One of our roles in the EU IMPROVEMENT EU IMPROVEMENT project is to develop accurate data-driven techniques for predicting the hourly market price of electricity in Portugal. By examining the literature, we have identified a large panoply of studies dealing with the forecast of energy prices. By deeply analyzing them, we realized that they fail to address unstable market conditions. It appears that each uses its own datasets and preparation processes to perform a prediction. To overcome those two major limitations, we propose ten (10) AI techniques for the forecast of electricity prices using the same dataset related to the Portuguese market between 2017 and 2022. Our techniques are belonging to statistical, Machine Learning (ML), and Deep Learning (DL) approaches, at several temporal granularities (hourly, daily, weekly, and monthly). The 10 techniques are implemented, tested, and evaluated with the MAE, RMSE, and CV metrics.
All algorithms have been implemented using Google Collab notebooks currently running :
| Tool | version |
|---|---|
| Python | 3.7.13 |
| tensorflow | 2.8.2 |
| keras | 2.8.0 |
| statsmodels | 0.13.2 |
| pandas | 1.3.5 |
| matplotlib | 3.2.2 |
| sklearn | 1.0.2 |
| numpy | 1.21.6 |
| seaborn | 0.11.2 |
| scipy | 1.7.3 |
This repository contains the code of the algorithms used to implement electricity price forecasting as described in in the Paper - AI Approaches for Electricity Price Forecasting in Stable/Unstable Markets: EU Improvement Project. We implemented and evaluate three classes of AI techniques: statistical (STAT), Machine Learning (ML), and Deep Learning (DL), and in each class we consider the following methods: (i) STAT: ARIMA, HoltWinter, (ii) ML: Multiple Linear regression, SVM, Extreme Gradient Boosting, Random Forest and (iii) DL: LSTM. All algorithms can be found in the src file.
The electricity prices of the Portuguese market are loaded from the Redes Energ ́eticas Nacionais (REN) website.
The hourly generation by source data are gathered from the ENTSO-E Transparency platform.
The data preprocessing we applied can be found here.
Here are all our results obtained with the differents algorithms using the entire dataset or only the 2017/2021 part.
Details the license agreement of StableUnstable_ElectricityPriceForecasting: LICENSE
- Florian CHAUVET (core developer)
- Ladjel BELLATRECHE