Forecasting for Wind Farm Energy Output in South Australia: A Comparative Analysis of Physical Methods and Deep Learning Methods

Abstract

To achieve the target of carbon zero” in 2050, the Australian government advocates the development of renewable energy technology to reduce CO2 emissions. Particularly, wind energy resources are rich in South Australia. With the development of wind farms, it is necessary to predict the energy output for the electricity market. This study compared two different methods for forecasting the wind energy output monthly. The first method is the physical method, using predicting weather data from Medium-Range Weather Forecasts (ECMWF). Another method is RNN-LSTM (Recurrent Neural Network-Long Short-Term Memory) by using Python to predict energy output. The result showed that the physical method can predict the trend of energy output value while RNN-LSTM is not suitable for monthly forecasting. This study proved that the deep learning methods should be utilized in the site that have numerous numbers of data resources. And it is better to use physical methods which consider the atmosphere, local terrain, and wind farm layout for wind farm energy outputs forecasting.