基于深度学习的示功图多混合故障诊断  

Multi-hybrid Fault Diagnosis of Dynamometer Cards Based on Deep Learning

作　　者：魏航信[1] 张青 Wei Hangxin;Zhang Qing(College of Mechanical Engineering,Xi′an Shiyou University,Xi′an 710065,China)

机构地区：[1]西安石油大学机械工程学院,西安710065

出　　处：《机电工程技术》2022年第3期112-116,共5页Mechanical & Electrical Engineering Technology

摘　　要：现有的油井地面示功图诊断方法只能诊断单一故障,为了提高抽油机的故障诊断性能,研究了一种改进型深度学习神经网络,可实现示功图可视化多混合故障诊断功能。改进型深度学习神经网络包括3层卷积神经网络和3层全连接神经网络。研究了改进型深度学习神经网络的前向学习算法和反向自适应权值修正算法,并提出了网络节点的PSO优化算法。对现场采集的地面示功图进行实验,优化后的卷积层节点数分别为64×64×20、28×28×16、10×10×16,池化层节点数分别为32×32×20、14×14×16、5×5×16。结果表明,示功图的平均识别时间为0.021 s,训练精度为99.4%,识别精度为94%,可以识别出两种混合故障,验证了该神经网络的可靠性和准确性,满足抽油机工况检测的诊断精度要求。该研究对于实现智慧采油具有重要的意义。The existing diagnosis method for surface dynamometer cards of oil well can only diagnose a single type of fault.Therefore,an improved deep learning neural network was studied to realize the visualized multi-hybrid fault diagnosis function of the dynamometer card.The neural network includes a 3-layer Convolutional Neural Network(CNN)and a 3-layer fully connected neural network.The forward learning algorithm and the backward adaptive weight correction algorithm of the improved deep learning neural network were studied.Particle Swarm Optimization(PSO)algorithm was used to optimize the number of neural network nodes.Diagnostic experients were performed for the ground dynamometer cards collected on site.The number of nodes in the optimized convolution layer was 64×64×20、28×28×16、10×10×16 respectively.The number of nodes in the pool layer was 32×32×20、14×14×16、5×5×16 respectively.The results show that the average recognition time of indicator diagram is 0.021s,the training accuracy is 99.4%,and the recognition accuracy is 94%.Two kinds of mixed faults can be identified,which verifies the reliability and accuracy of the neural network and meets the diagnostic accuracy requirements of pumping unit condition detection.The research is of great significance for realizing intelligent oil recovery.

关 键 词：深度学习 示功图 粒子群算法 卷积神经网络 

分 类 号：TE319[石油与天然气工程—油气田开发工程]