考虑上游来水影响的中长期径流预报OACSTPCD

As the limited number and uneven distribution of surface meteorological stations in Yalong River Basin,the precise rainfall data in the basin is difficult to obtain,and coupled with the limited generalization ability of traditional medium and long-term runoff forecasting models,there is a big bottleneck in medium and long-term runoff forecasting.Considering the temporal and spatial similarity of flow variation between upstream and downstream reservoirs,the principal component analysis(PCA)is conducted on the historical monthly runoff data of Jinping I Reservoir and Ertan Reservoir from 1957 to 2020,and three runoff prediction models are established by using three machine learning methods of BP artificial neural network,random forest and support vector regression,respectively.The prediction results of three runoff prediction models are evaluated by three indicators of coefficient of determination(R2),pass rate(QR),and mean relative error(MRE).The results show that the upstream reservoir has a significant influence on the inflow of the downstream reservoir,and all the three models exhibit excellent prediction performance in the medium and long-term runoff prediction of Ertan Reservoir(R2 greater than 0.8,QR greater than 0.7 and MRE less than 0.2).In general,the random forest model demonstrates superior simulation performance to BP artificial neural network and support vector regression model,and all the three models have good practicability,which can provide a data basis for the refined allocation and scientific management of river basin water resources.