Multi station water level prediction model for Shuifu to Yibin waterway based on graph neural network
Affected by Minjiang River and Hengjiang River,the water level variation characteristics of Shuifu-Yibin section in the lower reaches of Xiangjiaba are complex.The water level and flow data between the main and branch streams belong to multi-dimensional spatio-temporal data.The study selects spectral temporal graph neural network(StemGNN)for multi station water level prediction in the downstream of Xiangjiaba.The results show that the method is suitable for multi station water level prediction in the study area,and the prediction performance of the model in the next 1 hour and next 8 hours is better.The maximum prediction error at Xiangjiaba station,Yibin station and Lizhuang station is about 0.5 m.The characteristics of StemGNN are that it can automatically extract the river network structure information from the input data,reflecting the confluence characteristics of the study area.Hengjiang River flow has little impact on the water level and flow in the study area Xiangjiaba reservoir water level,Hengjiang water level and Gaochang water level represent the early stage water level of the study area.Gaochang flow,as a large flow input,has a great impact on the water level and flow of the study area.The research results can provide new ideas for multi station water level prediction in the river reach with complex water level variation characteristics,such as near dam section and tributary inflow.
water level forecastXiangjiaba ReservoirShuifu to Yibin waterwaygraph neural networkStemGNN
万方数据
维普
