Adaptive context modeling and situation awareness for wireless sensor networks

Environmental monitoring is becoming increasingly important. With advances in technology, sensors are now capable of sampling continuous high fidelity data. Context awareness is now realizable and necessary when we have complex situations such as distinguishing seismic activity, a rock falling, and an animal walking on a volcano. In this thesis, we developed and tested a novel adaptive context modeling and situation awareness architecture for detecting seismic tremors on volcanoes. This system allows wireless sensor networks to autonomously adapt to changes in the environment and in the network. Our system is comprised of two components an adaptable context modeling component and a dynamic situation awareness component. The purpose of our adaptable context modeling component is to provide an effective and intelligent context modeling framework that can adapt to changes in data quality. The purpose of this framework is not only to determine a specific context for a given situation, but to continually adjust itself in order to accurately model the environment at a specific instance in time based on the current state of the environment. We accomplished this by developing a three phase context model. In Phase I we developed a Bayesian network in order to determine the reliability of each data stream. We used a 0-1 knapsack optimization approach to choose the optimal subset of data. The optimal subset is input into Phase III, where we used a Hidden Markov Model (HMM) to determine if and where a tremor occurred. After careful evaluation, we were able to determine that our context model was able to more accurately determine if and where a tremor occurred, when compared to several existing algorithms. Once the context was determined, we used our situation awareness component to appropriately adjust the network resources to meet the application requirements. More specifically, within each context the priority of the individual sensor nodes, data and the current state of the network are used to ensure that relevant high priority data are collected. Our performance analysis shows improvement based in overall throughput and a decrease in packet loss for high priority data compared to traditional bandwidth scheduling algorithms