粮食供应链跨主体联邦协同优化计算方法

Optimization of Cross-subject Federal Collaborative Computing in Grain Supply Chain

摘要:

粮食供应链在质量溯源透明化、风险预警协同过程中面临严重的信息孤岛、数据碎片化问题，导致资源浪费与决策偏差，亟须构建多主体协同的信息生态系统。本文针对粮食供应链多主体协同中的数据隐私性和异构性问题，提出一种基于联邦学习的跨主体联邦协同优化计算方法，融合联邦学习算法与深度学习框架实现供应链多环节间的协作优化。针对粮食供应链各环节数据异构特点，构建了同环节不同主体的横向联邦和不同环节不同主体的纵向联邦协同计算架构。依据供应链中不同主体数据分布特征与应用场景异构问题，提出一种多策略融合的MSGWO-FedNova联邦学习协同计算算法，通过动态调整学习率，提升模型训练速度和协同计算精度。试验结果表明，联邦学习协同计算优化算法迭代周期缩短18%，损失函数降低23%，跨规模数据准确率提升7.2个百分点，优化后的协同计算评估准确性F1分数达95.3%，为粮食供应链精准需求预测、实时风险预警及多主体协同决策提供了技术支持。

Abstract:

The food supply chain faces significant challenges such as information silos and data fragmentation during quality traceability and risk warning processes, leading to resource waste and decision-making biases. There is an urgent need to establish a multi-agent collaborative information ecosystem. Aiming to address the issues of data privacy and heterogeneity among multiple agents in the food supply chain, a cross-subject Federated collaborative optimization computing method was proposed based on Federated learning. This method integrated Federated learning algorithms with deep learning frameworks to achieve collaborative optimization across multiple stages of the supply chain. Considering the heterogeneous data characteristics at different stages, the horizontal federated learning models for the same-stage different-subjects and vertical Federated learning models for different-stage different-subjects were constructed. Based on the logistic regression algorithm, the model parameters were rapidly adjusted. Experimental results indicated that the Federated learning collaborative computing optimization algorithm shortened the iteration cycle by 18%, reduced the loss function by 23%, and improved the accuracy of cross-scale data by 7.2 percentage points. The optimized collaborative computing evaluation achieved an F1 score of 95.3%, providing technical support for precise demand forecasting, real-time risk warning, and multi-agent collaborative decision-making in the food supply chain.