基于RBF神经网络的土壤侵蚀预测模型研究

Research on Soil Erosion Prediction Model Based on RBF Neural Network

土壤侵蚀的物理机理十分复杂,用数学方式难以描述。针对土壤侵蚀过程的模糊性、随机性、非线性等特点,将RBF神经网络的理论与方法应用到土壤侵蚀预测中。以杏木小流域为研究对象,应用RBF神经网络方法构建土壤侵蚀预测模型,以汛期降雨量、径流系数、土壤容量、有机质含量及孔隙度土壤侵蚀因子作为模型的输入层变量,输出层变量为年土壤侵蚀模数。通过模拟训练和预测,RBF神经网络取得的结果较好,能够有效地预测土壤侵蚀,且与常见的BP神经网络土壤侵蚀预测模型相比,RBF神经网络得到的预测结果精度更高。RBF神经网络模型将土壤侵蚀预测问题转化为影响因子和年侵蚀模数的非线性问题,该模型的模拟与预测为复杂的土壤侵蚀规律研究提供了新途径。

The physical mechanism of soil erosion is so complicated that it is difficult to be described by the mathematical mode. According to the characteristic of vagueness, randomness and nonlinear of soil erosion process, the RBF neural network theory and method are applied to soil erosion prediction. With Xingmu small watershed as the research case, the application of RBF neural network method was adopted to construct soil erosion prediction model, and flood season rainfall, runoff coefficient, soil capacity, organic matter con- tent and porosity and so on were used as input layer variables, and yearly soil erosion modulus were used as the output layer variable. Through the simulation training and the forecast, results obtained through RBF neural network were precise, RBF neural network could be used as soil erosion prediction model, compared with the traditional BP neural network, RBF neural network could give the higher accuracy prediction re- suits. RBF neural network model shifts soil erosion prediction problem into the impact factor and erosion modulus nonlinear problem, the model of the simulation and forecast provides a new way to complex law of soil erosion research.