基于D-S证据融合的风力发电机组的故障预测

发布时间：2018-10-12 21:12

【摘要】：随着不可再生能源的快速消耗,能源问题已经成为人类迫切需要解决的需求,风能因为持续可再生而成为备受注目的清洁能源。风力发电机是完成能量转换的关键部件,而风力发电机的故障诊断和维护是保障风机稳定正常运行的首要条件。风力发电机往往装在人迹罕至的极端环境或者海平面上,传统的设备维修都是等到风机损坏之后再派人过去维修,这样不仅浪费大量的人力物力,有时候会因为风机长期带病运行,最终造成严重的不可逆的设备故障,所以如何能对风机的故障进行早期预测成为一个值得研究的问题。本课题在大连驼山风场积累的历史数据和故障日志基础上,主要针对双馈异步发电机的常见故障进行故障预测,待识别的风机故障包括定子绕组短路,转子绕组短路,轴承损坏和转子偏心,前两个属于电气故障,而后两个属于机械故障。通过数据选取和小波包分解提取振动和电流频域特征向量,然后通过D-S证据融合理论建立故障预测模型。传统的故障诊断是通过分析正处在故障中机器运行参数建立诊断模型,因此所建立的诊断模型只适用于已经处于故障状态的风机。本文的方法是选取风机出现故障一个小时之前的运行数据,此时风机虽然仍然处于运行状态,但是振动参数和电流参数已经出现异常,属于带病运行状态,提早发现异常就可以提前停机,防止风机持续运行造成不可逆的损坏,同时预测出的故障类型对维修人员也有较大的参考。针对传统风机故障诊断中采用振动信号构造单个特征空间的故障预测的不足,本文将电流信号引入了故障预测,并引入了基于D-S证据融合的故障预测模型,首先在振动信号和电流信号上分别构造了两个后验概率支持向量机,将两个支持向量机的概率输出作为证据融合的基本概率分配,根据Dempster融合规则计算融合之后的概率分配,针对融合过程中证据之间的冲突因子太大容易导致融合失败的问题,本文提出了用局部可信度来修正融合之前的基本概率分配,局部可信度表示支持向量机对每种故障的预测准确率,实验证实经过局部可信度修正过基本概率分配的多个证据在融合过程中冲突因子更低,基于D-S证据融合模型相比于非融合模型对风机四种故障均有更高的预测准确率。
[Abstract]:With the rapid consumption of non-renewable energy, the energy problem has become an urgent need to be solved. Wind energy has become the focus of attention because of renewable energy. Wind turbine is the key component to complete the energy conversion, and the fault diagnosis and maintenance of wind turbine is the primary condition to ensure the stable and normal operation of wind turbine. Wind turbines are often installed in isolated extreme environments or at sea level. Traditional equipment maintenance is to wait until the fan is damaged before sending someone to repair it. This not only wastes a lot of manpower and material resources, Sometimes it is necessary to study how to predict the fault of fan in the early stage because the fan runs for a long time and finally causes the serious irreversible equipment failure. Based on the historical data and fault log accumulated in Dalian hump wind field, this paper mainly predicts the common faults of doubly-fed asynchronous generator. The fan faults to be identified include stator winding short circuit, rotor winding short circuit, and rotor winding short circuit. Bearing damage and rotor eccentricity, the first two electrical faults, the latter two mechanical faults. The eigenvector in vibration and current frequency domain is extracted by data selection and wavelet packet decomposition, and then the fault prediction model is established by D-S evidence fusion theory. The traditional fault diagnosis model is established by analyzing the operating parameters of the machine in the process of fault, so the diagnosis model is only suitable for the fan which is already in the fault state. The method of this paper is to select the operation data of the fan one hour before the failure. At this time, the fan is still in operation state, but the vibration and current parameters have been abnormal and belong to the diseased running state. Early detection of abnormal can stop the fan in advance to prevent irreversible damage caused by the continuous operation of the fan. At the same time, the predicted fault type also has a large reference for the maintenance personnel. Aiming at the shortcoming of using vibration signal to construct a single feature space for fault prediction in traditional fan fault diagnosis, the current signal is introduced into fault prediction, and a fault prediction model based on D-S evidence fusion is introduced. Firstly, two posterior probabilistic support vector machines are constructed on the vibration signal and the current signal respectively. The probability output of the two SVM is regarded as the basic probability distribution of evidence fusion, and the probability distribution after fusion is calculated according to the Dempster fusion rule. In view of the problem that the conflict factor between the evidence in the fusion process is too large to lead to the failure of fusion, this paper proposes to modify the basic probability allocation before fusion by using local credibility. The local reliability represents the prediction accuracy of each fault by support vector machine. The experimental results show that the conflict factor is lower in the fusion process when the local reliability is corrected for the basic probability allocation. Compared with the non-fusion model, D-S evidence fusion model has higher prediction accuracy for four kinds of fan faults.
【学位授予单位】：沈阳工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TM315

【参考文献】