Posters

Abstract

Accurate long-term yield assessment is critical for operational wind farms, yet significant methodological differences exist between deterministic and probabilistic approaches for post-construction Annual Energy Production (AEP) estimation. This study presents a comprehensive model-to-model comparison of Long-Term Yield Assessment frameworks for operational wind projects (LYTA-o), applying both deterministic and Monte Carlo probabilistic methodologies across multiple offshore and onshore wind farms with available SCADA data. We developed and cross-validated two distinct frameworks for operational yield assessment. The deterministic approach follows conventional industry practice, applying fixed uncertainty factors derived from measured power curves and historical variability to calculate combined uncertainties through root-sum-square methodology. The probabilistic framework employs Monte Carlo simulation, sampling from probability distributions of key input parameters to generate thousands of scenarios, producing full probability distributions of long-term AEP rather than estimates with attached uncertainties. Both methodologies begin with operational losses calculation using SCADA status codes, categorizing each timestamp to determine 100% availability production. This corrected production is then correlated with wind indices derived from multiple long-term reference datasets, including ERA5, MERRA-2, and commercial mesoscale products. We systematically evaluate the impact of different wind index implementations on the resulting P50 and P90 estimates, demonstrating that optimal dataset selection is highly site-specific. A key advantage of the Monte Carlo approach lies in its ability to propagate wind speed uncertainties directly through power production estimates without the need for the turbine sensitivity concept traditionally used in deterministic methods, whereby wind speed uncertainties are converted to power uncertainties via measured power curve gradients. Furthermore, the probabilistic framework naturally accommodates long-term adjustments for operational wear and performance degradation, which the deterministic approach struggles to capture without additional empirical corrections. We quantify the influence of SCADA data duration on uncertainty estimates, analysing how increasing years of operational data reduce the confidence intervals around P50 and P90 values. Results reveal that the associated uncertainty decreases substantially as operational periods extend from one to five years, with diminishing returns beyond this threshold. Losses categorization follows an IEC 61400-15 aligned framework, enabling transparent breakdown of gross-to-net production through standardized losses categories, and facilitating direct comparison between business plan assumptions and actual operational performance across different sites. Cross-validation across the portfolio of offshore and onshore wind farms yields distinct probability distributions of long-term AEP for each methodology. The P90/P50 ratios vary substantially between deterministic and probabilistic approaches, with the Monte Carlo framework consistently providing more realistic confidence intervals that better capture the correlations between uncertainty components often neglected in deterministic calculations. Results demonstrate that pre-analysis for selecting optimal long-term reference datasets is essential, as certain reference sources show superior correlation with actual performance depending on site characteristics. We show that business plan expectations require adjustment according to both reference data duration and quality, with systematic deviations correlating to specific loss categories varying by wind farm type. This study provides practitioners with actionable insights for operational yield assessment methodology selection and demonstrates that probabilistic approaches offer improved uncertainty quantification for both offshore and onshore applications.