Stratospheric airship fixed-time trajectory planning based on reinforcement learning

Large-scale movement over a fixed time is one of the unique tasks of stratospheric airships. In practical applications, stratospheric airships often need to arrive at the designated location on time when performing tasks such as monitoring and detection. Due to the large wind resistance and low ship speed, stratospheric airships are easily affected by wind during long-distance movement. Therefore, determining how to ensure that the airship arrives at the designated location within the target time under the influence of dynamic wind fields is an urgent problem to be solved. This paper proposes an innovative solution. Based on the dueling double deep Q-network (D3QN) architecture, a trajectory planning algorithm (named FTD3) for fixed-time large-scale maneuvers was constructed. By preprocessing the wind field data and reducing the amount of input data, all information about the future wind field can be retained without introducing the instantaneous wind field. A new reward function was designed to incorporate time and distance constraints into the same dimension through time–distance mapping. Comparative experiments with other architectures showed that in the test set verification, the success rate of FTD3 reached 78.3%, compared to 47.7% for the double deep Q-network (DDQN)-based algorithm. Compared to other algorithms, FTD3 could avoid overfitting problems with the same training step size and yielded good results in uncertain wind fields. In summary, FTD3 provides an effective solution for the trajectory planning of stratospheric airships for scheduled and large-scale movement in dynamic wind fields.