Reinforcement learning method for DEDS supervision

In this paper, based on a new way of specifying the system close-loop behavior, we propose a reinforcement learning method for discrete event dynamic system (DEDS) supervision. By means of the concept of subconnection neural networks we develop a new reinforcement learning structure which is adaptive to DEDS supervision, present the close relationship between reinforcement learning and the neural network, and build a foundation for the further development of our reinforcement learning based on neural network theory. Using two examples about the optimization and control of telecommunication networks, we have illustrated the application prospect of our method. Computer simulations have confirmed its effectiveness