基于支持向量机的地面驱动螺杆泵井工况诊断技术

地面驱动螺杆泵井举升工艺的故障频发,限制了其进一步发展.为了提高地面驱动螺杆泵井的经济效益和管理水平,提出了基于支持向量机的地面驱动螺杆泵井工况诊断技术.选取产量、动液面等8个表征油井工作状态的变量作为输入参数,常见10种螺杆泵井工况作为输出参数,以东胜公司金家油田已存故障地面驱动螺杆泵井为基础,建立诊断样本集,采用投票法建立C²₁₀=45个子分类器,基于网格寻优、遗传算法和粒子群寻优算法对C,g进行优化计算,借助Matlab调用Libsvm工具箱对支持向量机模型进行训练,利用东胜公司金家油田15口地面驱动螺杆泵井进行诊断验证,并与人工神经网络方法进行了对比.结果表明:采用支持向量机诊断方法诊断正确的有14口油井,诊断正确率为93.33%,与人工神经网络方法(88.90%)相比具有更高的精度,在小样本诊断问题中具有更强的优势,是一种切实可行的智能诊断方法.

Working conditions diagnosis of surface-driving progressive cavity pump wells based on support vector machine

The frequent faults of surface-driving progressive cavity pump(PCP)wells limit their further development. In order to improve the economic benefits and management level of surface-driving progressive cavity pump wells, a diagnosis method of working conditions of surface-driving PCP well based on support vector machine was proposed. The condition types of PCP wells were subdivided into 10 categories as outputs, and 8 variables were selected as inputs which can represent operation situation of oil wells. Sample sets were established based on the already existing failure oil wells of Jinjia oilfield in Dongsheng group company, and 45 classifiers were built using voting method. Best C and g were determined by three methods including grid optimization, genetic algorithm and particle swarm optimization. The Libsvm toolbox called by Matlab was used to establish and train the SVM model, and 15 PCP wells of Jinjia oilfield in Dongsheng group company were diagnosed to verify the SVM model, and the comparison between support vector machine and artificial neural network was conducted. The results show that the diagnosed working conditions of 14 PCP wells by the SVM method are in accord with their actual working conditions, with an accuracy of 93.33%. Compared to artificial neural network(88.9%), the SVM is more accurate, which is superior for small sample problems, and is a feasible diagnosis method for PCP wells.