Research And Application Of Hydrologic Data Mining In Taihu Basin Based On Deep Learning

Taihu basin is a pilot basin for the construction of smart basins,which aims to provide scientific and accurate decision-making support for the optimal allocation and dispatch of water resources in the basin.The perfect basin water resources dispatching system relies on the high Time-Spatial resolution of watershed water situation forecast and effective scheduling scheme formulation method.On this basis,this paper uses the hydrological data mining technology based on deep learning to carry out the joint research on the water situation prediction model and the search of similar dispatching scenarios in the Taihu basin,and provides strong decision support for the scientific dispatching of water resources in the Taihu basin through the forecast results and similar dispatching schemes.In order to accurately forecast the change process of water situation in the Taihu basin during the foreseeable period,the spatiotemporal data integration technology is proposed to render the rainfall and water level data into the spatiotemporal data of the whole basin,on this basis,the high-resolution water level prediction models of Taihu basin were constructed respectively with or without rainfall forecast considering the forecast period based on deep learning.According to the historical water level and rainfall data of Taihu basin,the highresolution level forecast models were evaluated,and verify the precision of forecasting model,and the forecast results show that with or without rainfall forecast,the model can accurately predict the water trend of the Taihu basin in the next three days.In order to provide the support of water resources scheduling scheme for the complex river network of Taihu Basin,a water resources scheduling support model based on hydrological similar scenarios is proposed in this paper.Firstly,the similarity degree of the overall rainfall situation and other key hydrological elements in the basin between the reference period and the forecast period were calculated by the structural similarity(SSIM)and the Euclidean distance method,on this basis,the model establishes the mapping relationship of the similarity of each hydrological feature between the reference period and the forecast period of the current hydrological scenario and the historical hydrological scenario through deep learning,finally,the BORDA number of each historical scene is calculated to determine the most similar historical scene,and provide scheduling scheme transplantation or reference.Through the overall evaluation index of the model and the calculation and visual comparison of the results of three typical cases in flood season,post-flood season and nonflood season,it is proved that the model can accurately find the most similar historical scenes corresponding to the forecast period.