Due to the inefficiency of Kalman filter-based methods for combining low-cost inertial navigation system data and global satellite navigation systems when satellite signals are outage, the use of artificial intelligence techniques in integrated architecture has become a common issue. Therefore, in this paper, while presenting an effective hybrid architecture, the generalized regression neural network is used to predict the required observations of the Kalman filter at the event of long-term outage of satellite signals. In the proposed model, for training the neural network, the velocities and positions of the inertial system are considered as inputs and also the velocities and positions of the global positioning system are considered as network outputs. This approach, while being practical and operational, has reduced computational time and increased the accuracy and speed of training and network estimation. The simulation results show that due to the simple yet robust structure of the proposed architecture and of course the selection of an efficient multi-input-multi-output neural network with the ability to detect the effective relationship between inputs and specified outputs and consequently correct errors related to speeds and situations, inertial navigation system can be used for real-time navigation, self-reliant, with high reliability and accuracy.