Light weight Intelligence-TinyML based Energy Usage Prediction for Resource-Constrained IoT Edges Nodes

Abstract

The extensive deployment of battery-powered and resource-constrained edge devices makes energy efficiency a major challenge in Internet of Things (IoT) systems. Accurate energy prediction is important to enable intelligent energy management. However, traditional machine learning models are usually computationally expensive and unsuitable for micro-controller based platforms. In this paper, we present a TinyML-based energy prediction framework for low-power IoT edge devices. The proposed approach employs lightweight machine learning models which are optimized for ultra-low memory and computation footprints, but still retain acceptable prediction accuracy. We collect energy consumption data from a real IoT testbed, and train and evaluate several TinyML compatible models. The experimental results show that the proposed TinyML-based predictor can provide reliable energy estimation with low inference latency and low memory overhead, and thus can be deployed on resource-constrained IoT devices. This work lays a fundamental foundation for intelligent and adaptive energy management in future IoT systems.