Deep reinforcement learning algorithm with long short-term memory network for optimizing unmanned aerial vehicle information transmission

Abstract

The optimization of information transmission in unmanned aerial vehicles (UAVs) is essential for enhancing their operational efficiency across various applications. This issue is framed as a mixed-integer nonconvex optimization challenge, which traditional optimization algorithms and reinforcement learning (RL) methods often struggle to address effectively. In this paper, we propose a novel deep reinforcement learning algorithm that utilizes a hybrid discrete–continuous action space. To address the long-term dependency issues inherent in UAV operations, we incorporate a long short-term memory (LSTM) network. Our approach accounts for the specific flight constraints of fixed-wing UAVs and employs a continuous policy network to facilitate real-time flight path planning. A non-sparse reward function is designed to maximize data collection from internet of things (IoT) devices, thus guiding the UAV to optimize its operational efficiency. Experimental results demonstrate that the proposed algorithm yields near-optimal flight paths and significantly improves data collection capabilities, compared to conventional heuristic methods, achieving an improvement of up to 10.76%. Validation through simulations confirms the effectiveness and practicality of the proposed approach in real-world scenarios.