摘要： 往复式注水泵作为油田注水作业的关键装备,由于恶劣的工作环境,易导致轴瓦、缸体和电机等部件故障,对其进行健康监测及健康度评价,以及时发现异常状态,保障安全可靠的注水作业十分必要.本文提出了一种基于支持向量数据描述的往复式注水泵健康度量化评价方法.首先,考虑到注水泵振动信号非线性、非平稳的特点,在有限条件下提取注水泵振动信号中的特征信息.针对注水泵核心部件多且间距小的问题分析振动信号的测点位置,以搭建数据采集系统.针对振动信号频率成分复杂的特点,利用变分模态分解提取振动数据的变分模态分量,在所有模态分量排列熵的基础上构建高维多域特征集,以描述注水泵的健康状态.其次,针对注水泵实际采集数据过程中,大多数设备处于健康工作状态,故障数据较少,容易造成样本不均衡的问题,利用单值分类方法支持向量数据描述在单值分类问题上的优势,仅使用健康状态运行下注水泵数据样本的特征向量进行支持向量数据描述的超球体模型构建,并引入粒子群优化算法对模型参数进行优化;通过计算注水泵不同健康状态的数据到超球体球心的距离,参考隶属度函数进行公式拟合,实现注水泵的健康度定量评价.最终,为验证该评价方法的适用性,对长庆油田现场注水泵的振动数据进行健康度评价计算,粒子群优化后模型准确率能达到95%,本文所提出的方法具有较好的准确性.

关键词: 健康度评价, 变分模态分解, 支持向量数据描述, 粒子群优化, 超球体模型

Abstract: The reciprocating water injection pump is the core equipment for the oilfield water injection operations.However,due to the harsh working environment,it is easy to cause frequent failures of bearings,cylinders,motors,and other components.Therefore,the health monitoring and the health evaluation of the reciprocating water injection pump are required to detect abnormal conditions in a timely manner to ensure safety and reliability for reciprocating water injection operation.This paper proposed a quantitative evaluation method for the health of the reciprocating water injection pump based on support vector data description.Firstly,considering the nonlinear and non-stationary characteristics of the reciprocating water injection pump vibra-tion signal,the characteristic information in the water injection pump vibration signal was extracted under limited conditions.Aiming at the problem that the water injection pump had many core components and small spacing,the measurement point locations of the vibration signals were analyzed to build a data acquisition system.In view of the complex frequency components of vibration signals,variational modal decomposition was used to extract the variational modal components of vibration data,and a high-dimensional multi-domain feature set was constructed based on the permutation entropy of all modal components to describe the health status of the water injection pump.Secondly,in view of the fact that during the actual data collection process of the water injection pump,most of the equipment was in a healthy working state and had less fault data,which easily caused sample imbalance.The single value classification method was used to support the advantages of vector data description in single value classification problems.Only the eigenvectors of the water injection pump data samples running in the healthy state were used to construct the hypersphere model described by the support vector data,and the particle swarm optimization algorithm was introduced to optimize the model parameters;the data of the different health states of the water injection pump were calculated to the hypersphere model distance from the center,and perform formula fitting with reference to the membership function to achieve quantitative evaluation of the health of the water injection pump.Finally,in order to verify the applicability of this evaluation method,the health evaluation calculation was performed on the vibration data of the on-site water injection pump in Changqing Oilfield.After particle swarm optimization,the model accuracy could reach 95%,which showed that the method proposed in this article has good practicability.

Key words: health evaluation, variational modal decomposition, support vector data description, particle swarm optimization, hypersphere model