Context-Aware Supporting Approximate Aggregation In Wireless Sensor Network

Wireless Sensor Network WSN, the network constituted by a number of low-energy sensornodes in an ad-hoc manner, had been widely used in the field of context monitoring. As a risingcalculation environment, it has its own advantages and disadvantages. Its advantage includes theinexpensiveness and redundancies of sensor nodes while its disadvantage includes energyconstraint and uncontrollable topology.The energy constraint of WSN leads to its considerably short life time and its uncontrollabletopology results in the intrinsic data uncertainty. Traditional research paid a lot of attention onthe energy limitation thus the paradigm focused on reducing energy consumption. It has becomea milestone since aggregation which originally belongs to the research of database wasintroduced to the field of WSN. The introduction of aggregation explicitly claims out thedatabase as a design philosophy of WSN middleware. In addition it highlights two issues: first itis the properties of objective world that interests us rather than the readings of specific nodes;second, energy consumption is proportional to the quantity of transmitted data so thataggregation can adopt in-network processing to reduce this transmission. According to these, aseries of sampling-based aggregation technology were developed.However, traditional research often neglects that second constraint, namely uncontrollabletopology. Since sensor nodes are often scattered to an area that are inaccessible to our humanbeings, topology or the distribution of WSN can not be predetermined. For a monitoring space,WSN actually acts like a measuring tool which should be calibrated before using. Differentspatial pattern will impair the accuracy of WSN to represent the true world and this thesis is tohandle this issue.(1) This thesis points out spatial pattern of sensor nodes has an effect on the accuracy ofWSN. Spatial pattern is modeled using single point extension function and information coveringfunction. Error caused by spatial pattern which is neglected by traditional sampling is discussedand modeled as the non-sampling error.(2) This thesis proposes stratified sampling to reduce the non-sampling error and theminimization of non-sampling error in stratified sampling is generalized into an optimizationproblem and an approximate solution is given. Bernoulli sampling is adopted to be a concreteimplementation in each stratum and the optimal sampling ratio is determined under the givenconfidence.(3) This thesis proposes X-connected-component based approximate aggregation algorithmand W-DBSCAN based approximate algorithm. These two algorithms are distributed and we carry out an experiment to verify their effectiveness and efficiency upon the simulation platformdeveloped by our team.(4) This thesis describe W-DBSCAN based approximate aggregation algorithm usingdeclarative language to implement the context awareness. Also we make an preliminaryexploitation to the pattern of logic language.