| As a new type of voltage source converter topology,modular multilevel converter(MMC)has many advantages,such as modular structure design,easy expansion,high output waveform quality,low operation loss,common DC bus,etc.It has been more and more widely used in medium and high voltage direct current transmission,new energy grid connection,high voltage electric drive and other occasions.MMC is formed by cascading a large number of sub-modules.Each sub-module uses insulated gate bipolar transistors(IGBT)and diodes as commutation devices.Compared with diode,IGBT has low over-voltage and over-current resistance and is more prone to failure.Therefore,sub-module failure is one of the common failure types of MMC.Sub-module failure will cause deviation of bridge arm output voltage from expectation,increase of interphase circulating current,increase of AC and DC side harmonics,and affect the safe and reliable operation of the MMC system seriously.Domestic and foreign scholars have conducted a series of studies on the DC side faults and AC side faults of the MMC system,but there are few studies on the fault diagnosis and location of the MMC submodules.The existing fault diagnosis methods for the submodules are based on the installation of each submodule Diagnosis and positioning are performed on the basis of a voltage sensor,and the actual number of sub-modules is large,which invisibly increases the cost of system hardware detection.Therefore,how to diagnose and locate the fault of the MMC sub-module based on the use of a small number of sensors has very important theoretical significance and engineering application value.This paper focuses on the diagnosis and location of MMC single sub-module faults and multi-sub-module composite faults,and proposes the use of signal processing and deep learning methods to detect and locate the open circuit faults of sub-module switching devices.The main research contents are as follows:1)When a sub-module fails,it is difficult to diagnose a specific fault type by using the time domain waveform directly because the characteristic information contained in the time domain waveform is limited.However,the spectrum components of MMC three-phase AC current and three-phase internal circulating current signal have undergone profound changes.Therefore,MMC three-phase AC current and internal circulating current are selected as the original fault detection signals,and a fault diagnosis method based on synchronous compressed wavelet transform and optimized convolution neural network is proposed.In this method,the original time-domain data is converted into two-dimensional time-frequency diagram data with rich time-frequency fault information by using synchronous compressed wavelet transform,which is used as the input data of the optimized convolution neural network model and outputs the fault diagnosis results.Compared with other traditional fault diagnosis methods,this method can automatically extract more abundant and deep fault features in time and frequency domain from image data,and achieve higher accuracy of fault diagnosis.The simulation results verify the effectiveness of the method.2)Aiming at the problem of MMC sub-module compound faulst,the output current waveform change is not very obvious,especially as the number of MMC levels increases,the time-domain waveform characteristics will become smaller and smaller.Therefore,a more accurate and sensitive fault detection method is needed.To this end,a composite fault diagnosis method based on an improved capsule network is proposed.This method improves the original single convolution feature extraction structure of capsule network,directly takes the original current time-domain signal of MMC as the input data of the model,and the model can automatically extract sufficient and detailed features from the time-domain signal.And compared with other three commonly used deep learning methods,analyze their fault diagnosis performance under different level numbers and under changing conditions,build an MMC simulation platform,collect simulation data,and verify the effectiveness of the method.Sexuality and correctness.3)When an open-circuit fault occurs in a sub-module of a certain bridge arm of the MMC,it is necessary to locate the specific faulty sub-module among the numerous sub-modules of the bridge arm.In order to explore a fault diagnosis system that does not need to be equipped with voltage sensors on all sub-modules,but only uses several voltage and current sensors,a fault location based on discrete Fourier transform(DFT)and high-frequency harmonic analysis of bridge arm voltage is proposed.method.In this method,the switching frequency harmonic component of the MMC bridge arm voltage is used as a fault detection signal,and the amplitude and phase angle of the switching frequency harmonic component of the bridge arm voltage are obtained through DFT transformation.The fault is detected by comparing the amplitude of the harmonic component of the switching frequency of the bridge arm voltage with the preset fault threshold,and then the phase angle obtained by the DFT transformation is used to locate the faulty sub-module.The simulation verifies the effectiveness of the proposed method. |
