土壤墒情预测模型对比
Comparative study on soil moisture content prediction model

为实现实时准确的墒情预报，以北京市延庆区为例，利用在该地区获取的2012-2016年5年的系列土壤墒情和气象数据，对土壤墒情预测模型进行了对比研究。通过相关性分析选取时段初墒值W0、降雨、湿度、气温、气压、地温和蒸发7种影响因子，对土壤墒情分别建立线性回归方程、基于主成分分析的径向基函数（PCA-RBF）神经网络和误差反向传导（BP）神经网络3种预测模型，并对3种模型预测结果进行了对比分析。结果显示：PCA-RBF神经网络模型精度最高，平均精度达到96.8%，线性回归模型和BP神经网络模型分别为94.6%和95.7%。研究认为，PCA-RBF神经网络具有稳定性好、精度高的特点，可以很好的实现土壤墒情预测。
Real-time and accurate prediction of soil moisture content is of great significance for irrigation management and drought resistance.In order to realize real-time accurate moisture prediction,a comparative study of soil moisture content prediction model was constructed in this study.Taking Yanqing District of Beijing as an example,a comparative study of the soil moisture prediction models were constructed by using the series of soil moisture and meteorological data obtained in this area from 2012 to 2016.Through the correlation analysis,initial moisture content W0,rainfall,humidity,temperature,barometric pressure,ground temperature and evaporation were selected to establish linear regression model,PCA-RBF Neural network model and error backtracking (BP) neural network model,and the prediction results of those three models were compared.The results showed that the accuracy of PCA-RBF neural network model was the highest,with an average precision of 96.8%.The linear regression model and BP neural network model were 94.6% and 95.7%,respectively.The study showed that the PCA-RBF neural network had the characteristics of good stability and high precision,which could well predict the soil moisture content.