基于BP神经网络的传感器网络动态采样模型研究

Dynamic Sampling Model Research of Sensor Network Based on BP Neural Network

摘要:

能耗控制对于农业环境监测无线传感器网络系统具有重要意义；基于误差反向传播的多层前馈神经网络预测和阈值分析建立了一种土壤温度传感器网络动态采样模型,实现了基于土壤温度周期变化特征的采样频率实时调整方法,达到减少网络冗余数据,降低网络功耗的目的；以环境温度和空气相对湿度为BP神经网络实测输入,土壤温度为预测输出,通过判断输出是否进入阈值区间动态调整采样周期；仿真实验结果表明,对于具有周期性特点的土壤温度,BP网络模型对其预测值和实测值之间的均方根误差RMSE及绝对误差AE分别为0.83 ℃和0.54 ℃；相比于连续采样,阈值分析动态采样次数减少30%。

Abstract:

Energy consumption control for agricultural environment monitoring wireless sensor network system is of great significance. A dynamic sampling model of soil temperature sensor network based on BP neural network prediction and threshold analysis,for real-time adjustment of sampling frequency in the light of the periodic variation features of soil temperature,was constructed to reduce the redundant sampling data and power consumption of the network. Using environment temperature and air humidity as the actual measure input of the BP neural network,soil temperature as predict output of the BP neural network,we adjust the sampling periods dynamically by observing whether the output of the BP neural network is entering the threshold interval. The simulation results show that,for the periodic soil temperature,the root mean squared error(RMSE) and absolute error(AE) between predicted values and measured values were 0.83 ℃ and 0.54 ℃. Comparing to continuous sampling,dynamic sampling reduce the data collection by 30%.