转炉吹炼终点碳锰预测数据驱动模型

For the carbon and manganese content at the end point of converter converting,data driven prediction models for carbon and manganese content at the end point of converter blowing were established by using big data and machine learning algorithm LGBM(Light Gradient Boosting Machine).The input variables of the model include the composition of hot metal,the weight of hot metal,the weight of scra,oxygen consumption and the addition amount of slagging formers.The learning curve parameter adjustment method was used to determine the optimal hyperparameter values of the two models.The research results indicate that under the premise of big data,balanced data distribution is beneficial for improving the predictive performance of data-driven models established by machine learning algorithms.Within the error range of(-0.025%,0.025%),(-0.020%,0.020%),and(-0.015%,0.015%),the hit rates of the endpoint carbon prediction model are 95.43%,94.06%,and 89.73%,respectively;The hit rates of the endpoint manganese prediction model were 97.00%,95.69%,and 90.61%,respectively.The root mean square error(RMSE)of the endpoint manganese prediction model is near to that of the endpoint carbon prediction model,but its coefficient of determination(R2)is higher and closer to 1 than that of the endpoint carbon prediction model.