喀斯特山区药用栽培植物泥沙减蚀量的神经网络模拟

Neural network modeling of erosion reduction of medicinal plants in karst mountain area

为了对药用植物的水土保持效益进行科学评价,在地处西南喀斯特腹地的毕节市七星关区岔河镇足纳村的黄壤坡地上,建立5个药用植物泥沙侵蚀监测小区,得出土壤侵蚀量与植物覆盖度、枝叶层厚度、根系条数的监测数据,以此建立3-32-1结构的BP神经网络模型。模型进行10 625次训练后达到标准误差平方和0.001的精度要求,侵蚀模数的实测值与模拟值的绝对误差在-0.02~0.03 t/hm2·a之间,相对误差基本上为零。对调查的20种药用植物进行模拟,泥沙减蚀率最少为11.8%,最大为58.73%,大部分为55%左右。通过BP神经网络,能将部分药用植物的水土流失径流小区监测结果扩展到多种药用植物,得出其水土流失减蚀量,模拟结果较为切合实际。

In order to scientifically evaluate the benefits of soil and water conservation for medicinal plants,data of soil erosion and plant coverage,leaf branch thickness,root number were obtained on 5 monitor plots of medicinal plant erosion were established on Yellow soil slop in Zuna village of Bijie city. Those plots are located in the hinterland of southwest karst area. BP neural network model of 3-32-1 structure is established accordingly. Standard error square sum of the model was up to 0. 001 accuracy requirements after 10 625 training,the absolute error between measured values and simulation of erosion modulus were between-0. 02 and0. 03 t/hm2·a,and the relative error was about zero. Investigation of 20 species of medicinal plants were simulated,the minimum erosion reduction rate was 11. 8%,and the maximum was 58. 73%,most of which was about 55%. Monitoring results of soil and water loss in runoff plots of some medicinal plants can be extended to a variety of medicinal plants through BP neural network,and the soil erosion reduction amount is obtained.