混沌背景中微弱信号检测的回声状态网络方法

对复杂非线性系统的相空间重构理论进行了研究分析，提出了混沌背景中微弱信号检测的回声状态网络方法。针对回声状态网络模型参数选取困难这一问题，采用遗传算法对其模型参数进行优化。将回声状态网络模型参数作为遗传算法的个体，混沌时间序列预测均方根误差的倒数作为适应度函数，通过选择、交叉、变异等操作获得适合数据特点的最优模型参数。根据回声状态网络强大的学习和非线性处理能力，利用得到的回声状态网络模型最优参数建立混沌背景噪声的单步预测模型，将淹没在混沌背景噪声中的微弱瞬态信号和周期信号从预测误差中检测出来。以Lorenz系统和实测的海杂波数据作为混沌背景噪声进行仿真实验，仿真结果表明，本文所提方法在预测精度和训练速度方面均优于支持向量机和神经网络模型，能够有效地检测出混沌背景噪声中的微弱目标信号，且具有较小的预测误差。 

Detection of Weak Signal Embedded in Chaotic Background Using Echo State Network

In this paper，the theory of phase space reconstruction of the complicated nonlinear system is analyzed, and a new method to detect weak signals from a chaotic background using echo state network (ESN) is put forward. Genetic algorithm is used to select and optimize the parameters which are hard to be selected in echo state network. In the method, the echo state network parameters and the reciprocal of prediction root mean square error of the chaotic time series are taken as individuals and the fitness function of genetic algorithm, respectively, then the optimal parameters for different data are obtained by selection, crossover and mutation of genetic algorithm. Through the powerful ability of learning and nonlinear processing, single-step predictive model for chaotic background noise is built by optimal parameters of echo state network model, then the weak transient signal and periodic signal which is embedded in the chaotic background noise can be detected from the predictive error. It is illustrated in the experiment, which is conducted to detect weak signals from Lorenz chaotic background and Sea Clutter, the proposed method in the paper is better than the support vector machine and neural network in the training speed and predictive accuracy. And this predictive model is also highly effective to detect weak signals from a chaotic background noise as well as possess minor predictive error.