Research On Analysis Technology Of Dynamometer Card For Pumping Wells Based On CNN-SVM

Petroleum,as a non-renewable energy source,is not only the number one energy factor in the world’s industry,but also a pivotal strategic energy source.Its output is related to the development of the country.In my country,90% of oil wells use rod pumps to extract oil.However,during long-term oil extraction operations,various failures may occur in the pumping wells,which not only affect production,but also cause safety hazards.Therefore,it is necessary to predict and diagnose the downhole working conditions of the pumping unit well and perform maintenance in advance.Analysis of dynamometer card can effectively understand the downhole working conditions of pumping wells.However,with the construction of the Internet of Things in the oilfield,traditional working condition diagnosis methods can no longer meet the needs,so this paper proposes a method that research on analysis technology of dynamometer card of pumping well based on deep machine learning.This paper first carried out the research on characteristic analysis of typical dynamometer card.Based on the dynamometer card data collected from the oilfield database,the correspondence relationship between the typical dynamometer card and the working conditions of the pumping unit well was analyzed,and a sample set of typical dynamometer card characteristics was established.Carried out the typical dynamometer card data preprocessing research,including data normalization,image binarization and refinement processing,to provide data for subsequent built models.Secondly,by optimizing the size of the model convolution kernel,the learning rate,and the number of iterations,the VGG-16(A visual geometric group network with 13 convolutional layers and 3 fully-linked layers)model is constructed as the basic model of CNN(Convolutional Neural Network)to identify and diagnose the dynamometer card.The accuracy of the model on the test set can reach 91.4%.Then,by combining the Adam algorithm(Adaptive moment estimation algorithm)with the SGD(Stochastic Gradient Descent)of the momentum algorithm,the algorithm of the built basic model is optimized.The stability and accuracy of the optimized model are improved,and the accuracy rate on the test set can reach 94%.Furthermore,because the support vector machine(SVM)as a classifier has a better application effect on image classification and recognition,this paper establishes a coupling model of the optimization model and the support vector machine to classify and recognize dynamometer card.The experimental results show that classification effect and accuracy of the model have been improved,and the accuracy of the coupled model on the test set can reach 97%.Finally,using Java as the programming language and Idea as the development tool platform,the “fault diagnosis system for pumping wells” has been development based on the B/S(Browser / Server)pattern,and the actual measurement dynamometer card of 80 wells in the oil field were analyzed,the recognition accuracy reach 86.25%.Practice has proved that the coupling model of the optimization model and SVM(Support Vector Machine)proposed in this paper can better meet the needs of oilfield pumping well condition analysis,and it also provides effective technical means for improving the real-time performance of oil well condition analysis under the conditions of the Internet of Things.