An open-source power grid dataset generator.

This tool generates a dataset of pandapower power grids datasets from a single snapshot.

It is compatible with powerdata-view for dataset visualization.

It randomly samples power grids built from a provided source power grid to generate datasets (i.e. directories filled with pandapower files). After the sampling, an AC power flow is run and a filtering is applied to make sure that the generated dataset satisfies certain conditions.

First, you need to clone the repository :

git clone https://github.com/bdonon/powerdata-gen.git

Then, go inside the project :

cd powerdata-gen

It is usually a good practice to have a virtual environment per project, so that any package installation that you do for one project will not alter the others. There are multiple ways of creating a virtual environment (virtualenv, conda or even your IDE).

In the following, we guide you through the creation of a virtual environment using the package virtualenv :

pip install virtualenv
virtualenv venv -p python3.10

Then, you need to activate the virtual environment :

source venv/bin/activate

Once your virtual environment activated, you will have to install the packages that powerdata-view requires :

pip install -r requirements.txt

To run powerdata-gen, you just need to run the following :

python main.py

The generated datasets are located in outputs/.

The configuration is defined in config/config.yaml. Here are the different fields :

• default_net_path: Path to the source pandapower .json file.
• n_train: Amount of sampled power grids in the Train dataset.
• n_val: Amount of sampled power grids in the Val dataset.
• n_test: Amount of sampled power grids in the Test dataset.
• seed: Random seed for the data generation process.
• sampling: Defines the sampling methods for the different components of the grid.
  • topology: Topology sampling process, cf. below.
  • total_total: Total active load sum sampling process, cf. below.
  • active_load: Individual active load sampling process, cf. below.
  • reactive_load: Individual reactive load sampling process, cf. below.
  • active_gen: Individual active generation sampling process, cf. below.
  • voltage_setpoint: Individual voltage set points sampling process, cf. below.
• powerflow: Pandapower AC power flow options.
• filtering: Defines the filtering step that rejects invalid samples, cf. below.

As shown in the figure above, the sampling process can be split into multiple parts. For each part, the configuration is required to provide a sampling method name, and a set of parameters (defined as a set of keyword arguments).

Sampling of the power grid topology.

In the random_disconnection mode, one should define a list of disconnection probabilities for generators, loads and lines. For instance, let us consider the following parameters :

  topology:
    method: "random_disconnection"
    params:
      line:
        probs:
          0: 0.2
          4: 0.8
        black_list: [25, 30]

There is a 20% probability that no lines are disconnected, and 80% probability that four of them (uniformly selected) are disconnected. Moreover, lines 25 and 30 are ``blacklisted'' and will not be disconnected at all.

Sampling of the total consumption of the grid, denoted as $P_{tot}^{new}$. The following sampling methods and corresponding params are available:

Sampling of the individual active loads in MW. The following sampling methods and corresponding params are available:

[^1] scaled so that it respects the total load.

where $n$ is the amount of loads and $S(\alpha) = \lbrace x | \sum x_i = 1, x_i \leq \alpha \rbrace $.

In the four last sampling methods, the parameter $beta \in [0,1]$ controls the spread of the distribution. Let us consider the case of the uniform_independent_values.

• If $beta = 0$, then the distribution is a dirac where all loads are the same.
• If $beta = 1$ then the distribution is uniform over a simplex.

WARNING: The marginal distributions displayed in the figures above are only valid if there are only two loads. When dealing with n loads, the uniform sampling will amount to sampling from a n-dimensional simplex, thus making the marginal distributions skewed towards low values.

Sampling of the individual reactive loads in MVAr. The sampling of the reactive load is performed after the sampling of active loads. In many of the proposed methods, the reactive power of a load depends on the new value of the active power. The following sampling methods and corresponding params are available:

Sampling of the individual active loads in MW. The following sampling methods and corresponding params are available:

In the dc_opf_disconnect mode, unused generators are disconnected.

Sampling of the individual voltage set points in p.u.. The following sampling methods and corresponding params are available:

After the sampling and the AC power flow step, each data sample is passed to a filtering function that checks a certain amount of conditions, and rejects invalid samples. Here are the currently implemented filtering options :

• max_loading_percent: Maximum branch loading percent. Rejects samples with overflow.
• max_count_voltage_violation: Maximum amount of voltage violations. Rejects samples with too many voltage violations.
• allow_disconnected_bus
• allow_negative_load
• allow_out_of_range_gen

If you want to define a different configuration file (e.g. config_2.yaml), make sure to place it inside the config/ directory, and use it using the following :

python main.py --config-name=config_2.yaml

If you have any questions, please contact me at [email protected]