Optimization Research on UAV Semantic Communication System Based on SVD-MADRL

The study optimizes the flight trajectory and power of multiple unmanned aerial vehicles with the deep deterministic policy gradient algorithm, and constructs a multi-unmanned aerial vehicle semantic communication optimization model based on singular value decomposition and multi-agent deep reinforcement learning. The results show that the relay system developed in this study is better than traditional algorithms in terms of coverage and flight path, with a coverage rate of up to 90%. Moreover, the energy consumption of the model to complete task transmission is only 2835 joules, and the delay time is only 15 seconds. In addition, compared with traditional algorithms, the semantic communication optimization model constructed in this study performs the best in terms of total reward, data collection efficiency, and accuracy. It has strong stability and convergence ability, with an accuracy close to 1, significantly improving the accuracy and efficiency of unmanned aerial vehicle communication transmission. The effectiveness of the two models designed for research is relatively high, both superior to traditional methods, providing a theoretical basis for optimizing the overall performance of unmanned aerial vehicle communication. It is of great significance in promoting the widespread application of unmanned aerial vehicle technology.