基于XGBoost算法的转炉吹炼终点预报

Endpoint prediction of converter steelmaking based on XGBoost algorithm

转炉吹炼终点温度和成分控制是转炉吹炼后期的重要操作,精确的温度和碳的预报十分重要,为了提高转炉吹炼终点碳、温命中率,利用特征相关性分析法确定模型的主要输入变量,基于XGBoost算法建立了转炉吹炼终点预报模型,并结合实际的转炉生产数据进行模型验证。通过与采用BP、优化BP神经网络建立的模型结果进行对比,表明XGBoost算法模型在保证收敛速度快的前提下,能够达到较高终点命中率,为进一步提高终点温、碳命中率,对XGBoost算法模型中几个重要建模参数值做出优化调整,最终XGBoost算法模型在出钢温度偏差为±15℃、±10℃时终点命中率分别为95.84%、91.69%;出钢碳质量分数偏差为±0.015%、±0.01%时终点命中率分别为93.31%、87.84%。

The control of end-point temperature and composition in converter steelmaking is an important operation in the later stage of blowing.The accurate temperature and carbon prediction is very important.In order to improve the end-point carbon and temperature hit rates of the converter blowing,main input variables were determined with characteristic correlation analysis method and the converter end-point prediction models based on the XGBoost algorithm were built up.The models were verified from the actual converter production data.In comparison with the results obtained from BP and optimized BP neural network models,it was found that the XGBoost algorithm model can achieve a high end-point hit rates under the premise of ensuring a fast convergence speed.In order to further improve the hit rates of end point temperature and carbon,the several important parameters in the XGBoost algorithm model were adjusted accordingly.Finally,when the temperature deviations were±15℃and±10℃,the hit rates of end point with XGBoost algorithm model were 95.84%and 91.69%,respectively.When the deviations of carbon mass fraction in molten steel were±0.015%and±0.01%,the end point hit rates were 93.31%and 87.84%,respectively.