基于迁移学习的示功图诊断方法

Diagnosis Method of Indicator Diagram Based on Transferring Learning

示功图是数字化分析抽油机作业状况的重要依据,不同形状的示功图代表着不同的作业状况。传统分析程序基于专家系统或统计学习方法对示功图进行分析,需要大量专家知识且鲁棒性较低。从深度学习的角度,提出了一种基于深度卷积神经网络的示功图检测方法,并通过迁移学习,大幅度减少了模型收敛所需样本数量。实验表明,该方法可以有效提高示功图分类的准确率,实现了真正的工业可用。

The indicator diagram is an important base for digital analysis of the operating conditions of the pumping unit.Different shapes of indicator diagrams represent different operating conditions.Traditional analysis programs analyze indicator diagrams based on expert systems or statistical learning methods,which require a large amount of expert knowledge,and the robust is low.From the perspective of deep learning,an indicator diagram detection method based on deep convolutional neural networks is put forward,and the number of samples required for model convergence is greatly reduced through transferring learning.Experiments show that the method can effectively improve the accuracy of the indicator diagram classification.True industrial application is realized.