To address multi-objective optimization scheduling and dynamic uncertainty issues for heterogeneous multi-agent systems in complex task environments, this paper proposes a dynamic task allocation method integrating adaptive dynamic graph neural networks (AD-GNN) with bio-inspired algorithms. First, by constructing an adaptive deep graph neural network, the network structure is dynamically adjusted based on task graph complexity to efficiently model intricate relationships between heterogeneous agents and task nodes. Then, a biomimetic optimization mechanism is introduced to simulate natural evolution and collective collaboration processes, enhancing the system’s robustness and global optimization capability under dynamic disturbances. This forms an intelligent decision-making framework with environmental adaptability. Finally, simulation results demonstrate that under dynamic conditions, the proposed method exhibits superior performance in task completion time, system energy consumption, and dynamic task coverage. It effectively addresses multi-agent task allocation challenges in dynamically uncertain environments, significantly enhancing the system’s overall capabilities in real-time decision-making, collaborative efficiency, and environmental adaptability.