有杆抽油系统故障递阶诊断的故障分辨研究

有杆抽油系统工况极其恶劣，发生故障的概率较高。针对有杆抽油系统的特点及其故障诊断目前存在的问
题，提出基于示功图的有杆抽油系统故障递阶诊断法，该方法分为故障分辨和故障识别两个阶段，即首先在故障分辨
阶段根据正常（或平稳状态）样本的统计规律把示功图分为故障类和非故障类；然后采用基于示功图的有杆抽油系统
故障识别的搜索树方法，对故障样本进行故障类型的识别。它提高了故障诊断系统的故障诊断性能，且不需要建立和
求解有杆抽油系统力学模型，也不存在训练集问题，并能反映出有杆抽油系统自身的基本特征。故障分辨过程包括：
在训练阶段，t 检验剔除异常数据后，通过X2拟合优度检验确定样本的随机分布形式及其参数，进而计算出正常区域
和故障区域；在分辨阶段，根据测试样本是否有特征量落入故障区域来判断其分类。实例表明故障分辨过程能从大量
的数据中筛选出故障样本，分辨的正确率比较高；训练样本的统计规律能反映油井实时的生产状况。

Hierarchical Fault Diagnosis of Rod Pumping System Based on Fault Distinguishing

There is higher fault probability under the bad work conditions of rod pumping system. Because of the large amount
of sample and the long-time steady status，this paper proposes hierarchical fault diagnosis method. Fault distinguishing and
fault identification are two phases in this method. In fault distinguishing，fault phase and normal phase are classified according
to statistical rule of normal sample in dynamometer card. Combining statistical theory and search-tree，fault identification is
processed in fault sample. Using this method，mechanical model of rod pumping system is not to be established and training
set of fault sample is not necessary. Fault distinguishing process includes：fault taring phase and fault distinguishing phase.
In fault training phase，stochastic distribution rule and parameters of samples are determined by X2 goodness-of-fit test after
abnormal samples are rejected，then normal area and fault area is calculated. In fault distinguishing phase，classification of
fault is estimated according to whether or not feature value of test sample is in fault area. Examples indicate fault samples can
be selected from large number of samples in distinguish fault and correct rate of fault distinguishing is high. Statistical rule of
training samples can reflect the actual production of oil well.