Intelligent Maritime Communications Enabled By Deep Reinforcement Learning

Nowadays, with maritime services grows exponentially, quality of service (QoS) for data transmission has become a bottleneck that restricts the development of maritime communication. In order to solve this problem, firstly, a software-defined maritime communication framework is proposed to overcome the difficulty of communication mode in heterogeneous environment. In addition, we propose a novel data transmission scheme with the enhanced deep Q-learning algorithm in this framework, which combines depth Q-network with softmax multiple classifier, also known as S-DQN algorithm. This scheme also mentions what the purpose of the optimization is (i.e., throughput, cost, energy). In our system, markov decision processes (MDPs) are used to implement the optimal strategy for network resource scheduling. The system employs the deep Q-network to establish the mapping relationship between the acquired information and the optimal strategy, and when the input data arrives, the optimal strategy can be as fast and accurate as possible due to a large amount of data self-learning. The simulation results show that the scheme is superior to other traditional schemes in different QoS, and the effectiveness of the proposed scheme is verified.