Abstract:

To overcome the limitations of static allocation models, the partially observable Markov decision processes (POMDP) based single target multi-stage decision model is proposed, which reflects the uncertainty in task execution and feedback information. Then, the model of dynamic resources allocation for multi-targets is put forward. The dual decomposition is used in off-line optimization processes to decouple the problem into POMDP sub-problems. The sub-gradients algorithm is used to offer the resources price so as to coordinate the resources consumption of policies constructed by sub-problems. In real-time decision, the actions for each target are selected base on the policies and execution states so as to satisfy the constraints. Simulation results illustrate the validity of the proposed method.