The integrated navigation system, which consists of the global navigation satellite system (GNSS) and inertial navigation system (INS), is widely used in various platforms. However, when the GNSS signal is unavailable, the GNSS/INS integrated navigation system will be converted into INS working alone. The error will gradually diverge. To solve this problem, this article constructs a hybrid neural network model composed of convolutional neural network (CNN) and gated recurrent unit (GRU). It combines the Pseudo-measurement information of GNSS predicted by the model with INS for integrated navigation to compensate for the interruption of GNSS and correct the error of INS. At the same time, considering that the predicted GNSS position information has a significant error, if the estimation is not accurate, the filtering accuracy of the system will decrease. Therefore, this article proposes an improved robust adaptive Kalman filter (IRAKF) algorithm to estimate the measurement noise covariance matrix for GNSS pseudo-measurement information. The actual road test results show that the addition of the CNN-GRU model and IRAKF algorithm improves the overall accuracy of the system.