Application Research Of Multi-scale Principal Component Analysis In Pattern Recognition Of Thermal System

The power generation process of thermal power units is a relatively complex process.The thermal system of the power plant is a typical multi-input,multi-output and multi-variable system.There is a strong coupling among each variable system,and the time scale difference of dynamic characteristics between different input and output signals is enormous.Moreover,there are strong nonlinear and time-varying between the input and output signals.So it is difficult to directly analyze the mechanism of the object.In view of the complex thermodynamic system,this thesis involves some data analysis methods,the statistical analysis to a great deal of unit historical operation data,discoveries of some specific rules,and the targeted mechanism analysis which in order to reduce the difficulty of mechanism modeling,soft sensor and state reconstruction of thermal power units.The common data analysis methods include principal component analysis,wavelet transform and so on.However,as the traditional principal component analysis method is only suitable for linear systems and static object processing,the single principal component analysis method is not applicable for complex thermal systems.The thermal system is non-linear as a whole,but it can be regarded as linear at a certain time scale.Therefore,wavelet transform can be used to decompose the signal of the thermal system in multi-scale.Then the principal component analysis is carried out on multiple scales to find out the rule of data that cannot be obtained by conventional methods.In this thesis,the data processing method of multi-scale principal component analysis is used to analyze and find the data rules of complex thermal systems.The change of coal feeding capacity caused by the increase of flexibility of thermal power units and the blending of low-quality coal are easy to cause the blocking of coal mills and affect the safe and stable operation of the unit.The multi-scale principal component analysis method is used to analyze the operating state of the coal mill,and the appropriate wavelet is selected to decompose the data of the relevant variables of the coal mill in the stable operation of the thermal power plant,and the principal component model is established on several scales respectively.By selecting real-time data and using multi-scale principal component model to detect the running state of coal mills,comparing the results of traditional principal component analysis and multi-scale principal component analysis,it is found that the detection results of multi-scale principal component analysis are more comprehensive.And the signal variables causing the change of the running state of coal mill are analyzed by contribution diagram method.The results of the analysis are consistent with the actual operation condition,which shows that the method of multi-scale principal component analysis is feasible.Because of the improvement of the flexibility of the thermal power unit and the unit is frequently operated under ultra-low load conditions,which is highly likely to cause unstable combustion and even cause the furnace to extinguish the fire,so it is very important to judge the combustion stability of boilers.The multi-scale principal component model is established by selecting the variable data related to combustion in thermal power units to judge the combustion stability of boilers.Compared with the results of multi-scale principal component analysis and traditional principal component analysis,it is found that the instability of combustion can be detected in advance at a high frequency scale,and the signal variables which cause the change of combustion state can be judged from the contribution diagram,which is consistent with the actual operation condition.Therefore,the multi-scale principal component analysis method can assist the operators to predict the combustion stability of the boiler.