基于鱼群神经网络的抽油机故障诊断研究

发布时间：2018-12-11 13:07

【摘要】：在油田开采生产中,采油设备如抽油机等因为大多数都是野外现场作业,不仅数量较多地理位置分散,而且周边环境恶劣,导致人工检测相对困难,由于抽油机井下的复杂环境导致故障时常发生,不利于对井下设备及开采过程进行实时监控,因此影响油田产量和效益。因此及时准确地诊断出抽油机的故障,有助于提高原油的开采效率,降低开采成本。首先,本文在分析原来的示功图特征提取方法的基础上,对示功图的理论进行介绍,并针对抽油机常见的多种故障类型,选取部分进行介绍。然后,对示功图进行预处理,并在最佳阈值下进行图像分割,得到边界为最大区域填充的图像。采用一种结合矩特征和傅里叶描述的方法对抽油机示功图进行故障诊断。计算预处理后图像的矩特征,获得表示物体形状的矩特征序列,再通过离散傅里叶变换得到具有平移、旋转及尺度不变性的归一化矩特征傅里叶描述法,并通过欧氏距离法判断该方法的快速和有效性。其次,介绍了人工鱼群算法和人工神经网络算法,然后针对BP神经网络收敛速度慢,容易陷入局部极值等问题,采用人工鱼群神经网络算法对抽油机进行故障诊断,运用Matlab仿真软件搭建人工鱼群BP神经网络模型,仿真结果验证本文方案的准确性和快速性。论文的研究,不仅丰富了抽油机示功图特征提取的方法,而且将鱼群—神经网络算法应用于抽油机故障诊断中进一步拓宽其应用领域,同时为实现抽油机故障诊断理论拓宽了思路。
[Abstract]:In the production of oil fields, oil production equipment such as pumping units, etc., because most of them are field operations, not only the number of geographical locations are scattered, but also the surrounding environment is bad, which makes manual detection relatively difficult. The complex environment of pumping unit often leads to failure which is not conducive to the real-time monitoring of downhole equipment and production process thus affecting the production and efficiency of the oil field. Therefore, it is helpful to improve the efficiency of crude oil extraction and reduce the exploitation cost to diagnose the faults of pumping unit timely and accurately. Firstly, based on the analysis of the original feature extraction method, this paper introduces the theory of indicator graph, and introduces the common fault types of pumping unit. Then, the indicator map is preprocessed and segmented under the optimal threshold value to obtain the image with the largest boundary. A method combining moment feature and Fourier description is used to diagnose the fault of the indicator diagram of the pumping unit. The moment feature sequence representing the shape of the object is obtained by calculating the moment feature of the pre-processed image, and then the normalized moment feature Fourier description method with translation, rotation and scale invariance is obtained by discrete Fourier transform. The Euclidean distance method is used to judge the speed and effectiveness of the method. Secondly, the artificial fish swarm algorithm and the artificial neural network algorithm are introduced. Then the artificial fish swarm neural network algorithm is used to diagnose the fault of the pumping unit in view of the slow convergence speed of the BP neural network, which is easy to fall into the local extremum. The artificial fish swarm BP neural network model is built by using Matlab simulation software. The simulation results verify the accuracy and rapidity of the scheme. The research in this paper not only enriches the method of feature extraction of power diagram of pumping unit, but also extends the application field of fish swarm neural network algorithm in fault diagnosis of pumping unit. At the same time, it broadens the train of thought for the fault diagnosis theory of pumping unit.
【学位授予单位】：东北石油大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TE933.1;TP183

【相似文献】