Research On River Water Level Prediction Based On Deep Learning

As an indispensable element in the study of hydrological changes,river level prediction plays a vital role in the production and life of human society.Based on historical experience,flood disasters occur frequently along rivers,and river level prediction can make early judgments on flood disasters and provide guiding data for disaster warning and prevention and control,which is beneficial to shipping management and water safety construction.In view of this,this paper analyzes the important factors affecting water level on the basis of previous water level prediction methods and establishes a river level prediction model by combining deep learning related technologies as follows.(1)A two-branch short-term rainfall prediction model based on Bi GRU-Caps Net and Transformer is proposed.The model performs comprehensive learning of time-series deep features by Bi GRU-Caps Net,while taking advantage of Transformer to establish remote dependency and multi-headed attention mechanism for feature extraction,and fuses the extracted features to achieve rainfall level prediction and lay the foundation for quantitative processing of meteorological data.Through ablation experiments and comparison experiments,the results show that the proposed model achieves better results in accuracy,precision,recall and F1 score.(2)A short-term river water level prediction model based on attention mechanism and Bi GRU-Caps Net is proposed to predict river water level on the basis of multivariate.Firstly,the quantitatively processed meteorological data and historical water level,flow rate and rainfall are used as model inputs,then deep features are extracted based on the previous and next moment states using Bi GRU,based on which influential features are extracted by the attention mechanism and input to the Caps Net for comprehensive learning of the extracted features,and finally the prediction results are output through the fully connected layer.The simulation experimental results show that the proposed model achieves better results in the short-term prediction of river water level.(3)A medium and long-term prediction model of river water level based on self-attentiveness mechanism is proposed for Conv LSTM and TCN.The model uses the self-attentive mechanism to record the location relationship between information and assign higher weights to the influential features;Conv LSTM and TCN network are used to learn the data temporal relationship while giving the model the ability to extract local features,and finally the prediction results are output through the fully connected layer.The simulation results show that the proposed model achieves good prediction results in RMSE and MAE,and can realize the medium and long-term prediction of river water level.