Effective Cluster Head and Routing Scheme Estimation Using Mixed Attention-Based Drift Enabled Federated Deep Reinforcement Learning in Wireless Sensor Networks

Abstract

Wireless Sensor Networks (WSN) serve as an efficient network for gathering and transmitting data in different application domains with the deployment of Internet of Things (IoT) devices. In the WSN, the selection of optimal cluster heads (CHs) and routing is an NP-hard problem. Nevertheless, the traditional routing protocols encounter challenges in handling frequent node relocations, scalability, long communication delays, energy constraints, security vulnerabilities, and dynamic network environments. Hence, this research proposes the Multi-objective Mixed Attention-based Drift applied Federated deep Reinforcement Learning (M2A-DFR) for selecting the best CH and optimal path for effective data transmission with minimum energy consumption and improved network lifetime. In addition, the M2A-DFR model offers adaptive and energy-efficient routing, taking into account message overhead, packet transfer, communication delay, and scalability. More specifically, the M2A-DFR model facilitates efficient data transmission by promptly detecting the drift occurrences and adapting the model to the dynamic changes in network behavior, thereby improving the overall network efficiency. Further, the federated learning-based global learning method periodically aggregates and updates the explorations of the local models. The simulation results reveal the effectiveness of the proposed approach in terms of energy efficiency, packet transfer ratio, latency, and scalability. Further, the proposed model outperforms other existing techniques, achieving the precision of 95.86%, sensitivity of 95.81%, and accuracy of 95.99%. In addition, the achieved mean square error value for the M2A-DFR model is reduced up to 0.827 when compared with the existing methods.