| With the decreasing of fossil fuels and the increasingly prominent global environmental problems,the development and utilization of clean energy has become a hot spot in recent years.Among them,wind power generation has been rapidly developed as the main form of wind energy utilization.Wind turbines are developing towards large-scale and intelligent development.Due to the remote location of wind turbines,most of the maintenance is at an altitude of 50-80 meters.The current SCADA system only records the operating parameters of the wind turbines,and a large amount of operating data has not been fully and effectively used,so the state evaluation of wind turbines based on big data analysis of wind farms has certain research value and practical engineering significance.The thesis first compares and analyzes the operating status and common faults of wind turbines.Based on the idea and method of centralized data fusion,the preprocessing of the historical SCADA data of wind turbines is completed.The decision tree algorithm model was optimized by using the grid search method.Through the feature-level data fusion strategy,the state evaluation model of wind turbines based on the decision tree algorithm was established,which realized the evaluation of the typical working state of wind turbines specified by the International Electro technical Commission.The average accuracy rates of the model on the test set and verification set’s state evaluation were 0.989 and 0.977,respectively,which verified that the wind turbine state evaluation model based on the decision tree algorithm has high accuracy and good generalization ability.By comparing with the logistic regression algorithm,it is proved that the decision tree model has a higher convergence speed and accuracy when the number of samples is relatively insufficient in the process of establishing the model.Based on the Pearson correlation coefficient method,the variable selection for the wind turbine gearbox state prediction model is carried out,the bidirectional LSTM algorithm and the two-layer LSTM model parameter optimization are carried out by the random grid search method,and the two algorithm models are calculated based on the " 3? " criterion The training and test residuals respectively determine the temperature thresholds used for the prediction of the gearbox state,and the sliding window is used to smooth the predicted value of the model to eliminate the influence of accidental errors on the prediction of the gearbox state.The comparison of the prediction results of the two algorithm models shows that the bidirectional LSTM algorithm model predicts the state of the wind turbine gearbox more accurately,and advances the actual time of fault occurrence by 2-5 h,which is safe,stable,economical,and economical for the wind turbine.Efficient operation is of great significance. |
