摘要：This research is conducted as a contribution to the design of future resilient global supply, subject to disruptions. In a collaborative network of enterprises (CNEs), companies laterally collaborate through information and resource sharing. A common practice of collaboration is through demand and capacity sharing (DCS), which aims to enable timely delivery to customers, despite uncertainties in the market and disruptions in available supply capacity. One key challenge of DCS implementation is to ensure that capacity-sharing proposals from one enterprise to another are generated efficiently. A slight delay in DCS protocols could lead to collaboration failures and backorder fees, prevalent if companies initiate the protocols when actual demand has been presented. This study aims to develop a new preemptive DCS, following CCT, Collaborative Control Theory protocol for DCS, in which machine learning is utilized to learn from historical market data and invite capacity-sharing proposals before actual demand occurs. Long Short-Term Memory (LSTM) Neural Network Autoencoders are used to detect anomalies in real-time. The detected anomalies become enablers of the capacity sharing proposal, which could initiate DCS preemptively. A case study of automobile manufacturers in Indonesia is used to validate the model. On average, for the given case study, preemptive protocols based on the learning protocols are capable of increasing the fulfillment rate to 91.03% in the demand-capacity sharing CNEs; better compared to the non-collaborative and non-preemptive demand-capacity sharing protocol (DCSP) baseline scenarios. Further research is being conducted for refinements.
关键词：collaborative control theory;collaborative enterprises network;demand;capacity sharing;learning protocols;preemptive protocols
