一种改进的复数BP神经网络算法研究

An Improved Complex-Valued BP Neural Network Algorithm

论文提出了复数BP神经网络的一种新结构和算法。算法的主要思想是将复值输入信号的实部和虚部分离,分别训练,使其达到稳定状态。其结构简单,易于实现,只需少量样本点却有很高收敛速度和精度。通过实验和仿真说明论文算法的有效性。

This paper presents a new structure and algorithm of complex-valued BP neural networks. The main idea of the algorithm is that the input signal is separated as the real part and imaginary part, which are trained respectively to attain the stable state. This structure is simple and easy to realize. It needs only a small amount of samples, but it has high convergence speed and accuracy. The experiment and simulation show the algorithm is effective.