Traction rectifier fault diagnosis method based on optimal multiband fuzzy entropy fusion

Abstract

The invention belongs to the field of electric locomotive fault diagnosis, and discloses a traction rectifier fault diagnosis method based on optimal multiband fuzzy entropy fusion, and the method comprises the steps: firstly, selecting an optimal wavelet function, carrying out the wavelet packet decomposition of rectifier signals under different working conditions, and obtaining a series of optimal multibands; secondly, calculating the fuzzy entropy of each optimal frequency band as a high-dimensional fault feature; and finally, fusing the high-dimensional features by using t-distributed random neighborhood embedding to obtain simple and sensitive fault features. The result shows that the optimal multi-band fuzzy entropy feature has high robustness to noise, and the proposed fusion algorithm can further improve the fault diagnosis rate. Compared with other methods, the fault diagnosis method provided by the invention has higher diagnosis rate and stronger robustness for output voltage, noise, training and testing.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of electric locomotive fault diagnosis, and particularly relates to a traction rectifier fault diagnosis method based on optimal multi-band fuzzy entropy fusion. BACKGROUND

[0002] The traction rectifier is the device with the highest failure rate in the electric locomotive traction drive system, and its failure causes current imbalance, harmonic increase and power conversion failure, resulting in accidents such as unexpected shutdown. Advanced rectifier fault diagnosis technology can improve operation and maintenance capability and reduce maintenance cost. According to the failure rate investigation of the electric locomotive traction drive system, the failure rates of the traction converter and the main circuit breaker are the highest in the entire traction drive system failure distribution, both being 32%. As the core component of locomotive energy conversion, the traction rectifier is mainly composed of a traction rectifier, a traction inverter and a train control unit, and its failure will cause the failure of the entire drive system. The traction rectifier is often accompanied by problems such as high temperature, frequent overvoltage and overcurrent, electromagnetic interference, etc. Due to the complex control strategy and harsh operating conditions, its failure rate is much higher than that of the traction inverter. Therefore, studying the fault technology of the traction rectifier is an effective means to ensure the reliability of electric locomotives.

[0003] At present, the fault diagnosis methods mainly include model-based methods, signal-based methods and data-driven methods. Among them, the model-based method has obvious advantages in diagnosis speed, but the establishment of the model and the determination of the parameters affect its applicability; the signal-based method improves the diagnosis rate, but it needs rich prior knowledge and theoretical analysis of the fault signal. In fact, the data-driven method is suitable for traction rectifier fault diagnosis under variable working conditions and strong noise conditions; the data-driven method, also known as the knowledge method, does not need accurate system model and signal symptoms, but uses historical data of the system to solve the problem of unknown system parameters and complex model. Compared with the other two methods, the data-driven method does not need accurate system model and strict signal analysis, but through data processing and feature extraction of the fault signal, finally uses pattern recognition technology to diagnose the fault equipment. Therefore, it has good robustness and dynamic characteristics.

[0004] However, the existing rectifier fault diagnosis technology ignores the complexity of system model establishment and the difficulty of signal analysis under multiple working conditions, variable loads and strong noise environment.

[0005] The traction rectifier fault signal is nonlinear and non-stationary. In order to extract reasonable fault features, not only the signal needs to be decomposed by multi-resolution, but also the quantitative description of fault features needs to be considered.

[0006] Wavelet packet can decompose the fault signal by multi-resolution, improve the signal processing capability, obtain more detailed low-frequency and high-frequency information, and mine accurate time-frequency features.

[0007] In the feature extraction, the entropy is utilized to represent the nonlinear features caused by the transient change of the system, thereby quantifying the dynamic change of the system fault signal and distinguishing different operating states. Therefore, the entropy-based feature extraction has the advantages of good clustering capability, high classification precision and strong robustness. The fuzzy entropy is highly sensitive to the non-stationarity and nonlinearity of the fault feature of the traction rectifier, can effectively capture the subtle dynamic change of the vibration or current signal of the traction rectifier under different fault states, has excellent anti-noise capability and can stably extract the fault feature in a complex electromagnetic interference environment, does not need to preset a signal model in the calculation process and is highly adaptive to the diversity and uncertainty of the fault signal. Meanwhile, the fuzzy degree is quantified to reflect the system confusion degree and fault severity, thereby providing comprehensive and robust feature basis for fault classification and early warning and finally improving the accuracy and reliability of fault diagnosis.

[0008] However, in the signal processing and feature extraction process, the problem is that the selection of the wavelet function affects the decomposition effect of the fault signal, and different wavelet functions have different decomposition effects on the same fault signal. Therefore, it is crucial to study the wavelet function suitable for the decomposition of the fault signal.

[0009] In the feature extraction process, there is a large amount of redundant information and feature conflict, so that the feature dimension is too large. SUMMARY

[0010] The embodiment of the present application provides a traction rectifier fault diagnosis method based on optimal multi-band fuzzy entropy fusion, to solve the problem of low fault diagnosis accuracy of simple sensitive fault features in the prior art. In order to have a basic understanding of some aspects of the disclosed embodiments, a brief summary is given below. This part is not a general review, nor is it intended to determine the key / important elements or delineate the scope of protection of these embodiments. Its only purpose is to present some concepts in a simple form as a prelude to the detailed description that follows.

[0011] According to a first aspect of the embodiment of the present application, a traction rectifier fault diagnosis method based on optimal multi-band fuzzy entropy fusion is provided.

[0012] In one embodiment, the method comprises the following steps: Collecting input voltage signals of fault modes under different working conditions to form an original sample set; Using different wavelet functions to perform wavelet packet multi-resolution decomposition on the input voltage signals of the fault modes, obtaining different frequency band information, and selecting a wavelet function suitable for the fault signal according to the energy-information entropy ratio of each group of frequency bands, the corresponding frequency band being the optimal frequency band; Calculate the fuzzy entropy of each frequency band coefficient, obtain the energy entropy feature of the optimal frequency band according to the fuzzy entropy of each frequency band coefficient, construct the high-dimensional feature vector corresponding to different working conditions and different fault modes, and use t SNE algorithm for high-dimensional feature fusion to obtain the feature vector. Use a support vector machine for mode recognition to obtain the fault diagnosis result of the traction rectifier.

[0013] On the basis of the above scheme, the input voltage signal of the fault mode is subjected to wavelet packet multi-resolution decomposition using different wavelet functions, different frequency band information is obtained, and the wavelet function suitable for the fault signal is selected according to the energy-information entropy ratio of each group of frequency bands. The corresponding frequency band is the optimal frequency band. The steps specifically include: (1) Calculate the multi-frequency band coefficient: given a fault signal , the fault signal is subjected to m layer multi-resolution decomposition using wavelet packet decomposition to obtain M frequency bands; let the m th coefficient of the n th frequency band node M , n after the r th layer decomposition be , then the low-pass filter coefficient is , and the high-pass filter coefficient is ; The wavelet packet coefficient recursive formula is: (1) After the fault signal is subjected to m layer decomposition, the frequency band coefficients are obtained.

[0014] (2) In the formula, L denotes the length of the n th frequency band; (2) Select the appropriate wavelet function according to the energy entropy ratio criterion: the energy entropy ratio criterion refers to the ratio of the energy value and the information entropy of each frequency band coefficient obtained after wavelet packet decomposition; the greater the energy value, the richer the fault information, and the smaller the information entropy, the smaller the uncertainty of the fault information. Therefore, the optimal wavelet function is selected according to the energy entropy ratio criterion: the energy , information entropy and energy information entropy ratio of each frequency band coefficient: (3) (4) In the formula, denotes the probability distribution. (5) The higher, the more energy selected wavelet function extracts from the analysis signal, the less information entropy, which shows that the selected wavelet function is suitable for the decomposition of the fault diagnosis signal, and the obtained frequency band is the optimal frequency band.

[0015] On the basis of the above scheme, the fuzzy entropy of each frequency band coefficient is calculated, the energy entropy feature of the optimal frequency band is obtained according to the fuzzy entropy of each frequency band coefficient, the high-dimensional feature vectors corresponding to different working conditions and different fault modes are constructed, and the t SNE algorithm is used for high-dimensional feature fusion to obtain the steps of the feature vector, specifically including: (1) Assuming that the coefficient of the optimal frequency band is , , the M dimensional vector is constructed: ( ) (6) (7) (2) Calculate the maximum distance between and : (8) In the formula, ; (3) Calculate the similarity between and : (9) In the formula, is the fuzzy membership function in exponential form. is the boundary gradient, is the similarity tolerance; (4) Calculate the average value of the similarity: (10) (5) Calculate the correlation dimension of and : (11) (6) Make M = M +1, repeat steps (3)-(6) to obtain ; (7) The fuzzy entropy is : (12) Therefore, after the wavelet packet m layer decomposition of the fault signal,M frequency bands, calculate the fuzzy entropy of each frequency band coefficient as the fault feature X H : (13) (8) Assuming high-dimensional fault characteristics ,but and The similar conditional probability is: (14) Where, yes Gaussian variance of (9) Calculate high-dimensional fault features The joint probability : (15) (10) Assume that the low-dimensional data is , Y Initialize low-dimensional data for dimension , using a t-distribution with 1 degree of freedom to calculate the joint probability : (16) (11) The similarity between the high-dimensional data Gaussian distribution P and the low-dimensional data t distribution Q is defined by the Kullback-Leiber divergence: (17) (12) Calculate the gradient To minimize the KL divergence: (18) (13) Obtain low-dimensional feature data : (19) Where, is the learning rate; is the momentum factor; k is the number of iterations; (14) Iterate steps (10)-(13) until k Reaching the maximum value K , then output low-dimensional data : (20) Based on the above solution, the step of using a support vector machine to perform pattern recognition and obtain the fault diagnosis result of the traction rectifier specifically includes: The support vector machine is used as the classifier of the fault feature, a radial basis function robust to noise is selected as the kernel function, Gaussian white noise is added in the collected original signal to simulate the real working environment of the traction rectifier, the optimized fuzzy entropy feature is taken as the input of the SVM classifier through signal decomposition and feature extraction, the test result is compared with the actual result of the trained SVM classifier to obtain the fault diagnosis result.

[0016] On the basis of the above scheme, the step of using the support vector machine as the classifier of the fault feature specifically comprises: The fault feature vector is randomly divided into training samples and test samples, the support vector machine classifier is used for fault diagnosis, the training samples are used for training the support vector machine, the test samples are used for verifying the performance of the support vector machine classifier after training, and thus the diagnosis result is obtained.

[0017] According to a second aspect of the embodiment of the present application, a traction rectifier fault diagnosis device based on optimal multi-band fuzzy entropy fusion is provided, and the device comprises: A data acquisition module is configured to acquire input voltage signals of fault modes under different working conditions to form an original sample set. A signal processing module is configured to use different wavelet functions to perform wavelet packet multi-resolution decomposition on the input voltage signals of the fault modes to obtain different frequency band information, select a wavelet function suitable for the fault signals according to the energy-information entropy ratio of each group of frequency bands, and take the corresponding frequency band as the optimal frequency band. A feature extraction module is configured to calculate the fuzzy entropy of each frequency band coefficient, obtain the energy entropy feature of the optimal frequency band according to the fuzzy entropy of each frequency band coefficient, construct a high-dimensional feature vector corresponding to different working conditions and different fault modes, and perform high-dimensional feature fusion on the high-dimensional feature vector by using the t SNE algorithm to obtain a feature vector. A mode recognition module is configured to use a support vector machine to perform mode recognition and obtain a fault diagnosis result of the traction rectifier.

[0018] According to a third aspect of the embodiment of the present application, a computer readable storage medium is provided.

[0019] In some embodiments, the computer readable storage medium comprises a computer program for saving, wherein the computer program is executed by a processor to implement the traction rectifier fault diagnosis method based on optimal multi-band fuzzy entropy fusion.

[0020] According to a fourth aspect of the embodiment of the present application, a computer device is provided.

[0021] In some embodiments, the computer device comprises 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.

[0022] The technical scheme provided by the embodiment of the application can include the following beneficial effects: In order to ensure the accuracy of diagnosis, the fusion technology is introduced to solve the conflict and redundancy between fault features, and the dimension of fault features is reduced to reduce the burden and calculation amount of the classifier, and specifically, a feature fusion algorithm based on optimal multi-band fuzzy entropy and t t-SNE (OMBFE-t-SNE) is proposed to improve the traction rectifier fault diagnosis technology. First, based on 9 Daubechies (dbN, N = 2, 3, … 10) wavelet functions, the wavelet packet is used to decompose the signals under different working conditions and different fault modes to obtain 9 groups of frequency band information for each fault mode. According to the energy information entropy ratio of each frequency band, the most suitable wavelet function is selected, and the corresponding series of frequency bands are the most suitable multi-band for fault signals. Second, the fuzzy entropy features of the optimal frequency band are calculated, and t-SNE is used to fuse the fault features to reduce the conflict and redundancy between the features and obtain simple and effective fault features. The wavelet packet multi-resolution decomposition is performed on the fault signals to obtain low-frequency and high-frequency fault information. According to the energy information entropy ratio, the optimal wavelet function suitable for fault signal decomposition is selected, and the corresponding frequency band is the optimal frequency band after fault signal decomposition. The optimal frequency band fuzzy entropy feature proposed in the application can quantify the dynamic changes of system fault signals, distinguish different operating states, and reflect the differences between different faults. The algorithm has strong clustering ability, high classification accuracy, and strong robustness to noise. The t-SNE algorithm is used to fuse the optimal frequency band fuzzy entropy features to improve the fault diagnosis rate and solve the redundancy and conflict problems between the features. At the same time, the robustness and stability of the algorithm are evaluated under different noise levels, operating conditions, training and test samples. Compared with other methods, this method has high diagnosis rate and strong robustness.

[0023] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and cannot limit the application. BRIEF DESCRIPTION OF DRAWINGS

[0024] The accompanying drawings, which are incorporated into and form part of the specification, illustrate embodiments consistent with the application and, together with the specification, serve to explain the principles of the application.

[0025] Figure 1 is a traction rectifier circuit structure schematic diagram according to an exemplary embodiment; Figure 2 is a voltage waveform before and after fault triggering according to an exemplary embodimentU ab waveform diagram, wherein, Figure 2 (a-1) is 1.5s when T1 open circuit fault occurs (no noise signal), Figure 2 (a-2) is 1.5s when T1 open circuit fault occurs (with noise signal), Figure 2 (b-1) is 1.5s when T1 and T3 open circuit fault occurs (no noise signal), Figure 2 (b-2) is 1.5s when T1 and T3 open circuit fault occurs (with noise signal), Figure 2 (c-1) is 1.5s when D1 open circuit fault occurs (no noise signal), Figure 2 (c-2) is 1.5s when D1 open circuit fault occurs (with noise signal); Figure 3 is an OMBFE-t-SNE algorithm flow chart according to an example embodiment; Figure 4 is a fault diagnosis method based on OMBFE-t-SNE algorithm according to an example embodiment; Figure 5 is a relationship diagram between energy ratio and fuzzy entropy according to an example embodiment; Figure 6 is a high-dimensional feature fusion after the SVM diagnosis result according to an example embodiment; Figure 7 is an intrinsic dimension of the data containing different signal-to-noise ratio according to an example embodiment; Figure 8 is a comparison of the diagnosis results of the data containing different signal-to-noise ratio according to an example embodiment; Figure 9 is a structural diagram of a computer device according to an example embodiment. DETAILED DESCRIPTION

[0026] The following description and drawings are illustrative of specific embodiments thereof and are not intended to limit the scope of the embodiments. Parts and features of some embodiments can be included or substituted in or for parts and features of other embodiments. The scope of the embodiments encompassed herein includes the whole scope of the claims together with all available equivalents of the claims. In this document, the terms "first", "second", etc. are used merely to distinguish one element from another, and do not require or imply any actual relationship or order between the elements. In fact, the first element can be referred to as the second element, and vice versa. Also, the terms "comprises", "comprising", or any other variations thereof are intended to cover a non-exclusive inclusion, such that a structure, device, or apparatus that comprises a list of elements does not include only those elements but can also include other elements not expressly listed or inherent to such structure, device, or apparatus. Without further limitation, an element defined by an "includes a" statement does not exclude the presence of additional identical elements in the structure, device, or apparatus that includes the element. Various embodiments are described in progressive stages, each of which focuses on the differences from other embodiments, and the same or similar parts between various embodiments can be referred to each other.

[0027] The terms "longitudinal", "lateral", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", and the like, as used herein, indicate relative positions or orientation relationships based on the positions or orientation relationships shown in the drawings, and are only used for the convenience of description herein and simplification of description, and do not indicate or imply that the devices or elements referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as limiting the present application. In the description herein, unless otherwise specified and limited, the terms "mount", "connect", "connection" should be understood broadly, for example, it can be a mechanical connection or an electrical connection, it can be a communication between two elements inside, it can be a direct connection, or an indirect connection through an intermediate medium, and the specific meaning of the above terms can be understood by those skilled in the art according to the specific circumstances.

[0028] In this document, the term "multiple" means two or more, unless otherwise specified.

[0029] In this document, the character " / " represents an "or" relationship between the objects before and after it. For example, A / B means: A or B.

[0030] In this document, the term "and / or" is a description of the relationship between the objects, which means that there can be three relationships. For example, A and / or B means: A or B, or, A and B, the three relationships.

[0031] In the case of no conflict, the embodiments in the present application and the features in the embodiments can be combined with each other.

[0032] Figure 1 An embodiment of the traction rectifier circuit structure of the present application is shown.

[0033] The traction rectifier in the drive system has the functions of energy transmission and feedback when the locomotive is in traction or braking state. In the traction driving state of the locomotive, the pantograph obtains 25kV power from the catenary, the traction transformer reduces the voltage from 25kV to 1450V as the input voltage of the rectifier, and charges the support capacitor at the output end of the rectifier. When the voltage of the support capacitor reaches 2000V, the traction rectifier starts to work and rectifies the 1450V AC to 2800V DC.

[0034] The main sources of traction rectifier faults are the open circuit and short circuit faults of IGBT (denoted as Ti) and power diode (denoted as Di) in the circuit structure, as shown in Figure 1 . The short circuit fault can be easily detected by hardware protection and monitoring devices. However, the open circuit fault does not cause obvious changes in current and voltage in a short time, and is often ignored, which is easy to cause greater losses in the system. This paper mainly studies the open circuit faults of IGBT and power diode.

[0035] The traction rectifier works in a noisy environment, and the rationality of fault signal acquisition affects the complexity of the diagnosis method.

[0036] The present application selects the rectifier input voltage U ab signal as the fault diagnosis signal. In order to simulate the noise environment, 10dB white noise is added to the collected fault diagnosis signal. It is stipulated that the current I N direction from the transformer secondary side to the rectifier, I N >0, otherwise I N <0. At 1.5s, an open circuit fault is injected, and the effects of IGBT and power diode faults on the traction rectifier are analyzed.

[0037] When T1 is open circuit fault, the voltage U ab waveform is shown in Figure 2 (a). When I N >0, the T1 open circuit fault does not affect the system, and the rectifier can still work normally. When I N <0, the support capacitor C dThe inductor cannot be charged, only the secondary winding of the traction transformer U N The inductor L N Charging, voltage waveform distortion, rectifier cannot work normally.

[0038] When T1 and T3 occur OC fault at the same time, the voltage U ab Waveform as Figure 2 (b) shown. Whether I N >0 or I N <0, the support capacitor C d The charging process of the inductor L N Fails, the secondary winding U N And the inductor L N Charges the support capacitor C d , or only the secondary winding U N Charges the inductor L N , the voltage waveform distortion, the rectifier cannot work normally. When D1 is OC fault, the voltage U ab Waveform positive half distortion, as shown in Figure 2 (c) and will cause IGBT overvoltage failure.

[0039] The present application mainly studies the open circuit fault of single IGBT and two IGBTs in IGBT and diode. There are 15 operating modes (1 normal state and 14 fault states), and the corresponding operating modes are shown in Table 1.

[0040] Table 1 Operating mode An embodiment of the traction rectifier fault diagnosis method based on optimal multi-band fuzzy entropy fusion of the present application.

[0041] In order to extract simple and effective features of nonlinear non-stationary signals in a noisy environment, the present application proposes a feature fusion algorithm based on optimal multi-band fuzzy entropy and t Distribution random field embedding (OMBFE-t-SNE).

[0042] Figure 3 An embodiment of the OMBFE-t-SNE algorithm flowchart of the present application is shown.

[0043] S1: Select the optimal wavelet function and perform wavelet packet decomposition on the rectifier signal under different working conditions to obtain a series of optimal multi-bands; Specifically, the following steps are included: (1) Calculate the multi-band coefficients. Given a fault signal , wavelet packet decomposition is used to analyze the fault signal m Layer multi-resolution decomposition, we get M frequency band. m After layer decomposition n Band nodes ( M , n ) r The coefficient is , then the low-pass filter coefficient is , the high-pass filter coefficient is .

[0044] The recursive formula for wavelet packet coefficients is: (1) Fault signal m After layer decomposition, the coefficients of each frequency band are obtained.

[0045] (2) Where, L Indicates the n The length of the frequency band.

[0046] (2) Select the appropriate wavelet function according to the energy entropy ratio criterion. The energy entropy ratio criterion refers to the ratio of the energy value of each frequency band coefficient obtained after wavelet packet decomposition to the information entropy. The larger the energy value, the richer the fault information, and the smaller the information entropy, the smaller the uncertainty of the fault information. Therefore, the optimal wavelet function is selected according to the energy entropy ratio criterion. The energy of each frequency band coefficient is , information entropy and the energy-information entropy ratio : (3) (4) Where, Represents a probability distribution.

[0047] (5) The higher it is, the more energy the selected wavelet function extracts from the analysis signal and the less information entropy it has, which means that the selected wavelet function is suitable for the decomposition of the fault diagnosis signal and the obtained frequency band is the optimal frequency band.

[0048] S2: Calculate the fuzzy entropy of each optimal frequency band as a high-dimensional fault feature; Specifically, the following steps are included: (3) Assuming the coefficient of the optimal frequency band is , , construct M dimensional vector: ( ) (6) (7) (4) Calculate the maximum distance between and : (8) In the formula, .

[0049] (5) Calculate the similarity between and : (9) In the formula, is an exponential fuzzy membership function. is the boundary gradient, is the similarity tolerance.

[0050] (6) Calculate the average value of the similarity.

[0051] (10) (7) Calculate the correlation dimension of and .

[0052] (11) (8) Let M = M +1, repeat steps (3)-(6) to obtain .

[0053] (9) The fuzzy entropy is : (12) Therefore, after the wavelet packet m layer decomposition of the fault signal, the M frequency bands are obtained. The fuzzy entropy of each frequency band coefficient is calculated as the fault feature X H .

[0054] (13) S3: Use t-distributed stochastic neighbor embedding (t-SNE) tDistributed Stochastic Neighbor Embedding, t SNE) to fuse high-dimensional features and obtain simple and sensitive fault features.

[0055] Specifically, the following steps are included: (10) Assuming high-dimensional fault features , and , the conditional probability similar to (14) where is the Gaussian variance of .

[0056] (11) Calculate the joint probability of high-dimensional fault features .

[0057] (15) (12) Assuming that the low-dimensional data is , Y and the dimension is , initialize the low-dimensional data , and calculate the joint probability using the t-distribution with 1 degree of freedom.

[0058] (16) (13) Define the similarity between the high-dimensional data Gaussian distribution P and the low-dimensional data t-distribution Q by Kullback-Leiber divergence.

[0059] (17) (14) Calculate the gradient to minimize the KL divergence.

[0060] (18) (15) Obtain the low-dimensional feature data .

[0061] (19) where is the learning rate; is the momentum factor; k is the number of iterations.

[0062] (16) Iterate steps (12)-(15) until k the maximum value K is reached, and output the low-dimensional data .

[0063] (20) Figure 4 An embodiment of the fault diagnosis method based on the OMBFE-t-SNE algorithm of the application is shown.

[0064] (1) Data collection: collect fault diagnosis signals of 15 kinds of operation modes under different working conditions to form an original sample set.

[0065] (2) Signal processing includes wavelet packet decomposition and optimal frequency band selection: Among them, the wavelet packet decomposition: select the Daubechies series wavelet suitable for signal decomposition and reconstruction, based on 9 Daubechies (db N , N = 2,3, … 10) wavelet functions, the wavelet packet decomposes the signals of different working conditions and different operation modes to obtain 9 groups of frequency band information of each mode; Optimal frequency band selection: select the optimal wavelet function according to the energy information entropy ratio of each frequency band coefficient, and the corresponding frequency band information is the optimal frequency band suitable for the fault mode signal.

[0066] (3) Feature extraction includes fuzzy entropy feature calculation and high-dimensional feature fusion; Fuzzy entropy feature calculation: calculate the fuzzy entropy feature of the optimal frequency band, and construct the high-dimensional feature vector corresponding to different working conditions and different fault modes; High-dimensional feature fusion: use t -SNE algorithm for high-dimensional feature fusion to obtain a simple and effective feature vector.

[0067] (4) Pattern recognition: randomly divide the fault features into training samples and test samples, and use the SVM classifier for fault diagnosis.

[0068] Fault diagnosis result analysis and comparison, specifically including the following steps: Establish a traction rectifier simulation model; the main parameters of the traction rectifier are shown in Table 2. In order to prove the robustness of the above method to the output voltage, the output voltage is changed from 2000V to 2800V, and the input voltage signal of 15 operation modes is collected every 10V U ab Therefore, there are 81 working conditions, each working condition contains 15 operation modes, and the entire sample set has 1215 samples.

[0069] Table 2. Main parameters of traction rectifier 1. Selection of optimal frequency band When the traction rectifier output voltage is 2800V, the rectifier input voltage under 15 operating modes is collected. U ab In order to meet the noise environment of traction rectifier, white noise with a signal-to-noise ratio of 10dB is added. Considering the degree of fault feature mining and the computational cost, this paper selects the commonly used signal decomposition and reconstruction method Daubechies wavelet function (db N ). 9 types of db are used respectively N The (2~10) wavelet function performs five-layer wavelet packet decomposition on the signal and calculates the energy-information entropy ratio of the coefficients in each frequency band.

[0070] Therefore, according to the principle of maximizing the energy-information entropy ratio, the optimal wavelet function suitable for signal decomposition under different working conditions and modes is selected. When the rectifier output voltage is 2800V, the optimal wavelet function selected under different modes is shown in Table 3.

[0071] Table 3. Optimal wavelet functions corresponding to different modes 2. Relationship between energy ratio and fuzzy entropy For 15 different operating modes, the optimal wavelet function in wavelet packet decomposition is used to obtain the optimal 32 frequency bands of the fault signal. The fuzzy entropy of each frequency band is calculated. Figure 5 is the traction rectifier output voltage U dc =2800V, when T1 fault occurs, the energy ratio of each frequency band (expressed as " ”) and fuzzy entropy (expressed by “ ” indicates the relationship between

[0072] exist Figure 5 Among the 32 frequency bands, the first frequency band has the largest energy contribution. The energy contribution of the other frequency bands does not decrease with increasing frequency band number, and some frequency bands exhibit energy spikes. The fuzzy entropy and energy contribution of each frequency band are essentially inversely proportional, with frequency bands with large energy contributions having smaller fuzzy entropies (e.g., frequency bands 5, 9, 13, and 25). Therefore, using fuzzy entropy as a fault signature is feasible.

[0073] 3. Feature extraction based on fuzzy entropy Under a signal-to-noise ratio (SNR) of 10 dB, the input voltage signals of the traction rectifier were collected for 15 operating modes under 81 operating conditions. Based on the optimal frequency band selection method, 32 optimal frequency bands were obtained. The fuzzy entropy of each frequency band coefficient was calculated, resulting in 32 fuzzy entropy features for each operating mode, i.e., 32-dimensional features. Pattern recognition was performed using a support vector machine (SVM), with a training sample to test sample ratio of 3:2. The diagnostic results are shown in Table 4.

[0074] Table 4. SVM diagnosis results based on fuzzy entropy features The results show that the feature extraction of fuzzy entropy is reasonable, but the diagnosis rate of some patterns is particularly low (less than 90%). In order to improve the accuracy of diagnosis, it is necessary to introduce fusion technology to eliminate the redundancy and conflict between features.

[0075] 4. Diagnosis results after high-dimensional feature fusion The initial dimension of the optimal multi-band fuzzy entropy feature containing 10 dB white noise is 32. According to the maximum likelihood estimation (MLE), the intrinsic dimension of this high-dimensional feature is 10.446. In order to not lose useful information, the largest integer is taken, and the intrinsic dimension is 11. The 32-dimensional fault feature of each operating mode is fused into an 11-dimensional fault feature as the input of SVM. According to the preset ratio of 3:2, the diagnosis results are shown in t The initial dimension of the SNE algorithm is set to 32, and the perplexity is set to 30; the target dimension is set to 11. That is, the 32-dimensional fault feature of each operating mode is fused into an 11-dimensional fault feature as the input of SVM. According to the preset ratio of 3:2, the diagnosis results are shown in Figure 6 .

[0076] As shown in Figure 6 , the fault diagnosis rate of this method reaches 99.1667%, which is improved by 7.0834% compared with the fault diagnosis rate based on fuzzy entropy features. The horizontal coordinate is the number of test samples, and the vertical coordinate is the operating mode after SVM classification. For the 220th test sample, the fault mode T3T4 is misdiagnosed as T1T2. For the 435th test sample, the fault mode D3 is misdiagnosed as D4; for the 472th and 473th test samples, the fault mode D4 is misdiagnosed as D3. The diagnosis rate of T3T4 and D3 fault modes is 96.875%. The diagnosis rate of D4 fault mode is 93.75%. The diagnosis rate of other fault modes corresponding to the test samples is 100%. The results show that this algorithm effectively fuses the optimal multi-band fuzzy entropy feature, and has strong robustness to operating condition changes (output voltage changes).

[0077] 5. Robustness analysis of noise White noise with signal-to-noise ratios of 5 dB, 10 dB, 15 dB, 20 dB, 25 dB, and 30 dB is added to the collected signals, and the fuzzy entropy features of the optimal frequency bands are extracted. According to the maximum likelihood estimation, the results of the intrinsic dimension are shown in Figure 7 .

[0078] To avoid losing useful information, we use the largest integer. Therefore, the intrinsic dimensions for different signal-to-noise ratios are 11, 6, 4, 3, 3, and 3. As the signal-to-noise ratio decreases, the intrinsic dimension gradually increases, indicating that the stronger the noise and the greater the interference, the larger the intrinsic dimension of the data.

[0079] use t The SNE algorithm fuses data with different signal-to-noise ratios to obtain corresponding low-dimensional features. SVM is used to identify fault types, running each data set 30 times at a preset ratio of 3:2. Figure 8 The average, maximum, and minimum values ​​of the diagnostic results are given.

[0080] When the signal-to-noise ratio (SNR) was 5 dB, the average, maximum, and minimum values ​​of the 30 diagnostic results were 88.8125%, 90.4167%, and 85.625%, respectively. As the SNR increased, the diagnostic results improved significantly beyond 5 dB. At 15 dB, the average reached 99.5208%, the maximum reached 100%, and the minimum reached 98.75%. This accurate diagnostic rate demonstrates the algorithm's robustness to noise.

[0081] 5.6 Analysis of the robustness of training and testing ratios The training and test sample ratios were set to 2:1, 3:2, 5:1, 10:1, 15:1, and 20:1. The average diagnostic rate after 30 runs of data with signal-to-noise ratios of 5dB, 10dB, 15dB, 20dB, 25dB, and 30dB was used as the evaluation metric. The results are shown in Table 5. Compared with other training and test ratios, the average diagnostic rate for data with different noise ratios was the lowest, at 3:2, with diagnostic results of 88.8125%, 98.2986%, 99.5208%, 99.8056%, 99.875%, and 99.9097%, respectively. The diagnostic results for other training and test ratios were all higher than 3:2. Therefore, this method is also robust to the training and test ratio.

[0082] Table 5. Robustness analysis of different training and testing ratios For the fault diagnosis of traction rectifier, a fault diagnosis device based on optimal multi-band fuzzy entropy fusion is proposed, which includes: data acquisition module, signal processing module, feature extraction module, feature optimization fusion module and pattern recognition module. The diagnostic block diagram is shown in the figure. Figure 5 shown.

[0083] (1) Data acquisition module: collects fault diagnosis signals of 15 operating modes under different working conditions to form an original sample set.

[0084] (2) Signal processing module: including wavelet packet decomposition and optimal frequency band selection; wavelet packet decomposition: selecting Daubechies series wavelet suitable for signal decomposition and reconstruction. Based on 9 Daubechies (db N , N = 2,3, … 10) wavelet functions, the wavelet packet decomposes signals of different working conditions and different operating modes to obtain 9 groups of frequency band information of each mode; optimal frequency band selection: selecting the optimal wavelet function according to the energy information entropy ratio of each frequency band coefficient, and the corresponding frequency band information is the optimal frequency band suitable for the fault mode signal.

[0085] (3) Feature extraction module: including fuzzy entropy feature calculation and high-dimensional feature fusion; fuzzy entropy feature calculation: calculating the fuzzy entropy feature of the optimal frequency band, and constructing the high-dimensional feature vector corresponding to different working conditions and different fault modes; high-dimensional feature fusion: using t-SNE algorithm for high-dimensional feature fusion to obtain a simple and effective feature vector.

[0086] (4) Mode recognition module: randomly dividing the fault feature vector into training samples and test samples, and using the SVM classifier 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, so as to obtain the diagnosis result.

[0087] In an embodiment, a computer device, which can be a server, has an internal structure diagram as shown in Figure 9 The computer device includes a processor, a memory and a network interface connected through 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 the 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 external terminals through network connection. The computer program is executed by the processor to implement the steps in the above method embodiments.

[0088] Those skilled in the art can understand that Figure 9 the structure shown in the above embodiment is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. A specific computer device can include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0089] In one embodiment, a computer device is also provided, including a memory and a processor, the memory has stored therein a computer program, and the processor implements the steps in the above method embodiments when executing the computer program.

[0090] In one embodiment, a computer readable storage medium is provided, having stored thereon a computer program, which, when executed by a processor, implements the steps in the above method embodiments.

[0091] A person of ordinary skill in the art can understand that all or part of the processes in the above method embodiments can be completed by a computer program instructing related hardware, and the computer program can be stored in a non-volatile computer readable storage medium, and when executed, can include the processes of the above method embodiments. Any reference to memory, storage, database or other medium used in the embodiments provided by the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory or optical memory, etc. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0092] In summary, for the open-circuit fault of IGBT and power diode in traction rectifier, an optimal frequency band fuzzy entropy fusion algorithm is proposed. The results show that the proposed method realizes the feature extraction of different nonlinear fault signals in the system under strong noise environment. The diagnosis rate of 15 kinds of operating modes is high, and the output voltage, noise, training and test ratio have good robustness, which shows that the proposed algorithm is suitable for fault diagnosis of traction rectifier under different working conditions in strong noise environment. Compared with other methods, the diagnosis effect of the proposed method is better than that of other methods.

[0093] The present application 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 application is limited only by the appended claims.

Claims

1. A traction rectifier fault diagnosis method based on optimal multi-band fuzzy entropy fusion, characterized by: The following steps are involved: Collect input voltage signals of fault modes under different working conditions to form an original sample set; Using different wavelet functions, the input voltage 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, and the corresponding frequency band is the optimal frequency band. Calculate the fuzzy entropy of each frequency band coefficient, obtain the energy entropy characteristics of the optimal frequency band according to the fuzzy entropy of each frequency band coefficient, and construct high-dimensional feature vectors corresponding to different working conditions and different fault modes. t -SNE algorithm performs high-dimensional feature fusion to obtain feature vectors; Support vector machine is used for pattern recognition to obtain the fault diagnosis results of traction rectifier.

2. The traction rectifier fault diagnosis method based on optimal multi-band fuzzy entropy fusion according to claim 1 is characterized in that: The steps of using different wavelet functions to perform wavelet packet multi-resolution decomposition on the input voltage signal of the fault mode to obtain different frequency band information, and selecting a wavelet function suitable for the fault signal based on the energy-information entropy ratio of each group of frequency bands, and the corresponding frequency band is the optimal frequency band, specifically include: (1) Calculation of multi-band coefficients: Given a fault signal , wavelet packet decomposition is used to analyze the fault signal m Layer multi-resolution decomposition, we get M frequency band; m After layer decomposition n Band nodes ( M , n ) r The coefficient is , then the low-pass filter coefficient is , the high-pass filter coefficient is ; The recursive formula for wavelet packet coefficients is: (1) Fault signal m After layer decomposition, the coefficients of each frequency band are obtained; (2) Where, L Indicates the n The length of the frequency band; (2) Select the appropriate wavelet function according to the energy entropy ratio criterion: The energy entropy ratio criterion refers to the ratio of the energy value of each frequency band coefficient obtained after wavelet packet decomposition to the information entropy; the larger the energy value, the richer the fault information, and the smaller the information entropy, the smaller the uncertainty of the fault information. Therefore, the optimal wavelet function is selected according to the energy entropy ratio criterion: the energy of each frequency band coefficient is the ratio of the energy value of each frequency band coefficient to the information entropy. , information entropy and the energy-information entropy ratio : (3) (4) Where, represents a probability distribution; (5) The higher it is, the more energy the selected wavelet function extracts from the analysis signal and the less information entropy it has, which means that the selected wavelet function is suitable for the decomposition of the fault diagnosis signal and the obtained frequency band is the optimal frequency band.

3. The traction rectifier fault diagnosis method based on optimal multi-band fuzzy entropy fusion according to claim 2 is characterized in that: The fuzzy entropy of each frequency band coefficient is calculated, and the energy entropy characteristics of the optimal frequency band are obtained according to the fuzzy entropy of each frequency band coefficient, and high-dimensional feature vectors corresponding to different working conditions and different fault modes are constructed. t -SNE algorithm for high-dimensional feature fusion, The steps to obtain the feature vector include: (1) Assume that the coefficient of the optimal frequency band is , , build M dimensional vector: ( ) (6) (7) (2) Calculation and Maximum distance between: (8) Where, ; (3) Calculation and Similarity between: (9) Where, is a fuzzy membership function in exponential form. is the boundary gradient, is the similarity tolerance; (4) Calculate the average similarity: (10) (5) Calculation and The correlation dimension of : (11) (6) Make M = M +1, repeat steps (3)-(6), and get ; (7) Fuzzy entropy is : (12) Therefore, the fault signal passes through the wavelet packet m After layer decomposition, we get M frequency bands, calculate the fuzzy entropy of each frequency band coefficient as the fault feature X H : (13) (8) Assuming high-dimensional fault characteristics ,but and The similar conditional probability is: (14) Where, yes Gaussian variance of (9) Calculate high-dimensional fault features The joint probability : (15) (10) Assume that the low-dimensional data is , Y Initialize low-dimensional data for dimension , using a t-distribution with 1 degree of freedom to calculate the joint probability : (16) (11) The similarity between the high-dimensional data Gaussian distribution P and the low-dimensional data t distribution Q is defined by the Kullback-Leiber divergence: (17) (12) Calculate the gradient To minimize the KL divergence: (18) (13) Obtain low-dimensional feature data : (19) Where, is the learning rate; is the momentum factor; k is the number of iterations; (14) Iterate steps (10)-(13) until k Reaching the maximum value K , then output low-dimensional data : (20)。 4. The traction rectifier fault diagnosis method based on optimal multi-band fuzzy entropy fusion according to claim 1 is characterized in that: The step of using a support vector machine to perform pattern recognition to obtain a fault diagnosis result of the traction 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 traction rectifier. Through signal decomposition and feature extraction, the optimized fuzzy 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.

5. The traction rectifier fault diagnosis method based on optimal multi-band fuzzy entropy 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.

6. A traction rectifier fault diagnosis device based on optimal multi-band fuzzy entropy fusion, characterized in that: include: The data acquisition module is used to collect input voltage 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 voltage 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. The feature extraction module is used to calculate the fuzzy entropy of each frequency band coefficient, obtain the energy entropy characteristics of the optimal frequency band according to the fuzzy entropy of each frequency band coefficient, and construct high-dimensional feature vectors corresponding to different working conditions and different fault modes. t -SNE algorithm performs high-dimensional feature fusion to obtain feature vectors; The pattern recognition module is used to perform pattern recognition using a support vector machine to obtain fault diagnosis results of the traction rectifier.

7. 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 traction rectifier fault diagnosis method based on optimal multi-band fuzzy entropy fusion according to any one of claims 1 to 5 is implemented.

8. 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 5 are implemented.

Images