Research on Measurement of Temperature Field of Furnace Flame Based on SVM
|School||Hefei University of Technology|
|Course||Detection Technology and Automation|
|Keywords||Temperature Field Measurement 3-d Reconstruction Support VectorMachine (SVM) Wavelet Kernel Function|
Electric Power Consumption has been everyone’s necessaries in life, as the increase in demand of using of electricity. How to make the power station boilers run safety is an important question for the power systems divisions. It has cooperative ties with everybody who uses the power. However, potential safety hazard of power station boilers combustion system is still being existence, because the temperature field measurement technology is not advanced. So advanced and effective temperature field measurement technology of power station boilers combustion system has a great science significance and practical value for the research of power station boilers system. Measuring temperature field of power station boilers truly is the core problem of combustion diagnosis. The traditional temperature field measurement method is to establish function relation between flames temperature and flame images and then solve the matrix equation. However, the actual industrial process is often shown for multi input-output, strong nonlinear, strong coupling characteristics, besides some of the parameters in the furnace (black smoke concentration distribution, the scattering of wall rate, etc) are difficult to be determined. Temperature field measurement method has limitations using the mechanism model. The support vector machine is a new type of machine learning algorithm based on statistical learning theory. It has good nonlinear function fitting and generalization ability. It has been widely used in system identification, intelligent modeling, etc. This paper tries to build mathematical model of boiler burning system by SVM identification method. Specific content shown below:(l)We discuss the characteristics of the power plant boiler combustion flame, the basic principles of flame radiation imaging and the temperature detection methods based on the mechanism modeling. Then we have further study of machine learning, statistical learning theory and support vector machine theory. At last we proof that the SVM applied in flame temperature field detection is feasible.(2)The flame temperature testing system is a multiple output system, however support vector regression machine generally apply to single output system.This paper puts forward a kind of multiple input multiple output SVM regression model algorithm. This paper shows us the linear and nonlinear regression function under the condition of the basic solution then knocked down a multiple output SVM regression algorithm.(3)This paper researches how to choose kernel function of SVM. At present gaussian kernel function is relatively commonly used kernel function. It made a good performance in pattern recognition and regression analysis. But the radial basis is not a complete basises, can not approximate any signal. Due to good time-frequency,multi-resolution characteristics and good approaching speciality this paper constructed wavelet kernel function and multi-scale kernel function. Through the simulation test analysis I find that it solved the problem that strong noise influence and coupling exist in measurement of the temperature field.Because traditional parameters optimization method has defects, this paper presents a new kind of parameters optimization method-the particle swarm optimization method. In this paper I introduce the particle swarm optimization algorithm in detail and train SVM model by particle swarm algorithm.