MMC submodule IGBT open circuit fault diagnosis method based on frequency domain analysis and readable storage medium

Abstract

The present invention relates to the field of electronic power technology and discloses a method for diagnosing open-circuit faults of MMC submodules and IGBTs based on frequency domain analysis and a readable storage medium. The method comprises the following steps: controlling a modular multilevel converter to be monitored by using a carrier phase-shift modulation method, collecting in real time the arm voltage signal of each arm on each phase in the modular multilevel converter to be monitored, dividing the arm voltage signal by using the power frequency period as the time domain division length, and obtaining a plurality of arm segment voltage signals; extracting the amplitude of harmonics within a range from a DC component to a preset multiple, and the phase angle of harmonics within a preset interval centered on the carrier frequency based on the complex representation of each frequency in the frequency domain information of each arm segment voltage signal, constructing the fault features of each arm segment voltage signal, and sequentially inputting them into a trained support vector machine model to obtain the fault label corresponding to each arm segment voltage signal, thereby completing the fault diagnosis.

Description

Technical Field

[0001] The present invention relates to the field of electronic power technology, and in particular to a method for diagnosing an open-circuit fault of an MMC submodule IGBT based on frequency domain analysis and a readable storage medium. Background Art

[0002] A modular multilevel converter (MMC) is a multilevel commutation topology consisting of three identical phase units. Each phase unit consists of two upper and lower bridge arms, each composed of multiple identical sub-modules (SMs) connected in series and a reactor. The upper and lower arms are completely symmetrical. Common sub-modules include half-bridge and full-bridge types, each containing an IGBT (Insulated Gate Bipolar Transistor) semiconductor switching device.

[0003] Since its introduction in the early 21st century, modular multilevel converters (MMCs) have been extensively studied by scholars both domestically and internationally, and are increasingly being applied in areas such as flexible direct current transmission, rail transit, power transmission, and distribution networks. Early research on MMCs focused on transient and steady-state mathematical modeling, modulation control strategies, capacitor voltage balancing algorithms, and strategies for suppressing intra-phase and inter-phase circulating currents. In recent years, with the increasing practical application of MMCs in engineering, research on their reliability has become a hot topic. In particular, the diagnosis of open-circuit faults in IGBTs (integrated bipolar transistors), which have relatively subtle fault characteristics, has attracted widespread attention from scholars both domestically and internationally. In recent years, open-circuit fault diagnosis of MMC submodules has gradually integrated fault detection, faulty submodule location, and internal submodule fault classification into a single, simultaneous process, significantly improving the effectiveness and reliability of fault diagnosis methods.

[0004] Currently, the diagnostic methods for MMC sub-module IGBT open circuit faults are mainly divided into two categories: signal processing-based methods and data-driven methods.

[0005] Signal processing-based fault diagnosis methods are based on the mathematical model of MMC. Through methods such as parameter estimation and observer establishment, the residuals of system parameters are calculated. These residuals are analyzed through methods such as setting thresholds to ultimately obtain diagnostic results. However, MMC is a complex system with high order, nonlinear, and strongly coupled functions, making it difficult to establish a concise and accurate mathematical model. Furthermore, the choice of threshold significantly impacts diagnostic effectiveness. A larger threshold reduces the probability of misjudgment, but also reduces system sensitivity. A smaller threshold increases system sensitivity but also increases susceptibility to interference from factors such as high-frequency noise in the system, leading to an increased probability of misjudgment. These factors contribute to the low robustness of signal processing-based fault diagnosis methods, resulting in significant limitations in practical applications.

[0006] The data-driven fault diagnosis method does not require precise modeling of the MMC. Instead, it pre-extracts fault characteristic variables under normal system operation and fault conditions. Combined with the adaptive classification capability of the data-driven method, a classifier is constructed by constructing methods such as Space Vector Machine (SVM) or Neural Network (NN). The MMC sub-module status is then directly diagnosed through the fault characteristic variables. Although this method does not require the construction of an MMC mathematical model, different selections of fault characteristic variables will lead to different diagnostic effects. At the same time, for high-level MMCs, the existing method needs to collect the sub-module status one by one to achieve fault location, which greatly increases the wiring complexity and input variable dimension, resulting in reduced algorithm robustness.

[0007] In summary, it is currently difficult to establish a concise and accurate mathematical model for MMC. The existing fault diagnosis methods that do not require modeling are greatly affected by the selected fault characteristic variables, and the status of each sub-module needs to be collected during fault location. The operation is complex and the feature dimension is high, which leads to poor robustness and low accuracy of the fault diagnosis results. Summary of the Invention

[0008] To this end, the technical problem to be solved by the present invention is to overcome the problem in the prior art that when locating the fault and identifying the fault type of the MMC, the status of each submodule needs to be collected, the required feature dimension is high, and the diagnostic result accuracy is low.

[0009] To solve the above technical problems, the present invention provides a method for diagnosing open-circuit faults of IGBTs in MMC submodules based on frequency domain analysis, comprising:

[0010] The modular multilevel converter to be monitored is controlled by using a carrier phase shift modulation method to provide a corresponding driving signal phase for each submodule in the modular multilevel converter to be detected;

[0011] Real-time acquisition of the bridge arm voltage signal of each bridge arm on each phase of the modular multilevel converter to be monitored;

[0012] The bridge arm voltage signal is divided using the power frequency period of the modular multilevel converter to be monitored as the time domain segmentation length to obtain multiple bridge arm segment voltage signals;

[0013] The bridge arm segment voltage signal is converted using a fast Fourier transform method to obtain frequency domain information corresponding to each bridge arm segment voltage signal;

[0014] Based on the complex representation of each frequency in the frequency domain information corresponding to each bridge arm segment voltage signal, the amplitude and phase angle corresponding to all frequencies in the frequency domain information are obtained, and the amplitude of the harmonics within the range from the DC component to a preset multiple, as well as the phase angle of the harmonics within a preset interval centered on the carrier frequency, are extracted to construct the fault feature corresponding to each bridge arm segment voltage signal;

[0015] The fault features corresponding to the voltage signal of each bridge arm segment are sequentially input into the trained support vector machine model to obtain the fault label corresponding to the voltage signal of each bridge arm segment; the fault label corresponds one-to-one to the system state of the bridge arm, and each system state includes the location of the fault submodule on the bridge arm and its fault type.

[0016] Preferably, the training process of the support vector machine model includes:

[0017] Acquire various system states of the bridge arms in the modular multilevel converter to be monitored, including normal operation, single-transistor fault, submodule open-circuit fault, dual-transistor fault in the same bridge arm, and dual-transistor fault in different bridge arms;

[0018] Collect the bridge arm voltage signal of any bridge arm of the modular multilevel converter to be monitored in each system state, convert the bridge arm voltage signal using the fast Fourier transform method to obtain the corresponding frequency domain information;

[0019] Based on the complex representation of each frequency in the frequency domain information corresponding to each system state, the amplitude of the harmonics within the range from the DC component to a preset multiple, and the phase angle of the harmonics within a preset interval centered on the carrier frequency are obtained to form the fault characteristics corresponding to each system state;

[0020] Assign corresponding fault labels to the fault characteristics corresponding to each system state;

[0021] A training set is formed based on the fault features and fault labels corresponding to each system state;

[0022] Build includes A support vector machine model with two binary classifiers; The total number of categories representing the system status;

[0023] Based on the training set, the K-fold cross-validation method is used to select the kernel function parameters and the penalty coefficient of the support vector machine model until the preset accuracy is reached to obtain the trained support vector machine model.

[0024] Preferably, after constructing the fault features corresponding to the voltage signals of each bridge arm segment, the method further includes: reducing the dimension of the fault features using principal component analysis to obtain reconstructed fault features after dimensionality reduction; and identifying the fault submodule and the corresponding fault type based on the reconstructed fault features.

[0025] Preferably, the mapping relationship between the reconstructed fault features and the fault features before dimensionality reduction is expressed as:

[0026] , ;

[0027] in, Indicates the Dimensional reconstruction of fault characteristics, Indicates the The transformation coefficient between the fault features reconstructed from the dimension and the fault features before dimension reduction, represents the first Dimensional fault characteristics.

[0028] Preferably, the bridge arm voltage signal is expressed as:

[0029] ;

[0030] in, and Respectively Phase upper bridge arm and The bridge arm voltage of the lower bridge arm, ; and Respectively Phase upper bridge arm and The first bridge arm in the phase The switch function of each submodule; and Respectively Phase upper bridge arm and The first bridge arm in the phase Capacitor voltage of each submodule; Indicates the total number of modules on each bridge arm.

[0031] Preferably, the bridge arm segment voltage signal is converted using a fast Fourier transform method to obtain frequency domain information corresponding to each bridge arm segment voltage signal, which is expressed as:

[0032] ;

[0033] in, Represents the preset rotation factor, the expression is , represents the frequency index coefficient, Indicates the total number of sampling points of Fourier transform; Represents the frequency domain representation after the even sampling point transformation, the expression is , ; Represents the frequency domain representation after the odd sampling point transformation, the expression is .

[0034] Preferably, after converting the bridge arm segment voltage signal using a fast Fourier transform method to obtain frequency domain information corresponding to each bridge arm segment voltage signal, the method further includes:

[0035] The frequency domain information is scaled, and all the eigenvalues in the scaled frequency domain information are converted proportionally to ( , ) interval, expressed as:

[0036] ;

[0037] in, represents the frequency domain information before scaling, represents the scaled frequency domain information, Indicates the preset data lower bound, Indicates the preset data upper bound, Represents the minimum value of this type of feature in all training data, Indicates the maximum value of this type of feature in all training data.

[0038] Preferably, constructing the fault feature corresponding to the voltage signal of each bridge arm segment includes:

[0039] Obtaining a complex number representation of each frequency in the frequency domain information corresponding to the voltage signal of each bridge arm segment;

[0040] Calculate the modulus and of the complex number for each frequency The ratio of the frequency domain information to the harmonics in the range of the DC component to the preset multiple is used as the amplitude feature of the voltage signal of each bridge arm segment; the amplitudes corresponding to all frequencies in the frequency domain information are obtained to form an amplitude set corresponding to the voltage signal of each bridge arm segment; and the amplitudes corresponding to the harmonics in the range of the DC component to the preset multiple are extracted as the amplitude features corresponding to the voltage signal of each bridge arm segment; Indicates the total number of sampling points of Fourier transform;

[0041] Calculate the phase corresponding to each frequency as the phase angle corresponding to the frequency, obtain the phase angles corresponding to all frequencies in the frequency domain information, form a phase angle set corresponding to each bridge arm segment voltage signal, and extract the phase angles corresponding to the harmonics within a preset interval centered on the carrier frequency as the phase angle feature corresponding to each bridge arm segment voltage signal;

[0042] Based on the amplitude characteristics and phase angle characteristics corresponding to the voltage signal of each bridge arm segment, the fault characteristics corresponding to the voltage signal of each bridge arm segment are composed.

[0043] Preferably, obtaining the system status based on the fault characteristics includes:

[0044] Based on the fault characteristics corresponding to the voltage signal of each bridge arm segment, the phase angle of the harmonics within a preset interval centered on the carrier frequency is used to locate the fault and identify the fault submodule of the bridge arm where the voltage signal of each bridge arm segment is located in the current time domain;

[0045] Based on the fault characteristics corresponding to the bridge arm voltage signal of each preset time domain length, fault diagnosis is performed using the amplitude of the harmonics within the range from the DC component to the preset multiple, and the fault type of the fault sub-module of the bridge arm where each bridge arm segment voltage signal is located in the current time domain is obtained.

[0046] This embodiment provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of the method for diagnosing an open-circuit fault of an MMC sub-module IGBT based on frequency domain analysis as described above are implemented.

[0047] The above technical solution of the present invention has the following beneficial effects compared with the prior art:

[0048] The MMC submodule IGBT open circuit fault diagnosis method based on frequency domain analysis described in the present invention uses a carrier phase-shift modulation method to control the monitored modular multi-level converter, so that different submodules have different drive signal phases, and at the same time collects the bridge arm voltage as the electrical parameter required for fault diagnosis; the magnitude of the bridge arm voltage is the sum of the capacitor voltages of all submodules in the open state on the bridge arm. Since different faults will change the port output characteristics of the submodule, causing the internal capacitor voltage of the submodule to change to varying degrees, thereby affecting the bridge arm voltage waveform, combined with the phase of the voltage signal, the faulty submodule can be located and the fault type can be diagnosed. There is no need to collect the capacitor voltage of each submodule, the number of signal collections is small, and the number of sensors required is also correspondingly small. There is no need to install a voltage sensor in each submodule, which saves hardware costs and is applicable to MMCs with different power levels and levels.

[0049] The present invention performs FFT processing on the bridge arm voltage signal to convert the discrete time domain signal into a discrete frequency signal. After the conversion, the noise and accidental interference in the time domain will be eliminated, thereby improving the characterization accuracy of the fault characteristics and the robustness of the fault diagnosis results.

[0050] The present invention uses principal component analysis to perform principal component analysis and feature reconstruction on fault features, deleting some variables with strong linear correlation with other variables to reduce the input dimension of the support vector machine model. While speeding up the operation, it also reduces the interference of excessive feature dimensions on classification, greatly improving the diagnosis rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] In order to make the content of the present invention more clearly understood, the present invention is further described in detail below based on specific embodiments of the present invention in conjunction with the accompanying drawings, wherein:

[0052] Figure 1 This is a flowchart of the steps of the MMC submodule IGBT open circuit fault diagnosis method based on frequency domain analysis provided by the present invention;

[0053] Figure 2 It is a system structure diagram of fault diagnosis;

[0054] Figure 3 It is a schematic diagram of the composition of the fault characteristic parameters of a single sample;

[0055] Figure 4 It is a schematic diagram of the SVM classification principle;

[0056] Figure 5 This is a flow chart of fault detection based on SVM;

[0057] Figure 6 This is the flow chart of SVM parameter optimization under K-fold cross validation;

[0058] Figure 7 This is the CV verification result diagram. DETAILED DESCRIPTION

[0059] The present invention will be further described below with reference to the accompanying drawings and specific embodiments so that those skilled in the art can better understand the present invention and implement it. However, the embodiments are not intended to limit the present invention.

[0060] Reference Figure 1 As shown in FIG, the flowchart of the method for diagnosing an open-circuit fault of an MMC submodule IGBT based on frequency domain analysis provided by the present invention includes the following steps:

[0061] S101: Controlling the modular multilevel converter to be monitored using a carrier phase shift modulation method to provide a corresponding driving signal phase for each submodule in the modular multilevel converter to be detected;

[0062] S102: collecting in real time the bridge arm voltage signal of each bridge arm on each phase of the modular multilevel converter to be monitored, expressed as:

[0063] ;

[0064] in, and Respectively Phase upper bridge arm and The bridge arm voltage of the lower bridge arm, ; and Respectively Phase upper bridge arm and The first bridge arm in the phase The switch function of each submodule; and Respectively Phase upper bridge arm and The first bridge arm in the phase Capacitor voltage of each submodule; Indicates the total number of modules on each bridge arm.

[0065] S103: Dividing the bridge arm voltage signal using the power frequency period of the modular multilevel converter to be monitored as a time domain segmentation length to obtain a plurality of bridge arm segment voltage signals;

[0066] S104: Convert the bridge arm segment voltage signal using a fast Fourier transform method to obtain frequency domain information corresponding to each bridge arm segment voltage signal, expressed as:

[0067] ;

[0068] in, Represents the preset rotation factor, the expression is , represents the frequency index coefficient, Indicates the total number of sampling points of Fourier transform; Represents the frequency domain representation after the even sampling point transformation, the expression is , ; Represents the frequency domain representation after the odd sampling point transformation, the expression is .

[0069] S105: For the complex representation of each frequency in the frequency domain information corresponding to each bridge arm segment voltage signal, obtain the amplitude and phase angle corresponding to all frequencies in the frequency domain information, and extract the amplitude of the harmonics within a range from the DC component to a preset multiple, and the phase angle of the harmonics within a preset interval centered on the carrier frequency, to construct a fault feature corresponding to each bridge arm segment voltage signal;

[0070] S106: Input the fault features corresponding to the voltage signal of each bridge arm segment into the trained support vector machine model in sequence to obtain the fault label corresponding to the voltage signal of each bridge arm segment; the fault label corresponds one-to-one to the system state of the bridge arm, and each system state includes the location of the fault submodule on the bridge arm and its fault type.

[0071] The training process of the support vector machine model includes:

[0072] Acquire various system states of the bridge arms in the modular multilevel converter to be monitored, including normal operation, single-transistor fault, submodule open-circuit fault, dual-transistor fault in the same bridge arm, and dual-transistor fault in different bridge arms;

[0073] Collect the bridge arm voltage signal of any bridge arm of the modular multilevel converter to be monitored in each system state, convert the bridge arm voltage signal using the fast Fourier transform method to obtain the corresponding frequency domain information;

[0074] Based on the complex representation of each frequency in the frequency domain information corresponding to each system state, the amplitude of the harmonics within the range from the DC component to a preset multiple, and the phase angle of the harmonics within a preset interval centered on the carrier frequency are obtained to form the fault characteristics corresponding to each system state;

[0075] Assign corresponding fault labels to the fault characteristics corresponding to each system state;

[0076] A training set is formed based on the fault features and fault labels corresponding to each system state;

[0077] Build includes A support vector machine model with two binary classifiers; The total number of categories representing the system status;

[0078] Based on the training set, the K-fold cross-validation method is used to select the kernel function parameters and the penalty coefficient of the support vector machine model until the preset accuracy is reached to obtain the trained support vector machine model.

[0079] Specifically, in step S105, the construction of the fault characteristics includes:

[0080] S105-1: Obtain a complex number representation of each frequency in the frequency domain information corresponding to the voltage signal of each bridge arm segment;

[0081] S105-2: Calculate the modulus and of the complex number for each frequency The ratio of the frequency domain information to the harmonics in the range of the DC component to the preset multiple is used as the amplitude feature of the voltage signal of each bridge arm segment; the amplitudes corresponding to all frequencies in the frequency domain information are obtained to form an amplitude set corresponding to the voltage signal of each bridge arm segment; and the amplitudes corresponding to the harmonics in the range of the DC component to the preset multiple are extracted as the amplitude features corresponding to the voltage signal of each bridge arm segment; Indicates the total number of sampling points of Fourier transform;

[0082] S105-3: Calculate the phase corresponding to each frequency as the phase angle corresponding to the frequency, obtain the phase angles corresponding to all frequencies in the frequency domain information, form a phase angle set corresponding to each bridge arm segment voltage signal, and extract the phase angles corresponding to the harmonics within a preset interval centered on the carrier frequency as the phase angle feature corresponding to each bridge arm segment voltage signal;

[0083] S105 - 4 : Based on the amplitude characteristics and phase angle characteristics corresponding to the voltage signal of each bridge arm segment, a fault characteristic corresponding to the voltage signal of each bridge arm segment is formed.

[0084] According to the fault feature analysis, the phase angle of the harmonics within the preset interval centered on the carrier frequency mainly contains the difference information between the faults of different submodules. At the same time, the amplitude of the harmonics from the DC component to the preset multiple range mainly contains the difference information between different types of faults (including the open circuit fault type and the fault bridge arm location). After the fault feature sample is constructed, the two types of information are fused into a single fault feature sample to simultaneously detect and locate the fault. Based on the above fault features, the system status is obtained, including:

[0085] Based on the fault characteristics corresponding to the voltage signal of each bridge arm segment, the phase angle of the harmonics within a preset interval centered on the carrier frequency is used to locate the fault and identify the fault submodule of the bridge arm where the voltage signal of each bridge arm segment is located in the current time domain;

[0086] Based on the fault characteristics corresponding to the bridge arm voltage signal of each preset time domain length, fault diagnosis is performed using the amplitude of the harmonics within the range from the DC component to the preset multiple, and the fault type of the fault sub-module of the bridge arm where each bridge arm segment voltage signal is located in the current time domain is obtained.

[0087] The MMC submodule IGBT open circuit fault diagnosis method based on frequency domain analysis described in the present invention uses a carrier phase-shift modulation method to control the monitored modular multi-level converter, so that different submodules have different drive signal phases, and at the same time collects the bridge arm voltage as the electrical parameter required for fault diagnosis; the magnitude of the bridge arm voltage is the sum of the capacitor voltages of all submodules in the open state on the bridge arm. Since different faults will change the port output characteristics of the submodule, causing the internal capacitor voltage of the submodule to change to varying degrees, thereby affecting the bridge arm voltage waveform, combined with the phase of the voltage signal, the faulty submodule can be located and the fault type can be diagnosed. There is no need to collect the capacitor voltage of each submodule, the number of signal collections is small, and the number of sensors required is also correspondingly small. There is no need to install a voltage sensor in each submodule, which saves hardware costs and is applicable to MMCs with different power levels and levels.

[0088] Specifically, after constructing the fault features corresponding to the voltage signals of each bridge arm segment, this embodiment further includes: using principal component analysis to reduce the dimension of the fault features to obtain reconstructed fault features after dimensionality reduction; based on the reconstructed fault features, identifying the fault submodule and the corresponding fault type. The mapping relationship between the reconstructed fault features and the fault features before dimensionality reduction is expressed as:

[0089] , ;

[0090] in, Indicates the Dimensional reconstruction of fault characteristics, Indicates the The transformation coefficient between the fault features reconstructed from the dimension and the fault features before dimension reduction, represents the first Dimensional fault characteristics.

[0091] Specifically, after obtaining the frequency domain information corresponding to the voltage signal of each bridge arm segment, the embodiment of the present invention further includes: scaling the frequency domain information, and converting all characteristic values in the scaled frequency domain information into ( , ) interval, expressed as:

[0092] ;

[0093] in, represents the frequency domain information before scaling, represents the scaled frequency domain information, Indicates the preset data lower bound, Indicates the preset data upper bound, Represents the minimum value of this type of feature in all training data, Indicates the maximum value of this type of feature in all training data.

[0094] The present invention performs FFT processing on the bridge arm voltage signal, converting the discrete time-domain signal into a discrete frequency signal. This conversion eliminates noise and incidental interference in the time domain, thereby improving the accuracy of fault feature characterization and the robustness of fault diagnosis results. Furthermore, the present invention utilizes principal component analysis (PCA) to perform principal component analysis and feature reconstruction on the fault features, deleting some variables with strong linear correlations with other variables to reduce the input dimension of the support vector machine model. This speeds up the model while also reducing the interference of excessive feature dimensions on classification, significantly improving the diagnostic rate.

[0095] Based on the above embodiment, in the embodiment of the present invention, the MMC submodule IGBT open circuit fault diagnosis method based on frequency domain analysis proposed by the present invention is used for fault diagnosis; Figure 2The figure shows the system structure diagram of fault diagnosis, which mainly includes five modules: sample division, FFT conversion, fault feature parameter construction, principal component analysis dimensionality reduction, and SVM fault classification. First, the electrical parameters containing fault characteristics in the MMC system are collected, and the selected parameters are divided in the time domain, and divided into samples of the same length. Then, the amplitude and phase angle of the key frequency bands containing more fault information in each sample are obtained through FFT, and the combination is constructed into the original fault feature parameters. The fault feature parameters are then subjected to principal component analysis, and the feature vector is reconstructed based on the size of the eigenvalue to achieve dimensionality reduction of the fault feature samples. Finally, the diagnosis of specific fault types and modules is achieved through SVM. The specific steps include:

[0096] S201: Fault electrical variable selection and sample division;

[0097] Commonly used electrical quantities for fault detection in MMC systems include: submodule capacitor voltage value, bridge arm current value, output phase voltage value, output phase current value, circulating current value, and bridge arm voltage value. This embodiment does not want to install a voltage sensor at each submodule to increase system cost and potential fault points. Therefore, it is chosen to analyze the remaining electrical quantities and specifically select electrical quantities that contain fault characteristics when a submodule open circuit fault occurs for fault diagnosis.

[0098] As can be seen from the MMC topology diagram, the submodules are connected in series within the MMC bridge arm. Therefore, when the same fault type occurs in different modules, the fault characteristics of the output phase voltage and output phase current are essentially the same. While integrating fault detection and location is desirable, these two cannot be used to diagnose submodule open-circuit faults. While the bridge arm current and circulating current primarily contain fault arm characteristics, meaning they can detect the faulty bridge arm, it is difficult to locate the faulty submodule using only these two. Furthermore, considering that current values are susceptible to load changes, the fault diagnosis model must be adjusted after each load change, making practical use inconvenient. Therefore, the ideal fault characteristic signals are the bridge arm voltage signals of the upper and lower bridge arms. The voltage of a single bridge arm is the sum of the output voltage of the submodule in that bridge arm and the voltage drop across the reactor. When the reactor voltage drop is ignored, the bridge arm voltage is the sum of the submodule output voltages. This voltage is related to the switching state of each submodule in the bridge arm, as expressed by:

[0099] ;

[0100] in, and Respectively Phase upper bridge arm and The bridge arm voltage of the lower bridge arm, ; and Respectively Phase upper bridge arm and The first bridge arm in the phase The switch function of each submodule; and Respectively Phase upper bridge arm and The first bridge arm in the phase Capacitor voltage of each submodule; Indicates the total number of modules on each bridge arm.

[0101] Clearly, the bridge arm voltage at any moment is the sum of the capacitor voltages of all submodules in the on state. Based on the mechanism of a submodule IGBT open-circuit fault, the fault first changes the output characteristics of the module's port, causing an increase in the internal capacitor voltage, which in turn affects the bridge arm voltage waveform of the faulty bridge arm, further impacting the AC output of that phase. Therefore, the bridge arm voltage waveform itself can distinguish faulty bridge arms. When different types of faults occur, the submodule port voltage output patterns differ, and the impact on the bridge arm voltage also varies, allowing the bridge arm voltage to distinguish different types of faults.

[0102] Under the carrier phase-shift modulation selected in this embodiment, each submodule has a unique drive signal phase. When a submodule of the same type fails at a different position, the phase of the voltage abnormality generated on the bridge arm voltage is also related to the drive phase of the specific faulty module. This allows the bridge arm voltage to also distinguish different faulty modules.

[0103] In summary, this embodiment uses the bridge arm voltage as the fault characteristic variable. After data preprocessing such as frequency domain changes, it is constructed into fault characteristic parameters that meet the input requirements of the machine learning method to diagnose various faults that may occur in the MMC system.

[0104] After selecting the fault characteristic variables, we need to determine the time domain length of each sample and divide the raw data. Ideally, the upper and lower bridge arms have the same waveform amplitude, opposite phase, and a period that matches the power frequency period. To facilitate frequency domain analysis, the power frequency period is used as the basis for sample segmentation. In this embodiment, each 0.02s-long bridge arm voltage waveform is divided into one sample.

[0105] S202: Sample frequency domain conversion and fault characteristic parameter construction based on FFT;

[0106] In order to increase the difference in fault characteristics between different faults and improve the accuracy and robustness of the diagnosis method, FFT transformation is performed on the divided time domain samples to convert them into more targeted frequency information to avoid potential noise interference.

[0107] The rotation factors are defined as: , the calculation formula of FFT is expressed as:

[0108] ;

[0109] in, , ;

[0110] Represents the preset rotation factor, the expression is , represents the frequency index coefficient, Indicates the total number of sampling points of Fourier transform; Represents the frequency domain representation after the even sampling point transformation, the expression is , ; Represents the frequency domain representation after the odd sampling point transformation, the expression is .

[0111] After the time domain signal passes through FFT, we get As a result, each point corresponds to a frequency. The specific representative frequency is related to the sampling frequency and the total number of sampling points. Suppose the sampling frequency is The total number of sampling points is , then The frequencies represented by the points are: .

[0112] Each point of the FFT result can be represented by a complex number, and the modulus of the complex number is divided by That is the amplitude of the harmonic corresponding to the point (for DC components, it should be divided by ), the complex phase angle is the phase of the signal at the corresponding frequency.

[0113] Based on the simulation model, the bridge arm voltage was collected for several cycles during normal MMC system operation and when an open-circuit fault occurred in upper bridge arm SMT1. FFT processing was performed to convert the discrete time-domain data into the amplitude and phase angle of each harmonic. Analysis of the amplitude information revealed that when the system transitions from normal operation to a fault, the second harmonic content of the faulty bridge arm increases. Simultaneously, a large harmonic component appears near the carrier frequency, and the DC component increases slightly. Therefore, by comparing the harmonic amplitudes near low frequencies with those near the carrier frequency, it is possible to distinguish between normal system operation and an open-circuit fault in submodule T1.

[0114] The distinction between T1 open circuit fault and T2 open circuit fault can also be made through the amplitude of the harmonic component. Compared with the T1 open circuit fault, the increase of the DC component is more obvious during the T2 open circuit fault, and a higher fourth harmonic appears at the same time. The harmonic components near the carrier frequency are also symmetrical around the carrier frequency, showing a harmonic amplitude characteristic different from that of the T1 open circuit fault.

[0115] However, the harmonic amplitude-based approach described above is difficult to use when distinguishing between different modules experiencing the same fault. Simulation results show that the FFT results of the upper-arm voltage when an open-circuit fault occurs in upper-arm SMT1 (SMT2) are almost identical in amplitude to those in the case of an upper-arm SMT1 fault, making direct differentiation difficult. Considering that when using carrier phase-shift modulation for MMC control, the switching signal timing of each submodule is fixed and the phase difference between them is fixed, abnormal voltages are superimposed on the arm voltage at different times when a fault occurs. This manifests itself in the frequency domain as harmonic phase angle differences. Therefore, the differences in the harmonic phase angles at specific frequencies can be used to distinguish different faulty submodules. Table 1 shows the harmonic phase angles near the carrier frequency when an open-circuit fault occurs in upper-arm SMs 1 and 2.

[0116] Table 1 Harmonic phase angles near the carrier frequency when T1 open circuit fault occurs in different submodules of the upper bridge arm

[0117]

[0118] As shown in Table 1, after several cycles of the fault, the harmonic phase angle near the carrier frequency tends to be stable, and there is a relatively stable difference in the harmonic phase angles between different modules, and the difference is close to the phase difference of the carriers of different submodules.

[0119] In summary, the bridge arm voltage after FFT contains characteristic information that can be used to diagnose and locate the open circuit fault of the MMC submodule. Among them, the amplitude of the harmonics in the frequency band from DC to the carrier frequency can be used to diagnose the fault type, and the phase angle of the harmonics near the carrier frequency can be used to locate the specific fault module. The amplitude of the harmonics in the range of DC component (0Hz) to 60 times the fundamental frequency (3000Hz) is combined with the phase angle of the 23rd harmonic (1150Hz) to the 27th harmonic (1350Hz) near the carrier frequency (1250Hz) to construct the fault characteristic parameters of a single cycle sample. Figure 3 As shown in FIG, a schematic diagram of the composition of fault characteristic parameters of a single sample.

[0120] S203: Data standardization and principal component analysis dimensionality reduction;

[0121] In the harmonic amplitude results obtained by FFT, the proportions of the harmonics vary greatly. Direct comparison can easily cause the features with smaller proportions but more fault information to be masked by the low-frequency features with larger proportions. Therefore, a scaling algorithm is needed to standardize the feature parameters after FFT. The scaling formula is:

[0122] ;

[0123] in, represents the data before scaling, represents the scaled data, Indicates the lower bound of the data specified in the parameter, Indicates the upper bound of the data specified in the parameter, Represents the minimum value of this type of feature in all training data, Represents the maximum value of this type of feature in all training data; let the upper bound of the data be 1 and the lower bound be -1. After scaling, all feature values will be proportionally converted to the interval (-1, 1) according to the size relationship of this type of feature. This format is conducive to data processing by machine learning algorithms such as neural networks or SVM to achieve better classification results.

[0124] In addition, since the harmonic amplitudes of various frequencies in different power frequency cycles are slightly different, in order to better represent the amplitude changes, all harmonic amplitudes are divided by the fundamental frequency amplitude to convert them into harmonic content ratios relative to the fundamental frequency, expressed as: ;

[0125] in, Indicates the harmonic frequency, represents the original harmonic amplitude, represents the fundamental frequency harmonic amplitude, Indicates the relative harmonic content rate after conversion.

[0126] Since all samples use the fundamental frequency amplitude of this period as a reference, the fundamental frequency characteristics of all samples are 1, and the maximum and minimum values are the same. In this case, the scaled characteristic value is 0. Therefore, this column can be deleted in actual processing.

[0127] Taking the normal operation cycle of the system as an example, the fault characteristic parameters of a single sample before and after data standardization are listed as shown in Table 2. The representative sample characteristic parameters, for example, is the DC component of the upper arm voltage, is the fundamental frequency component of the upper arm voltage.

[0128] Table 2 Comparison of fault characteristic parameters before and after data standardization

[0129]

[0130] After using FFT and data normalization, the time-domain bridge arm voltage waveform is converted into frequency-domain samples based on a single sinusoidal cycle. For each sinusoidal cycle, the harmonic amplitude ratios of the upper and lower bridge arm voltages within the first 3000 Hz (60 times the fundamental frequency) and the phase angles of the harmonics near the carrier frequency (1250 Hz) are collected as sample feature parameters. After data normalization and removal of the fundamental frequency components of the upper and lower bridge arms, a single sample still contains 128 features. Excessive features not only increases the algorithm's diagnostic time but can also affect classification accuracy. Therefore, principal component analysis (PCA) can be used to remove highly correlated variables to reduce the input feature dimensionality, improving classification accuracy while also reducing training time.

[0131] Principal component analysis (PCA) is a statistical method for dimensionality reduction. Using orthogonal transformations, it transforms originally correlated random vectors into new, uncorrelated random variables. It also extracts a few minimally correlated variables based on actual needs to maximize the information contained in the original features. The specific steps are: removing the means of each category; calculating the covariance matrix; calculating the eigenvalues and eigenvectors of the covariance matrix; setting a covariance threshold; removing eigenvectors with eigenvalues less than the threshold; calculating a transformation matrix based on the remaining eigenvectors and eigenvalues; and using the transformation matrix to reduce the dimensionality of the input feature matrix.

[0132] After the PCA dimension reduction, the data obtains fewer feature dimensions by reconstructing features. At the same time, since the principle of PCA is to make the sample points obtain the maximum variance on the newly constructed coordinate axis vector, it can also play a certain role in removing noise.

[0133] The scaled fault characteristic parameters obtained in the above process are analyzed using PCA, and the obtained eigenvalues are arranged from large to small. The cumulative proportion of the eigenvalues is shown in Table 3.

[0134] Table 3 Comparison of fault characteristic parameters before and after data standardization

[0135]

[0136] Obviously, the amount of information contained in different dimensions is different. By selecting some dimensions, we can achieve feature dimensionality reduction while retaining most of the information in the original sample. Here, we take the first 100 dimensional features as an example. The selected features have 28 fewer dimensions than the original fault features, but have covered 99.85% of the information in the original features. The mapping relationship between dimensional features and original features is:

[0137] , ;

[0138] in, Indicates the dimensional reconstruction features, Indicates the The transformation coefficient between the dimensional reconstructed features and the original features, Indicates the Dimensional original features.

[0139] Some of the feature transformation coefficients before and after PCA are shown in Table 4:

[0140] Table 4 Feature transformation coefficients before and after PCA (partial)

[0141]

[0142] The coefficient matrix size is 128*100. After the above steps, the original 128-dimensional features can be reconstructed into new 100-dimensional features.

[0143] S204: classifying fault samples using SVM;

[0144] Based on data preprocessing, a support vector machine (SVM) is used to identify and classify MMC neutron module faults. SVM is a machine learning method that can achieve the best possible generalized classification results using a relatively small number of samples. Compared to neural networks, it has advantages such as requiring fewer learning samples, fast learning speed, and the ability to obtain globally optimal solutions. Its basic concept is based on the VC dimension theory and the principle of structural risk minimization in statistical learning theory. It searches for support vectors based on limited sample information and seeks a compromise between model complexity and learning ability, hoping to achieve optimal generalization capabilities.

[0145] The core of the SVM algorithm is to construct the optimal classification surface based on the condition that the sample points are linearly separable. Although SVM was developed as a linear classifier, in fact, through the kernel function method, SVM becomes a nonlinear classifier in essence. Linear separability refers to the The data points in the dimensional real space can be found The two are separated by a dimensional hyperplane. The calculation process of SVM is the process of finding the classification hyperplane. Since there is often more than one hyperplane that meets the classification conditions, the optimization process of SVM is the process of finding the hyperplane (optimal hyperplane) that can maximize the distance between different types of data points. In order to achieve this process, SVM needs to find several sample points in different categories that can help it divide the hyperplane. The hyperplane composed of these sample points achieves the maximum classification interval between different categories. It can be considered that these sample points "support" the classification hyperplane. These points are also called support vectors. Figure 4 As shown in Figure 1, it is a schematic diagram of the SVM classification principle, which shows the relationship between the support vector, classification hyperplane and optimal hyperplane.

[0146] The traditional SVM structure is only for binary classification problems, that is, it can only distinguish whether the sample point belongs to the target class. However, the fault detection of MMC is divided into two steps: detecting the fault type and finding the specific fault module. In order to achieve detection and location at the same time, the combination of submodules and fault types is often directly marked with fault labels in sequence, and then multiple categories of data are simultaneously input into the classifier for multi-classification training. This embodiment uses a one-to-one model to construct a multi-classification SVM. When the sample categories have a common When planting, build There are two binary classifiers, each classifier is used to distinguish between two types.

[0147] Specifically, the training process of SVM can be simplified as the process of dividing hyperplanes in multidimensional space. It is a type of supervised learning. By covering the training set samples of all fault types, support vectors are selected, and the classification hyperplane with the largest sum of the distance between the support vector and the hyperplane is divided. After the division is completed, the relationship between the sample to be classified and the hyperplane is compared to obtain the hyperspace to which the sample belongs and the fault label corresponding to the hyperspace. The correspondence between the system status and the fault label of the upper bridge arm is shown in Table 5 below. The fault label setting method of other bridge arms and other phases can be deduced in the same way. Based on Table 5, it can be seen that the system status and the fault label are in a one-to-one correspondence. Based on the fault label output by the support vector machine model, the system status corresponding to the input fault feature can be obtained, that is, the position of the fault submodule on the bridge arm, and the fault type of the fault submodule.

[0148] Table 5 Correspondence between upper arm system status and fault labels

[0149]

[0150] Reference Figure 5Figure 2 shows a flowchart for fault detection based on an SVM. The steps for fault diagnosis using an SVM include data acquisition, data preprocessing, classifier training, and validation and testing. As can be seen from the flowchart, fault diagnosis generally consists of two phases: training and validation. During training, simulated data of the MMC bridge arm voltage under various fault conditions is first collected and processed using the FFT algorithm to convert the discrete time-domain signals into discrete frequency signals. This conversion eliminates noise and random interference in the time domain, allowing the high-frequency features to better characterize the MMC open-circuit fault. After this data processing, the original bridge arm voltage time-domain signals collected from the upper and lower bridge arms are converted into harmonic amplitude ratios and phase angles with a 50Hz sine wave fundamental frequency and a 3000Hz upper frequency limit. Each feature type is then standardized using all samples as a benchmark. Principal component analysis and feature reconstruction are then performed on the data. Variables with strong linear correlations with other variables are removed to reduce the input dimensionality of the SVM, accelerating its execution while minimizing the interference of excessive feature dimensionality on classification.

[0151] In this embodiment, the kernel function uses the radial basis kernel function (RBF), and the SVM type selects C-SVM, that is, the penalty coefficient is used. As the SVM with optimized parameters, the kernel function parameters need to be adjusted before training. and the penalty coefficient Use K-fold Cross Validation to select the penalty coefficient of SVM. and kernel function parameters Make a selection, refer to Figure 6 The figure shows the flow chart of SVM parameter optimization under K-fold cross validation. To prevent the training time from being too long, this embodiment uses 3-fold CV accuracy as feedback to find the parameters that can make CV accuracy the highest. and On this basis, the parameters are adjusted according to the classification performance of the model on the actual test set.

[0152] The present invention only collects the bridge arm voltage as the electrical parameter required for fault diagnosis in the MMC system, which can realize the diagnosis of the fault type and the location of the specific fault module. The number of signal acquisitions is small, and the number of sensors required is also reduced accordingly. There is no need to install voltage sensors in each submodule, which can save hardware costs; and it has strong scalability and can adapt to MMC systems with different power levels and levels; its data processing flow and algorithm design are applicable to the current MMC system while also being able to smoothly migrate to systems with higher or lower power levels without significantly modifying the algorithm. At the same time, compared with the method of using artificial intelligence approaches such as neural networks as classifiers, the fault diagnosis algorithm proposed by the present invention significantly reduces the number of samples and training time required. The fault samples can also be reduced in dimension through principal component analysis. The sample size and training time can be further adjusted according to the actual accuracy and computing power constraints without almost reducing the diagnostic accuracy, which has strong flexibility.

[0153] Based on the above examples, to verify the effectiveness of the present invention, a simulation model was constructed based on a half-bridge submodule structure. The MMC bridge arm voltage waveforms under normal conditions and various fault conditions were collected and processed according to the method proposed in this invention. A corresponding SVM classifier was constructed, yielding verification results. The samples included 40 cycles under normal operating conditions, 40 cycles under 64 single-transistor fault conditions, 40 cycles under 32 submodule T1 and T2 simultaneous open-circuit fault conditions, and 40 cycles under dual-transistor fault conditions in both the same and different bridge arms, totaling 106 fault types.

[0154] Use 3-fold CV validation to obtain SVM parameters for the training set samples and , first use a coarse grid with a larger interval to get the approximate range of the parameters, and then subdivide it to get the corresponding parameters. Figure 7 As shown in the figure, it is the CV verification result. The average verification accuracy is 89.15%, which corresponds to is 190, g is 0.015, and the training time, training set accuracy, and test set accuracy of the corresponding model are shown in Table 6.

[0155] Table 6 Fault diagnosis results before and after using PCA

[0156]

[0157] Table 6 shows that compared with existing neural network-based diagnostic methods, the proposed strategy significantly reduces model training time, reducing it from hundreds of seconds or even minutes to seconds, while maintaining high fault diagnosis accuracy. Specifically, for the method without PCA, the model training time was only 1.993 seconds. When PCA was used to reduce the data dimension, the SVM training time was further reduced to 1.840 seconds, while maintaining high diagnostic accuracy despite a 22% reduction in sample size.

[0158] Based on the above embodiments, an embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of the MMC sub-module IGBT open circuit fault diagnosis method based on frequency domain analysis as described above are implemented.

[0159] The present invention discloses a method for diagnosing open-circuit faults of MMC submodules based on frequency domain analysis. The method uses a carrier phase shift modulation method to control the monitored modular multilevel converter, so that different submodules have different drive signal phases, and simultaneously collects the bridge arm voltage as the electrical parameter required for fault diagnosis. The bridge arm voltage is the sum of the capacitor voltages of all submodules in the open state on the bridge arm. Since different faults change the port output characteristics of the submodule, the internal capacitor voltage of the submodule changes to varying degrees, thereby affecting the bridge arm voltage waveform. Combined with the phase of the voltage signal, the fault submodule can be located and the fault type can be diagnosed. There is no need to collect the capacitor voltage of each submodule, the number of signal collections is small, the number of sensors required is also small, and there is no need to install a voltage sensor in each submodule, saving hardware cost. The method is applicable to MMCs of different power levels and levels. The present invention performs FFT processing on the bridge arm voltage signal to convert the discrete time domain signal into a discrete frequency signal. After the conversion, the noise and accidental interference in the time domain are eliminated, thereby improving the characterization accuracy of the fault characteristics and improving the robustness of the fault diagnosis results. The present invention uses principal component analysis to perform principal component analysis and feature reconstruction on fault features, deleting some variables with strong linear correlation with other variables to reduce the input dimension of the support vector machine model. While speeding up the operation, it also reduces the interference of excessive feature dimensions on classification, greatly improving the diagnosis rate.

[0160] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0161] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0162] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0163] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 The steps for the function specified in one or more boxes.

[0164] Obviously, the above embodiments are merely examples for clarity of explanation and are not intended to limit the implementation methods. Those skilled in the art will appreciate that other variations or modifications can be made based on the above description. It is not necessary and impossible to enumerate all implementation methods here. Obvious variations or modifications arising therefrom remain within the scope of protection of the present invention.

Claims

1. A method for diagnosing open-circuit faults of MMC submodule IGBTs based on frequency domain analysis, characterized in that: include: The modular multilevel converter to be monitored is controlled by using a carrier phase shift modulation method to provide a corresponding driving signal phase for each submodule in the modular multilevel converter to be detected; Real-time acquisition of the bridge arm voltage signal of each bridge arm on each phase of the modular multilevel converter to be monitored; The bridge arm voltage signal is divided using the power frequency period of the modular multilevel converter to be monitored as the time domain segmentation length to obtain multiple bridge arm segment voltage signals; The bridge arm segment voltage signal is converted using a fast Fourier transform method to obtain frequency domain information corresponding to each bridge arm segment voltage signal; Based on the complex representation of each frequency in the frequency domain information corresponding to each bridge arm segment voltage signal, the amplitude and phase angle corresponding to all frequencies in the frequency domain information are obtained, and the amplitude of the harmonics within the range from the DC component to a preset multiple, as well as the phase angle of the harmonics within a preset interval centered on the carrier frequency, are extracted to construct the fault feature corresponding to each bridge arm segment voltage signal; The fault features corresponding to the voltage signal of each bridge arm segment are sequentially input into the trained support vector machine model to obtain the fault label corresponding to the voltage signal of each bridge arm segment; the fault label corresponds one-to-one to the system state of the bridge arm, and each system state includes the location of the fault submodule on the bridge arm and its fault type.

2. The MMC submodule IGBT open circuit fault diagnosis method based on frequency domain analysis according to claim 1, characterized in that: The training process of the support vector machine model includes: Acquire various system states of the bridge arms in the modular multilevel converter to be monitored, including normal operation, single-transistor fault, submodule open-circuit fault, dual-transistor fault in the same bridge arm, and dual-transistor fault in different bridge arms; Collect the bridge arm voltage signal of any bridge arm of the modular multilevel converter to be monitored in each system state, convert the bridge arm voltage signal using the fast Fourier transform method to obtain the corresponding frequency domain information; Based on the complex representation of each frequency in the frequency domain information corresponding to each system state, the amplitude of the harmonics within the range from the DC component to a preset multiple, and the phase angle of the harmonics within a preset interval centered on the carrier frequency are obtained to form the fault characteristics corresponding to each system state; Assign corresponding fault labels to the fault characteristics corresponding to each system state; A training set is formed based on the fault features and fault labels corresponding to each system state; Build includes A support vector machine model with two binary classifiers; The total number of categories representing the system status; Based on the training set, the K-fold cross-validation method is used to select the kernel function parameters and the penalty coefficient of the support vector machine model until the preset accuracy is reached to obtain the trained support vector machine model.

3. The MMC submodule IGBT open circuit fault diagnosis method based on frequency domain analysis according to claim 1, characterized in that: After constructing the fault features corresponding to the voltage signals of each bridge arm segment, the method also includes: using the principal component analysis method to reduce the dimension of the fault features to obtain the reconstructed fault features after dimensionality reduction; based on the reconstructed fault features, identifying the fault submodules and the corresponding fault types.

4. The MMC submodule IGBT open circuit fault diagnosis method based on frequency domain analysis according to claim 3 is characterized in that: The mapping relationship between the reconstructed fault features and the fault features before dimensionality reduction is expressed as: ， ； in, Indicates the Reconstruct fault characteristics, Indicates the The transformation coefficient between the fault features reconstructed from the dimension and the fault features before dimension reduction, represents the first Dimensional fault characteristics.

5. The MMC submodule IGBT open circuit fault diagnosis method based on frequency domain analysis according to claim 1, characterized in that: The bridge arm voltage signal is expressed as: ； in, and Respectively Phase upper bridge arm and The bridge arm voltage of the lower bridge arm, ; and Respectively Phase upper bridge arm and The first bridge arm in the phase The switch function of each submodule; and Respectively Phase upper bridge arm and The first bridge arm in the phase Capacitor voltage of each submodule; Indicates the total number of modules on each bridge arm.

6. The MMC submodule IGBT open circuit fault diagnosis method based on frequency domain analysis according to claim 1, characterized in that: The fast Fourier transform method is used to transform the voltage signal of the bridge arm segment to obtain the frequency domain information corresponding to the voltage signal of each bridge arm segment, which is expressed as: ； in, Represents the preset rotation factor, the expression is , represents the frequency index coefficient, Indicates the total number of sampling points of Fourier transform; Represents the frequency domain representation after the even sampling point transformation, the expression is , ; Represents the frequency domain representation after the odd sampling point transformation, the expression is .

7. The method for diagnosing an open-circuit fault of an MMC submodule IGBT based on frequency domain analysis according to claim 1, characterized in that: After converting the bridge arm segment voltage signal using the fast Fourier transform method to obtain frequency domain information corresponding to each bridge arm segment voltage signal, the method further includes: The frequency domain information is scaled, and all the eigenvalues in the scaled frequency domain information are converted proportionally to ( , ) interval, expressed as: ； in, represents the frequency domain information before scaling, represents the scaled frequency domain information, Indicates the preset data lower bound, Indicates the preset data upper bound, Represents the minimum value of this type of feature in all training data, Indicates the maximum value of this type of feature in all training data.

8. The method for diagnosing an open-circuit fault of an MMC submodule IGBT based on frequency domain analysis according to claim 1, characterized in that: Construct the fault characteristics corresponding to the voltage signal of each bridge arm segment, including: Obtaining a complex number representation of each frequency in the frequency domain information corresponding to the voltage signal of each bridge arm segment; Calculate the modulus and of the complex number for each frequency The ratio of the frequency domain information to the harmonics in the range of the DC component to the preset multiple is used as the amplitude feature of the voltage signal of each bridge arm segment; the amplitudes corresponding to all frequencies in the frequency domain information are obtained to form an amplitude set corresponding to the voltage signal of each bridge arm segment; and the amplitudes corresponding to the harmonics in the range of the DC component to the preset multiple are extracted as the amplitude features corresponding to the voltage signal of each bridge arm segment; Indicates the total number of sampling points of Fourier transform; Calculate the phase corresponding to each frequency as the phase angle corresponding to the frequency, obtain the phase angles corresponding to all frequencies in the frequency domain information, form a phase angle set corresponding to each bridge arm segment voltage signal, and extract the phase angles corresponding to the harmonics within a preset interval centered on the carrier frequency as the phase angle feature corresponding to each bridge arm segment voltage signal; Based on the amplitude characteristics and phase angle characteristics corresponding to the voltage signal of each bridge arm segment, the fault characteristics corresponding to the voltage signal of each bridge arm segment are composed.

9. The method for diagnosing an open-circuit fault of an MMC submodule IGBT based on frequency domain analysis according to claim 8, characterized in that: Obtain system status based on fault characteristics, including: Based on the fault characteristics corresponding to the voltage signal of each bridge arm segment, the phase angle of the harmonics within a preset interval centered on the carrier frequency is used to locate the fault and identify the fault submodule of the bridge arm where the voltage signal of each bridge arm segment is located in the current time domain; Based on the fault characteristics corresponding to the bridge arm voltage signal of each preset time domain length, fault diagnosis is performed using the amplitude of the harmonics within the range from the DC component to the preset multiple, and the fault type of the fault sub-module of the bridge arm where each bridge arm segment voltage signal is located in the current time domain is obtained.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for diagnosing an open circuit fault of an MMC sub-module IGBT based on frequency domain analysis are implemented.

Images