The Research On Temperature Fields Reconstruction Algorithm Based On Neural Network

The measurement and reconstruction of coal boiler flame temperature fields is a difficult problem at all times. This paper discuss with the technique of temperature fields reconstruction of coal boiler, and carry out research work on some key issue of this area. The paper have completed following works.The paper has summarized the actuality, difficulty and evolution trend of technology of the coal boiler temperature fields measurement.The paper put forward a new temperature fields reconstruction algorithm based on Radical Basis Functions (RBF) Nrueal Network, and carry out some simulation expriments and research work. For these single-peak temperature model, asythetic single-peak temperature model and double-peak temperature model, the reconstruction results are indicated by maximum relative error, relative mean error and mean square error as 6.1% 1.2% 1.5%; 8.1% 3.0% 3.7% and 5.7% 0.7% 1.1% respectively. Result shows that the algorithm has prior reconstruction precision.Regressive analysis of random samples reconstruction results have been carried out, and result of maen of 100 reconstruction results show that maximum relative error, relative mean error and mean square error as 5.7% 1.2% and 1.5%. Results show that the result of temperature fields reconstruction tally with original temperature model well.Serve as researching work of measure error impress on temperature fields reconstruction results, the paper look upon measure error as random noise with normal distribute, and carried out temperature fields reconstruction under level of 40dB, 30dB and 24dB. The main square error of reconstruction results are 1.5% 4.0%, 1.1%; 1.6% 4.4% 1.1%and 2.2% 5.4%, 1.5% respectively. The results shows that the algorithm can reconstruct temperature fields under these noise level, thus it has anti-noise ability.Asymmetrical burning flame and symmetrical burning flame temperature fields reconstruction experiments have been carried out using the algorithm and addition with measurement system. The reconstruction results have been compared with FouierRegularizing Algorithm, Least-square Algorithm and Gaussian Function Regularizing Algorithm. The result shows that the new algorithm have prior reconstruction quallity than others.