Energy-Efficient AI Models for Large-Scale Technical Data Collection in Cyber-Physical Systems

Large-scale Cyber-Physical Systems (CPS) produce diverse, high-velocity technical data that necessitate constant acquisition, processing, and transmission. When using traditional fixed-rate data collection methods, energy is wasted, communication is slowed down, and sampling is done too often, especially in distributed sensor networks. This paper presents an energy-efficient AI-driven data acquisition framework that incorporates Reinforcement Learning (RL), adaptive sampling, and lightweight deep learning models to dynamically optimize sampling rates according to sensed context, signal variability, and operational constraints. When tested on a publicly avaialble data set Intel Lab Data with 120 sensor nodes used to monitor temperature, the proposed model uses 41.6 % less energy, 37.2 % fewer redundant data samples, and 22.4 % more accurate data than static sampling. The framework shows promise for being used on a large scale in smart grids, the industrial Internet of Things, precision agriculture, and strong CPS infrastructures.