基于神经网络的特征结构并行提取算法研究

为并行实时地提取数据协方差矩阵的信号特征结构,从特征结构并行提取的约束优化问题表示入手,利用梯度方法和迭代法构建了可实时并行提取信号特征向量矩阵的直接神经网络求解法和基于能量函数的神经网络求解算法,并形成了相关迭代学习算法.理论分析表明当数据样本足够大时,算法的迭代结果就是数据协方差矩阵信号特征结构的一个良好估计,同时计算机仿真亦验证了算法的有效性.另外,仿真试验亦表明可以通过调节加权矩阵D的对角元来控制算法的收敛速度.

An Algorithm for Eigenstructure Parallel Extraction Based on Neural Networks

To extract the signal eigenstructure parallelly on line of the data covariance matrix,the expression of the eigenstructure extraction with limitation is discussed firstly,then two kinds of algorithms are constructed based upon neural networks for eigenvectors matrix extraction with gradient method and iteration method.They are direct neural network algorithm and the neural network algorithm based upon energy function.Theoretical analysis shows that the iteration solution of the algorithms presented in the p...