A comparative analysis of intelligent techniques to predict energy generated by a small wind turbine from atmospheric variables.

The article focuses on predicting the short-term power generation of a small wind turbine installed in a bioclimatic house in northwest Spain using atmospheric and operational data collected over one year. It analyzes 19 atmospheric variables through principal component analysis, identifying five key factors—wind strength, temperature, solar power, wind direction, and atmospheric pressure—with wind strength being the most influential on power output. Various regression techniques, including regression trees, support vector regression (SVR), multilayer perceptron (MLP), self-organizing maps (SOM), and Cubist models, were tested for prediction accuracy, with the Cubist method consistently outperforming others. The study concludes that accurate short-term power prediction is feasible and suggests future work incorporating temporal data to improve model performance, which could benefit smart grid energy management.