Meteorological Subset Selection for Multidimensional Wind Speed Prediction

The global demand for a clean, green environment and sustainable energy supply led to the rapid rise of renewable energy such as wind resources. Nevertheless, plant managers and operators face the significant challenge of accurately forecasting wind power which is a crucial aspect for wind energy generation. The amount of wind power generated is highly dependent on wind speed. As a result, accurately quantifying wind speed predictions is highly essential for the wind energy production process. Meteorological features such as wind direction, temperature, relative humidity and others, are recognized by researchers as contributors to the unexpected fluctuations in wind speed, hence including these features as wind speed prediction model inputs can improve predictions. This study aimed to determine, identify and select the relevant meteorological features influencing wind speed, with the goal of integrating them as model inputs, to reduce dimensionality. ReliefF feature selection was applied as an inherent preprocessing stage to determine the feature subset that demonstrate a noteworthy correlation to wind speed. Findings revealed that in both experimental datasets employed, two features had demonstrated minimal impact on variations in wind speed. Identifying and incorporating only the meteorological subset that exhibit substantial correlation with wind speed reduces complexity and enhances model performance.