基于BP神经网络的淤地坝次降雨泥沙淤积预测

Sediment deposition prediction of warping dam in single event rainfall based on BP neural network

为了探求淤地坝在次降雨情况下的泥沙淤积量,以黄土高原丘陵区花梁坝实测数据为例,引用3层前馈型BP网络建模方法,对侵蚀性降雨条件下淤地坝泥沙淤积量进行了研究。在模型输入层变量分别为最大30min降雨强度(mm/min)、降雨总量(mm)、平均降雨强度(mm/min)和降雨侵蚀力(mm2.min),输出层变量为淤地坝泥沙淤积量,根据降雨资料和淤积信息对应关系所计算的实际资料,对网络进行了训练,并运用训练后的网络进行模拟和预测。结果表明,BP网络的绝对拟合误差和相对拟合误差均较低,绝对拟合误差最大为-0.0061万t,相对拟合误差最大为-1.2946%。同时,BP网络还具有较高的预测精度,泥沙淤积预测的绝对误差最大为-0.039万t,相对误差最大为-5.5901%。该模型的建立为土壤侵蚀产沙规律的研究提供了一条新途径。

In order to find sediment deposition of warping dam in single event rainfall, sediment deposition under erosive rainfalls was studied with Hualiang warping dam at the hilly and gully regions of the Loess Plateau as example by using BP network of three layers. The input variables of BP network were the maximum rainfall intensity in thirty minutes, the total rainfalls, the mean rainfall intensity, and rainfall erosivity;the output variable was sediment deposition. BP network was trained based on the real data calculated by using rainfall and deposition data. Sediment deposition was simulated and predicted by the trained BP network. Results show that sediment deposition simulation error is low. The maximum absolute simulation error is --0. 0061 （104 t） ;the maximum relative simulation error is --1. 2946 %;at the same time, sediment deposition prediction error is low too. The maximum absolute prediction error is --0. 039 （104 t） ;the maximum relative prediction error is --5. 5901%. The establishment of BP neural network model provides a new way to study soil erosion and sediment yield rules.