Analysis of a Wind Speed Prediction Model Based on Decision Trees

Predicting wind speed is crucial for efficient wind farm management, aiming to optimize en-ergy production and ensure grid stability. This paper addresses wind speed prediction using data from a LIDAR sensor at five measurement points at different heights (40 to 260 meters). A model based on decision trees, specifically the Random Forests (TreeBagger), has been applied to predict wind speed at 10 to 60-minute intervals. The model's performance has been evaluated using the Root Mean Square Error (RMSE), Mean Absolute Percentage Error (MAPE), and Coeffi-cient of Determination (R2) indices. The point closest to the coast (PO) yielded the best results, with RMSE ranging from 0.15 to 0.22 m/s, MAPE ranging from 0.69% to 1.51%, and R2 of 0.99. As the measurement points moved away from the coast, errors increased slightly due to variations in temperature and surface roughness. Data collection at 10-minute intervals provided a robust model training and evaluation dataset. The evaluation metrics reveal the quality of predictions regarding accuracy and generalization ability. The adopted approach provides valuable information for improving wind en-ergy prediction under specific conditions, as found at the investigated site.