Research On Health Status Assessment And Failure Warning Of Wind Turbine Based On Data Fusion

In recent years,due to environmental pollution,world energy shortage and other problems,wind power,as a pollution-free and renewable energy,has gradually become a hot industry that countries all over the world are competing to develop.China has formulated many favorable policies and measures to encourage the development of wind power industry.Therefore,the industry is developing by leaps and bounds,and the installed capacity of wind power is gradually increasing,which has entered the forefront of the world.However,due to the long-term operation of wind turbine,complex equipment structure and harsh location environment,components of wind turbine may be vulnerable to damage,and the failure rate is high,resulting in high maintenance costs.Therefore,the analysis is made from three aspects: state parameter prediction,state evaluation and fault warning,in order to provide a basis for the safe and stable operation of wind power equipment,reduce the failure rate and reduce the maintenance cost.This paper studies the health status assessment and early warning of generator sets,and the main research contents are as follows:(1)Aiming at the prediction of wind turbine state parameters,a prediction method based on LSTNet network is proposed.First,in view of the lack of theoretical basis and feature redundancy in the selection of observation vectors,the grey correlation analysis is used to screen out the feature parameters with strong correlation with the state parameters,and then the average influence value of each parameter is calculated in combination with the MIV index to further compare the influence of the feature parameters on the state parameters.Furthermore,the selected feature parameters are involved in the model prediction.Finally,the LSTNet multivariate time series framework is used to integrate GRU The structure characteristics of RNN and LSTM networks,the LSTNet network prediction model is established and compared with SVR,RNN and LSTM prediction methods.The results show that the multi-step prediction method of long-term and short-term time series based on LSTNet model has a high prediction accuracy,which is obviously superior to other methods in this paper,and effectively improves the prediction accuracy.(2)Aiming at the problem of health status assessment of wind turbine,a health status assessment method of wind turbine based on ELM-SOM is proposed.Firstly,outliers are filtered and parameter features are extracted from Supervisory Control And Data Acquisition(SCADA);Then several prediction models of Extreme Learning Machine(ELM)are established,and residual data and prediction model values are extracted from actual operation data to evaluate their characteristics.Self organizing mapping(SOM)is used to fuse multiple residual features to reflect the operation status of wind turbines.The best matched unit(BMU)determines the threshold value,and finally establishes the health indicators to realize the status recognition of wind turbine generator units.Finally,the validity of the proposed method is verified by comparing single feature and different classification methods with experimental data.(3)Aiming at the early warning problem of wind turbine operation state,a fault early warning method based on GRU Attention model is proposed.Firstly,the data of the wind farm SCADA system is preprocessed,and the monitoring quantity that can represent the key operation status of the wind turbine generator is selected as the output quantity.The Kalman filter is used for data fusion,and the monitoring quantity with high correlation with the output quantity is selected as the input parameter according to the correlation analysis;Then the GRU Attention network model is constructed according to the feature selection characteristics and parameter nonlinear characteristics,and the output predictive value and residual error are statistically analyzed,and the adaptive threshold is set to monitor the trend change of the abnormal state of the wind turbine.The early warning model is applied to the analysis of a wind farm,which can effectively warn the early failure of key components of wind turbine.