Digital Twin-Enabled Explainable Multi-Agent Deep Reinforcement Learning for Adaptive Cross-Layer Hybrid IoT Communication Networks

Abstract

The explosive growth of heterogeneous Internet of Things (IoT) applications has posed great challenges in realizing energy-efficient, reliable, and adaptive communication in highly dynamic network environments. Traditional communication optimization techniques usually suffer from static decision making, poor scalability and lack of joint optimization of multiple communication parameters over network layers. To deal with these challenges, this paper presents a Digital Twin-Enabled Explainable Multi-Agent Deep Reinforcement Learning (DT-XMADRL) framework for adaptive cross-layer hybrid IoT communication networks. Specifically, the proposed framework integrates a real-time Digital Twin that continuously reflects the physical communication environment to: (i) accurately monitor the network states; and (ii) safely evaluate the communication strategies before their actual implementation. An Explainable Multi-Agent Deep Reinforcement Learning model is proposed to cooperatively optimize routing, transmission power, spectrum allocation, interface selection, and traffic scheduling over heterogeneous communication technologies such as Wi-Fi, LoRa, ZigBee, BLE, and 5G. In addition, an explainable artificial intelligence (XAI) module offers interpretable decision-making by pinpointing the key factors driving communication policies, thus promoting transparency and trust. A multi-objective reward function is formulated to maximize the energy efficiency, throughput, reliability, and network lifetime, and minimize the latency and communication overhead. Extensive simulations under different IoT scenarios show that the proposed DT-XMADRL framework outperforms the traditional reinforcement learning and optimization methods in terms of communication reliability, energy consumption, packet delivery ratio, throughput and latency. The proposed framework offers a scalable, intelligent, and explainable communication architecture suitable for next generation smart cities, industrial IoT, healthcare, precision agriculture, and intelligent transportation systems.