Enhanced Multiagent Reinforcement Learning Driven QoS and Power‐Efficient IIoT Relaying and Transmission Control Protocol

The reinforcement learning (RL) algorithm has emerged as a viable machine learning algorithm towards proactive routing and transmission control in wireless networks. Despite efficacy, the confined learning ability with known node's state‐action information makes classical RL algorithm like Q‐learning inferior over run‐time industrial IoT (IIoT) network. Additionally, dependency on standalone parameter‐based approximation confines their suitability towards IIoT. To alleviate it, the use of multiobjective optimization can improve decision accuracy to meet run‐time network demand. Here, a novel and robust enhanced multiagent reinforcement learning (MARL)–based QoS and energy‐efficient routing and transmission control protocol (eMARL‐IIoT) is proposed. The eMARL‐IIoT protocol is designed as a stochastically optimized postdecision state learning driven MARL that in conjunction with the multiobjective optimization (MOO) performs proactive routing and transmission control protocol to meet IIoT network demands. The eMARL‐IIoT model exploits channel state information to derive goodput, overflow, and throughput along with the stochastically derived buffer cost, which are applied for routing and transmission control. The use of MOO cost function with Lagrange relaxation assisted Markov model achieved higher packet delivery rate, lower packet loss, negligible delay and power consumption, and high resource utilization. It makes the proposed model suitable towards run‐time IIoT network.