摘要
抽油机故障诊断对于保障油气田的稳定运行至关重要.针对已有基于深度学习的故障诊断模型参数量大导致应用范围受限的问题提出一种基于空洞卷积和惩罚机制的卷积神经网络模型.该模型在浅层神经网络部署不同空洞卷积率的空洞残差模块高效获取示功图轮廓特征的同时降低了模型参数量.其次将惩罚机制融入Softmax损失函数增强模型诊断气体影响等难分样本的故障准确率.采用抽油机实况数据集进行实验验证结果表明该模型参数量为0.94 M浮点型计算量为165.24 M.与MobileNetV3相比改进后的算法模型在准确率同为96.6%的前提下参数量减少了3.30 M浮点型计算量减少了52.22 M更易部署在资源受限的故障诊断平台.
Pumping unit fault diagnosis is crucial to ensure the stable operation of oil and gas fields.The current fault diagnosis of pumping unit based on deep learning model has the problem that the number of parameters is too large and it is difficult to be widely used in actual production.Considering the real demand for reducing system resource usage of the fault diagnosis model a novel convolutional neural network is established based on dilated convolution and penalty mechanism in this study.In this model dilated convolution residual blocks of different dilated convolution rates are deployed in the shallow neural network to efficiently acquire the contour features of the dynamometer card and reduce the number of model parameters.Moreover the penalty mechanism is integrated into the Softmax loss function to enhance the influence of indistinguishable samples such as gas influence on the fault diagnosis model.Experimental validation is conducted with the data set made from actual working conditions of the pumping unit.When the accuracy rate is 96.6%the number of parameters acquired by this model is 0.94 M which is decreased by 3.30 M in MobileNetV3 model.Similarly the floating-point operations calculated by this model is 165.24 M which are also decreased by 52.22 M in MobileNetV3 model.In conclusion the convolutional neural network holds potentially promising in the resourceconstrained platform of actual production.
作者
吴昊臻
许燕
周建平
谢欣岳
彭东
WU Haozhen;XU Yan;ZHOU Jianping;XIE Xinyue;PENG Dong(College of Mechanical Engineering Xinjiang University Urumqi,Xinjiang,830039 China;Xinjiang Golden Calf Energy IOT Technology Co.Ltd.Karamay,Xinjiang,834008 China)
出处
《燕山大学学报》
北大核心
2024年第1期30-38,共9页
Journal of Yanshan University
基金
国家自然科学基金资助项目(52265061)
新疆维吾尔自治区重点研发专项(2020B02016)。
关键词
卷积神经网络
抽油机
故障诊断
空洞卷积
损失函数
convolutional neural network
pumping unit
fault diagnosis
dilated convolution
loss function
