PyPSA Network Analyzer

A post-processing and analysis toolkit for PyPSA hindcast simulations, developed to support the paper:

"Can Optimal Dispatch Models Recreate Reality? A Retrospective Analysis of Europe's Energy Crisis Using PyPSA-Eur"
Presented at EEM26, Trondheim - https://www.ntnu.edu/eem26

The tool compares simulated dispatch, prices, generation mix, and CO $_2$ emissions against historical benchmarks. While the repository is set up to reproduce the paper results, it is designed to work with any PyPSA .nc network file.

Solved PyPSA network files relative to the publication are hosted at DTU data

Quickstart

Clone the repository locally:

git clone https://github.com/marco-saretta/pypsa-network-analyzer.git
cd pypsa-network-analyzer

Setup the environment using uv (recommended):

uv sync

or using conda:

conda create -n pypsa-course python=3.11 && conda activate pypsa-course
pip install -e .

Reproducing the Paper Results

The paper simulated three scenarios:

Scenario	Fuel prices	Solving approach
Hindcast standard	Static	Full horizon at once
Hindcast dynamic	Dynamic	Full horizon at once
Hindcast dynamic rolling horizon	Dynamic	Two-week rolling horizon

Each scenario has hourly granularity, solved yearly from 2020 to 2024. This results in 5 network files per simulation, and totalling 15 files for all scenarios.

1. Download the solved network files

The 15 solved PyPSA network files are available at:

DTU data

Place the downloaded .nc files in:

data/network_files/     <-- Place the 15 ".nc" files downloaded here

2. Configure the run

configs/default_config.yaml is pre-configured with the correct file names. There is no need to edit the default .yaml file.

However, to run a subset of scenarios, comment/uncomment the relevant entries under network_files:

network_files:
  - "hindcast_standard_2020.nc"
  - "hindcast_standard_2021.nc"
  # ...

3. Run the analysis

First, run main.py to process the network files and produce the results CSVs:

# with uv
uv run main.py          

# with conda
python main.py          

Then generate the paper figures:

# with uv
uv run src/pypsa_network_analyzer/plot_figures_paper.py     

# with conda
python src/pypsa_network_analyzer/plot_figures_paper.py     

Figures are saved to figures_paper/.

Repository Structure

├── .github/                        # GitHub Actions and Dependabot
│
├── configs/                        # Hydra configuration files
│   ├── default_config.yaml         # Main configuration entry point
│   └── config_results_concat/
│       └── config_results_concat.yaml
│
├── data/
│   ├── benchmark/                  # Benchmark data for scoring
│   └── network_files/              # Place PyPSA .nc files here
│
├── docs/                           # Documentation
├── figures_paper/                  # Output figures for the paper
├── logs/                           # Log files
├── notebooks/                      # Jupyter notebooks for exploration
├── outputs/                        # General output files
├── results/                        # Analysis results (CSV)
├── results_concat/                 # Concatenated multi-year results
│
├── src/
│   └── pypsa_network_analyzer/
│       ├── utils/                  # Utilities and helpers
│       ├── __init__.py             # Import packages
│       ├── entsoe_retrieval.py     # ENTSO-E data fetching
│       ├── network_analyzer.py     # Core analysis per network file
│       ├── plot_figures_paper.py   # Figure generation for the paper
│       └── score_analyzer.py       # Scoring vs benchmark
│
├── .gitignore
├── .pre-commit-config.yaml
├── .python-version
├── LICENSE
├── main.py                         # Main entry point
├── pyproject.toml                  # Project config and dependencies
├── README.md                       # This file :-)
└── uv.lock

Other Repository Functionalities

Running the network analyzer on a single file

To analyze a single PyPSA network file directly, without going through main.py:

uv run src/pypsa_network_analyzer/network_analyzer.py      # with uv
python src/pypsa_network_analyzer/network_analyzer.py      # with conda

Using the tool with your own networks

Place your .nc files in data/network_files/ and update configs/default_config.yaml:

network_files:
  - "your_network_file.nc"

Then run main.py as above. The tool will produce dispatch comparisons, price diagnostics, energy-mix breakdowns, CO₂ emissions analysis, and benchmark scores for any valid PyPSA network.

Fetching live ENTSO-E data

Scripts that retrieve live data from ENTSO-E are available in utils/ via entsoe_retrieval.py. An API key is required and can be set in configs/default_config.yaml:

api_key: YOUR-ENTSOE-API-KEY

Configuration

The project uses Hydra for configuration management.
The main entry point is:

configs/default_config.yaml

Default configuration structure

defaults:
  - config_results_concat: config_results_concat
  - _self_

network_files:
  - "YOUR_NETWORK_FILE_HERE.nc"
  - "YOUR_NETWORK_FILE_HERE_1.nc"
  
paths:
  root: "${hydra:runtime.cwd}"
  log: "./logs"

api_key: YOUR-ENTSOE-API-KEY

plot_export_format: "pdf"

metrics:
  - mae
  - rmse
  - smape

Key fields

• network_files
  List of PyPSA .nc files to analyze.
  Comment or uncomment blocks to switch between hindcast variants.

• paths.root
  Base directory for relative paths. By default, this is the runtime working directory.

• paths.log
  Directory for log files.

• api_key
  Optional ENTSO-E API key. Required only for scripts that fetch live data.

• exclude_countries
  Countries to omit from analysis due to data quality or model issues.

• metrics
  Error metrics computed when comparing simulations.

Data

Data relative to the solved network files is available at DTU data.

Input data required for the hindcast analyses(load, generation, prices, capacities, emissions) are either:

• already included, or
• expected to be locally available in preprocessed form.

ENTSO-E API tokens are not included.
Scripts that require live ENTSO-E access are optional and clearly separated in utils/.

Authors

• Lukas Karkossa and Marco Saretta
  Technical University of Denmark
  {lalka, mcsr}@dtu.dk

• Frederik Erhard Gullach
  Aarhus University
  fegu@cowi.com

Template

Created using DTU_mlops_template, a cookiecutter template for getting started with Machine Learning Operations (MLOps).