Wind Power Forecasting Using Machine Learning: State of the Art, Trends and Challenges

The future challenges in the power grid have become more real the last decade. The wind power production increases rapidly. Having compatible wind turbines in the electricity spot market, forces conventional powerplants to shut down. This affects the reserve markets whose cost increases as the wind power capacity grows. By having wind turbines participate in the reserve markets, the costs could be reduced. Wind turbines are now excluded from the Danish markets due to low reliability of forecasts. Wind power forecasts must reflect the reality if the TSOs are to rely on the availability of the wind turbines. A State-of-the-Art analysis of four machine learning methods, Neural Network, Support Vector Machine, k Nearest Neighbor and Random Forest, investigates the challenges and advantages of the algorithms within wind power forecasting. The State-of-the-Art results showed that Neural Network and Support Vector Machine are the most common algorithms within the field. By investigating the algorithms, it was found that Neural Network and Support Vector Machine have several parameters, which will increase errors, if tuned poorly. Further it was found that due to the many parameters, the algorithms can be modified to fit many specific cases. There is a growing trend in the general use of machine learning in order to digitalize wind power forecasts. A more stable, automatic, and human-error-free prediction of wind power will bring wind turbines one step closer in participating in the reserve markets.