摘要
为克服已有抽油机异常诊断方法中存在的不足,提出一种多尺度结合全局池化的AlexNet模型(简记为AGINet模型)。首先,AGINet模型在已有AlexNet模型中使用批归一化代替原来的局部响应归一化,接着加入Inception模块,最后使用全局平均池化层代替原有的全连接层。基于相同的输入输出和抽油机示功图数据集,分别完成对AGINet模型、AlexNet模型、LeNet5模型、VGG11模型、卷积神经网络和支持向量机的训练和测试。测试结果表明,与其它深度学习模型和支持向量机相比,AGINet模型的分类准确率、召回率的宏平均值、F1的宏平均值、参数量和所占内存大小均有一定程度的改善,具体值分别为99.9%、99.9%、99.9%、338649和1334KB。AGINet模型为油田抽油机的异常诊断提供了重要的技术参考,促进了先进计算机技术在石油工业中的应用。
In order to overcome the shortcomings of existing abnormal diagnosis methods for pumping units,an AlexNet model with multi-scale and global pooling(abbreviated as AGINet model)has been proposed in this paper.First,the batch normalization was used to replace the original local response normalization in the AGINet model;Second,the inception module was added;Last,the global average pooling layer was used to replace the original full connection layer.Based on the same input,output and thesame dataset of pumping unit indicator diagrams,the training and testing of the AGINet model,AlexNet model,LeNet5 model,VGG11 model,convolutional neural network and support vector machine were completed,respectively.The testing results show that compared with other deep learning models and support vector machine,the classification accuracy,the macro average of Recall,the macro average of F1,the amount of parameters and the size of the memory occupied by the AGINet model have been improved to a certain extent.The specific values are 99.9%,99.9%,99.9%,338649 and 1334 KB,respectively.The AGINet model provides an important technical reference for abnormal diagnosis of pumping units,and promotes the application of advanced computer technologies in the petroleum industry.
作者
潘少伟
王树楷
张航
秦国伟
PAN Shao-wei;WANG Shu-kai;ZHANG Hang;QIN Guo-wei(School of Computer Science,Xi'an Shiyou University,Xi'an Shaanxi 710065,China;College of Petroleum Engineering,Xi'an Shiyou University,Xi'an Shaanxi 710065,China)
出处
《计算机仿真》
2024年第5期88-93,167,共7页
Computer Simulation
基金
国家自然科学基金项目(52174027)。
关键词
抽油机
异常诊断
示功图
深度学习模型
Pumping unit
Abnormal diagnosis
Indicator diagram
Deep learning model
