基于ANN的多路径传输WSNs数据丢包率预测

摘要

In order to improve the accuracy of WSN data transmission, a method based on artificial neuron network ( ANN ) is designed for the prediction of data packet loss ( DPL) rate. Through predicting the packet loss rate of WSN data packets in multipath transmission,the optimal data transmission path is selected. Firstly, a topologic structure of the network with 150 nodes that are distributed randomly in a fixed area is established,and hierarchical clustering of nodes is conducted by PSO algorithm. Then,the DPL of data transmission is predicted by using a three-layer feed-forward ANN. A momentum correction factor is introduced in BP network learning algorithm,to avoid ANN training process traps into local minima. Three of the inputs of ANN are respectively related to the node number of transmission path,types of node application environment,and the node sampling rate. Finally,the multipath transmission planning method based on BWAS is used;the optimal transmission path is selected with the transmission path data packet loss rate as the evaluation parameter. From the simulation verification,it is found that the method proposed obtains higher accuracy of data transmission;it has obvious advantages than that of the FEC encoding multipath transmission strategy.%为了提高无线传感器网络(WSNs)数据传输的准确率,设计了一种基于人工神经元网络(ANN)的数据丢包(DPL)率预测方法.通过预测WSN数据包在多个传输路径上的丢包率,选择出最优数据传输路径.首先,建立1个150节点的网络拓扑结构,使各节点在固定区域内随机分布,并采用粒子群算法对各节点进行层级聚类.然后,采用1个3层前向人工神经元网络来预测数据传输的丢包率.在BP网络学习算法中引入动量修正因子,避免了ANN训练陷入局部极小的问题.ANN的3个输入分别为传输路径上的节点数目、节点应用环境类型和节点采样频率.最后,采用基于最优最差蚂蚁系统(BWAS)的多传输路径规划方法,以传输路径数据丢包率作为评价参数,选择最优传输路径.经仿真验证可知,本方法能够获得较高的数据传输准确率,与FEC纠删编码多路径传输策略相比具有明显优势.