Research On Fault Early Warning Method Of Wind Turbine Pitch System Based On Data Driven

As offshore wind power development continues to advance,the individual capacity and complexity of wind turbines continue to rise,and capacity losses and maintenance costs due to failures continue to increase.On the other hand,along with the promotion of affordable wind power and the end of various wind power feed-in subsidies,the profitability of wind power development has been further compressed.Based on this,it is necessary to reduce the operation and maintenance costs of wind turbines,improve the efficiency of wind turbine utilization,and enhance the profitability of wind power operation by strengthening the monitoring and fault warning capability of wind turbines.In this paper,we take the pitch system with high failure rate as the research object.Based on the in-depth analysis of its working principle,failure mechanism and failure characteristics,we study the data-driven fault warning method of wind turbine pitch system based on the historical data and fault simulation data of wind turbines,and the main research contents are as follows.（1）Combining the overall structure and working principle of wind turbine and pitch system,describe the common faults of wind turbine pitch system and give the analysis of the causes leading to various faults.The data sources used in the pitch system fault warning method are introduced,and the characteristics of the existing data sets are summarized.The data set is pre-processed using filtering and normalization,and the correlation analysis of the data is performed based on the correlation coefficients to digitally represent the correlation of the parameters related to the wind turbine pitch system faults.（2）In order to achieve reliable early warning of the early failure of the pitch system,a fault warning method of the pitch system based on kernel density estimation is proposed.First,the input and output variables of the kernel density estimation model are determined on the basis of data correlation analysis.Secondly,the kernel density estimation model is used to explore the conditional dependence relationship between the observed variables and the associated variables,and to determine the distribution of the observed variables of the pitch system under different operating conditions;on this basis,the confidence level is specified to estimate the distribution interval of the observed variables,and the preliminary judgment of the operating status of the pitch system is made by observing the relationship between the actual sample values and the estimated interval.Finally,based on an empirical statistical conclusion that the samples have a high probability of falling within the estimated interval under normal conditions,hypothesis testing and sliding window statistics are used to determine the abnormality identification results and give the final warning information.（3）A fault diagnosis method based on softmax classification is proposed for the problem that the actual wind turbine SCADA system provides few fault samples,which makes it difficult to achieve accurate diagnosis of fault types in the pitch system.First,the method simulates different faults of the pitch system by the high-fidelity wind turbine simulation software Bladed and constructs the corresponding fault data set;and then,a softmax fault classification model based on the improved gradient descent method and the cross-entropy loss function value is constructed,which achieves the diagnosis of different fault types by maximizing the probability of different categories of faults;finally,the classification results based on the simulation data.Finally,the classification results based on simulation data verify the feasibility of the proposed method and validate the effectiveness of the method based on actual SCADA data.