Research On The Fault Early Warning Methods Of Wind Turbine Generator Under The Big Data

As a clean energy,wind energy plays an increasingly important role in the improvement of China's energy structure in recent years, and wind power has been greatly supported and developed rapidly. With the operation of large scale wind turbine, because of the special location of the wind farm and the unstable load, many units have been running fault, which greatly affects the safety and economic benefits of the wind farm. As the key part of the high failure rate of the wind turbine, the condition monitoring and fault diagnosis of the wind generator is very important, and it can greatly be found in a timely manner and troubleshooting and maintenance and the cost of the power generation, so as to improve the competitiveness of the wind power plant. With the expansion of the application of the information collection system, the breadth and depth of the condition monitoring of the wind turbine is continuously strengthened, and the generated data is characterized by a large amount of data. How to store and deal with the continuous growth of massive state monitoring data, the rapid and effective fault diagnosis and early warning has become an important issue. Under this background, this paper studies the above problems.(1) According to the massive state monitoring data, the model of on-line fault diagnosis and early warning of wind power units combined with Storm real-time stream data processing and Spark memory batch processing technology is presented to solve the problem of rapid processing in the case of the guarantee accuracy.(2) For most electric power data is continuous, a combination of the fuzzy set and rough set theory of attribute reduction algorithm RDD-AD is proposed. Using fuzzy c-means clustering algorithm(FCM) of every power characteristics of fuzzy, through hierarchical computing attribute dependent degree to realize redundant power attribute reduction, improves the accuracy and efficiency of fault diagnosis, provide effective reference for fault diagnosis of the data preprocessing stage.(3) For the reduction of the data, Naive Bayesian algorithm has the advantages of simple method, high accuracy, fast speed, and more suitable for the classification of large amounts of data. The RDD-NB algorithm is designed for the effective diagnosis.(4) According to the special environment and complex and changeable operation conditions of wind turbine, the method RDD-DBNS-MA combined with DBN, BP algorithm and distributed and multi-agent thought is designed to predict the fault characteristic value of wind turbine generator. The restricted Boltzmann machine(RBM) of DBNS greedily layer by layer pre training, optimize the initial weights of the network and accelerate the convergence speed and prediction accuracy of training, to solve the problem that traditional vibration analysis method is difficult to accurately in a timely manner to extract fault characteristics and scalarization of early warning indicators.Finally, experimental tests and numerical examples are carried out. The real operation data of a wind farm is chosen to test the performance of the proposed algorithms in the cloud computing cluster. The RDD(elastic distributed data set) programming model is adopted in the Spark technology to design the parallel algorithms, which improve the ability of processing massive high-dimensional data. The experimental results demonstrate the effectiveness of the designed algorithms and the good parallel performance.