Machine Learning-Driven Lightweight Protocol for Energy-Efficient IoT Communication

The Internet of Things (IoT) ecosystems have grown at a very high rate and this has increased the demand of energy-efficient and scalable communication protocols especially in smart grid and advanced metering infrastructure (AMI) applications. Because of their high energy consumption, latency bottlenecks and lack of adaptability, traditional IoT communication techniques like MQTT, CoAP as well as heuristic duty cycling are frequently inappropriate for large-scale installations. The hybrid multi-layered system proposed in this research integrates Bayesian Optimisation (BO) at the cloud level, Federated Reinforcement Learning (FRL) in the gateway level and Tiny Machine Learning (TinyML) on the node level to overcome these limitations. The methodology uses TinyML to reduce duplicate data transmissions by predicting local usage and detecting anomalies. While cloud-based Bayesian Optimisation constantly adjusts hyperparameters like learning rates along with duty cycles to maintain the network running efficiently, gateways use FRL to learn the most effective way to schedule transmissions in a way that reduces latency while maximising energy savings. AMI data was used to compare the proposed method to standard protocols and centralised machine learning models. Contrasting experimental findings to traditional approaches, it is shown that there is a 30% decrease in bandwidth, a 45% reduction in energy and an improvement in latency performance. The findings underline that diffusion of intelligence across the IoT hierarchy enables scalable, versatile and energy efficient communication besides safeguarding of privacy, as well as cut in communication overhead. This study opens the way to more sustainable IoT infrastructure of smart cities and manufacturing systems.

Hali tarjima qilinmagan