Posters

Abstract

Calculating wind farm wake losses has applications in yield assessment, control, and wind farm design. These applications demand wake models that are both fast and accurate, typically achieved with analytical formulations. Unlike higher-fidelity approaches, analytical wake models require calibration. Calibration against SCADA data leverages field data that capture site-, farm-, and turbine-specific effects, but it also presents challenges due to sensor uncertainty and the stochastic nature of wind. A robust calibration framework is therefore needed.  When calibrating wake models against SCADA data, several methodological choices arise. Calibration may be performed using time-series data or binned data. The target variable may be active power or hub-height wind speed, and selecting an appropriate cost function is non-trivial. Other factors also influence outcomes, notably turbulence intensity: should it be calculated using IEC standards, which model freestream turbulence intensity as a function of freestream wind speed, or derived from measurements of the freestream turbines? A further challenge is whether the estimated freestream wind conditions truly reflect farm-wide conditions or should instead be treated as uncertain. This work outlines best practices and common pitfalls in wake model calibration using SCADA data from multiple wind farms, and addresses the following questions: How do results vary with different cost functions and target variables? What is the recommended approach for modeling freestream turbulence? Should the model freestream wind speed and wind direction be included in the calibration, and what is the impact of neglecting this step?    This work also examines the generalization of calibration outputs. Time-series calibration yields time-series outputs, whereas binned calibration yields bin-averaged outputs. Methods are proposed to generalize these outcomes and to handle lower-quality data.