Link Quality Estimator Based On Adaptive General Regression Neural Network

Wireless Sensor Networks(WSNs)are composed of a large number of micro,low-power,dynamic cooperative sensor nodes,which transmit information by multi-hop mode.Accurate link quality estimation enables communication applications to make more efficient use of network resources,thus improving network performance in terms of delivery rate,bandwidth,throughput,and so on.It plays an important role in reducing packet error rate,delay and ensuring the reliability of communication.Firstly,under different link quality levels,this thesis analyzes the change of link quality indicator,signal to noise ratio and received signal strength indicator with time and determines candidate set of link quality metrics.On the basis of the frequent fluctuation of links in WSNs,adaptive general regression neural network(AGRNN)is used as the base classifier.In the thesis,we use back propagation to optimize smoothing factors of AGRNN.According to the smoothing factors can express the importance of link quality metrics in candidate set,an algorithm for selecting link quality metrics is proposed.Link quality estimation model based on AGRNN is constructed by using the balanced cascade algorithm.A series of base classifiers are trained until the data set samples tend to balance,and the test set samples are estimated by integration method.Indexes,F-score,G-mean,MAUC and generalization error are used to evaluate the link quality estimation model.Three typical experimental scenarios are designed,including indoor,park and motorway.The results of the link quality estimation model based AGRNN using different class imbalance processing algorithms are analyzed,and the experiment shows that the link quality estimation model based on AGRNN is constructed by using the balanced cascade algorithm which can effectively solve the problem of class imbalance,and it is necessary to study class imbalance in link quality estimation.Compared with link quality estimation models based on support vector classifier,and the link quality estimation model based on random forest,and the relation model of weighted euclidean distance and packet reception ratio respectively,the experiment results show that this link quality estimation model can achieve better evaluate performance and generalization capability.