Failure Diagnosis Method of Oil Pump Well Based on MobileNet V3

Aiming at the fault diagnosis problem of indicator diagram of pumping Wells, a lightweight convolutional neural network is proposed, and the attention mechanism is replaced to improve the network diagnostic performance under the condition of reducing the number of parameters and calculation. MobileNet-V3 was used as the network infrastructure, and the SE Attention Module in the inverted residual module of MobileNet-V3 was replaced by the Efficient Channel Attention Module. In the inverted residual module, ECA multiplied the generated channel attention recalification weight with the input feature map from the channel dimension, and finally obtained the feature map with attention weighting. Based on the proposed method, the complexity of the network model structure is reduced under the premise of ensuring the accuracy of the model diagnosis. The experimental results show that the diagnosis accuracy of the improved MobileNet-V3 reaches 89.20%, which meets the actual production needs of oil fields.