A machine learning framework for dynamic and balanced computing resource allocation in 5G networks

Abstract

The swift expansion of fifth-generation (5G) networks has heightened the difficulty of distributing computing and transmission resources amidst the demands for extensive connectivity, ultra-low latency, and high throughput. This paper presents an innovative hybrid framework that combines deep learning (DL) with bird swarm optimization (BSO) to achieve dynamic and balanced resource allocation in mobile edge–cloud environments. A DL model based on long short-term memory (LSTM) forecasts user demand and channel conditions, while BSO enhances offloading and power distribution to reduce latency, energy usage, and expenses. In a setup utilizing non-orthogonal multiple access (NOMA) and mobile edge computing (MEC), the proposed DL–BSO approach demonstrates an impressive improvement of up to 54% compared to heuristic methods in simulations that reflect realistic traffic and channel conditions. The framework demonstrates a strong ability to adjust to different loads, rendering it ideal for applications that require low latency, including autonomous driving and augmented reality. The constraints involve dependence on precise forecasts and scalability issues in extensive implementations, which will be tackled in forthcoming research focused on 6G advancements.