Remote Fault Detection of Anemometers Using Machine Learning on Simulated Wake Models

Wind energy is one of the most prominent forms of renewable energy currently being implemented to decrease our reliance on fossil fuels and curb the emission rates of greenhouse gases. The growth of wind energy is challenged by the unpredictability of the wind and the power fluctuations accompanying variable resources. To account for the wind speed fluctuations, wind turbines have control systems to optimize the power generated. Turbine controllers are dependent on accurate measurements of the wind speed, usually made by devices called anemometers. Faults in anemometers can lead to suboptimal turbine performance and additional mechanical strain on the turbine. There are a wide range of methods employed towards the fault detection of anemometers, including data driven approaches, machine learning techniques, set value observations, and parameter estimation models. This paper proposes using knowledge of wind field and wake interactions at an entire wind farm to detect faults in anemometers by applying machine learning to simulated wake fields of wind farms. This approach was able to detect artificial errors of 5% and greater in wind speed readings with an accuracy of 94.1%.