VoltGNN: A graph neural network for autonomous renewable energy redistribution in IoT-driven smart grids
ENERGY AND AI, vol.25, 2026 (ESCI, Scopus)
- Publication Type: Article / Article
- Volume: 25
- Publication Date: 2026
- Doi Number: 10.1016/j.egyai.2026.100841
- Journal Name: ENERGY AND AI
- Journal Indexes: Emerging Sources Citation Index (ESCI), Scopus, Compendex, INSPEC, Directory of Open Access Journals
- Azerbaijan State University of Economics (UNEC) Affiliated: No
Abstract
Conventional graph neural networks-based approaches applied to smart grid management are primarily designed for passive tasks such as load forecasting, anomaly detection, and demand prediction. This paper introduces VoltGNN, a generative graph intelligence framework that transforms a standard spatio-temporal graph neural network into a graph-conditioned policy synthesis system for autonomous renewable energy distribution. Rather than predicting future demand or generation, VoltGNN directly generates an energy redistribution policy, namely, a capacity-bounded power-flow allocation over every transmission edge of a dynamically constructed grid graph. VoltGNN ingests multi-source internet of things streams from smart meters, solar panels, wind turbines, electric vehicle charging stations, and battery storage units; constructs dynamic energy graphs whose node states and edge weights evolve at every sampling interval; and learns spatio-temporal dependencies through a graph attention network coupled with a temporal graph network backbone. A dedicated generative decision layer, implemented as a graph-conditioned transformer decoder with reinforcement-guided policy refinement, synthesizes explicit energy redistribution strategies. An autonomous optimization layer then resolves conflicting objectives such as minimizing transmission loss, overload probability, and distribution latency while maximizing renewable utilization. Ablation studies confirm the individual contributions of each component. Experiments on two real-world field datasets and one synthetic but physically consistent dataset, together with IEEE 33-Bus and IEEE 118-Bus simulation environments show that, within these simulated and reconstructed settings, VoltGNN raises the overload-reduction rate, improves renewable utilization, and keeps distribution decision latency relatively low. Project details are available at https://github.com/AraoufBh/VoltGNN.