采用预测模型与模糊理论的风电机组状态参数异常辨识方法

为提高风电机组的停运预警能力,基于风电场数据采集与监控(SCADA)系统数据提出了一种风电机组状态参数的异常辨识方法。对参数进行划分,针对与环境因素密切相关的状态参数,采用神经网络建立了状态参数预测模型。采用本机组近期SCADA样本、本机组历史样本和其他机组近期样本分别作为预测模型的训练数据,对比分析了基于3类样本建立的模型的预测精度。采用平均绝对误差对基于本机组历史样本和其他机组近期样本建立的预测模型进行选择。定义了异常程度指标量化预测残差的异常程度。为了提高异常辨识的精度,采用模糊综合评判对筛选出的预测模型的异常辨识结果进行融合。最后,以国内某风场的1.5 MW风电机组为例进行了异常分析,并与传统的风电机组状态参数异常检测方法进行了对比,实例分析结果表明所提出的异常辨识方法具有更高的准确性。

Wind turbine status parameter anomaly detection based on prediction models and fuzzy theory

In order to improve the ability of early WT(Wind Turbine) outage warning, a method for the anomaly detection of WT status parameters is proposed based on the SCADA(Supervisory Control And Data Acquisition) system data of wind farm. The parameters are grouped and the neural networks are applied to develop the prediction models for the environmentally-sensitive status parameters. The resent and historical SCADA data samples of this WT and the resent SCADA data samples of other WTs are used as the training data and the prediction accuracies of the prediction models based on the three sample types are compared. The MAE(Mean Absolute Error) is used to select the prediction models trained by the historical SCADA data sample of this WT and the resent SCADA data samples of other WTs. An anomaly index is proposed to quantify the anomaly level of the residual error of status parameter prediction. In order to improve the accuracy of anomaly detection, the fuzzy synthetic evaluation is applied to integrate the results of anomaly detection by the selected prediction models. The proposed anomaly detection method is applied to a domestic 1.5 MW WT and the results show that it has higher accuracy than the traditional methods.