基于数据驱动的风电机组变桨系统故障诊断与健康状态预测研究

Research on Fault Diagnosis and Health State Prediction of Wind Turbine Variable Pitch System Based on Data Drive

文中首先聚焦于风电机组桨叶零位偏移故障,利用GH-Bladed风机仿真软件仿真不同工况下的桨叶零位偏移,研究零位偏差对运行机组叶轮转速、叶根弯矩的影响,并采用风电机组轴向加速度1P谐波幅值和3P谐波幅值之比拟合桨叶零位偏差判定曲线图,构建桨叶零位偏差判定模型。再基于神经网络技术,分析机组实际运行数据采集与监视控制系统(Supervisory Control And Data Acquisition,SCADA)历史数据,完成变桨故障特征提取和数据分析处理,训练添加注意力机制的长短期记忆神经网络(Long Short-term Memory Neural Network,LSTM)模型,构建AT-LSTM变桨健康状态预测模型,并从多个分类模型指标,将AT-LSTM与循环神经网络(RNN)和长短期记忆神经网络(LSTM)进行对比,证明了添加注意力机制对于神经网络带来的提升。

Due to the randomness of wind,the frequent variable pitch action of wind turbine leads to frequent failures of variable pitch system.The research on fault diagnosis and fault warning of variable pitch system of wind turbine is more and more important.In this paper,we first focus on the blade zero offset fault of wind turbine,use GH-Bladed fan simulation software to simulate blade zero offset under different working conditions,and study the effect of zero deviation on the impeller speed and blade root bending moment of the operating unit,and use the comparison of wind turbine axial acceleration 1P harmonic amplitude and 3P harmonic amplitude to determine the blade zero deviation curve.The blade zero deviation determination model is constructed.Then,based on the neural network technology,the historical data of SCADA(Supervisory Control And Data Acquisition)in the actual operation of the unit was analyzed,the feature extraction and data analysis and processing of variable pitch fault were completed,and the LSTM(Long short-term memory neural network)model with attention mechanism was trained to build the AT-LSTM(attention based LSTM)health state prediction model of variable pitch.By comparing AT-LSTM with recurrent neural network(RNN)and long short-term memory neural network(LSTM),we prove that adding attention mechanism can improve neural network.