Study On Data Fusion Based On Compressed Sensing For Meteorological Sensor Networks

Meteorological Sensor Networks have characteristics of small size, low cost, easy to deploy, less impact for environment. It is suitable for meteorological environment monitoring to collect meteorological information of monitoring objects. Meteorological elements have characteristics of time-space continuum such as temperature, humidity and nodes are usually deployed densely. So, the information collected by sensors involves data of huge volumes, high spatial-temporal correlation and high redundancy. Nodes have characteristics of limited energy and bandwidth resources, nodes will cost lots of resources if transmit all redundant sensed data. Data fusion is a effective way to decrease amount of data transmission. Compressed sensing has opened a new avenue of data fusion in meteorological sensor network. The major works of this paper are about data fusion based on compressed sensing for meteorological sensor networks by exploiting the Spatial-temporal correlation of sensed data.The structure of meteorological sensor networks and data fusion technology is firstly summarized. The function of data fusion in meteorological sensor networks is analyzed. Then, Compressed sensing is summarized. Besides, existing data fusion technology based Compressed Sensing and its performance is analyzed. These studies provide preparation for the next step research of data fusion based on compressed sensing.Aiming at the spatial-temporal correlation of sensed data, Compressed Sensing based Spatial-temporal Data Fusion model for meteorological sensor network (CSSDF) is proposed. The model extends sensed data in the cluster from spatial to temporal domain and fuse sampling data in several rounds in cluster heads by building block-diagonal random measurement matrices to decrease the amount of data transmission and latency. Then, the impact on compression and reconstruction of sparse abnormal data is analyzed and Corresponding processing method is been proposed. Simulation result shows that comparing with compressed sensing with spatial correlation and state without processing, CSSDF obtain higher compression performance and probability of recovery. Sparse abnormal data could also be processed in time.Aiming at the spatial-temporal correlation of sampling data and limited node bandwidth resources, Compressed Sensing based Energy Balanced Data Fusion algorithm (CSEBDF) is proposed. The fused result could be got by selecting nodes randomly in every cluster in each sampling round. Residual energy of nodes is taken into account to get balanced energy consumption. Abnormal data could also be processed by this algorithm. All results of fusion in several sampling rounds are reconstructed jointly based on spatial-temporal correlation of sensed data. The results of experiments show that comparing with CS-LEACH, CSEBDF could achieve greater recovering performance less nodes in every round and longer network lifetime in the same condition.