Adaptive Resource Management Framework for Profit Optimisation in 5G Network Slicing

The fifth‐generation (5G) network slicing paradigm promises a customized service delivery through virtualized, isolated network slices. However, its full potential is hindered by inefficient and static resource allocation strategies that often fail to adapt to dynamic traffic and network conditions. This paper proposes a novel two‐phase optimization framework to address this challenge. First, an Integer Linear Programming (ILP) model is developed to prioritize high‐revenue slice admission, revoke underutilized slices, and reallocate resources for profitability. Simulations using real‐world traffic data demonstrate that this proposed approach outperforms static and reactive approaches, achieving up to 24.8% higher resource utilisation and 98.99% higher profitability than the baseline method. The framework also adapts to dynamic traffic patterns and network conditions, balancing profit maximisation with reconfiguration costs. Second, to further improve performance, the paper introduces a deep reconfiguration agent (DRA), a Deep Reinforcement Learning (DRL) model that learns policies for slice admission, resource allocation and reconfiguration, and predicts network slice resource demands and consumption, enabling adaptive reconfiguration based on future demands and long‐term profit. The results show that the DRA‐based strategy increases the InP's profit by up to 5 times and boosts resource utilisation by 43.71% compared to the ILP model alone and also converges by 38.89% faster compared to using only the DRL model.