PCA在非线性系统传感器故障检测和重构中的应用

由于主成分分析(PCA)方法是一种线性算法,基于PCA的故障检测方法若直接运用于非线性系统的传感器故障检测和重构,会导致明显的故障误报和数据重构错误。为了使基于PCA的传感器故障检测和重构方法适用于非线性较严重的热工过程,对该方法进行了有效的改进。应用不同负荷下的历史数据,分别建立机组不同负荷下的局部PCA模型,再根据机组当前实际运行负荷选择相应的PCA模型进行传感器故障检测和重构,并结合相邻负荷PCA模型的计算结果进行数据融合,从而进一步提高了故障检测的准确性和重构精度。理论分析和现场实际应用表明,该算法能够对非线性较为严重的电厂热工过程进行精确的传感器故障检测和重构。

Nonlinear sensor fault detection and data rebuilding based on principle component analysis

Fault detection method based on principle component analysis(PCA) which uses linear arithmetic is not appropriate for nonlinear systems because of inaccuracies in the fault mismatch detection and data rebuilding.An improved PCA arithmetic model is applicable to nonlinear plant thermal processes using local PCA model for various power loads based on the historical data.Then an appropriate PCA model was selected for each power load.Finally,data from adjacent model outputs was rebuilt through data fusion.The method improves the accuracy of fault detection and data rebuilding.Theoretical analyses and practical results show that the arithmetic can be accurately used for sensor fault detection and data rebuilding for a wide range of plant thermal processes.