混沌噪声背景下弱谐波信号的GRNN检测

针对BP(Back Propagation)神经网络方法存在训练时间长,收敛性能不理想;RBF(Radial BasisFunction)神经网络的隐层结构对鲁棒性影响大的问题,将广义回归神经网络GRNN(GeneralizationRegression Neural Network)引入混沌背景下的弱谐波信号检测中,提出了一种提取混沌噪声背景下微弱谐波信号的GRNN检测方法.该方法利用GRNN建立噪声混沌背景的最优一步预测模型,再结合频域处理预测误差提取微弱信号,以Duffing系统产生混沌时序作为混沌背景,使用该方法用MATLAB6.1验证在没有噪声、存在高斯白噪声和存在色噪声情况下的混沌背景下的弱谐波信号检测.实验结果表明,谐波对混沌的信噪比达到-36dB时仍然可以检测出谐波.

Generalization regression neural network method for detecting weak harmonic signal under background of noisy chaos

Aiming at the problems that BP(Back Propagation) network which needs long training time and weak astringency, and that the robustness of RBF(Radial Basis Function) neural network is susceptible to the structure of hidden layer, the GRNN(Generalization Regression Neural Network) method is presented to detect weak harmonic signal under the noisy chaotic background. In this method, GRNN is used to build one-step predictive model for chaos background first, and then the frequency way is combined to process predictive error for detection of the weak harmonic signal. The chaotic series of Duffing equation is utilized as chaotic background to experiment this method under the noiseless background, white noise and colored noise through MATLAB6\^1. The results show that this method can detect weak harmonic signal at SNR(Signal-to-Noise Radio)of harmonic to chaos background up to -36 dB.