Mobile sensor networks for environmental monitoring

摘要: Vulnerability to natural disasters and the human pressure on resources have increased need for environmental monitoring. Proper decisions, based real-time information gathered from environment, are critical protecting lives resources. To this end, mobile sensor networks, such as wireless promising sensing systems flexible autonomous gathering of information. Mobile networks consist geographically deployed sensors very close a phenomenon interest. The autonomous, self-configured, small, lightweight low powered, they become when attached objects robots, people or bikes. Research has focused primarily using mobility reduce main network limitations in terms topology, connectivity energy conservation. However, how could improve monitoring still remains largely unexplored. Addressing requires consideration two aspects: sampling constraints. Sampling is about where should be moved, while constraints movements handled, considering context which carried out. This thesis explores approaches within use achieve goal, four sub-objectives were defined: Explore metadata describe dynamic status networks. Develop constraint model infer behaviour. method adapt spatial mobile, constrained sensors. Extend developed adaptive highly phenomena. Chapter 2 A was proposed general situation consists types contexts: sensor, network, organisation, each contexts describes different perspective. Metadata, descriptors observed data, configurations functionalities, used parameters what happening contexts. results reveal that suitable describing reporting back other components, users. 3 develops inferring three components: first, typology contexts; second, graph, modelled Bayesian encode dependencies same contexts, well among behaviour; third, contextual rules dependent expected constrain Metadata values monitored properties feed graph. They propagated through graph structure, most simulate behaviour obtain coverage high fire risk scenarios. It shown successfully inferred behaviour, sleeping sensors, moving deploying more enhance coverage. 4 strategy with according value (EVoI) EVoI allows decisions made location observe. minimises costs wrong predictions spatially aggregated criterion. Mobility allow move. cost-distance criterion minimise unwanted effects itself, depletion. assessed by comparing it random selection sample locations combined minimum Euclidian distance demonstrate enables informative locations, provide needed selection. 5 extends case deciding optimisation maximise new deployment at time step. demonstrated scenario simulated chemical factory released polluted smoke into open air. plume varied space because variations atmospheric conditions only partially predicted deterministic dispersion model. In-situ observations acquired considered predictions. comparison previous without performing shows optimised reduced caused poor 6 synthesises findings presents my reflections implications findings. relevant improving selecting deliver improves quality managing strategies. traditional field computer sciences mainly leads self-protection rather than protection beings By contrast, monitoring, above all performance, even thought might produce negative coverage, consumption. Thus, useful reducing constraining strategy. Although now mature technology, not yet widely surveyors scientists. operational geo-information therefore needs further stimulated. focuses informal also smart phones, volunteer citizens web. Finally, following recommendations given research: extend phenomena take account constraints; develop strategies decentralised approach; focus related data transmission; elicit expert knowledge preferences under applications; validate implementations.