Summary of the main code snipplets used during my master's thesis "Embedding Neural Networks into Dynamic Power System Simulators"
Here is an example workflow to navigate through this repository. For now, it only explains how to use a NN in FORTRAN/RAMSES. Updates about the whole PowerPINN pipeline will be added in a later commit:
-
First, the neural network should be trained and exported in .onnx (Open Neural Network Exchange) format. In Python, using PyTorch: torch.onnx.export(model, init_cond, "NN.onnx"), where init_cond is an example input of the neural network.
-
Then, the tutorial roseNNa_tutorial should be followed to transform the .onnx files to those readable by roseNNA.
-
Finally, in the FORTRAN code, the functions from the roseNNa library are added to the solution. Then, the following lines are added to call the Neural Network: program NN_program use rosenna ... ! Define and initialise values for the nn inputs and outputs real(real64), dimension(1,10) :: nn_input ! Example input with 10 features real(real64), dimension(1,9) :: nn_output ! Example output with 9 features
! Initialise rosenna call initialize_nnx()
! NN forward pass call use_model(pinn_input, pinn_output)
Once the code is written, the program NN_program should be added into the workflow of the code. in RAMSES, the power system model should be added to the lists of models.
- A further update of this repository will include the whole pipeline from PINN training to RAMSES simulations.