Abstract: The beam pumping unit is the most widely used component in the rod pump system, and analyzing the indicator diagram of the rod pump is an important means to judge the downhole working conditions of the pumping unit. Traditional indicator diagram recognition method relies on expert experience and manual feature extraction, which leads to low recognition accuracy when dealing with similar indicator diagrams. Through the application of deep learning convolutional neural network in the field of image recognition, a LeNet-based convolutional neural network model was proposed, which realized the automatic recognition of the indicator diagram, and the model built simplified the model structure while considering 15 common downhole conditions of pumping units, and a dropout layer and a local response normalization layer are introduced to prevent the model from overfitting and improve the generalization ability of the model. The experimental results showed that the model not only converged quickly, but also had an average accuracy of 94.68% for diagnosing working conditions, which met the accuracy requirements of pumping unit working conditions detection. The research provides a basis for the construction of an intelligent monitoring and early warning system for pumping wells, and is of great significance to the construction of smart oilfields and efficient production of oilfields.

Key words: pumping unit, indicator diagram, convolutional neural network, LeNet, working condition diagnosis