基于人工智能计算技术的区域性土壤水盐环境动态监测

根据区域水环境监测与评价的要求，采用人工智能计算技术的BP神经网络模型进行二维区域性土壤水盐环境动态监测，仿真计算了河套灌区内两个实验区的区域土壤水盐动态值，绘制出水盐等值线图与三维透视图，将两年的预测成果对比分析。结果表明：隆胜实验区2002、2004两年春季耕作层土壤水分平均值稳定为21.0%，电导率平均值分别为0.35 ms/cm、0.42 ms/cm，有少量积盐。沙壕渠实验区两年春季耕作层土壤水分平均值分别为25.8%、21.8%，水分减少4%，而电导率明显增加，从0.44 ms/cm增至0.66 ms/cm，土壤耕层积盐明显，值得引起重视。由于BP神经网络技术对原始数据无参数及分布要求，不涉及特异值处理问题，可消除普通克立格法的平滑效应，具有较强的非线性拟合智能，对采样系统布置无严格要求，计算程序简单实用，是对常用地质统计学Kriging传统预测方法的改进，有独特的优点，可应用于大面积土壤水盐动态监测工作。

Dynamatic monitoring of zonal soil water-salt environment based on artificial intelligent technique

In order to forecast zonal water environment, the authors used BP neural network model of artificial intelligent technique to monitor two dimensional zonal soil water-salt environment dynamatically, simulated soil water-salt values of two experiment areas in Hetao Irrigation Zone, and obtainedisogram and tri-dimensional maps, analyzed the results of two years. Results show that the average water values among arable layer in the spring of 2002 and 2004 year are both 21% at Longsheng experimental areas, average EC values are 0.35 and 0.42 ms/cm, with a small amount of salt accumulation. The averagewater valuesare 25.8% and 21.8% at Shahaoqu experimental areas, with the decrease of water by 4%, but the salt accumulation is remarkable, it increased from 0.44 to 0.66ms/cm, salt accumulated obviously, this must be emphasized. The BP neural network technique had no requirement of parameter and distribution of raw data, not involving the question of outlier disposal. It can eliminate glabrous effect of ordinary krigingmethod, with strong nonlinear fitting function, and has no strict demands for sampling system, calculative program is easy and practical. Meanwhile it can improve conventional GS such as Kriging, and has special advantages. The BP neural network technique can be used to finish large area soil water-salt environment monitoring dynamatically.