Fault diagnosis based on fully-correlated kernel partial least squares for pumping unit

DOI编号  10.7641/CTA.2020.90531
2020,37(9):2039-2046

 作者 单位 E-mail 汪波 重庆科技学院 981325938@qq.com 夏钦锋 中石化重庆涪陵页岩气勘探开发有限公司 钱龙 重庆科技学院 彭军 重庆科技学院 周伟 重庆科技学院

针对油田抽油机生产数据存在强非线性和强耦合性, 导致故障诊断困难的问题, 本文提出一种全相关动态 核偏最小二乘(FCDKPLS)故障诊断方法. 首先, 构建抽油机生产数据自回归模型, 反映数据变量间的动态特性; 其 次, 分析了KPLS算法中输出变量与输入变量残差子空间的相关性, 为此, 在输出模型上构建一个辅助矩阵, 从而表 征输入变量与输出变量的全相关性, 建立输入变量和输出变量之间更直接的联系. 最后, 将提出的全相关动态偏最 小二乘方法应用于抽油机过程故障诊断, 实验结果表明本文提出方法的有效性.

Fault diagnosis of pumping unit system is a challenging issue owing to the system that exhibits strong nonlinearity and strong coupling of the production parameters. In this paper, a fault diagnosis method based on fully- correlated dynamic kernel partial least squares (FCDKPLS) is developed for pumping unit system. First, auto regressive model of the production data is constructed to obtain the dynamic performance between the production data of pumping unit. Then, the correlation between the output variable and the input residual subspace is studied by the KPLS method. To address this issue, an auxiliary matrix based on the output model is developed to represent the fully-correlated between the input variable and the output variable. In particular, a more direct link between the input variable and the output variable can be obtained. The proposed FCDKPLS algorithm is applied to the pumping unit system, and experimental results show the effectiveness of the proposed approach.