Posters

Abstract

Abstract summary Data quality should remain a critical concern in the accurate analysis of wind turbine performance, especially when using SCADA data. Despite the current landscape of "data-driven decision-making," there is often a prevailing assumption that even unfiltered data can yield valuable insights when processed through seemingly magical "black-box" models. This study introduces a simple and pragmatic methodology designed to detect North offset biases in large wind turbine fleets. Developed to function across a broad range of turbine types, the algorithm utilizes data from neighboring turbines within the same wind farm and across one's portfolio. Currently operational for a significant portion of the i4SEE fleet, this algorithm provides monthly updates to account for potential temporal shifts due to human intervention, software changes, or other variables. By correcting these offsets, we lay the foundation for more reliable, future data analyses, thereby addressing the long-standing "garbage in, garbage out" issue. Method We developed an algorithm capable of detecting the North offset for individual wind turbines by leveraging data from neighboring turbines within the same wind farm and across our i4SEE portfolio. This approach is augmented by weather information sourced from third-party providers. Importantly, the algorithm is transparent and deliberately designed to be understandable and replicable by experienced data analysts. It has undergone rigorous validation across a diverse range of wind turbine types in our i4SEE fleet to accurately identify the "North offset." In the study, we will delineate the assumptions underpinning this algorithm, providing insights into its current limitations and avenues for future improvements. Results Our specialized algorithm consistently identifies varying North offsets across the extensive i4SEE fleet. It is designed to offer monthly monitoring to account for any temporal changes caused by human intervention, software updates, or other variables. Through this dynamic tracking, we have flagged turbines with multiple North offset changes over time, thereby enabling more accurate and fine-tuned data corrections for future analysis. For each wind turbine owner, we provide a specific corrective coefficient, giving priority to turbines that are most severely affected by North offset inaccuracies. With turbine owners' consent, we aim to present a comprehensive statistical analysis that categorizes all the calculated offsets by both manufacturer and specific turbine type. Conclusions The algorithm we've developed is a reliable mechanism for automated North offset detection across large wind turbine fleets. It is currently running in production for a substantial part of the i4SEE portfolio. Although North offsets do not usually affect wind turbine performance, they can, in specific cases, alter turbine operations. Consequently, their correction is not a trivial matter but a requisite step for guaranteeing the validity of both current and future data analyses. Our algorithm serves as a real-world, pragmatic solution to the long-standing "garbage in, garbage out" problem inherent to data analysis. Learning Objectives (required) Upon reviewing this work, delegates will gain valuable insights into the issue of North offset bias and its potential impact on both wind turbine data analysis and actual performance.