动力系统混沌解的机器学习识别研究

Research on Machine Learning Recognition for Chaotic Solution of Dynamical System

摘要: 针对混沌时间序列的分类问题，本文提出了一种基于孪生神经网络的混沌时间序列分类新方法(SNN-SVM)，在孪生神经网络学习混沌特征的基础上，使用支持向量机进行识别分类，对时间序列是否为混沌做出精准判断。结果表明，我们提出的模型具有较高的泛化能力，可以准确学习一个混沌系统的特征，泛化到多个系统的混沌特征识别上，实现对混沌与非混沌时间序列的精准分类，并且无需选取特定的混沌系统。

Abstract: Aiming at the classification of chaotic time series, we proposed a new method for classification of chaotic time series based on siamese neural network (SNN-SVM), by imploying the chaotic features generated from siamese network, used Support Vector Machines to identify and classify, and make accurate judgments on whether the time series is chaotic. The results show that our proposed mod-el had high generalization ability, can accurately learn the characteristics of a chaotic system, and generalize to the chaotic feature recognition of multiple systems, without selecting a specific chaotic system. The accurated classification of chaotic and non-chaotic time series can be achieved.