Adaptive Hybrid Routing for Wireless Mesh Networks in Smart Grid: An ML-Driven Framework Integrating RPL and GPSR
TL;DR
An Adaptive Hybrid Routing Framework that integrates RPL and GPSR under a machine learning (ML)-driven decision engine that offers a resilient and energy-efficient routing solution for next-generation smart grid neighborhood area networks is proposed.
Abstract
The integration of communication networks into smart grids introduces stringent requirements for reliability, low latency, scalability, and energy efficiency. Existing routing protocols — the Routing Protocol for Low-Power and Lossy Networks (RPL) and Greedy Perimeter Stateless Routing (GPSR) exhibit complementary strengths and weaknesses across varying network conditions. This paper proposes an Adaptive Hybrid Routing Framework (AHRF) that integrates RPL and GPSR under a machine learning (ML)-driven decision engine. The system dynamically selects the most suitable protocol based on real-time network features including link quality, node degree, residual energy, queue occupancy, and traffic load. We present rigorous mathematical models of both protocols, formulate a composite utility function capturing trade-offs among packet delivery ratio (PDR), end-to-end delay, throughput, and energy consumption, and integrate a Random Forest classifier for adaptive protocol selection. The framework is validated through a custom discrete-event packet-level simulator implementing log-distance path loss with shadowing over a 500×500 m wireless mesh with up to 200 randomly deployed nodes across four operational scenarios. Results demonstrate that the proposed Hybrid-ML framework achieves PDR improvements of up to 8.6% over standalone RPL in the density scenario and up to 110.7% over GPSR under node failure conditions, while achieving 27.4% lower energy consumption per packet than GPSR in dense deployments. The Random Forest classifier achieves 95.6% cross-validation accuracy. Feature importance analysis reveals that average SNR (29.9%), SNR standard deviation (20.8%), and path diversity (15.9%) are the dominant predictors of optimal protocol selection, providing interpretability to the ML component. The findings demonstrate that ML-based hybridization of complementary routing protocols offers a resilient and energy-efficient routing solution for next-generation smart grid neighborhood area networks.