This tool assists researchers to export electrical quantities such as power flows, injections and others to featured datasets that can be used in data-driven modelling such as neural networks.

Using Pandapower http://www.pandapower.org/, the electrical quantities are first imported from a test power grid such as IEEE-118. Loads in the power grid are then varied and power flow analysis is conducted for each load scenario. The resulting data is csv tabulated with proper headers that fully describe the electrical quantities. For instance, qfl45_35_33 is the reactive power flowing in a transmission line from bus 35 to bus 33.

Since this code was developed as part of research related to Power State Estimation, states are generated in a separate file. However, all quantities are linked with an ID indicating the corresponding load scenario. Rows are duplicated to allow for adding noise. If not needed, duplicate rows may be easily dropped when imported by Pandas. The three generated files (source_clean_meas.csv, source_noisy_meas and target_gt_states.csv) are all found in the subfolder datasets.
The underlying functions are all defined in the python script file gen_scen_fnctns.py. The loads are varied using a uniform distribution. If needed, the function gen_clean_scen_unfrm may be easily modified to allow for other projected load patterns.

The tool further generates the DC measurement matrix H sorted in the same order of the selected measurements. Clearly, the H matrix does not include entries for voltage magnitudes (all 1 p.u. for DC model assumptions). This matrix can be found under the subfolder base_net. Note: The H matrix is tested only for IEEE-118 case.

1. Install conda https://docs.conda.io/projects/conda/en/latest/user-guide/install/index.html

2. Open conda ternminal in the project root and create conda environment using:conda env create -f pdv1_env.yml. Note: Change the prefix in the yml file as per your convenience. You may also choose to install the listed dependencies manually.

3. Activate pdv1_env using: conda activate pdv1_env

4. Install panda power using: pip install pandapower

5. Run jupyter notebook from the terminal

Please make sure you have the following main folders in your project:

• notebooks

• runs

• src

You may need to manually create ’runs’ folder if not available

1. Open generate_meas_template notebook.

2. Select a power grid e.g. Enter net=nw.case118(). Note: This tool is tested only for IEEE118.

3. Run the notebook.

4. Go to and open ..runsmeas_template.csv. The file lists all available measurements in the selected grid.

5. Set the measurement you need to include in your target dataset to TRUE. Save the file and close it.

6. Open generate_load_scenarios notebook.

7. Select the base power grid. This should match the one used to generate the measurement template e.g. net=nw.case118

8. Enter your parameters and run the notebook. The notebook will construct a timestamped run folder with three subfolders, namely base_net, datasets, and grid.

9. Check the generated datastes under the subfolder datasets. Other subfolders contain useful information about the grid, e.g. buses, lines..etc.

Note that you will find useful comments embedded in the code. All units of measurement are inline with Pandapower except the angles where they are given in radians instead of degrees.

1. Q: It is mentioned in the readme document that "qfl45_35_33 is the reactive power flowing in a transmission line from bus 35 to bus 33.". What’s the meaning of 45 here, is it the line number?
   A: It is the transmission line index used by panda power. The same applies to the rest of quantities. For instance, pft1_25_24 refers to active power flow from bus 25 to bus 24 in the transformer with panda power index 1. I used the following naming convention:

   • Pf: for active power flow

   • qf: for reactive power flow

   • p: for active power injection

   • q: for reactive power injection

   • l: for transmission line

   • t: for transformer

   • vm: is bus voltage magnitude in pu

   • va :is bus voltage angle in radians

2. Q: In the csv data, what is the scenario number? Can I think they are the independent measurement at different time points?
   A: Yes, for sure. Load scenarios can be thought of as snapshots in time.

3. Q: Each scenario (e.g.,A0) is copied 10 (or any number) times in the clean dataset, I think your purpose is for later use to add 10 different noises. Is it correct?
   A: Yes, correct.

4. Q: I noticed there are three H matrices under the ’base_net’ repository, which one should be used in DC power flow model?
   A: You need to use h_no_slack.csv. The first row in the file shows the bus indices.

For any queries, please contact Samer Kababji at [email protected]