DYNAMIC BAYESIAN NETWORK FOR WEATHER FORECAST AND EVALUATION OF RENEWABLE RESOURCES AVAILABILITY

摘要

The authors present a Bayesiannetwork capable to estimate the weather parametersrelated, not only, to renewable resources: wind speedand solar irradiation. A large and systematic database about simple and composed weather indicesregistered during four years, 2013-2016, was used toconstruct the data-driven Bayesian structure and tolearn and validate its parameters. It includes 9weather indices collected, minute by minute, by aprofessional Davis Instrument Pro 2 Plus weatherstation. The extremely large initial data base, over 1.8million records, was discretized in 4 classes makingpossible to use a very simple algorithm like Bayesiansearch to establish the most suitable networkstructure fitting the data. The main and first usefulresults mean the probability of wind speed and solarirradiance classes. Both parameters can betransformed in electrical power considering a givenwind generator and a solar panel.