Abstract: The core temperature uniformity control of billet in heating furnace is very important to the stability of product quality,due to the high temperature environment in heating furnace, it is always a difficult problem to predict the core temperature of billet with high precision. In order to solve this problem, in this paper a temperature measurement method based on billet embedded thermocouple black box is established to effectively obtain the actual temperature distribution of billet at different positions in the heating furnace. Based on the experimental data of black box temperature measurement, the methods such as data cleaning,data smoothing and standardization areapplied, based on the data-driven neural network, random forest and XGBoost model, the unmeasured core temperature of billet is predicted by using the measurable gas temperature in the heating furnace. The prediction results of core temperature of GCr15 steel 150 mm x 150 mm billet show that the regression prediction effect of XGBoost model is the best, and the relative errors are mainly distributed in the range of 0%-5.4%. The absolute error of 97.1% of the sample points in the model is less than 10 °C , the RMSE error is 4. 1345 ℃ , and the MAPE error is 0.47%. The method of billet core temperature prediction based on billet embedded thermocouple black box temperature measurement + XGBoost model is proposed.

Key words: GCr15 Steel Billet Core Temperature , Black Box Experiment , Neural Network , Random Forest , XGBoost

摘要： 加热炉钢坯的心部温度均匀性控制对产品质量稳定性至关重要，由于加热炉中的高温环境，对钢坯心部温 度高精度预测始终是一个难题。为了解决这个难题，本实验建立了一种基于钢坯埋偶黑匣子温度测量方法，有效 获知加热炉内钢坯不同位置实际温度分布情况。基于黑匣子测温实验数据，采用数据清洗、数据平滑与标准化等 预处理方法，釆用基于数据驱动的神经网络、随机森林与XGBoost模型,利用加热炉中可测的炉气温度对不可测的 钢坯心部的温度进行预测。预测GCr15钢150 mm x 150 mm坯心部温度，结果表明:XGBoost模型回归预测效果最 好,相对误差主要分布在0%~5.4%，模型中97.1%的样本点绝对误差小于10 ℃，其RMSE误差为4.1345 ℃ , MAPE误差为0.47%。提出了钢坯埋偶黑匣子测温+ XGBoost模型预测钢坯心部温度的方法。

关键词: GCr15钢坯心部温度 , 黑匣子实验 , 神经网络 , 随机森林 , XGBoost