Colloquia

Summary

According with the International Energy Agency (IEA), wind power generation has increased worldwide by 17% in 2021 compared to the previous year, the highest growth among renewable sources. To further increase energy production, countries with high wind power potential started investing in this type of energy. However, the wind potential often goes with harsh climate conditions, and thus one of the main issues in generating this type of energy is the process of the formation and accretion of ice in the blades. In the presence of ice, the shape of the turbine's blade changes, consequently decreasing power generation. Next, the ice may cause mechanical damage to blades, lowering their life span and leading to potential safety risks such as ice throwing and ice falling near urban areas. Therefore, during an ice event, the usual practice is to hold up the energy generation by temporarily shutting down the affected turbines. This action, however, may have a significant effect on the market side. The non-delivered power from the wind turbines may cause imbalances in the trading operations and thus may increase the energy price tremendously. In this aspect, correctly anticipating icing events from safety and financial savings is paramount. The present work aims to use Machine Learning (ML) techniques to predict the availability of a turbine due to ice events based on the local atmospheric conditions. The atmospheric phenomena are analyzed to extract the most significant environmental variables that cause the ice events, which is done with the help of the ERA5 reanalysis model that provides global, hourly estimates of atmospheric variables at a horizontal resolution of 31km and 137 vertical levels from the surface to 0.1hPa. Once the corresponding variables from this model are selected, they are further used as input features for the ML model. Finally, a prediction model using the data from the Numerical Weather Predictions (NWP) from Global Forecast System (GFS) weather models are inputs to an ML model to predict ice events.