土壤墒情预测相关性数据耦合性试验分析

Coupling test analysis of correlation data for soil moisture prediction

为提高墒情预测的精确度,对影响墒情变化的各个变量进行试验分析,提出在剔除与墒情变化的相关性较小变量后的一种预测方法。采用皮尔逊相关系数法对影响墒情变化的蒸发量、地温、降水、气压、日照时数、气温、风速这些变量进行分析,得出影响墒情变化的较大变量,然后针对有无降水两种状态进行二分类处理,通过ROC曲线分析,分别得出在有无降水两种状态下每个变量的阈值,而后通过卡方分析,进一步筛选出在这两种状态下影响墒情变化的主要变量。最后通过线性回归分析和BP神经网络对墒情进行预测对比,结果表明,在剔除与墒情变化相关性较小的变量后,线性回归分析预测的标准偏差为12.8883,BP神经网络预测的平均误差为0.0232,二者的预测误差均低于未剔除相关性较小变量时的预测误差。

In order to improve the accuracy of moisture prediction,an experimental analysis was conducted on various variables that affect the changes in moisture content,and a prediction method that eliminates the less relevant variables with the changes in moisture content was proposed.In this paper,the Pearson correlation coefficient method was used to analyze the evapotranspiration,ground temperature,precipitation,air pressure,sunshine hours,air temperature,wind speed and other variables that affect the change of moisture content,and obtain the larger variables that affect the change of moisture content.The state was classified into two categories.Through ROC curve analysis,the threshold of each variable in the presence or absence of precipitation was obtained,after which chi-square analysis was used to further filter the main variables that affect the changes in moisture content in these two states.Finally,the prediction of moisture content was compared through linear regression analysis and BP neural network.The results showed that after eliminating variables with less correlation with moisture changes,the standard deviation of linear regression analysis and prediction was 12.8883,and the average error of BP neural network prediction was 0.0232.The prediction errors of both were lower than the prediction error when the less relevant variables were not excluded.