基于改进广义回归神经网络玉米田中无线传感器网络信号路径衰减预测

Path-loss prediction for radio frequency signal of wireless sensor network in corn field based on improved generalized regression neural network

为解决玉米田中无线传感器网络(WSN)节点快速合理布局问题,本文提出基于改进广义回归神经网络(GRNN)预测玉米田间无线信号路径损耗。试验中,选取433MHz和2.4GHz载波频率,依据无线电传输特性测量玉米三个不同生育期的路径衰减。以衰减值为输出期望值,生育期、发射天线高度、接收天线高度、天线增益、载波频率、通信距离这6个影响因素为输入向量建立GRNN预测模型。为提高预测精度,应用粒子群算法寻找GRNN模型最优光滑因子。所建模型的预测值与实测值的最大残差为3.50dB,最大预测标准差为1.53dB。试验结果表明,预测准确度较高,为合理部署无线传感网络节点提供依据。

Proposed in order to rationally distribute WSN nodes in corn field. Two cartier frequencies 915 MHz and 2.4 GHz were selected. Path-loss was measured by using radio frequency signal at different corn growth stages. GRNN prediction model was established with attenuation as expected output and growth period, height of emission antenna, receiving antenna height, antenna gain, carrier frequency and communication distance as inputs respectively. In order to improve the prediction precision, PSO algorithm was used to find the optimal smoothing factor of GRNN model. The maximum value of residual error difference between the model predicted and the measured values is 3.50 dB, and the maximum value of standard deviation is 1.53dB. GRNN prediction model provides possibly good guidance for WSN system configuration and node distribution in citrus orchards.