基于模型预测的传感器网内通信数据约减方法

Traffic data mining method based on model prediction in wireless sensor networks

针对平稳感知对象和能耗敏感的一类无线传感器网络,提出了一种基于自回归模型的时间序列数据预测生成方法.通过采用少量感知对象的测量值作为样本来估计传感器节点AR模型的参数,并使用模型参数在汇聚节点内重构传感器节点的测量值,实现网内节点通信数据的约减.在误差可控情况下,减少传感器节点与汇聚节点间的通信数据量,从而降低网络整体能耗.通过实测数据仿真验证该方法的有效性.

Aiming at a class of wireless sensor network applications,which are sensitive to energy consumption and surveillance target with stationary state change process,a new method based on AR model was proposed to forecast time series data. A small amount of measurements would be taken as samples to estimate the parameters of AR model in sensor nodes,and used to forecast sensor node＇ s observation values by sink node. Under the control of errors,such observation values would be used to substitute ground truth measurements of the sensor nodes.Traffic data between nodes and the overall energy consumption in the network would be reduced. Third party data were used to simulate and verify the method to demonstrate its validity.