Research On Fault Diagnosis For Wind Turbine Based On K-nearest Neighbor And Rule Mining

The exploitation of clean and renewable energy and development of low-carbon economy have become the focus of global energy strategy and human sustainable development.As a new energy power generation method with advantages of mature technology,large development scale and good commercial development,wind power has been one of the world fastest-growing green energy.The rapid increasing of the wind power capacity in China has caused overcapacity and quality problems.Since the wind turbine’s operating environment is harsh,the quality wind resources are getting less,and the wind power industry tends to be "emphasizing manufacturing over management",under this situation,the faults of key components of wind turbines occur frequently,which seriously affected the operation efficiency of wind turbines and the safety of power system.In view of the above background and situation,data mining technology was applied and the fault diagnosis method of wind turbine based on data-driven method was carried out in this paper,to realize the working state features mining,abnormal state detection,fault identification and fault probability analysis.This study provides the theoretical basis for the research and development of wind turbine fault diagnosis system,the main work was summarized as follows:(1)A dynamic feature mining method of wind turbines based on mutual information is presented due to coupling and dynamic correlation among the monitoring signals of wind turbines.The instant features and time-delayed features were used to constitute the augmented matrix,the accumulation of mutual information among different features were used to bulid the dynamic feature matrix of wind turbine monitoring signals,and the feature parameters in the dynamic feature matrix were used as the input of the wind turbine component fault detection model.The proposed feature mining method was verified by comparing the influence of different feature processing methods on the wind turbine fault detection performance.This method considered both the coupling and trending relationship among features,can obviously reduce the influence of irrelevant features on the model output and avoid the complex model calculation due to excessive features.(2)To solve the problem of wind turbine fault detection under complex working conditions,a weighted k-nearest neighbor fault detection method based on dynamic feature matrix was proposed.To decrease the influence on the accuracy of the detection model in complex working conditions,the k-nearest neighbor fault detection method was used as a framework,the dentification of abnormal state for online sample depends on its nearest neighbor distance.The cumulative mutual information value among features were considered into account to achieve feature weighting,to increase the separated ability on abnormal satate.A dynamic threshold setting method based on nearest neighbor samples is proposed to avoid false alarm and missed alarm caused by abrupt changes of working state.The dynamic threshold of fault detection was obtained by combining the threshold value at a given confidence level with the mean value of the nearest neighbor distance.The proposed method is respectively applied to the FAST simulation model and the real fault of pitch system to verify its validity.(3)For the fault mining of wind turbine,a fuzzy rule mining method based on intelligent optimization is proposed.In the proposed method,the initial rule set was generated by combining fuzzy C-means clustering and heuristic learning to avoid the influence of noise samples.The diversity of population and overall searching performance were improved by means of multi-population quantum coding and hybrid evolution strategy.And the reconstruction of contradictory rules was introduced to deal with low performance rules.After verifying the search ability and noise tolerance of the proposed rule mining method through standard data sets,the proposed method was combined with ReliefF feature selection method,and applied to the fault rule mining and fault type identification,to verify the availability of the proposed method for mining fault knowledge of wind turbines.(4)To deal with the problem of potential fault analysis of wind turbines,the fault probability analysis method and the backtracking method of key abnormal variable were proposed.The proposed method applied a neighborhood faults rule selection strategy to to evaluate the fault status of wind turbine component,On this basis,the potential probability sort was obtained by the steps of fault filtering,multi-rule competition and fault probability calculation,and the key abnormal variables were identified by variable filtering and contribution evaluation according to the neighbor fault rules.The validity of proposed methond was verified by applying the proposed method to the fault root analysis of NREL-5MW wind turbine.