基于改进代价敏感支持向量机的风电机组齿轮箱轴承故障诊断研究

发布时间：2018-10-18 10:13

【摘要】：随着大容量、高参数的风力发电机组投入商业运行,对机组设备故障诊断的实时性、准确性及有效性的要求也越来越高,而故障诊断是保证机组安全可靠运行的重要方法之一。风速频繁变化、冲击大、变载荷的运行特点,导致风电机组故障类型多且频率高。而齿轮箱是风电机组最重要传动部件之一,也是故障高发部件,且造成风电机组的停机时间也最长。本论文分析了传统风电机组齿轮箱故障诊断方法的特点及存在的问题,尝试将可有效解决类别不平衡问题的代价敏感学习应用于风电机组齿轮箱故障诊断,探索齿轮箱故障诊断的新方法。主要研究成果如下:针对代价敏感支持向量机(Cost-sensitive Support Vector Machine,CSVM)在样本数据量较大时训练速度过慢的问题,提出增量代价敏感支持向量机(Incremental Cost-sensitive Support Vector Machine,ICSVM)。该算法有效利用KKT条件,对增量样本集中的样本进行有效的选取,剔除对下一步训练无效的样本,得到边界支持向量集。在UCI标准数据集上进行仿真实验,验证的ICSVM的有效性。给出了基于ICSVM风电机组齿轮箱轴承故障诊断方法的具体实现过程,该试验结果表明,该方法平均误分类代价最低;故障类识别率更高;该方法具有训练速度快,非常适合风电机组的在线故障诊断。针对最小二乘支持向量机(Least Squares Support Vector Machine,LSSVM)不具有代价敏感性的问题,提出代价敏感最小二乘支持向量机(Cost-sensitive Least Square Support Vector Machine,CLSSVM)。在LSSVM原始的优化问题上嵌入不同的误分类代价参数,以平均误分类代价最小为优化目标,详细推导了CLSSVM算法。最后将其应用于UCI标准数据集和风电机组齿轮箱轴承故障诊断。试验结果表明,该方法的平均误分类代价最低,克服LSSVM不具有代价敏感性的问题,能够提高故障类样本的正确率;CLSSVM的训练时间短,非常适合风电机组的在线诊断。
[Abstract]:With the large capacity and high parameter wind turbine being put into commercial operation, the requirements of real-time, accuracy and effectiveness of fault diagnosis are becoming higher and higher, and fault diagnosis is one of the important methods to ensure the safe and reliable operation of the unit. The frequent variation of wind speed, high impact and variable load result in many fault types and high frequency of wind turbine. The gearbox is one of the most important transmission parts of wind turbine, and is also the part with high fault rate, which results in the longest downtime of wind turbine. In this paper, the characteristics and problems of the traditional gearbox fault diagnosis method of wind turbine are analyzed, and the cost sensitive learning, which can effectively solve the problem of class imbalance, is tried to be applied to the fault diagnosis of the gearbox of wind turbine. To explore a new method of gearbox fault diagnosis. The main research results are as follows: aiming at the problem of slow training speed of cost sensitive support vector machine (Cost-sensitive Support Vector Machine,CSVM) when the sample data is large, an incremental cost sensitive support vector machine (Incremental Cost-sensitive Support Vector Machine,ICSVM) is proposed. Using the KKT condition effectively, the algorithm selects the samples in the incremental sample set effectively, removes the samples that are not valid for the next training, and obtains the boundary support vector set. The effectiveness of the ICSVM is verified by simulation experiments on the UCI standard data set. The realization process of bearing fault diagnosis method for wind turbine gearbox based on ICSVM is presented. The experimental results show that the method has the lowest average misclassification cost, higher fault class recognition rate and faster training speed. It is very suitable for on-line fault diagnosis of wind turbine. To solve the problem that least squares support vector machine (Least Squares Support Vector Machine,LSSVM) is not cost sensitive, a cost sensitive least squares support vector machine (Cost-sensitive Least Square Support Vector Machine,CLSSVM) is proposed. Different misclassification cost parameters are embedded in the original optimization problem of LSSVM. The CLSSVM algorithm is deduced in detail with the minimum average misclassification cost as the optimization objective. Finally, it is applied to UCI standard data set and wind turbine gearbox bearing fault diagnosis. The experimental results show that this method has the lowest average misclassification cost and can improve the accuracy of fault samples by overcoming the problem that LSSVM is not sensitive to cost, and the training time of CLSSVM is short, so it is very suitable for on-line diagnosis of wind turbines.
【学位授予单位】：长沙理工大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TM315

【参考文献】