If you want to reproduce the results, you need to gather the data (see Data Availability). A self-contained example using synthetic data is contained in the synthetic_example_gbt.ipynb notebook.
The data that was used is publicly available via the following links:
- ENTSO-E https://transparency.entsoe.eu/
- carbon prices https://carbonpricingdashboard.worldbank.org/compliance/price
- gas prices https://fred.stlouisfed.org/series/PNGASEUUSDM
- oil prices https://fred.stlouisfed.org/series/DCOILBRENTEU
- temperature https://open-meteo.com/en/docs/historical-weather-api
The Shapley flow implementation is available at https://github.com/nathanwang000/Shapley-Flow
The increasing complexity of modern machine learning models presents a challenge to their interpretability. In response, the field of Explainable Artificial Intelligence (XAI) has emerged as a means of making these so-called black-box models explainable and providing insights into the relevance of input features. Previous research in the energy system domain has extensively utilized Shapley additive explanations (SHAP), although its inherent feature independence assumption can lead to difficulties when explaining models that rely on data that violates this assumption. Consequently, several approaches based on Shapley values have been proposed with the aim of incorporating causal knowledge and improving SHAP. As accurate electricity price forecasting is crucial for maintaining the stability of the electricity grid, we develop machine learning models based on neural networks and gradient boosted trees for predicting day- ahead electricity prices in France based on a set of publicly available techno-economic features. We generate explanations using the novel Shapley flow approach, through which we uncover hidden relationships that are not visible with the commonly employed SHAP framework.