GreenCurve

GreenCurve is a machine learning tool designed to predict the 24-hour renewable energy production curve for the upcoming day using historical data from US and ERCOT sources. Leveraging multiple forecasting models—including Facebook's Prophet, SARIMAX, XGBoost, and a moving average approach—GreenCurve provides robust predictions and insights into future energy trends. It also includes visualization tools to compare model forecasts against real test data.

Table of Contents

• Features
• Installation
• Usage
• Configuration
• Model Overview
• Examples
• Contributing
• License
• Contact

Features

• Multi-Model Forecasting: Predict renewable energy production using several methods (Prophet, SARIMAX, XGBoost, and moving average) for robust and comparative performance.
• Data Loading & Preprocessing: Built-in functions to load, clean, and preprocess historical energy data from US and ERCOT sources.
• Time Series Formatting: Automatically format your data for time series analysis, handling resampling, imputation, and feature engineering.
• Visualization: Generate detailed plots comparing forecasts from different models against real energy data, including error metrics such as RMSE and MAE.
• Easy Integration: Use GreenCurve as a standalone tool or as a module within your Python projects.

Installation

Ensure you have Python 3.7 or later installed. Clone the repository and install the required dependencies:

git clone https://github.com/yourusername/GreenCurve.git
cd GreenCurve
pip install -r requirements.txt

Run the test suite to ensure everything works:

pytest -s tests/test.py

Usage

GreenCurve can be used both as a standalone script and as an importable module. Example usage:

from greencurve import predict_energy_curve

# Predict the renewable energy production curve for a given day
forecast = predict_energy_curve(
    country="US",
    extra_data={},         # Optionally provide override data
    current_date="2024-01-12",
    days=1,                # Forecast for 24 hours (1 day)
    plot=True              # Display forecast plots
)

print(forecast.head())

Configuration

GreenCurve offers flexibility in configuring forecasting parameters:

• Training Period: Set the number of days used for training (history_days).
• Forecast Horizon: Define the number of forecast steps (e.g., forecast_steps = 24 for a 24-hour forecast).
• Model Tuning: You can adjust parameters for each forecasting method directly in the code:
  • Prophet Parameters: Configure changepoint_prior_scale, seasonality_prior_scale, and Fourier orders.
  • SARIMAX Parameters: Set order (p, d, q) and seasonal_order (P, D, Q, s).
  • XGBoost Parameters: Customize n_estimators, learning_rate, etc.

You may also manage configuration through external config files if needed.

Model Overview

GreenCurve integrates multiple forecasting approaches to improve prediction robustness:

• Prophet: Designed for time series data with strong seasonal effects. Detects changepoints and captures seasonality.
• SARIMAX: Seasonal ARIMA with support for exogenous variables. Suitable for trending and seasonal time series.
• XGBoost: Gradient boosting with engineered time-based features. Captures nonlinear relationships.
• Moving Average: Baseline method using historical hourly averages.

Each model's predictions are evaluated using:

• RMSE (Root Mean Squared Error)
• MAE (Mean Absolute Error)

This multi-model strategy enables performance comparison and model selection.

Examples

Coming soon.

License

This project is licensed under the MIT License.

Contact

For inquiries, please contact: nathalia.wolf@inria.fr