Run the optimal power flow on AC network with augmented data

[TurtleCamera]

Change in code

• [x] create a new branch ac_opf
• [x] make changes in h_data['y'] with the optimal pg from gen
  • [x] without the optimal results
  • [x] inputs from config file
  • [ ] ~with the optimal results~
  • [ ] ~update/read node/edge features from result~
  • [ ] ~branch has qf, qt, pt, pf~
  • [ ] ~gen has qg~
  • [ ] ~bus has va, vm~

Change in training

• [x] split the generated data into train/val/test as 80/10/10
• [x] train the model in mini-batch
  • [x] build data from DataLoader

Update the model as the regression model to predict the pg/qg/va/vm in generator/bus Keep the DC network as it is.

• [ ] verify the performance of HeteroGNN

[cshjin]

Load the result to data only when the termination_status is locally_solved. @TurtleCamera you can move forward with processing the JSON file into dataset.py

[cshjin]

Notes:

• load on bus with pd and qd values
• pg and qg on gen
• pd and pd are the perturbed values for load in buses, but with a scalar mpc['bus']['Pd'] = 100 * load_data['net']['load'][k]['pd']

[cshjin]

AC_OPF setting is considered as reg problem. Ignore for now.