Real-Time Optimization Algorithm for Unmanned Aerial Vehicle Trajectory Based on Improved Recurrent Neural Network

Aiming at the real-time problem of unmanned aerial vehicle (UAV) trajectory optimization, this paper constructs a decision system based on an improved Recurrent Neural Network (RNN). First, it compares the differences and connections between the trajectory optimization process and deep learning. Furthermore, a deep learning network is proposed to capture the nonlinear relationship between trajectory optimization parameters and the optimal trajectory, and then a real-time decision model is established. To address the gradient explosion and vanishing problems of the RNN used in the model, the Relative Support Vector Machine (SVM) and correlation coefficient are employed to determine the active data group, thereby improving network performance. Finally, the algorithm of the proposed decision system is compared with existing research results. It can be concluded that this method can reduce the time required for trajectory planning by an order of magnitude. The research presented in this paper plays an important role in enhancing the real-time performance of trajectory planning.