. Due to the difficulty to dynamically modeling occupancy change, an approach combining gray box models with information fusion techniques should overcome this issue. This study proposes a fusion of two complementary estimators. First estimator, is based on a physical model of CO2 dynamics, which has proven effective for occupancy estimation, but suffers from a certain time lag. Second estimator, is based on PIR sensor, known for their sensitivity to occupancy dynamics, but less effective when occupancy becomes static. Two widely used fusion algorithm approach has been compared: Kalman filter and Dempster-Shafer theory. In addition, four black box models were developed using the same approach in order to compare techniques. Information fusion algorithms produce similar results and provide a better representation of room occupancy than basic estimators or black box models. With a global accuracy of 89% and an MAE of 0.14 occupant.