针对回声状态网络(ESN)算法在训练中只能调节输出权值的缺陷,提出了动态回声状态网络(DESN)研究方法。为了提高Sigmoid函数的非线性映射能力,该算法将映射区间因子引入到激活函数中,不仅对激活函数的位置和形状上进行了调整,而且对映射区间范围也作出调整。通过基准数据集仿真实验,验证了提出算法的有效性和可行性;并利用其预测地空导弹生存能力,获得了满意的结果。

In this paper, a new learning algorithm called dynamic echo state network(DESN) is proposed to solve the problem that ESN algorithm adjusts only output weight in the process of training. In order to improve nonlinear mapping capability of Sigmoid activation function, the mapping range factor is introduced into activation function, it not only adjusts position and figure of activation function, but also adjusts mapping range. Simulation results verified the effectiveness and feasibility of this method by using some real-world datasets. And it is applied to develop a prediction model for the surface-to-air missile survivability, and the result was satisfied.