An Optimized Resource Allocation Method Based on Deep Reinforcement Learning for the Ad Hoc Network of TACS

The ad hoc network is suggested to be used for train autonomous circumambulate system (TACS) due to its high survivability and low latency of communication. The resource allocation is critical to reducing interference within the network. In this paper, we propose a multi-dimensional resource allocation method to maximize network capacity and minimize the latency of data transmission. Models to analyze the capacity and latency of multi-path transmission are set up, which consider the location of nodes, the interference among concurrent multi-path transmission, the periodicity of communication and the waiting of data at each hop, etc. The multi-path transmission resource allocation based on deep reinforcement learning (MTRARL) algorithm is proposed to obtain a set of communication paths and the corresponding resource allocation scheme to optimize the capacity and latency of train-to-train (T2T) and train-to-ground (T2G) communication. Simulation results show that the proposed method can support the communication of 4 trains in a cluster. The minimum bandwidth and the maximum latency of communication for trains in a cluster are 17.8 Mbps and 9.1 ms, which satisfy the requirements of TACS.