基于数据驱动的碗窑水库洪水预报  

Wanyao Reservoir Flood Forecasting Based on Data Driven

作　　者：舒全英 郭磊 孙甜 王青青 SHU Quan-ying;GUO Lei;SUN Tian;WANG Qing-qing(Zhejiang Design Institute of Water Conservancy&Hydro-Electric Power,Hangzhou 310002,China;Zhejiang Dayu Information Technology Limited Company,Hangzhou 310002,China)

机构地区：[1]浙江省水利水电勘测设计院,浙江杭州310002 [2]浙江大禹信息技术有限公司,浙江杭州310002

出　　处：《水电能源科学》2021年第11期115-118,共4页Water Resources and Power

基　　金：浙江省水利厅科技重点项目(RB2010)。

摘　　要：为实现碗窑水库洪水智能化预报,分别建立了基于洪水场次数据驱动的BP、LSTM神经网络模型,完成了碗窑水库在不同预见期的入库流量预报。模型以过去5h降水和入库流量为输入,非滚动预报模型的有效预见期在4h以内,预报效果随预见期的增长逐渐降低;滚动预报模型的预见期相对较长,但预报效果稍逊。整体上,LSTM模型比BP模型的预报性能更优,非滚动预报模型的短期预报和滚动预报模型的长期预报相结合,能够满足碗窑水库实际洪水预报作业需求。In order to realize the intelligent flood forecast of Wanyao Reservoir,back propagation(BP)and long-short term memory(LSTM)neural network models driven by flood data are respectively established to complete the inflow forecast of Wanyao Reservoir in different forecast periods.The precipitation and inflow in the past 5 hours are chosen as the input of the model.The effective forecast period of non-rolling forecast model is no more than 4 hours,and the forecast effect decreases gradually with increase of forecast period.The forecast period of rolling forecast model is relatively long,but the forecast effect is lower.The LSTM model has better forecasting performance than the BP model on the whole.The combination of short-term forecast of non-rolling forecast model and long-term forecast of rolling forecast model can meet the demand of actual flood forecast operation of Wanyao Reservoir.

关 键 词：洪水预报 数据驱动 BP神经网络模型 LSTM模型 

分 类 号：TV122[水利工程—水文学及水资源]