【摘要】 轧后冷却直接影响热轧钢板的质量，在冷却过程中冷却集管流量是至关重要的控制参数。由于不能对流量进行实时监测会导致实际总流量与设定总流量出现偏差，因此将自学习模型应用在轧后冷却系统的冷却流量预测中可以提高流量的预测准确度。采用ETR自学习模型，以钢板宽度、钢板冷却时间、冷却集管开启组数等8个变量作为影响因素，以AH36钢为例进行冷却流量回归预测并得到预测结果。同时，将网格搜索超参数优化的ETR模型与随机森林模型、RNN神经网络模型、LightGBM模型、默认参数的ETR模型以及DNN神经网络模型的预测结果进行对比，结果表明：使用网格搜索进行超参数优化的ETR模型的预测结果均方根误差和平均绝对误差均较小，决定系数更接近于1,相对于其他模型具有更高的预测准确度，适合应用于轧后冷却系统的冷却总流量预测。更多还原

【Abstract】 Cooling process after rolling directly affects the quality of hot rolled steel plates, and the flow rate of cooling headers is a crucial control parameter during the cooling process. The inability to monitor the flow rate in real time leads to deviations between the actual total flow rate and the set total flow rate. Therefore, applying a self-learning model to the cooling flow rate prediction in the cooling system after rolling can improve the prediction accuracy. The ETR self-learning model is adopted, with eight variables such as plate width, cooling time, and the number of opened cooling header groups as influencing factors. Taking AH36 steel as an example, regression prediction of the cooling flow rate is conducted, and the prediction results are obtained. Meanwhile, the prediction results of the ETR model optimized by grid search hyperparameter tuning are compared with those of the random forest model, RNN neural network model, LightGBM model, default-parameter ETR model, and DNN neural network model. The results show that the ETR model optimized by grid search hyperparameter tuning has smaller root mean square error and mean absolute error in its prediction results, with the coefficient of determination closer to 1, indicating higher prediction accuracy compared to other models. Thus, it is suitable for predicting the total cooling flow rate in cooling systems after rolling.更多还原