土壤侵蚀产沙量的人工神经网络模拟

在分析土壤侵蚀产沙过程和神经网络模型特点具有某些相似的基础上 ,采用三层前馈网络模型 (BP算法 ) ,模型的第一层有 5个结点 ,分别代表降雨强度、降雨历时、降雨量、前期降雨量 (用前 1 0天降雨总量表示 )、径流深 ;第三层只有一个结点 ,表示土壤侵蚀产沙量 ;隐层的结点数采用“试错法”确定为 3个。利用四川某地水土保持试验观测资料 ,对土壤侵蚀产沙量进行模拟及预测 ,通过分析比较 ,显示了具有较好的模拟预测效果

ARTIFICIAL NEURAL NETWORK MODELING OF THE SOIL EROSION AND SEDIMENT YIELD

An artificial neural network (ANN) is a mathematical tool which is capable of modeling complex nonlinear relationships between input and output data sets and has been proved useful and efficient in many fields. The process of soil erosion and sediment yield is derived from interactions between the characteristics of soil and external factors (including natural and artificial factors); there are obviously nonlinear relationships among them. The quantities of soil erosion and sediment yield have been studied by many researchers by means of linear or nonlinear regressive methods, and then, some empiric formulas which are consisted of one or many factors have been obtained. However, the models based on physical concept can not ideally express the complex process of soil erosion and sediment yield because of many empiric simplifications. The nonlinear ANN approach can deal with the complex relationships between input and output data sets according to the collected data. After the similarity between the process of soil erosion and artificial neural networks models have been analyzed, a three layer feed forward ANN model has been founded. The structure of the model has five input nodes including precipitation intensity, duration, quantity, preceding quantity (preceding ten days) of precipitation and depth of runoff, one output nodes for the sediment yield of soil erosion, and three hidden nodes obtained by tril and error method. This model provides input output simulation and forecasting of the soil erosion and sediment yield according to the water and soil conservation experimental observation data in different fields. The results are better efficient after the relationship between simulation and observation have been analyzed.