Research On Efficient Continuous Data Collection Techniques For Wireless Sensor Networks

Wireless Sensor Networks (WSNs), which are self-organizing and self-managing by widely distributed sensor nodes in a monitoring region, have become a brand new information acquisition and processing approach, and been applied in various applications, including environment mon-itoring, smart city management, monitoring and preventive conservation for heritages and so on. Data collection acts as the foundation of various WSN-based applications, and is an important ap-proach to connect the physical world and the cyber world. Efficient data collection can improve the data collection efficiency of WSNs, and has been one of the hot research topics for WSNs.With the intensive studies and widely use of WSNs, the requirements of data collection in WSNs become diverse. Various network deployment scenarios ask for high performance of data collection methods on terms of energy-efficiency, scalability and dynamic adaptability. Based on the review and analysis of related works on data collection in WSNs, this thesis has proposed a set of data collection methods for different network deployment scenarios (i.e., single-hop WSNs or multi-hop WSNs) and network sensing patterns (i.e., dense sampling or sparse sampling). The main contributions of this thesis can be summarized as follows:Firstly, this thesis studies the representative node based data collection problem for single-hop and dense sampling WSNs. Due to the high spatial correlation in WSNs, nearby sensor nodes usu-ally have the similar sensing data, and thus they can be represented for each other to report sensing data to the Sink for the purpose of energy saving. Though some attempts have been made for such idea, they still suffer from poor performances, such as high energy consumption, redundant similar node groups, uneven distribution of similar node groups and the low internal data correlation of each group. Thus, this thesis proposes a measure model of sensor node’s representation capabili-ty and a representative node clustering algorithm, for solving the problems of distributed sensing similarity measure among sensor nodes and clustering of similar nodes. With our measure model, any node can assess its representation capability in a certain region based on local information exchange. By ranking sensor nodes on their representation capabilities, our clustering algorithm can quickly group the similar nodes into exclusive clusters in a distributed manner. Furthermore, a randomized intra-cluster scheduling and data restoration strategy have also been designed ac-cordingly. Simulation results show that our method outperforms existing methods on the terms of representative node clustering, energy-efficiency and data accuracy.Secondly, this thesis studies the time series analysis based data collection problem for multi-hop and dense sampling WSNs. The temporal correlation among sensing data makes the sensor data predictable, which offers a valuable opportunity for designing energy-efficient data collection method. Some existing works based on complicated probabilistic models have high complexity and uncontrollable data accuracy, while some pure time series analysis based methods overlook the spatial correlation in WSNs. This thesis thus proposes a data collection method by combining both time series analysis and spatial correlation, for solving the node similarity measure and clus-tering of similar nodes in multi-hop networks. We adopt the AutoRegressive (AR) model rather than the complicated probabilistic models to capture the data distribution of sensor nodes. By leveraging the lightweight AR models, we design a node similarity measure metric which takes both magnitude and trend similarity of sensing data into consideration. The proposed hierarchical clustering algorithm groups the most similar nodes with the node similarity measure metric, along a pre-built data collection tree. By exploiting the AR models and similar node clusters, we pro-pose a dual-prediction based data collection method. Simulations results show the efficiency of our method on the terms of energy consumption and data accuracy.Thirdly, this thesis studies the compressive sensing based data collection problem for sparse sampling WSNs (either single-hop or multi-hop networks). The major problem for sparse sampling networks, formed due to either network deployment cost or duty cycling for energy saving purpose, is the completeness of data collection. Therefore, this thesis proposes a compressive sensing based data collection framework, which solves the above problem by exploiting the recent advances in compressive sensing theory. This thesis adopts the Multiple Linear Regression (MLR) model to capture the correlation among sensor nodes, and thus designs a representation basis to sparsely represent the global sensing information. Through timely processing of sparse samplings, this thesis constructs the compressive sensing based global information recovery system, and restores the sensing data for all sensor nodes with high accuracy. Taking the traffic condition estimation of urban city as an example, this thesis demonstrates the application of our proposed method. Extensive simulations based on the practical traffic data set show that our method can recover the global and detailed sensing information from only sparse samplings, and meanwhile possesses the properties of high accuracy and good scalability.