Abstract: The GNSS/INS integrated navigation system can provide long-term, high-precision navigation information for mobile carriers. However, in adverse environments where filter measurement vectors cannot be obtained, it leads to rapid divergence in navigation positioning results. To address this issue, an increasing number of researchers are employing artificial neural networks to directly fuse information in the integrated navigation system. However, the inherent characteristics of the inertial navigation system (INS) result in errors in previously trained network models, and inertial navigation errors continue to accumulate during interruption periods. Therefore, an intelligent positioning algorithm for GNSS interruptions is proposed. This algorithm utilizes backpropagation (BP) neural networks to train filter measurement vectors and then updates the Kalman filter (KF) by incorporating diagonal-loaded reconstructed measurement noise covariance matrices. This approach reduces the impact of neural network training errors on the integrated navigation algorithm, enabling the navigation system to maintain relatively reliable navigation performance even during prolonged GNSS signal interruptions.