A Research On New Class Of Fault Diagnosis In Wind Power System Based On Deep Learning

With the continuous depletion of non-renewable energy sources such as coal and oil,many renewable energy sources such as wind,solar and tidal energy have been developed rapidly.Wind energy is not only a renewable energy source but also extremely clean and can provide a constant source of energy for power generation systems.However,wind energy resources are very unstable and wind power systems need to serve in extremely harsh climatic environments,which causes great damage to wind power systems and makes the operation and maintenance costs increase significantly.A timely diagnosis of minor faults in wind power systems would greatly reduce O&M costs.However,the wind power generation system has many components and the fault locations are extremely complex,which puts high demands on the fault diagnosis technology.The traditional fault diagnosis algorithm can only identify fixed types of faults,i.e.,the algorithm is designed to limit the types of faults it can identify,and it cannot make corresponding algorithm adjustments according to the actual types of faults.In the actual operation of the wind power system,there may be fault types that are not considered in the algorithm design,and if such faults occur,the algorithm cannot identify the fault types.In this thesis,we establish fault diagnosis algorithms for existing faults in wind power generation systems and then rely on the learning ability of the algorithms to continuously add new types of faults,so that the algorithms can diagnose more types of faults.In this thesis,we conduct an in-depth study on the diagnosis of new types of faults in wind power generation systems,the main contents of which are as follows.(1)A new type of fault diagnosis method based on semi-supervised deep learning is proposed for the traditional fault diagnosis methods that cannot identify the newly emerged types of faults in wind power generation systems.Firstly,multiple wind power generation system signals are used as inputs,and features are extracted by Convolutional Autoencoder.Secondly,initialization models are established with compressed features as classifier’s inputs and error feature maps as detector’s inputs.Finally,the detector will put new type fault instances into the cache,and the algorithm starts to update when the cache overflows,thus achieving the purpose of identifying new type faults.The simulation experimental results show that the method can effectively solve the new type fault identification problem of wind power generation systems,and various indexes are better than other algorithms.(2)The new type of fault diagnosis framework for wind power generation systems cannot balance the update speed of the model with the recognition accuracy,and the detectors are subject to false detection.A new type of fault diagnosis method based on a generative adversarial network and multi-label threshold detection algorithm is proposed,which Accelerated update speed and greatly reduces the false detection rate.The simulation experimental results show that the algorithm greatly accelerates the update speed and reduces the false detection rate with high accuracy.