现有回声状态网络(Echo State Networks,ESNs)的通信话务量预测方法只考虑了历史通信话务量对预测性能的影响,较少涉及多个输入变量的通信话务量预测问题.文中首先针对ESNs用于实际多元时间序列预测任务时训练效率低,输入数据维数较多时计算复杂度大的问题,提出用改进的交替方向乘子算法(IAD-ESNs算法)训练ESNs;针对单一输入变量不能提供更加全面的预测信息,提出了改进ESNs的多变量预测模型(MP-IADMM-ESNs).以真实通信话务量数据进行仿真实验,结果表明,提出的预测模型MP-IADMM-ESNs对多变量通信话务量预测有较高的预测精度和预测效率.
Abstract
Based on the fact that the existing traffic communication prediction methods for Echo State Net-works(ESNs)only consider the impact of historical traffic on the prediction performance,and rarely in-volve the traffic prediction problem of multiple input variables.Therefore,this paper firstly proposes to use an improved alternating direction method of multipliers algorithm(I AD-ESNs algorithm)to train ESNs for solving the problems of low training efficiency and large computational complexity when ESNs are used in ac-tual multivariate time series forecasting tasks.Secondly,as single input variable cannot provide more compre-hensive prediction information,an improved multivariate prediction model for ESNs(MP-IADMM-ESNs)is proposed.Simulation experiments with real communication traffic data show that the proposed prediction mod-el MP-IADMM-ESNs has higher prediction accuracy and prediction efficiency for multiple input variables.
关键词
多元时间序列/回声状态网络/时间序列预测/交替方向乘子算法/通信话务量
Key words
multivariate time series/Echo State Networks/time series prediction/alternating direction method of multipliers/communication traffic
维普
万方数据