关键词: 风电主轴承, 故障诊断, 粒子群优化, 随机森林, 特征选择, 机器学习

Abstract: A novel fault diagnosis method based on particle swarm optimization ( PSO) algorithm and random forest ( RF) model was proposed for fault detection of wind turbine main bearings. The feature subset was optimized by combining the feature selection capability of PSO algorithm with the high accuracy and stability of RF model, significantly improving the accuracy and reliability of fault diagnosis. The experimental results show that the PSO-RF model had a diagnostic accuracy of 92. 42% and an F1 value of 75. 38% in wind power main bearing fault detection tasks. All performance indicators were superior to the unoptimized RF model, XGBoost model and K-nearest neighbor algorithm. The PSO-RF model demonstrated high tolerance for outliers and noise in robustness testing, as well as good cross condition generalization ability. This method not only improves the diagnostic ability of the model, but also provides a new technical means for the maintenance and fault warning of wind power equipment, which helps to improve the operational efficiency and equipment reliability of wind farms.

Key words: wind power main bearings, fault diagnosis, particle swarm optimization ( PSO), random forest ( RF), feature selection, machine learning