基于频带特性的小波神经网络天线辐射场建模

Wavelet Network Modeling Method Based on Radiation Characteristics of Array Antenna

针对外场复杂电磁环境下,传统神经网络方法在建模时极易受噪声影响而陷入局部最优,发生过拟合等问题,提出一种采用采样理论算法的小波神经网络建模方法。通过对小波神经网络和阵列天线辐射场频带特性的分析,证明了提出的改进方法可以有效降低噪声对模型影响,防止神经网络发生过拟合。仿真将改进方法与基于正则化技术的RBF神经网络建模方法进行对比,实验结果表明,改进的建模方法具有更强的抗噪能力,并对噪声的变化具有鲁棒性。

Since in the complex electromagnetic environment, the traditional neural network is easy to fall into the local optimum and becomes over fitting that affected by the noise when modeling, a sort of wavelet network modeling method based on sampling theory is proposed in the paper. Through the analysis of the frequency characteristics of wavelet network and array antenna radiation field, it was proved that the improved method proposed can effectively re- duce the influence of noise on the model and prevent the neural network from over fitting. In the simulation, the im- proved method was compared with the method of RBF neural network that based on regularization techniques. The ex- perimental results show that the improved modeling method has stronger anti-noise ability and is more robust to noise.