Charging module open-circuit fault rapid diagnosis method and system based on improved migration light gradient elevator

Abstract

The invention discloses a charging module open-circuit fault rapid diagnosis method and system based on an improved migration light gradient elevator, and the method comprises the steps: firstly carrying out the segmentation of operation data of a charging power module through a mobile sliding window, carrying out the time domain feature extraction, and constructing a feature vector set; a LightGBM is used as a basic classifier, a sample weight dynamic updating mechanism is designed, a mixed loss function is constructed in combination with the maximum mean value difference and classification errors, and a migration fault classification model is constructed; inputting a source domain data set and a small amount of target domain data sets into the model for iterative training, and dynamically adjusting sample weights of a source domain and a target domain; and inputting the remaining target domain data set into the classification model for testing. The method is suitable for quickly and accurately diagnosing the open-circuit fault of the power switch tube of the charging power supply module under different working conditions.

Description

Technical Field

[0001] The present invention belongs to the technical field of fault diagnosis, and particularly relates to a method and system for quickly diagnosing open - circuit faults of a charging module based on an improved transfer light gradient boosting machine. Background Art

[0002] In recent years, the growing market demand for new - energy electric vehicles has driven the wide deployment of charging piles, resulting in a steady annual increase in the number of charging piles. As one of the most important components of a charging pile, the safety of the power module has received increasing attention in recent years. Among them, the power - switch device has the highest failure probability in the power module. Fault location and diagnosis of power - switch tubes are beneficial to the online detection of the state of the charging power module and the maintenance after damage, which has important economic significance for the charging operation service industry.

[0003] The fault modes of the switching device mainly manifest as short - circuit faults and open - circuit faults. Short - circuit faults are destructive, but can be detected by hardware protection and converted into open - circuit faults by disabling the switch. On the contrary, open - circuit faults are usually not destructive, but will reduce the overall system performance. Since the evolving open - circuit faults are not obvious in characteristics, more serious faults are likely to occur if they are not diagnosed and repaired in time.

[0004] Currently, data - driven diagnostic methods have shown good performance in the fault diagnosis of power - switch tubes in power modules. However, power modules often operate under changing working conditions, and the existing diagnostic models trained only by the original working - condition database have low generalization ability under changing working conditions. Summary of the Invention

[0005] The main purpose of the present invention is to overcome the disadvantages and deficiencies of the prior art, and provide a method and system for quickly diagnosing open - circuit faults of a charging module based on an improved transfer light gradient boosting machine, which solves the technical problems of poor transfer performance and low diagnostic - result accuracy of data - driven diagnostic methods under changing working conditions.

[0006] To achieve the above purpose, the present invention adopts the following technical solutions:

[0007] In the first aspect, the present invention provides a method for quickly diagnosing open - circuit faults of a charging module based on an improved transfer light gradient boosting machine, including the following steps:

[0008] Obtain the operation data of the charging power module, where the operation data includes original working - condition data and changing - working - condition data. Take the operation data under the original working condition as the source - domain data, and the operation data under the changing working condition as the target - domain data; perform sample segmentation on the operation data to obtain multiple short - data samples;

[0009] Extract time-domain features based on short data samples, construct a set of feature vectors, combine the feature vector sets corresponding to the source domain data and a set number of target domain data as the training set, and use the feature vector sets corresponding to the remaining target domain data as the test set;

[0010] Use LightGBM as the basic classifier, set a dynamic update mechanism for sample weights, construct a hybrid loss function by combining the maximum mean discrepancy and classification error, and build a transfer fault classification model; input the training set into the fault classification model for iterative training, dynamically adjust the sample weights of the source domain and the target domain until the set number of iterations or the sample weights converge, and obtain the trained transfer fault classification model;

[0011] Use the test set to diagnose open circuit faults for the trained transfer fault classification model, generate corresponding fault diagnosis data, and verify its transfer performance on the target domain test set.

[0012] As a preferred technical solution, set the acquisition time according to the fault category, set the moving window length and window interval, and use a sliding window to segment the operation data into samples to obtain multiple short data samples; the short data samples include the number of data points and multiple data points.

[0013] As a preferred technical solution, the time-domain feature extraction based on short data samples includes:

[0014] Calculate the maximum value using the data points to generate the maximum value corresponding to the short data sample;

[0015] Calculate the minimum value using the data points to generate the minimum value corresponding to the short data sample;

[0016] Calculate the mean value using the number of data points and the data points to generate the mean value corresponding to the short data sample;

[0017] Calculate the variance using the mean value, the number of data points and the data points to generate the variance corresponding to the short data sample;

[0018] Calculate the peak-to-peak value using the maximum value and the minimum value to generate the peak-to-peak value corresponding to the short data sample;

[0019] Calculate the kurtosis value using the mean value, the variance, the number of data points and the data points to generate the kurtosis value corresponding to the short data sample;

[0020] Calculate the root mean square using the number of data points and the data points to generate the root mean square corresponding to the short data sample;

[0021] Calculate the crest factor using the maximum value and the root mean square to generate the crest factor corresponding to the short data sample;

[0022] Using the maximum value, the number of data points, and the data points to calculate the pulse factor, and generating the pulse factor corresponding to the short data sample;

[0023] Using the root mean square, the number of data points, and the data points to calculate the form factor, and generating the form factor corresponding to the short data sample;

[0024] Using the maximum value, the number of data points, and the data points to calculate the marginal factor, and generating the marginal factor corresponding to the short data sample;

[0025] Using the maximum value, the minimum value, the mean value, the variance, the peak-to-peak value, the kurtosis value, the root mean square, the crest factor, the pulse factor, the form factor, and the marginal factor as the feature vector to construct a feature vector set.

[0026] As a preferred technical solution, a sample weight dynamic update mechanism is set based on the transfer learning method, including the following steps:

[0027] According to the classification of the source domain data and the target domain data, the training set is divided into subsets to obtain the source domain data subset and the target domain data subset, and each subset includes samples of multiple fault categories;

[0028] Traverse all samples to update the sample weights. For the i-th sample x i , the weight at the k-th iteration is updated as follows:

[0029]

[0030] where K is the total number of iterations, h t (x i ) ∈ {0, 1} is the probability that the i-th sample is predicted as the fault category c, c t (x i ) ∈ {0, 1} is the true probability that the i-th sample is of the category c, X c S is the sample set of the fault category c in the source domain data subset, n s represents the number of source domain data samples in the training set, n t represents the number of target domain data samples in the training set, is the sample set of the fault category c in the target domain data subset, D c is the maximum mean difference between the sample sets of the fault category c in the source domain data subset and the target domain data subset respectively, and the calculation formula is as follows:

[0031]

[0032] where \(k(x, y)\) is set as a Gaussian kernel function, and its calculation formula is as follows:

[0033]

[0034] In the formula, \(\|x - y\|\) 2 represents the square of the Euclidean distance between two data samples, \(\sigma\) is the bandwidth parameter of the kernel function; \(d\) is the distribution difference threshold, \(\beta\in(0, 1)\), which is a constant related only to the number of iteration times and the number of source domain samples, and \(\beta\) t is the change coefficient of the target domain sample weight, which is determined by the weighted error rate of the target domain training samples, is the sample weight of the \(i\)-th sample \(x\) i at the \(k\)-th iteration.

[0035] As a preferred technical solution, the combined maximum mean discrepancy and classification error are used to construct a hybrid loss function as follows:

[0036]

[0037] where \(n\) s represents the number of source domain data samples in the training set, and \(n\) t represents the number of target domain data samples in the training set, \(p\) n,c is the predicted probability of the \(n\)-th sample in the \(c\)-th category, and \(y\) n,c is the true probability of the \(n\)-th sample in the \(c\)-th category, \(\lambda\) is the weight parameter of the maximum mean discrepancy loss, and \(k(\cdot)\) is the kernel function, respectively represent the predicted class probability distributions of the \(i\)-th and \(j\)-th samples in the source domain, respectively represent the predicted class probability distributions of the \(i\)-th and \(j\)-th samples in the target domain.

[0038] As a preferred technical solution, inputting the training set into the fault classification model for iterative training and dynamically adjusting the sample weights of the source domain and the target domain includes:

[0039] S31. Initialize LightGBM and set hyperparameters, and define the objective function as the hybrid loss function;

[0040] S32. Input the training data set into LightGBM, use one-sided sampling to retain high-gradient samples, randomly sample low-gradient samples to reduce the number of samples, and use mutually exclusive feature bundling to merge mutually exclusive features to reduce the feature dimension;

[0041] S33. Discretize the features into multiple intervals, form a histogram from these intervals, obtain the input sample histogram, and use the leaf-wise growth strategy with depth limit to preferentially split the leaf node with the largest gain;

[0042] S34. Repeat steps S31 - S33 until the preset iteration threshold is reached or the performance no longer improves, then stop training.

[0043] As a preferred technical solution, step S31 is specifically:

[0044] Initialize the sample weights, and the calculation formula is as follows:

[0045]

[0046] where, is the initial weight of each sample, and n s represents the number of source - domain data samples in the training set, and n t represents the number of target - domain data samples in the training set;

[0047] Set the total number of weight iterations, and set various hyperparameters, including the learning rate, the maximum depth of the tree, and the number of trees.

[0048] As a preferred technical solution, step S32 is specifically:

[0049] Sort the samples according to the absolute value of the gradient. Set the retention rate a of high - gradient samples and the retention rate b of low - gradient samples. Retain the top a×100% of high - gradient samples, and randomly sample b×100% of low - gradient samples from the remaining (1 - a)×100% of samples for retention, and assign an inflation coefficient ω to the sampled low - gradient samples to compensate for the bias caused by sampling. The inflation coefficient is as follows:

[0050] Calculate the conflict rate between features through mutual - exclusion analysis. The conflict rate is defined as the proportion of samples in which two features both take non - zero values. If the conflict rate is lower than the preset threshold, the mutually - exclusive features are determined and merged to form a new feature space.

[0051] In a second aspect, the present invention also provides a fast open - circuit fault diagnosis system for a charging module based on an improved transfer light gradient boosting machine, which is applied to the fast open - circuit fault diagnosis method for a charging module based on an improved transfer light gradient boosting machine, and includes

[0052] a sample generation module, a feature vector set construction module, a model construction and training module, and a fault diagnosis data generation module;

[0053] The sample generation module is used to obtain the operation data of the charging power module. The operation data includes original working - condition data and changing - working - condition data. The operation data under the original working condition is used as source - domain data, and the operation data under the changing working condition is used as target - domain data; the operation data is segmented into samples to obtain multiple short - data samples;

[0054] The feature vector set construction module is used to extract time-domain features based on short data samples, construct a feature vector set, and use the feature vector sets corresponding to the source domain data and a set number of target domain data as the training set, and the feature vector sets corresponding to the remaining target domain data as the test set;

[0055] The model construction and training module is used to use LightGBM as the basic classifier, set up a dynamic update mechanism for sample weights, construct a hybrid loss function by combining the maximum mean discrepancy and classification error, and construct a transfer fault classification model; input the training set into the fault classification model for iterative training, dynamically adjust the sample weights of the source domain and the target domain until the set number of iterations or the sample weights converge, and obtain the trained transfer fault classification model;

[0056] The fault diagnosis data generation module is used to use the test set to perform open-circuit fault diagnosis on the trained transfer fault classification model, generate corresponding fault diagnosis data, and verify its transfer performance on the target domain test set.

[0057] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0058] The present invention uses the sliding window data segmentation technology to segment the operation data and extract time-domain features, which is beneficial to data analysis under changing working conditions of the model, significantly improves the diagnosis speed, and meets the real-time diagnosis requirements. By designing a hybrid loss function and a sample weight update mechanism, the LightGBM model is improved, and the generalization and transfer capabilities of the model under different working conditions are improved. Even when the number of target domain samples is small, a high diagnosis accuracy can be maintained. Description of the Drawings

[0059] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0060] Figure 1 It is the flowchart of the open-circuit fault rapid diagnosis method for the charging module based on the improved transfer light gradient boosting machine in the embodiment of the present invention;

[0061] Figure 2 It is the flowchart of the sample weight dynamic update mechanism in the embodiment of the present invention;

[0062] Figure 3 It is the flowchart of the improved transfer LightGBM model training in the embodiment of the present invention;

[0063] Figure 4This is a schematic diagram of the topological structure of a charging module according to an embodiment of the present invention;

[0064] Figure 5 The confusion matrix of the migration fault diagnosis learning model on the test set in the embodiment of the present invention;

[0065] Figure 6 It is a structural schematic diagram of a charging module open circuit fault rapid diagnosis system based on an improved migration light gradient lifting machine according to an embodiment of the present invention. DETAILED DESCRIPTION

[0066] In order to enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without making creative work are within the scope of protection of the present application.

[0067] Reference to "embodiments" in this application means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described in this application may be combined with other embodiments.

[0068] Example 1.

[0069] See also Figure 1 This embodiment provides a method for quickly diagnosing an open circuit fault of a charging module based on an improved migration light gradient lifting machine, comprising the following steps:

[0070] S1. Under different fault conditions, obtain the three-phase input current and resonant capacitor voltage signals of the original working condition and the changed working condition for a period of time, set the length and interval of the moving sliding window, and divide the multi-dimensional waveform data of a period of time into multiple short data samples of the same length, where each short data sample contains several data points and their quantity values.

[0071] S2. Perform time domain feature extraction based on short data samples, construct a feature vector set, use the feature vector set corresponding to the source domain data and a set number of target domain data as a training set, and use the feature vector set corresponding to the remaining target domain data as a test set.

[0072] In particular, the time domain feature extraction based on short data samples includes:

[0073] Performing maximum value calculation using the data points to generate a maximum value corresponding to the short data sample;

[0074] Using the data points to perform minimum value calculation, generating a minimum value corresponding to the short data sample;

[0075] Performing mean calculation using the number of data points and the data points to generate a mean corresponding to the short data sample;

[0076] Performing variance calculation using the mean, the number of data points, and the data points to generate a variance corresponding to the short data sample;

[0077] Performing peak-to-peak value calculation using the maximum value and the minimum value to generate a peak-to-peak value corresponding to the short data sample;

[0078] Calculating the kurtosis value using the mean, the variance, the number of data points and the data points to generate the kurtosis value corresponding to the short data sample;

[0079] Performing root mean square calculation using the number of data points and the data points to generate a root mean square corresponding to the short data sample;

[0080] Calculating a crest factor using the maximum value and the root mean square to generate a crest factor corresponding to the short data sample;

[0081] Calculating a pulse factor using the maximum value, the number of data points, and the data points to generate a pulse factor corresponding to the short data sample;

[0082] Calculating a waveform factor using the root mean square, the number of data points, and the data points to generate a waveform factor corresponding to the short data sample;

[0083] Calculating a marginal factor using the maximum value, the number of data points, and the data points to generate a marginal factor corresponding to the short data sample;

[0084] The maximum value, the minimum value, the mean value, the variance, the peak-to-peak value, the kurtosis value, the root mean square, the crest factor, the pulse factor, the waveform factor and the marginal factor are used as feature vectors to construct a feature vector set.

[0085] S3. Use LightGBM as the basic classifier, set a dynamic update mechanism for sample weights, build a hybrid loss function based on the maximum mean difference and classification error, and build a migration fault classification model. Input the training set into the fault classification model for iterative training, and dynamically adjust the sample weights of the source domain and target domain until the set number of iterations or the sample weights converge, and obtain the trained migration fault classification model.

[0086] This embodiment improves the LightGBM model. First, a dynamic update mechanism for sample weights is designed based on the transfer learning method. By adjusting the sample weights, the prediction performance of the model in the target domain is improved. The dynamic update mechanism for sample weights is specifically as follows:

[0087] According to the classification of source domain data and target domain data, the training set is divided into subsets to obtain source domain data subsets and target domain data subsets, each subset includes samples of multiple fault categories;

[0088] like Figure 2 As shown, all samples are traversed to update the sample weights. For the i-th sample x i , the weight at the kth iteration Update as follows:

[0089]

[0090] Among them, K is the total number of iterations, h t (x i )∈{0,1} is the probability that the i-th sample is predicted to be fault category c, c t (x i )∈{0,1} is the true probability that the i-th sample belongs to category c, is the sample set with fault category c in the source domain data subset, n s Represents the number of source domain data samples in the training set, n t represents the number of target domain data samples in the training set, is the sample set with fault category c in the target domain data subset, D c is the maximum mean difference between the sample sets of fault category c in the source domain data subset and the target domain data subset, and the calculation formula is as follows:

[0091]

[0092] Where k(x,y) is set as the Gaussian kernel function, and the calculation formula is as follows:

[0093]

[0094] Where ||xy|| 2 represents the square of the Euclidean distance between two data samples, σ is the bandwidth parameter of the kernel function; d is the distribution difference threshold, β∈(0,1), which is a constant that is only related to the number of iterations and the number of source domain samples, β t is the coefficient of variation of the target domain sample weight, which is determined by the weighted error rate of the target domain training samples. is the i-th sample x i The sample weight at the kth iteration.

[0095] Secondly, in order to further satisfy the generalization ability of the model under different working conditions, this embodiment designs a hybrid loss function in combination with the maximum mean difference and the classification error. The hybrid loss function is a weighted combination of the classification cross entropy error and the maximum mean difference, as shown in the following formula:

[0096]

[0097] Among them, n s Represents the number of source domain data samples in the training set, n t represents the number of target domain data samples in the training set, p n,c is the predicted probability of the nth sample in the cth category, y n,c is the true probability of the nth sample in the cth category, λ is the weight parameter of the maximum mean difference loss, k(·) is the kernel function, Respectively represent the predicted category probability distribution of the i-th sample and the j-th sample in the source domain, They represent the predicted category probability distribution of the i-th sample and the j-th sample in the target domain respectively.

[0098] After the model is built, this embodiment inputs the training set into the migration fault classification model for training, wherein LightGBM is used as the classifier. LightGBM is a fast, distributed, high-performance gradient boosting framework based on the decision tree algorithm. Figure 3 As shown, the training process specifically includes the following steps:

[0099] S31. Initialize LightGBM and set hyperparameters, defining the objective function as a hybrid loss function;

[0100] S32, use unilateral sampling to retain high-gradient samples, randomly sample low-gradient samples to reduce the number of samples, and use mutually exclusive feature bundling to merge mutually exclusive features to reduce feature dimensions;

[0101] S33. Input the training data set into LightGBM, construct the input sample histogram, and use the leaf-oriented growth strategy with depth restriction to preferentially split the leaf nodes with the largest gain;

[0102] S34. Repeat steps S31-S33 until the preset iteration threshold is reached or the performance stops improving, and then stop training.

[0103] Further, step S31 is specifically:

[0104] Initialize the sample weights, and the calculation formula is as follows:

[0105]

[0106] in, is the initial weight for each sample, n s Represents the number of source domain data samples in the training set, n t Represents the number of target domain data samples in the training set;

[0107] Set the total number of weight iterations and various hyperparameters, including the learning rate, the maximum depth of the tree, and the number of trees.

[0108] Furthermore, in step S32, high gradient samples are retained by using single-side sampling, and low gradient samples are randomly sampled to reduce the number of samples, specifically:

[0109] The samples are sorted according to the absolute value of the gradient, and the high gradient sample retention rate a and the low gradient sample retention rate b are set. The high gradient samples of the first a×100% are retained, and the low gradient samples of the last (1-a)×100% are randomly sampled and retained. The expansion coefficient ω is assigned to the sampled low gradient samples to compensate for the deviation caused by sampling, wherein the expansion coefficient is as follows:

[0110] The conflict rate between features is calculated through mutual exclusivity analysis, where the conflict rate is defined as the proportion of samples in which two features take non-zero values ​​at the same time. If the conflict rate is lower than the preset threshold, it is determined to be a mutually exclusive feature, and the mutually exclusive features are merged to form a new feature space.

[0111] S4. Use the test set to perform open circuit fault diagnosis on the trained migration fault classification model, generate corresponding fault diagnosis data, and verify its migration performance on the target domain test set.

[0112] Example 2.

[0113] Step 1: Under different fault conditions, obtain the three-phase input current and resonant capacitor voltage signals of the original working condition and the changed working condition for a period of time, set the length and interval of the moving sliding window, and divide the multi-dimensional waveform data for a period of time into multiple short data samples of the same length.

[0114] The topology of the charging power module is as follows: Figure 4 As shown, the front stage AC-DC is a three-phase VIENNA rectifier structure, and U a , U b , U c is the three-phase grid voltage, L a , L b , L c For the same filter inductance, C 1 , C 2 is a filter capacitor, and the subsequent DC-DC is a full-bridge LLC resonant converter structure. r , C r are the resonant inductor and the resonant capacitor, C3 is the output filter capacitor of the subsequent stage, R 2 is the load resistance, Q 1 ~Q 10 These are the 10 power switch tubes in the charging power module circuit.

[0115] Since in the VIENNA rectifier, the bidirectional switches composed of two series power tubes in the same phase are driven by the same signal, if one of the bidirectional switches is damaged, the other switch in series with it will soon have an open circuit fault. Therefore, we regard the open circuit faults of the bidirectional switches as one type. Similarly, the diagonal switch pairs in the active bridge of the DC-DC converter are driven by the same signal, and an open circuit fault in any of them will have the same effect on the current flowing through the LLC resonant circuit. Therefore, the faults of the diagonal switches are regarded as one type. The specific classification of open circuit faults is shown in Table 1.

[0116] Table 1 Classification of charging module open circuit faults

[0117]

[0118] The experimental device used in the present invention includes a PC controller, a three-phase AC power supply, a charging module, an eight-channel oscilloscope and a load. The switch device in the AC-DC circuit is XNS40N60TH, and the switch device in the DC-DC circuit is CREEC3M0040120K. The main controller is a digital signal processor (DSP) chip TI TMS320F28034, which is used to control the operating state of the charging module.

[0119] The experimental setup includes a power supply with a rated voltage of 220V (RMS), a charging pile power module, and a 12.8Ω resistive load. The switching devices in the AC-DC rectifier and DC-DC converter are XNS40N60TH and C3M0040120K, respectively. The main controller is a digital signal processor (DSP) chip TI TMS320F28034, which is used to control the operating status of the charging module.

[0120] In order to verify the migration capability of the fault diagnosis model, data under three different working conditions were collected for experiments, namely, normal load 20kW, light load 12kW, and unbalanced grid voltage, where the three-phase voltages are 200V, 220V, and 240V, respectively. The above three working conditions are represented by labels 0, 1, and 2, respectively.

[0121] In an embodiment of the present invention, an open circuit fault of a power switch tube is simulated by setting a driving signal to zero. For the 10 fault types analyzed in Table 1, an oscilloscope is used to capture the three-phase input current and resonant capacitor voltage waveforms from normal to faulty at a sampling frequency of 20 kHz. A four-dimensional waveform signal with a collection time of 1 s for each fault category is collected. The moving window length is set to 10 ms, and the window interval is 0.5 ms. The above waveform data is segmented, and multiple short data samples with a time length of 10 ms are obtained for each fault category, and each sample contains 200×4 data points.

[0122] Step 2: Extract the time domain features of the short data samples obtained in step 1 and calculate the maximum value X of each dimension of the short data samples MAX , minimum value X MIN , mean X MEAN , Variance X VAR , peak-to-peak value X PPV , Kurtosis value X KV , RMS X RMS , Crest Factor X CF , pulse factor X IF , Form Factor X WF and the marginal factor X MF There are 11 time domain features in total, that is, each short data sample is converted into a 1×44-dimensional feature vector. The calculation formula is as follows:

[0123] The calculation formula is as follows:

[0124] X MAX =max(x i );

[0125] X MIN =min(x i );

[0126]

[0127] X PPV =X MAX -X MIN

[0128]

[0129] Where N is the number of data points in each dimension of a short data sample, that is, 200, x i represents the i-th data point.

[0130] In the target domain dataset, 10 samples are randomly selected for each fault category, and together with the source domain dataset containing 1000 samples for each fault category, they constitute the training set, and the remaining target domain samples are used as the test set.

[0131] Step 3: Initialize sample weights.

[0132] The total number of weight iterations is set to 5, the hyperparameters of the LightGBM model are initialized, the learning rate is set to 0.1, the maximum depth of the tree is set to 10, the number of trees is set to 150, and the objective function is defined as a hybrid loss function consisting of the maximum mean difference of the training samples in the source domain and the target domain and the classification cross entropy loss, where the maximum mean difference weight coefficient is set to 0.5.

[0133] Step 4: Input the training data set into the LightGBM model initialized in step 3 for training. Figure 3 and Figure 4 Based on the fast computing capability of the LightGBM model, the LightGBM model containing a hybrid loss function is used as the basic classifier. Combined with the transfer learning method, a dynamic sample weight update mechanism is designed to improve domain adaptability.

[0134] The source domain dataset X S and the target domain dataset X T Divide into subsets by category Calculate the maximum mean difference D between the source domain sample subset and the target domain sample subset under each category c ; C is the number of fault categories.

[0135] Traverse all samples to update sample weights. For the i-th sample x i , the weight at the kth iteration renew.

[0136] Repeat model training and sample weight update until the number of iterations reaches the set total number of iterations, and output the generated migration fault diagnosis model.

[0137] Step 5: Input the test set into the migration fault diagnosis model outputted in step 4 to perform open circuit fault diagnosis, generate corresponding fault diagnosis data, and verify the performance of the model through confusion matrix and accuracy.

[0138] In order to verify the performance superiority of the proposed charging power module open circuit fault diagnosis method based on improved migration LightGBM under cross-operating condition diagnosis, the model described in the present invention is compared with several other typical transfer learning models, and the same training samples and test samples are trained and tested under the same migration task.

[0139] The migration tasks are described as follows:

[0140] Task 1: The data of working condition 0 is used as the source domain, and the data of working condition 1 is used as the target domain;

[0141] Task 2: The data of working condition 0 is used as the source domain, and the data of working condition 2 is used as the target domain;

[0142] Task 3: The data of working condition 1 is used as the source domain, and the data of working condition 2 is used as the target domain;

[0143] The test set confusion matrix of the transfer fault diagnosis learning model on Task 2 is as follows: Figure 5 The experimental results of different migration tasks are shown in Table 2.

[0144] Table 2 Comparison of the accuracy of different transfer learning models

[0145]

[0146]

[0147] Through experimental verification, the method can accurately diagnose single switch and multiple switch failures, maintain high-precision diagnosis under different working conditions, and achieve an average accuracy of 99.20% in different migration tasks even when the number of target domain samples is small.

[0148] It should be noted that, for the sake of convenience, the aforementioned method embodiments are all expressed as a series of action combinations, but those skilled in the art should know that the present invention is not limited to the described order of actions, because according to the present invention, certain steps can be performed in other orders or simultaneously.

[0149] Based on the same idea as the method for rapid diagnosis of open circuit faults of charging modules based on an improved migration light gradient hoist in the above-mentioned embodiment, the present invention also provides a system for rapid diagnosis of open circuit faults of charging modules based on an improved migration light gradient hoist, which can be used to execute the above-mentioned method for rapid diagnosis of open circuit faults of charging modules based on an improved migration light gradient hoist. For ease of explanation, the structural schematic diagram of the embodiment of the system for rapid diagnosis of open circuit faults of charging modules based on an improved migration light gradient hoist only shows the parts related to the embodiment of the present invention, and those skilled in the art can understand that the illustrated structure does not constitute a limitation on the device, and may include more or fewer components than shown in the diagram, or a combination of certain components, or a different arrangement of components.

[0150] See also Figure 6 In another embodiment of the present application, a charging module open circuit fault rapid diagnosis system 10 based on an improved migration light gradient lifting machine is provided, the system comprising a sample generation module 11, a feature vector set construction module 12, a model construction and training module 13 and a fault diagnosis data generation module 14;

[0151] The sample generation module 11 is used to obtain the operation data of the charging power supply module, wherein the operation data includes original working condition data and changed working condition data, and the operation data under the original working condition is used as the source domain data, and the operation data under the changed working condition is used as the target domain data; the operation data is sample segmented to obtain multiple short data samples;

[0152] A feature vector set construction module 12 is used to extract time domain features based on short data samples, construct a feature vector set, use the feature vector set corresponding to the source domain data and a set number of target domain data as a training set, and use the feature vector set corresponding to the remaining target domain data as a test set;

[0153] The model building and training module 13 is used to use LightGBM as the basic classifier, set a dynamic update mechanism for sample weights, construct a hybrid loss function combining the maximum mean difference and classification error, and build a migration fault classification model; input the training set into the fault classification model for iterative training, dynamically adjust the sample weights of the source domain and the target domain, and obtain the trained migration fault classification model until the set number of iterations or the sample weights converge;

[0154] The fault diagnosis data generation module 14 is used to perform open circuit fault diagnosis on the trained migration fault classification model using the test set, generate corresponding fault diagnosis data, and verify its migration performance on the target domain test set.

[0155] It should be noted that the charging module open circuit fault rapid diagnosis system based on the improved migration light gradient hoist of the present invention corresponds one to one with the charging module open circuit fault rapid diagnosis method based on the improved migration light gradient hoist of the present invention. The technical features and beneficial effects described in the above-mentioned embodiment of the charging module open circuit fault rapid diagnosis method based on the improved migration light gradient hoist are applicable to the embodiment of the charging module open circuit fault rapid diagnosis method based on the improved migration light gradient hoist. For specific contents, please refer to the description in the embodiment of the method of the present invention, which will not be repeated here. This is hereby declared.

[0156] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0157] The above embodiments are preferred implementation modes of the present invention, but the implementation modes of the present invention are not limited to the above embodiments. Any other changes, modifications, substitutions, combinations, and simplifications that do not deviate from the spirit and principles of the present invention should be equivalent replacement methods and are included in the protection scope of the present invention.

Claims

1. A rapid diagnosis method for open circuit fault of charging module based on improved migration light gradient lifting machine, characterized in that: The steps include: Acquire operation data of the charging power module, wherein the operation data includes original working condition data and changed working condition data, use the operation data under the original working condition as source domain data, and use the operation data under the changed working condition as target domain data; perform sample segmentation on the operation data to obtain multiple short data samples; Extract time domain features based on short data samples, construct a feature vector set, merge the feature vector sets corresponding to the source domain data and a set number of target domain data as a training set, and use the feature vector sets corresponding to the remaining target domain data as a test set; LightGBM is used as the basic classifier, a dynamic update mechanism for sample weights is set, and a hybrid loss function is constructed by combining the maximum mean difference and classification error to build a migration fault classification model. The training set is input into the fault classification model for iterative training, and the weights of source domain and target domain samples are dynamically adjusted until the set number of iterations or the sample weights converge, and the trained migration fault classification model is obtained. The test set is used to perform open circuit fault diagnosis on the trained migration fault classification model, generate corresponding fault diagnosis data, and verify its migration performance on the target domain test set.

2. According to claim 1, the method for rapid diagnosis of open circuit fault of charging module based on improved migration light gradient lifting machine is characterized in that: The acquisition time is set according to the fault category, the moving window length and the window interval are set, and the operating data is sampled using a sliding window to obtain multiple short data samples; the short data samples include the number of data points and multiple data points.

3. According to claim 2, the method for rapid diagnosis of open circuit fault of charging module based on improved migration light gradient lifting machine is characterized in that: The time domain feature extraction based on the short data sample includes: Performing maximum value calculation using the data points to generate a maximum value corresponding to the short data sample; Using the data points to perform minimum value calculation, generating a minimum value corresponding to the short data sample; Performing mean calculation using the number of data points and the data points to generate a mean corresponding to the short data sample; Performing variance calculation using the mean, the number of data points, and the data points to generate a variance corresponding to the short data sample; Performing peak-to-peak value calculation using the maximum value and the minimum value to generate a peak-to-peak value corresponding to the short data sample; Calculating the kurtosis value using the mean, the variance, the number of data points and the data points to generate the kurtosis value corresponding to the short data sample; Performing root mean square calculation using the number of data points and the data points to generate a root mean square corresponding to the short data sample; Calculating a crest factor using the maximum value and the root mean square to generate a crest factor corresponding to the short data sample; Calculating a pulse factor using the maximum value, the number of data points, and the data points to generate a pulse factor corresponding to the short data sample; Calculating a waveform factor using the root mean square, the number of data points, and the data points to generate a waveform factor corresponding to the short data sample; Calculating a marginal factor using the maximum value, the number of data points, and the data points to generate a marginal factor corresponding to the short data sample; The maximum value, the minimum value, the mean value, the variance, the peak-to-peak value, the kurtosis value, the root mean square, the crest factor, the pulse factor, the waveform factor and the marginal factor are used as feature vectors to construct a feature vector set.

4. According to claim 1, the method for rapid diagnosis of open circuit fault of charging module based on improved migration light gradient lifting machine is characterized in that: Setting up a dynamic update mechanism for sample weights based on the transfer learning method includes the following steps: According to the classification of source domain data and target domain data, the training set is divided into subsets to obtain source domain data subsets and target domain data subsets, each subset includes samples of multiple fault categories; Traverse all samples to update sample weights. For the i-th sample x i , the weight at the kth iteration Update as follows: Among them, K is the total number of iterations, h t (x i )∈{0,1} is the probability that the i-th sample is predicted to be fault category c, c t (x i )∈{0,1} is the true probability that the i-th sample belongs to category c, is the sample set with fault category c in the source domain data subset, n s Represents the number of source domain data samples in the training set, n t represents the number of target domain data samples in the training set, is the sample set with fault category c in the target domain data subset, D c is the maximum mean difference between the sample sets of fault category c in the source domain data subset and the target domain data subset, and the calculation formula is as follows: Where k(x,y) is set as the Gaussian kernel function, and the calculation formula is as follows: Where ||xy|| 2 represents the square of the Euclidean distance between two data samples, σ is the bandwidth parameter of the kernel function; d is the distribution difference threshold, β∈(0,1), which is a constant that is only related to the number of iterations and the number of source domain samples, β t is the coefficient of variation of the target domain sample weight, which is determined by the weighted error rate of the target domain training samples. is the i-th sample x i The sample weight at the kth iteration.

5. According to claim 1, the method for rapid diagnosis of open circuit fault of charging module based on improved migration light gradient lifting machine is characterized in that: The hybrid loss function is constructed by combining the maximum mean difference and the classification error, as shown in the following formula: Among them, n s Represents the number of source domain data samples in the training set, n t represents the number of target domain data samples in the training set, p n,c is the predicted probability of the nth sample in the cth category, y n,c is the true probability of the nth sample in the cth category, λ is the weight parameter of the maximum mean difference loss, k(·) is the kernel function, Respectively represent the predicted category probability distribution of the i-th sample and the j-th sample in the source domain, They represent the predicted category probability distribution of the i-th sample and the j-th sample in the target domain respectively.

6. According to claim 1, the method for rapid diagnosis of open circuit fault of charging module based on improved migration light gradient lifting machine is characterized in that: The step of inputting the training set into the fault classification model for iterative training and dynamically adjusting the weights of source domain and target domain samples includes: S31. Initialize LightGBM and set hyperparameters, defining the objective function as a hybrid loss function; S32. Input the training data set into LightGBM, use unilateral sampling to retain high-gradient samples, randomly sample low-gradient samples to reduce the number of samples, and use mutually exclusive feature bundling to merge mutually exclusive features to reduce feature dimensions; S33, discretizing the feature into multiple intervals, forming a histogram of these intervals, obtaining an input sample histogram, and using a leaf-oriented growth strategy with depth restriction to preferentially split the leaf node with the largest gain; S34. Repeat steps S31-S33 until the preset iteration threshold is reached or the performance stops improving, and then stop training.

7. The method for rapid diagnosis of open circuit fault of charging module based on improved migration light gradient lifting machine according to claim 6 is characterized in that: The step S31 is specifically as follows: Initialize the sample weights, the calculation formula is as follows: in, is the initial weight for each sample, n s Represents the number of source domain data samples in the training set, n t Represents the number of target domain data samples in the training set; Set the total number of weight iterations and various hyperparameters, including the learning rate, the maximum depth of the tree, and the number of trees.

8. The method for rapid diagnosis of open circuit fault of charging module based on improved migration light gradient lifting machine according to claim 6 is characterized in that: The step S32 comprises: The samples are sorted according to the absolute value of the gradient, and the high gradient sample retention rate a and the low gradient sample retention rate b are set. The high gradient samples of the first a×100% are retained, and the low gradient samples of the last (1-a)×100% are randomly sampled and retained. The expansion coefficient ω is assigned to the sampled low gradient samples to compensate for the deviation caused by sampling, wherein the expansion coefficient is as follows:

9. The method for rapid diagnosis of open circuit fault of charging module based on improved migration light gradient lifting machine according to claim 6 is characterized in that: The step S32 comprises: The conflict rate between features is calculated through mutual exclusivity analysis, where the conflict rate is defined as the proportion of samples in which two features take non-zero values ​​at the same time. If the conflict rate is lower than the preset threshold, it is determined to be a mutually exclusive feature, and the mutually exclusive features are merged to form a new feature space.

10. A charging module open circuit fault rapid diagnosis system based on an improved migration light gradient hoist, characterized in that: A method for rapid diagnosis of open circuit faults in a charging module based on an improved migration light gradient lifting machine applied to any one of claims 1 to 9, comprising a sample generation module, a feature vector set construction module, a model construction and training module, and a fault diagnosis data generation module; A sample generation module is used to obtain the operation data of the charging power supply module, wherein the operation data includes original working condition data and changed working condition data, and the operation data under the original working condition is used as the source domain data, and the operation data under the changed working condition is used as the target domain data; the operation data is sample segmented to obtain multiple short data samples; A feature vector set construction module is used to extract time domain features based on short data samples, construct a feature vector set, use the feature vector set corresponding to the source domain data and a set number of target domain data as a training set, and use the feature vector set corresponding to the remaining target domain data as a test set; The model building and training module is used to use LightGBM as the basic classifier, set a dynamic update mechanism for sample weights, build a hybrid loss function based on the maximum mean difference and classification error, and build a migration fault classification model; input the training set into the fault classification model for iterative training, dynamically adjust the sample weights of the source domain and the target domain, and obtain the trained migration fault classification model until the set number of iterations or the sample weights converge; The fault diagnosis data generation module is used to perform open circuit fault diagnosis on the trained migration fault classification model using the test set, generate corresponding fault diagnosis data, and verify its migration performance on the target domain test set.

Images