A Simulation-Driven Reliability-Aware Hybrid Deep Learning Framework for Energy-Efficient Routing in Dynamic IoT Wireless Sensor Networks
Keywords:
wireless sensor networks, IOT, energy-aware routing, CNN-BiLSTM, attention; uncertainty estimation, PPO, NS-3, network lifetimeAbstract
Energy-constrained Internet of Things (IoT) wireless sensor networks (WSNs) require routing strategies that balance battery consumption, link reliability, and communication delay under changing network conditions. This study proposes RALiWSN (Reliability-Aware Learning for Intelligent Wireless Sensor Networks), a simulation-driven framework combining CNN-BiLSTM-attention-based network-state prediction with uncertainty-aware Proximal Policy Optimization (PPO). The CNN extracts local relationships among network features, BiLSTM models temporal dependencies, and attention weights informative observations. An ensemble-based uncertainty estimate controls the fusion of predicted and currently observed states before PPO selects a feasible next hop. The proposed evaluation plan uses NS-3-generated traces and compares conventional routing, energy-aware routing, PPO using current observations, and the complete RALiWSN approach. Publicly available WSN datasets are identified for complementary reference and external validation where feature definitions are compatible. To illustrate the reporting procedure, this manuscript includes a reproducible, seeded synthetic pilot dataset and synthetic comparative plots. These pilot values are explicitly illustrative and are not NS-3 measurements. Final claims require packet-level simulation and independent-seed statistical validation.