Application Of Data Mining In Wind Turbine Fault Diagnosis

As one of the new energy, wind energy is an important source of power supply.Because of the harsh operating environment of the wind turbine, maintenance costs are very high. Traditional reactive maintenance program brought lower yielding, and the preventive maintenance program is limited to technical pressure, which needs a major breakthrough. In this paper, based on the big data of wind farms and long-term historical data, data mining techniques and methods of data mining are used to find the law and to establish and optimize the forecasting model. Compared to traditional fault diagnosis methods, such as signal processing, mathematical statistics, the advantages of data mining in fault diagnosis of wind turbines can make better use of the law and the value of data, which greatly enhance fault prediction timeliness. As long as the data amount is large enough and the quality of data cleaning is high,choosing a higher-degree matched training and learning algorithm can obviously improve the accuracy of fault diagnosis.This paper first introduces the common faults of wind turbine, analyzes the causes of failure, and then discusses the use of data mining in the wind turbine fault prediction processes, including the pre-analysis of the big data from wind turbine by clustering, data cleansing, data preprocessing, feature Selection, model implementation, model evaluation, and the model application in the actual operation.For wind turbine gearbox temperature prediction, using the residuals between actual and predictive values and combining with statistical analysis of characteristic can decide the point of failure.The innovation of the article is that based on the special nature of the wind turbine operating conditions, K-means clustering method is used in the input data preprocessing, and the BP neural network is selected to train the model. Compared with the prediction effect by a direct traditional use of neural network, the proposed method can improve the accuracy rate by 3.5%, which effectively improve the accuracy of the determination of the fault point. The model is applied to the actual practical application, and the fault alarm can bring forward 3 to 5 days in advance compared to the existing reactive maintenance program, thus greatly reducing maintenance costs and improving wind farm efficiency.