A collaborative forecasting platform for the energy sector
Predico is a collaborative forecasting platform, built by INESC TEC Centre for Power and Energy Systems (CPES), for energy time-series forecasting.
It improves accuracy and robustness by combining forecasts from multiple suppliers using ensemble methods, rather than relying on a single model or vendor. Check the Official Platform Website for more details.
Note
Predico is currently being tested as an operational proof-of-concept for Elia, Belgium's Electricity System Operator. For more information about the project and its applications, please visit Elia Innovation Press Release or contact us directly (see Contacts section below).
This repository contains the Forecast Engine, the core algorithmic layer of Predico that:
- Orchestrates forecasting sessions (a.k.a. market sessions)
- Runs ensemble strategies to combine individual forecasts
- Scores forecast quality with multiple metrics
- Provides a standalone simulator to develop and benchmark strategies locally
Important
Predico follows an open core approach. This repository contains exclusively the open-source Forecast Engine, that supports the full Predico platform (RESTful API, Web UI, marketplace features). Contact us for more details.
This project is currently in development. Core functionalities are working, but improvements are ongoing. The available base ensemble strategies and scoring mechanisms are production-ready while experimental features (e.g., advanced ensemble strategies, forecast contribution) are under active development and testing.
See the Releases page for version history, changelogs, and downloadable snapshots.
| Category | Technologies |
|---|---|
| Language | Python 3.12+ |
| ML & Data | scikit-learn, pandas, NumPy, joblib |
| Outlier Detection | DTAIDistance (Dynamic Time Warping) |
| Database | SQLAlchemy, psycopg2 (PostgreSQL) |
| CLI | Python Fire |
| Logging | Loguru |
| Progress | tqdm |
| Package Management | Poetry |
This project requires Python 3.12+. All dependencies are managed via pyproject.toml, with Poetry.
Install Poetry (if not already installed), via pip install poetry or see Poetry Installation Guide.
Then, install all dependencies with poetry install.
.
├── src/
│ ├── strategies/ # Ensemble algorithms
│ ├── features/ # Feature engineering (lags, diversity metrics, rolling stats)
│ ├── assessment/ # Scoring metrics (RMSE, Pinball Loss, Winkler)
│ ├── market/ # Session orchestration and forecaster management
│ ├── io/ # Database and API I/O helpers
│ ├── core/ # Interfaces, configuration, exceptions
│ └── MarketController.py # Production orchestration (session lifecycle)
├── simulator/
│ └── community/ # Public simulator for strategy development
├── tests/ # Unit tests (pytest)
│ ├── strategies/ # Strategy-specific tests
│ ├── features/ # Feature engineering tests
│ ├── assessment/ # Scoring tests
│ └── validation/ # Data validation tests
├── tasks.py # Fire CLI for production market session loop
└── conf/ # Configuration settings
Tip
See src/strategies/README.md for available strategies and how to add new ones.
This repository includes a standalone forecasting simulator to facilitate the development, testing, and benchmarking of ensemble strategies without requiring access to the full Predico platform (API, database, etc.).
Tip
A complete guide to using the Community Simulator is available in the simulator/community/README.md
# Navigate to the community simulator
cd simulator/community/
# Generate synthetic datasets with configurable forecaster behaviors
# (interactive wizard — use 'quickstart' for a non-interactive alternative)
python simulate.py generate_dataset
# Or run a quick non-interactive demo
python simulate.py quickstart
# Run simulations and compare strategies
python simulate.py run --strategies="weighted_avg,median"
# Visualize results
python simulate.py plot output/<dataset>/<timestamp>/Predico's community simulator allows you to develop and test ensemble forecasting strategies locally, without needing access to the full platform backend.
# Clone the repository
git clone https://github.com/INESCTEC/predico-collabforecast.git
cd predico-collabforecast
# Install dependencies
poetry install
# Run a quick simulation
cd simulator/community
poetry run python simulate.py quickstartNote
You can also activate the virtual environment with poetry shell and then run commands directly (e.g., python simulate.py quickstart). Using poetry run is recommended as a single-step alternative.
This generates a synthetic dataset, with predefined forecaster supplier archetypes, and runs a simulation to demonstrate how ensembles can outperform individual predictors.
You can easily add your own ensemble forecasting strategies.
First, add your strategy code in src/strategies/, following the existing structure.
In this example, we create a simple averaging strategy, leveraging on our SimpleStrategy base class.
# src/strategies/my_strategy.py
from src.strategies import SimpleStrategy, StrategyRegistry
@StrategyRegistry.register("my_strategy")
class MyStrategy(SimpleStrategy):
@property
def name(self) -> str:
return "my_strategy"
def combine(self, forecasts, **kwargs):
# forecasts: pandas.DataFrame, rows=time, columns=forecasters
return forecasts.mean(axis=1)Then, import your strategy in src/strategies/__init__.py so the @register decorator is triggered at load time:
# src/strategies/__init__.py
from .my_strategy import MyStrategy # This import triggers @register
__all__ = [
# ... Other strategy imports ...
"MyStrategy",
]Finally, run the community simulator using your new strategy:
cd simulator/community
python simulate.py run --strategies="my_strategy" --n_sessions=10| Resource | Description |
|---|---|
simulator/community/README.md |
Getting started with local simulations |
src/strategies/README.md |
Available algorithms + how to create new ones |
simulator/community/docs/strategy-guide.md |
Step-by-step strategy development |
simulator/community/docs/advanced-strategies.md |
Advanced base classes and techniques |
simulator/community/docs/cli-reference.md |
CLI options |
simulator/community/docs/dataset-format.md |
Input data specifications |
simulator/community/docs/evaluation-metrics.md |
RMSE, Pinball Loss, Winkler scoring |
For platform integration and API documentation: https://predico.inesctec.pt/docs/
Below is an overview of the main components and data flow within the Predico Forecast Engine. Besides the ensemble strategies, Predico also includes modules for outlier detection, feature engineering, session management, and scoring.
No major issues are currently documented. For bug reports or feature requests, please open an issue in the repository or contact the maintainers.
Contributions are welcome (strategies, metrics, docs, tests). If you add a new strategy, include:
- A minimal unit test (see
tests/) - A short note about your strategy in
src/strategies/README.md
Running Tests:
poetry run pytestSee INESC TEC's Contributing Guidelines for more details on how to contribute.
This project follows the INESC TEC organisation-wide governance documents:
Please review these documents before contributing to the project.
This project is developed and maintained by INESC TEC.
Licensed under the GNU Affero General Public License v3.0 (AGPL-3.0).
- See:
LICENSE - Commercial licensing is available — contact INESC TEC Technology Transfer Office (TTO) at tech-transfer@inesctec.pt
- José Andrade — jose.r.andrade@inesctec.pt
- André Garcia — andre.f.garcia@inesctec.pt
- Carla Gonçalves — carla.s.goncalves@inesctec.pt
- Ricardo Bessa — ricardo.j.bessa@inesctec.pt