Thermal Power Plant Sensor Measurement Data To Test And Fault Diagnosis,

With the development of power plant information control system, the requirement of the accuracy and reliability of sensors become gradually higher. As the working environment of sensor is bad, sensor failures happen easily with different kinds of faults. Through the analysis and modeling towards history data, sensor fault can be separated, identified and corrected with a substitute value, which is of great importance to the control and performance optimization of power plant.In this paper, Several key problems in the process of developing PLS model are discussed. The principal on how to select related variables, cross validation on confirm the numbers of the components and the forecast regression function are given in the following dissertation. The research work is based on the 24 hours data from PI server database of two units (135MW). Partial least square method (PLS) is adopted to develop the simulating model to detect and diagnosis the sensor faults. This approach performs well when multicollinearity of variables exists and has the virtues of fast calculating speed and strong real-time performance.The principals and methods of PCA are discussed, and several typical statistical indexes which are used in PCA is proposed.Four familiar forms of sensor faults are presented. Fault diagnosis and identification is pursued according to the variance between the residuals. Fault threshold and diagnosis window length is studied. The performances of diagnosis program in different forms and magnitude of faults are tested in this paper.The programs on this project are developed with the platform of Microsoft Visual C++. It was tested by real-time data in power plant to assure its reliability and accuracy.