基于监督机器学习的车载协作通信中继选择

Relay selection for vehicular cooperative communication based on supervised machine learning

为提高车载协作通信链路的可靠性与连通性,提出了一种在下行非正交多址接入(NOMA)的中继网络中基于监督机器学习算法的中继选择方案,通过构建反向传播(BP)神经网络预测模型,对候选中继集的中断概率进行预测,以此来进行基于最小中断概率的中继选择,提高了下行链路的连通性。在Matlab中,利用样本数据进行网络训练后,得出的预测值与理论值的相关系数为0.99944,预测的平均相对误差为0.57%,表明该中继选择方案模型能够对多个中继进行中断概率的预测,从而实现基于最小中断概率的中继选择。数值仿真结果表明,基于监督机器学习的中继选择方案相较于基于增强学习的中继选择方案中断概率下降了60%,能明显提高中断性能。

To improve the reliability and connectivity of vehicular cooperative communication links,a relay selection scheme based on supervised machine learning algorithm was proposed in the downlink Non-Orthogonal Multiple Access(NOMA)relay network.The outage probability of the candidate relay set was predicted by constructing Back Propagation(BP)neural network prediction model.In the scheme,the relay with the minimum outage probability was selected to improve the connectivity of the downlinks.In Matlab,after using the sample data for network training,the correlation coefficient is 0.99944 between the predicted value and the theoretical value,and the corresponding average relative error is 0.57%,and thus the validity of the proposed scheme was confirmed.Moreover,numerical simulation results demonstrate that the relay selection scheme based on supervised machine learning reduces the outage probability by 60%compared with the relay selection scheme based on reinforcement learning,so that its interruption performance is better.