Research on Fault Analysis Model of Lightweight Pumping Unit Based on Classical Convolutional Neural Network

In the conventional sucker rod pumping system, the pumping unit often be produced many types of faults that due to the influence of sucker rod, pump, and other accessories, as well as oil well paraffinication, gas interference, sand production and other environmental impacts. Using indicator diagram to analyze the fault diagnosis of pumping units is a common method. In this paper, a lightweight model was designed based on the classical convolutional neural network, and a comparative experiment was used to optimize the model from four perspectives: learning rate, convolution kernel size, batch size, and optimization algorithm. Finally, the average accuracy achieved 95.5%.