基于复合神经网络的数据分布模式仿真识别

采用复合结构神经网络对数据概率分布模式进行了仿真识别。首先提取数据的样本峰度、偏度、分位数和累积概率等数字特征 ,构建数据概率分布模式和仿真识别训练样本 ,然后采用Kohonen网络对概率分布进行聚类 ,接着运用BP网络对各聚类分别进行分类识别 ,最后对任意数据流进行了仿真识别。仿真识别结果表明 ,同其它模式识别方法相比 ,该方法更能够反映数据流概率分布的特征 ,从而得到了满意的模式识别效果。

Data Distribution Simulation Identification Using Combined Structure Neural Networks

This paper studies the data distributions using combined structure neural networks. First, the paper sets up data distribution modes and simulation identification training set based on statistical data such as kurtosis, skewness, quantile and cumulative probability. And then, it clusters distribution functions using Kohonen self organizing map. Furthermore, it classifies the clusterred modes respectively using BP neural networks. Finally, it identifies lots of random data series through computer simulation. The result indicates that the proposed approach is more successful for selecting input data distributions compared with other methods.