Research On Diagnosis Methods Of Working Conditions Of Electric Submersible Pump Wells Based On OCSVM

The electric submersible pumps are the important oil extraction mechanical equipments,and it is of great significance to monitor their operation status.With the accelerating pace of oilfield informatization construction,the oilfield production command system can obtain realtime data related to electric submersible pump wells,which provides data support for the diagnosis and analysis of the working conditions of electric submersible pump wells,but there is still a lack of corresponding data processing methods.How to use a lot of data to realize the diagnosis and early warning of the working conditions of electric submersible pump wells is the focus of the thesis.The main tasks completed in the thesis are as follows:(1)A two-level classification model for identifying abnormal conditions is established.First of all,in view of the problem of a lot of normal operating conditions data,few abnormal operating conditions data,and even missing some abnormal operating conditions data,it is proposed to apply One-Class Support Vector Machine(OCSVM)to the diagnosis of working conditions of electric submersible pump wells.The first-level classification model of electric pump wells is established,and it only need normal working condition data to train the model to realize the identification of normal working conditions and abnormal working conditions.Secondly,based on the current data of existing working conditions and the current change characteristics of a large number of paper current cards,for 6 common abnormal conditions,a feature matching recognition algorithm for the types of conditions is designed,a feature library of abnormal conditions is established,and the recognition of the types of abnormal conditions is realized,and the second level of classification is realized on the basis of the first level of work condition classification.(2)An early warning model of abnormal working conditions based on long and short-term memory neural network(LSTM)is established.In response to the slow-changing abnormal working conditions,the working condition data is resampled on the hourly time scale,and LSTM is used to predict the operating current of electric submersible pumps.The OCSVM model and predicted value are used to realize the early warning of abnormal working conditions of electric submersible pump wells.(3)The designed model is verified by the measured data.Firstly,the first-level classification model is verified by the measured data of 30 electric submersible pump wells.According to the verification results,the accuracy reaches 93%,indicating that the OCSVM model has good robustness.The second-level classification model is verified by the data of 8electric pump wells under abnormal working conditions,and the results showed that the identification accuracy reaches 87%.The verification results of the early warning model show that for slow-changing conditions such as sand production,the alarm time can be advanced by1 hour,shortening the identification cycle of abnormal conditions.(4)The working condition diagnosis system software of electric submersible pump is designed.Real-time data acquisition is achieved by reading the Oracle database in real time.The data is preprocessed,and the OCSVM model,feature matching recognition algorithm and LSTM model are used to realize the real-time monitoring and early warning of the operation status of electric submersible pump wells.