PUC-Rio's Time Series (IND2616) project on wind speed prediction

PUC-Rio IND2616

Data

We analyse two time series of wind speed in two locations with relevant eolic energy production in Brazil. The data was obtained on a software, designed by the enterprise PSR, called Time Series Lab (TSL). The data consistis on a hourly dataset from ERA5. We evaluate the average of the daily wind speed and consider daily data on our approach. We analyse 5 years, from 2016 to 2021, a total of 1825 days.

The first location represents a wind farm, called Usinas Eólicas São Fernando 1, 2, 3 e 4, in the state of Rio Grande do Norte. In 2020, Rio Grande do Norte was the second state with the highest eolic energy generation in Brazil.

The second location represents a wind farm, called Usina Eólica Elebrás Cidreira 1, in the state of Rio Grande do Sul. In 2020, Rio Grande do Sul was the fifth state with the highest eolic energy generation in Brazil.

Both states were chosen not only because of the significant contribution to the Brazilian electricity mix but also because they are located in different parts of the country. The local in Rio Grande do Norte in the littoral of the Northeast region of Brazil while the local in Rio Grande do Sul is in the littoral of the South region.

Objective

We aim to realize time series predictions. For that, we laverage on three methods, and we want to compare the predictions with each of them.

1) SARIMA models: from Box & Jenkins, the classic models;

2) Vector Autoregressive Model (VAR): it is a multivariate model that analyse more than one time series at the same time;

3) Neural Networks: flexible blackbox models.

Moreover, for each location and each method, we make predictions for two horizons: 14 days ahead and 122 days ahead.

Files

conda create -p .venv python=3.10 # must be compatible with dependencies such as pytorch
# from https://pytorch.org/
conda install pytorch pytorch-cuda=11.7 -c pytorch -c nvidia # GPU

conda install pandas seaborn
conda install -c anaconda scikit-learn
conda install scikit-learn-intelex

conda install -c conda-forge pytorch-lightning
conda install -c conda-forge pmdarima # `diff` and `diff_inv`

pip install reorder-python-imports
conda install -c conda-forge black