陈一梅, 张梦成. 基于SVM的丁坝群束水攻沙功能预测[J]. 水利水运工程学报, 2019, (3): 25-31. DOI: 10.16198/j.cnki.1009-640X.2019.03.004
引用本文: 陈一梅, 张梦成. 基于SVM的丁坝群束水攻沙功能预测[J]. 水利水运工程学报, 2019, (3): 25-31. DOI: 10.16198/j.cnki.1009-640X.2019.03.004
CHEN Yimei, ZHANG Mengcheng. Predicting the function of spur-dike group restricting rivers based on SVM[J]. Hydro-Science and Engineering, 2019, (3): 25-31. DOI: 10.16198/j.cnki.1009-640X.2019.03.004
Citation: CHEN Yimei, ZHANG Mengcheng. Predicting the function of spur-dike group restricting rivers based on SVM[J]. Hydro-Science and Engineering, 2019, (3): 25-31. DOI: 10.16198/j.cnki.1009-640X.2019.03.004

基于SVM的丁坝群束水攻沙功能预测

Predicting the function of spur-dike group restricting rivers based on SVM

  • 摘要: 丁坝群坝田为回流区,主流区束窄流速增加,形成束水攻沙之势。在理论分析基础上,提出反映丁坝群束水攻沙功能的指标可用设计最低通航水位时的河面宽度以及宽深比表示。基于回归支持向量机理论,建立了丁坝群束水攻沙功能指标的预测模型,模型的输入因子为反映来水来沙量及变化过程、水面比降、河床形态及床沙组成等因素,输出因子为功能指标;采用试算法确定模型不灵敏参数ε、惩罚常数C和核函数参数σ。以张南水道下浅区为例,采集模型中需要的数据,基于MATLAB编程实现SVM模型训练,训练样本显示模型精度符合要求,验证得到的结果相对误差在10%以下,SVM预测模型较BP人工神经网络模型效果更佳,模型具有实用性。

     

    Abstract: The Groin field of spur-dike group is a recirculation region. In the main flow region, the width of cross-section is narrow, the flow velocity increases, and the sediment transport is accelerated.The function indexes of the spur-dike group restricting rivers are: the width of the river and the ratio of width to mean depth in section(\sqrtB / H)at the desiged lowest navigable stage. A prediction model for the function indexes of the spur-dike group was established based on the theory of regression support vector machine. Input factors of the model were the indexes of incoming water and sediment, water surface gradient, riverbed morphology and bed sediment composition, and the output factor was the function indexes of spur-dike group. A trial method was used to determine the insensitivity parameter ε, the penalty constant C and the kernel function parameter σ of this model. The shallow area of Zhangnan waterway was taken as an example and data were collected. The realization of SVM Model training was based on MATLAB programming. The accuracy of the sample display model met the requirements. The verification results show that the relative error of the simulation results is below 10%, which indicates that the SVM prediction model is more effective than the BP artificial neural network model, and the SVM model is practical.

     

/

返回文章
返回