In this paper, we consider an energy harvesting device-to-device (D2D) communication system, where D2D transmitter can use mode A to directly communicate with D2D receiver or use mode B as a relay to assist cellular communication while communicating with D2D receiver by adopting non-orthogonal multiple access technology. Firstly, the outage probability expression in two modes is derived, and the communication mode is determined according to outage performance. Then, assuming that the full system information is available, the channel allocation and relay selection are completed by Kuhn–Munkres algorithm, and the offline power allocation of D2D users is realized by reinforcement learning. Next, the offline optimization results are taken as the training data set to train the neural network, and the optimal model of the transmission power is obtained. Considering the transmission power constraint, the online power allocation optimization algorithm is further proposed. Numerical results demonstrate the accuracy of derived outage probability, and the proposed resource allocation algorithm can improve the performance of hybrid system.

在本文中，我们考虑一个能量收集设备到设备（D2D）通信系统，其中D2D发射机可以使用模式A与D2D接收机直接通信或使用模式B作为中继，通过采用非正交多址技术与D2D接收器进行通信时有助于蜂窝通信。首先，推导两种模式下的中断概率表达式，并根据中断性能确定通信模式。然后，假设有完整的系统信息可用，则通过Kuhn–Munkres算法完成信道分配和中继选择，并通过强化学习来实现D2D用户的离线功率分配。接下来，将离线优化结果作为训练神经网络的训练数据集，并获得最优的传输功率模型。考虑到发射功率的约束，进一步提出了在线功率分配优化算法。