On multi-scale event-based optimization

For most practical discrete event dynamic systems (DEDSs), the system structures are hierarchical and contain multiple subsystems, which leads to the fact that state transitions within different scales are differentiated. Directly applying standard Markov decision process (MDP) or event-based optimization (EBO) methods to model and optimize such problems is inefficient and often encounter issues like the curse of dimensionality. In this paper, a multi-scale event-driven optimization (MEBO) method is proposed. To accurately capture system state transitions at different scales, MEBO defines macro and micro events, respectively. With proper definition, the size of event space can be constant or grows only linearly with the problem size. Furthermore, two policy iteration algorithms are proposed to solve MEBO problems and the local optimality of the algorithms is proved theoretically. an example based on queueing system. Finally, the effectiveness of the MEBO methods are demonstrated by an example of admission control in queuing system. It is hoped that this method can be used as a supplement to EBO method and provide a new perspective for performance optimization of multi-scale DEDSs