| dc.description.abstract | Predicting the weather is important for a lot of fields including agriculture, construction and
hydro-power and flood management. Currently mechanistic meteorology predictions are
generated using heavy computing based 3D Navier-Stokes models. Therefore, it is of interest to
develop models that can predict weather conditions faster than traditional meteorological
models. The field of machine learning has received much interest from the scientific community.
Due to its applicability in a variety of fields, it is of interest to study if the use of artificial neural
networks can be a good candidate for prediction of weather conditions. Machine learning
methods benefit from large datasets. A python interface was developed to make it easier to
obtain weather data from free sources, the python interface works well, but is more user-friendly
when used with Met supplier compared with Netatmo supplier. Four separate models where
trained to predict the temperature 1, 3, 6 and 12 hours ahead. In the first experiment, only
temperature was used as input to the networks. This constitutes an auto-regressive neural
network(ARNN). In the second experiment, precipitation data was introduced into the network,
forming an autoregressive neural network with exogenous inputs (ARX-NN). The results show
that the inclusion of precipitation had a negligible effect on accuracy for temperature prediction.
Out of the four model types, 1-hour prediction has the best prediction results for both the ARNN
and the ARX-NN. | nb_NO |
