主元数量对故障传感器重构精度影响分析

Evaluation of PCA-based Reconstruction Accuracy for Faulty Sensor in Different Number of Principal Components

传感器正常工作是冷水机组安全运行和优化节能的必要条件,但随着使用年限的增加,各类传感器故障时常发生。以Q统计量为检测指标的主元分析方法,常用于传感器故障检测、诊断与数据重构。由于主元数量的选取对于主元空间和残差空间的投影过程的建立具有较大影响,分析了不同主元数量对重构精度的影响规律。结果表明,主元数量越多,传感器重构数据的精度越高。

The normal working condition of sensors is the essential condition for the safety operation and optimalconservation for chillers. Unfortunately, the sensor faults are occurred easily and identified hardly as the increasing ofoperation period. A method based on Principal Component Analysis by Q-statistics is usually employed for the sensorfault detection, diagnosis and data reconstruction. The selection of the number for the Principal Component is the key forthe projection of the Principal Component Subspace and the Residual Subspace. The influence for the reconstructionresults by different numbers for the Principal Component is demonstrated in this paper. It＇s demonstrated that highernumber for the Principal Component results in more accurate ratios of the repaired data for the faulty sensor.