French Energy Consumption Forecasting

This project explores statistical models for analyzing the energy consumption in France and improving peak demand management. To do so we focused on seasonality and trend identification and volatility forecasting. We use autoregressive models (ARIMA, GARCH) and Copula-based methods to model temporal and spatial dependencies, providing both short-term forecasts and insights into inter-regional correlations.

Data

This project uses publicly available datasets from the French government:

• https://www.data.gouv.fr/fr/datasets/consommation-quotidienne-brute/
• https://www.data.gouv.fr/fr/datasets/consommation-quotidienne-brute-regionale/

The dataset is provided as a .zip file in the repository. Before running the notebooks, make sure to unzip it: unzip consommation-quotidienne-brute-regionale.zip

Table of Contents

1. Analysis using ARIMA Models
2. Analysis using GARCH Processes
3. Modeling Regional Dependence with Copulas

Analysis using ARIMA Models

In the arima.ipynb notebook, we focused on time-series forecasting of energy consumption using ARIMA-based models.
After removing long-term trends and seasonal components, we fitted and compared several ARIMA and SARIMA configurations to capture short-term dynamics in the data.

The goal was to obtain a model that performs well both in-sample and out-of-sample, while maintaining interpretable components for trend, seasonality, and residual noise.

Key Steps:

• Loaded and cleaned the daily national energy consumption dataset.
• Decomposed the series into trend, seasonality, and residual components.
• Performed ACF/PACF analysis to identify suitable ARIMA parameters.
• Compared ARIMA and SARIMA models on multiple forecasting horizons.
• Evaluated model performance using AIC, BIC, RMSE, and diagnostic plots (residual normality, autocorrelation, etc.).
• Produced short-term forecasts of French energy consumption and visualized them against observed data.

Even though the forecasting using solely this anaylis isn't great, we then reemployed this baseline for the volatility modeling performed with GARCH and the overall forecasting.

Analysis using GARCH Processes

ARCH and GARCH Models

In the garch.ipynb notebook, we employed ARCH (Autoregressive Conditional Heteroskedasticity) and GARCH (Generalized ARCH) models to forecast volatility. The GARCH(1,1) specification yielded significant results, confirming that current volatility is influenced by past returns. To capture asymmetric effects, particularly the heightened impact of sharp increases in energy consumption, we further applied more advanced models such as EGARCH (Exponential GARCH), which account for differential responses to positive and negative shocks.

McLeod-Li Test

We performed the McLeod-Li test to detect the presence of ARCH effects in the residuals, with results suggesting significant effects, leading us to continue our analysis with the GARCH model.

Summary of the work done:

• Loaded and preprocessed energy consumption data.
• Investigated different models to separate volatility from trend and seasonality.
• Fitted GARCH-based models to the volatility.
• Forecast energy consumption using trend, seasonality and volatility forecast.
• Evaluated our models using AIC, BIC, QQ-Plot, McLeod-Li Test

Modeling Regional Dependence with Copulas

The copulas.ipynb notebook applies copula models to analyze inter-regional dependencies in energy consumption. Copulas enable us to capture complex, non-linear relationships between variables that go beyond what traditional correlation measures can reveal. We explore several copula types (Gaussian, t-Copulas, Gumbel,...) to assess joint behavior across regions. These models are particularly useful for managing peak consumption scenarios and guiding strategic decisions on prioritizing energy distribution.

Summary of the work done:

• Loaded and preprocessed regional energy consumption data.
• Investigated correlation patterns between selected regions with varying climatic characteristics (both geographically close and distant).
• Selected and fitted appropriate copula models to capture interdependencies.
• Analyzed and visualized the resulting dependency structures to interpret joint behavior and inform energy distribution strategies.

Structure of the Repository

├── arima.ipynb                                   # Time series modeling with ARIMA/SARIMA
├── consommation-quotidienne-brute-regionale.zip  # data in a .zip file
├── copulas.ipynb                                 # Regional Dependence modeling with Copulas
├── dataset_exploration.ipynb                     # Data exploration and visualization
├── figures
│   ├── garch_forecast.png
│   └── occ_hdf_copula.png
├── fr.json                                       # GeoJSON for map plotting
├── garch.ipynb                                   # Volatility modeling using GARCH
├── README.md
└── requirements.txt