基于LIBSVM的融合傅里叶幅值与相位的示功图识别方法

An identification method of Indicator Diagram based on LIBSVMFusion Fourier Amplitude and Phase Information

摘要:

通过示功图来诊断抽油机井工况,是确保油井安全高效生产的一种重要手段。针对现有示功图特征提取只利用其离散傅里叶变换(Discrete Fourier Transform,DFT)的幅度谱而忽略了其相位谱,从而导致识别率较低的问题,提出了一种融合DFT的幅度谱与相位谱的示功图识别方法。首先,将示功图数据组成复数序列,取其DFT的幅度谱与相位谱构造特征向量；其次,对已知故障种类的示功图的特征向量进行训练,构造多分类支持向量机(Support Vector Machines,SVM)分类判别模型；最后,通过LIBSVM分类识别方法对未知类别的示功图进行分类识别,从而诊断抽油井的工况。实测结果表明,与只利用DFT幅度谱的方法相比,该方法能够有效地提高示功图的识别率,能为油井工况的准确分析、诊断与实时优化控制提供技术支撑。

Abstract:

It is an important means to ensure safe and efficient production of oil wells to diagnose pumping wells by means of indicator diagram. In view of the problem that the feature extraction of the present indicator diagram only uses the amplitude spectrum of its Discrete Fourier Transform (DFT) and ignores its phase spectrum, which leads to the low recognition rate, presents a method to identify the amplitude spectrumand phase spectrum of the fused DFT. Firstly, the DFT of the indicator diagram sequence is calculated and linearly independent and orthogonal eigenvectors are constructed according to its DFT amplitude spectrum and phase spectrum. Secondly, The characteristic vectors of the indicator diagram of the known fault-type are trained, and the Multi-class Support Vector Machines classification model Multi-class Support Vector Machinesis constructed. Finally, the LIBSVM identification method is used to classify the unknown type of indicator diagram to diagnose the working condition of the pumping well. The measured results show that this method can effectively improve the identification rate of the indicator diagram and provide technical support for accurate analysis, diagnosis and real-time optimization control of the oil well condition, compared with the method of using only the amplitude spectrum of DFT.