| As we all know,oil resources are the material basis for the survival and development of human beings.With the rapid development of science and technology,it is our goal to realize the digitalization and intellectualization of petroleum industry at an early date.Crude oil production process commonly used pumping unit,most pumping units work in unattended state,and the pumping unit underground condition is complex,and the load varies with the oiling process,which leads to the malfunction of the pumping unit is often affect the output and benefit of oil field.In this dissertation,a fault diagnosis method based on intelligent BP neural network is presented to solve the problem that the accuracy of pumping unit fault diagnosis is not high and intelligence is not strong.Firstly,in this dissertation,the pumping unit indicator diagram fault diagnosis research status at home and abroad as well as the general structure and basic principle of pumping unit are briefly introduced.The normal dynamometer and the 20 main types of fault diagram are learned from the oil and gas product IOT system(A11)of China National Petroleum Corporation.And these 20 kinds of fault indicator diagrams obtained from the system will be used as teachers signal in BP neural network training.Secondly,on the basis of the study of the relevant basic principles,combined with the actual oil production process,the fault diagnosis model of pumping unit indicator diagram is established.And the model is analyzed in theory.The model is divided into three steps: The first step is to use the appropriate sensor for basic data collection.The second step is to draw the indicator diagram by the indicator instrument for each pumping unit.The third step is the analysis of the fault diagnosis method of the indicator diagram,which is also the main research content of this dissertation.Finally,it focuses on the analysis of the fault diagnosis method of the indicator diagram,which is divided into three stages.In the first stage,a new method of combining Shape invariant moments with Fourier descriptors is selected to extract the characteristic parameters of indicator diagram,the combination of the two improves the speed of operation,reduces the amount of computation.The new method has the advantages of stability of graph rotation,scale and translation invariance.In the second stage,the classification of fault diagnosis is demonstrated.According to the shortcomings of BP neural network training,the weight of BP neural network is optimized by iterative learning control,which makes the algorithm have good performance and shorten the training time and training speed.And it has the characteristics of high recognition accuracy and strong generalization ability.It overcomes the shortcomings of the traditional methods which need to manually extract the characteristics and the network is easy to fall into local minimum.The third stage is to compare the results of the diagnosis with the diagnostic results of the expert diagnosis method and the diagnostic results of the slope diagnosis method.If the diagnosis results are consistent,the fault type can be determined.If the test results are inconsistent,the recognition judgment needs to be repeated.The final diagnosis results are obtained.The MATLAB software was used to simulate the model training.From the test process and the results,it is shown that the performance of the diagnostic method proposed in this dissertation has certain advantages.In the same test sample cases,the recognition time is improved by about 9 ms relative to the support vector machine identification method,the accuracy of the diagnosis has also increased significantly.So the intelligent fault diagnosis method presented in this dissertation is more suitable for real-time fault diagnosis of large on-site indicator diagram datas in oil field.It not only makes up for the shortcomings of existing diagnostic methods,but also has good comprehensive performance.It has a broad application prospects. |
PDF Full Text Request