Extreme Learning Machine-based Link Quality Estimation Method For Wireless Sensor Networks

Wireless sensor networks(WSNs)are a type of multi-hop and self-organized network composed by a large number of sensor nodes deployed in the monitoring area.Effective link quality evaluation provides a reference for routing protocols to select high quality links for data forwarding.It can not only guarantee the stability of data transmission,improve the network throughput,but also reduce the number of retransmission data,and then reduce the node energy consumption and prolong the network lifetime.The research is supported by the National Natural Science Foundation,aiming at research on wireless sensor networks link quality estimation.This thesis studies the link characteristics of wireless sensor networks,and analyzes the existing link quality estimation methods.Considering the timeliness and low energy consumption of link quality estimation,the method of evaluating the link quality based on hardware parameters is determined.Based on the data set,this thesis study the correlation between hardware parameters and packet received rate,and choose received signal strength indica tor,link quality index and signal noise rate as link quality comprehensive evaluation parameters.The link quality level is divided according to the packet received rate,which is used as a link quality evaluation index.On this basis,a link quality evaluation model based on the extreme learning machine(ELM)is proposed.In view of the noise in the link quality sample,a denoising method based on kalman filter is used.Aiming at the imbalance of link quality samples,the sample weighted and regularization method are used to optimize ELM,and propose a data-driven parametric adaptive model training algorithm based particle swarm optimization.G_mean value and precision are used to evaluate the link quality estimation model.In the thesis,large numbers of link quality samples,which we collect from experiments,are used to analyze the temporal and spatial characteristics of link in different scenarios,including squares,campus forests,and indoor corridors.The experimental results show that the evaluation performance of the model based on the Sigmoid activation function is superior to the model based on Sine or Hardlim activation functio ns.ELM,Weighted ELM(WELM),and Improved WELM algorithms are employed to construct link quality estimation model.The experimental results show that the model with improved WELM algorithm achieves best evaluation performance.Compared with the model based on support vector classification machine,the model proposed in the thesis has higher precision and G_mean value.