嵌入局部模型的SOM网络对混沌时间序列预测研究

Prediction of chaotic time series based on self-organizing map embedded local models

针对混沌时间序列特征空间多变性的特点 ,在 SOM自组织神经网络中嵌入局部线性回归模型 ,用于混沌时间序列的预测。该方法融合了局部线性预测的优点以及 SOM网络数据快速聚类能力、可视化特征识别性质和拓扑保留映射特点 ,既可减少运算时间和存储空间 ,又能适应混沌时间序列的多变特征 ,取得了较高的预测精度。

Anewself-organizingmapmodel with local linear models is presented, which is used to predict chaotic time series. The novel approach fuses linear prediction and capabilities of the SOM which is good at clustering, visualization analysis and topology preservation. The simulation shows that the approach achieves good results that reduce much computation time and large memory requirement, and also it can adapt characteristic variety of chaotic time series throughout the feature space.