PowerFlowNet

Leveraging Message Passing GNNs for High-Quality Power Flow Approximation.

PowerFlowNet's distinctiveness, compared to existing PF GNN approaches, lies in its adept utilization of the capabilities from message-passing GNNs and high-order GCNs in a unique arrangement called PowerFlowConv, for handling a trainable masked embedding of the network graph. This innovative approach renders PoweFlowNet remarkably scalable, presenting an effective solution for the PF problem.

The PowerFlowNet Paper can be found at: link

Description

PowerFlowNet transforms the PF into a GNN node-regression problem by representing each bus as a node and each transmission line as an edge while maintaining the network's connectivity.

Instructions

To train a model run train.py with the desired arguments. For example:

python3 train.py --cfg_json ./configs/standard.json\
                --num-epochs 2000\
                --data-dir ./data/
                --batch-size 128\
                --train_loss_fn mse_loss\
                --lr 0.001\
                --case 118v2\
                --model MaskEmbdMultiMPN\
                --save

Datasets

Follow the links below to download the datasets and the trained models used in the paper.

Dataset link

Trained models link

File Structure

runnable files:

• train.py trains the model
• results.py plots the results
• and more scripts to generate results and plots ...

Useful Information

First two dimensions out of seven in edge_features are from_node and to_node, and they are indexed from $1$. This is processed in the PowerFlowData dataset class. It is reindexed from $0$ and the from_node and to_node are removed from the edge_features tensor.

Raw data format:

Number	Description
N	number of nodes
E	number of edges
Fn = 9	number of features per node
Fe = 5	orginally 7, first two dims are from_node and to_node number of features per edge
Fn_out = 8	number of output features per node

Tensor	Dimension
Data.x	(batch_size*N, Fe)
Data.edge_index	(2, E)
Data.edge_attr	(E, Fe)
Data.y	(batch_size*N, Fn)

Citation

If you use parts of this framework, datasets, or trained models, please cite as:

@article{LIN2024110112,
  title = {PowerFlowNet: Power flow approximation using message passing Graph Neural Networks},
  journal = {International Journal of Electrical Power & Energy Systems},
  volume = {160},
  pages = {110112},
  year = {2024},
  issn = {0142-0615},
  doi = {https://doi.org/10.1016/j.ijepes.2024.110112},
  author = {Nan Lin and Stavros Orfanoudakis and Nathan Ordonez Cardenas and Juan S. Giraldo and Pedro P. Vergara},
}