Research On Fault Detection Of Wind Turbine Pitch System Based On Margin Distribution Machine Optimized By State Transition Algorithm

With the rapid development of wind power industry,the pressure of wind turbine maintenance is also increasing.Pitch system is an important component of wind turbine.Once the system fails,it will directly affect the power generation rate of wind turbine,even lead to the damage of wind turbine equipment and cause huge economic losses.Therefore,it is of great significance to carry out the research on effective fault detection and diagnosis of pitch system of wind turbine.The main research work is as follows:(1)Aiming at the problem of redundant features in the fault data characteristics of wind turbine pitch system,in the aspect of feature selection of wind turbine pitch system fault data,in order to eliminate redundant and invalid features from the fault data of pitch system and extract an effective and simplified feature set,A feature selection method for wind turbine pitch system based on random forest(RF)importance ranking method and Pearson correlation analysis method is proposed.The random forest importance ranking is used to screen the features of the data,extract the representative features in the data,reduce the feature scale,and then the Pearson correlation analysis method is used to analyze the correlation between the screened features and the features,so as to filter the redundant features.L1 regularization,Spearman correlation coefficient method and the proposed method are introduced to select the features of the fault data of the pitch system.Compared with the experimental results,the results show that the feature set obtained by the proposed method is simpler without affecting the integrity of data information,and the accuracy of model fault detection is improved to a certain extent.(2)Aiming at the problem that it is difficult to optimize the parameters of pitch system fault detection model,a fault detection method of wind turbine pitch system(LDM-STA)based on state transition algorithm(STA)to optimize large margin distribution machine(LDM)is proposed.The fault detection performance of large margin distribution machine model largely depends on regularization parameters,mean and variance coefficients.The three parameters of the model are set as a threedimensional vector,the classification accuracy of the model is taken as the fitness function,and the global optimal solution of the model parameters is obtained through the state transition algorithm,so as to meet the needs of parameter optimization of the fault detection model of large margin distribution machine.Compared with grid search method,particle swarm optimization algorithm and genetic algorithm,the experimental results show that the proposed method has lower false negative rate and false positive rate.(3)Aiming at the problem that it is difficult to obtain the optimal parameters of the pitch system fault detection model,a wind turbine pitch system fault detection method(ssODM-DSTA)based on dynamic state transition algorithm(DSTA)optimized semi supervised optimal margin distribution machine(ssODM)is proposed.In this method,the three parameters of the semi supervised optimal margin distribution learning machine are input into the dynamic state transition algorithm as three-dimensional vectors,so as to obtain the global optimal parameters of the model to improve the performance of the fault detection model.Compared with grid search method,particle swarm optimization algorithm and genetic algorithm,the experimental results show that the proposed method has lower false negative rate and false positive rate.(4)Development of fault detection system for wind turbine pitch system.The fault detection system of wind turbine pitch system is composed of data and model parameter adjustment module,training model module and model prediction module.The data and model parameter adjustment module can realize the data reading,preprocessing and visual monitoring of the pitch system,and the parameter optimization algorithm is used to optimize the fault detection model.The training model module can build and train the optimized model.The function of the model prediction module is to output the fault detection results and use the evaluation index to evaluate the performance of the model.The experimental data are the actual operation data of wind turbine.The method proposed in the research content is verified by using multiple types of pitch system fault data.