Study On Link Quality Prediction Mechanism For Wireless Sensor Networks Based On Support Vector Machine

Wireless sensor networks (WSNs) are composed of large numbers of cheapminiature sensor nodes which are deployed in the monitoring area. All sensor nodes aredeployed randomly and constitute a network in wireless self-organization manner. It hasa broad application prospect in medical, military, industrial, agricultural andenvironmental protection field. Link quality for WSNs communication service is afundamental issue, which can make upper layer protocols get the status of the wirelesscommunication link timely, so that it can select the better quality of wirelesscommunication link for data transmission and improve delivery success rate of the datatransmission using less energy costs, thereby increase the network lifetime.In this paper, we give a concrete analysis about the related works on WSNs linkquality. The traditional link quality prediction method is often make the current linkquality evaluation value as the link quality prediction value of the next moment,however, good current link quality can not guarantee the link quality is also good in thenext moment. Consider the internal correlation of links, representative predictiontheories like EWMA have been proposed successively. Although it can predict linkquality in a way, nonlinear fitting capability for the curve of time changing links isextremely limited and the communication links are time variable, directional andasymmetric, which often can not obtain satisfactory results.This article is dedicated to research link quality on intelligent prediction strategy toimprove the quality of the sensor network communication. A link quality predictionmodel based on SVM (LQP-SVM) is proposed according to the existing intelligent linkquality prediction theories. The LQP-SVM module can be divided into two layers:assessment layer and prediction layer. In LQP-SVM assessment layer, in order toprovide basic data and technical support for the LQP-SVM prediction layer, the linkquality assessment concept is converted to link level classification by using theadvantages of SVM on classification problem. In the LQP-SVM prediction layer,according to the historical link quality assessment results, we can predict a time unitbased on phase space reconstruction and SVR theory. Finally, we can use slidingwindow method to improve our LQP-SVM module. Experimental results show that the LQP-SVM mothod we proposed can providetechnical support for the upper routing protocols with greater precision rate and canalso improve energy cost, extend the network lifetime and achieve better real-timingperformance than traditional counting-based (PRR) link quality assessment metric. Themodel can be more accurately to predict the quality of the communication link for thenext time unit compared to the intelligent prediction method based on artificial neuralnetwork (ANN) at the same time.