sunchaesk / ThermoGNN

Thermal Zone modelling using GNN

Building Modelling using GNN

• the most accurate model is defined in file gcn2.py

gcn2.py

• uses GNN-RNN architecture and was able to achieve on par performance to a typical mlp (RNN) based model
• gcn2 uses a GCN for generating a representation of the graph nodes, and the nodes are passed into a RNN layer

gnn model/training pipeline

1. data_collect.py: collects system-excited (randomly set actuation value) -> the data is not in graph format
2. data_scale.py: data dimension goes over the scikitlearn.StandardScaler capability
3. data_process.py: loop over the collected data, embed the vector data in graph format
4. gcn2.py: include PyG dataloader + model + training code