Graph toolbox simplifies building graph neural networks for electricity forecasting
GraphToolbox: A Configurable Python Framework for Graph Neural Network Forecasting
Machine LearningArtificial Intelligence
Summary
Electricity forecasting involves predicting signals that are related by location, which can be naturally represented using graphs. The authors created GraphToolbox, a Python tool that combines many complicated steps needed to build graph-based forecasting models into one easy pipeline. It supports many different graph neural network types and adds features like mixing model predictions and interpreting results. The tool was tested on French electricity data, showing better accuracy than traditional methods and helping compare many types of graph models systematically.
What this means in practice
- •For energy forecasting teams: Build, train, and evaluate graph neural network models for regional electricity demand prediction using a unified Python framework.
- •For environmental monitoring groups: Implement graph-based forecasting of spatially distributed signals such as pollution or weather data to improve regional prediction models.
Authors
Eloi Campagne, Yvenn Amara-Ouali, Yannig Goude, Argyris Kalogeratos
Abstract
Electricity forecasting often involves spatially related signals observed over regions, substations, and feeders, and Graph Neural Networks (GNNs) provide a natural way to represent these relations. Building a complete GNN forecasting experiment is nonetheless laborious, because graph construction, model selection, training, aggregation, and interpretation sit in incompatible tools. We present GraphToolbox, an open-source Python framework that unifies these stages in one configurationdriven pipeline built on PyTorch Geometric. It offers data-driven graph construction, an adapter that instantiates and trains 51 of the 65 PyTorch Geometric convolutions together with the recurrent cells of PyTorch Geometric Temporal, online expert aggregation, forecasting interpretability, and significance testing on cached forecasts. We evaluate the pipeline in two case studies. On French regional load, the 48 convolutions included in the complete forecasting sweep fall in a band from 1.14% to 1.60% error, online aggregation lowers this to 0.98%, and the graph models improve on classical additive and boosting baselines. On net-load, direct graph models are less accurate than a classical additive model, while forecasting each physical component separately improves them without closing that gap. Both comparisons use the same experimental interface, illustrating the role of GraphToolbox in systematic architectural evaluation.