中文摘要：

针对在线应用中回声状态网络（echo state network,ESN）的储备池适应性和训练算法效率问题,文中提出一种基于扩展卡尔曼滤波（extended kalman filter,EKF）的ESN在线训练算法。该算法以ESN的储备池参数以及输出连接权矩阵为目标参数,利用EKF对其进行联合训练提高储备池适应性,并能够有效地克服交叉验证参数选择导致的ESN训练效率下降问题。Lorenz混沌时间序列以及移动通信话务量时间序列预测实验证明,新方法可显著提升ESN算法的总体计算效率。

英文摘要：

A novel on-line training algorithm base on Extended Kalman Filter（EKF） is proposed for the adaptation of reservoir and training efficiencies Echo State Network（ESN）.The output weight-matrix and parameters of reservoir are selected as objective parameters and trained simultaneously by EKF.The adaptive property of reservior is improved by simultaneously training the objective parameters,which also can overcome the poor training efficiencies resulted from cross validation for parameter selection.The simulation experiments on benchmark data set Lorenz and mobile traffic time series show that the method proposed in this paper can improve the training efficiencies of the ESN when the parameters selection of reservoir are considered.