基于KECA-PLS的风电机组健康状态评估

A kernel Entropy PLS Algorithm and its Application for Wind Turbine Performance Degradation PrognosticMA Liang1, REN Chao2,3, LIU Yuewen2,3, HAO Zengxiao1

摘要:

为了及时而准确地预测风电机组整机性能的退化趋势，提出一种基于核熵主成分分析和偏最小二乘（KECA-PLS）的风电机组健康状态评估算法。首先采用前置局部异常因子算法（LOF）对数据进行预处理，并运用高斯混合模型（GMM）对风电场数据进行聚类，将聚类结果作为后续分类的标签；之后利用核熵成分分析（KECA）对数据进行降维和特征提取，并采用SPE统计量监测风电机组状态；鉴于SCADA数据的不平稳性和非线性，将PLS算法引入到核熵空间，用于故障预测，通过预测残差变化趋势来确定报警限，实现了对故障的及早预警；最后使用模糊评判法并绘制风电机组退化状态雷达图，评价风电机组的退化性能。将该方法应用于某风场实际风电机组运行数据中，结果表明该方法能够准确评估风电机组的当前健康状态，可视化出风机故障的变化过程。

Abstract:

In order to timely and accurately predict the degradation trend of the whole performance of wind turbine,a wind turbine health state assessment algorithm based on kernel entropy principal component analysis and partial least squares (KECA-PLS) is proposed. Firstly, the data are preprocessed using the front local anomaly factor algorithm (LOF), and the wind farm data are clustered using Gaussian mixture model (GMM), and the clustering results are used as labels for subsequent classification; after that, nuclear entropy principal component analysis (KECA) is utilized for the degradation of the data and the extraction of features, and the SPE statistics are used to monitor the state of WTGs; in view of the unsteadiness of the SCADA data and the nonlinearity, the PLS algorithm is introduced into the kernel entropy space for fault prediction, and the alarm limit is determined by predicting the trend of the residual change, which realizes the early warning of faults; finally, the fuzzy judgment method is used and the radar map of the degraded state of the WTGs is plotted to evaluate the degraded performance of the WTGs. The method is applied to the actual wind turbine operation data of a wind farm, and the results show that the method can accurately assess the current health state of wind turbines and visualize the change process of wind turbine failures.