论文部分内容阅读
基于VC维和结构化风险最小理论的支持向量机方法因具有较好的学习和泛化能力而在预测预报领域得到广泛的应用。文中选取当月平均降水量、上月平均降水量以及当月平均相对湿度、平均最高气温和平均最低气温五个预报因子,采用Gridsearch算法优化参数,建立了基于支持向量机的月径流预报模型,并将其应用于石羊河流域八个子流域,定量分析其适用性。结果表明:模型在率定期和验证期模拟的平均Nash-Sutcliffe效率系数分别为0.831和0.806,相对误差分别在6%和5%以内;除个别峰值模拟较小之外,流量序列整体模拟效果较好;模型在丰水时段模拟值小于实测值,枯水时段模拟值大于实测值,在平水时段和枯水时段的模拟效果要优于丰水月份。因此,支持向量机模型在石羊河流域具有较好的适用性,可用于该流域的中长期水文预报。
SVM based on VC dimension and structured risk minimization has been widely used in the field of forecasting because of its good learning and generalization ability. In this paper, we select five forecasting factors, such as average precipitation in the month, average precipitation in the previous month, average relative humidity in the current month, mean maximum temperature and mean minimum temperature. Gridflow algorithm is used to optimize the parameters and a monthly runoff forecast model based on support vector machine It is applied to eight sub-basins in Shiyang River Basin to quantitatively analyze its applicability. The results show that the average Nash-Sutcliffe efficiency coefficients of the model are 0.831 and 0.806, respectively, with the relative errors of 6% and 5%, respectively. In addition to the small individual peak simulation, the overall simulation results of the flow series The simulation value of the model is less than the measured value in the wet season, the simulated value of the dry season is greater than the measured value, and the simulation effect is better than the wet month in the flat water period and the dry water period. Therefore, the SVM model has good applicability in the Shiyang River Basin and can be used for mid-long term hydrological forecast of the basin.