Research On Data-driven Remaining Useful Life Prediction Of ESP Wells

The ESP is the mainstream core equipment of artificial lift.The research on the health state management of the complex system of the ESP can improve the operation reliability of the ESP,improve the intelligent management level of the oil field,and avoid the malfunction caused by the failure of the ESP.The remaining life prediction technology discussed in this thesis,as the core technology of health management,can help oilfields accurately discover the operating status of ESP wells and predict the time of failure in time,which is of great practical significance for improving the level of intelligent management in oilfields.The traditional ESP health management can only rely on expert experience and shutting down well to check the pump,and it is difficult to use the rich real-time production data accumulated by the rich sensors on the ESP.Traditional methods cannot utilize the high-dimensional and in-depth information in the real-time data of ESPs,so they cannot accurately predict the remaining life of ESPs,which has become a major problem in ESP fault warning and health management.In this thesis,We use the remaining useful life prediction method of mechanical equipment is used,and the deep learning algorithm and survival analysis method to process and model the real-time production data of the ESP,to predict the remaining useful life of the ESP well.The main tasks of this thesis are as follows:1.Data preprocessing,including outlier cleaning of real-time data,data enhancement,time-series resampling,data normalization,etc.,make high-quality real-time production data in the database into custom-loadable ESP Well remaining life prediction dataset.2.The remaining life prediction model adopts the common time series modeling network LSTM to extract the features of the time series and predict the remaining life.On the basis of the bidirectional LSTM,the LSTM prediction network based on the encoderdecoder architecture and the Survival＿LSTM with the introduction of survival analysis are implemented.Realize the prediction technology from the perspective of oil field and equipment manufacturer.3.The remaining life research framework divides the prediction of the remaining life of ESP wells into data modules and model modules,and the prediction process is decoupled.The data module can customize the data form,and the model module realizes the unification of common model input and model training,evaluation,and deployment.And show the effect of different combinations on the prediction accuracy.