针对海上漂浮式风力机在长期的海洋环境作用下，受到风、浪、流等复杂载荷的影响，发生腐蚀、蠕变和失效等故障问题，基于深度学习理论，提出了一种改进的完全集合经验模态分解（Improved Complete Ensemble EMD，ICEEMDAN）结合卷积神经网络（Convolutional Neural Network， CNN）的故障诊断方法，用于对海上漂浮式风力机的系泊系统进行故障识别。该方法基于平台艏摇响应信号状态，计算系泊蠕变与失效阶段，并分析不同位置系泊对漂浮式风力机稳定性的影响，诊断出系泊是否产生蠕变以及系泊蠕变位置。研究结果表明：改进后的方法能够较好地识别系泊蠕变到失效过程，挖掘了纵荡、横荡、横摇及艏摇等因素对风力机稳定性的影响，其在不同信噪比下均可有效地诊断出系泊状况与不同位置的蠕变，且准确率最高可达99.83%。

In the longterm presence of the ocean environment, floating offshore wind turbines (FOWT) are subjected to complex loads such as wind, waves, and currents, which may result in fault problems such as corrosion, creep and failure. To address this issue, this study proposes an improved complete ensemble empirical mode decomposition with adaptive noise (ICEEMDAN) combined with convolutional neural network (CNN) method based on deep learning theory for fault diagnosis of the mooring system of offshore floating wind turbines. The method calculates the creep and failure stages of the mooring system based on the platform′s yaw response signal status, analyzes the impact of mooring at different positions on the stability of floating wind turbines, and diagnoses whether creep occurs in the mooring system and its location. The research results show that the improved method can effectively identify the creep to failure process of the mooring system, explore the impact of surge, sway, roll and yaw factors on the stability of wind turbines, and can accurately diagnose the mooring condition and creep at different positions under different signaltonoise ratios (SNR), with a highest accuracy rate of 99.83%.

TH133

国家自然科学基金（51976131，52006148，52106262）