Wind turbines are increasing in popularity as a power source around the world, and are among the cheapest sources of energy. They introduce problems related to the volatile nature of the wind, and the uncertainty of the timing and amount of energy production. Current methods generally use a physical or statistical approach for predicting power generated from wind turbines. Our thesis explores combining these approaches for a hybrid approach, using computational fluid dynamics as the physical approach, and machine learning models for the statistical approach. Six different machine learning models have been tested, resulting in six different hybrid approaches. The data consisted of historical wind condition forecasts and the historical power production, for five different wind farms in Norway. The output was the predicted power production for the next 24 hours, given in hourly measures of megawatt hours. The hybrid approaches are tested against the standard physical and statistical approach, and their performance is measured and compared. The results show that for four of the five wind farms, the hybrid approach outperformed both the statistical and physical approaches.