电力物联网中基于深度学习的无蜂窝接入点选择算法

Deep Learning based Cell-free Access Point Selection Algorithm in Power Internet of Things

针对传统电力物联网中的设备间干扰和高能耗问题,提出一种基于深度神经网络(deep neural network,DNN)的无蜂窝接入点(access point,AP)选择算法。首先,考虑无蜂窝网络下,将中央处理单元连接的多个协作AP随机分布在大规模电力设备周围,从而拉近AP与设备间的距离以降低能耗;然后,基于深度神经网络算法对AP进行有效筛选,通过训练静态随机设备的大尺度衰落系数将AP组合进行标签分类,使得平均频谱效率(spectrum effectiveness,SE)达到最大;最后,利用测试集进行测验,仿真结果表明,算法准确率能够达到95%,且相比于传统全AP传输和基于信道特征排序的算法,所提出的算法能够进一步提升无蜂窝网络多设备的平均SE,并具有更低能耗。

To cope with the problem of inter-device interference and high energy consumption in traditional power Internet of things,a deep neural network(DNN)based access point(AP)selection algorithm was proposed.Firstly,it is considered that the AP without cellular was densely distributed around power equipment,so as to shorten the distance among the AP and equipments to reduce energy consumption.Secondly,based on the DNN the AP combination was effectively screened and labeled by training the large-scale fading coefficient of static random users to maximize the average spectral efficiency(SE).Finally,the test set was used to test.Simulation results show that the proposed algorithm can further improve the average spectral efficiency of the non-cellular network multi-device and reduce energy consumption compared with the traditional all-AP transmission and channel feature-based sorting algorithm.