Meta-Learning Enhanced Multi-Agent Cooperative Optimization for Dynamic Resource Allocation in Multi-Band Networks

With the development of multi-band communication technologies, resource allocation and power control in heterogeneous networks have become increasingly critical, necessitating solutions to throughput degradation in complex environments. This paper first constructs a joint optimization model to address this challenge, transforming multi-band resource allocation into a collaborative optimization task considering THz link characteristics and multi-connectivity constraints. Rapid adaptive scheduling is achieved by embedding meta-learning into a multi-agent deep reinforcement learning (ML-MADDPG) architecture. Simulation results demonstrate significant improvements in network throughput, showcasing excellent convergence and robustness.