Conclusion

This paper presented a hybrid PLC–RF communication system designed to enhance the reliability and adaptability of smart metering networks within Advanced Metering Infrastructure. The architecture was validated using physical LX200V50 PLC modules configured as a mesh network and TI CC1352R RF modules operating in a star topology under the TI 15.4 stack, demonstrating link rerouting, fault tolerance, and reliable end-to-end data transmission to a central Data Concentrator Unit. A cloud-based web application was integrated for real-time monitoring of power consumption, network connectivity, and active communication paths. To enable intelligent channel selection, a Proximal Policy Optimization (PPO) reinforcement learning agent was implemented and benchmarked against tabular Q-Learning, DDQN, and DDPG, dynamically switching between PLC and RF based on real-time latency, packet loss, and SNR observations. Evaluated over 5, 000 deterministic test cases against a reward-derived oracle, the PPO agent achieved an accuracy of 93.78% and an F1 score of 87.74%, attaining the highest single-run performance among all evaluated agents, with statistically significant advantages over Q-Learning and, for F1 score, over DDPG. Multi-seed training across ten distinct initializations confirmed superior convergence stability with an accuracy standard deviation of only 8.74%, mitigating the policy collapse risks observed in DDQN. Edge AI profiling on an Arm Cortex-M4 microcontroller platform demonstrated that full INT8 quantization compresses the model by 2.36× to a footprint of 64.7 KB with a deterministic inference latency of 1.93 ms, preserving >99.9% of the baseline accuracy. A reward function sensitivity analysis under 16 coefficient perturbations confirmed the robustness of the learned policy across varying weight configurations. Finally, a simulation-based network-scale evaluation across 500 to 5, 000 nodes confirmed scale-invariant performance with a Packet Delivery Ratio above 92% and per-node batch latencies of 2.27–22.9 _μ_s, while stress tests under node outages, traffic overload, and channel degradation demonstrated consistent resilience advantages over static baselines. These results collectively provide simulation-based evidence of the viability of AI-driven adaptive communication for resilient, large-scale smart grid infrastructures.