基于BP神经网络的药用栽培植物泥沙减蚀量模拟

Simulation of Reduction Erosion for Medicinal Plants Based on the BP Neural Network

选择药用植物栽培种类,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,得到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,对药用植物的水土保持效益进行科学评价。模型进行10625次训练后达到标准误差平方和0.001的精度要求,侵蚀模数的模拟值与实测值的绝对误差在-0.02~0.03 t hm^(-2) a^(-1)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。

The monitoring data of soil erosion and plant coverage, litter thickness, and root number were obtained from 5 monitoring plots of medicinal plants established on slope of Yellow Soil hill, Zuna Village, Bijie City, Southwest Karst. Then, the BP neural network model of 3-32-1 structure was established to evaluate the benefits of water and soil conservation for medicinal plant cultivation. The sum of squares of standard error in the model was at the accuracy of 0.001 after 10625 training, the absolute error between the measured and simulation values of erosion modulus was between-0.02 and 0.03 t hm^-2 a^-1, and the relative error was close to zero. For the 20 species of medicinal plants, the erosion reduction rate ranged from 11.8% to 58.73%, with an average of about 55%. The amounts of soil erosion could be predicated based on the monitoring results of soil and water loss in some medicinal plants by using the method of BP neural network, and these results would be suited with actual values.