Research Of Link Quality Prediction Mechanism Based On Bayesian Networks

In wireless sensor network, the link quality prediction is a basic issue in guarantying reliable data transmission and upper network protocol performance. High quality communication link selected by using link quality prediction ensures reliable transmission of data and raises the whole network throughput. Meanwhile, it also reduces energy consumption of nodes and prolongs the network lifetime.The research is supported by the National Natural Science Foundation, aiming at research on wireless sensor networks link quality prediction. The characteristic of the wireless network and existing link quality prediction method are analyzed, and parameters of the link quality prediction are classified. Considering volatility and asymmetry of the link, the parameters of link quality stability and link asymmetry level are redefined in this thesis. The link quality is evaluated on link signal quality, link stability and symmetry three aspects. The K-means clustering algorithm is used to discrete the parameters, and the weight of each parameter is determined by the entropy value method. Besides, closeness analysis method is employed to construct the comprehensive link quality level indicators. With the advantage of Bayesian Networks in dealing with uncertainty problem and its good performance on classification, link quality prediction mechanism based on Bayesian network is proposed. A link quality estimation model based on BN and a link quality prediction model based on Dynamic Bayesian Networks are constructed, respectively.In this thesis, WSNs are deployed in different scenarios as campus woods, roads and interior corridors to verify the validity of the proposed link quality prediction model. Large numbers of link quality samples, which are collected by our link quality test bed, are used in training the proposed model and prediction analysis. Experimental results show that the link quality level division in term of the closeness analysis is reasonable. Compared with link quality estimation method basd on Support Vector Machines, the proposed link quality estimation model based on BN has higher accuracy and generalization ability in different scenarios. Moreover, compared with the BP neural network time sequence prediction model, the proposed model based on DBN has higher accuracy.