Collective Risk Learning for Supply Chain Disruption Preparedness

各种政府调查强调了对更高效、更有弹性的供应链的需求，并因英国脱欧和新冠肺炎等近期世界事件而被放大。随着企业将生产外包给彼此，它们创造了规模经济并降低了价格，但如果供应链的任何一个成员被破坏，也会增加破坏级联的风险。通常，在预测延误、中断和决定安全清单时，组织会单独行动，而不是集体行动。然而，单个组织可以收集和分析的中断数据很小，不平衡，并且完全偏向于自己的观点。当不确定性增加时，这种个人主义的方法会导致股票通胀和缺货之间的混乱振荡。许多研究证明，增加数据共享和集体决策将增加弹性，但这并不可信，因为供应链成员担心产能和过剩库存等信息可能被客户“推断”出来，并被投机地用于降低成本。有两个关键的紧急方法可以帮助改变这种状况。首先是开发低成本平台，促进中小企业及其买家的数据共享，我们将在本项目中使用这些平台，使中小企业能够获得集体学习。二是人工智能技术的出现。在这个项目中，将开发集体学习（CL）方法，这将使组织代理能够协作开发共享的预测模型。在这里，如果一个组织能够预测中断，它的知识可以共享，防止其他组织缺货。由于该方法可以自动化，因此可以避免手动编排的成本。CORES方法将被集成到低成本的数据集成平台中，并在航空航天领域进行试验。

The need for more efficient, resilient, supply chains has been highlighted by various government inquiries and amplified by recent world events including Brexit and Covid-19.As organisations outsource production to one another they create economies-of-scale and reduce prices but also increase risk of disruption cascades if any member of the chain is disrupted. Typically, organisations act alone, rather than collectively, when predicting delays, disruptions and deciding on safety inventories. However, disruption data an individual organization can collect and analyse is small, imbalanced, and partial entirely to its own view. When uncertainties increase, this individualistic approach results in chaotic oscillations between stock inflation and stock-outs. Numerous studies proved that increased data sharing and collective decision making would increase resilience, but this has not been plausible as members of the chain fear that information such as capacity and excess stock can be "inferred" by clients, and used opportunistically for cost reduction. Two key emergent approaches can help change this state of affairs. First is the development of low cost platforms that facilitate data sharing for SMEs and their buyers, which we will use in this project to enable SME access to collective learning. The second is the emergence of AI technology. In this project Collective learning (CL) approaches will be developed, which will enable organizational agents to collaboratively develop a shared prediction model. Here, if one organization is able to predict a disruption, its knowledge can be shared, preventing others from stock outs. As the approach can be automated, costs of manual orchestration are avoided. CORES approaches will be integrated into low cost data integration platforms and trialled within the Aerospace sector.