Theories And Methods Of Data Collection Considering Energy Saving And Accuracy In Wireless Sensor Networks

Wireless sensor networks (WSNs) have been paid much world-wide attention for a long time, and widely used for many application scenarios such as environmental science, intelligent household, health care, factory monitoring and disaster warning, and so on. Especially with the development of sensor technology, embedded computing and wireless communication technology in the field of smart sensor evoke the growing enthusiasm in scholars. A WSN typically has little or no infrastructure. It consists of a large number of sensor nodes working together in a self organizing way to monitor a region to collect data about the environment. The users can get support by receiving the data from WSNs. As a consequence, data collection is one of the most main works in WSNs. According to the needs of the users, data collection in WSNs can be divided into three categories:event-driven, query-driven and periodic (time-driven) that is one of the most widely used approaches.Due to restrictions in the environment and type of sensors, reducing energy consumption to prolong the network life is one of the most important topics in WSNs. Energy saving and data accuracy in periodic data collection applications are mainly considered with a higher application value in this dissertation. In the process of implementation, the unnecessary energy consumption produced by the continual data communication is reduced from three aspects:improving the prediction accuracy, removing the temporal and spatial redundancy, and maintaining the structure fidelity. The main innovation of this dissertation are as follows:(1) Based on the traditional Least Mean Square (LMS) algorithm, the hierarchical Least Mean Square (HLMS) prediction algorithm is designed, the value of the desired signal is defined by utilizing the temporal correlation of periodic data, and the significant improvements are obtained in prediction accuracy, convergence speed, and algorithm stability.(2) A cluster-based data collection framework integrating prediction, compression, and recovery to a whole is proposed for matching the clustered network architecture, spatial redundancy of nodes and temporal redundancy of data are reduced, satisfactory results are obtained both in guaranteeing users accuracy requirement for data error and decreasing network traffic.(3) The structural similarity (SSIM) based on the image quality assessment approach is introduced into the sensor nodes work/sleep scheduling, a structure fidelity data collection (SFDC) framework is proposed, the structural accuracy requirement is meet, and the amount of working nodes is also reduced to save energy.The major works of this dissertation are listed as follows.(1) The mechanism of HLMS is described in details, the optimal weight vector and mean square error on two levels HLMS are analyzed, and the HLMS dual prediction algorithms implemented respectively on the sink and the source, and an interactive data transmission protocol between them are designed. Based on theoretical analysis, the real temperature data from Intel laboratory is used to evaluate the performance including prediction accuracy, convergence speed, and algorithm stability.(2) The optimal step size least mean square (OSSLMS) prediction algorithm based on the clustered WSNs is proposed, the dual prediction protocol performed on cluster head and member nodes is designed, data compression and recovery on cluster head and sink are realized by using principal component analysis technology, and communication cost and mean square error during data transmission are analyzed. By evaluating prediction accuracy, coverage speed, saving in communication cost, mean square error and reconstructed real error, the proposed data collect framework is proved to be workable.(3) Based on structural similarity and characteristics of WSNs, a complete clustering algorithm including cluster construction and cluster head selection is designed, principles and methods for selecting the active nodes during the learning period are proposed, and the strategy by adjusting the data collection period dynamically is adopted. The simulation based on synthetic and real world datasets is implemented, the results verify the effectiveness of SFDC framework both on energy saving for large scale dense WSNs and structure fidelity requirement.