基于BP神经网络与SVM模型的黄尾河径流预测比较分析

比较分析BP神经网络与SVM模型在径流预测应用中的性能特征。以降雨量为预报因子，采用BP人工神经网络模型和SVM模型对大别山黄尾河流域40 a时长的同期径流过程进行数值模拟，并对二者的预测性能进行比较与评价。结果表明，黄尾河流域BP模型模拟的总体相对误差为14.43%，合格率为77.5%，确定性系数为0.76，预报精度等级为乙级；SVM模拟的总体相对误差为12.41%，合格率、确定性系数及预报精度等级与BP模型相同。SVM模型模拟结果较BP模型而言更集中于较小的误差范围内。BP模型的累积误差>SVM模型，并且随着误差自由度的增大，这种差距有扩大的趋势，表明SVM模型的误差范围较小，误差间隔小于BP模型，模拟性能较BP模型更稳定。

Comparative Analysis of Huangwei River Runoff Prediction Based on BP Neural Network and SVM Model

To compare and analyze the performance characteristics of back propagation (BP) neural network and support vector machines (SVM) model in runoff prediction.With rainfall as the forecast factor,BP artificial neural network model and SVM model were used to simulate the runoff process in the Huangwei River Basin of Dabie Mountain for 40 years.The prediction performances of the two models were compared and evaluated.The overall relative error of the BP model was 14.43%,the percentage of pass was 77.5%,the deterministic coefficient was 0.76,forecasting accuracy level was b; the overall relative error of the SVM model was 12.41%,and the percentage of pass,the deterministic coefficient and forecasting precision level were the same as the BP model.The SVM simulation results compared with the BP model were more focused on small error range.The cumulative error of BP model was greater than the SVM model,and with the increase of error of degree of freedom,the gap had a tendency to expand,showing that the SVM model of the error range,smaller error interval were less than BP model,the simulation performance was more stable than BP model.