Development of a short-term wind forecasting model for wind farms, based on Time Series, Statistical Techniques and Machine Learning.

Abstract

This degree work presents the development of wind forecasting models based on SARIMA and autoregressive models with application to wind farms. The model was validated for the Villonaco wind power plant, considering among its most important data, the historical data of wind speed, active power and temperature. The Python programming environment was used to process the information and in general for the development of the work. Initially, the data set was extracted to the Python environment, followed by an analysis of the characteristics of the data set, in order to determine the presence of anomalous data and the type of data for each variable. Complementarily, a statistical analysis is performed, which allows to know the behavior of the data and its respective distribution. For the development of the forecasting models, an analysis of series decomposition and the use of the Dickey-Fuller test is carried out, so that an initial panorama of the characteristics of the time series in relation to the base data is obtained, to later carry out the training of the time series based on the SARIMA model and the autoregressive ones, to finally carry out the respective training and forecasting tests, besides defining their accuracy based on the quadratic error index.