A generalized model for detecting the incipient wind turbine (WT) faulty condition based on the data collected from wind farm supervisory control and data acquisition (SCADA) system is proposed in this paper. The linear combination prediction approach and the information entropy are integrated to develop the generalized model, in which the linear combination prediction approach improves the accuracy and generalization performance of the model, and the information entropy of prediction residual quantifies the abnormal level of the condition parameter. SCADA datasets were selected to establish the prediction models of WT condition parameters that are dependent on environmental conditions such as ambient temperature and wind speed. The combination prediction models of WT condition parameters were developed based on different data mining algorithms such as Back propagation neural network (BPNN) algorithm, radial basis function neural network (RBFNN) algorithm and least square support vector machine (LSSVM) algorithm. The information entropy was utilized to extract useful information from residuals of the prediction models for WT faulty condition detection. Finally, the proposed method has been used for real 1.5 MW WTs with doubly fed induction generators (DFIG). Through investigation of cases of actual WT faults, the effectiveness of the proposed WT imminent fault identification approach was verified.

中文翻译：

---

使用组合预测方法和信息熵的风力涡轮机故障状态检测的广义模型

本文提出了一种基于从风电场监控和数据采集 (SCADA) 系统收集的数据来检测初始风力涡轮机 (WT) 故障状态的通用模型。将线性组合预测方法和信息熵相结合开发广义模型，其中线性组合预测方法提高了模型的准确性和泛化性能，预测残差的信息熵量化了条件参数的异常程度。选择 SCADA 数据集来建立 WT 条件参数的预测模型，这些参数取决于环境温度和风速等环境条件。基于反向传播神经网络（BPNN）算法、径向基函数神经网络（RBFNN）算法和最小二乘支持向量机（LSSVM）算法等不同的数据挖掘算法，开发了WT条件参数的组合预测模型。信息熵用于从预测模型的残差中提取有用信息，用于 WT 故障条件检测。最后，所提出的方法已用于具有双馈感应发电机 (DFIG) 的实际 1.5 MW WT。通过对实际WT故障案例的调查，验证了所提出的WT即将故障识别方法的有效性。信息熵用于从预测模型的残差中提取有用信息，用于 WT 故障条件检测。最后，所提出的方法已用于具有双馈感应发电机 (DFIG) 的实际 1.5 MW WT。通过对实际WT故障案例的调查，验证了所提出的WT即将故障识别方法的有效性。信息熵用于从预测模型的残差中提取有用信息，用于 WT 故障条件检测。最后，所提出的方法已用于具有双馈感应发电机 (DFIG) 的实际 1.5 MW WT。通过对实际WT故障案例的调查，验证了所提出的WT即将故障识别方法的有效性。