基于SVM的河道洪峰水位校正预报方法

针对平原河网地区河道洪峰水位预报中经验模型可靠性不足的问题,提出一种基于SVM的河道洪峰水位校正预报方法。采用谱系聚类法对历史洪水过程数据按降雨特性分类,选择与预报降雨过程最接近的历史数据训练预报模型;采用PCA对输入数据降维以提取有效特征;基于支持向量回归机建立河道洪峰水位预报模型;采用滚动模式对洪峰水位预报,每小时根据最新水位以及降水信息预报未来洪峰水位,不断提高预报精度。通过对京杭运河拱宸桥站的洪峰水位实例预测验证了该研究方法的有效性。

Correction of the Flood Water Level Forecasting Based on SVM

Due to the reliability shortfall of the empirical model for flood water level prediction of plain river network,a correction forecasting method based on support vector machine(SVM)is introduced to forecast the flood water level.Firstly,the historical flood data are classified according to rainfall characteristics by hierarchical clustering,and the historical data which has the most similar characteristics to the predicted rainfall process is chosen to train the forecast model.Then,the dimension of input data is reduced with PCA to extract effective features.Finally,the prediction model of flood water level is established based on support vector regression machine.In practical water level forecast,the peak water level is continuously forecasted according to latest water level and rainfall information every hour,which can continuously improve forecast accuracy.An experimental example on the prediction of Gongchen Bridge Station of Beijing-Hangzhou Canal is presented to demonstrate the efficacy of proposed method.