On The Fault-Tolerant Event Region Detection Scheme With User-Specified Threshold In Wireless Sensor Networks

Wireless sensor networks (WSN) are the current hot research field which is in the international arena, which involves multi-disciplinary knowledge and highly comprehensive applications of communication and computer technologies. One of the most important sensor network applications is to monitor the inaccessible environment The noisy environment and energy constraints, however, challenge the event detection problem. Most of recently proposed fault-tolerant event detection algorithms could be categorized into the centralized and the distributed scheme in accordance with the way where to make up the final decision on the event. Basically, the distributed one is more suitable for the energy-constrained, bandwidth-limited wireless sensor networks. However, two known real environmental values (i.e., the normal value and the event value) are assumed in the previously proposed distributed event detection schemes, and the optimal threshold is set to be the intermediate value of the real values. However, the environmental values to be detected are diverse and the definition of events often depends on the user interest. Apparently, the fault-tolerant event region detection scheme with user-specified threshold is very important in wireless sensor networks.In this thesis, it is shown that the conventional temporal-correlation (TS), together with the temporal-spatial correlation (TSC) based fault tolerant event detection schemes could also be extended to the event region detection application with user specified threshold. The user-specified threshold and its impact on the fault-tolerant event detection performance are addressed. This thesis analyzes the performance which is in different probabilistic distribution properties of the real values, the results show that:the user-specified threshold and probabilistic distribution properties of the real values will affect the detection performance:the greater the probability distribution which further from the user-specified threshold is, the better the detection performance has and when the user-specified value is exactly the intermediate value, the different distribution does not affect detection performance because of the complementary. Besides, a new performance metric of detection confidence is proposed to provide more information without any increase in the communication overhead. It is also validated that, the temporal-spatial correlation based scheme outperforms the temporal-correlation based scheme in terms of fault-tolerant detection performance with some increase in the energy consumption. It is highlighted that, there exists an optimal threshold zone in both event detection schemes. More specifically, a larger threshold zone is preferable with the increase in noise variance and the variation range in the real values.In order to explore the event detection in time varying environment, the dynamic event detection with user-specified threshold based on the static event detection and a localized fault-tolerant event boundary base on detection confidence is addressed, the environmental characteristics of which are a time-varying. There also exists an optimal threshold parameter according to the simulation. Besides, it is highlighted that, when the user-specified threshold is outside the environmental characteristics function, the performance is getting better with the time-varying environmental characteristics is closer to the threshold; when the user-specified threshold is inside the environmental characteristics function, the performance of the threshold farther from the starting point is obviously better than the closer. And this is an interesting observation for the real deployment the wireless sensor networks for event detection in practical applications.