AI-Driven Energy-Efficient Routing and Lifetime Optimization in Large-Scale Wireless Sensor Networks

Abstract Large-scale wireless sensor networks (WSNs) are increasingly deployed in critical applications such as environmental monitoring, smart agriculture, industrial automation, and smart city infrastructures. However, the severe energy constraints of sensor nodes and the need for long- term, reliable data delivery pose significant challenges to network design. Traditional energy- efficient routing protocols rely on static heuristics and localized decision- making, which limits their adaptability to dynamicnetworkconditions, heterogeneous node states, and varying traffic patterns. To address these challenges, this paper proposes an AI-driven routing framework that jointly optimizes energy efficiency and network lifetime in large-scale WSNs. The proposed approach models the WSN as a dynamic graph and integrates graph neural network (GNN)–based state representations with a deep reinforcement learning (DRL) routing agent. Node-level features such as residual energy, queue length, link quality, and hop distance to the sink are encoded into compact embeddings using a lightweight GraphSAGE architecture. Based on these embeddings, a Double Deep Q-Network (Double-DQN) with a dueling architecture selects energy- aware next-hop routing decisions. To ensurescalabilityandreducecommunication overhead in large deployments, the routing policy is trained using a federated learning strategy that aggregates local model updates without requiring raw data exchange. Extensive packet-level simulations are conducted under diverse scenarios involving varying network sizes, node densities, and traffic patterns. The proposed framework is evaluated against widely used routing protocols, including LEACH, PEGASIS, HEED, and energy- aware shortest-path routing. Simulation results demonstrate that the proposed method significantly improves first-node- death time and overall network lifetime, while achieving better energy balance, higher packet delivery ratio, and lower end-to-end delay. These results confirm the robustness and scalability of the proposed framework, making it a promising solution for long- term and large-scale WSN deployments.