Generating Probabilistic Forecasts from Arbitrary Point Forecasts Using a Conditional Invertible Neural Network

This repository contains the Python implementation of the approach to generate probabilistic forecasts from arbitrary point forecasts. This approach is presented in the following paper:

Kaleb Phipps, Benedikt Heidrich, Marian Turowski, Moritz Wittig, Ralf Mikut, and Veit Hagenmeyer. 2024. Generating Probabilistic Forecasts from Arbitrary Point Forecasts Using a Conditional Invertible Neural Network. Applied Intelligence. Springer.

Repository Structure

This repository is structured in a few key folders:

• base_pipelines: This folder contains the code used to create the pipelines that are then executed for each data set.
• data: This folder contains the data used for the analyses in our paper.
• modules: This folder contains multiple pyWATTS modules that are included in the pipelines.
• pipelines: This folder contains the pipelines which can be executed to recreate the results from Chapter 4.

Installation

Before the proposed approach can be applied using a pyWATTS pipeline, you need to prepare a Python environment and download energy time series (if you have no data available).

1. Setup Python Environment

Perform the following steps:

• Set up a virtual environment of Python 3.10 using e.g. venv (python3.10 -m venv venv) or Anaconda (conda create -n env_name python=3.10).
• Possibly install pip via conda install pip.
• Install tensorflow via pip install tensorflow or if using a mac pip install tensorflow-macos.
• Install the dependencies with pip install -r requirements.txt.
• Install tensorflow-addons via pip install tensorflow-addons.

2. Download Data (optional)

We provide the open source data to replicate our price, mobility, and solar results in the folder data. However, if you want to replicate our electricity results, you have to download the ElectricityLoadDiagrams20112014 Data Set
from the UCI Machine Learning Repository and save it as elec.csv in the data folder.

Execution

If you are interested in running code, you should navigate to the appropriate pipeline in the pipelines folder and run the respective pipeline from there.

If you are interested in applying our method to your own data, you will need to create a new pipeline. You can use the existing pipelines in the pipelines folder as orientation for any pipeline you create.

Funding

This project is supported by the Helmholtz Association’s Initiative and Networking Fund through Helmholtz AI and by the Helmholtz Association under the Program “Energy System Design”.

License

This code is licensed under the MIT License.