Graph Neural Networks for PV Power Forecasting on Smart Meters
A study explores deploying graph neural networks on edge intelligent meters to forecast photovoltaic power generation in microgrids. The research introduces ONNX and ONNX Runtime for model deployment, focusing on GCN and GraphSAGE models. A custom ONNX operator was developed for GCN. Using real datasets from a village microgrid, the models were successfully deployed and executed on a smart meter, with performance compared between PC and the meter.
Key facts
- Graph neural networks used for PV power forecasting on smart meters
- ONNX and ONNX Runtime adopted for deployment
- GCN and GraphSAGE models trained and deployed
- Custom ONNX operator developed for GCN
- Real datasets from a village microgrid used
- Performance compared on PC and smart meter
- Successful deployment and execution on smart meter
Entities
—