基于LS-SVM-RBF网络模型的洋河水库入库径流量模拟

针对采用传统的时间序列等方法模拟入库径流量精度不高的问题,基于最小二乘向量机(LS-SVM)与径向基神经网络(RBF)的相似性结构,利用最小二乘向量机优化径向基神经网络结构,构建了LS-SVM-RBF网络模型,模拟研究了洋河水库入库径流量的变化趋势,并选取BP网络模型和RBF网络模型作为对比模型,以决定系数、合格率、平均相对误差、最大相对误差、均方根绝对误差5个统计学参量作为模型性能评价指标。结果表明,LS-SVM-RBF模型的泛化能力与精度较BP网络模型和RBF网络模型均有较大提高,其决定系数和合格率均为最高,而平均相对误差、最大相对误差、均方根绝对误差三个指标均为最低,检验结果与训练效果相吻合,可见LS-SVM-RBF网络模型误差最小,且具有较高的可靠性。

Simulation of Inflow Runoff of Yanghe Reservoir Based on LS-SVM-RBF Network Model

The runoff simulation accuracy was not high with the traditional time series methods. Based on the similarity structure between least square support vector machine (LS-SVM) and radial basis function neural network (RBF), the structure of radial basis function neural network was optimized through least squares support vector machine. And then LS-SVM-RBF network model was built. The inflow runoff trend of Yanghe Reservoir was simulated. The BP and RBF network model was selected for comparison. The evaluation index of the model performance were selected from statistical parameters such as coefficient of determination, pass rate, average relative error, maximum relative error and absolute error of root mean square. The results show that the generalization ability and accuracy of LS-SVM-RBF model were better than that of BP and RBF model; its coefficient of determination and pass rate were the highest; the average relative error, maximum relative error and absolute error of root mean square were the lowest; test results were consistent with training; the error of LS-SVM-RBF network model was the smallest and its reliability was high.