Research On Dynamic Diagnosis Method For Oil Wells' Production Failure Based On Deep Learning

With the continuous deepening of oilfield production and development,the polymer flooding technology has been increasingly applied.As the most basic unit of crude oil production by polymer flooding,polymer flooding wells play an important role in ensuring the stable production of oilfields.Due to the more complex structure and complicated production process of polymer flooding wells compared with other production wells,there are many factors to be considered in the diagnosis of production failures of polymer production wells,and the diagnosis is extremely difficult.At present,the intelligent analysis and failure diagnosis method based on oilfield production data has problems such as inaccurate mining results of failure occurrence patterns,difficulty in fuzzy and uncertain reasoning and poor adaptability to abnormal data environments.These problems result in the low failure diagnosis rate,misdiagnosis and missed diagnosis,which directly leads to the high maintenance cost and the serious result that affects the safe and stable production.Therefore,it is urgent to study an efficient and accurate dynamic intelligent diagnosis method for production failures of polymer flooding wells,in order to alleviate the contradiction of crude oil production by polymer flooding and improve the excellent development and management level of tertiary oil recovery.In recent years,deep learning technology has developed rapidly.It shows a strong pattern mining ability in environments with a large number of features,low sensitivity and complex data types.It has been widely used in such fields as image recognition and speech recognition.At the same time,as the knowledge engineering system constantly improves,the fuzzy reasoning and uncertain reasoning methods become more mature.Given this,after analyzing the essential problems of the existing methods,we made an in-depth analysis of the structure,production process and data characteristics of the polymer flooding well,and proposed a dynamic intelligent diagnosis method for production failures of polymer flooding wells based on deep learning by combining the advantages of the above methods and means.The main research contents are described as follows.1.Overall model design: From the perspective of business and data,aiming at production failures of polymer flooding wells,we analyzed the causes,classification and data change characteristics,concluded the general diagnosis mechanism,and summarized the intelligent diagnosis workflow.On this basis,a production failure prediction model of polymer flooding well based on deep learning was constructed,and the key problems to be solved in the model were analyzed.2.Itemized research on key issues(1)Aiming at the problems of many failure characterization factors,serious diagnosis difficulty and inaccurate pattern mining results,we followed the research paradigm of the ?data-sample-algorithm?.First,we collected the oilfield production data that can characterize and describe the characteristics of production failures of polymer flooding wells(referred to as failure diagnosis data of polymer flooding wells),and deeply analyzed its structure,inline relationship and change laws.Second,we constructed a production failure diagnosis sample model of polymer flooding wells,and put forward the data integration method.Third,based on the convolutional neural network method in the deep learning method,we introduced the long short-term memory(LSTM)to improve the time series data mining ability,and designed a mining algorithm for the production failure occurrence pattern of polymer flooding wells based on CNN-BLSTM.Finally,a series of comparative experiments were designed to verify its apparent advantages in expression ability,running speed and other aspects.(2)As to the uncertainty and fuzziness of knowledge inference,we proposed the intelligent reasoning and dynamic diagnosis mechanism of production failures of polymer flooding wells.First,we provided an abstract expression of knowledge using a ?triple? structure,broke down knowledge into meta-concepts,and expressed the scope of knowledge and the failure occurrence pattern with the object-oriented OOP structure and the frame structure.Second,we gave the representation model of knowledge and the physical structure of the knowledge base,and designed the intelligent diagnosis and reasoning mechanism.Third,the Bayesian method for argument accumulation was used to solve the uncertainty of knowledge inference.Fourth,a fuzziness calculation method was designed based on the semantic restriction for the fuzziness of knowledge expression.Finally,an experiment was designed to verify the effectiveness of the mechanism proposed in this paper.(3)After completing the research on the theoretical method,considering the fact that the oilfield production data is abnormal,we employed the concept of ?reducing the impact of data rather than correcting the data?,to study the data processing method under the abnormal data environment.First,we analyzed the causes and classification of data anomalies,to determine the scope of missing data and the types of data errors that seriously affect failure diagnosis.Second,we used the Markov chain model to repair the missing time series data,and proposed the first-order MCM and higher-order MCM methods for different degrees of missing.Third,according to the impact of the IPR productivity curve on the failure diagnosis process,a multi-point productivity calculation method was proposed,based on which an MCC-based IPR curve correction method was presented.Finally,by artificially constructing an abnormal data environment,we experimented on the processing method of data missing and data errors to verify the effectiveness of the method.3.Research and development of a dynamic intelligent diagnosis system for production failures of real polymer flooding wellsBased on the application background of the dynamic intelligent diagnosis of production failures of polymer flooding wells,the PMP-DL model,and the purpose of improving the intelligent level of the polymer flooding production management department in real-time dynamic analysis and failure diagnosis,the ?tertiary oil recovery production development and abnormality diagnosis management platform? was designed and developed.Then,we detailed the platform overview,overall design,key technologies and other related contents,and further demonstrated the effectiveness of the method proposed in this paper by testing and analyzing the actual data in the real environment.The research results show that the dynamic intelligent diagnosis method for production failures of polymer flooding wells based on deep learning can better solve the problems in the current diagnosis of production failures of polymer flooding wells.At the same time,through extensive research and analysis,this method provides a new idea for dealing with other types of production wells in oilfields,and even for the application in other industrial fields.