Model And System Of Working Condition Diagnosis For Electric Submersible Pump Well Based On Multi-Sensor Data Fusion

Electric submersible pump(ESP)well plays a very important role in oil and gas lift.At present,the working condition diagnosis theory of ESP well mainly analyzes a single parameter,such as current signal timing sequence analysis,vibration signal conversion analysis,current card pattern matching,comparative analysis of pressure holding curve,threshold discrimination of different parameters,etc.Among them,there are few methods that can realize fault diagnosis,and in the application process of similar working conditions,it is impossible to extract features in detail for slight changes of graphic features,and it does not specifically consider the comprehensive parameters of diagnosis and cannot reasonably use the downhole companion data of ESP.The established method for diagnosis of operating condition of ESP is only based on off-line mode,and the fault cause has not been identified in real time.In view of the above problems,this paper analyzes and preprocesses the companion data of downhole ESP and wellhead data from the research of multiple data sources of ESP unit,for a single current value,take the current data collected in the filed as the source data,extract the feature of current card under typical working condition,artificial simulate the sample set of current card under different working condition;for multiple parameters,combined with the field application at home and abroad,summarizes kinds of corresponding to the plates of rules for working condition of kinds of characteristic parameters,and established the data sets of curve trend.Then,this paper utilizes convolutional neural network to extract the detailed features of the current card and the multi-parameter curve graph,and established the production condition diagnosis model of ESP well,and the optimization analysis of the number of iterations,the number of batch samples,the learning rate,different optimization algorithms,and whether to perform Dropout have improved the average accuracy of current card diagnosis to 90.23%,and the average accuracy of curve trend recognition to 87.84%,after some examples analysis,it is confirmed that the theory of this paper has good feasibility for the oil field.Finally,on the basis of the above theories,the production monitoring and working condition diagnosis module of the software of “ Intelligent Management System for Oil and Gas Lift ” are developed,and realized the objectives of economic reliability,environmental friendliness,intelligence and fine management of oil and gas production.