Occupancy sensor networks for improved lighting system control

In automatic lighting control systems, there can be considerable uncertainty associated with the determination of occupancy using single-point sensing, resulting in lights switching off when the space is still occupied. Current control systems compensate for this uncertainty with a time delay setting of 20 to 30 minutes to ensure user satisfaction, which can double the operating time of the lighting system controlled. This dissertation describes three studies investigating the performance of a network of linked occupancy sensors, and establishes that a network of sensors offers more accurate occupancy measurement, and greater energy savings than can be achieved with a single sensor. The accuracy of occupancy detection can be greatly enhanced by fusing the data stream from a sensor network with an analysis algorithm. Eight data fusion methods, from the simple logical functions to more sophisticated Bayesian belief network and neural network methods, were applied in this research. The most advanced of these methods, that incorporate knowledge of detector performance and typical occupancy patterns, can self-diagnose and identify faulty sensor(s), and improve the accuracy, reliability and robustness of the whole control system. The accuracy of occupancy detection (in terms of ϕ correlation, where ϕ=1 means perfect match) can be improved from 0.68 by a single sensor to 0.88 using a sensor network with a neural network algorithm. Because of the increase in accuracy of occupancy detection, much shorter time delays (e.g. 5 minutes) can be applied to the sensor network without sacrificing user satisfaction, and the operating time of occupancy-based building systems can be reduced by an extra 20%. Economic analysis shows the simple payback period of the wireless sensor network is less than 2 years, which is less than the conventional single-sensor systems, due to the much lower labor costs and the decreasing price for simple wireless sensors