为保证石油管道系统的安全运行，需准确判断管道法兰盘堵塞工况，因此提出一种基于卷积神经网络的石油管道法兰盘堵塞诊断方法。首先，通过递归图对采集到的石油管道法兰盘的振动数据进行递归得二维图像，再将图像数据输入融合了ECA（efficient channel attention）高效通道注意力机制的AResNet（adaptive deep residual network）自适应深度残差网络中进行训练。该网络能够自适应地增强对故障敏感特征的关注度，同时削弱与故障相关性差的特征，以此取代原始的残差网络。最后，将测试集数据输入到模型中进行验证。结果表明：融合递归图和ECA-AResNet的石油管道法兰盘堵塞诊断模型与现有的方法相比具有更优的准确率、鲁棒性和计算性能。

In order to ensure the safe operation of the oil pipeline system, it is necessary to accurately determine the clogging condition of the pipeline flange, so a convolutional neural network-based oil pipeline flange clogging diagnosis method was proposed. Firstly, the vibration data of the oil pipeline flange collected by recursive map was recursively obtained as a two-dimensional image, and then the image data was inputted into adaptive deep residual network (AResNet) that integrated efficient channel attention(ECA) and efficient channel attention mechanism for training. This network was able to adaptively enhance the attention to fault-sensitive features while weakening the features that were poorly correlated with faults, thus replacing the original residual network. Finally, the test set data was fed into the model for validation. The results show that the oil pipeline flange blockage diagnostic model incorporating recurrence plot and ECA-AResNet has superior accuracy, robustness and computational performance compared to existing methods.