Abstract:

For the variable working conditions of wind turbine and difficulty in fault extraction and warning index quantization, a new wind turbine fault warning method was proposed based on k-neighbor anomaly detection. Firstly, non-stationary time-domain signals were converted to stationary or quasi-stationary angle-domain signals. Secondly, new dimensionless amplitude domain parameters were constructed and the early failure characteristics of order proportion sampling angle-domain signals were extracted. Finally, mapping the angle-domain series into multi-dimensional feature vectors, the potential abnormal information was excauated by means of k-neighbor anomaly detection, the early fault warning was realized. Finally,the paper verified  the effectiveness of the proposed method.

Key words: order resampling, dimensional parameter, k-neighbor, abnormality detection

摘要：

针对风电机组运行工况复杂多变,难以实现故障特征提取和预警指标量化的特点,提出基于k邻近度异常检测技术的风电机组故障预警方法:首先利用阶比重采样技术将时域非平稳信号转换为角域的平稳或准平稳信号;其次构建出新量纲一幅域特征值,提取阶比重采样角域信号早期故障特征;最后将振动角域序列映射成多维特征向量，通过基于k邻近度的异常点检测技术挖掘机组潜在异常信息,实现机组的早期故障预警。试验仿真验证了该方法的有效性。

关键词: 阶比重采样, 量纲一幅域参数, k邻近度, 异常检测