| As the main control building in the process of water transfer,the control gate can control the change of water volume in the channel by dynamically adjusting its opening and closing amplitude to ensure the safety of the project operation.The traditional control gate regulation method is sound,but there are disadvantages such as low decision-making efficiency and high labor intensity.With the rapid development and application of emerging technologies such as deep learning and artificial intelligence,the traditional control gate scheduling method is transformed to intelligent scheduling.In order to improve the automatic control level of the sluice group of the middle route of the Southto-North Water Transfer project and promote the process of "intelligent water transfer",this thesis takes the real-time control of the sluice group as the main research objective,and uses the method of machine learning to construct a real-time control prediction model based on historical monitoring data,so as to realize the automatic control of the sluice group under conventional working conditions.The main research results are as follows:(1)In view of the abnormal phenomenon of inverted flow of the monitoring station in the dispatching operation,the inverted data cleaning model of the flow monitoring station was constructed based on the principle of water dynamic balance and the longest sequence method of interval flow.Taking the Baihe River to Huanghe River section of the middle line project as an example,based on the flow data after model cleaning as the upper boundary condition,a hydrodynamic model was constructed to verify the cleaning effect of the model.The results show that the cleaning model solves the inverted phenomenon of flow monitoring data and improves the quality of flow monitoring data.It is found that the main reasons for this phenomenon are the monitoring data deviation of the control gate flow monitoring equipment and the interference of gate regulation.The hydrodynamic numerical simulation using the flow data after model cleaning is significantly improved compared with that before cleaning.The mean absolute error of water level in front of gate is reduced by 0.0757 m,and the root mean square error is reduced by 0.0895 m.(2)Using the control gate regulation data in the historical monitoring data,the realtime control and prediction models of gate group based on random forest algorithm and BP neural network are respectively constructed,and the control scheme generated by the prediction model is tested by constructing a one-dimensional hydrodynamic model to verify the reliability of the prediction model.The results show that the prediction accuracy of the two prediction models for the control state of the gate is up to 80%,and the R2 of the variable amplitude prediction of the gate opening is up to 0.8.The prediction effect is good.The daily fluctuation of water level in front of sluice gate by hydrodynamic numerical simulation meets the design requirements and does not exceed the design water level.The control scheme generated by real-time control prediction model is safe and reliable.(3)Based on mathematical statistics and deep learning methods,the water level prediction model after regulation of the control gate is constructed respectively to verify the feasibility and effectiveness of regulation amplitude of the control gate.The prediction results show that the prediction effect of the water level prediction model based on deep learning is better than that of the mathematical statistics method.The average absolute error of the water level before the control gate is reduced by 0.058 m and the average R2 is increased by 0.1.Aiming at the poor prediction effect of the above model,the water level prediction model driven by both data and mechanism is constructed,and the prediction effect of the improved model is tested.The prediction results show that the prediction model can solve the problem of under-fitting of some prediction models of the control gate,and improve the prediction accuracy. |
PDF Full Text Request