Research On Prediction Of Converter Steelmaking Endpoint Temperature Based On Flame Image And Spectral Characteristic

Converter steelmaking is the most important mode of steel production in China.The temperature and composition of molten steel at the end of steelmaking determine the quality of final tapping.At present,most small and medium-sized converters in China still use manual fire watching to determine the end temperature,and the determination accuracy needs to be improved.The change of furnace mouth flame in the steelmaking process reflects the change of temperature in the furnace.By extracting the key features of flame image and spectrum,this paper establishes the mapping model between furnace mouth flame features and molten steel temperature based on neural network method,and realizes the high-precision prediction of endpoint temperature.In this paper,an improved colorimetric temperature measurement method is used to calculate the temperature corresponding to the flame image;The first moment,second moment and third moment of color are used to describe the color characteristics of the image;The texture features of the image are expressed by the corner second-order moment,contrast and other feature descriptors of the color co-occurrence matrix.A competitive adaptive reweighted sampling algorithm combined with the characteristic variable selection method of random leapfrog algorithm is proposed to select the flame spectral wavelength and obtain the characteristic wavelength variable of the spectrum.The extracted image features and spectral features are fused and analyzed.Aiming at the problem that the full feature collection dimension is high and includes variables with low correlation with molten steel temperature,this paper uses Pearson correlation coefficient method to screen the flame features and obtain a new feature set.Taking these feature sets as input,the prediction model of converter steelmaking end-point temperature is established based on BP neural network and PSO-BP neural network,and the prediction effects of different models are compared and analyzed.The test of the data in the actual production process of converter steelmaking shows that when the prediction model of converter steelmaking end-point temperature is established by PSO-BP neural network and the feature set extracted by Pearson correlation coefficient method is used as the input variable of the model,the model has higher prediction accuracy and can meet the production requirements of steelmaking site,which is of great significance to the prediction of converter steelmaking end-point temperature.