Surrogate‐Accelerated Optimization of Virtual High‐Mass ATMD for Multifrequency Vibration Control in 15‐MW Wind Turbines

The upscaling of wind turbines to ultra‐large capacities, such as 15 MW, introduces important structural control challenges because the tower natural frequency may interact with the dominant blade‐passing (3P) frequency, reducing the effectiveness of conventional passive tuned mass dampers (TMDs) under detuned operating conditions. To address this issue, this paper proposes an active control strategy in which a virtual tuned mass damper (VTMD) is realized by a lightweight active tuned mass damper (ATMD). This control strategy is further integrated with a surrogate‐assisted optimization framework based on Gaussian process regression (GPR). The proposed control law enables the lightweight actuator to approximate the inertial effect of an idealized high‐mass VTMD, thereby broadening the effective vibration suppression bandwidth. To reduce the computational cost of tuning this high‐dimensional nonlinear system, the GPR surrogate is trained and validated using high‐fidelity OpenFAST simulations. The surrogate model captures the nonlinear response characteristics and the feasibility boundary associated with the actuator stroke constraint, achieving a prediction error of less than 0.7% with a limited training dataset. Based on the surrogate‐assisted optimization framework, the optimized ATMD reduces the root mean square (RMS) tower‐top acceleration by 37.70% and the peak acceleration by 36.77% under the selected operating condition, whereas the passive TMD increases the RMS response by 5.32%. Once trained, the surrogate evaluates candidate parameter combinations within milliseconds, whereas each high‐fidelity OpenFAST simulation requires approximately 20 min, indicating its potential value for rapid parameter assessment and future adaptive‐control studies of ultra‐large wind turbines.