The source code of the paper is named as: Solar Irradiance forecasting using dynamic ensemble selection. Four data sets from Salvador, Fortaleza, Florianópolis and São Paulo are considered, which are disponible at: https://portal.inmet.gov.br/dadoshistoricos

General information:

• There is a jupyter notebook to execute every single model (TODO: ARIMA execution);
• After executing all single model, is possible to use the notebook called by heterogen_ensemble.ipynb, to create the Dynamic Selection experiments;

Files information:

• The file models_configuration_60_20_20.json is responsible to configure the time series paths, test and validation splits, series names, and time series codes;
• All time-series data are inside of ./data/;
• All models execution will be persisted in ./solar_rad/;
• The ARIMA model are persisted in ./arima_models/;

Python files description:

• fit_predict_models.py: responsible for training and predicting scripts for Sklearn based models;
• heterogen_ensemble.py: responsible for creating dynamic selection approaches;
• time_series_functions.py: scripts for persisting and load results, and compute performance metrics;