Wind Turbine Health Assessment And Fault Prediction Based On Diffusion Mapping And Neural Network

In recent years,the scale of wind farm construction is expanding,but the high-frequency failure of wind turbine components will hinder the development of wind power industry,so it is urgent to reduce the operation and maintenance cost by optimizing the maintenance strategy.The operation and maintenance of wind power plant only by manpower is not only time-consuming and laborious,but also the efficiency and accuracy are not guaranteed.Therefore,by using the data of the Supervisory Control and Data Acquisition(SCADA)system of the wind farm to assess the health and predict the failure mode of the wind turbine,this project guides the maintenance scheme of the wind turbine,reduces the loss caused by the failure of the wind turbine,and then reduces the operation and maintenance cost.The specific research content of this paper is as follows:Firstly,this paper summarizes the research progress of condition monitoring and fault detection of wind turbine based on the data collected by SCADA system in recent years,and analyzes the structure and working principle of wind turbine.This paper describes the source,collection mode and database interaction of SCADA data used in this paper.The operation data of the wind turbine collected by the SCADA system is displayed in the Graphical User Interface(GUI),which realizes the one-to-one correspondence of the database parameters and the search and display of the wind turbine operation data.Secondly,the high-dimensional and massive wind turbine operation data collected by SCADA system can’t directly reflect the health degree of the wind turbine.In this paper,the Diffusion Mapping(DM)data dimensionality reduction technology is used to reduce the wind turbine operation data.The wind turbine presents different operation curves under different health degree operation,and then the standard curve fitted after the wind turbine standard health state operation data dimensionality reduction is compared with the number to be evaluated According to the distance of new data generated after dimension reduction,the health degree of the wind turbine is evaluated.In this paper,the Bidirectional Recurrent Neural Networks(BRNN)algorithm is selected to predict the 23 operating parameters of the wind turbine in the sub-health state.Through the comparison of different algorithms and evaluation indexes,it is proved that the BRNN algorithm selected in this paper is suitable for the wind turbine studied,and the 23 predicted operating parameters will be used for the subsequent fault mode prediction.Finally,through the fault research and statistics of 33 wind turbines in recent three years,it analyzes the components that should be monitored during shipment maintenance.The historical operation data of the four kinds of typical faults that lead to the shutdown of the wind turbine for more than 48 hours are extracted from the SCADA system.By mapping different fault types to the same two-dimensional plane through the dimension reduction technology of diffusion mapping data,it can be seen that there are four kinds of curve trends.At the same time,it is proved that there are four corresponding modes of these four kinds of faults in the high-dimensional space.In order to reduce the calculation cost,this paper proposes that Through the training of Probabilistic Neural network(PNN),four kinds of prediction models of wind turbine failure modes are established.The prediction parameters of wind turbine in sub-health state are input into the model to realize the fast prediction of wind turbine failure modes.