A method for locating and identifying a grounding fault of a power supply system of an electric locomotive

Abstract

The application provides a power locomotive power supply system grounding fault positioning and identification method based on timing characteristics, relates to the technical field of power locomotive traction transmission systems, constructs two characteristic vectors with fault positioning distinguishability based on historical system signal data of the power locomotive power supply system, extracts characteristic indexes of the characteristic vectors in a sliding window, makes the characteristic index data have timing characteristics, improves the distinguishability of different types of faults, reduces the influence of data noise and other interferences, constructs a grounding fault positioning and identification model through machine learning, mines a nonlinear mapping relationship between the characteristic indexes and fault types, realizes power supply system grounding fault positioning, solves the problems of single detection characteristics of the over-limit alarm method, limitations and original data waveform noise interference, reduces false positives, missed reports and other problems, improves the precision and accuracy of grounding faults, and guarantees the operation efficiency and safety of the power locomotive.

Description

Technical Field

[0001] This invention relates to the technical field of electric locomotive traction drive systems, and more specifically, to a method for locating and identifying grounding faults in the power supply system of an electric locomotive. Background Technology

[0002] Currently, most passenger trains in my country use a centralized DC 600V power supply with distributed converters. The locomotive provides a DC 600V bus voltage, and converters in each carriage convert the DC 600V voltage into three-phase 380V and single-phase 220V AC power to supply power to the carriage's electrical equipment (such as air conditioning, heating, and lighting). Therefore, when a short circuit occurs in the DC 600V system, the current cannot be quickly cut off, requiring timely identification, location, and fault resolution to ensure stable train operation and passenger safety.

[0003] Currently, grounding detection in locomotive and passenger car power supply systems still relies on over-limit alarm methods, primarily targeting the main lines of DC 600V systems. This method is limited in its detection methods and characteristics, and heavily depends on engineering experience to set threshold limits, making it prone to misdiagnosis. Furthermore, this method can only detect grounding faults that have occurred, not their precise location, causing difficulties for subsequent maintenance and impacting efficiency. In actual engineering practice, due to factors such as operating conditions, untimely maintenance, and the limitations of over-limit alarm detection, false alarms and missed alarms can occur. If the accuracy and precision of grounding fault location are not improved, it will severely affect the efficiency and safety of train operations.

[0004] Existing technology discloses a ground fault diagnosis method for a DC 600V train power supply system. This method acquires fault waveform sample sets of the DC 600V train power supply system under different ground fault types, trains these sample sets to obtain a fault classification model, and when a ground fault occurs in the target DC 600V train power supply system, it collects the current fault waveform in real time and diagnoses the ground fault type corresponding to the current fault waveform based on the fault classification model. However, in this scheme, the original waveform data does not undergo feature extraction, making it susceptible to noise interference. To ensure the accuracy of ground fault identification, an external leakage current detection device is required, and half-voltage and leakage current are combined to determine whether a ground fault has occurred in the power supply system, increasing the data acquisition burden and affecting the efficiency of ground fault location and identification. Summary of the Invention

[0005] To address the issues of low efficiency, limited accuracy, and limited precision in current electric locomotive power supply system ground fault location and identification methods, this invention proposes a ground fault location and identification method for electric locomotive power supply systems. This method constructs two new feature vectors associated with the location of the ground fault from the original data. By extracting feature indicators from these feature vectors within a sliding window, ground fault location and identification are performed. This improves the distinguishability of different types of faults and reduces the impact of data noise and other interference, resulting in high efficiency and accuracy.

[0006] To achieve the above-mentioned technical effects, the technical solution of the present invention is as follows:

[0007] A method for locating and identifying grounding faults in the power supply system of an electric locomotive includes:

[0008] S1. Collect historical system signal data of the electric locomotive power supply system, and construct a feature vector with fault location discrimination based on the historical system signal data;

[0009] S2. Perform sliding window segmentation on the feature vectors to obtain the feature indices of the feature vectors. Construct a dataset from the feature indices of the feature vectors and divide the dataset into training set and test set.

[0010] S3. Construct a ground fault location and identification model, train the ground fault location and identification model using the training set, and test the ground fault location and identification model using the test set to obtain a trained ground fault location and identification model;

[0011] S4. Collect real-time system signal data of the electric locomotive power supply system, extract the real-time feature vector of the real-time system signal data, and determine whether a ground fault has occurred in the electric locomotive power supply system based on the real-time feature vector. If yes, proceed to step S5; otherwise, continue to collect real-time system signal data.

[0012] S5. Extract cached data segments from the real-time feature vector and calculate the feature indicators corresponding to the cached data segments;

[0013] S6. Input the feature indicators corresponding to the cached data segment into the trained ground fault location and identification model to perform fault type analysis and output the fault location and identification results.

[0014] Preferably, in step S1, the historical system signal data of the electric locomotive power supply system includes: the AC side current I of the electric locomotive power supply system. s Ground fault detection voltage U jd and DC side voltage U dc .

[0015] Preferably, in step S1, a feature vector I with fault location discrimination is constructed. x1 I x2The formula is as follows:

[0016] I x1 =U jd -0.5U dc

[0017] I x2 =I x1 ×I s =(U jd -0.5U dc )I s

[0018] Here, based on the principle of grounding faults, two feature vectors with fault location discrimination are constructed, which reduces the probability of false alarms and missed alarms and improves the monitoring and maintenance efficiency of the power supply system.

[0019] Preferably, in step S2, a window of length L is used to process the feature vector I of the accompanying time series of length H. x1 and I x2 Perform sliding segmentation and set a sliding step size S to make the window slide along the time series direction and slide at the set sliding step size. Each slide obtains a data segment of length L. When the remaining data segment of the feature vector is less than the set sliding step size, stop sliding and obtain the feature index of the feature vector after sliding segmentation.

[0020] Preferably, in step S2, for feature vector I x1 I x2 After performing sliding window segmentation, the feature vector I is calculated for each power grid frequency AC cycle. x1 I x2 The mean and variance are calculated using the following formulas:

[0021]

[0022]

[0023]

[0024]

[0025] Where k is the start time of the sliding window segmentation process, N is the number of feature vectors within the sliding window, and i = 1, 2, ..., N, J E1 For the feature vector I x1 The average value of J D1 For the feature vector I x1 The variance, J E2 For the feature vector I x2 The average value of J D2 For the feature vector I x2The variance of J; E1 and J D1 As the feature vector I x1 The characteristic indicators of J E2 and J D2 As the feature vector I x2 Characteristic indicators.

[0026] Here, by extracting feature indices from feature vectors within a sliding window, the feature vector data is given temporal features, resulting in feature indices for feature vectors within a period. This further improves the accuracy of ground fault detection and enhances the generalization ability of the ground fault location and identification model.

[0027] Preferably, before performing step S3, the feature indicators in the dataset are divided into sample data of different grounding fault categories, including: DC positive terminal grounding fault, DC negative terminal grounding fault, reactor front end grounding fault, AC positive terminal grounding fault, and AC negative terminal grounding fault.

[0028] The sample data for each fault category are transposed and normalized, with normalization parameters of 0 and 1.

[0029] Preferably, in step S3, a ground fault location and identification model is constructed based on a machine learning algorithm. The model parameters of the ground fault location and identification model include: machine learning input layer parameters and hidden layer parameters. The hidden layer input weights and node thresholds are optimized. By mining the nonlinear mapping relationship between feature indicators and fault types through machine learning, it is beneficial to accurately locate ground faults.

[0030] Preferably, in step S4, the real-time system signal data of the electric locomotive power supply system includes: the real-time AC side current I′ of the electric locomotive power supply system. s Real-time ground fault detection voltage U′ jd and real-time DC side voltage U′ dc Based on the real-time system signal data of the electric locomotive power supply system, extract the real-time feature vector I′ of the real-time system signal data. x1 、I′ x2 .

[0031] Preferably, in step S4, the method for determining whether a ground fault has occurred in the electric locomotive power supply system is as follows:

[0032] If I' x1 =0 indicates that the electric locomotive power supply system is normal, and real-time system signal data is collected again for a new round of real-time feature vector extraction; if I′ x1 If ≠0, it indicates that a grounding fault has occurred in the power supply system of the electric locomotive.

[0033] Preferably, in step S5, for the real-time feature vector I′ x1 、I′ x2 Perform sliding window segmentation to obtain a buffered data segment I′ of length L. sw1 、I′ sw2 ;

[0034] Calculate cached data segment I′ sw1 The average value J′ E1 Sum of variance J′ D1 And used as cached data segment I′ sw1 The characteristic indicators are used to calculate the cached data segment I′. sw2 The average value J′ E2 Sum of variance J′ D2 And used as cached data segment I′ sw2 Characteristic indicators.

[0035] Compared with the prior art, the beneficial effects of the technical solution of the present invention are:

[0036] This invention proposes a ground fault location and identification method for electric locomotive power supply systems based on time-series features. It constructs two feature vectors with fault location discrimination based on historical system signal data of the electric locomotive power supply system, and extracts feature indicators within a sliding window from these feature vectors. This imbues the feature indicator data with time-series characteristics, improving the distinguishability of different types of faults and reducing the impact of data noise and other interference. A ground fault location and identification model is constructed through machine learning, exploring the nonlinear mapping relationship between feature indicators and fault types to achieve ground fault location in the power supply system. This method addresses the limitations of single detection features in over-limit alarm methods and the interference of noise in the original data waveform, reducing false alarms and missed alarms, improving the accuracy and precision of ground fault detection, and ensuring the operating efficiency and safety of electric locomotives. Attached Figure Description

[0037] Figure 1 This diagram illustrates the main circuit schematic of the electric locomotive power supply system proposed in Embodiment 1 of the present invention.

[0038] Figure 2 This is a schematic diagram of the ground fault location and identification method for electric locomotive power supply system proposed in Embodiment 1 of the present invention;

[0039] Figure 3 This diagram illustrates the principle of DC-side positive terminal grounding fault in the electric locomotive power supply system proposed in Embodiment 2 of the present invention.

[0040] Figure 4 This diagram illustrates the principle of DC-side negative terminal grounding fault in the electric locomotive power supply system proposed in Embodiment 2 of the present invention.

[0041] Figure 5This diagram illustrates the principle of grounding fault at the front end of the reactor in the electric locomotive power supply system proposed in Embodiment 2 of the present invention.

[0042] Figure 6 This diagram illustrates the principle of AC positive terminal grounding fault in the electric locomotive power supply system proposed in Embodiment 2 of the present invention.

[0043] Figure 7 This diagram illustrates the principle of AC side negative terminal grounding fault in the electric locomotive power supply system proposed in Embodiment 2 of the present invention.

[0044] Figure 8 This diagram shows the structure of the SSA-ELM model proposed in Embodiment 3 of the present invention.

[0045] Figure 9 This is a flowchart illustrating the localization process of the SSA-ELM model proposed in Embodiment 3 of the present invention. Detailed Implementation

[0046] The accompanying drawings are for illustrative purposes only and should not be construed as limiting the scope of this patent.

[0047] To better illustrate this embodiment, some parts of the accompanying drawings may be omitted, enlarged, or reduced, and do not represent the actual dimensions;

[0048] It is understandable to those skilled in the art that some well-known details may be omitted from the accompanying drawings.

[0049] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0050] The positional relationships depicted in the accompanying drawings are for illustrative purposes only and should not be construed as limiting this patent.

[0051] Example 1

[0052] In this embodiment, as Figure 1 As shown, the main circuit of the electric locomotive power supply system includes a traction transformer, a main circuit breaker VCB, and an electric locomotive power supply system converter. The main structure of the electric locomotive power supply system converter consists of a single-phase bridge semi-controlled rectifier circuit, an intermediate DC circuit, and a grounding detection circuit.

[0053] Single-phase AC 25kV alternating current flows into the locomotive body via the pantograph, main circuit breaker VCB, and the primary winding of the traction transformer. After being stepped down by the traction transformer, it supplies single-phase AC power to the electric locomotive power supply system converter through the secondary winding. Under the action of the electric locomotive power supply system converter, the single-phase AC power is converted into DC 600V direct current to power the electric locomotive load; for example... Figure 1 As shown, Figure 1The serial numbers ① to ⑤ in the table represent five different grounding fault locations, namely DC positive terminal grounding, DC negative terminal grounding, reactor front end grounding, AC positive terminal grounding, and AC negative terminal grounding.

[0054] like Figure 2 As shown in the figure, this embodiment proposes a method for locating and identifying grounding faults in the power supply system of electric locomotives, including the following steps:

[0055] S1. Collect historical system signal data of the electric locomotive power supply system, and construct a feature vector with fault location discrimination based on the historical system signal data;

[0056] The historical system signal data of the electric locomotive power supply system includes: the AC side current I of the electric locomotive power supply system. s Ground fault detection voltage U jd and DC side voltage U dc ;

[0057] Construct a feature vector I with fault location discrimination. x1 I x2 The formula is as follows:

[0058] I x1 =U jd -0.5U dc

[0059] I x2 =I x1 ×I s =(U jd -0.5U dc )I s

[0060] S2. Perform sliding window segmentation on the feature vectors to obtain the feature indices of the feature vectors. Construct a dataset from the feature indices of the feature vectors and divide the dataset into training set and test set.

[0061] Using a window of length L to analyze the feature vector I of an accompanying time series of length H. x1 and I x2 Sliding segmentation is performed, and a sliding step size S is set. In this embodiment, the length of each sliding window is set to 400, the step size is set to 10, and 6 groups are collected, each with 1960 feature index samples. The window is moved along the time series direction and slides according to the set sliding step size. Each slide obtains a data segment of length 400. When the remaining data segment of the feature vector is less than the set sliding step size 10, the sliding stops, and the feature index of the feature vector after sliding segmentation is obtained.

[0062] For the eigenvector I x1 I x2After performing sliding window segmentation, the feature vector I within one cycle of the power frequency of 50Hz is calculated. x1 I x2 The mean and variance are calculated using the following formulas:

[0063]

[0064]

[0065]

[0066]

[0067] Where k is the start time of the sliding window segmentation process, N is the number of feature vectors within the sliding window, and i = 1, 2, ..., N, J E1 For the feature vector I x1 The average value of J D1 For the feature vector I x1 The variance, J E2 For the feature vector I x2 The average value of J D2 For the feature vector I x2 The variance of J; E1 and J D1 As the feature vector I x1 The characteristic indicators of J E2 and J D2 As the feature vector I x2 Characteristic indicators.

[0068] In this embodiment, the feature indicators in the dataset are divided into sample data of different grounding fault categories. The fault categories include: DC positive terminal grounding fault, DC negative terminal grounding fault, reactor front-end grounding fault, AC positive terminal grounding fault, and AC negative terminal grounding fault. The sample data of each fault category are transposed and normalized, wherein the normalization parameters are 0 and 1.

[0069] S3. Construct a ground fault location and identification model, train the ground fault location and identification model using the training set, and test the ground fault location and identification model using the test set to obtain a trained ground fault location and identification model;

[0070] In this embodiment, a ground fault location and identification model is constructed based on a machine learning algorithm. The model parameters of the ground fault location and identification model include: machine learning input layer parameters, hidden layer parameters, and optimized hidden layer input weights and node thresholds.

[0071] S4. Collect real-time system signal data of the electric locomotive power supply system, extract the real-time feature vector of the real-time system signal data, and determine whether a ground fault has occurred in the electric locomotive power supply system based on the real-time feature vector. If yes, proceed to step S5; otherwise, continue to collect real-time system signal data.

[0072] Real-time system signal data of the electric locomotive power supply system includes: real-time AC side current I′ of the electric locomotive power supply system. s Real-time ground fault detection voltage U′ jd and real-time DC side voltage U′ dc Based on the real-time system signal data of the electric locomotive power supply system, extract the real-time feature vector I′ of the real-time system signal data. x1 、I′ x2 The following are methods for determining whether a ground fault has occurred in the power supply system of an electric locomotive:

[0073] If I' x1 =0 indicates that the electric locomotive power supply system is normal, and real-time system signal data is collected again for a new round of real-time feature vector extraction; if I′ x1 If ≠0, it indicates that a grounding fault has occurred in the power supply system of the electric locomotive.

[0074] S5. Extract cached data segments from the real-time feature vector and calculate the feature indicators corresponding to the cached data segments;

[0075] For the real-time feature vector I′ x1 、I′ x2 In this embodiment, a buffered data segment I′ with a length of 400 and a movement step of 10 is obtained for sliding window segmentation. sw1 、I′ sw2 ;

[0076] Calculate cached data segment I′ sw1 The average value J′ E1 Sum of variance J′ D1 And used as cached data segment I′ sw1 The characteristic indicators are used to calculate the cached data segment I′. sw2 The average value J′ E2 Sum of variance J′ D2 And used as cached data segment I′ sw2 Characteristic indicators.

[0077] S6. Input the feature indicators corresponding to the cached data segment into the trained ground fault location and identification model to perform fault type analysis and output the fault location and identification results.

[0078] Example 2

[0079] In this embodiment, see Figure 1The grounding detection circuit includes resistors R1, R2, and R3, and a voltage sensor VH2. One end of resistor R1 is connected to one end of voltage sensor VH2, indicating the location of a DC positive terminal grounding fault. The other end of resistor R1 is connected to one end of resistor R2 and one end of resistor R3. The other end of resistor R3 is connected to the other end of voltage sensor VH2 and grounded. The other end of R2 indicates the location of a DC negative terminal grounding fault.

[0080] Based on Example 1, in step S1, historical system signal data of the electric locomotive power supply system is collected. Based on the historical system signal data, a feature vector with fault location discrimination is constructed using ground fault mechanism analysis. The ground fault mechanism analysis is as follows:

[0081] S11. Collect the AC side voltage U of the electric locomotive power supply system. s AC side current I s Ground fault detection voltage U jd and DC side voltage U dc ;

[0082] S12. When the electric locomotive is running normally, the equivalent grounding resistance value R at each location is... jd The voltage is in the MΩ range. Since the fixed discharge resistors R1 and R2 have equal resistance values, the ground detection voltage U collected by the voltage sensor VH2 is... jd U is the DC side voltage dc Half of the grounding detection voltage U jd U with DC side voltage dc Constructing feature vector I x1 , when I x1 When the value is 0, the electric locomotive operates normally without faults;

[0083] S13. When a DC-side positive terminal ground fault occurs in the power supply system, the ground fault detection voltage U jd It is a direct flow, and is related to U dc Proportional to the grounding insulation, as the grounding insulation gradually deteriorates, the equivalent grounding resistance gradually decreases to 0, and the grounding detection voltage also decreases from 0.5U. dc Gradually decreasing to 0, the feature vector I x1 Maintain -0.5U dc x1 <0, and moving towards -0.5U dc Decreasing in direction; at this time, such as Figure 3 As shown, the ground detection voltage U at the positive terminal of the DC side jd The expression is as follows:

[0084]

[0085] S14. When a DC-side negative terminal ground fault occurs in the power supply system, the ground detection voltage U​jd For direct flow and with U dc Proportional to the grounding insulation, as the grounding insulation gradually deteriorates, the equivalent grounding resistance gradually decreases to 0, and the grounding detection voltage decreases from 0.5U. dc Gradually increase to U dc eigenvector I x1 Keep 0 x1 <0.5U dc and toward 0.5U dc Increasing in direction; at this time, such as Figure 4 As shown, the ground detection voltage U at the negative terminal of the DC side jd The expression is as follows:

[0086]

[0087] S15. When a ground fault occurs at the front end of the reactor in the power supply system, the ground detection voltage U jd with U dc U s I s Related, ground detection voltage U jd The DC component corresponding to the grounding detection voltage grounded to the negative terminal of the DC side, and the DC component corresponding to U s I s The strongly correlated communication component consists of two parts, where the communication part is based on I. s The piecewise function distinguishes the direction, and the proportion of the AC component increases as the ground fault worsens; at this time, such as Figure 5 As shown, the grounding detection voltage U at the front end of the reactor jd The expression is as follows:

[0088]

[0089]

[0090] S16. When an AC positive terminal ground fault occurs in the power supply system, the ground detection voltage U jd with U dc U s I s Related, ground detection voltage U jd The DC component corresponding to the grounding detection voltage grounded to the negative terminal of the DC side, and the DC component corresponding to U s I s The strongly correlated communication component consists of two parts, where the communication part is based on I. s Piecewise functions that differentiate directions, if I s If the value is less than 0, its AC component is 0; at this time, such as Figure 6 As shown, the grounding detection voltage U at the positive terminal of the AC side jd The expression is as follows: ​

[0091]

[0092]

[0093] S17. When a ground fault occurs on the negative terminal of the AC side of the power supply system, the ground detection voltage U jd with U dc U s I s Related, ground detection voltage U jd The DC component corresponding to the grounding detection voltage grounded to the negative terminal of the DC side, and the DC component corresponding to U s I s The strongly correlated communication component consists of two parts, where the communication part is based on I. s Piecewise functions that differentiate directions, if I s If it is not less than 0, its AC component is 0; at this time, such as Figure 7 As shown, the grounding detection voltage U at the negative terminal of the AC side jd The expression is as follows:

[0094]

[0095]

[0096] Based on the above analysis of the grounding fault mechanism, a feature vector I with fault location discrimination is constructed. x1 I x2 , among which, I x1 =U jd -0.5U dc I x2 =I x1 ×I s =(U jd -0.5U dc )I s .

[0097] In this embodiment, the grounding fault simulation of the electric locomotive power supply system is performed using MATLAB-Simulink. The data of the system under normal conditions and grounding faults are used as the raw data, including the normal state and 5 types of grounding fault states, for a total of 6 sets of data, with 20,000 samples in each set.

[0098] Extract the feature vector I from each set of original data. x1 I x2 Generate new feature vector sample data, also in 6 groups, each with 20,000 samples, and construct the feature vector sample data into a dataset.

[0099] Example 3

[0100] Based on Example 1, a ground fault location and identification model is constructed using machine learning algorithms. In this example, as shown... Figure 8 As shown, a ground fault location and identification model is constructed based on the Sparrow Search algorithm Extreme Learning Machine (SSA-ELM). The ground fault location and identification model includes an input layer, a hidden layer, and an output layer. Feature indicators are used as input. The input layer parameters of the machine learning model are set to a 4-dimensional vector, and the number of hidden layer nodes is 10. The Sparrow Search algorithm is used to optimize the hidden layer input weights ω. i and node threshold b i The output layer is 6-dimensional, the activation function is sig; the sparrow size is 5, the maximum number of iterations is 100, the discoverers account for 20% of the population, and the alarm threshold ST is 0.8.

[0101] The ground fault location and identification model is trained using the training set and tested using the test set to obtain the trained ground fault location and identification model.

[0102] like Figure 9 As shown, the dataset is divided into a training set p_train and a test set p_test. The ground fault location and recognition model is trained using the training set p_train: ELM parameters, SSA parameters, weights, etc., are initialized, the global optimal fitness value and optimal population are obtained, the optimization algorithm iteratively calculates the fitness values ​​of all sparrows, updates the optimal fitness value and population, and determines whether the optimization termination condition is met. If the condition is not met, the fitness value is re-optimized. If the condition is met, the ground fault location and recognition model obtains the optimal weights and thresholds of ELM internally, and the ground fault location and recognition model training is completed. The test set p_test is then input into the trained ground fault location and recognition model for detection and recognition.

[0103] In other embodiments, other optimization algorithms can be used to construct the ground fault location and identification model, such as Random Forest, Gaussian Mixture Model with Hidden Markov Chain (GMM-HMM), Cumulative Sum Algorithm (CUSUM), etc.

[0104] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the implementation of the present invention. Those skilled in the art can make other variations or modifications based on the above description. It is neither necessary nor possible to exhaustively describe all embodiments here. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the claims of the present invention.

Claims

1. A method for locating and identifying grounding faults in the power supply system of an electric locomotive, comprising: S1. Collect historical system signal data of the electric locomotive power supply system, and construct a feature vector with fault location discrimination based on the historical system signal data; S2. Perform sliding window segmentation on the feature vectors to obtain the feature indices of the feature vectors. Construct a dataset from the feature indices of the feature vectors and divide the dataset into training set and test set. S3. Construct a ground fault location and identification model, train the ground fault location and identification model using the training set, and test the ground fault location and identification model using the test set to obtain a trained ground fault location and identification model; S4. Collect real-time system signal data of the electric locomotive power supply system, extract real-time feature vectors from the real-time system signal data, and determine whether a ground fault has occurred in the electric locomotive power supply system based on the real-time feature vectors. If yes, proceed to step S5; otherwise, continue collecting real-time system signal data. S5. Extract cached data segments from the real-time feature vector and calculate the feature indicators corresponding to the cached data segments; S6. Input the feature indicators corresponding to the cached data segment into the trained ground fault location and identification model to perform fault type analysis and output the fault location and identification results; In step S1, the historical system signal data of the electric locomotive power supply system includes: the AC side current of the electric locomotive power supply system. Ground fault detection voltage and DC side voltage ; In step S1, a feature vector with fault location discrimination is constructed. , The formula is as follows: 。 2. The method for locating and identifying grounding faults in the power supply system of an electric locomotive according to claim 1, characterized in that, In step S2, using a length of L The window has a length of H Feature vectors of the accompanying time series and Perform sliding segmentation and cutoff, and set the sliding step size. S This causes the window to slide along the time series direction, with a set sliding step size. Each slide obtains a segment of length [missing information]. L When the remaining data segment of the feature vector is less than the set sliding step size, the sliding stops, and the feature index of the feature vector after sliding segmentation is obtained.

3. The method for locating and identifying grounding faults in the power supply system of an electric locomotive according to claim 1, characterized in that, In step S2, for the feature vector , After performing sliding window segmentation, the feature vectors within one power frequency AC cycle of the power grid are calculated respectively. , The mean and variance are calculated using the following formulas: in, k This marks the start time of the sliding window segmentation process. N This represents the number of feature vectors within the sliding window. , For feature vectors The average value, For feature vectors variance For feature vectors The average value, For feature vectors The variance; and As feature vectors The characteristic indicators will and As feature vectors Characteristic indicators.

4. The method for locating and identifying grounding faults in the power supply system of an electric locomotive according to claim 1, characterized in that, Before executing step S3, the feature indicators in the dataset are divided into sample data of different ground fault categories. The fault categories include: DC positive terminal ground fault, DC negative terminal ground fault, reactor front end ground fault, AC positive terminal ground fault, and AC negative terminal ground fault. The sample data for each fault category are transposed and normalized, with normalization parameters of 0 and 1.

5. The method for locating and identifying grounding faults in the power supply system of an electric locomotive according to claim 1, characterized in that, In step S3, a ground fault location and identification model is constructed based on a machine learning algorithm. The model parameters of the ground fault location and identification model include: machine learning input layer parameters, hidden layer parameters, and optimized hidden layer input weights and node thresholds.

6. The method for locating and identifying grounding faults in the power supply system of an electric locomotive according to claim 1, characterized in that, In step S4, the real-time system signal data of the electric locomotive power supply system includes: the real-time AC side current of the electric locomotive power supply system. Real-time ground fault detection voltage and real-time DC side voltage Based on the real-time system signal data of the electric locomotive power supply system, extract the real-time feature vector of the real-time system signal data. , .

7. The method for locating and identifying grounding faults in the power supply system of an electric locomotive according to claim 1, characterized in that, In step S4, the method for determining whether a ground fault has occurred in the electric locomotive power supply system is as follows: like If the signal is positive, it indicates that the electric locomotive's power supply system is normal, and real-time system signal data will be collected again for a new round of real-time feature vector extraction; if... This indicates that a grounding fault has occurred in the power supply system of the electric locomotive.

8. The method for locating and identifying grounding faults in the power supply system of an electric locomotive according to claim 1, characterized in that, In step S5, for the real-time feature vector , Perform sliding window segmentation to obtain a cached data segment of length L. , ; Compute cache data segment average and variance And used as a cached data segment Characteristic indicators for calculating cached data segments average and variance And used as a cached data segment Characteristic indicators.

Images