论文部分内容阅读
以广东省增城市为研究对象,采集了全市内200个土壤样点,利用BP神经网络插值方法对研究区土壤的氮和磷进行空间插值预测,将插值结果与土壤样点实测值进行对比,得到预测数据的误差均方根。结果表明,BP神经网络的插值精度比克里格高,在样点较少的情况下,BP神经网络的插值结果克服了克里格插值方法的平滑效应。BP神经网络对插值的样本数据的分布类型没有要求,比传统插值方法有更强的泛化能力,是一种可替代的插值方法。
Taking the city of Zengcheng in Guangdong province as the research object, 200 soil samples were collected from the whole city. The interpolation of BP neural network was used to predict the nitrogen and phosphorus in the soil in the study area. The interpolation results were compared with the measured data of soil samples. The root mean square error of the prediction data is obtained. The results show that the interpolation accuracy of BP neural network is higher than that of Kriging, and the interpolation result of BP neural network overcomes the smoothing effect of Krieger interpolation method with fewer samples. BP neural network has no requirement on the distribution type of the interpolated sample data and has more generalization ability than the traditional interpolation method. It is an alternative interpolation method.