Research On Fault Diagnosis And Early Warning Of ESP Well Based On Physical Constraint And Data-driven Fusion

Electric Submersible Pump(ESP)well plays an important role in oil and gas fields.It is one of the most important lifting wells in artificial lifting wells.However,the stability and reliability of ESP well still need to be improved,and unplanned shutdown caused by failure occurs frequently.At present,the fault diagnosis and early warning of ESP well is mainly realized based on mechanism model or data-driven,but there are some deficiencies in a single mechanism model or data-driven.For example,the mechanism model has poor real-time performance and complex model;Data driven is pure data driven,without considering the physical laws and scientific theories behind the problem,and the accuracy of the results is poor.Aiming at the above problems,based on the concept and method of fusion model,this paper carries out the research on fault diagnosis and early warning of ESP well based on physical constraint and data-driven fusion.Firstly,starting from the original data,a data preprocessing model of ESP well based on physical constraints and machine learning fusion(PC-ML)is established based on box diagram,empirical method,moving average method and random deep forest algorithm.Secondly,the fault characteristics of ESP well are studied,and the physical constraints of ESP well fault diagnosis,namely multi parameter rule chart,statistical process control(SPC)expansion rules and working condition weight factor table,are established.Based on these physical constraints,the ESP well fault diagnosis model of physical constraints and real data-driven fusion(PCRDD)is established.Thirdly,the main control factors of production parameters are analyzed based on Pearson correlation coefficient and XGboost algorithm,the main control parameters affecting the fault of ESP well are found,and the health index of ESP well is established based on principal component analysis(PCA).Finally,combining the energy conservation equation of the inlet and outlet of the ESP with the calculation formula of the health index,the control equation of physical constraints is established.Using this control equation,the long-term and short-term memory network(LSTM)is constrained in the training process,and the fault early warning model of the ESP well with the integration of physical constraints and LSTM(PC-LSTM)is established.Through the above research,the fault diagnosis and early warning model of ESP well based on the integration of physical constraints and data drive is established,which solves the key scientific problems such as poor real-time and accuracy in the current diagnosis and early warning.It has important practical significance for guiding field production,reducing the number of faults,prolonging the pump inspection cycle,reducing oil production costs and improving the economic benefits of oilfield development.It also provides a new research idea for fault diagnosis and early warning of other oil,gas and water wells.