基于LSTM 的水文站流量短期预测建模差异性研究.

Current hydrological prediction model studies lack analysis of the differential selection of modeling parameters for neural networks in different watersheds, and the adaptability of the model is low, which is not conducive to the promotion and application of the model. In this study, three sub-basins of the Yellow River Basin were selected as the research objects, and the parameter differences of hydrological stations in different characteristic sub-basins were modeled based on the discrete wavelet algorithm and Long Short-Term Memory (LSTM) model. The aim was to improve the adaptability of hydrological prediction models. The results show that for hydrological stations with good periodicity of historical flow and less human influence, a prediction model can be built based on the historical flow of the station. For mainstream hydrological stations with greater influence from upstream flow, the predictive power of the model based solely on the historical flow of the station is lower. Combining the upstream flow, it can improve the prediction accuracy of the model. Moreover, the farther the upstream hydrological station considered is from the predicted hydrological station, the longer the flow can be predicted by the model. For basins with small river flow and significant influence from groundwater and precipitation, the predictive power of the model based solely on the historical flow of the station is lower, but incorporating rainfall can improve the prediction accuracy. [ABSTRACT FROM AUTHOR]