Research On Fault Diagnosis Of Offshore Doubly-fed Wind Turbine Based On Data Drive

Offshore wind power has become an important strategic layout for global renewable energy development.With the rapid development of offshore wind power,European offshore wind power powerhouses have begun to move towards large-scale and deep-sea development.my country's offshore wind farms are also advancing in the direction of large-scale and far-reaching marine development.As one of the mainstream models of Shenyuanhai wind turbines,doubly-fed wind turbines face problems of poor accessibility,high failure rate,and difficult maintenance in the deep sea operating environment,which are more prominent than offshore,and may even cause casualties..In order to reduce the major economic losses or safety accidents that may be caused by the failure and shutdown of the deep sea wind turbines,there is an urgent need for accurate offshore doubly-fed wind turbine fault warning and diagnosis technology.My country's offshore wind power development has a history of ten years.The SCADA system of large-scale wind turbines can obtain massive real-time operating data of the entire wind farm,which is a data-driven offshore dual the research on fault diagnosis of the fed wind turbine provides data support.With the rapid development of artificial intelligence and deep learning,it provides theoretical support for data-driven offshore doubly-fed wind turbine fault diagnosis research.Therefore,based on the actual operating data of an offshore wind farm in China,this paper uses a data-driven method to diagnose the fault of DFIGs.Firstly,This paper considers the correlation changes between state variables,predicts abnormal operating conditions through reconstruction probability,and proposes a new method for predicting abnormal operating conditions of wind turbines based on variational autoencoding.First,use the generator SCADA data under normal operating conditions to train the VAE network,and calculate the reconstruction error matrix of the input variables.Then,the robust Mahalanobis distance of the reconstructed error matrix is used as the monitoring index of abnormal working conditions,and the monitoring index alarm threshold value under normal operation state is calculated by the kernel density estimation method,and the criterion for the abnormal working condition of the wind turbine is obtained.Finally,the SCADA data of a 3MW doubly-fed wind turbine in an offshore wind farm is analyzed as an example.The results show that the method proposed in this paper can effectively predict the abnormal working conditions of offshore wind turbines and provide technical reserves for the development of deep sea wind power.Afterwards,in order to solve the problem of insufficient warning time and difficulty in obtaining fault samples in the SCADA system of offshore wind farms,a new method of early warning of wind turbines based on the GRA-LSTM model was proposed.First,use GRA to analyze the SCADA data of the normally operating wind turbines.Considering that the temperature characteristics have the characteristics of thermal inertia and strong anti-interference,they can effectively reflect the gradual trend of generator failures,and extract the relationship between the generator temperature highly correlated state variables are used as input to the LSTM temperature prediction model.Then calculate the residual absolute value of the predicted value of the output temperature and the actual value,use the method of probability distribution fitting to set the alarm threshold,and provide early warning of the generator's early failure.Finally,taking the measured operating data of domestic offshore wind farms in operation as an example,the results show that the method in this paper can identify early failures of offshore wind turbines29-72 hours in advance.Finally,in order to reduce the occurrence of wind turbine shutdowns caused by generator failures,improve the accuracy of fault diagnosis,and ensure the safe and reliable operation of wind turbines,a new method of wind turbine fault diagnosis based on stacking fusion algorithm is proposed.First of all,the data from the early warning time to the shutdown is used as the original data set of stacking fusion algorithm.Then divide the state parameters of the fault period,use K-fold cross-validation to train the first-level basic learner of the fusion algorithm,and output the first-level diagnosis results in the form of probability.Then combine the probabilities output by the first layer into a new data set,train the second layer meta-learner,and output the final accurate diagnosis result.Finally,it is verified with actual offshore wind farm data,and the results show that the method proposed in this paper improves the accuracy of generator fault diagnosis by 6.12% on average.