Yiming Ke

Precise measurement of the structural loads of wind turbines is crucial for condition monitoring and failure prevention; however, traditional methods often rely on expensive strain sensors or complex physical models. This article proposes a new framework for soft measurement of wind turbine structural loads using multidimensional time-series data measured in wind power supervisory control and data acquisition (SCADA) systems, combined with variable importance-based dimension reduction and enhanced meta-learning. First, a fusion model combining a global time-series variable attention mechanism (VAM) with a residual network (RN) is designed to achieve dimension reduction of time-series data while preserving the physical coupling characteristics of the data. Second, a regression model stacked with CatBoost and multilayer perceptron (MLP) is constructed to improve the estimation accuracy of the source wind turbine structural loads. Crucially, in response to the challenges brought by the diverse types of wind turbines and inconsistent measurement data space in actual application environments, an enhanced meta-learning framework is proposed to achieve multitask evaluation for different wind turbines. The experimental results show that compared with other methods, the proposed method significantly improves the load estimation accuracy of wind turbines and performs well in noisy and small-sample migration scenarios. It has the potential to provide a general solution with high accuracy, high robustness, and low data dependence for the structural load assessment of wind turbines.