Prediction of runoff based on the multiple quantity index of SVM
Nan Zhang · Journal of Hydraulic Engineering · 2010
A runoff prediction model was developed based on the multiple quantity index of SVM(support vector machine)method.According to the evaluation index of reservoir volume and evaporation,different temperature and rainfall parameters can be tested in the model training processes.The data at Cuntan Gauge Station in the period from 1981 to 2000 in the Upper Yangtze River were used for the model training,and 15 schemes were established.Grid search algorithm was used to find optimal regularization factors and kernel bandwidth,and to forecast the monthly runoff in 2001～2006.The results indicate that scheme 3,14,12 are in high precisions,and that of scheme 3 is the highest(RMSRE 0.11,R 2 0.89,I A 0.88,inputs of this scheme are predicted evaluation index of average temperature,reservoir volume,average rainfall and evaporation).The comparison of these 15 schemes shows that Qmax、Qmin、Tmin、Tmax will reduce the precision of forecast,the influence of Vkr is larger than Ezf.The evaluation index SVM forecasting model which is based on influence factors of multi-plans,the unification of realiges precision and usability,provides a new forecast method to the lack-data watershed.