融合轧制机理和深度学习的带钢精轧宽度预测

宽度精度是热轧带钢成形过程的重要指标,准确预测精轧宽度有助于及时修正粗轧宽度设定模型,提高成品带钢的宽度精度。然而,依据轧制机理建立的宽度预测模型偏离实际工况从而精度较低,依据神经网络建立的模型由于过程黑箱导致可信度低。为此,提出了一种融合轧制机理和人工神经网络的热轧带钢精轧宽度组合预测模型,以基于Hill公式的机理模型计算精轧宽度的预测基准值,以基于深度置信网络(DBN)的深度学习模型预测精轧宽度的修正值。选取实际生产的2 730组数据中的49个特征值作为试验数据进行建模分析,结果表明:该组合模型预测精度高、稳定性好且预测时间短,其均方根误差为0.428 15 mm,相比机理模型降低了79.6%,相比神经网络模型降低了6.2%,实现了精轧宽度的高精度预测。

Hot finishing rolling strip width predicting model based on rolling mechanism and deep learning

In the metal forming process of hot rolling, accurate prediction of finishing rolling strip width is helpful to modify roughing rolling model and improve product quality.However, the strip width prediction model based on rolling mechanism has low accuracy due to the changing actual condition and the model based on neural network has low credibility due to its black box.Therefore, a combined prediction model of hot rolled strip width based on rolling mechanism and artificial neural network was proposed to solve the above problems.The mechanism model based on Hill formula was used to calculate the prediction benchmark value of finishing rolling strip width, and the deep learning model based on deep belief network (DBN) was used to predict the correction value of finishing rolling strip width.49 features of 2 730 data sets in actual production were selected as experimental data for modeling and analysis.The experimental results show that the combined model has high prediction accuracy and favorable stability.Its root mean square error is 0.428 15 mm, which is 79.6% lower than the mechanism model and 6.2% lower than the neural network model, the high-precision prediction of finishing rolling strip width is realized.