基于RBF神经网络的河川年径流量预测

文章以新疆开都河年径流量为研究对象,选用能够模拟输入与输出层非线性关系的径向基函数(radial basis function,RBF)神经网络,构建了河流年径流量预测模型。研究结果表明:通过自相关系数法,选用河流自身前1~5 a径流量作为输入层,当前年径流量作为输出层,利用Matlab软件建立RBF神经网络模型,预测开都河2008—2012年径流量,预测值最小相对误差为3.22%,最大相对误差为7.61%,平均相对误差为5.19%,相关系数为0.863;通过对预测样本实测值与模拟值进行经典统计学分析,2组数据间无显著性差异。这说明RBF人工神经网络模型用于模拟预测河川年径流量是可行的。

Prediction of Annual Runoff Volume in River Based on RBF Neural Network

With the annual runoff volume of the Kaidu River in Xinjiang as the study object, in the paper, the RBF neural network which can simulate the non-linear relationship of the input and output levels is applied to build the model for the predication of the annual runoff volume of the river. The study results show that, through autocorrelation coefficient method, the runoff volume of the river in the first 1 1.5 a is selected as the input level and the runoff volume of the current year as the output level. By application of Matlab software, RBF neural network model is established to predict the runoff volume of 2008--2012 of the Kaidu River. The minimum relative error of the prediction value is 3.22%, the maximum relative error of the prediction value is 7.61%, the average relative error is 5.19% and the correlation coefficient is 0.863. Through the typical statistics analysis on the measured and simulated values of the predication samples, no outstanding difference between two groups of values is available. This demonstrates that it is feasible to simulate the predication of the annual runoff value of the river by application of RBF artificial neural network model.