| 摘要: |
| 【目的】将遗传算法（GA）与支持向量机回归（SVR）2种算法结合，构建GA-SVR模型，并采用该模型对径流进行预报，为制定防洪抗旱与水资源调度方案提供依据。【方法】以陕西府谷县黄甫川水文站1979-2003年实测资料作为拟合样本，2004-2008年资料作为检验样本，选取降水量、蒸发量为输入量，径流为输出量，通过GA优化SVR的结构和参数，建立GA-SVR预报模型，进而进行径流预报，同时与基于误差反向传播算法的人工神经网络（BP-ANN）、投影寻踪回归（PPR）模型的预报结果进行对比分析。【结果】应用GA-SVR、BP-ANN、PPR 3个模型在径流拟合阶段的预报精度较检验阶段有所下降，但是预报精度均达到了乙级水平，其中以GA-SVR的预报精度最高，效果最好。【结论】GA-SVR模型实现了SVR参数自动化选取，较好地解决了高度非线性、小样本、过学习等问题，模型可行有效，为径流预报提供了一种新途径。 |
| 关键词: 支持向量机回归 遗传算法 径流预报 精度等级 |
| DOI： |
| 分类号: |
| 基金项目:国家高技术研究发展计划(“863”计划)项目(14110209)；国家重大科技支撑计划项目(2006BAD11B05)；西北农林科技大学博士科研启动基金项目( 01140504)；西北农林科技大学科研专项(08080230) |
|
| Mid-long term runoff forecast based on GA-SVR |
|
|
| Abstract: |
| 【Objective】In order to provide some significant references for the program of flood control and drought resistance and water scheduling,the GA-SVR forecast model established by integrating GA with SVR was utilized to forecast runoff.【Method】Hydrological data from 1979 to 2003 were chosen as a training sample and the data from 2004 to 2008 as a test sample in Huangpuchuan station in Fugu county,and then precipitation and evaporation were selected as input variables,runoff as a output variable.Structure and parameters of support vector regression were optimized by genetic algorithm,then the GA-SVR forecast model was established and runoff forecast got under way.The results were compared and analyzed with GA-SVR,BP-ANN and PPR.【Result】The results obtained by GA-SVR,BP-ANN and PPR showed that the precision of fitting was better than test,but both reached grade B.Meanwhile,the precision of GA-SVR was best and its effect was remarkable.【Conclusion】The GA-SVR achieved parameters automatic selection in application of support vector regression,solved highly nonlinear,small sample and learning problems.Overall,its model and method were feasible and effective,which provided a new way for runoff forecast. |
| Key words: support vector regression genetic algorithm runoff forecast accuracy class |
