Converters Fault Diagnosis Of DFIG-Based Wind Turbine Systems Based On Data Driven Methods

The increasing use of fossil energy has brought about problems of environmental pollution.As renewable energy,wind energy has received more and more attention,and its penetration rate in the power generation field continues to rise.Therefore it is necessary to improve the stability and reliability of wind power generation technology.Converters are the most vulnerable parts in the energy systems.In order to reduce failures and unscheduled shutdowns of the wind turbine system,effective diagnosis scheme for the converter is regarded as essential and mandatory.This paper proposes fault diagnosis algorithms for sensor faults and opencircuit switch faults in the back to back converter of doubly-fed wind turbine system.Firstly,the topology and mathematical model of the grid-connected doubly-fed wind power generation system is introduced.The structure,mathematical model and control scheme of back-to-back PWM are stuided.The simulation platform of wind turbine system is MATLAB/Simulink and RTDS,and the parameter setting scheme of the simulation model is given,which provides a foundation for simulation research on fault diagnosis strategy.Secondly,methods of data cleaning and feature engineering are introduced according to the characteristics of doubly-fed wind power system.Pearson coefficient was used to evaluate the correlation between different parameters in fault scenarios.The missing data signals were filled in by mean interpolation.Mathematical morphology is used to eliminate the noise.The problem of imbalance data was solved by SMOTE oversampling.Data cleaning and feature engineering are applied to get rid of the redundant information and keep the valuable data.In addition,two data-driven fault diagnosis scheme for the converter of doubly-fed wind power generation system is proposed,which are fault detection method based on XGBoost and fault location method based on LSTM.XGBoost is a highly extensible integrated decision tree scheme,which is able to process multi-dimensional feature in parallel and determines the failure state of the system.LSTM mines the hidden relationship in time sequence and locate the faulty components.Afterwards,a fusion model of XGBoost and LSTM is proposed.Finally,the feasibility and reliability of the fault diagnosis methods for doubly-fed wind power generation system are verified by simulation.A fault detection scheme based on XGBoost for converters current sensors is designed,which is able to distinguish bias fault,gain fault and constant value fault.A fault locating scheme based on LSTM for multiple open-circuit switch faults of converter is constructed.The effectiveness of the fusion model is also discussed.Simulation results have demonstrated that,the proposed method excavates the deep information of the fault signal with high diagnosis accuracy and strong robustness with short time delay.