本文提出了一种新的自递归神经网络结构.这种网络结构由全递归网络改造而成,只有一个隐层,而且隐单元仅存在自递归.研究了这种网络的学习算法.为了保证快速学习收敛,应用Lyapunov函数得到一种自适应学习率方法.用这种方法对一两层建筑结构响应进行在线预测.计算机仿真结果表明,这种网络学习算法是有效的,并且是可行的 This paper proposes a new self-recursive neural network structure. This network structure is transformed from a fully recursive network with only one hidden layer, and the hidden unit only has self recursion. The learning algorithm of this network is studied. In order to ensure fast Learning convergence, using Lyapunov function to obtain an adaptive learning rate method. This method is used to predict online response of a two-story building. Computer simulation results show that this kind of network learning algorithm is effective and feasible.