Posters

Abstract

Within the wind turbine industry, curtailment controls are more and more common. These rely often on curtailment schemas implemented in the turbine software by the service providers of the original equipment manufacturer (OEM). Errors and low response time are common between the turbine's operator and the OEM which can lead to production losses and delays. Furthermore, operators strive to assess curtailment losses with the utmost accuracy. This task proves challenging as certain curtailments lack clear labels in the Supervisory control and data acquisition (SCADA) alarm list. Challenges emerge when various types of curtailments (bat, noise, or grid) are scheduled concurrently, introducing additional complexity into the calculation of production losses, particularly when the specific curtailment applied is unknown. Ultimately, the available data may vary from site to site, and there is a possibility that only 10-minute aggregated data is available. This circumstance necessitates approximations in the annual energy production (AEP) loss calculation process. This presentation introduces a new approach to enhance AEP calculation specifically for noise curtailment events when limited to 10-minute data. Additionally, it outlines a methodology to detect abnormal curtailment behaviors through the utilization of a curtailment plan. The method can be easily extended to other types of curtailment such as bat, grid, or wind sector management. The analysis incorporates machine learning techniques to forecast the turbine's specific power curve. This model is then leveraged for precise AEP loss calculations, contributing to a more nuanced understanding of the turbine's performance under varying conditions. Using synthesized data, the demonstration reveals that depending on 10-minute averaged data for AEP loss calculation may introduce significant discrepancies compared to higher frequent data such as 1 second data, emphasizing the potential for notable variations. Furthermore, the findings highlight the necessity of considering external factors, such as air density, in the computation of losses attributed to curtailments. Employing a statistical approach, significantly enhanced the precision of AEP loss calculations within the context of 10-minute aggregated data, resulting in a clear improvement. This underscores the importance of incorporating comprehensive external factors for a more refined analysis of curtailment-related losses. The presentation introduces a robust approach to ascertain whether turbines adhere to the communicated curtailment plan from the original equipment manufacturer (OEM). By intentionally introducing variations into the data, the model effectively pinpoints abnormal data instances that warrant further investigation. In conclusion, the presented analysis offers a valuable tool for producers seeking to enhance AEP loss calculations during curtailment events through the utilization of machine learning models with 10-minute aggregated data. Additionally, the presentation introduces a solution for promptly identifying any deviations from the curtailment plan, empowering producers to respond swiftly and effectively.