本文基于对多层前向神经网络学习训练获得最优权集合过程看成是非线性动态系统模型参数自组织、自学习的辨识过程，阐述了基于多层前向网络描述体系的定常和时变非线性动态系统的ＧＢＰ（广义反向传播算法）自适应递推辨识算法和模型的校验．ＧＢＰ递推算法包括在采样时间段上的纵向参数辨识过程和时序上的横向滑动辨识过程，它是现有多层网络学习算法的拓广，仿真研究表明该算法的有效性。 In this paper, based on the training of multi-layer forward neural network to obtain the optimal weight set process as a nonlinear dynamic system model parameters self-organization, self-learning identification process, described based on multi-layer forward network description system of stationary and time-varying non- GBP (Generalized Backward Propagation Algorithm) Adaptive Recursive Identification Algorithm and Model Validation for Linear Dynamic Systems. The GBP recursive algorithm includes the longitudinal parameter identification process and the time-series lateral sliding identification process in the sampling time period. It is an extension of the existing multi-layer network learning algorithm. The simulation results show the effectiveness of the algorithm.