222042015/scopf-approx

SCOPF surrogate

1. generate the scenario data
2. evaluate the solutions to generate the training data for the approximation network

The evaluation of the solutions is done by the evaluate.py and data.py script. The script is based on https://github.com/GOCompetition/Evaluation.git

Repository structure

- src
    - generate_data.py  \\generate the scenario data
    - random_load_profile.py    \\generate the random load profile
    - generate_data_parallel.py and scenario_operations.py  \\used to generate the scenario data in parallel
    - get_stage2_pen.py     \\gather the second stage penalty from each detail file, and splitting the data for training the approximation neural network.
    - generate_result.py    \\generate solving time and solution results into the the ./result folder
- model
    folder to store the trained model weights
- result    \\store the result csv files
    - result.ipynb and time_result.ipynb    \\generate the result table and plots
- load_model.py     \\load the trained model and write the weights of the neural network in to txt files in the model folder
- test.py, test.sh      \\the function to call the evalutation.run method to evaluate the solutions
- examples
    - case2 # an example datafile 

Generate scenarios

python generate_data.py

python random_load_profile.py

Get the penalization for collected base case solutions

bash test.sh
python get_stage2_pen.py

Train the model

python approx_model.py