Health State Evaluation Of Electric Submersible Pump

As an efficient and reliable artificial lift tool,the Electrical Submersible Pump is widely used in the petroleum industry.Electrical submersible pumps are the second most common method of artificial lift in the world.It is estimated that the number of oil wells deployed with ESP is between 150,000 and 200,000 worldwide,and the crude oil production of ESP wells accounts for about 10% of the world’s crude oil production.However,the ESP system has a complex structure and a high degree of integration.Once a failure occurs,it will lead to production interruption,which in turn will lead to huge economic losses.Therefore,it is very necessary for the production management of ESP wells to prolong the operation cycle of ESP and improve the safety performance of ESP systemAt present,the means of prolonging the service life of the ESP are mainly to diagnose the fault of the ESP through expert experience,manual calculation and machine learning methods.Most of these measures are taken after the ESP fails,and it is impossible to give an accurate evaluation of the risk level and health of the ESP production system before the ESP fails.How to evaluate the health status of the ESP before the failure of the ESP,and rise the alarm when the health status of the ESP is in a dangerous range are the key to extending the life of ESP.In this thesis,based on the field production data of offshore oil and gas fields,the abnormal data and missing data of ESP are preprocessed,and the fault database of ESP is formed.;analyzes the failure trend of each parameter and the correlation between each parameter and the health state,and selects a batch of parameters that can characterize the health state of ESP.On the basis of these parameters,using the feature extraction ability and time dependence of Long Short-Term Memory(LSTM)network,a health state representation model of ESP is constructed.By reducing the dimension of the data,adjusting the structure of the model and optimizing the hyperparameters of the model,the overfitting phenomenon of the model and the performance of the health state characterization model is improved.Finally,the characterization of the ESP health state is finally realized,and a method for evaluating the health status of electric submersible pumps based on machine learning method is formed.