基于深度学习的热连轧轧制力预测

摘要：轧制力预报一直是热连轧过程控制模型的核心，浅层神经网络对复杂函数的表示能力有限，而深度学习模型通过学习一种深层非线性网络结构，实现复杂函数逼近。利用深度学习框架TensorFlow，构建了一种深度前馈神经网络轧制力模型，采用BP算法计算网络损失函数的梯度，运用融入Mini batch策略的Adam优化算法进行参数寻优，采用Early stopping、参数惩罚和Dropout正则化策略提高模型的泛化能力。基于上述建模策略，针对宝钢1880热连轧精轧机组的大量轧制历史数据进行了建模实验，对比分析了4种不同结构的前馈网络预测精度。结果表明，相比于传统SIMS轧制力模型，深度神经网络可实现轧制力的高精度预测，针对所有机架的预测精度平均提升21.11%。

Prediction of hot-rolled roll force based on deep learning

Roll force prediction has always been the core of the hot rolling process control model. The shallow neural network has limited ability to express complex functions，while the deep learning model realizes complex function approximation by learning a deep nonlinear network structure. A deep feedforward neural network roll force model is constructed by using the deep learning framework called TensorFlow. The BP algorithm is used to calculate the gradient of the network loss function. The Adam optimization algorithm integrated into the Mini batch strategy is used to optimize the parameters. Early stopping，parameter penalties，and Dropout regularization strategies are used to improve the generalization capabilities of the model. Based on the above mentioned modeling strategy，a large number of rolling historical data of Baosteel 1880 hot rolling finishing mill were used to carry out modeling experiments，and the prediction accuracies of four different structures of feedforward network were compared and analyzed. The experimental results show that compared with the traditional SIMS roll force model，the deep neural network can achieve high precision prediction of roll force. The prediction accuracy of roll force for steel of all stands is improved by 21.11% on average。