在海上风力机设计阶段，通常使用相关海域的统计数据来确定风速工况，以分析风机的气动载荷。然而，这种方法导致设计载荷与实际载荷之间存在不一致性，从而增加了结构疲劳评估的不确定性，给风机结构带来了安全风险。针对上述的问题，本文提出一种海上重力式风机塔筒气动载荷监测方法，通过监测塔筒结构上的应变响应，结合动态载荷识别技术反演气动载荷分量的功率谱密度函数。而载荷识别的数学模型往往受到病态问题的困扰，为了降低或完全消除数学模型的病态问题，本文提出一种测点优化方法-半枚举法，通过优化结构测点位置，进而降低数学模型的条件数，使共振频率处的频响函数矩阵的条件数从1600降低到了28.45。然后，采用优化后的无病态数学模型匹配逆伪激励法识别气动载荷，并分别以原始载荷与识别载荷为输入，分析塔筒底部的疲劳损伤，分析结果表明当监测信号的信噪比为40dB以上时，损伤的相对误差不超过2%。

During the design phase of offshore wind turbines, statistical data from relevant offshore areas are typically utilized to determine wind speed conditions for analyzing the aerodynamic loads on the turbine. However, this approach often leads to inconsistencies between the design loads and actual loads, thereby increasing uncertainty in structural fatigue assessment and posing safety risks to wind turbine structures. To address these issues, this paper proposes a method for monitoring aerodynamic loads on offshore gravity-based wind turbine towers. This method involves monitoring strain responses on the tower structure and employing dynamic load identification techniques to infer the power spectral density function of aerodynamic load components. Nevertheless, the mathematical models for load identification are frequently afflicted by ill-conditioned problems. To mitigate or completely eliminate the ill-conditioned issues in the mathematical models, this paper introduces a semi-empirical method for sensor placement optimization. By optimizing the locations of structural measurement points, the condition number of the mathematical model is reduced, thereby decreasing the condition number of the frequency response function matrix at resonance frequencies from 1600 to 28.45. Subsequently, the optimized non-ill-conditioned mathematical model is utilized with a matching pursuit inverse pseudo-excitation method to identify aerodynamic loads. The original loads and the identified loads are then inputted separately to analyze fatigue damage at the bottom of the tower. The analysis results indicate that when the signal-to-noise ratio of the monitoring signal exceeds 40 dB, the relative error in damage does not exceed 2%.