Adaptive Sequence Importance Sampling Method for Online Wind Fields Assimilation Using Stratospheric Airship State Estimation

Stratospheric airships have attracted broad interest due to the high-altitude, long-endurance, station-keeping flight capabilities. However, accurate airspeed measurement is a major challenge due to the low cruising speeds and harsh environmental conditions, including low temperatures and pressures at these altitudes. In this study, we propose the Square-root Scaled Spherical Kalman Filter (S4F) for estimating the kinematic states and wind inflow velocity of airships. The S4F, which uses only $n+2$ sigma points, offers robust performance with reduced computational complexity. Furthermore, Sequential Importance Sampling (SIS) combined with S4F is employed to integrate wind forecasts from UNet, a neural network, and the sparse indirect measurements. An adaptive strategy is introduced to optimally fuse measurements and predictions. Simulations conducted across various locations and durations demonstrate a performance improvement of 10% to 54% over ensemble average and elite selection strategies. This method also shows potential for application in airship fleet operations.