Main Article Content

Hybrid multi-agent Q-learning and intelligent water drops framework for adaptive campus bandwidth allocation


O. S. Adeoye
M. B. Abdullahi
O. S. Adebayo
O. A. Ojerinde
M. Danlami

Abstract

Efficient bandwidth allocation in campus networks remains challenging because of dynamic traffic patterns, heterogeneous user demands, congestion, and fairness constraints. This study aims to develop and evaluate a hybrid Intelligent Water Drops Multi-Agent Q-Learning (IWDAMA) framework for adaptive bandwidth allocation in software-defined campus networks. The proposed framework integrates Multi-Agent Q-Learning with the Intelligent Water Drops algorithm to enhance exploration, convergence, and adaptive scheduling. Performance evaluation was conducted using MATLAB simulations in a software-defined campus network under five traffic scenarios involving 200–2000 users with a fixed link capacity of 2000 Mbps. The proposed model was compared with standalone Multi-Agent Q-Learning (MAQL), Q-learning, and First-Come-First-Served (FCFS) scheduling using throughput, delay, bandwidth utilization, fairness index, and packet loss as evaluation metrics. Experimental results showed that IWDAMA achieved throughput ranging from 392 Mbps to 1764 Mbps, bandwidth utilization of 0.99, fairness index of 0.995, packet loss between 0.0159 and 0.0600, and delay between 0.0636 ms and 0.2019 ms, consistently outperforming MAQL, Q-learning, and FCFS across all scenarios. The findings demonstrate that integrating heuristic optimization with cooperative reinforcement learning provides a scalable and robust solution for intelligent bandwidth management in campus and enterprise software-defined networks.


Journal Identifiers


eISSN: 2437-2110
print ISSN: 0189-9546