Abstract: To realize the joint optimization of cross-enterprise supply chains with carbon trading,a multi-network bi-directional cooperative integer programming model and a situational transfer learning algorithm were proposed.Aiming at the cooperative decision-making between independent supply chains of different enterprises,the proposed model employed cooperative cost,partner sharing,collaborative transportation and horizontal logistics to establish objective functions on economic and environmental bottom-lines,which were quantified and combined the objectives through quota-based carbon trading.The proposed algorithm integrated the Markov decision process with an information accumulation mechanism,which could effectively accumulate solution experience in different situations and improve the solving efficiency and accuracy.The experimental results based on real data showed that the solving speed of the proposed method was 92.68 times faster than that of the CPLEX solver.The transfer learning method was proven to be superior to the independent learning in both efficiency and accuracy under different scale instances.Moreover,the proposed algorithm showed good compatibility and robustness,and could process large-scale instances effectively.

Key words: supply chain, multi network cooperation, carbon trading, situational transfer learning, network optimization

摘要： 为解决碳交易背景下跨企业供应链网络的联合优化问题,提出一种碳交易供应链多网双向合作整数规划模型以及一种情景迁移学习算法。前者针对跨企业供应链网络间的协同决策,引入合作成本、伙伴共享、协同运输及横向物流,建立针对经济侧和环境侧的目标函数,并通过基于配额的碳交易手段将二者进行量化整合。后者融合马尔可夫决策过程和信息积累机制,可有效学习不同情景下的求解经验,实现了模型的高效率、高精度可迁移求解。基于真实企业数据的实验结果表明,所得方法的求解速度优于CPLEX求解器92.68倍。迁移学习模式在不同规模算例下的效率和精度上均优于独立学习模式,且具有良好的兼容性和鲁棒性,能够有效处理大规模算例。

关键词: 供应链, 多网合作, 碳交易, 情景迁移学习, 网络优化