A method for diagnosing open-circuit faults in IGBT power devices of a single-phase PWM rectifier based on RVFL

Abstract

This invention discloses a method for diagnosing open-circuit faults in IGBT power devices of a single-phase PWM rectifier based on RVFL. This method obtains the original database of IGBT open-circuit faults based on an experimental platform and its digital simulation system, designs fault feature vectors, trains a fault classifier using an RVFL classifier, adjusts the number of nodes and activation function of the network model to save computational resources and time, and uses the trained network model to classify and identify online acquired data samples. This solves the problem of real-time and rapid diagnosis of IGBT open-circuit faults in single-phase PWM rectifiers and is beneficial for the real-time online operation of fault diagnosis algorithms.

Description

Technical Field

[0001] This invention relates to the field of IGBT fault diagnosis, and specifically to a method for diagnosing open-circuit faults in IGBT power devices of a single-phase PWM rectifier based on RVFL. Background Technology

[0002] Single-phase pulse width modulation (PWM) rectifiers, as energy converters between AC and DC sides, are widely used in photovoltaic power generation, railway electric traction transportation, battery charging, and aerospace systems. For example... Figure 2 As shown, the insulated-gate bipolar transistor (IGBT) is a crucial component of a single-phase PWM rectifier, and its operating condition directly affects the energy conversion process. It has been reported that IGBTs have the highest failure rate among components in power electronic systems, and fault diagnosis methods are rapidly developing. Among these methods, detecting and diagnosing IGBT open-circuit faults, which exhibit latent characteristics, is a vital step in ensuring the safe operation of the system and also provides a reference for precise fault-tolerant control after system failures.

[0003] Traditional signal-based fault diagnosis methods extract fault features by processing current and voltage signals in the time and frequency domains to obtain fault information. These methods are often susceptible to noise and load variations, leading to misdiagnosis. Analytical model-based methods typically use mathematical models and physical principles to simulate the actual system and design fault detection and diagnosis strategies based on the residual information between the actual measured output and the analytical model's predicted output. These methods rely on an accurate mathematical model of the system, and parameter changes can affect fault diagnosis performance. Data-driven methods utilize intelligent algorithms (such as neural networks) to learn fault knowledge from large amounts of historical data, obtaining the mapping relationship between fault features and fault modes to achieve fault identification. They have advantages such as being unaffected by load and not relying on an accurate system model, and have the potential to solve the problems of signal-based and analytical model-based diagnostic methods. However, they require a large amount of raw data samples and feature information, and the implementation of intelligent algorithms often requires significant computational resources and long execution times. Furthermore, IGBT open-circuit fault diagnosis requires both real-time speed and high-precision diagnosis, and the control board of a single-phase PWM rectifier system is typically designed only for control and protection, with very limited computational resources and capabilities. Summary of the Invention

[0004] To address the aforementioned shortcomings in existing technologies, this invention provides a method for diagnosing open-circuit faults in single-phase PWM rectifier power devices IGBTs based on RVFL, which solves the problem of long processing time in existing data-driven IGBT open-circuit fault diagnosis methods.

[0005] To achieve the above-mentioned objectives, the technical solution adopted by this invention is as follows:

[0006] A method for diagnosing open-circuit faults in IGBT power devices of a single-phase PWM rectifier based on RVFL is provided, which includes the following steps:

[0007] S1. Based on the experimental prototype platform of the single-phase PWM rectifier system, a digital simulation system was built to obtain grid-side current and DC-side voltage data under different IGBT single-tube open-circuit fault modes, and a historical data training set was constructed.

[0008] S2. Train the RVFL classifier using historical data training set to obtain the trained RVFL classifier;

[0009] S3. Obtain the average value of the grid-side current over half a fundamental cycle, and determine whether the absolute value of the average current is less than 0.1. If yes, proceed to step S4; otherwise, proceed to step S5.

[0010] S4. If the half-cycle of the fundamental frequency is a positive half-cycle, then T2 and T3 are determined to have open circuit faults; if the half-cycle of the fundamental frequency is a negative half-cycle, then T1 and T4 are determined to have open circuit faults.

[0011] S5. Collect and construct a test set based on the grid-side current and DC-side voltage data of the target single-phase PWM rectifier within a quarter of the fundamental frequency cycle. Input the test data into the trained RVFL classifier and obtain the fault labels output by the trained RVFL classifier.

[0012] S6. Determine if the current fault label is greater than 0. If yes, proceed to step S7; otherwise, determine that there is no fault.

[0013] S7. Determine whether the current fault label and the previous fault label are both greater than 0 and not equal. If so, determine that the two non-paired transistors have a double open circuit fault; otherwise, determine that a single transistor fault has occurred.

[0014] Among them, T1 and T2 are located in the upper and lower arms of the a-phase bridge arm, respectively, and T3 and T4 are located in the upper and lower arms of the b-phase bridge arm, respectively.

[0015] Furthermore, the specific method of step S1 includes the following sub-steps:

[0016] S1-1. Based on the experimental prototype platform of the single-phase PWM rectifier system, a digital simulation system is built to obtain grid-side current and DC-side voltage data under different IGBT single-tube open-circuit fault modes, and initial sample data is obtained.

[0017] S1-2. Standardize the initial sample data under the same fault mode to obtain the processed sample data;

[0018] S1-3. The processed sample data is constructed into three different feature vectors according to the following criteria: containing only grid-side current, containing only DC-side voltage, and containing both grid-side current and DC-side voltage. Each feature vector is set with n different dimensions to obtain 3n different feature subsets.

[0019] S1-4. Obtain the feature subset with the highest test accuracy for each fault mode using existing classification algorithms to obtain the training set.

[0020] Furthermore, the specific method for constructing the eigenvector that simultaneously contains grid-side current and DC-side voltage in steps S1-3 is as follows:

[0021] The grid-side current data and DC-side voltage data corresponding to the y-th fault are spliced ​​together after h-dimensional normalization to obtain a 2h-dimensional feature vector of the y-th fault. Then, d 2h-dimensional feature vectors under each fault are obtained. All feature vectors of the same dimension corresponding to the same fault constitute the attribute value of the fault.

[0022] Furthermore, the existing classification algorithms in steps S1-4 include support vector machine classification model, decision tree classification model, random forest classification model, random nearest neighbor classification model, and neural network classification model; the feature subset with the highest accuracy is the feature subset with dimension 200 that simultaneously contains grid-side current and DC-side voltage.

[0023] Furthermore, the specific method of step S2 includes the following sub-steps:

[0024] S2-1, According to the formula:

[0025]

[0026] Construct the input matrix H1 of the RVFL classifier; where x Nd [x] represents the d-th feature vector of the N-th fault sample in the training set; N1 ,x N2 ,...,x Nd [x] is the attribute value of the Nth fault sample. N ;

[0027] S2-2, According to the formula:

[0028]

[0029] Construct the hidden layer matrix H2 of the RVFL classifier; where ω j b represents the weight of the j-th input; j g represents the bias of the j-th hidden layer; g(·) represents the activation function;

[0030] S2-3, According to the formula:

[0031] H = [H1 H2]

[0032] Construct the output matrix H;

[0033] S2-4, Randomly initialize ω j and b j According to the formula:

[0034]

[0035] Obtain the output weight β; where Let H be the Moore-Penrose generalized inverse of the output matrix H; (·) T Y represents the transpose of the matrix; Y is the true label matrix;

[0036] S2-5. Use the current input weights, hidden layer biases, and output weights as parameters for the trained RVFL classifier to obtain the trained RVFL classifier.

[0037] Furthermore, step S5 specifically includes the following sub-steps:

[0038] S5-1. Obtain the grid-side current and DC-side voltage data of the target single-phase PWM rectifier and normalize them;

[0039] S5-2. The grid-side current data and DC-side voltage data of the target single-phase PWM rectifier after 100-dimensional normalization are spliced ​​together to obtain the test set.

[0040] S5-3, According to the formula:

[0041]

[0042]

[0043] Obtain the fault label label(x) output by the trained RVFL classifier; where f(X) is the set of values ​​corresponding to all fault types. i (x) represents the value corresponding to the i-th fault type in f(X); X is the test set, x j For the j-th sample in the test set; β j This represents the j-th element in the output weights; (·) T This represents the transpose of the matrix; D is the total number of hidden nodes in the RVFL classifier. This indicates the fault label corresponding to the fault type with the highest value.

[0044] Furthermore, when a single-tube fault is determined:

[0045] If the fault label label(x) is 1, it corresponds to a fault in single tube T1;

[0046] If the fault label label(x) is 2, it corresponds to a single tube T2 fault;

[0047] If the fault label (x) is 3, it corresponds to a single tube T3 fault;

[0048] If the fault label label(x) is 4, it corresponds to a single tube T4 fault.

[0049] Furthermore, when it is determined that a double-tube open-circuit fault has occurred between two non-paired tubes:

[0050] If the current fault label is 2 and the previous fault label is 1, or the current fault label is 1 and the previous fault label is 2, then it corresponds to an open circuit fault in the non-paired transistors T1 and T2 in the dual-transistor configuration.

[0051] If the current fault label is 3 and the previous fault label is 1, or the current fault label is 1 and the previous fault label is 3, then it corresponds to an open circuit fault in the non-paired transistors T1 and T3 in the dual-transistor configuration.

[0052] If the current fault label is 4 and the previous fault label is 2, or the current fault label is 2 and the previous fault label is 4, then it corresponds to an open circuit fault in the non-paired transistors T2 and T4 in the dual-transistor configuration.

[0053] If the current fault label is 4 and the previous fault label is 3, or the current fault label is 3 and the previous fault label is 4, then the corresponding fault is an open circuit fault in the non-paired transistors T3 and T4 in the dual-tube configuration.

[0054] The beneficial effects of this invention are as follows:

[0055] 1. Based on the experimental platform and its digital simulation system, the original database of IGBT open-circuit faults was obtained. Fault feature vectors were designed, and the fault classifier was trained using the RVFL classifier. The number of nodes and activation function of the network model were adjusted to save computing resources and time. The trained network model was used to classify and identify the data samples acquired online, which solved the problem of real-time and rapid diagnosis of IGBT open-circuit faults in single-phase PWM rectifiers and facilitated the real-time online operation of fault diagnosis algorithms.

[0056] 2. This method has high diagnostic accuracy and can be applied to different load conditions and operating conditions. Attached Figure Description

[0057] Figure 1 This is a flowchart illustrating the method.

[0058] Figure 2 This is a topology diagram of a single-phase PWM rectifier;

[0059] Figure 3 This is a structural diagram of the RVFL classifier model of the present invention;

[0060] Figure 4 This diagram illustrates the changes in fault labels output by the RVFL classifier under varying grid voltage in an experimental prototype system.

[0061] Figure 5 This diagram illustrates the changes in fault labels output by the RVFL classifier under varying load conditions in an experimental prototype system.

[0062] Figure 6 This diagram illustrates the changes in the converter output grid-side voltage and current, DC-side voltage waveform, positive and negative half-cycle current flags, average current value, RVFL classifier recorded value, RVFL classifier output fault label, and fault switch flag before and after a fault in the experimental prototype system.

[0063] Figure 7 This diagram illustrates the changes in the grid-side voltage and current, DC-side voltage waveform, positive and negative half-cycle current flags, average current, RVFL classifier recorded values, RVFL classifier output fault labels, and fault switch flags of the converters before and after simultaneous faults of T1 and T2 in the experimental prototype system.

[0064] Figure 8 This diagram illustrates the changes in the converter output grid-side voltage and current, DC-side voltage waveform, current operating positive and negative half-cycle flags, current average value, RVFL classifier recorded value, RVFL classifier output fault label, and fault switch flag before and after simultaneous faults of T1 and T4 in the experimental prototype system. Detailed Implementation

[0065] The specific embodiments of the present invention are described below to enable those skilled in the art to understand the present invention. However, it should be understood that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, various changes are obvious as long as they are within the spirit and scope of the present invention as defined and determined by the appended claims. All inventions utilizing the concept of the present invention are protected.

[0066] like Figure 1 As shown, the method for diagnosing open-circuit faults in IGBT power devices of a single-phase PWM rectifier based on RVFL includes the following steps:

[0067] S1. Based on the experimental prototype platform of the single-phase PWM rectifier system, a digital simulation system was built to obtain grid-side current and DC-side voltage data under different IGBT single-tube open-circuit fault modes, and a historical data training set was constructed.

[0068] S2. Train the RVFL classifier using historical data training set to obtain the trained RVFL classifier;

[0069] S3. Obtain the average value of the grid-side current over half a fundamental cycle, and determine whether the absolute value of the average current is less than 0.1. If yes, proceed to step S4; otherwise, proceed to step S5.

[0070] S4. If the half-cycle of the fundamental frequency is a positive half-cycle, then T2 and T3 are determined to have open circuit faults; if the half-cycle of the fundamental frequency is a negative half-cycle, then T1 and T4 are determined to have open circuit faults.

[0071] S5. Collect and construct a test set based on the grid-side current and DC-side voltage data of the target single-phase PWM rectifier within a quarter of the fundamental frequency cycle. Input the test data into the trained RVFL classifier and obtain the fault labels output by the trained RVFL classifier.

[0072] S6. Determine if the current fault label is greater than 0. If yes, proceed to step S7; otherwise, determine that there is no fault.

[0073] S7. Determine whether the current fault label and the previous fault label are both greater than 0 and not equal. If so, determine that the two non-paired transistors have a double open circuit fault; otherwise, determine that a single transistor fault has occurred.

[0074] Among them, T1 and T2 are located in the upper and lower arms of the a-phase bridge arm, respectively, and T3 and T4 are located in the upper and lower arms of the b-phase bridge arm, respectively.

[0075] The specific method of step S1 includes the following sub-steps:

[0076] S1-1. Based on the experimental prototype platform of the single-phase PWM rectifier system, a digital simulation system is built to obtain grid-side current and DC-side voltage data under different IGBT single-tube open-circuit fault modes, and initial sample data is obtained.

[0077] S1-2. Standardize the initial sample data under the same fault mode to obtain the processed sample data;

[0078] S1-3. The processed sample data is constructed into three different feature vectors according to the following criteria: containing only grid-side current, containing only DC-side voltage, and containing both grid-side current and DC-side voltage. Each feature vector is set with n different dimensions to obtain 3n different feature subsets.

[0079] S1-4. Obtain the feature subset with the highest test accuracy for each fault mode using existing classification algorithms to obtain the training set.

[0080] The specific method for constructing the eigenvector that simultaneously contains grid-side current and DC-side voltage in step S1-3 is as follows:

[0081] The grid-side current data and DC-side voltage data corresponding to the y-th fault are spliced ​​together after h-dimensional normalization to obtain a 2h-dimensional feature vector of the y-th fault. Then, d 2h-dimensional feature vectors under each fault are obtained. All feature vectors of the same dimension corresponding to the same fault constitute the attribute value of the fault.

[0082] In steps S1-4, the existing classification algorithms include support vector machine classification model, decision tree classification model, random forest classification model, random nearest neighbor classification model, and neural network classification model; the feature subset with the highest accuracy is the feature subset with dimension 200 that simultaneously contains grid-side current and DC-side voltage.

[0083] The specific method of step S2 includes the following sub-steps:

[0084] S2-1, According to the formula:

[0085]

[0086] Construct the input matrix H1 of the RVFL classifier; where x Nd [x] represents the d-th feature vector of the N-th fault sample in the training set; N1 ,x N2 ,...,x Nd [x] is the attribute value of the Nth fault sample. N ;

[0087] S2-2, According to the formula:

[0088]

[0089] Construct the hidden layer matrix H2 of the RVFL classifier; where ω j b represents the weight of the j-th input; j g represents the bias of the j-th hidden layer; g(·) represents the activation function;

[0090] S2-3, According to the formula:

[0091] H = [H1 H2]

[0092] Construct the output matrix H;

[0093] S2-4, Randomly initialize ω j and b j According to the formula:

[0094]

[0095] Obtain the output weight β; where Let H be the Moore-Penrose generalized inverse of the output matrix H; (·) T Y represents the transpose of the matrix; Y is the true label matrix;

[0096] S2-5. Use the current input weights, hidden layer biases, and output weights as parameters for the trained RVFL classifier to obtain the trained RVFL classifier.

[0097] Step S5 specifically includes the following sub-steps:

[0098] S5-1. Obtain the grid-side current and DC-side voltage data of the target single-phase PWM rectifier and normalize them;

[0099] S5-2. The grid-side current data and DC-side voltage data of the target single-phase PWM rectifier after 100-dimensional normalization are spliced ​​together to obtain the test set.

[0100] S5-3, According to the formula:

[0101]

[0102]

[0103] Obtain the fault label label(x) output by the trained RVFL classifier; where f(X) is the set of values ​​corresponding to all fault types. i (x) represents the value corresponding to the i-th fault type in f(X); X is the test set, x j For the j-th sample in the test set; β j This represents the j-th element in the output weights; (·) T This represents the transpose of the matrix; D is the total number of hidden nodes in the RVFL classifier. This indicates the fault label corresponding to the fault type with the highest value.

[0104] When the fault is determined to be a single tube:

[0105] If the fault label label(x) is 1, it corresponds to a fault in single tube T1;

[0106] If the fault label label(x) is 2, it corresponds to a single tube T2 fault;

[0107] If the fault label (x) is 3, it corresponds to a single tube T3 fault;

[0108] If the fault label label(x) is 4, it corresponds to a single tube T4 fault.

[0109] When it is determined that two non-paired transistors have a double-transistor open-circuit fault:

[0110] If the current fault label is 2 and the previous fault label is 1, or the current fault label is 1 and the previous fault label is 2, then it corresponds to an open circuit fault in the non-paired transistors T1 and T2 in the dual-transistor configuration.

[0111] If the current fault label is 3 and the previous fault label is 1, or the current fault label is 1 and the previous fault label is 3, then it corresponds to an open circuit fault in the non-paired transistors T1 and T3 in the dual-transistor configuration.

[0112] If the current fault label is 4 and the previous fault label is 2, or the current fault label is 2 and the previous fault label is 4, then it corresponds to an open circuit fault in the non-paired transistors T2 and T4 in the dual-transistor configuration.

[0113] If the current fault label is 4 and the previous fault label is 3, or the current fault label is 3 and the previous fault label is 4, then the corresponding fault is an open circuit fault in the non-paired transistors T3 and T4 in the dual-tube configuration.

[0114] In one embodiment of the present invention, such as Figure 3 As shown, the RVFL classifier model of this invention includes an input layer, a hidden layer, and an output layer. The dimension of the feature vectors corresponding to the training set can be set to 50, 100, 150, and 200 dimensions, thus constructing 12 different feature subsets. Through existing classification algorithms, the 200-dimensional feature vector that simultaneously includes grid-side current and DC-side voltage achieves the highest accuracy; therefore, this type of feature vector is used to construct the feature subset.

[0115] Pre-trained RVFL classifier models Model 1, Model 2, Model 3, and Model 4, with the activation function "sigmoidal" and hidden layer nodes of 20, 40, 80, and 160 respectively, were compiled into the controller. The computation times for the first three models were calculated to be 22.38 μs, 41.7 μs, and 79.48 μs, respectively. Model 4 could not be computed due to insufficient memory. Pre-trained models Model 5, Model 6, and Model 7, with 20 hidden layer nodes and activation functions "sine", "radbas", and "tribas" respectively, had computation times of 22.74 μs, 22.99 μs, and 20.11 μs, respectively. It can be seen that the computation time doubles with the increase in the number of hidden layer nodes, while the differences between activation functions are relatively small.

[0116] By adjusting the number of hidden layer nodes and the activation function parameters of the RVFL classifier model, the problem of insufficient computational resources and capabilities can be solved. The basic idea is to use a five-fold cross-validation method to obtain different pre-trained models based on a selected training set, and then test them on the same test set to calculate their test accuracy. Considering both test accuracy and computation time requirements, the final number of hidden nodes is chosen to be 20, and the activation function is "sigmoidal". At this point, the RVFL classifier model can achieve a test accuracy of 99.5%, while also reducing the computational load for the real-time execution of subsequent diagnostic strategies. The RVFL classifier model trained at this point is defined as the RVFL classifier.

[0117] When a single-phase PWM rectifier experiences an open-circuit fault in its IGBT, it disrupts the bidirectional energy transfer balance between the AC and DC sides, causing distortion in the grid-side current and DC-side voltage. Based on the differences in fault characteristics, single-transistor and dual-transistor open-circuit faults in IGBTs can be categorized into three fault scenarios:

[0118] The first fault scenario is a single-tube fault, where the fault characteristics only appear during the positive or negative half-cycle of the current waveform.

[0119] Fault scenario two is an open circuit fault in two non-paired tubes in a dual-tube configuration, including T1&T2, T1&T3, T2&T4, and T3&T4. Its fault characteristics are the superposition of the fault characteristics of the two single-tube configurations in fault scenario one.

[0120] The third fault scenario involves open-circuit faults in two pairs of transistors in a dual-transistor configuration, including T1 & T4 and T2 & T3. An open-circuit fault in T1 & T4 causes the negative half-cycle of the current waveform to disappear, while an open-circuit fault in T2 & T3 causes the positive half-cycle of the current waveform to disappear. Therefore, diagnosis can be made by calculating the average current value within the sampling window and determining whether the current waveform is in a positive or negative half-cycle.

[0121] During real-time sampling, the sampling frequency is 20kHz. The sampling start point within one fundamental cycle is obtained by determining the zero-phase point of the grid-side current. After data sampling within a quarter of the fundamental cycle is completed, the grid-side current and DC-side voltage data are processed into a test set for diagnosis. The fault diagnosis logic and process designed based on three fault scenarios are used to comprehensively output the diagnostic results, which are updated after the next data window sampling is completed.

[0122] In the specific implementation process, the hardware structure used is a hardware test platform with an RT-Box controller and a physical main circuit. Programmable power supplies, power modules and their drive circuits, resistors, capacitors, and other components are used according to... Figure 2The main circuit model of the topology shown is built. The control circuit of the single-phase PWM rectifier is built on the RT-Box platform. The control strategy is transient current control. The switching frequency is set to 1KHz, the effective value of the grid side voltage is 60V, and the DC side reference voltage is 120V. The trained RVFL classifier and fault diagnosis algorithm are pre-embedded in the RT-Box for online fault diagnosis.

[0123] An open-circuit fault in the switching transistor is simulated by shielding the drive signal in the RT-Box controller. The sampling frequency is 20kHz. During real-time sampling, the zero-phase point of the grid-side current is used as the sampling start point within one fundamental cycle. The grid-side current and DC-side voltage data are collected and stored. After the data sampling within a quarter of the fundamental cycle is completed, the grid-side current and DC-side voltage data are normalized, the average current value is calculated, and subsequent diagnostic procedures are performed.

[0124] Test results of single-phase PWM rectifier under normal operation and different switching transistor faults are as follows: Figure 4 , Figure 5 , Figure 6 , Figure 7 and Figure 8 As shown, where Figure 4 The test results for the system outputting fault labels under load changes from 20Ω to 40Ω and then back to 20Ω are given. Figure 5 The test results show the system outputting fault tags under the condition of mains voltage changes from 60V to 70V and then back to 60V. Figure 6 The results of the diagnostic test for an open-circuit fault in T1 under rated power conditions are as follows. Figure 7 The results are diagnostic test findings for open circuits of T1 and T2 under rated power conditions. Figure 8 The results show the diagnostic test findings for open-circuit faults in IGBTs T1 and T4. It can be seen that this method can effectively diagnose IGBT open-circuit faults and meets the real-time requirements of online diagnosis.

[0125] In summary, this invention obtains the original database of IGBT open-circuit faults based on an experimental platform and its digital simulation system, designs fault feature vectors, trains a fault classifier using the RVFL classifier, adjusts the number of nodes and activation function of the network model to save computational resources and time, and uses the trained network model to classify and identify the online data samples. This solves the problem of real-time and rapid diagnosis of open-circuit faults in single-phase PWM rectifiers and is beneficial for the real-time online operation of fault diagnosis algorithms.

Claims

1. A method for diagnosing open-circuit fault of power device IGBT of single-phase PWM rectifier based on RVFL, characterized in that, The method comprises the following steps: S1, a digital simulation system is built based on a single-phase PWM rectifier system experimental prototype platform, and grid-side current and DC-side voltage data under different IGBT single tube open circuit fault modes are obtained to construct a historical data training set; S2, the RVFL classifier is trained through the historical data training set to obtain the trained RVFL classifier; S3, the average value of the grid-side current in a half fundamental period is obtained, and whether the absolute value of the current average value is less than 0.1 is determined, if yes, step S4 is entered; otherwise, step S5 is entered; S4, if the half fundamental period is a positive half cycle, then determine and an open circuit fault has occurred; if the half fundamental period is a negative half cycle, then determine and an open circuit fault has occurred; S5, the grid-side current and DC-side voltage data of the target single-phase PWM rectifier in a quarter of a fundamental period are collected and used as test data to construct a test set, and the test data are input into the trained RVFL classifier to obtain the fault label output by the trained RVFL classifier; S6, whether the current fault label is greater than 0 is determined, if yes, step S7 is entered; otherwise, it is determined that there is no fault; S7, whether the current fault label and the previous fault label are both greater than 0 and not equal is determined, if yes, it is determined that double tube open circuit faults occur in two non-pairs of tubes; otherwise, it is determined that single tube faults occur; wherein, and are located in a the upper arm and the lower arm of the phase bridge arm, and are located in b the upper arm and the lower arm of the phase bridge arm; The specific method of step S1 comprises the following sub-steps: S1-1, a digital simulation system is built based on a single-phase PWM rectifier system experimental prototype platform, and grid-side current and DC-side voltage data under different IGBT single tube open circuit fault modes are obtained to obtain initial sample data; S1-2, the initial sample data under the same fault mode are normalized to obtain processed sample data; S1-3, the processed sample data is constructed into three different feature vectors according to containing only grid-side current, containing only DC-side voltage, and containing both grid-side current and DC-side voltage, and each feature vector is set n to different dimensions, and three n different feature subsets are obtained; S1-4, the feature subset with the highest test accuracy under each fault mode is obtained through an existing classification algorithm to obtain a training set; The specific method for constructing the feature vector containing the grid-side current and the DC-side voltage in step S1-3 is: The first y Type of fault corresponding h Grid-side current data after standardization and h The DC-side voltage data after being normalized to a specific value are spliced ​​together to obtain the first... y One of the two types of faults h 3D feature vectors are used to obtain the features under each fault condition. d 2 h A dimensional feature vector; where all feature vectors of the same dimension corresponding to the same type of fault constitute the attribute value of that fault; The existing classification algorithm in step S1-4 includes a support vector machine classification model, a decision tree classification model, a random forest classification model, a random neighbor classification model and a neural network classification model; the feature subset with the highest accuracy is a feature subset containing the grid-side current and the DC-side voltage and having a dimension of 200.

2. The RVFL-based single-phase PWM rectifier power device IGBT open-circuit fault diagnosis method according to claim 1, characterized in that, The specific method of step S2 comprises the following sub-steps: S2-1, according to the formula: Input matrix for constructing the RVFL classifier ; wherein denotes the i-th feature vector of the j-th fault sample in the training set N d N ;​​​​ S2-2, according to the formula: a hidden layer matrix of a constructed RVFL classifier ; wherein denotes the j input weight; denotes the j bias of the denotes an activation function; S2-3, according to the formula: Constructing an output matrix ; S2-4, Random initialization and , according to the formula: obtaining output weights ; wherein is a Moore-Penrose generalized inverse of the output matrix ; denotes the transpose of a matrix; is a true label matrix; S2-5, the current input weight, the bias of the hidden layer and the output weight are taken as the trained RVFL classifier parameters to obtain the trained RVFL classifier.

3. The RVFL-based single-phase PWM rectifier power device IGBT open-circuit fault diagnostic method according to claim 2, characterized in that, Step S5 specifically comprises the following sub-steps: S5-1, the grid-side current and DC-side voltage data of the target single-phase PWM rectifier are obtained and normalized; S5-2, the 100-dimensional normalized grid-side current data of the target single-phase PWM rectifier and the 100-dimensional normalized DC-side voltage data of the target single-phase PWM rectifier are spliced to obtain a test set; S5-3, according to the formula: obtaining a fault label output by the trained RVFL classifier ; wherein is a set of values corresponding to all fault types, is a value corresponding to the i th fault type in is a test set, is the j th sample in the test set; is the j th element in the output weight; denotes the transpose of a matrix; D is the total number of hidden nodes in the RVFL classifier; denotes taking the fault label corresponding to the fault type with the largest value.

4. The RVFL-based single-phase PWM rectifier power device IGBT open-circuit fault diagnostic method according to claim 3, characterized in that, When it is determined to be a single tube fault: if the fault flag is 1, then the corresponding single tube fault; if the fault label is 2, then the corresponding single tube fault; if the fault label is 3, then the corresponding single tube fault; If the fault flag is 4, then the corresponding single tube fails.

5. The RVFL-based single-phase PWM rectifier power device IGBT open-circuit fault diagnostic method according to claim 4, characterized in that, When it is determined that double tube open circuit faults occur in two non-pairs of tubes: If the current fault label is 2 and the previous fault label is 1, or the current fault label is 1 and the previous fault label is 2, then the non-mate of the corresponding pair of tubes and open circuit fault; If the current fault label is 3 and the previous fault label is 1, or the current fault label is 1 and the previous fault label is 3, then the non-mate of the corresponding pair of tubes and open circuit fault; If the current fault label is 4 and the previous fault label is 2, or the current fault label is 2 and the previous fault label is 4, then the non-mate of the corresponding pair of tubes and open circuit fault; If the current fault tag is 4 and the previous fault tag is 3, or the current fault tag is 3 and the previous fault tag is 4, then the non-mate of the corresponding pair of tubes is open circuit fault. and open circuit fault.

Images