LLMGA: A Large Language Model-Guided Genetic Algorithm for Dynamic Parameter Control in Multi-UAV Target Assignment

The multi-UAV target assignment problem holds significant importance across various real-world applications, with its core focus on efficiently assigning multiple unmanned aerial vehicles (UAVs) to target areas in order to optimize task performance and minimize costs. Traditional optimization methods often rely on fixed or manually tuned parameters, requiring extensive repeated experiments to determine suitable parameter combinations, which results in low efficiency and flexibility. To address these limitations, we propose a novel Large Language Model-Driven Genetic Algorithm (LLMGA) framework, which leverages the capabilities of LLMs to dynamically adjust crossover and mutation rates. LLMGA introduces a new meta-prompt mechanism to extract population-level information and employs a discrete output strategy to generate adaptive parameters. Experimental results on multi-UAV target assignment tasks demonstrate that LLMGA outperforms traditional approaches in both solution quality and robustness. Ablation studies further validate the significance of the framework's core components. This research highlights the potential of integrating LLMs with adaptive optimization techniques for solving complex real-world problems.