Research On The Working Condition Diagnosis Of Pumpingunit Wells Based On PNN Neural Network

Indicator diagram is the main basis for the diagnosis of condition of pumping unit wells,and extract the characteristic values from dynamometer is the key step in diagnosis.Currently,the major oil field is still collected by manual analysis dynamometer,often influenced by artificial subjective factors,leading to the deviation of diagnosis.Because of the pumping unit downhole working environment is very complex,pumping equipment are often destroyed,fault judgment has become more difficult.If we can timely understand and grasp the operation condition of the sucker rod pumping system,to achieve remote automation management and scientific monitoring of the pumping unit has very important significance.According to the current graph,use the time-current instead of load-displacement description indicator diagram,namely through the current method of indirectly measuring the equivalent indicator diagram.Based on the law of conservation of energy,using complex vector method to analyze the motion law of beam pumping unit,the establishment of the electric current,the polished rod load and suspension displacement between the mathematical model.In terms of the extraction of characteristic value of indicator diagram,this paper proposes the use of Freeman chain code on the equivalent dynamometer characteristic parameter extraction,preprocessing,build pumping conditions typical chain code sample library features.Diagnosis of working conditions of pumping wells based on PNN network,established probabilistic neural network model of pumping wells condition diagnosis.Firstly,this thesis introduces the working principle of pumping unit and the related concepts of indicator diagram,describes the formation process of indicator diagram.Secondly,introduce the basic principle of current method of indirect measurement indicator diagram,by establishing a mathematical model of how to draw up the equivalent dynamometer.Then the thesis provides an introduction of the related concepts of Freeman chain code and the extraction methods of Freeman chain code eigenvalue.Finally,analyzes the characteristics of BP neural network and PNN neural network,compared the disadvantages of the BP neural network,using PNN neural network to determine the classification of fault sample training.Use Freeman chain code as a feature vector, the model of probabilistic neural network was set up and trained by MATLAB. The experiments show that Freeman chain code can accurately represent the characteristics of the indicator diagram, and this network features fast learning, high diagnostic accuracy for real time detection and diagnostics of pumping unit operating conditions.