On-Meter Graph Machine Learning: A Case Study of PV Power Forecasting for Grid Edge Intelligence

2026-04-23 · Source: cs.LG updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Internet of Things (IoT) & Connected Devices, Renewable Energy Systems · Depth: Advanced, long

Summary

This study details the deployment of graph neural networks (GNNs) on edge intelligent meters for photovoltaic (PV) power forecasting within a microgrid. It introduces the problem of "cloud-less" village microgrids requiring local forecasting without real-time meteorological data, relying solely on historical power measurements. The paper focuses on training and deploying two GNN models, GCN and GraphSAGE, using ONNX and ONNX Runtime. A key contribution is the development and deployment of a customized ONNX operator for GCN layers, which are not natively supported. The hardware specifications of the smart meter, an ARM-based quad-core Cortex-A53 CPU with 984 MB RAM, are described. A case study using real datasets from a village microgrid from January to May 2024 demonstrates successful deployment and execution, comparing model performance on both PC and the smart meter.

Key takeaway

For AI Scientists and Research Scientists developing edge computing solutions, this work demonstrates that complex GNN models can be effectively deployed on resource-constrained smart meters. You should consider developing custom ONNX operators for GNN layers not natively supported to ensure cross-platform compatibility and maintain model accuracy, especially for applications like PV power forecasting in remote microgrids where cloud connectivity is unreliable.

Key insights

GNNs can be deployed on resource-constrained smart meters for local PV forecasting using ONNX and custom operators.

Principles

• Edge autonomy reduces cloud reliance.
• GNNs excel with graph-structured data.
• ONNX enables cross-platform model deployment.

Method

The method involves defining custom ONNX operators for unsupported GNN layers (like GCN), modifying batch processing for models like GraphSAGE for ONNX compatibility, and implementing backend kernels for custom operators within ONNX Runtime.

In practice

• Develop custom ONNX operators for unsupported GNN layers.
• Rewrite complex batch operations for ONNX export.
• Utilize ONNX Runtime for efficient edge inference.

Topics

• Graph Neural Networks
• PV Power Forecasting
• Grid Edge Intelligence
• ONNX Runtime
• Customized ONNX Operators

Best for: AI Scientist, Research Scientist, Machine Learning Engineer, AI Engineer, AI Hardware Engineer

Related on AIssential

• Outage Detection in Self-Healing Smart Grids Using Reinforcement Learning with Spectral Graph Neural Networks — Spectral GNNs in RL enable real-time, scalable outage management in smart grids by capturing global network topology.
• What Is ONNX? (And Why Transformers.js Uses It) — ONNX provides an open standard for neural network models, enabling cross-platform portability and efficient execution via specialized runtimes.
• Learning to Approximate Uniform Facility Location via Graph Neural Networks — A differentiable, unsupervised MPNN bridges classical approximation algorithms and neural networks for combinatorial optimization, offering provable guarantees.
• Step-adaptive multimodal fusion network with multi-scale cloud feature learning for ultra-short-term solar irradiance forecasting — A multi-source fusion model with adaptive compensation improves ultra-short-term solar irradiance forecasting accuracy.
• Q&A: Developing a sustainable power grid in the era of AI — AI infrastructure growth is bottlenecked by electric power, necessitating grid modernization for both AI and decarbonization.
• Optimizing Energy-based Neural Network Training with Coherent Ising Machine — Coherent Ising Machines can train energy-based neural networks, enhanced by Adam, for energy-efficient AI hardware.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.LG updates on arXiv.org.