Research On Fault Diagnosis Of Wind Turbine Drive Chain Based On Big Data And Artificial Intelligence

According to the latest statistics of the National Energy Administration,the cumulative grid-connected capacity of wind power nationwide by the end of 2018 is about 184 million kilowatts.It is expected that by the end of 2020,the cumulative installed capacity of wind power will reach more than 210 million kilowatts.With the implementation of non-subsidized and parity grid access for wind power and photovoltaic power generation,this policy adjustment has forced the entire wind power industry to start from its own healthy operation to increase corporate efficiency.The wind turbine drive chain is an indispensable part of energy conversion,mainly including bearings and gearboxes.As a key part of the whole machine,it is of vital importance to the healthy operation of the wind turbine.According to the existing fault data,the downtime caused by the failure of the wind turbine drive chain is about 80%,which is mainly caused by gearbox and bearing failure.Therefore,early warning and fault diagnosis of key parts of wind turbines have practical application value for the development of the whole industry.First,the data types and sizes generated by the wind turbines are counted,and structured and unstructured data such as production and operation data,production management data,and on-site monitoring video of the wind turbine are sorted out.Using HDFS in open source Hadoop as the underlying distributed file system,using HBse as the database and Hive as the data warehouse,and Zoo Keeper to provide coordinated scheduling to establish the wind turbine large data storage and management system.Secondly,for the traditional threshold alarm,the fault has already occurred in the alarm,and the damage is irreversible and there are misstatements and false positives.It is proposed to construct a fault warning system based on big data.Through the big data storage system to capture a large amount of historical monitoring data for data analysis,verify the normal distribution of the sample data in the early stage,further fit the Beta distribution,and then the shape parameters are obtained by optimizing the parameters of the estimated probability density function.Finally,the intelligent knowledge algorithm of self-learning threshold is used to determine the expert knowledge base,thus completing the construction of the big data-based fault early warning system.In the whole process,the correlation of speed,wind speed and output power is considered comprehensively,realizing the real-time monitoring of the running state of the wind turbine.Finally,the problem of fault signal decomposition and fault feature extraction is limited and the nonlinear expression ability is poor for traditional time-frequency analysis and shallow neural network.A feature extraction model of convolutional neural networks is proposed by using the characteristics of deep nonlinear learning and the ability to automatically select fault features.At the same time,the fuzzy theory is applied to the multi-classifier to design a fuzzy multi-classifier to realize the refinement of the feature state.The experimentally simulated fault data is used to train the entire model to obtain the model parameters,and finally to accurately diagnose the fault type and fault level.