基于PNN神经网络的抽油机井工况诊断研究
发布时间：2019-06-24 23:54
【摘要】：示功图是进行抽油机井工况诊断的主要依据,而示功图的特征值提取是诊断的关键步骤。目前各大油田现场仍然通过对采集到的示功图进行人工分析,往往受到人工主观因素的影响,导致诊断结果有偏差。由于抽油机井下工作环境非常复杂,抽油设备会经常遭到破坏,故障的判断越来越难。如果能够及时了解和掌握有杆抽油系统的工况,对实现抽油机的远程自动化管理和科学监控具有很重要的意义。根据电流曲线图,用时间 电流关系代替位移 载荷描述示功图,即通过电流法间接测量等效示功图。根据能量守恒定律,利用复变矢量法对游梁式抽油机的运动规律进行分析,建立了电流和光杆载荷及悬点位移之间的数学模型。在示功图特征值的提取方面,使用Freeman链码对等效示功图提取特征参数,进行预处理,建立抽油机典型工况的链码特征样本库。利用PNN网络对抽油机井工况进行诊断,建立了抽油机井工况诊断的概率神经网络模型。本文首先介绍了抽油机的工作原理和示功图的相关概念,描述了示功图的形成过程;然后介绍电流法间接测量示功图的基本原理,以及通过建立起的数学模型如何绘制出等效示功图;接着介绍Freeman链码的相关概念,以及示功图的预处理和Freeman链码特征值的提取方法;最后分析了BP神经网络和PNN神经网络的特点,比较了BP神经网络的不足,利用PNN神经网络对样本训练确定分类故障。将Freeman链码作为特征向量,利用MATLAB对网络进行训练。实验结论:用Freeman链码可以准确表达出示功图的特征,并且该PNN网络模型学习速度快、诊断准确率高,可用于抽油机井工况的实时监测和诊断。
[Abstract]:Indicator diagram is the main basis for working condition diagnosis of pumping well, and eigenvalue extraction of indicator diagram is the key step of diagnosis. At present, the field of major oil fields is still through the manual analysis of the collected indicator diagram, which is often affected by artificial subjective factors, which leads to the deviation of diagnosis results. Because the underground working environment of pumping unit is very complex, the pumping equipment will often be destroyed, so it is more and more difficult to judge the fault. If we can understand and master the working condition of rod pumping system in time, it is of great significance to realize the remote automatic management and scientific monitoring of pumping unit. According to the current curve, the time current relation is used instead of the displacement load to describe the indicator diagram, that is, the equivalent indicator diagram is measured indirectly by the current method. According to the law of conservation of energy, the motion law of beam pumping unit is analyzed by using complex variable vector method, and the mathematical model between current and beam load and suspension point displacement is established. In the aspect of characteristic extraction of indicator diagram, Freeman chain code equivalent indicator diagram is used to extract feature parameters, preprocessing is carried out, and the chain code feature sample database of typical working conditions of pumping unit is established. The PNN network is used to diagnose the working condition of the pumping well, and the probabilistic neural network model of the working condition diagnosis of the pumping well is established. This paper first introduces the working principle of pumping unit and the related concepts of indicator diagram, describes the formation process of indicator diagram, then introduces the basic principle of indirect measurement of indicator diagram by current method, and how to draw the equivalent indicator diagram through the established mathematical model, and then introduces the related concepts of Freeman chain code, as well as the preprocessing of indicator diagram and the extraction method of Freeman chain code eigenvalues. Finally, the characteristics of BP neural network and PNN neural network are analyzed, and the shortcomings of BP neural network are compared. PNN neural network is used to determine the classification faults by sample training. The Freeman chain code is used as the eigenvector, and MATLAB is used to train the network. The experimental results show that the Freeman chain code can accurately express the characteristics of the power map, and the PNN network model has the advantages of fast learning speed and high diagnostic accuracy, so it can be used for real-time monitoring and diagnosis of pumping well working conditions.
【学位授予单位】：安徽工业大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TE933.1;TP183
本文编号：2505469
[Abstract]:Indicator diagram is the main basis for working condition diagnosis of pumping well, and eigenvalue extraction of indicator diagram is the key step of diagnosis. At present, the field of major oil fields is still through the manual analysis of the collected indicator diagram, which is often affected by artificial subjective factors, which leads to the deviation of diagnosis results. Because the underground working environment of pumping unit is very complex, the pumping equipment will often be destroyed, so it is more and more difficult to judge the fault. If we can understand and master the working condition of rod pumping system in time, it is of great significance to realize the remote automatic management and scientific monitoring of pumping unit. According to the current curve, the time current relation is used instead of the displacement load to describe the indicator diagram, that is, the equivalent indicator diagram is measured indirectly by the current method. According to the law of conservation of energy, the motion law of beam pumping unit is analyzed by using complex variable vector method, and the mathematical model between current and beam load and suspension point displacement is established. In the aspect of characteristic extraction of indicator diagram, Freeman chain code equivalent indicator diagram is used to extract feature parameters, preprocessing is carried out, and the chain code feature sample database of typical working conditions of pumping unit is established. The PNN network is used to diagnose the working condition of the pumping well, and the probabilistic neural network model of the working condition diagnosis of the pumping well is established. This paper first introduces the working principle of pumping unit and the related concepts of indicator diagram, describes the formation process of indicator diagram, then introduces the basic principle of indirect measurement of indicator diagram by current method, and how to draw the equivalent indicator diagram through the established mathematical model, and then introduces the related concepts of Freeman chain code, as well as the preprocessing of indicator diagram and the extraction method of Freeman chain code eigenvalues. Finally, the characteristics of BP neural network and PNN neural network are analyzed, and the shortcomings of BP neural network are compared. PNN neural network is used to determine the classification faults by sample training. The Freeman chain code is used as the eigenvector, and MATLAB is used to train the network. The experimental results show that the Freeman chain code can accurately express the characteristics of the power map, and the PNN network model has the advantages of fast learning speed and high diagnostic accuracy, so it can be used for real-time monitoring and diagnosis of pumping well working conditions.
【学位授予单位】：安徽工业大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TE933.1;TP183
【参考文献】
相关硕士学位论文 前2条
1 王秋勤;基于概率神经网络的发动机故障诊断研究[D];西南林业大学;2010年
2 王巨轮;有杆泵抽油系统的智能故障诊断及远程监控的研究[D];浙江大学;2009年
，本文编号：2505469
本文链接：https://www.wllwen.com/kejilunwen/shiyounenyuanlunwen/2505469.html
