Intelligent Transmission Scheduling Based on Deep Reinforcement Learning

Abstract

With the increasing diversification of ship users’ communication services, the QoS of data transmission has become the limitation of the development of maritime communication. The software-defined maritime communication networks are proposed to solve the problem of communication mode obstacles in heterogeneous networks. Based on this framework, we propose a transmission scheduling scheme based on improved deep Q learning algorithm which combines the deep Q network with softmax classifier (also known as S-DQN algorithm) to improve throughput, balance delay and energy consumption. First of all, the Markov decision process (MDP) is used to realize the optimal scheduling strategy. In addition, the mapping relationship between the optimal policy and the obtained information is established by using the deep Q network in the system. When the input data arrives, after the amounts of data self-learning, the optimal strategy is made as quickly and accurately as possible. The simulation results show that the scheme is better than other traditional schemes under the different quality of service, which verifies the effectiveness of the scheme.