油井结蜡参数检测与智能判别方法研究

发布时间：2018-07-29 10:28

【摘要】：石油是当今世界不可或缺的非再生能源,提高石油的采集效率,及时识别采油设备系统的故障显得尤为重要。目前,多数油田采用有杆泵抽油系统采集石油,该设备具有操作简单,成本较低等优点。在开采的过程中,设备经常出现结蜡故障,使抽油机载荷增加,电动机电流增大,严重影响有杆抽油系统的采油效率。当前已有很多识别结蜡故障的方法,但识别准确率并不理想。本文运用统计学习理论的支持向量机方法对有杆抽油系统结蜡故障进行智能识别,该方法的泛化能力强,尤其适合小样本的模式识别。论文主要进行以下研究:(1)介绍目前国内外有杆抽油系统故障诊断的背景,对有杆抽油系统的结构,工作原理进行阐述,研究示功图形成,并利用载荷传感器对油井的载荷参数进行提取。示功图含有诸多信息,可以根据它的图像特征了解油井的生产状况,所以选择示功图作为油井故障诊断的依据。(2)采集地面示功图后,对系统建立数学模型,将地面示功图转化为井下泵功图,更利于了解井下工况,并用MATLAB软件进行计算,得到所需泵功图。再将泵功图进行MATLAB图像处理,得知图像的具体类型,运用大律法进行阈值分割,得到最佳的阈值后转化为二值图像,所得图像利用数学形态学里的膨胀、腐蚀、细化、收缩进行处理,最终获得所需图像。(3)将处理好的泵功图进行特征参数的提取,提取的参数应具备区分性、聚类性以及独立性等,因此,选用不变矩理论提取泵功图的7个不变矩参数来描述各种油井故障的情况,为分类器模式识别提供数据样本。(4)选择支持向量机这一方法对有杆抽油系统的故障进行智能识别,尤其是结蜡故障。对支持向量机进行理论分析并用MATLAB软件进行仿真验证。为了得到更好的识别效果,分别运用了交叉验证法、粒子群算法和遗传算法对支持向量机的参数进行寻优。由于不同核函数识别效果不同,因此,对不同核函数的识别效果进行比较,寻得最优参数,达到良好的智能识别效果。
[Abstract]:Petroleum is an indispensable non-renewable energy in the world today. It is very important to improve the efficiency of oil collection and identify the fault of oil recovery equipment system in time. At present, the rod pump pumping system is used to collect oil in most oilfields, which has the advantages of simple operation and low cost. In the process of exploitation, the equipment often appears wax deposit fault, which makes the load of pumping unit increase and the electric current of motor increase, which seriously affects the oil recovery efficiency of the rod pumping system. At present, there are many methods to identify wax deposit fault, but the accuracy is not ideal. In this paper, the support vector machine (SVM) method based on statistical learning theory is used to identify waxing faults in rod pumping system. This method has strong generalization ability and is especially suitable for pattern recognition with small samples. The main contents of this paper are as follows: (1) the background of fault diagnosis of rod pumping system at home and abroad is introduced, the structure and working principle of rod pumping system are expounded, and the formation of indicator diagram is studied. The load parameters of oil well are extracted by load sensor. The indicator diagram contains a lot of information and can be used to understand the production condition of the oil well according to its image characteristics. Therefore, the indicator diagram is selected as the basis for fault diagnosis of the oil well. (2) after collecting the surface indicator diagram, the mathematical model of the system is established. If the surface indicator diagram is converted into the underground pump work diagram, it is more helpful to understand the downhole working conditions. The required pump power diagram is obtained by using MATLAB software. Then the pump work graph is processed by MATLAB image, and the specific type of image is obtained. The threshold value is segmented by using the great law, and the optimal threshold value is obtained. The image is transformed into a binary image by using the expansion, corrosion and thinning of mathematical morphology. The shrinkage is processed and the desired image is obtained. (3) the extracted parameters should be distinguished, clustered and independent, and so on. The moment invariant theory is used to extract 7 invariant moment parameters of pump power diagram to describe various oil well faults, and to provide data samples for classifier pattern recognition. (4) selecting support vector machine (SVM) to identify faults of sucker rod pumping system intelligently. Especially the waxing fault. The support vector machine is theoretically analyzed and simulated by MATLAB software. In order to obtain better recognition effect, the parameters of support vector machine are optimized by cross validation, particle swarm optimization and genetic algorithm respectively. Because different kernel functions have different recognition effects, the recognition effects of different kernel functions are compared, and the optimal parameters are found to achieve a good intelligent recognition effect.
【学位授予单位】：沈阳工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TE358.2

【参考文献】