Ensemble machine learning model for forecasting wind farm generation

Abstract

The use of renewable energy sources is the only way to avoid emissions from energy production and to decide pollution of ecology. Despite the fact that renewable energy has become clean energy, which called green energy, the issue of using it is quite difficult for the control and regulate the energy system. Efficient use of renewable energy requires information on primary sources. If looking case of large-scale integration, this requirement will be significantly felt. Thus, to ensure the normal operating modes of the energy system, it is necessary to predict the generation of renewable sources with an acceptable error. PURPOSE. To forecast the generation of wind farms. METHODS. This study is carried out by ensemble algorithms, such as Random Forest, AdaBoost and XGBoost, which are one of the machine learning approaches. The software implementation is made using the Python programming language. As initial inputs historical data on windspeed and generating of some windfarms in Mongolia by 2019-2021 were used. RESULTS. The proposed method predicted daily production schedules at three wind farms with an error of 2.4 to 3.4 MW or 5.0 to 7.0 percent of the installed capacity of the corresponding wind farm. Also, normalized MAE was 12,3 to 13.3 percent. CONCLUSIONS. Ensemble methods of machine learning made it possible to determine non-linear and non-stationary dependencies of the time series, and also can be implemented in the problem of predicting the daily production schedule. Increasing the accuracy of wind energy forecasting will affect positively the operation and planning of the power systems.

Keywords

• forecasting
• generate
• windspeed
• ensemble method
• wind farm