Sediment content in nandong underground river based on levenberg-marquardt algorithm and BP neural network

Nandong Underground River is a large-scale karst subterranean streamin southwest China. The basin witnessed a high degree of rocky desertification, a wide distribution area and serious soil erosion. The underground river is the carrier of water and soil loss in the area. The sediment concentration of the underground river can directly reflect the soil erosion in the river basin. In this paper, rainfall is taken as a given input variable, and the suspended sediment content is taken as the unknown variable to be predicted. Based on the BP neural network improved by the Levenberg-Marquardt algorithm, the suspended sediment concentration in Nandong underground river basin is modeled and predicted.Through repeatedly adjustment and optimization, three-layer BP neural network structure model with 4 (input layer dimension) -10 (number of hidden layer node) -1 (single output variable) is determined and simulated verification iss conducted. The simulation results show that the model can predict the sediment concentration with high accuracy. By analyzing the sample of error value, it is found that most of them are samples whose rainfall value is zero. It is deduced that the asymmetry of input and output is the reason of system error. The sediment concentration model proposed by this paper is a simple and effective method to quantitatively study the erosion and sediment yield. It can provide data reference for the local rocky desertification control project.