基于信息几何的统计回馈神经网络非线性自适应预测控制
Recurrent Neural Network Prediction Controller Based on the Information Geometry
-
摘要: 研究统计回馈神经网络(SRNN)的非线性自适应预测控制.基于混合统计模型,利 用信息几何的处理方法,将SRNN的参数估计转化为一般的线性ARMA系统的最小均方误 差参数估计算法,最终获得SRNN参数估计.获得RNN预测的参数估计以后,可以十分方便 地利用线性ARMA系统的控制规律来设计SRNN的预测控制规律,解决了非线性SRNN预 测参数估计、复杂非线性系统控制规律设计等问题.在研究单隐元SRNN的基础上,进一步 探讨了多隐元SRNN的自适应预测控制问题.Abstract: The paper primarily investigates the adaptive predication controller modeled by the statistic; recurrent neural network(SRNN). Using a mixed statistic model to approximate single neuron of the SRNN, one can easily obtain both parameter estimation and control rule of the adaptive controller based on the procedure of information geometry. On the basis of adaptive controller of single hidden unit, the paper further investigates the controller problems of the complicated SRNN with multiple hidden units.
-
点击查看大图
计量
- 文章访问数: 2098
- HTML全文浏览量: 156
- PDF下载量: 861
- 被引次数: 0
下载:
