EIL-CNN: Embodied Intelligence-Driven Lightweight CNN for Representation Learning and Intelligent Extraction of Power Metering Information

Accurate and efficient extraction of power metering information is essential for modern smart grids and energy management systems. Traditional methods often suffer from high computational cost and limited adaptability to dynamic electrical environments, especially when processing high-dimensional, multi-source, and non-stationary load data, where manually designed features may lead to increased prediction errors and delayed inference. In this paper, we propose EIL-CNN, an Embodied Intelligence-Driven Lightweight Convolutional Neural Network framework for representation learning and intelligent extraction of power metering data. By integrating embodied intelligence principles, EIL-CNN enables adaptive feature selection and contextual awareness in metering data streams, allowing the network to focus on high-value information. A lightweight convolutional architecture further ensures computational efficiency, enabling real-time deployment on edge devices such as smart meters and gateways. Extensive experiments on multiple real-world power datasets demonstrate that EIL-CNN achieves superior performance in both representation learning and intelligent extraction tasks, while significantly reducing computational overhead compared to conventional deep learning approaches. The proposed framework paves the way for intelligent, edge-deployable power metering solutions in future smart grids.