The integrated INS/optical flow navigation method has found wide applications in many occasions.The accuracy of optical flow has direct influence on the performance of navigation parameters. When the optical flow has great error due to low illumination or other reasons, the accuracy of navigation parameters will decline greatly and thus cannot continue to navigate.To overcome this problem, a method for velocity forecasting based on Elman neural network is proposed.When the situation is suitable, the neural network model is trained online;and under special condition when the reliability of optical flow is very small, the neural network model that has already been trained is used for forecasting the velocity of the platform.In addition, the KF based on the INS dynamic error model can obtain the error vector by fusing the data of the INS with the optical flow, which is then used for correcting the navigation parameters. The trial based on the small quad-rotor air vehicle shows that, the forecasting of velocity from the neural network can closely approximate the true value in a short time, which demonstrates the correctness and effectiveness of the method. By comparing with the real value, the average error of attitude, velocity and position are 0.1%, 1%, and 2.4% respectively.