Fault Diagnosis For Control Systems Of Thermal Power Plant Based On Multivariate Statistical Analysis

Along with the continuous developments of automatic control technique, the control strategy has gradually matured in thermal power plant control systems and forms a comparatively perfect system. So the research on control theory of thermal power plant is gradually developed to a higher level. One of the important directions is fault diagnosis in thermal power plant control systems. As scale expanding and complexity increasing, the number of fault points in control systems has increased and they are more difficult to detect, which sets higher demands on reliability and maintenance of control system. The traditional maintenance of alarm and manual point-check obviously couldn't meet the requirements, thus it is necessary to research on the advanced fault diagnosis technology for the thermal power plant control systems.Considering the characteristics of large scale and complex structure in the thermal power plant control system, the multivariate statistical analysis method is studied in the paper, which doesn't rely on mathematical models. Two multivariate statistical analysis tools, principal component analysis (PCA) and Fisher discriminant analysis (FDA), are used to implement control system fault diagnosis by means of process data analysis. The PCA method is mainly used to research on fault detection while FDA to fault isolation and identification. The two methods interact and make up a complete fault diagnosis system. The main work of the paper includes following aspects:First, the superiority and feasibility are analyzed when applying multivariate statistical method to fault diagnosis in thermal power plant control systems. It is clarified that the scope,objects and theoretical methods of fault diagnosis in thermal power plant control systems.Second, fault detection in steady state is studied when using PCA in thermal power plant control system. It emphasizes on the connotative analysis of PCA fault detection and a conclusion is presented which could guide the fault detection. Data preprocessing in PCA fault detection is studied and the PCA fault detection method for the sensor and actuator is proposed.Third, against the varying load condition characteristic existed in thermal power plant process, a fault detection method based on dynamic multi principal component model is proposed. The dynamic principal component model is obtained by a fuzzy system, which can solve the working condition compatibility problem of principal...