Occupant sensing in buildings: methods for occupant number, location and activity
Date of Issue2017-05-05
School of Electrical and Electronic Engineering
Since buildings account for some 40% of total energy usage in the world, great attentions have been paid on energy efficient buildings. To achieve this objective, occupant sensing is a key factor, which includes knowing the number of occupants, their locations and their performed activities in buildings. In this dissertation, we attempt to give solutions to the three corresponding questions of occupant sensing: How many occupants are there in a zone? Where are they located? What are they doing? Occupancy, i.e. the number of occupants, is a coarse information for the control of energy efficient buildings. We develop novel inhomogeneous Markov chain models which utilize the incremental information of occupancy for building occupancy modeling under two scenarios of multi-occupant single-zone (MOSZ) and multi-occupant multi-zone (MOMZ). In this way, we can dramatically simplify the calculation of model parameters, i.e. transition probability matrices. The proposed models have been evaluated using actual occupancy data. After that, we explore the more valuable information of real-time occupancy estimation that can be used for real-time building environment control. Since occupancy models can provide the information of occupancy pattern, we present a fusion framework of combining occupancy models with data-driven models for occupancy estimation using environmental parameters. Real experiments showed the effectiveness of the proposed approach. For some applications, such as smart scheduling, pre-heating and pre-cooling, they require the knowledge of occupancy level in future, known as occupancy prediction. We compare the prediction performance of existing occupancy models with some popular linear and non-linear data mining approaches. The experimental results allow us to deduce a guideline on how to choose a proper method for the prediction of occupancy in buildings under different prediction horizons. One of the detailed aspects in occupant sensing is the locations of occupants in buildings. Outdoor localization can be resolved using GPS (Global Positioning System). However, since GPS signals are blocked in indoor environments, the performance of GPS is greatly degraded. The most widely used technique for indoor localization is based on WiFi technology. We propose a fusion framework of WiFi, smartphone sensors and landmarks for indoor localization. Instead of using the popular WiFi fingerprinting approach which requires a labor intensive and time consuming site survey of the environment, we apply a weighted pass loss model for WiFi localization. We formulate the fusion in a linear perspective. Then, the Kalman filter which is computational light can be applied. Moreover, we use landmarks such as turns, stairs and elevators, which can be detected using smartphone sensors to further improve the performance of the proposed system. Significant improvements were demonstrated in real experiments. In situations where WiFi signals are not available, or the application is sensitive to power consumption, which makes WiFi based approaches not suitable because of the power hungry property of WiFi scanning, we present another localization and tracking system using smartphone sensors with occasional iBeacon corrections. Based on the detailed analysis of iBeacon technology, we define a calibration range where the extended Kalman filter is formulated. Real experiments have been conducted in two different environments. The experimental results demonstrated the effectiveness of the proposed approach. We also tested the localization accuracy with respect to the number of iBeacons. Another detailed aspect in occupant sensing is the activity performed by occupants. This activity information can be used to identify landmarks for indoor localization and, more importantly, to determine the metabolic rate which is a key parameter in calculating human comfort index. We propose an orientation independent activity recognition system based coordinate transformation and principle component analysis using smartphone acceleration data. The experimental results indicated that the proposed approach significantly improve the detection accuracy with regard to orientation variations. Moreover, the results also showed some improvements of our proposed approach on placement and subject variations.