工业工程

工业工程 ›› 2024, Vol. 27 ›› Issue (5): 161-171.doi: 10.3969/j.issn.1007-7375.240218

• 工业工程应用案例 • 上一篇    

基于LSTM的供应链全生命周期碳足迹测度与预测研究

俞春华2, 佘程熙1, 李金霞2, 陈琦2, 温富国2, 吴义男1   

  1. 1. 南京大学 工程管理学院，江苏 南京 210008;
    2. 国网江苏省电力有限公司物资分公司，江苏 南京 210036
  • 收稿日期:2024-06-24 发布日期:2024-11-05
  • 通讯作者: 佘程熙(2001—)，男，湖南省人，硕士研究生，主要研究方向为可信机器学习等。Email: 522023150104@Smail.nju.edu.cn E-mail:522023150104@Smail.nju.edu.cn
  • 作者简介:俞春华(1973—)，男，江苏省人，高级经济师，主要研究方向为供应链管理等。Email: yuch@js.sgcc.com.cn

Research on Measuring and Predicting the Carbon Footprint of Supply Chain Throughout its Life Cycle Based on LSTM

YU Chunhua2, SHE Chengxi1, LI Jinxia2, CHEN Qi2, WEN Fuguo2, WU Yinan1   

  1. 1. School of Management & Engineering, Nanjing University, Nanjing 210008, China;
    2. State Grid Jiangsu Electric Power Co., Ltd. Material Branch, Nanjing 210036, China
  • Received:2024-06-24 Published:2024-11-05

PDF

7

摘要： 碳足迹测量与估计是低碳供应链评估的重要指标，目前缺乏统一的碳足迹衡量标准与界限，同时传统的碳足迹测量方法需要大量的计算成本。因此，提出一种先核算后预测的两阶段全生命周期碳足迹估算方法。在第1阶段，电网物资供应链被划分为5个阶段，并构建相应的测算模型，实现对碳足迹的定量描述与评估；在第2阶段，以电缆产品作为碳源，构建基于长短时记忆神经网络(long short-term memory neural network，LSTM)的供应链全生命周期碳排放量预测模型。基于2020 ~ 2023年电网供应链的碳足迹管理数据进行了数值实验，预测准确率为99.3%。通过与BP神经网络和GABP神经网络构建的模型对比，证明模型的准确性与优越性，实现对碳足迹的有效核算与预测。

关键词: 低碳供应链, 碳足迹测度, 碳排放量预测, 全生命周期法, LSTM神经网络

Abstract: Carbon footprint measurement and estimation are important indicators for low-carbon supply chain assessment, but currently there is a lack of unified carbon footprint measurement standards and boundaries, and traditional carbon footprint measurement methods require significant computational costs. Therefore, a two-stage full lifecycle carbon footprint estimation method that calculates first and predicts later has been proposed. In the first stage, the power grid material supply chain is divided into five stages, and corresponding calculation models are constructed to achieve quantitative description and evaluation of carbon footprint; In the second stage, a supply chain lifecycle carbon emission prediction model based on long short-term memory neural network (LSTM) was constructed using cable products as carbon sources. Finally, numerical experiments were conducted based on carbon footprint management data of the power grid supply chain from 2020 to 2023, with a prediction accuracy of 99.3%. By comparing the models constructed with BP neural network and GABP neural network, the accuracy and superiority of the models have been demonstrated, achieving effective accounting and prediction of carbon footprint.

Key words: low carbon supply chain, carbon footprint measurement, carbon emission prediction, life cycle assessment, long short-term memory neural network (LSTM)

中图分类号: