Abstract: Based on continuous-time Markov decision processes,a power management policy optimization approach is proposed.First an event-driven stochastic switching model is introduced for power-managed systems.Under this model,the problem of dynamic power management is formulated as a constrained policy optimization problem.Then an efficient synthetic gradient-based policy optimization algorithm is presented. This algorithm can be employed to on-line optimization as well as off-line numerical computation.Finally,a simulation example is given to illustrate the effectiveness of the proposed approach.