A Probabilistic Baseline Learning Framework for SCADA-Based Wind Turbine Aging Diagnosis and Multi-Scale Performance Degradation Analysis

Abstract

Accurate diagnosis of wind turbine aging from SCADA data is essential for reliable long-term operation, yet conventional
wind-speed-binning methods often fail to capture the nonlinear and condition-dependent nature of power degradation. To
address this issue, this study proposes a probabilistic baseline learning framework for SCADA-based wind turbine aging
diagnosis and multi-scale degradation analysis. A two-stage mean–variance XGBoost model with uncertainty calibration
is developed to estimate both the healthy-reference power output and its predictive uncertainty under varying operating
conditions. The deviation between measured power and the probabilistic healthy baseline is then used to quantify degrada-
tion across temporal, wind-speed, directional, and joint windspeed–direction dimensions. The results show that turbine
aging is cumulative but strongly condition-dependent, with the most severe degradation concentrated in specific operating sec-
tors rather than uniformly distributed across all inflow states. Furthermore, some directional degradation hotspots appear
to correspond to potential upstream turbine positions, suggesting a possible wake-related contribution, although further
verification is still required. More importantly, the proposed framework not only improves aging diagnosis accuracy, but
also provides wind farm operators with interpretable information on the conditions under which degradation is most severe,
thereby supporting targeted inspection, maintenance planning, and future evaluation of wake-aware mitigation strategies such
as active wake steering.