The Research Of Multi-source Data Aggregation Model For Wireless Sensor Networks

The Wireless Sensor Networks (WSNs), recognized as one of the mostinfluential technologies in21st century, is a self-organized distributed network systemthat composed of a large number of micro-sensors. A wireless sensor network alwaysoperates in an unattended environment, and is energy-constrained of sensor nodes.This motivates the study of techniques to minimize energy consumption, enhance dataaccuracy and utilize wireless sensor nodes effectively. Data aggregation is one such atechnique.Focus on data aggregation in WSNs, this thesis developes a common aggregationmodel under multi-source data environment, and then the specific method for thepixel-level fusion between homogeneous sensors is studied with the guidance of theproposed model. Besides, the power-saving mechanism is also explored.(1) In WSNs, multi-source data obtained form different nodes representredundancy or complement property. The competition and cooperation between themcan be well explained by Game theory. However, there are some flaws of data fusionmodel based on the classic game for the “perfect rationality” of each node.Considering the possibility of elements to change over time during the whole processand the limited intial information of each sensor, the evolutionary game thought isadopted here. The improved model is given, including the formal definition, functionstructure as well as the generl description of the aggregation process.(2) Adaptive weighting method is a favorate way to address the pixel-level dataaggregation with multiple homogeneous sensors. In this thesis, based on the proposedmodel, the adaptive adjustment of weights is treated as the dynamic process ofsearching for the game equilibrium among sensors. First, we get the intial weightsrandomly. Then, each sensor can adjust its own strategy according to the “bestresponse”. After Several rounds of Game, a reasonable weight distribution is availablefinally. Priori knowledge is unnessesary for this data aggregation algorithm. And italways performs good precision and fault-tolerance.(3) For lower power consumption, we develop an improved BP neural networkoptimized by particle swarm optimization (PSO-BPNN) to predict the future according to historical data for single sensor. PSO-BPNN is deployed at both thesender and the receiver, only when the deviation between the actual and the predictedvalue at the sender exceeds a certain threshold, the sampling value and new model aresent to the receiver. The experiments show that the mechanism can effectively reducethe frequency of data transmissions.In a word, after analyzing the multi-source data environment of WSNs, thisthesis constructs a general data aggregation model, studies the aggregation algorithmfor specific level and structure under the model, and then brings a energy savingmechanism based on prediction of single node. This thesis employs the evolutionarygame theory and intelligent algorithms into the research of data aggregation for WSNs,which improves the information precision, reduces the total data transfer amount,helps to achieve the efficient data aggregation process.