The rapid evolution of new service models and interactive applications is being driven by the development of B5G/6G networks. To address the diverse requirements of these networks, researchers have proposed deep reinforcement learning (DRL) based solutions. However, these current solutions often struggle to effectively handle the dynamic and unpredictable nature of traffic flow in real-time network environments. In this study, we propose a novel approach that integrates meta-learning with deep reinforcement learning to improve the effectiveness and adaptability of scheduling algorithms for different traffic patterns. By introducing a latent dynamic variable as a state variable, our approach enables adaptive responses to network changes and user requirements. The experimental results demonstrate that our proposed meta-learning strategy outperforms the second-best algorithm and related joint allocation schemes by 24.5 % and 12.8 % on unseen scenarios, respectively.