Abstract: A new neural network model named state delay input dynamical recurrent neural network is presented. The model with new topological structure and learning algorithm has significance for weight matrices and makes training process of weights become more distinct and straightforward. Simulation shows that the speed of learning and convergence is improved by input of the priori input-output state knowledge. The new neural network is then applied to identification of robotic dynamics through strategy of combining information obtained from simplified model with identification of unknown nonlinear dynamical information. This method reduces computing time and accelerate the identification speed. Simulation results show the efficiency of the new neural network and the strategy.