基于改进DBSCAN算法的风机故障诊断研究

Research on wind turbine fault diagnosis based on improved DBSCAN algorithm

针对风电集控中心需监控多个类型风机并对故障的风机进行故障诊断问题,通过运用DBSCAN聚类算法对风机运行数据进行密度聚类,判定齿轮箱和主轴方面的故障,并针对DBSCAN算法中需人为设定参数的确定进行了改进。首先在KNN分布曲线上假定陡增点,采用循环迭代的方法进行分段拟合计算出最优参数Eps;然后利用数学统计原理分析计算MinPts,实现聚类全过程的自动化,减小了根据经验判断参数的误差。最后利用风场实际数据进行试验,提取并分析聚类结果中的噪声点,通过数据异常值进行故障诊断,验证了此方法的可行性和有效性。

Since the wind power centralized control center monitors various types of wind turbine,and diagnoses the fault of wind turbine,a DBSCAN clustering algorithm is used to perform the density cluster for the operating data collected from the wind turbine,and judge the fault of the gearbox and main axis.The item that the parameters should be set artificially in DBSCAN algorithm is improved.A steep increasing point is assumed on the KNN distribution curve,then the loop iteration method is used for segmentation fitting to calculate the optimal parameter Eps,and the mathematical statistics theory is used to analyze and calculate MinPts to realize the automation of whole clustering process and reduce the parameter error according to the experience-based judgment.The experiment was carried out with wind site practical data,and the noise point in clustering result was extracted and analyzed.The feasibility and effectiveness of the method are verified by means of the fault diagnosis of abnormal data.