融合LSTM预测需求的线型材料多批次优化下料方法  

An Approach Based on Demand Prediction with LSTM for Solving Multi-batch One-dimension Cutting Stock Problems

作　　者：庞凯民 朱波 张宏硕 刘宁 张连富 PANG Kai-min;ZHU Bo;ZHANG Hong-shuo;LIU Ning;ZHANG Lian-fu(Kunming University of Science and Technology,Kunming 650500,China;Shenyang Machine Tool(Group)Co.,Ltd.,Shenyang 110142,China)

机构地区：[1]昆明理工大学机电工程学院,云南昆明650500 [2]沈阳机床(集团)有限责任公司,辽宁沈阳110142

出　　处：《软件导刊》2021年第12期30-36,共7页Software Guide

基　　金：国家重点研发计划项目(2017YFB1400301);云南省人才培养项目(KKSY201401007)。

摘　　要：针对小批量多批次生产模式下的线型材料优化下料问题,提出利用长短期记忆网络(LSTM)对后续批次的下料需求进行预测的方法。按照集中下料思想,将预测得到的多个批次下料需求整合成一个较大规模的优化下料问题进行求解,并通过补偿下料对预测误差产生的缺料进行补偿以满足实际下料需求。基于LSTM和经典的列生成法构建型材优化下料模型,根据收集到的下料需求历史数据对模型进行训练和测试。仿真实验结果表明,该模型对切割零件需求预测精度较高(平均决定系数R2达到0.93),相比批次下料方法和基于库存下料方法,原材料利用率分别提升了0.14%和0.03%,原材料成本分别减少了14963.4元和14332.4元,证明了模型的有效性。Aiming at the cutting stock problem of linear material under the production mode of multiple small-lot production,inspired by the idea of centralized cutting,proposes to predict the cutting demands of subsequent batches first,and then integrate them into a large-scale optimization cutting problem to solve,accompanied by compensating the probable lack of demand resulted from prediction error by greedy method.A model is established,which uses the Long Short-Term Memory network(LSTM)to act as a predictor of cut⁃ting part demands and the classical column generation method to act as solver of the integrated cutting stock problem.After being trained with some simulated data,which is generated by Mont-Carlo simulation method based on historical data collected from an actu⁃al enterprise,this model is verified by some experiments.Simulation experiment showed that the model has high accuracy in predicting the demand of cutting parts,gives higher material utilization rate than those of the in batches cutting method and the inventory-based cutting method(0.14%and 0.03%respectively),and lower cost than those of the latter two(14963.4 yuan and 14332.4 yuan respec⁃tively),which proves the effectiveness of the proposed model.

关 键 词：线型材料优化下料 多批次下料 列生成法 需求预测 长短期记忆人工神经网络(LSTM) 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]