基于CNN-SVR的作业车间订单完工周期预测方法  

Method of predicting order-completion time in workshops based on CNN-SVR

作　　者：于嘉惠 李铁克[1,2] 王柏琳[1,2] 张文新[1,2] 张卓伦 袁帅鹏 YU Jia-hui;LI Tie-ke;WANG Bai-lin;ZHANG Wen-xin;ZHANG Zhuo-lun;YUAN Shuai-peng(School of Economics and Management,University of Science and Technology Beijing,Beijing 100083;Engineering Research Center of MES Technology for Iron&Steel Production,Ministry of Education,Beijing 100083)

机构地区：[1]北京科技大学经济管理学院,北京100083 [2]钢铁生产制造执行系统技术教育部工程研究中心,北京100083

出　　处：《机械设计》2023年第2期57-64,共8页Journal of Machine Design

基　　金：国家自然科学基金资助项目(71701016,71231001);教育部人文社会科学研究青年基金项目资助(17YJC630143);北京市自然科学基金项目(9174038);中央高校基本科研业务费资助项目(FRF-BD-20-16A)。

摘　　要：经典作业车间生产环境包含多产品、多机器和多工序,提高其完工周期的预测精度对于企业提高客户满意度、优化生产调度等方面都具有重要意义。因此,针对作业车间订单完工周期预测问题提出一种卷积神经网络-支持向量回归(CNN-SVR)方法。首先,将订单完工周期影响因素分类为订单信息与车间实时状态信息,并分析得到其中的关键特征因素。进而,采用ReLU激活函数训练卷积神经网络,对生产数据特征进行自适应提取,并将结果输入至支持向量回归模型中进行预测。最后,设计FlexSim作业车间仿真模型生成车间订单生产数据,确定评价指标并进行试验验证。结果表明,相较于其他对比模型,CNN-SVR预测方法在拟合优度和预测误差等方面均有很好表现,能够得到理想的预测效果。The traditional workshop environment involves multiple products,machines and processes.High accuracy in predicting the order-completion time is of great significance for enterprises to improve customer satisfaction and optimize production scheduling.Therefore,in this article the Convolution Neural Network-Support Vector Regression(CNN-SVR)method is proposed to predict the order-completion time in the workshop.Firstly,the influencing factors of the order-completion time are classified into the information on the order and the information on the workshop real-time status;and the key characteristic factors are analyzed.Furthermore,the convolution neural network is trained by means of ReLU activation function,in order to extract the features of production data in an adaptive manner,and the results are input into the SVR model for prediction.Finally,the Flexsim workshop simulation model is designed to generate the order-production data,determine the evaluation indicators and conduct experimental verification.It’s shown that compared with other models,the CNN-SVR prediction model performs well in fit goodness and error prediction,and as a result,ideal prediction results are obtained.

关 键 词：订单完工周期 完工周期预测 卷积神经网络 支持向量回归 仿真模型 

分 类 号：TH164[机械工程—机械制造及自动化]