Modelos avanzados para la predicción a corto plazo de la producción eléctrica en parques eólicos

Resumen

The penetration of wind farms into the power systems (nearly 9% of the electric energy consumed in Spain in 2006 was generated in wind farms) requires effective short-term forecasting tools of the electrical production in these plants. These tools are needed by the electric network Operator and by the managers of wind farms. The electric network Operator needs to know beforehand the value of the electric power generation and demand that is going to exist in each one of the buses of the System to obtain the correct operation. The manager of a wind farm needs to predict the hourly electric power generation generated in the wind park, to sell the energy produced to the electrical market. By means of the application and development of suitable tools, based on meta-heuristic methods known as "soft-computing" techniques, and using real data of wind parks, a set of new forecasting models have been developed to be used for a variety of cases with different types and amounts of data available. Also a global forecasting system has been developed, including a set of models, each of which is specialized in a range of the forecasting horizons, and obtains a total range from 0.5 to 72 hours. The global forecasting system uses sophisticated techniques such as a new Kalman filter to improve the predictions of the average wind speed provided by a numerical weather prediction model; such as the combination of the forecasts from different models by means of fuzzy inference systems; and the selection of the suitable forecasting model based on the moment in which the forecasts are carried out. The performance of all the developed models has been evaluated by comparing the results obtained in the prediction of the electrical production, with those from other models described in international literature, using always the same data. In all the cases, the results obtained with the new forecasting models are better than those obtained with the rest of the models.