主管:中国科学院
主办:中国优选法统筹法与经济数学研究会
   中国科学院科技战略咨询研究院

中国管理科学 ›› 2023, Vol. 31 ›› Issue (11): 341-348.doi: 10.16381/j.cnki.issn1003-207x.2022.2692

• • 上一篇    

制造业多价值链协同数据空间产品知识属性与销量预测关联性研究

张键1,2,谢庭玉1,2,彭鹏1,王宏伟1()   

  1. 1.浙江大学伊利诺伊大学厄巴纳香槟校区联合学院, 浙江 海宁 314400
    2.浙江大学计算机科学与技术学院, 浙江 杭州 310000
  • 收稿日期:2022-12-14 修回日期:2023-06-16 出版日期:2023-11-15 发布日期:2023-12-05
  • 通讯作者: 王宏伟 E-mail:hongweiwang@zju.edu.cn
  • 基金资助:
    国家重点研发计划资助项目(2020YFB1707803)

Research on the Correlation between Product Knowledge Attributes and Sales Forecast in Multi-Value Chain Collaborative Data Space of Manufacturing Industry

Jian ZHANG1,2,Ting-yu XIE1,2,Peng PENG1,Hong-wei WANG1()   

  1. 1.Zhejiang University-University of Illinois at Urbana-Champain Institute, Haining 314400, China
    2.College of Computer Science and Technology, Zhejiang University, Hangzhou 310000, China
  • Received:2022-12-14 Revised:2023-06-16 Online:2023-11-15 Published:2023-12-05
  • Contact: Hong-wei WANG E-mail:hongweiwang@zju.edu.cn

摘要:

制造业领域多价值链协同数据空间知识引擎在本体建模的过程中,通常会出现由于产品知识不完备所导致的部门间知识管理与知识共享效率低下的问题。因而,需要系统地分析本体、属性与特征之间的关联与约束关系,提出结合产品生产、销售、供应、服务等多价值链数据协同的智能本体构建模型。制造业多价值链协同数据知识引擎本体的构建,是知识引擎应用场景的关键环节之一。针对数据空间中特征关联性挖掘的需求,本文构建了基于轻量梯度提升机、极度梯度提升与随机森林的树集成销量预测模型,并提出了以销量预测任务为中心的制造业多价值链协同数据空间的特征关联分析方法。对企业网上销售平台的销售数据进行预测实验。在树集成销量预测模型实验结果基础上,计算SHAP值,分析各项特征与预测结果之间关联性,并在消融实验中验证了SHAP值获得的特征关联性对树集成销量模型的影响。本文提出的树集成销量预测模型与SHAP值的特征关联度分析法为制造业知识引擎的建模过程的本体与属性的自动化筛选提供了可靠的理论与数据支撑。

关键词: 销量预测, 机器学习, 知识引擎, 数据空间, 多价值链

Abstract:

During the processing of ontology modeling multi-value chain collaborative data space knowledge engine in the manufacturing field, it should be aimed at the problem of low efficiency of knowledge management and sharing caused by incomplete product knowledge. It is necessary to analyze the relationship among ontology, attributes and features. This is to complete the ontology modeling method of multi-value chain collaborative data space, including production, sales, supply and service. As one of the key links in knowledge engine application scenarios, the construction of multi-value chain collaborative data knowledge engine ontology in manufacturing industry needs to fully mine the characteristic relationship in data space. The tree integration sale forecast model is constructed based on LightGBM, XGBoost and Random Forest, and a feature correlation analysis method of manufacturing multi-value chain collaborative data space is proposed. On the prediction results of the tree integration model, the correlation between the characteristics and the influence of the characteristics on the sales prediction results are analyzed through the SHAP value. The effect of feature correlation obtained by using SHAP value on sales forecast in tree ensemble model is demonstrated through ablation experiments. The tree-integrated sales forecast model proposed in this paper and the feature correlation analysis method of SHAP value provide reliable theoretical and data support for the automatic screening of ontology and attributes in the modeling process of manufacturing knowledge engine.

Key words: sales forecast, machine learning, knowledge graph engine, data space, multi-value chain

中图分类号: