基于尺度UKF小波网络的混沌时间序列预测  

Prediction of Chaotic Time Series based on Scaled UKF Wavelet Neural Network

作　　者：赵鲲鹏[1] 丛伟[2] 赵亮[2] 

机构地区：[1]西京学院,西安710123 [2]空军工程大学工程学院,西安710038

出　　处：《火力与指挥控制》2010年第5期114-116,120,共4页Fire Control & Command Control

摘　　要：根据小波网络和Kalman滤波的特点,将小波网络的权重、平移因子和伸缩因子作为Kalman滤波的状态变量进行估计,提出一种训练速度快、泛化能力强的小波网络的尺度无轨迹Kalman滤波学习算法。然后,利用Rǒssler时间序列和变参数的Ikeda时间序列对该算法进行了验证。仿真结果表明,该算法预测混沌时间序列收敛速度快,预测精度高,而且在混沌时间序列的嵌入维数未知时也能取得较好的预测效果。可以胜任混沌时间序列的精确建模和预测任务。According to the characteristics of wavelet neural network（WNN） and Kalman filter（KF）,proposed an scaled unscented KF learning algorithm for WNN with the merits of fast training speed and good generalization capability which regards weight,translation factor and dilation factor as state variable of KF to estimate.Then,the chaotic time series （CTS） of Hénon,Rǒssler and Ikeda map whose parameter values change with time are applied to validate this algorithm,respectively.The simulation indicate that the proposed algorithm can predict CTS accurately with fast convergence speed even if the embedding dimension is unknown,and is feasible for accurate modeling and prediction of CTS.

关 键 词：小波 神经网络 卡尔曼滤波 混沌时间序列 

分 类 号：O545[理学—物理]