Four-quadrant rectifier fault diagnosis method and device based on energy entropy optimization multi-feature fusion, medium and equipment

Abstract

The present application belongs to the field of high-speed train fault diagnosis, and discloses a four-quadrant rectifier fault diagnosis method, device, medium and equipment based on energy entropy optimization multi-feature fusion, which comprises the following steps: firstly, wavelet packet decomposition is carried out on the fault signal by using different wavelet functions, and the optimal frequency band is selected according to the energy-information entropy ratio of the frequency band coefficient; secondly, the energy entropy of each coefficient in the optimal frequency band is calculated to obtain multiple energy entropy features; finally, the multiple fault features are locally arranged in the space position to optimize the data distribution of the multiple features, reduce the abnormal values and eliminate the redundancy among the multiple features. The results show that the average diagnosis accuracy of the four-quadrant rectifier is 99.0625% through the optimization and fusion of the multiple features. Compared with other methods, the fault diagnosis method has higher diagnosis rate and stronger robustness in output voltage, noise, training and test ratio.

Description

Technical Field

[0001] The present invention relates to the field of high-speed train fault diagnosis, and in particular to a four-quadrant rectifier fault diagnosis method, device, medium and equipment based on energy entropy optimization and multi-feature fusion. Background Art

[0002] The four-quadrant rectifier is an important component of the traction drive system of heavy-duty trains. It uses the pulse width modulation (PWM) method to enable the rectifier to work in both rectification and inverter states. [1] , the power factor is close to 1, which reduces the power consumption and improves the conversion rate and utilization rate of power in the process of traction and regenerative braking. During the operation of heavy-load trains, the four-quadrant rectifier is often subjected to high temperature, high voltage, overcurrent and other problems, which are prone to failure. [2] Such failures can lead to unstable intermediate voltage in the traction converter, distorted input current, increased harmonics, and failure of regenerative braking, seriously impacting the safe operation of the train. Therefore, efficient fault diagnosis methods can effectively prevent catastrophic train accidents and improve train operation and maintenance capabilities.

[0003] Fault diagnosis of four-quadrant rectifiers falls within the scope of fault diagnosis of power converters. The main sources of faults are insulated gate bipolar transistors (IGBTs) and power diodes. Existing fault diagnosis methods mainly include model-based methods, signal-based methods, and data-driven methods.

[0004] The model-based approach is a method that aims to establish a mathematical model of the system and make the diagnostic system have a faster response speed by detecting the difference between the actual signal and the signal predicted by the model. [3] . Reference [4] combines the system's hybrid logic dynamic model with a sliding mode observer to establish a state estimator, completing the detection and location of the open-circuit fault of the switching device. Reference [5] uses four observers to diagnose the open-circuit faults of four IGBTs in the rectifier. Although these methods have obvious advantages in diagnostic speed, the establishment of the model and the determination of parameters limit their applicability. The signal-based method does not require a mathematical model of the system. It extracts the symptoms and characteristics of the measured signal through signal processing technology to obtain the diagnostic results. [6]Reference [7] uses the analytical redundancy relationship under rectifier fault conditions to determine the threshold of each residual by the mean and variance level of each residual under normal conditions, and completes the fault diagnosis of IGBT. Reference [8] analyzes the instantaneous input voltage of the rectifier to detect and locate the open circuit fault of IGBT under all switching frequencies and different load conditions. Although these methods improve the diagnosis rate, they require rich prior knowledge and theoretical analysis of fault signals.

[0005] With the development of artificial intelligence and big data, fault diagnosis methods have begun to focus on data-driven methods. Data-driven methods avoid dependence on system models and analysis of prior knowledge of signals. They diagnose fault types through signal processing, feature extraction, and pattern recognition, and have strong robustness and dynamic characteristics. [9] . Reference

[10] uses a convolutional neural network to directly use the excitation current as the input feature to achieve the extraction and classification of rectifier fault features. Reference

[11] uses the particle swarm algorithm to optimize the deep belief network structure for the fault feature extraction of the generator rotating rectifier. Although references

[10] and

[11] have achieved a high fault diagnosis rate, they both use deep learning theory to directly train and identify fault signals, ignoring the problems of large computational complexity and complex model design. Therefore, in the data-driven method, simple and sensitive fault features can reduce the computational burden of pattern recognition and improve the fault diagnosis rate.

[0006] Reference

[12] uses wavelet packet decomposition to calculate the wavelet packet entropy features of the frequency band coefficients, optimizes the fault features through principal component analysis (PCA), and realizes the open circuit fault diagnosis of IGBT. However, PCA has a good effect on the fusion of linear data, but a poor fusion effect on nonlinear data. Reference

[13] fuses the energy spectrum entropy features through kernel principal component analysis (KPCA), and finally uses the wavelet neural network method to identify the fault category. Although KPCA can be used for the fusion of nonlinear data, it ignores the influence of the choice of kernel function in KPCA on the fusion effect. Reference

[14] uses the positioning analysis method to fuse the energy entropy features and identifies the fault category through the stacked sparse autoencoders (SSAE) network method. However, the setting of the positioning parameters is complex and uncertain, and the use of deep learning algorithms increases the difficulty of the fault diagnosis method.

[0007] The fault signal of the four-quadrant rectifier of a heavy-duty train has the characteristics of nonlinearity and non-stationarity. In order to extract efficient fault features, it is necessary to consider not only the effective decomposition of the signal but also the sensitivity of the features. Wavelet packet decomposition can not only achieve multi-resolution decomposition in the time-frequency domain to obtain high-frequency and low-frequency components, but also has a certain robustness to noise. Although references

[12] and

[14] use wavelet packet decomposition, they ignore the different decomposition effects of different wavelet functions on the signal, which affects the fault diagnosis results. Entropy-based features can be used to characterize the nonlinear characteristics caused by transient changes in the system, thereby quantifying the dynamic changes of the fault signal and distinguishing different operating states. Therefore, entropy-based features have the advantages of good clustering ability, high classification accuracy, and strong robustness. Although references

[12] to

[14] all use entropy features, the method for obtaining simple sensitive fault features needs to be improved. Summary of the Invention

[0008] Embodiments of the present invention provide a four-quadrant rectifier fault diagnosis method, apparatus, medium, and device based on energy entropy optimization and multi-feature fusion to address the low fault diagnosis accuracy of simple, sensitive fault signatures in the prior art. To provide a basic understanding of some aspects of the disclosed embodiments, a brief summary is provided below. This summary is not intended to be a comprehensive review, identify key / important elements, or delineate the scope of protection of these embodiments. Its sole purpose is to present some concepts in a simplified form as a prelude to the detailed description that follows.

[0009] According to a first aspect of an embodiment of the present invention, a four-quadrant rectifier fault diagnosis method based on energy entropy optimization and multi-feature fusion is provided.

[0010] In one embodiment, the method comprises the following steps:

[0011] Collect input current signals of fault modes under different working conditions to form an original sample set;

[0012] Using different wavelet functions, the input current signal of the fault mode is decomposed by wavelet packet multi-resolution to obtain different frequency band information. The wavelet function suitable for the fault signal is selected according to the energy-information entropy ratio of each frequency band. The corresponding frequency band is the optimal frequency band, and the energy entropy of each frequency band coefficient is calculated.

[0013] According to the energy entropy of each frequency band coefficient, the energy entropy characteristics of the optimal frequency band are obtained, and the energy entropy characteristics corresponding to different working conditions and different fault modes are constructed;

[0014] Perform local tangent space arrangement optimization on multiple redundant and conflicting energy entropy features in the data space to obtain the optimal energy entropy feature;

[0015] Support vector machine is used for pattern recognition to obtain the fault diagnosis results of four-quadrant rectifier.

[0016] Based on the above scheme, different wavelet functions are used to perform wavelet packet multi-resolution decomposition on the input current signal of the fault mode to obtain different frequency band information. The wavelet function suitable for the fault signal is selected based on the energy-information entropy ratio of each group of frequency bands. The corresponding frequency band is the optimal frequency band. The steps of calculating the energy entropy of each frequency band coefficient specifically include:

[0017] (1) Calculation of multi-band coefficients of different wavelet functions: Assuming the fault signal ,Using 9 Daubechies (dbN, N= 2,3,…10) wavelet functions commonly used for signal decomposition and reconstruction, the fault signal is decomposed by wavelet packet to obtain 9 groups of frequency band information;

[0018] (1)

[0019] Where, Indicates the use of db N Wavelet function, signal passes through wavelet packet r After layer decomposition, we get m In the frequency band n The coefficients of the frequency bands;

[0020] (2) Select a suitable wavelet function:

[0021] The frequency band energy value corresponding to the wavelet function after wavelet packet decomposition:

[0022] (2)

[0023] After a system failure occurs, the uncertainty of each failure mode is related to the number of times the corresponding failure mode occurs. If a failure mode of the rectifier is Y i , the probability of this failure mode occurring is P ( Y i ), then the information entropy of the frequency band coefficient obtained after the signal of the fault mode is decomposed is :

[0024] (3)

[0025] The energy-information entropy ratio of the corresponding frequency band is R :

[0026] (4)

[0027] The larger the energy-information entropy ratio is, the more suitable the selected wavelet function is for the fault signal, and the frequency band obtained after decomposition is the optimal frequency band;

[0028] (3) Calculate the energy entropy characteristics of the optimal frequency band: the fault signal is decomposed by wavelet packet r After layer decomposition, m frequency band coefficients, and calculate the energy entropy of each frequency band coefficient in the frequency band respectively;

[0029] (5)

[0030] (6)

[0031] (7)

[0032] In the formula, n The energy value of the frequency band coefficient is E ( n ), the probability distribution is p (n), energy entropy is T (n);

[0033] Therefore, based on the optimal wavelet function, the energy entropy characteristics of the optimal frequency band obtained by decomposing the fault signal through wavelet packets are: T ( n ):

[0034] (8)

[0035] (4) Constructing the domain matrix of energy entropy characteristics: Due to the energy entropy characteristics T for m dimensional matrix, T n is any data point in the matrix, m is the number of data points, and the number of nearest neighbors for each data point is determined as k , forming a neighborhood matrix;

[0036] (9)

[0037] Where, For the n The data point j Nearest neighbors, n =1,2,…, N ; j =1,2,…, k ;

[0038] (5) Calculate local coordinates: At the data point T n Find the orthogonal basis in the field matrix Qi to constitute g dimensional tangent space, in the neighborhood, calculate each point T nj Orthogonal projection to tangent space λ nj , then find the neighborhood matrix T mn The local coordinates of n :

[0039] (10)

[0040] (6) Establish global coordinates: Set global coordinates , for energy entropy characteristics T ( n ) is arranged in tangent space to obtain the global coordinate system. The global coordinates reflect the geometric structure of the local coordinate matrix and should meet the following requirements:

[0041] (11)

[0042] Where, is the mapping matrix to be determined; is the reconstruction error; yes The centralization matrix;

[0043] (7) Calculate the permutation matrix: Assume , S is a 0-1 selection matrix, is the energy entropy characteristic data point T n The weight matrix of , the permutation matrix is :

[0044] (12)

[0045] (8) Minimize the reconstruction error: Solve the minimum eigenvalue of the permutation matrix. The eigenvector corresponding to the second to (g+1) minimum eigenvalues ​​obtained by calculating the permutation matrix is ​​the energy entropy feature. T ( n ) is the global coordinate of each data point in the , which is also the feature vector after the energy entropy feature is optimized by local tangent space arrangement. :

[0046] (13).

[0047] Based on the above solution, the step of performing local tangent space arrangement optimization on multiple redundant and conflicting energy entropy features in the data space to obtain the optimal energy entropy feature specifically includes:

[0048] (1) Selection of optimal wavelet basis function:

[0049] Considering the degree of fault feature mining and computational cost, the fault signal is decomposed by wavelet packet decomposition. The Daubechies wavelet system, which is commonly used for signal decomposition and reconstruction, is selected. The wavelet basis functions of the Daubechies wavelet system are used to decompose the signal. Multiple frequency bands are obtained for each sample and each fault mode. The energy-information entropy ratio of each frequency band is calculated and compared. The largest ratio indicates that the selected wavelet function is optimal and the frequency band obtained from the decomposition of the corresponding fault signal is optimal.

[0050] (2) Extraction of energy entropy features and data analysis:

[0051] For each operating condition, the multiple fault modes are decomposed into wavelet packets using their respective optimal wavelet functions to obtain multiple frequency bands, and the energy entropy characteristics of each frequency band coefficient are calculated.

[0052] (3) Optimization of energy entropy characteristics:

[0053] During the feature extraction process, the feature data presents a normal distribution, and the extracted multiple energy entropy feature data are arranged in a local tangent space.

[0054] Based on the above solution, the step of using a support vector machine to perform pattern recognition and obtain a fault diagnosis result of a four-quadrant rectifier specifically includes:

[0055] Support vector machine is used as the classifier of fault features, in which the kernel function selects the radial basis function which is robust to noise. Gaussian white noise is added to the collected original signal to simulate the real working environment of the four-quadrant rectifier. Through signal decomposition and feature extraction, the optimized energy entropy feature is used as the input of the classifier SVM. The test results are compared with the actual results of the trained SVM classifier to give the fault diagnosis results.

[0056] Based on the above solution, the step of using a support vector machine as a classifier for fault features specifically includes:

[0057] The fault feature vectors are randomly divided into training samples and test samples, and the support vector machine classifier is used for fault diagnosis. The training samples are used to train the support vector machine, and the test samples are used to verify the performance of the trained support vector machine classifier, thereby obtaining the diagnosis results.

[0058] According to a second aspect of an embodiment of the present invention, a four-quadrant rectifier fault diagnosis device based on energy entropy optimization and multi-feature fusion is provided.

[0059] In one embodiment, the apparatus comprises:

[0060] The data acquisition module is used to collect input current signals of fault modes under different working conditions to form an original sample set;

[0061] The signal processing module is used to perform wavelet packet multi-resolution decomposition on the input current signal of the fault mode using different wavelet functions to obtain different frequency band information. The wavelet function suitable for the fault signal is selected based on the energy-information entropy ratio of each frequency band. The corresponding frequency band is the optimal frequency band, and the energy entropy of each frequency band coefficient is calculated;

[0062] The feature extraction module is used to obtain the energy entropy characteristics of the optimal frequency band based on the energy entropy of each frequency band coefficient, and construct the energy entropy characteristics corresponding to different working conditions and different fault modes;

[0063] The feature optimization fusion module is used to perform local tangent space arrangement optimization on multiple redundant and conflicting energy entropy features in the data space to obtain the optimal energy entropy feature;

[0064] The pattern recognition module is used to perform pattern recognition using a support vector machine to obtain a fault diagnosis result of a four-quadrant rectifier.

[0065] According to a third aspect of embodiments of the present invention, a computer-readable storage medium is provided.

[0066] In some embodiments, the computer-readable storage medium includes a computer program for storing a computer program, wherein when the computer program is executed by a processor, the four-quadrant rectifier fault diagnosis method based on energy entropy optimization and multi-feature fusion is implemented.

[0067] According to a fourth aspect of embodiments of the present invention, a computer device is provided.

[0068] In some embodiments, the computer device includes a memory and a processor, the memory stores a computer program, and the processor implements the steps of the above method when executing the computer program.

[0069] The technical solution provided by the embodiment of the present invention may have the following beneficial effects:

[0070] This paper proposes a fault diagnosis method based on energy entropy optimization and multi-feature fusion for four-quadrant rectifiers, which have a high failure rate in heavy-duty train traction drive systems. First, the fault signal is decomposed into wavelet packets using different wavelet functions. The optimal frequency band is selected based on the energy-information entropy ratio of the frequency band coefficients. Second, the energy entropy of each coefficient in the optimal frequency band is calculated to obtain multiple energy entropy features. Finally, the multiple fault features are spatially arranged locally in tangent space, optimizing the data distribution of the features, reducing outliers, and eliminating redundancy between multiple features. Results show that the optimized fusion of multiple features achieves an average diagnostic accuracy of 99.0625% for four-quadrant rectifiers. Compared with other methods, this method exhibits higher diagnostic accuracy and stronger robustness for output voltage, noise, and training-to-test ratio.

[0071] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0073] Figure 1 is a schematic diagram of a single-phase PWM rectifier circuit structure according to an exemplary embodiment;

[0074] Figure 2 is an open circuit fault simulation model shown according to an exemplary embodiment;

[0075] Figure 3 is the current before and after the fault is triggered according to an exemplary embodiment I The waveform diagram of Figure 3 (a) is the waveform of VT1 open circuit fault at 1.5s. Figure 3 (b) At 1.5s, VT1 and VT2 simultaneously experience open circuit failures. Figure 3 (c) The waveform of VD1 open circuit fault at 1.5s;

[0076] Figure 4 is a flow chart of a multi-energy entropy feature optimization fusion algorithm according to an exemplary embodiment;

[0077] Figure 5 1 is a schematic structural diagram of a fault diagnosis device for energy entropy optimization and multi-feature fusion according to an exemplary embodiment;

[0078] Figure 6 is a schematic diagram showing energy entropy characteristics of 15 failure modes according to an exemplary embodiment;

[0079] Figure 7 is a schematic diagram showing the distribution of energy entropy characteristic data according to an exemplary embodiment;

[0080] Figure 8 is a schematic diagram showing an optimized energy entropy feature according to an exemplary embodiment;

[0081] Figure 9 is a schematic diagram showing the distribution of energy entropy characteristic data after optimization according to an exemplary embodiment;

[0082] Figure 10 is a schematic diagram showing an optimization effect of energy entropy characteristics according to an exemplary embodiment;

[0083] Figure 11 is a schematic diagram showing a comparison of robustness to noise according to an exemplary embodiment;

[0084] Figure 12 is a schematic diagram showing a robustness comparison of training and test ratios according to an exemplary embodiment;

[0085] Figure 13 is a schematic diagram showing a comparison of different methods according to an exemplary embodiment;

[0086] Figure 14 The figure is a schematic diagram showing the structure of a computer device according to an exemplary embodiment. DETAILED DESCRIPTION

[0087] The following description and accompanying drawings sufficiently illustrate the specific embodiments herein to enable those skilled in the art to practice them. Portions and features of some embodiments may be included in or substituted for portions and features of other embodiments. The scope of the embodiments herein includes the entire scope of the claims, including all available equivalents thereof. Herein, the terms "first," "second," and the like are used solely to distinguish one element from another and do not require or imply any actual relationship or order between these elements. In practice, the first element can also be referred to as the second element, and vice versa. Furthermore, the terms "comprise," "comprising," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a structure, device, or apparatus comprising a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such structure, device, or apparatus. Without further limitation, an element defined by the phrase "comprising a..." does not preclude the presence of other identical elements in the structure, device, or apparatus comprising the element. The various embodiments herein are described in a progressive manner, with each embodiment focusing on its differences from the other embodiments. Similar or identical parts between the various embodiments can be referenced to each other.

[0088] The terms "longitudinal", "transverse", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like used herein to indicate orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, are intended only to facilitate the description of this document and simplify the description, and are not intended to indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be construed as a limitation on the present invention. In the description herein, unless otherwise specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, they can be mechanical or electrical connections, or they can be internal connections between two elements, they can be directly connected, or they can be indirectly connected through an intermediate medium. For those of ordinary skill in the art, the specific meanings of the above terms can be understood according to the specific circumstances.

[0089] As used herein, unless otherwise specified, the term "plurality" means two or more.

[0090] In this document, the character " / " indicates that the preceding and following objects are in an "or" relationship. For example, A / B means: A or B.

[0091] In this article, the term "and / or" is used to describe the association relationship between objects, indicating that three relationships can exist. For example, A and / or B means: A or B, or, A and B.

[0092] In the absence of conflict, the embodiments of the present invention and the features thereof may be combined with each other.

[0093] Figure 1 An embodiment of the four-quadrant rectifier circuit structure of the present invention is shown.

[0094] The four-quadrant rectifier is an important device for power conversion of the heavy-load train HXD3 electric locomotive. The structure of the transmission system consists of a pantograph, a main circuit breaker, a main transformer, a four-quadrant rectifier, a traction inverter, a traction motor, a wheelset and a gearbox. When the intermediate DC side voltage is zero, the 1450V / 50Hz AC voltage drawn from the secondary winding of the traction transformer supplies power to the intermediate support capacitor through the charging resistor. When the intermediate DC voltage reaches 2000V, the charging contactor cuts off the charging resistor, the intermediate circuit pre-charging is completed, the gate blockade of the rectifier switching device is released, and the traction winding directly supplies power to the four-quadrant rectifier. When the intermediate DC voltage reaches 2800V, the inverter starts to work.

[16] .

[0095] The four-quadrant rectifier is also called a single-phase PWM rectifier. The circuit structure is as follows: Figure 1As shown, the main sources of failure are open circuit and short circuit failures of IGBT (represented by VT) and power diode (represented by VD). Among them, short circuit failure can be easily detected by the system monitoring and protection devices.

[17] Open circuit faults do not cause significant changes in the waveforms of current and voltage, and cannot be detected for a long time. They are often ignored and can easily lead to secondary failures of other components, causing greater losses to the system.

[18] Therefore, the fault diagnosis of four-quadrant rectifiers usually targets open-circuit faults of IGBTs and power diodes.

[0096] In order to analyze the impact of open-circuit faults of IGBT and power diode on the system, an open-circuit fault simulation model is established, such as Figure 2 As shown, an open-circuit fault is simulated by removing the IGBT gate signal and the power diode anode current. For example, at 1.5 seconds, the VT1 gate signal and VD1 anode current are removed, respectively, to simulate an open-circuit fault in both VT1 and VD1. Without adding additional sensors, the present invention utilizes the existing input current sensor signal for fault acquisition, specifying that the input current I flowing toward the rectifier is I>0, and vice versa.

[0097] At 1.5s, an open circuit fault is injected into VT1, and the current before and after the fault is triggered is I The waveform is as Figure 3 (a) When I > 0, due to the antiparallel effect of diode VD1, current flows through VD1, and the VT1 open-circuit fault has no effect on the rectifier's operation. Therefore, during the first half of the cycle after the fault occurs, the waveform of current I remains unchanged. When I < 0, current I cannot flow through diode VD1, and the waveform is distorted during the second half of the cycle after the fault occurs. Therefore, the symptoms of VT1's open-circuit fault only appear when I < 0, and the system cannot operate normally.

[0098] At 1.5s, an open circuit fault is injected into VT1 and VT3 at the same time. The current before and after the fault is triggered is I The waveform is as follows Figure 3 (b) When I > 0, VT3 provides a charging channel for the traction winding UN to charge the intermediate capacitor Cd. When VT3 experiences an open-circuit fault, the current I becomes distorted. When I < 0, the open-circuit faults of both VT1 and VT3 hinder the energy feedback process. Therefore, when VT1 and VT3 experience an open-circuit fault, the waveform of the current I is distorted, the amplitude is reduced, and fault symptoms appear in both the positive and negative half-cycles.

[0099] At 1.5s, an open circuit fault is injected into VD1, and the current before and after the fault is triggered is I The waveform is as Figure 3(c) When VD1 experiences an open-circuit fault, the rectifier branch fails to conduct, distorting the positive half-wave shape of the input current I. Therefore, an open-circuit diode failure affects the normal operation of the rectifier, making it necessary to accurately diagnose the diode failure.

[0100] During actual train operation, IGBTs or power diodes may experience open-circuit faults. Different combinations of switching devices result in different failure modes and symptoms. This application studies 15 open-circuit fault modes (normal mode and 14 failure modes) of four-quadrant rectifiers. The corresponding fault modes and their fault labels are shown in Table 1.

[0101]

[0102] An embodiment of the present invention is a four-quadrant rectifier fault diagnosis method based on energy entropy optimization and multi-feature fusion.

[0103] Simple and sensitive fault features help reduce the computational burden of the classifier and improve fault diagnosis results. Wavelet packets are used to perform multi-resolution decomposition of noisy, nonlinear, and non-stationary fault signals, avoiding the problem of missing fault information in high-frequency signals. The energy-information entropy criterion is proposed to select the wavelet function suitable for fault signal decomposition, thereby obtaining the optimal multi-band information and solving the problem of different fault signals being suitable for different wavelet functions.

[19] Since the change of fault mode will lead to the change of signal energy value, the entropy theory is used to evaluate the probability of energy change. The smaller the energy entropy, the greater the certainty of the energy value change, that is, the greater the possibility of fault mode change. Therefore, the energy entropy feature is used to simultaneously characterize different fault modes in terms of numerical and probability distribution. Multiple energy entropy features are arranged in local tangent space in the data space.

[20] Optimization reduces outliers in multi-feature data. The data is spatially dispersed, and the data distribution is approximately normal. The spatial optimization of multi-feature data enables redundant and conflicting multi-features to be optimized and fused to obtain simple and sensitive fault features.

[0104] Figure 1 An embodiment of the multi-energy entropy feature optimization fusion algorithm flow chart of the present invention is shown.

[0105] S1: Using different wavelet functions, perform wavelet packet multi-resolution decomposition on the input current signal (fault signal) of the fault mode to obtain different frequency band information. Based on the energy-information entropy ratio of each frequency band, a wavelet function suitable for the fault signal is selected. The corresponding frequency band is the optimal frequency band, and the energy entropy of each frequency band coefficient is calculated.

[0106] Specifically, the following steps are included:

[0107] (1) Calculation of multi-band coefficients of different wavelet functions. Assume that the fault signal ,Using 9 Daubechies (dbN, N= 2,3,…10) wavelet functions commonly used in signal decomposition and reconstruction, the fault signal is decomposed by wavelet packet to obtain 9 groups of frequency band information.

[0108] (1)

[0109] Where, Indicates the use of db N Wavelet function, signal passes through wavelet packet r After layer decomposition, we get m In the frequency band n The coefficients of the frequency bands.

[0110] (2) Select an appropriate wavelet function. Different wavelet functions have different decomposition effects on the same fault signal, which directly affects the effect of feature extraction. It is crucial to select a wavelet function that is suitable for the fault signal. The energy-information entropy ratio is the ratio of the energy of the frequency band coefficient to the information entropy. It also takes into account the energy value changes of different fault modes and the size of the probability of occurrence, that is, the size of the uncertainty.

[0111] The frequency band energy value corresponding to the wavelet function after wavelet packet decomposition:

[0112] (2)

[0113] After a system failure occurs, the uncertainty of each failure mode is related to the number of times the corresponding failure mode occurs. If a failure mode of the rectifier is Y i , the probability of this failure mode occurring is P ( Y i ), then the information entropy of the frequency band coefficient obtained after the signal of the fault mode is decomposed is :

[0114] (3)

[0115] The energy-information entropy ratio of the corresponding frequency band is R :

[0116] (4)

[0117] The larger the energy-information entropy ratio is, the more suitable the selected wavelet function is for the fault signal, and the frequency band obtained after decomposition is the optimal frequency band.

[0118] (3) Calculate the energy entropy characteristics of the optimal frequency band. The fault signal is decomposed by wavelet packet r After layer decomposition, mfrequency band coefficients, and calculate the energy entropy of each frequency band coefficient in the frequency band. n The energy value of the frequency band coefficient is E ( n ), the probability distribution is p (n), energy entropy is T (n):

[0119] (5)

[0120] (6)

[0121] (7)

[0122] Therefore, based on the optimal wavelet function, the energy entropy characteristics of the optimal frequency band obtained by decomposing the fault signal through wavelet packets are: T ( n ):

[0123] (8)

[0124] (4) Construct the domain matrix of energy entropy characteristics. T for m dimensional matrix, T n is any data point in the matrix, m is the number of data points, and the number of nearest neighbors for each data point is determined as k , forming a neighborhood matrix.

[0125] (9)

[0126] Where, For the n The data point j Nearest neighbors, n =1,2,…, N ; j =1,2,…, k

[0127] (5) Calculate local coordinates. At the data point T n Find the orthogonal basis in the field matrix Q i to constitute g dimensional tangent space, in the neighborhood, calculate each point T nj Orthogonal projection to tangent space λ nj , then find the neighborhood matrix T mn The local coordinates of n:

[0128] (10)

[0129] (6) Establish global coordinates. Set global coordinates , for energy entropy characteristics T ( n ) is arranged in tangent space to obtain the global coordinate system. The global coordinates reflect the geometric structure of the local coordinate matrix and should meet the following requirements:

[0130] (11)

[0131] Where, is the mapping matrix to be determined; is the reconstruction error; yes The centralization matrix.

[0132] (7) Calculate the permutation matrix. Assume , S is a 0-1 selection matrix, is the energy entropy characteristic data point T n The weight matrix of , the permutation matrix is :

[0133] (12)

[0134] (8) Minimize the reconstruction error. This is to solve the minimum eigenvalue of the permutation matrix. The eigenvectors corresponding to the second to (g+1) minimum eigenvalues ​​obtained by calculating the permutation matrix are the energy entropy characteristics. T ( n ) is the global coordinate of each data point in the , which is also the feature vector after the energy entropy feature is optimized by local tangent space arrangement. .

[0135] (13)

[0136] S2: Perform local tangent space arrangement optimization on multiple redundant and conflicting energy entropy features in the data space to obtain efficient fault features;

[0137] To verify the robustness of the proposed method to output voltage, a system simulation model was established. The main parameters of the four-quadrant rectifier are shown in Table 2. The output voltage was set to vary from 2000V to 2800V. The rectifier input current signals were collected at 10V intervals under 15 different fault modes as fault acquisition signals. To simulate a noisy environment, white noise with a signal-to-noise ratio of 50dB was added to the fault acquisition signals. Each fault mode had 81 samples, and the 15 fault modes had 81 × 15 = 1215 samples.

[0138]

[0139] Specifically, the following steps are included:

[0140] (1) Selection of optimal wavelet basis function

[0141] The present invention considers the degree of mining fault features and the computational cost, performs 5-layer wavelet packet decomposition on the fault signal, and selects Daubechies (db N ) wavelet system, respectively using db N Nine wavelet basis functions (2-10) were used to decompose the signal using wavelet packets, resulting in nine frequency bands for each sample and each fault mode. The energy-information entropy ratio of each frequency band was calculated and compared. The maximum ratio indicated that the selected wavelet function was optimal and that the frequency bands obtained from the decomposition of the corresponding fault signal were optimal. Table 3 shows the results of selecting the optimal wavelet function for different fault modes when the output DC voltage was 2800 V.

[0142]

[0143] (2) Extraction of energy entropy characteristics and data analysis

[0144] 81 operating conditions and 15 failure modes for each condition were analyzed using their respective optimal wavelet functions. A five-layer wavelet packet decomposition of the fault signals was performed, resulting in 32 frequency bands. The energy entropy of each frequency band coefficient was calculated, resulting in 32 energy entropy features for each failure mode. Line graphs and box plots were used to illustrate the distribution of the energy entropy features for each failure mode. Line graphs depict the patterns and trends of energy entropy changes across all frequency bands. Box plots are statistical graphs used to reflect the distribution characteristics of raw data. They provide information on the concentration trend and dispersion of the data, allowing for the determination of data skewness and severity, thereby revealing whether the data conform to a normal distribution.

[0145] When the rectifier output voltage is 2800V, the line graph of the energy entropy characteristics of 15 fault modes is as follows: Figure 6As shown in the figure, the energy entropy characteristics of different fault modes in the 1st to 5th frequency bands and the 13th frequency band are greater than those in other frequency bands. However, the energy entropy of many fault modes overlaps with each other in the corresponding frequency bands, making it impossible to accurately identify the fault type.

[0146] The energy entropy characteristic data distribution of the rectifier is analyzed by box plot, such as Figure 7 As shown in the figure, the energy entropy feature data contains many outliers, all of which are on the larger side, indicating that the energy entropy feature data distribution is right-skewed and does not follow a normal distribution. The characteristic data box plots for each fault mode are very short, indicating that many data points are distributed in a small range, each data point is similar, and the data is relatively concentrated, which is consistent with the analysis of the line chart.

[0147] (3) Optimization of energy entropy characteristics and effect analysis

[0148] During the feature extraction process, the feature data is as normally distributed as possible, so that statistical inference and hypothesis verification can be better performed. Although the energy entropy feature obeys a skewed distribution, the dispersion and deviation of the data points are changed by fusion technology, and the difference between the data points is increased, so that the fused data can obey a normal or approximately normal distribution. Therefore, the multiple energy entropy feature data extracted are arranged in a local tangent space. In order to avoid different selections of neighboring points of different dimensions, the computational cost and data portability are taken into consideration. It is known that the number of rectifier fault types is C=15. According to the literature

[21] , its inherent dimension is defined as D=C-1, that is, the intrinsic dimension of the feature data fusion is 14 dimensions. After multiple experiments, it is verified that the neighboring points of the selected data are 400. The line graph of the optimized energy entropy feature is as follows. Figure 8 As shown in Figure 3, the optimized energy entropy characteristics are numerically dispersed.

[0149] The data distribution of the optimized energy entropy characteristics is analyzed by box plot, such as Figure 9 As shown in the figure, compared to the feature data before optimization, outliers are significantly reduced, and the feature data follows an approximately normal distribution. After optimization, the energy entropy features, with the exception of the normal, VT2VT4, and VD1 models, each of which has one or two outliers, have significantly reduced these outliers. Furthermore, the optimization process eliminates outliers in other fault modes, optimizing the feature data distribution for each fault mode. This results in longer box plots, increased standard deviation and variance, reduced data skewness, and greater differences between data points, shifting from a concentrated state to a dispersed state. Regarding feature data symmetry, with the exception of the normal, VT1VT3, VT2VT4, VD1, and VD2 models, where the upper and lower whisker lengths differ significantly, the upper and lower whisker lengths of the remaining fault modes are approximately equal, indicating a symmetrical feature data distribution.

[0150] The optimization process improves the data distribution of energy entropy features. The improvement of data distribution is verified by the scatter plot of three principal component features, i.e. three-dimensional features, in each fault mode, as shown in the following example: Figure 10 As shown in Figure 3, the optimized energy entropy signatures for the 15 traction rectifier fault modes have indeed been improved, resulting in satisfactory clustering. However, the energy entropy signatures for each fault mode are very similar in value, resulting in overlap between samples within each fault mode. Due to the large number of fault modes and the close proximity of the fault samples, classification by observation is difficult. Therefore, pattern recognition technology is required to input the optimized energy entropy signatures into a classifier for accurate fault diagnosis.

[0151] S3: Using Support Vector Machine (SVM) technology

[15] Perform pattern recognition to realize fault diagnosis of four-quadrant rectifier.

[0152] This application adopts support vector machine as the classifier of fault features, in which the kernel function selects the radial basis function (RBF) that is robust to noise. A certain amount of Gaussian white noise is added to the collected original signal to maximize the simulation of the real working environment of the rectifier. Through signal decomposition and feature extraction, the optimized energy entropy feature is used as the input of the classifier SVM, and the test results are compared with the actual results of the trained SVM classifier to give a diagnostic result.

[0153] Fault diagnosis result analysis and comparison include the following steps:

[0154] (1) Robustness analysis of output voltage

[0155] To verify the robustness of the proposed method to output voltage variations, the output voltage was set from 2000V to 2800V. Input current signals of different modes were collected at 10V intervals. 50dB Gaussian white noise was added to the collected currents at different output voltages. Each mode had 81 samples, totaling 1215 samples for the 15 fault modes. Using the SVM classifier, 729 samples were randomly selected as training samples. After training, the remaining samples were used as test samples and compared with the actual output results to obtain the final diagnosis rate. The classifier was trained and tested 30 times, with the average diagnostic accuracy as the evaluation indicator. After 30 runs, the average diagnostic accuracy was 99.0625%. The average diagnostic results for each fault mode are shown in Table 4. Among the 486 random test samples, for the open-circuit fault of IGBT, except for VT2 (99.7917%), the average diagnostic accuracy of the rest was 100%, indicating that the proposed method has good fault diagnosis effects for both single open-circuit fault and double open-circuit fault of IGBT. Although the fault diagnosis rate for diode is lower than that of IGBT, the fault diagnosis result is still high, indicating that the proposed method has high diagnostic results for open-circuit faults of IGBT and power diode.

[0156]

[0157] (2) Robustness analysis against noise

[0158] In order to verify the diagnostic rate of the selected diagnostic method in different noise environments, white noise with signal-to-noise ratios of 30dB, 35dB, 40dB, 45dB, and 50dB was added to the original signal. According to the proposed method, feature extraction and optimization were performed to obtain simple and sensitive features. Then, the classifier SVM was used to run each set of features 30 times. The average, maximum, and minimum values ​​of the diagnostic accuracy of the 15 fault modes are shown in Figure 2. Figure 11 shown.

[0159] When the signal-to-noise ratio is preset to 30dB and the ratio is 3:2, the average value after 30 diagnoses is 94.5903%, the maximum value is 96.6667%, and the minimum value is 93.1250%. As the signal-to-noise ratio increases, the diagnosis results are better than 30dB, indicating that the proposed method is robust to noise.

[0160] (3) Robustness analysis of training and testing ratio

[0161] In order to verify the influence of the ratio of selected training samples and test samples on the diagnosis rate, the training and test ratios were set to 2:1, 3:2, 5:1, 10:1, 15:1, and 20:1. The average diagnosis rate was used as the evaluation index after running the data containing 30dB, 35dB, 40dB, 45dB, and 50dB white noise 30 times. The diagnosis results are shown in the figure below. Figure 12 As shown in the figure, when the training / test ratio is 3:2, the average diagnosis results for all data containing white noise are the lowest, but the corresponding diagnosis rate is relatively high. The diagnosis rates for data containing different white noise at a 3:2 ratio are 94.5903%, 97.5833%, 98.7917%, 98.8889%, and 99.0625%, respectively. Therefore, this method is robust to the training / test ratio.

[0162] (4) Comparison of different methods

[0163] In order to illustrate the superiority of the proposed method, data with signal-to-noise ratios of 30dB, 35dB, 40dB, 45dB, and 50dB, and a training-to-test ratio of 3:2 were used. The average value of the diagnostic results after 30 runs of the proposed method was used as the evaluation index for each set of data using energy features, energy entropy features, and optimized features. The results are shown in the figure below. Figure 13 shown.

[0164] When energy and energy entropy are used as features, their fault diagnosis accuracy is low due to the similarity of their feature data. However, energy entropy, because it considers the spectral distribution of energy, makes the diagnosis of faults based on energy entropy higher than that based on energy. After local tangent space permutation, energy and energy entropy features are arranged, redundancy and conflict between features are eliminated, and the distribution of feature data changes from an approximate and overlapping state to a dispersed state, more closely approaching a normal distribution. The dimension of the feature data decreases from 32 to 14, resulting in simple and sensitive features that reduce the computational burden of the classifier. Both fault features significantly improve the fault diagnosis rate after local tangent space permutation optimization. The improvement in the diagnosis rate of faults based on energy entropy is greater than that of faults based on energy. As the signal-to-noise ratio increases, the average diagnostic accuracy reaches 94.59%, 97.58%, 98.79%, 98.89%, and 99.06%, respectively. Compared with the pre-optimization results, the diagnostic rates increase by 39.27%, 41.95%, 44.32%, 43.97%, and 43.69%, respectively.

[0165] In terms of algorithmic complexity, evaluation metrics were feature fusion time, training time, diagnosis time, computational burden, and the average diagnostic accuracy over 30 diagnoses, as shown in Table 5. Regarding computational burden, several fusion algorithms all involve the target dimension. Although PCA does not require additional parameters, it performs poorly on nonlinear data. KPCA and the ISOMAP algorithm are suitable for dimensionality reduction of nonlinear data, but KPCA relies heavily on the kernel function, which affects diagnostic accuracy. Furthermore, since the data is categorical and rarely embedded in a single mainstream, the dimensionality reduction effect is relatively poor. Compared to the pre-optimized energy entropy feature, all three algorithms reduced training and diagnosis time. Although training times were similar, the local tangent space permutation algorithm had the shortest diagnosis time. In terms of fusion time, the local tangent space permutation algorithm, while taking a longer time, was able to accurately fuse fault features, resulting in the highest average fault diagnosis accuracy.

[0166]

[0167] For the fault diagnosis of four-quadrant rectifier, a fault diagnosis device with energy entropy optimization and multi-feature fusion is proposed, which includes: data acquisition module 11, signal processing module 12, feature extraction module 13, feature optimization and fusion module 14 and pattern recognition module 15. The diagnostic block diagram is shown in the figure. Figure 5 shown.

[0168] (1) Data acquisition module 11: Collect input current signals of 15 fault modes under different working conditions to form an original sample set.

[0169] (2) Signal processing module 12 includes wavelet packet decomposition and optimal frequency band selection. Wavelet packet decomposition selects the Daubechies series wavelet suitable for signal decomposition and reconstruction. Based on nine Daubechies (dbN, N = 2, 3, ...) 10 wavelet functions, the wavelet packet decomposes the signals of different working conditions and different fault modes, obtaining nine sets of frequency band information for each fault mode. Optimal frequency band selection selects the optimal wavelet function based on the energy-information entropy ratio of each frequency band coefficient. The corresponding frequency band information is the optimal frequency band suitable for the fault mode signal.

[0170] (3) Feature extraction module 13. Calculate the energy entropy characteristics of the optimal frequency band and construct energy entropy characteristics corresponding to different working conditions and different fault modes.

[0171] (4) Feature optimization fusion module 14. The energy entropy features are spatially arranged locally and then fused to obtain the optimal energy entropy features.

[0172] (5) Pattern recognition module 15. The fault feature vector is randomly divided into training samples and test samples, and the SVM classifier is used for fault diagnosis. The training samples are used to train the SVM, and the test samples are used to verify the performance of the trained SVM classifier, thereby obtaining the diagnosis results.

[0173] References are as follows

[0174] [1] Pan Pengyu, Hu Haitao, Xiao Donghua, et al. Frequency coupling interference mechanism and suppression strategy in “sweep frequency” dq impedance measurement of high-speed train converter [J]. Transactions of China Electrotechnical Society, 2022, 37(4): 990-999.

[0175] Pan Pengyu, Hu Haitao, Xiao Donghua, et al. Frequency couplinginterference mechanism and suppression strategy for frequency-sweeping-baseddq impedance measurement of high-Speed ​​train converter[J]. Transactions ofChina Electrotechnical Society, 2022, 37(4): 990-999.

[0176] [2] Tian Zisi, Ge Xinglai. An on-line fault diagnostic method based on frequency-domain analysis for IGBTs in traction PWM rectifiers[C] / / 2016IEEE 8th International Power Electronics and Motion Control Conference(IPEMC-ECCE Asia), Hefei, China, 2016: 3403-3407.

[0177] [3] Xu Shuiqing, Huang Wenzhan, He Yigang, et al. Open circuit fault diagnosis of neutral point clamped three-level grid-connected inverter based on adaptive sliding mode observer [J]. Transactions of the Chinese Society of Electrotechnical Engineering, 2023, 38(4): 1010-1022.

[0178] Xu Shuiqing, Huang Wenzhan, He Yigang, et al. Open-circuit faultdiagnosis method of neutral point clamped three-level grid-connected inverterbased on adaptive sliding mode observer[J]. Transactions of ChinaElectrotechnical Society, 2023, 38(4): 1010-1022.

[0179] [4] Xie Dong, Ge Xinglai. A state estimator-based approach for open-circuit fault diagnosis in single-phase cascaded H-bridge rectifiers[J]. IEEETransactions on Industry Applications, 2018, 55(2): 1608-1618.

[0180] [5] Hu, Keting, Liu Zhigang, Francesco Iannuzzo, et al. Simple and effective open switch fault diagnosis of single-phase PWM rectifier[J]. Microelectronics Reliability, 2018, 88: 423-427.

[0181] [6] Wu Hong, Wang Yue, Xue Yinglin, et al. Diagnosis strategy of open circuit fault of MMC submodule under multi-power nearest level modulation[J]. Transactions of China Electrotechnical Society, 2024, 39(1): 233-245.

[0182] Wu Hong, Wang Yue, Xue Yinglin, et al. A diagnosis strategy for open-circuit submodule faults in MMCs under nearst level modulation suitable fordifferent powers[J]. Transactions of China Electrotechnical Society, 2024, 39(1): 233-245.

[0183] [7] Li Xueming, Xu Jiamin, Chen Zhiwen, et al. Real-time faultdiagnosis of pulse rectifier in traction system based on structural model[J].IEEE Transactions on Intelligent Transportation Systems, 2020, 23(3): 2130-2143.

[0184] [8] M. Arehpanahi, A. M. Entekhabi. A new technique for online openswitch fault detection and location in single-phase pulse width modulationrectifier[J]. International Journal of Engineering, 2022, 35(9): 1759-1764.

[0185] [9] Wang Yanxin, Yan Jing, Wang Jianhua, et al. Small sample GIS insulating defect intelligent diagnosis method based on domain adversarial transfer convolutional neural network[J]. Transactions of China Electrotechnical Society, 2022, 37(9): 2150-2160.

[0186] ​​​​Wang Yanxin, Yan Jing, Wang Jianhua, et al. Intelligent diagnosis for GIS with small samples using a novel adversarial transfer learning inconvolutional neural network[J]. Transactions of China ElectrotechnicalSociety, 2022, 37(9): 2150-2160.

[0187]

[10] Cui Jiang, Guo Ruidong, Zhang Zhuoran, et al. Fault feature extraction technology of generator rotating rectifier based on improved DBN[J]. Proceedings of the CSEE, 2020, 40(7): 2369-2376.

[0188] Cui Jiang, Guo Ruidong, Zhang Zhuoran, et al. Generator rotatingrectifier fault feature extraction technique based on improved DBN[J]. Proceedings of the CSEE, 2020, 40(7): 2369-2376.

[0189]

[11] Liang Zhengqiu, Hao Liangliang, Zhou Yanzhen, et al. Fault diagnosis of rotating rectifier in nuclear power multiphase brushless excitation system based on convolutional neural network[J]. Transactions of China Electrotechnical Society, 2023, 38(20): 5458-5472.

[0190] Liang Zhengqiu, Hao Liangliang, Zhou Yanzhen, et al. Fault diagnosis of rotating rectifier in nuclear multi-phase brushless excitation system based on convolutional neural network[J]. Transactions of China Electrotechnical Society, 2023, 38(20): 5458-5472.

[0191]

[12] Zhang Guoheng, Gao Fengyang, Shi Yan, et al. Multi-feature fusion diagnosis method for open circuit fault of traction inverter based on Bayesian network[J]. Journal of Railway Science and Engineering, 2020, 17(3): 732-740.

[0192] Zhang Guoheng, Gao Fengyang, Shi Yan, et al. Multi-feature fusiondiagnosis method of open circuit fault for traction inverter based onBayesian network[J]. Journal of Railway Science and Engineering, 2020, 17(3):732-740.

[0193]

[13] Li Bing, Cui Jiebing, He Yigang, et al. Fault diagnosis of active neutral point clamped three-level inverter based on energy spectrum entropy and wavelet neural network[J]. Transactions of the Chinese Society of Electrotechnical Engineering, 2020, 35(10): 2216-2225.

[0194] Li Bing, Cui Jiebing, He Yigang, et al. Fault diagnosis of activeneutral point clamped three-level inverter based on energy spectrum entropyand wavelet neural network[J]. Transactions of China ElectrotechnicalSociety, 2020, 35(10): 2216-2225.

[0195]

[14] Zhang Ruicheng, Bai Xiaoze, Dong Yan, et al. Open circuit fault diagnosis of wind power converter based on LMD energy entropy and positioning analysis[J]. Acta Energiae Solaris Sinica, 2023, 44(06): 484-494.

[0196] Zhang Ruicheng, Bai Xiaoze, Dong Yan, et al. Open-circuit faultdiagnosis of wind power converter based on LMD energy entropy and location analysis[J]. Acta Energiae Solaris Sinica, 2023, 44(6): 484-494.

[0197]

[15] Wu Xin, Yan Meng, Guo Yifan, et al. Non-intrusive load identification based on combined support vector machine with structured feature graph[J]. Automation of Electric Power Systems, 2022, 46(12): 210-219.

[0198] Wu Xin, Yan Meng, Guo Yifan, et al. Non-intrusive load identification by combined support vector machine based on structured characteristic spectrum[J]. Automation of Electric Power Systems, 2022, 46(12):210-219.

[0199]

[16] Zhang Shuguang. HXD3 electric locomotive[M]. Beijing: China Railway Publishing House, 2009.

[0200] Zhang Shuguang. HXD3 electric locomotive[M]. Beijing: China RailwayPublishing House, 2009.

[0201]

[17] Wen Yang, Yang Yuan, Ning Hongying, et al. Review of SiC MOSFET short-circuit protection technology[J]. Transactions of the Chinese Society of Electrotechnical Engineering, 2022, 37(10): 2538-2548.

[0202] Wen Yang, Yang Yuan, Ning Hongying, et al. Review on short-circuitprotection technology of SiC MOSFET[J]. Transactions of ChinaElectrotechnical Society, 2022, 37(10): 2538-2548.

[0203]

[18] Yang Heya, Xing Wenshuo, Xiang Xin, et al. MMC submodule open circuit fault diagnosis method based on multivariate Gaussian distribution anomaly detection model [J]. Transactions of the Chinese Society of Electrotechnical Engineering, 2023, 38(10): 2744-2756.

[0204] Yang Heya, Xing Wenshuo, Xiang Xin, et al. A sub-module open-circuitfault detection and location strategy for modular multilevel converters based on multivariate gaussian distribution[J]. Transactions of ChinaElectrotechnical Society, 2023, 38(10): 2744-2756.

[0205]

[19] He Hong, Tan Yonghong, Wang Yuexia. Optimal base wavelets selection for ECG noise reduction using a comprehensive entropy criterion[J].Entropy, 2015, 17(9): 6093-6109.

[0206]

[20] Lu Na, Zhang Guangtao, Xiao Zhihuai, et al. Feature extraction based on adaptive multiwavelets and LTSA for rotating machinery faultdiagnosis[J]. Shock and Vibration, 2019(1): 1201084.

[0207]

[21] He, Qingbo. Time–frequency manifold for nonlinear feature extraction in machinery fault diagnosis[J]Mechanical Systems and SignalProcessing, 2013, 35(1-2): 200-218.

[0208] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Figure 14 As shown. The computer device includes a processor, a memory, and a network interface connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The database of the computer device is used to store static information and dynamic information data. The network interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, the steps of the above-mentioned method embodiment are implemented.

[0209] Those skilled in the art will understand that Figure 14 The structure shown in the figure is merely a block diagram of a portion of the structure related to the solution of the present invention and does not constitute a limitation on the computer device to which the solution of the present invention is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0210] In one embodiment, a computer device is further provided, including a memory and a processor. The memory stores a computer program, and the processor implements the steps in the above method embodiment when executing the computer program.

[0211] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above method embodiment are implemented.

[0212] Those skilled in the art will appreciate that all or part of the processes in the above-described method embodiments can be implemented by instructing the relevant hardware using a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes in the above-described method embodiments. Any reference to memory, storage, database, or other media used in the various embodiments provided herein may include at least one of non-volatile and volatile memory. Non-volatile memory may include read-only memory (ROM), magnetic tape, floppy disk, flash memory, or optical storage. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM can take various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0213] In summary, a fault diagnosis method based on energy entropy optimization and multi-feature fusion was proposed for open-circuit faults in four-quadrant rectifiers of heavy-duty trains. Results show that the proposed method spatially approximates a normal distribution for multi-feature data, reduces outliers in the data, eliminates redundancy and conflict between multiple features, and achieves good clustering. Fifteen fault modes have high diagnostic rates at various signal-to-noise ratios and exhibit good robustness to DC voltage, noise, and the training-to-test ratio. Compared with other methods, the proposed method achieves superior diagnostic performance.

[0214] The present invention is not limited to the structures described above and shown in the drawings, and various modifications and changes can be made without departing from the scope thereof. The scope of the present invention is limited only by the appended claims.

Claims

1. A four-quadrant rectifier fault diagnosis method based on energy entropy optimization and multi-feature fusion, characterized in that: The following steps are involved: Collect input current signals of fault modes under different working conditions to form an original sample set; Using different wavelet functions, the input current signal of the fault mode is decomposed by wavelet packet multi-resolution to obtain different frequency band information. The wavelet function suitable for the fault signal is selected according to the energy-information entropy ratio of each frequency band. The corresponding frequency band is the optimal frequency band, and the energy entropy of each frequency band coefficient is calculated. According to the energy entropy of each frequency band coefficient, the energy entropy characteristics of the optimal frequency band are obtained, and the energy entropy characteristics corresponding to different working conditions and different fault modes are constructed; Perform local tangent space arrangement optimization on multiple redundant and conflicting energy entropy features in the data space to obtain the optimal energy entropy feature; Support vector machine is used for pattern recognition to obtain fault diagnosis results of four-quadrant rectifier; The method uses different wavelet functions to perform wavelet packet multi-resolution decomposition on the input current signal of the fault mode to obtain different frequency band information. The wavelet function suitable for the fault signal is selected based on the energy-information entropy ratio of each frequency band. The corresponding frequency band is the optimal frequency band. The energy entropy of each frequency band coefficient is calculated, which specifically includes the following steps: (1) Calculation of multi-band coefficients of different wavelet functions: Assuming the fault signal ,Using 9 Daubechies wavelet functions commonly used in signal decomposition and reconstruction, the fault signal is decomposed by wavelet packet to obtain 9 groups of frequency band information; （1） Where, Indicates the use of db N Wavelet function, dbN, N= 2,3,…10, the signal passes through the wavelet packet r After layer decomposition, we get m In the frequency band n The coefficients of the frequency bands; (2) Select a suitable wavelet function: The frequency band energy value corresponding to the wavelet function after wavelet packet decomposition: （2） After a system failure occurs, the uncertainty of each failure mode is related to the number of times the corresponding failure mode occurs. If a failure mode of the rectifier is Y i , the probability of this failure mode occurring is P ( Y i ), then the information entropy of the frequency band coefficient obtained after the signal of the fault mode is decomposed is : （3） The energy-information entropy ratio of the corresponding frequency band is R : （4） The larger the energy-information entropy ratio is, the more suitable the selected wavelet function is for the fault signal, and the frequency band obtained after decomposition is the optimal frequency band; (3) Calculate the energy entropy characteristics of the optimal frequency band: the fault signal is decomposed by wavelet packet r After layer decomposition, m frequency band coefficients, and calculate the energy entropy of each frequency band coefficient in the frequency band respectively; （5） （6） （7） In the formula, n The energy value of the frequency band coefficient is E ( n ), the probability distribution is p (n), energy entropy is T (n); Therefore, based on the optimal wavelet function, the energy entropy characteristics of the optimal frequency band obtained by decomposing the fault signal through wavelet packets are: T ( n ): （8） (4) Constructing the domain matrix of energy entropy characteristics: Due to the energy entropy characteristics T for m dimensional matrix, T n is any data point in the matrix, m is the number of data points, and the number of nearest neighbors for each data point is determined as k , forming a neighborhood matrix; （9） Where, For the n The data point j Nearest neighbor points, n =1,2,…, N ; j =1,2,…, k ; (5) Calculate local coordinates: At the data point T n Find the orthogonal basis in the field matrix Q i to constitute g dimensional tangent space, in the neighborhood, calculate each point T nj Orthogonal projection to tangent space λ nj , then find the neighborhood matrix T mn The local coordinates of n : （10） (6) Establish global coordinates: Set global coordinates , for energy entropy characteristics T ( n ) is arranged in tangent space to obtain the global coordinate system. The global coordinates reflect the geometric structure of the local coordinate matrix and should meet the following requirements: （11） Where, is the mapping matrix to be determined; is the reconstruction error; yes The centralization matrix; (7) Calculate the permutation matrix: Assume , S is a 0-1 selection matrix, is the energy entropy characteristic data point T n The weight matrix of , the permutation matrix is : （12） (8) Minimize the reconstruction error: Solve the minimum eigenvalue of the permutation matrix. The eigenvector corresponding to the second to (g+1) minimum eigenvalues ​​obtained by calculating the permutation matrix is ​​the energy entropy feature. T ( n ) is the global coordinate of each data point in the tangent space, which is also the feature vector of the energy entropy feature after optimization of the local tangent space arrangement. : （13）。 2. The four-quadrant rectifier fault diagnosis method based on energy entropy optimization and multi-feature fusion according to claim 1 is characterized in that: The step of performing local tangent space arrangement optimization on multiple redundant and conflicting energy entropy features in the data space to obtain the optimal energy entropy feature specifically includes: (1) Selection of optimal wavelet basis function: Considering the degree of fault feature mining and computational cost, the fault signal is decomposed by wavelet packet decomposition. The Daubechies wavelet system, which is commonly used for signal decomposition and reconstruction, is selected. The wavelet basis functions of the Daubechies wavelet system are used to decompose the signal. Multiple frequency bands are obtained for each sample and each fault mode. The energy-information entropy ratio of each frequency band is calculated and compared. The largest ratio indicates that the selected wavelet function is optimal and the frequency band obtained from the decomposition of the corresponding fault signal is optimal. (2) Extraction of energy entropy features and data analysis: For each operating condition, the multiple fault modes are decomposed into wavelet packets using their respective optimal wavelet functions to obtain multiple frequency bands, and the energy entropy characteristics of each frequency band coefficient are calculated. (3) Optimization of energy entropy characteristics: During the feature extraction process, the feature data presents a normal distribution, and the extracted multiple energy entropy feature data are arranged in a local tangent space.

3. The four-quadrant rectifier fault diagnosis method based on energy entropy optimization and multi-feature fusion according to claim 1 or 2 is characterized in that: The step of using a support vector machine to perform pattern recognition to obtain a fault diagnosis result of a four-quadrant rectifier specifically includes: Support vector machine is used as the classifier of fault features, in which the kernel function selects the radial basis function which is robust to noise. Gaussian white noise is added to the collected original signal to simulate the real working environment of the four-quadrant rectifier. Through signal decomposition and feature extraction, the optimized energy entropy feature is used as the input of the classifier SVM. The test results are compared with the actual results of the trained SVM classifier to give the fault diagnosis results.

4. The four-quadrant rectifier fault diagnosis method based on energy entropy optimization and multi-feature fusion according to claim 1 is characterized in that: The step of using a support vector machine as a classifier of fault features specifically includes: The fault feature vectors are randomly divided into training samples and test samples, and the support vector machine classifier is used for fault diagnosis. The training samples are used to train the support vector machine, and the test samples are used to verify the performance of the trained support vector machine classifier, thereby obtaining the diagnosis results.

5. A four-quadrant rectifier fault diagnosis device based on energy entropy optimization and multi-feature fusion, characterized in that: The four-quadrant rectifier fault diagnosis method based on energy entropy optimization and multi-feature fusion according to any one of claims 1 to 4 comprises: The data acquisition module is used to collect input current signals of fault modes under different working conditions to form an original sample set; The signal processing module is used to perform wavelet packet multi-resolution decomposition on the input current signal of the fault mode using different wavelet functions to obtain different frequency band information. The wavelet function suitable for the fault signal is selected based on the energy-information entropy ratio of each frequency band. The corresponding frequency band is the optimal frequency band, and the energy entropy of each frequency band coefficient is calculated; The feature extraction module is used to obtain the energy entropy characteristics of the optimal frequency band based on the energy entropy of each frequency band coefficient, and construct the energy entropy characteristics corresponding to different working conditions and different fault modes; The feature optimization fusion module is used to perform local tangent space arrangement optimization on multiple redundant and conflicting energy entropy features in the data space to obtain the optimal energy entropy feature; The pattern recognition module is used to perform pattern recognition using a support vector machine to obtain a fault diagnosis result of a four-quadrant rectifier.

6. A computer-readable storage medium, characterized in that Used to store a computer program, wherein when the computer program is executed by a processor, the four-quadrant rectifier fault diagnosis method based on energy entropy optimization and multi-feature fusion as described in any one of claims 1 to 4 is implemented.

7. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 4 are implemented.

Images