基于AOS-ELM的高铁频谱预测策略  

High-speed railway spectrum prediction strategy based on AOS-ELM

作　　者：白天晟 陈永刚[1] BAI Tiancheng;CHEN Yonggang(School of Automatic&Electrical Engineering,Lanzhou Jiaotong University,Lanzhou 730070,China)

机构地区：[1]兰州交通大学自动化与电气工程学院,甘肃兰州730070

出　　处：《铁道科学与工程学报》2020年第6期1366-1375,共10页Journal of Railway Science and Engineering

基　　金：国家自然科学基金资助项目(61763023)。

摘　　要：针对高铁无线通信环境中频谱利用率低且网络环境复杂这一现实问题,提出一种基于自适应在线极限学习机(Adaptive Online Sequence Extreme Learning Machine,AOS-ELM)的频谱状态预测模型。利用计算机产生与实际环境相符且在一定时间内满足指数分布的主用户到来时间及满足正态分布的持续时间,建立频谱状态模型。提出基于自适应神经元构造法和Cholesky分解的AOS-ELM,通过二者对模型的优化,提高模型灵活性及泛化能力,简化计算复杂度。将一维数据利用交互信息法和Cao氏计算法分别计算延迟时间和嵌入维数,构造相应样本,并送入ELM计算相对较优的初始隐层节点数,进而利用AOS-ELM进行频谱状态的预测,并与ELM和在线序列ELM(Online Sequence ELM,OS-ELM)等模型进行对比。研究结果表明:该模型可用于预知频谱状态,指导信道择优分配,提高频谱利用率。在提高预测精度的同时,显著降低了频谱预测时间,具有一定的适用性及实用性。Aiming at the practical problem of low spectrum utilization and complex network environment in high-speed rail wireless communication environment,a spectrum state prediction model based on Adaptive Online Sequence Extreme Learning Machine(AOS-ELM)was proposed.It can be used to predict the spectrum status and guide the channel preferential allocation to improve spectrum utilization.First,a spectrum state model was established by using a computer to generate a primary user arrival time that satisfies the actual environment and meets the exponential distribution within a certain period of time and a duration satisfying the normal distribution.Second,AOS-ELM based on adaptive neuron construction method and Cholesky decomposition was proposed.The model flexibility and generalization ability were improved and the computational complexity was simplified by optimizing the model.Finally,the one-dimensional data was used to calculate the delay time and the embedding dimension respectively by the adaptive method and the Cao calculation method and the corresponding samples was constructed.The corresponding initial hidden node number was sent to the ELM to calculate the relatively good initial hidden layer nodes,and then the spectrum was performed by using the AOS-ELM.The prediction of the state was compared with the ELM and the online sequence ELM(OS-ELM).The results showed that the model which has certain applicability and practicability can significantly reduce the spectrum prediction time while improving the prediction accuracy.

关 键 词：高速铁路 认知无线电 频谱预测 在线序列ELM 自适应神经元 CHOLESKY分解 

分 类 号：TN929.5[电子电信—通信与信息系统]