Optimizing power allocation in contemporary IoT systems: A deep reinforcement learning approach

Abstract

The study presents an advanced optimization framework for power allocation in contemporary Internet of Things (IoT) systems, integrating multiple-input multiple-output (MIMO) technologies with non-orthogonal multiple access (NOMA). A novel deep reinforcement learning (DRL) approach is developed, incorporating an improved African Bison Optimization (IABO) algorithm to enhance system efficiency. Unlike existing methods, which primarily focus on either minimizing energy consumption or reducing information age, the proposed framework jointly optimizes both metrics, ensuring a balanced and adaptive power distribution strategy. The optimization framework leverages a DRL-driven approach to dynamically allocate power while addressing interference management in IoT networks. The IABO introduces an adaptive mechanism that refines the trade-off between exploration and exploitation, ensuring enhanced convergence and system stability. A key novelty of this work is the integration of discrete and continuous action spaces within the DRL model, allowing for efficient resource allocation in real-time scenarios. Extensive simulations validate the superiority of the proposed approach over conventional algorithms, such as genetic algorithms and random allocation methods. The results indicate a significant reduction in both energy consumption and information age, demonstrating improved transmission efficiency and overall network performance.