Research On Application Of Convolutional Neural Network In Condition Monitoring Of Pumping Units

With the rapid development of society,the dependence on oil is growing.Oil occupies a decisive role in the stable development of economy and society.The safe and stable operation of production well equipment is the basic guarantee for oil extraction and the safe operation of pumping units.It is particularly important for the stability and safety of oilfield production and the economic benefits of the oilfield.Pumping units are mainly located in sparsely populated places,and work 24 hours a day,24 hours a day,7 days a week,and the working environment is particularly poor.Maintenance is also a major problem,and it is impossible to find all equipment problems.Therefore,a more reasonable method is needed,which does not affect the normal operation of the pumping unit,but also can monitor its status,which has a positive effect on the safe operation of the oil field and the national energy strategy.In recent years,with the development of technology,the method adopted has gradually changed from subjective judgment to dynamometer to convolutional neural network,but the monitoring effect has not been very satisfactory,and the detection results and actual conditions are great.The difference in actual production has always been the main reason that affects safe production.In response to this situation,this paper collects signals during the operation of the pumping unit,and judges the running state of the pumping unit through a convolutional neural network to achieve real-time monitoring of the equipment status.The main research contents of this article:(1)An improved wavelet packet threshold denoising algorithm is proposed.The operating conditions of the pumping unit are complicated,and the collected signal contains a lot of noise.Traditional algorithms cannot effectively process the signal of the pumping unit.This paper proposes an improved wavelet packet threshold denoising algorithm.After experimental comparison and analysis.The improved algorithm retains the energy characteristics and details of the original signal better.(2)A method for condition monitoring of pumping units based on short-time Fourier transform technology and convolutional neural network is proposed.First,the signal is converted into a time-frequency map by short-time Fourier transform technology.Then,build a convolutional neural network suitable for pumping unit condition monitoring.By selecting high-quality nodes and compressing some neurons in the fully connected layer,it is used to improve the model learning rate and the ability to recognize the state of the pumping unit.After experimental comparison,the success rate of state identification of the pumping unit after optimization is 99%.(3)A set of pumping unit condition monitoring system is designed and implemented.The system uses the five functions of basic data management,data management and analysis,fault diagnosis and analysis,log alarm query,and user login management to realize the analysis of the pumping unit operation status by analyzing the collected pumping unit signals.aims.