The Application Of Principal Component Analysis In Power Plant Fault Detection And Its Effect Study

Power plant technical transformation is now one of the mainstream directions of technical and economic research, among them, the technical transformation is mainly by improving their physical structure or adds new devices, such as increasing the desulfurization device, or reconstructs the coal mill. However, technical transformation is not only the physical transformation; it should also include the improvement of electronic technology, such as the control algorithm and improved detection methods. In these areas, the domestic related research is not much. In this paper, by studying the power plant operation, found in fault detection, the vast majority of domestic thermal power plants are using fixed control limit to an error, the actual operation process is not continuous steady operation of power plant, which leads to the increase or decrease of load in power plant system misstatement or loss of data, seriously affecting the stability of the monitoring, which affects the efficiency of power plant.Method of principal component analysis（PCA） as a kind of multivariate statistical method, its fault detection and diagnosis methods of process control is not dependent on the mathematical model of the system, so it is widely used in the field of industrial statistics detection method. Process detection method based on principal component analysis method, by fully using principal component analysis algorithm in dealing with a linear correlation data dimension reduction function, makes the detection of multivariable process can be realized in low-dimensional variable control. As a result, using principal component analysis method in fault detection may help. The PCA method in the fault detection principle should be used in steady state while during data collection, using a fixed control limit of SPE, T₂ and T_H². However, for those transient process（e.g., power plant operation system）, which with a fixed control limit, the method cannot be used directly. Or that will seriously affect the stability of the monitor system, and lead to unnecessary economic losses. Based on that, the principal component analysis method was improved to make it used in dynamic steady state, which often appeared in power plant operation system.This paper summarizes the field in the thermal power plant fault detection in the domestic and foreign research results, points out the problems which need to be studied further in this field, puts forward the improved method based on adaptive control limit of variance, and thus to overcome in the transient process of the problem of false positives due to conditions changes. Selecting data from thermal power plant, the method make use of matlab software simulation, verify the feasibility of the technical transformation. Through the establishment of evaluation index, using the analytic hierarchy process（ahp） to determine the index weight and the fuzzy evaluation method to plan for comprehensive evaluation, the result shows that the technical reform can be gained for the thermal power enterprise economic benefits and technical improvements.