Abstract: Aiming at lack of fault samples and plenty of normal samples in the running process of wind turbine gearbox, a novel fault diagnosis of wind turbine gearbox method based on Incremental Cost-sensitive Support Vector Machine（ICSVM） is proposed. Due to lack of the fault samples, ICSVM fault diagnosis model which aims to minimize misclassification costs is built. In the training period of incremental cost-sensitive support vector machine, ICSVM using KKT conditions is trained by incremental samples and initial samples. Experimental results show that the proposed method reduces average misclassification costs and training time, and increases recognition rate of gearbox fault.

Key words: wind turbine, gearbox, cost-sensitive support vector machine, incremental learning, fault diagnosis

摘要： 针对风电机组齿轮箱运行过程中故障样本缺乏、正常样本充裕的特点，提出基于增量代价敏感支持向量机（Incremental Cost-sensitive Support Vector Machine，ICSVM）的风电机组齿轮箱故障诊断方法。由于齿轮箱故障样本缺乏，建立以误分类代价最小化为目标的代价敏感支持向量机故障诊断模型；在增量训练代价敏感支持向量机阶段，利用KKT条件，以增量样本和初始样本训练增量代价敏感支持向量机。实验结果表明，该方法能有效地减少平均误分类代价和训练时间，提高齿轮箱故障识别率。

关键词: 风电机组, 齿轮箱, 代价敏感支持向量机, 增量学习, 故障诊断