Carbon Content Measurement of BOF by Radiation Spectrum Based on Support Vector Machine Regression

Accurate on-line prediction of endpoint carbon is of significance for achieving control of end points, improving the quality of steel products, reducing energy consumption and reducing exhaust emissions. In order to solve the problems of endpoint control and carbon content measurement in converter smelting, a novel non-contact on-line method for detecting carbon content was proposed in this paper. The method realized BOF endpoint prediction and carbon content detection based on radiation spectrum with support vector regression. Firstly, a remote spectrum acquisition system was adopted to obtain flame information. Changes of flame radiation spectrum in smelting process were analyzed and spectral width and background radiation peak which characterize the overall spectral and intensity values of wavelength 600, 630 and 775 nm corresponding to emission peaks were extracted respectively and then used as inputs of support vector machines, combining decarburization theory and measured carbon value, the decarburization curve was reconstructed as output of support vector machine. The relationship model between spectral distribution and carbon content was established by support vector regression. The optimal parameters of the model were determined by training the sample set and the test set. The designed instrument and the optimized model have been installed in the converter production control room, field experiment results show that the accuracy of end-point carbon content prediction is 90.2%, and the measurement time is 0.3 s. It can be detected online in real time, and meet the production needs. The method provides an important basis for the precise endpoint control of the BOF.