Research On Fault Feature Extraction And Early Warning Method Of Electric Submersible Pump

Electric submersible pump is the key equipment in artificial lifting technology,which is widely used in the field of oil drilling and production,and its working state is very important to ensure the safety of crude oil production.The working environment of electric submersible pump is bad.Once it fails,it will cause the unit to stop and affect the production efficiency,or the wellhead will be scrapped.Therefore,it is necessary to realize the monitoring and early warning of its operation status to coordinate the whole production process.In this thesis,aiming at the problem of fault early warning of electric submersible pump in offshore oilfield,based on the research of slow feature extraction of ESP operation parameter data,the ESP monitoring and early warning model is established,and the following research work is carried out:(1)Aiming at the problem that the fault data of electric submersible pump is scarce and the data samples are unbalanced,this thesis proposes a data enhancement method of electric submersible pump operating parameters based on generating countermeasure network.With this method,new virtual data is generated on the basis of the original electric submersible pump fault data,which can be used as a supplement to the original data set.Experiments are designed,according to which the method reconstructs the data on the basis of learning the feature distribution of the original data,and the experimental results prove that the enhanced data can improve the accuracy of the classification algorithm.(2)Aiming at the problems that the key features of ESP are difficult to extract in the potential abnormal state and the time margin of early warning is small in traditional methods.In this thesis,the feature extraction method of time series based on SFA is adopted,and the selection method of control limits is improved..And constructs the running state monitoring model of ESP combined with key statistics reflecting dynamic steady-state characteristics.The experiment was designed by using the real operating parameters of electric submersible pump.The results show,this method can give early warning before the operating parameters of electric submersible pump are distorted,and it has a large margin of early warning time.(3)Aiming at the problems of strong coupling of ESP fault states,difficult discrimination and low discrimination of traditional diagnosis and early warning to fault conditions,this thesis proposes a monitoring and early warning method of ESP based on slow feature analysis and XGBoost.On the basis of monitoring the running state of potential faults of electric submersible pump,this method makes a distinction and predictive distinction between the types of potential faults.The method is verified by real industrial field data,and the experimental results show that the method has strong early warning ability for five kinds of faults and is feasible in engineering.(4)A software of electric submersible pump early warning system based on Django framework is developed,and the software of research results is realized.Combined with the actual situation of software and hardware in the digitalization of the petroleum industry,the specific requirements of modular development of the software of ESP early warning system are analyzed.After the joint deployment of u WSGI and Nginx,the software was put online and tested,and the test results proved that the software met the requirements in terms of stability and compatibility.