基于深度学习的棒材轧机电机负荷预测模型  

Load forecasting model of bar mill motor based on deep learning

作　　者：王文汇 赵宪明[1] 张令华 倪晓东 翁莉 WANG Wenhui;ZHAO Xianming;ZHANG Linghua;NI Xiaodong;WENG Li(State Key Laboratory of Rolling and Automation,Northeastern University,Shenyang 110819,China;Technology Center,Fushun New Steel Co.,Ltd.,Fushun 113001,China)

机构地区：[1]东北大学轧制技术及连轧自动化国家重点实验室,辽宁沈阳110819 [2]抚顺新钢铁有限责任公司技术中心,辽宁抚顺113001

出　　处：《冶金自动化》2023年第2期82-88,共7页Metallurgical Industry Automation

摘　　要：为了确保有效利用轧机设备能力,以某钢厂棒材生产线的轧机电机负荷数据为研究对象,利用PyTorch搭建基于长短期记忆(long short term memory,LSTM)神经网络的预测模型,定义模型初始网格结构参数,选定单元结构激活函数,并针对模型超参数的选择问题,采用自适应学习算法(adaptive moment estimation,Adam)进行参数优化,迭代降低损失值,提高模型的预测精度。通过试验设计,采用生产两种规格棒材的轧机负荷数据进行验证,结果表明,与未优化的负荷预测模型对比,均方误差SME分别降低了3.28、1.76,证明了所建立模型的预测效果更好,具有较高的稳定性。In order to ensure the effective use of rolling equipment capacity,taking the mill motor load data of a bar production line in a steel plant as the object of study,PyTorch was used to build a prediction model based on long short term memory(LSTM)network,define the initial grid structure parameters of the model,and select the cell structure activation function.For the problem of model hyper-parameters selection,the adaptive moment estimation(Adam)algorithm was used to optimize the parameters,iteratively reduce the loss value and improve the prediction accuracy of the model.Through the experimental design,the mill load data for two bar sizes were used to verify,and that the mean square error SME is reduced by 3.28 and 1.76,respectively,compared with the unoptimized load prediction models.The results show that the established models have better prediction effect and higher stability.

关 键 词：负荷预测 长短期记忆神经网络 深度学习 超参数 预测模型 轧机 电机负荷 

分 类 号：TG333[金属学及工艺—金属压力加工] TP18[自动化与计算机技术—控制理论与控制工程]