Handling method for GPS outages based on PSO-LSTM and fading adaptive Kalman filtering

To mitigate the degradation in GPS/INS integrated navigation performance during GPS signal outages, a PSO-optimized LSTM method is proposed to predict the pseudo position. The PSO algorithm is utilized to optimize two hyperparameters, neuron count and learning rate, which are essential to improve the training efficiency and prediction accuracy in the LSTM model. Considering that the predicted pseudo-position may contain outliers or accumulated errors, a robust algorithm is employed to mitigate its impact on correcting INS errors. Therefore, a Fading Adaptive Kalman Filter is introduced, which incorporates a dynamic fading factor to adaptively adjust the observation noise covariance matrix. This mitigates the impact of observation anomalies, further refining the filtering process. Experimental results demonstrate that the proposed PSO-LSTM method effectively reduces positional errors associated with inertial navigation during GPS outages and enhances the reliability of positioning. Compared to the conventional Extended Kalman Filter (EKF), the Fading adaptive EKF further improves three-dimensional positioning accuracy by up to 23.6%, 18.3%, and 22.7%, respectively.