改进GAN模型在基站流量预测及5G节能中的应用

Application of GAN Model with DE-GWO Optimized LSTM for 5G Energy Consumption Control

【目的】为了更精准地预测5G基站的流量,分析潮汐现象,提出一种优化的生成对抗网络(generative adversarial network,GAN)模型流量预测方法,并将其用于实际基站的定时控制中。【方法】GAN的生成器利用差分演化灰狼算法优化长短时记忆网络(long short term memory networks,LSTM),判别器使用门控循环神经网络(gated recurrent unit,GRU)进行判别,生成器和判别器利用不断地对抗训练达到均衡从而提高了5G基站流量的预测精度;其次,利用改进人工蜂群优化k-means++算法,将其用于输出最优基站定时时间,达到最大限度节能的目的。【结果】实验结果表明,与现有模型相比,所提预测模型有更高的预测精度,定时控制功能可极大地节约能耗。

【Purposes】In order to predict the traffic of 5G base stations more accurately and analyze the tidal phenomenon,a traffic prediction method of GAN model with differential algorithm is proposed to improve the gray wolf optimized LSTM.And the modified GAN model is used in the timing control of actual base stations,which can effectively save the energy consumption.【Methods】First,since GAN is not adaptable to while LSTM is suitable for time series problems,by combining them,the GAN generator optimizes LSTM by differential evolution grey wolf algorithm.Discriminator uses GRU for discriminating,through continuous adversarial training,the generator and discriminator get equilibrium,thus improving the prediction accuracy of 5G base station traffic.Second,because of the poor global search capability of k-means++algorithm,the k-means++algorithm is optimized by using an improved artificial bee colony,and is used to output the optimal base station timing time to achieve the maximum energy saving.【Findings】The experimental results show that the proposed model has higher prediction accuracy compared with existing models,and the timing control function can greatly save energy consumption.