基于相空间重构和卷积神经网络的混沌信号识别方法

A chaotic signal recognition method based on phase space reconstruction and convolutional neural network

针对卷积神经网络在处理一维信号时会由于网络模型参量过多导致算法收敛慢和过拟合问题,提出了一种基于相空间重构和卷积神经网络的混沌振动信号智能识别方法。首先,利用时间延迟法将一维混沌振动信号重构为二维吸引子图;然后,通过扫描转换法将其转换为标准像素图输入卷积神经网络模型;最后,借助卷积神经网络强大的图像分类能力,实现仿真和试验混沌振动信号的智能识别。结果表明:该方法能对含噪声的混沌振动信号进行有效识别,在信噪比超过10 dB时分类准确率仍可达100%,不仅具有良好的泛化性、稳定性和通用性,还消除了训练的过拟合现象,能较好地应用于工程实际中。

Aiming at the problem of slow convergence and over-fitting for convolutional neural net-work(CNN)algorithm when processing one-dimensional signals due to too many network model pa-rameters,an intelligent recognition method of chaotic vibration signal based on phase space recon-struction(PSR)and CNN was proposed.Firstly,one-dimensional chaotic vibration signals were re-constructed into two-dimensional attractors by time-delay method.Secondly,they were converted into standard pixel images by scanning transformation method and input into CNN model.Finally,the powerful image classification ability of CNN was used to realize the intelligent recognition of chaotic vibration signals in simulation and experiment.The results show that the proposed method can even effectively identify noisy chaotic vibration signals where the classification accuracy can reach 100%when the SNR exceeds 10dB.It is not only good in terms of generalization,stability and universality,but also reduces the over-fitting phenomenon of training,and can be well applied in engineering prac-tice.