【摘要】 本文以塿土和黄绵土作为实验材料,尝试使用BP神经网络方法(Back-Propagation neural network)模拟人工降雨条件下,间隔覆盖坡面的产流产沙状况。通过设置不同坡度、降雨强度、面积比,获得各种因素不同水平组合下的实测数据;以实际降雨强度、坡度、面积比、径流起始时间和初始含水率5个因子为输入变量、坡面产流量和产沙量为输出变量,利用BP神经网络模型与多元线性回归模型对数据进行模拟分析,并检验其模拟效果。研究结果表明:训练样本集平均相对误差为18.23%,预测样本集平均相对误差为5.21%;与多元线性回归模型相比,BP神经网络模型拟合精度较高,拟合效果更理想,表现出更强的预测能力。另外,比较不同土质坡面产流量与产沙量模拟效果,塿土优于黄绵土。从本研究的结果看,BP神经网络模型应用于坡面产流产沙模拟预测,省时省力,方便快捷,具有一定的应用潜力,但其实际的模拟预测能力尚需进一步探索。更多还原

【Abstract】 Back-Propagation(BP) neural network was used to simulate runoff and sediment yield of Lou soil and Loess soil on partially covered slope exposed to indoor artificial rainfall events. Output variables including runoff and sediment yield data were obtained for treatments with different combinations of slope, rainfall intensity, and coverage area ratio.Actual rainfall intensity, slope, covered area ratio, runoff starting time, and initial water content consisted of the input variables. Runoff and sediment yield were simulated using the BP neural network and multiple linear regression. The results showed that the average relative errors of the training samples and predicting samples were 18.23% and 5.21%,respectively. The BP neural network model outperformed the multiple linear regression model in terms of fitting accuracy and predictability. Relatively, Lou soil showed higher accuracy than that of Loess soil. The BP neural network has potential in simulating runoff and sediment yield on sloping land, and the simulation capability needs to be verified in the field.更多还原