Data Modeling And Clustering Update Based On Model Driven In Sensor Network

Wireless Sensor Network (WSN) as a new mode of data acquisition and processing, has been widely used in military aviation, environmental monitoring, health care and other fields. However, sensor nodes are usually relying on energy limited battery power supply. When battery power runs out of time, sensor node will stop working. As a result, the energy problem becomes the bottleneck of the network performance. So far, when the proposed sensor network energy saving method being designed, there are two aspects mainly considered: energy consumption and data accuracy. Data from the sensor nodes in the consecutive perception stage usually have high data correlation, it shows that the continuous data sequences have redundant data, led to unnecessary data transfer and energy consumption. And there will be nodes in the actual application of deployment and repeat coverage and so on. And so forth, closer nodes tend to have high correlation data, in the process of data collection in sensor networks, is bound to generate a large number of redundant data result in an increase in energy consumption of the network.In order to solve the problem mentioned above, this thesis puts forward data modeling method and clustering update method which both based on model driven in wireless sensor network, which mainly includes two aspects:First, a novel prediction-based data collection framework is proposed to reduce redundant data transmission, where a dual regression model is designed for compressing data series. By using single variable model with the state space expression and using Kalman Filter to recursively compute the estimator of model parameter the model for describing the data overall trend is obtained. Also, an adjusting model is dealed with the short time fast changing part of the data sensor node collected. Compared with the traditional parameter transfer scheme, the framework transmits the real-time adaptive trend model and adusting model, in result that the model is more precise than the direct prediction model. The frequency of communication can be greatly reduced and hence energy can be saved by reducing data transmission and reducing the quantity of data acquisition by probability.Secondly, this thesis proposes an clustering update method based on model driven method. The nodes with strong data correlation form a group. There is a high data correlation between every two nodes. The nodes in one node group as a representative in the form of rotation, on behalf of the other member nodes for data transfer and environmental monitoring. In this paper, according to the change of the trend model of node data, the node in whether to be detected is determined. And then the region representative nodes on behalf of will be updated. While reducing the amount of communication and energy consumption at the same time, this method ensure the consistency of the node data and the accuracy of the models for the nodes in network.