基于广义回归神经网络的风电机组性能预测模型及状态预警

Wind Turbine Performance Prediction Model and Early Warning of Abnormal Condition Based on GRNN

提出一种基于广义回归神经网络(generalized regression neural network,GRNN)的风力发电机组性能预测及异常状态预警方法。通过分析运行中影响风机主轴转速和发电功率的主要因素,确定了性能预测模型的输入和输出参数。运用监控与数据采集(supervisory control and data acquisition,SCADA)系统的真实历史数据,采用广义回归神经网络(GRNN)建立了风电机组的性能预测模型,通过比较模型的预测精度对GRNN的平滑因子进行了优选。以此模型为基础,采用滑动数据窗方法实时计算风电机组转速和功率的残差评价指标,当评价指标连续超过预先设定的阈值时,则可判断风电机组状态异常。采用某实际风电机组若干历史故障发生前后的真实SCADA数据进行模拟,验证了方法的有效性。

A wind turbine performance prediction and abnormal condition early warning approach based on generalized regression neural network(GRNN)proposed.By analyzing main factors affect fan speed and unit power of the wind turbine,the input and output parameters of the performance prediction model determined.Based on the historical data of generalized regression neural network(SCADA)system,the performance prediction model of wind turbine established with GRNN,and the smooth factor of the GRNN model optimized by comparing the model prediction accuracy.On this basis,the sliding data window model adopted to calculate the residual evaluation indexes of the generator rotating speed and power of the wind unit.When the evaluation indexes continuously exceed the preset thresholds,the wind turbine abnormal.By taking the real SCADA data before and after the occurrence of several historical faults of the actual wind unit as simulation examplesthe effectiveness of the proposed method verified.