Research On Diagnosis Of Rod Pumping System Condition Based On Deep Learning

Rod pumping is the most widely used artificial lifting method in China.Due to the complex underground conditions and bad working conditions,it is of great significance for the normal production of the oilfield to timely grasp the downhole fault conditions and take appropriate measures.At present,it is a common method to diagnose the working condition by combining the computer with the surface indicator diagram.However,because the surface indicator diagram contains less information,it can only diagnose the downhole working condition and other problems,so it is difficult to further develop the diagnosis method by using the surface indicator diagram.With the development of artificial neural network,convolutional neural network stands out from many types of artificial neural network because of its advantages in image recognition.However,due to the lack of research and improvement on convolutional neural network,the existing convolutional neural network model is not completely suitable for the recognition and classification of surface indicator diagram,which limits the rod pumping system Research on intelligent diagnosis method.In this dissertation,the idea of fault diagnosis based on a variety of information synthesis is used to improve the ground indicator diagram,and the current combination indicator diagram including load,displacement and current is established.The current combination indicator diagram can effectively make up for the lack of information in the ground indicator diagram,and can identify the ground fault conditions by using the current information.In this dissertation,the classical convolutional neural network is improved,and a D-Net network model is proposed,which has seven layers of network.The D-Net network model can be applied to the diagnosis and identification of current combination indicator diagram.Combined with the current combination indicator diagram and D-Net network model,an intelligent monitoring system for oil well working condition is developed.Based on the above research,combined with the oilfield field data,this dissertation establishes the current combination indicator diagram atlas,and classifies nine classic working conditions;the improved convolution neural network model D-Net network model includes three convolution layers,two pooling layers,and two fully connected layers,and applies it to the current combination indicator diagram.The test shows that the accuracy of D-Net network reaches 99.92%,and the application effect is satisfactory It is better than the other three kinds of network,which proves the superiority of D-Net network;four kinds of network models are used to compare the ground indicator diagram and the current combination indicator diagram,and the test results of the current combination indicator diagram are better than the ground indicator diagram,which proves the advantage of the current combination indicator diagram;after being put into use,the developed intelligent software for fault diagnosis of rod pumping system can effectively identify the fault The accuracy of each condition is more than 90%,which proves that the field application effect is excellent.