Water Hammer And Prediction Of Pump Full Flow Characteristics Based On Machine Learning Method

In the long-distance water transmission system or hydropower station,the changes in flow state may induce water hammer,which may lead to pipeline rupture,property loss and casualties.Due to the serious consequences of water hammer,the prediction and prevention of it in water transmission system have always been a major area of focus in hydraulics research.It is an important part of water hammer research to simulate the operational state of pipeline systems and predict the occurrence of water hammer through accurate mathematical model and reliable boundary conditions.After almost a century of research,the physical mechanism behind water hammer induction is becoming increasingly clear.As a result,numerous mathematical models have been proposed to predict water hammer.Despite the numerous mathematical models developed to predict water hammer,the complexity of the problem remains a challenge.Accurately simulating water hammer,particularly with regards to the matching boundary conditions,such as dynamic modeling of pumps and valves,is still inadequate.As a result,the existing methods for predicting the water hammer process have low accuracy,leading to water hammer accidents during production processes.In all the various reported water hammer accidents,the water hammer accidents caused by the change of pump or turbine working conditions account for a large proportion.In order to improve the prediction accuracy of water hammer model,this thesis carries out the following research work:(1)The existing calculation methods of the water hammer model are summarized and analyzed,with particular emphasis on the wave characteristic method(WCM),which is currently the most commonly used method for simulating water hammer.Through testing and comparison to laboratory experiments,it is found that the results of the numerical method are satisfactory when the first pressure shock occurs.(2)The changes in pump operating conditions is a significant factor in inducing water hammer in pipeline systems.However,theoretical calculations of water hammer require the full flow characteristic curve of the pump,which is challenging to experimentally measure.To address this issue,this thesis proposes a method that uses a machine learning approach to extrapolate the normal operational characterisitic curves of the pump to the full flow characteristic curves.The study demonstrates that the machine learning model can effectively predict the complete characteristic curve(CPC)based on data data from the normal operation zone.In the prediction experiment of specific speed of three different classes of centrifugal pumps,the maximum root mean square error(RMSE)value obtained is 0.032.(3)The WCM is used to study the transient of the pump-valve system under different valve types,valve coefficients and closing modes,with CPCs of Ns =8.06,15.7 and 20(the CPCs closest to Ns = 8.06).The valve types considered include fast,linear,and slow opening valves.The results of the transient analysis highlight significant differences between the results obtained using different CPCs and combinations of valve coefficients and closing times,underscoring the importance of selecting appropriate CPCs and valve Settings for accurate transient prediction.