Online bipartite matching methodology for resources with major epidemics: adaptive time window based on reinforcement learning

A time-window-based multi-stage online dynamic bipartite matching model is formulated, which can be transformed into a Markov decision process. The paper proposes adaptive time window batch bipartite matching algorithm based on reinforcement learning with nearest neighbor first heuristic strategy to allocate anti-epidemic resources.

The findings indicate that while the matching rate continues to rise, the average waiting time exhibits a tendency of first reducing and then increasing as the matching time window gets longer. This implies that a manager should adjust the matching time window in accordance with the epidemic's development tendency and the availability of resources. It demonstrates that keep exploring the matching time window at a high level with taking the acceptable waiting time into account.