高精度短期降雨预报对洪水预报和水库调度极为重要。本文以TIGGE资料中心的ECMWF、CMA及NCEP三个集合预报中心发布的桓仁水库流域预报降雨数据为基础，利用ANN、ELM以及SVM模型对桓仁水库流域未来1-3天降雨进行多模式集成预报，并从绝对平均误差、均方根误差、相对误差、纳什系数、预报确率等多个方面分析了集成预报的效果。试验结果表明，基于SVM和ELM的多模式集成预报模型预报效果均优于单一模式，基于ANN的集成预报模型在输入因子选择合适的情况下，其预报效果也优于单一模式。三种模型中，SVM模型对降雨预报精度改善最为明显。

High accuracy of rainfall forecasting is of great importance to flood forecasting and reservoir dispatching. Based on the rainfall forecasts for Huanren reservoir basin conducted by ECMWF, CMA and NCEP in the TIGGE dataset, the ANN, SVM and ELM models were developed to simulate and forecast the rainfall of Huanren reservoir basin in the next 1to 3 days, and the effect of the forecasting results was analyzed from the aspects of MAE, RMSE, Bias, NSE and prediction accuracy. Results showed that the integrated forecasting models based on SVM and ELM were better than the single models, and the integrated models based on ANN were better than the single models when the input factors were selected properly. Among the three integrated models, SVM model had the most obvious improvement in rainfall forecasting accuracy.

国家重点研发计划