基于机器学习模型的河道含沙量预测OA

Accurate prediction of sediment content in rivers plays a significant role in water intake for irrigation,river navigation and sediment discharge from reservoirs.A lightweight architecture machine learning model named WCNN(Wavelet Convolutional Neural Network)with parallel computing capabilities for river sediment concentration prediction is employed in this study,and applied to the Waizhou Hydrological Station on the Nanchang reach of the lower Ganjiang River in China.Results demonstrate that,for the sediment concentration forecasting at Waizhou Station during 2017-2018,the WCNN model outperforms the commonly used recurrent neural networks LSTM and GRU.Particularly for 1-day-ahead predictions,it achieves RMSE and MAE values of 0.003 1 kg/m3 and 0.002 0 kg/m3,with NSE and R reaching 0.887 2 and 0.946 1 respectively,indicating high predictive accuracy.Concurrently,the parameter count of the WCNN model is reduced by 54.8%and 40.2%compared to LSTM and GRU models,while training time decreased by 44.3%and 27.9%.Overall comparative analysis demonstrates that the WCNN model delivers superior prediction performance with significantly reduced computational time.