DEVELOPMENT OF AN ENHANCED MULTI-HOP PROTOCOL FOR OPTIMIZATION OF NETWORK EFFICIENCY IN WIRELESS SENSOR NETWORK USING REINFORCEMENT LEARNING

Abstract

The most prominent recent developments in distributed computing have utilized Wireless Sensor Networks (WSNs) for real-time data acquisition and monitoring of environmental, industrial, health, and urban environments. Despite such innovative applications, their continued use is hampered by limitations such as energy efficiency, maximization of network lifetime, and reliable multi-hop communication. A large part of existing research on WSN routing protocols is heuristic based and static, which, in turn, facilitates the wasteful use of energy in the network, quick depletion of the operating nodes of the network, and overall loss of network efficiency. This research proposes the development of a multi-hop routing protocol and network reinforcement learning WSN routing protocol. The new protocol uses Q-learning and energy-aware routing metrics to make decisions at each sensor node to create an optimal balance of energy, latency, and packet delivery ratio. The protocol was tested on the NS-3 network simulator in direct comparison to LEACH, PEGASIS, and AODV routing protocols, and demonstrated 35% network lifetime extension, 42% reduction in end-to-end delay, and 28% improvement in packet delivery ratio. Contributions of this study include the construction of a new state-space model for WSN routing, a distributed reinforcement learning framework that adapts to the resource limitations of the sensor nodes, and the development of an adaptive reward function that adjusts to the state of the network. The results confirm the validity of using AI techniques combined with standard networking protocols to tackle the complex problems of WSN deployments.