Dynamic allocation and optimization strategy of communication network resources driven by reinforcement learning

Modern communication networks are becoming increasingly complex as they accommodate rising traffic from mobile users, expanding IoT ecosystems, and delay-sensitive services. These requirements pose significant challenges for traditional static and heuristic-based allocation methods, which often fail to adjust effectively to rapid changes in traffic patterns, channel conditions, and heterogeneous network infrastructures. To overcome these limitations, this study introduces a reinforcement learning (RL) driven dynamic resource allocation framework. The framework integrates an adaptive sailfish optimizer with the deep deterministic policy gradient algorithm (ASF-DDPG), enabling the agent to learn resource allocation policies in real time, including routing, channel assignment, bandwidth distribution, and power control. Network behavior is modelled as a Markov decision process (MDP) to support long-term reward optimization, enhancing throughput while reducing latency, energy consumption, and the dismissal rate. The design also supports distributed scenarios such as heterogeneous radio access technologies (RAT), edge intelligence, and network slicing, which are crucial in emerging 6G architectures. Extensive Python-based simulation experiments show that ASF-DDPG achieves higher training accuracy, improved convergence behavior, lower latency and power usage, and a reduced dismissing rate compared with contemporary reinforcement learning-based approaches. These performance improvements reflect the model’s ability to balance exploration and exploitation, address non-stationary network conditions, and adapt to large-scale environments without stability loss. Overall, the findings demonstrate that combining RL with adaptive meta-heuristic optimization offers a promising direction for scalable, resilient, and energy-efficient network resource management in future wireless systems.