AC / DC integrated grounding fault location method and system

Abstract

The present application relates to an integrated AC / DC grounding fault location method and system, comprising: collecting potential change data of a DC power supply system and frequency characteristic data of an AC power supply system to determine the electrical signal characteristics of stray current corrosion; decomposing the electrical signal using wavelet transform to extract DC and AC components; determining the fault type based on the component amplitudes; analyzing the potential gradient and stray current flow direction in combination with the line topology to determine the fault point location; measuring resistance to determine the grounding loop and analyzing the correlation between corrosion distribution and electrical parameter characteristics; evaluating the impact of corrosion on the grounding path in combination with material resistivity; sending pulse signals and analyzing the reflected waveform to determine the fault distance. The present invention can accurately identify faults, locate fault points, evaluate the degree of corrosion and its impact on the system, and provide effective support for the safe operation and maintenance of subway power supply systems.

Description

Technical Field

[0001] The present application relates to the technical field of measuring electrical variables, and in particular to an AC / DC integrated ground fault location method and system. Background Art

[0002] The subway traction power supply system uses direct current, while the auxiliary system relies on alternating current. This coexistence of AC and DC is particularly significant in complex electromagnetic environments. However, stray currents caused by ground faults can corrode metal pipes and equipment casings, threatening driving safety. Existing methods mostly rely on traditional electrical detection technologies, such as voltage monitoring or current distribution analysis, but it is difficult to accurately distinguish the source of the fault under strong electromagnetic interference. Especially in AC / DC hybrid systems, fault characteristics are easily masked by interference signals, resulting in insufficient positioning accuracy. Existing technologies find it difficult to effectively separate the fault characteristics of the traction network and signal system under strong electromagnetic interference environments, resulting in low fault diagnosis efficiency and reduced system reliability. Therefore, how to use the AC / DC characteristics of the subway power supply system in a strong electromagnetic interference environment to accurately distinguish ground faults between the traction network and signal system and accurately locate the fault point has become a key issue in improving system safety and operational stability. Summary of the Invention

[0003] In order to solve the problems existing in the above-mentioned prior art, the purpose of this application is to provide an AC / DC integrated grounding fault location method and system.

[0004] The present invention provides an AC / DC integrated ground fault location method, comprising the following steps:

[0005] Collect potential change data of the subway traction DC power supply system and frequency characteristic data of the AC power supply system to determine the electrical signal characteristics of stray current corrosion;

[0006] In a strong electromagnetic interference environment, the db4 wavelet basis function is used to perform a 4-layer wavelet decomposition on the electrical signal to extract the AC component eigenvalues in the high-frequency component and the DC component eigenvalues in the low-frequency component.

[0007] The amplitude of the DC component characteristic value is calculated. If it exceeds the preset DC amplitude threshold, a DC ground fault is determined. The amplitude, waveform distortion, and phase offset angle of the AC component characteristic value are calculated. If the amplitude exceeds the preset AC amplitude threshold and the waveform distortion is greater than 3% or the phase offset angle is greater than 30°, an AC ground fault is determined.

[0008] Based on the topological structure of the subway line, the finite element method is used to iteratively calculate the potential values of the grid nodes, generate the potential gradient distribution map, calculate the current density between the nodes through the current continuity equation, and determine the stray current flow vector based on the potential gradient direction;

[0009] Based on the potential gradient distribution and stray current flow direction vector, the physical location interval of the fault point is determined. If the physical location interval of the fault point is determined to exist, the following operations are performed: Measure the resistance between the pipeline or equipment casing and the preset grounding point. Combine the potential gradient distribution and stray current flow direction vector to extract corrosion distribution data, obtain material resistivity, analyze the current density distribution and potential gradient changes in the corrosion area, and evaluate the degree of local resistance increase, impedance discontinuity, and signal attenuation;

[0010] A narrow pulse signal with a pulse width of ≤1μs is sent to the traction network and signal system. The initial fault distance is calculated based on the propagation delay and amplitude attenuation data of the reflected waveform. The waveform interference offset is corrected based on the evaluation results of local resistance increase, impedance discontinuity and signal attenuation, and the calibrated fault point coordinates are output.

[0011] The present application discloses an AC / DC integrated ground fault location system, comprising:

[0012] The potential acquisition module is used to collect potential changes in the subway traction DC power supply system and frequency characteristics in the AC power supply system to determine the electrical signal characteristics of stray current corrosion;

[0013] The signal decomposition module is used to perform 4-layer wavelet decomposition on the electrical signal using the db4 wavelet basis function in a strong electromagnetic interference environment, extracting the AC component eigenvalues in the high-frequency component and the DC component eigenvalues in the low-frequency component;

[0014] A fault determination module is used to calculate the amplitude of the DC component characteristic value. If it exceeds a preset DC amplitude threshold, it is determined to be a DC ground fault. The amplitude, waveform distortion, and phase offset angle of the AC component characteristic value are calculated. If the amplitude exceeds the preset AC amplitude threshold and the waveform distortion is greater than 3% or the phase offset angle is greater than 30°, it is determined to be an AC ground fault.

[0015] The fault location module is used to iteratively calculate the potential values of grid nodes based on the finite element method according to the subway line topology, generate a potential gradient distribution map, calculate the current density between nodes through the current continuity equation, and determine the stray current flow direction vector based on the potential gradient direction;

[0016] The corrosion analysis module is used to determine the physical location interval of the fault point based on the potential gradient distribution map and the stray current flow direction vector. If the fault point physical location interval is determined to exist, the module performs the following operations: measuring the resistance between the pipeline or equipment casing and the preset grounding point, extracting corrosion distribution data based on the potential gradient distribution and the stray current flow direction vector, obtaining the material resistivity, analyzing the current density distribution and potential gradient changes in the corrosion area, and evaluating the degree of local resistance increase, impedance discontinuity, and signal attenuation;

[0017] The fault calibration module is used to send narrow pulse signals with a pulse width of ≤1μs to the traction network and signal system, calculate the initial fault distance based on the propagation delay and amplitude attenuation data of the reflected waveform, correct the waveform interference offset based on the evaluation results of local resistance increase, impedance discontinuity and signal attenuation, and output the coordinates of the calibrated fault point.

[0018] The advantages of the integrated AC / DC grounding fault location method and system described in the present application are that by collecting the potential and frequency characteristics of the subway traction power supply system, using wavelet transform to decompose the electrical signal, extracting the DC and AC components, judging the fault type based on the component amplitude, analyzing the potential gradient and stray current flow direction in combination with the line topology structure, determining the fault point location, determining the grounding loop by measuring resistance, analyzing the correlation between corrosion distribution and electrical parameter characteristics, evaluating the impact of corrosion on the grounding path in combination with material resistivity, sending pulse signals and analyzing the reflected waveform to determine the fault distance, the present invention can accurately identify stray current corrosion faults, locate the fault point, and assess the degree of corrosion and its impact on the system, providing effective support for the safe operation and maintenance of the subway power supply system. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 This is the process of an AC / DC integrated ground fault location method described in this application Figure 1 . DETAILED DESCRIPTION

[0020] like Figure 1 As shown, the AC / DC integrated ground fault location method described in this application includes the following steps:

[0021] like Figure 1 As shown, in step S101, potential change data is collected in the subway traction DC power supply system and frequency characteristic data is collected in the AC power supply system to determine the electrical signal characteristics caused by stray current corrosion, where the electrical signal characteristics include the DC potential offset amplitude and duration, and judge the corrosion occurrence interval.

[0022] Specifically, in step S101, a distributed data acquisition unit is used to collect raw DC potential data in real time from the track foundation and grounding grid nodes of the subway traction DC power supply system. The sampling interval is set to 100 milliseconds, and the potential fluctuation value is recorded according to the preset DC potential sampling frequency.

[0023] Set the acquisition duration and interval according to the substation monitoring requirements, perform potential sampling at the spatially distributed locations of different measurement points, and establish a DC potential data cache queue;

[0024] The original DC potential data is subjected to a 4-layer wavelet decomposition to remove high-frequency noise, and a median filter window with a length of 5 is used to eliminate sudden changes to obtain a smooth pre-processed DC potential data sequence. For example, the original potential of a measurement point is 850 mV, which is stabilized at 820 mV after processing.

[0025] In the AC power supply system, a signal analyzer is used to collect the voltage signal according to the AC frequency sampling period. A 1024-point fast Fourier transform is performed in the frequency range of 0 to 1000 Hz to generate the AC frequency characteristic spectrum. The signal is then smoothed using a Hanning window function to reduce spectrum leakage.

[0026] During substation operation, the frequency spectrum shows a fundamental frequency of 50 Hz and its harmonic components, with the third harmonic amplitude accounting for 15% of the fundamental and the fifth harmonic for 8%. Through time series correlation analysis, the preprocessed DC potential data sequence is matched with the AC frequency spectrum to extract the electrical signal characteristics.

[0027] The offset amplitude of the pre-processed DC potential data sequence is detected, with the positive threshold set to 600 mV and the negative threshold set to 1100 mV. When the potential offset exceeds the threshold and lasts for more than 300 seconds, the stray current calculation is triggered.

[0028] Using the sliding time window method, with the track loop resistance of 0.01 ohm as the benchmark, the stray current amplitude is obtained by dividing the potential difference by the resistance. For example, when the potential difference is 200 millivolts, the stray current is 20 amperes, and the amplitude sequence is recorded;

[0029] A three-layer BP neural network mapping model was further constructed. The input layer contained 12 nodes, including the stray current mean, standard deviation, crest factor, and AC harmonic content characteristics. The hidden layer had 16 nodes. The output layer was the potential anomaly degree judgment value. The sigmoid activation function was used to train the model.

[0030] In actual applications, the output value of the neural network is compared with the preset abnormal threshold of 0.85. If the judgment value exceeds the standard, it is confirmed that stray current corrosion has occurred. For example, in a certain test, the potential deviation lasted for 875 seconds and the maximum amplitude reached 350 millivolts. This occurred simultaneously with the increase of the AC third harmonic, indicating the presence of corrosion risk.

[0031] Record the potential offset amplitude and duration characteristics within the corrosion interval to provide data support for subsequent fault location. The embodiment of the present invention does not impose too many restrictions on the specific sampling frequency or filtering parameters, which can be adjusted by technicians according to actual scenarios.

[0032] Through distributed data acquisition and signal processing technology, the electrical signal characteristics in a mixed AC / DC environment can be effectively extracted. Neural network analysis further improves the accuracy of corrosion interval identification, laying the foundation for fault type judgment. Compared with traditional methods, this method shows higher robustness in strong interference environments, ensuring the reliability of fault diagnosis.

[0033] In practical applications, the corrosion interval is triggered when the potential abnormality judgment value output by the neural network exceeds the preset threshold.

[0034] In one embodiment, the stray current is calculated as follows: the potential difference value refers to the actual potential difference between the track and the grounding grid, and the resistance refers to a preset standard measurement resistance (0.01Ω).

[0035] In one embodiment, the corrosion interval is defined as a period during which the potential excursion lasts for more than 300 seconds.

[0036] like Figure 1 As shown, in step S102, wavelet transform technology is used to decompose the electrical signal characteristics in a strong electromagnetic interference environment to obtain high-frequency components and low-frequency components, from which the DC component of the DC power supply system and the AC component of the AC power supply system are extracted.

[0037] Specifically, in step S102, in a strong electromagnetic interference scenario, a high-precision data collector is used to sample the electrical signal in a time sequence, and an electromagnetic field strength sensor is used to monitor the environmental interference intensity in real time. The sampling frequency is adaptively adjusted according to the interference intensity to obtain the original sampled signal;

[0038] When the interference intensity exceeds 100 microtesla, the sampling frequency is automatically increased to 40 kHz to ensure the capture of high-frequency interference details;

[0039] The original sampling signal contains mixed AC / DC characteristics and noise interference. To improve signal quality, a bandpass filter is used with a passband range of 5 Hz to 5 kHz to filter out low-frequency power frequency interference and high-frequency random noise. At the same time, a pulse detection circuit is used to identify sudden interference points. For example, after detecting a pulse interference with an amplitude of 2 volts and a duration of 50 microseconds, a signal elimination algorithm is used to replace the interference segment with the mean of adjacent sampling points to generate a preprocessed sampling signal.

[0040] The preprocessed sampled signal is subjected to wavelet transform processing. The Daubechies 4 wavelet basis function is selected due to its superior time-frequency resolution ability. The signal is layered into multiple frequency band sub-signals through a four-layer decomposition. The first layer extracts the high-frequency sub-signal of 2 to 5 kHz, the second layer extracts the frequency band signal of 1 to 2 kHz, the third layer extracts the frequency band signal of 500 Hz to 1 kHz, and the fourth layer separates the frequency band signal of 250 to 500 Hz and the low-frequency sub-signal below 250 Hz.

[0041] To ensure the frequency band separation effect, calculate the energy proportion of each frequency band. For example, a measurement shows that the high-frequency sub-signal energy accounts for 35% and the low-frequency sub-signal accounts for 65%. Based on this frequency band separation, set the high-frequency threshold to 0.5 volt and the low-frequency threshold to 0.8 volt. Perform soft threshold processing on the high-frequency and low-frequency sub-signals respectively, retaining the effective components greater than the threshold and removing weak noise to obtain the high-frequency and low-frequency components after filtering;

[0042] The AC power supply system characteristics were extracted from the filtered high-frequency components. Spectral analysis was used to identify the fundamental frequency of 50 Hz and its harmonic components, such as an amplitude of 0.3 volts at 100 Hz, 0.2 volts at 150 Hz, and 0.1 volts at 200 Hz. The DC power supply system characteristics were extracted from the filtered low-frequency components. The detected DC voltage value was 750 volts with low-frequency fluctuations below 20 Hz.

[0043] The extracted eigenvalues are integrated into a 16-dimensional eigenvector, including 4 AC frequency parameters, 4 phase parameters, 4 amplitude parameters and 4 DC parameters;

[0044] A support vector machine is used to classify and train feature vectors using a radial basis kernel function. Under normal conditions, feature vectors are clustered in a high-dimensional space. When an abnormality occurs, they deviate from the clustering area. For example, when the DC voltage fluctuation exceeds 50 volts and the 150 Hz harmonic amplitude suddenly increases to 0.4 volts, it is judged as an abnormal state and the electrical signal feature discrimination result is output;

[0045] To meet the signal decomposition requirements in strong electromagnetic interference environments, the wavelet transform effectively separates high- and low-frequency components through multi-scale analysis. Compared with the traditional Fourier transform, it is more suitable for non-stationary signal processing. The combination of bandpass filtering and signal elimination algorithms ensures the purity of the pre-processed signal and provides a reliable foundation for subsequent feature extraction. The support vector machine classification and discrimination transforms the signal characteristics under complex interference into recognizable patterns through high-dimensional feature space mapping, thereby improving the discrimination accuracy.

[0046] In practical applications, if the electromagnetic interference intensity fluctuates frequently, the data collector can dynamically adjust the sampling interval according to the interference intensity. For example, when the interference intensity rises to 150 microtesla, the sampling frequency is further increased to 50 kHz;

[0047] During the wavelet decomposition process, the calculation of frequency band separation can introduce an adaptive threshold adjustment mechanism to optimize the high and low frequency threshold settings according to the real-time energy ratio to ensure the accuracy of the decomposition components;

[0048] The construction of feature vectors can be expanded to 20 dimensions, and new time-domain statistical features such as peak factor and skewness are added to further enhance classification robustness;

[0049] In some scenarios, if abnormally prominent harmonics are detected in the high-frequency component, such as an amplitude exceeding 0.4 volts at 150 Hz and lasting for more than 1 second, an AC component abnormality warning can be directly triggered;

[0050] If the DC voltage fluctuation amplitude in the low-frequency component exceeds 50V and is accompanied by low-frequency oscillation, it can be preliminarily determined that the DC component is abnormal;

[0051] A multi-class classification model can be introduced during support vector machine training to distinguish between normal state, DC abnormality, AC abnormality, and mixed abnormality, and output more detailed discrimination results to guide subsequent fault analysis;

[0052] Through adaptive sampling and multi-level signal processing, the AC and DC component features can be accurately extracted in complex electromagnetic environments. The application of support vector machines not only improves the discrimination efficiency, but also provides data support for the subsequent determination of fault types. In actual operation, this method can flexibly adjust parameters according to environmental changes to ensure applicability under different interference intensities.

[0053] In one embodiment, the electrical signal sampling object is to sample the track potential signal of the subway traction DC power supply system and the voltage signal of the AC power supply system.

[0054] In one embodiment, the support vector machine classifier uses a radial basis kernel function (RBF), a penalty factor C=1.0, a kernel parameter γ=0.01, and training data is 200 groups of historical fault samples (100 groups of normal states, 50 groups of DC faults, and 50 groups of AC faults).

[0055] like Figure 1 As shown, in step S103, the amplitudes of the DC component and the AC component are calculated respectively and compared with preset thresholds to determine whether there is a DC or AC grounding fault and to determine the fault type.

[0056] Specifically, in step S103, a high-precision data acquisition device is used to sample the voltage signal at the connection point. The sampling period is set to 100 microseconds. The original voltage signal is processed using an 8th-order Butterworth digital filter to eliminate high-frequency interference components above 10 kHz. The filtered signal is then subjected to spectrum analysis using a 1024-point fast Fourier transform to obtain the DC voltage amplitude and AC voltage amplitude, respectively, providing a data basis for subsequent fault determination.

[0057] In DC component processing, the DC voltage amplitude is smoothed using a 10-degree sliding window to reduce the impact of instantaneous fluctuations. Considering the drift characteristics of the measuring equipment under different ambient temperatures, an amplitude correction factor is introduced to compensate for errors. For example, the correction factor is set to 1.02 at 35 degrees Celsius.

[0058] The preset DC voltage threshold range is 1200V to 1800V. If the corrected DC voltage amplitude exceeds this range, for example, if it drops to 1150V in a certain measurement, the DC voltage threshold flag is recorded and the threshold threshold timestamp is marked.

[0059] This method ensures the stability and accuracy of DC amplitude calculation through dual processing of smoothing and compensation, and avoids misjudgment caused by noise or equipment errors;

[0060] In AC component analysis, the fundamental 50 Hz component is extracted through spectrum analysis and the AC voltage amplitude is calculated. The waveform distortion and phase shift angle are also calculated to characterize signal abnormalities.

[0061] During normal operation, the distortion is less than 3% and the phase shift angle is close to zero. The preset AC voltage threshold is 10 volts. If the measured amplitude exceeds this value, for example, reaching 12 volts and the phase shift angle exceeds 30 degrees, the AC voltage limit flag is recorded;

[0062] This process can effectively identify abnormal fluctuations in AC signals through multi-dimensional feature extraction, providing rich information for subsequent fault type determination;

[0063] The ground resistance measurement device is used to obtain the actual ground resistance value based on the four-wire method. The measurement current is set to 10 amperes. The resistance measurement compensation algorithm eliminates the influence of ambient temperature and contact resistance. For example, the normal resistance value should be less than 0.1 ohm. If the measurement rises to 0.3 ohm in a certain period, it indicates that the ground path is abnormal.

[0064] Perform a time-series correlation analysis on the DC voltage over-limit flag, AC voltage over-limit flag, and corrected ground resistance value to calculate the duration of the over-limit. For example, if both over-limits occur simultaneously and last for more than one second during a test, it indicates a significant fault characteristic.

[0065] In fault type identification, a Gaussian mixture clustering algorithm is used to cluster the feature vectors containing the duration. The feature vectors include the DC voltage over-limit flag, the AC voltage over-limit flag, and the over-limit duration. This generates three types of fault feature distributions, including normal state, minor abnormality, and severe fault.

[0066] A fault identification model was further established using the random forest algorithm. The model consists of 50 decision trees, and its input features include DC voltage deviation, AC voltage amplitude, waveform distortion, phase offset angle, and ground resistance.

[0067] Random forests use a multi-tree voting mechanism to map fault feature relationships. For example, during a certain operation, the DC voltage dropped to 1150 volts, the AC voltage rose to 12 volts, and the ground resistance was 0.3 ohms. The model output indicated a double ground fault, with the DC fault located on the negative bus and the AC fault involving the phase A winding.

[0068] The combination of a Butterworth filter and Fourier transform ensures accurate signal processing. The multi-step process of amplitude calculation and threshold determination improves the reliability of fault detection. Real-time measurement of ground resistance provides auxiliary evidence for fault location. The combined application of Gaussian mixture clustering and random forest algorithms enables accurate identification of complex fault types through feature clustering and classification prediction.

[0069] In practical applications, if the DC voltage amplitude fluctuates frequently, the sliding window length can be dynamically adjusted, for example, increasing it to 15 sampling points in a high-interference environment to enhance the smoothing effect;

[0070] For AC signals, if an abnormal increase in harmonic components is detected, additional features such as harmonic content ratio can be introduced to further refine the distortion analysis;

[0071] The fault discrimination model can also regularly update the decision tree weights based on operating data to adapt to the long-term changing characteristics of the subway power supply system, thereby improving the robustness of the discrimination.

[0072] Through comprehensive multi-parameter analysis, this method can quickly distinguish fault types in AC / DC hybrid power supply systems. Compared with traditional single threshold judgment, the random forest model significantly improves the discrimination accuracy by using multi-dimensional feature mapping, and is particularly suitable for complex ground fault scenarios.

[0073] In one embodiment, the waveform distortion and phase shift angle are calculated for the AC voltage signal, where:

[0074] The waveform distortion is calculated by extracting the fundamental and harmonic components through fast Fourier transform; the phase offset angle is obtained by comparing the reference waveform with the cross-correlation algorithm;

[0075] If the AC component amplitude exceeds the preset AC amplitude threshold, and the waveform distortion is greater than 3% or the phase offset angle is greater than 30°, an AC ground fault is determined. This clarifies that the calculation object is the AC voltage signal rather than the amplitude.

[0076] In one embodiment, the AC over-limit flag determination logic is that the determination of the AC over-limit flag must simultaneously meet the following conditions: AC component amplitude > 10V, and waveform distortion > 3% or phase shift > 30°.

[0077] In one embodiment, the resistance value example logic is that after eliminating the influence of ambient temperature through the compensation algorithm, if the ground resistance value is greater than 0.1Ω (for example, the measured value after compensation is 0.3Ω), then it is determined that the ground path is abnormal.

[0078] In one embodiment, a random forest fault discrimination model is used: the number of decision trees is 50, the feature splitting rule uses Gini impurity, and the input feature weight distribution is DC voltage deviation (weight 0.3), AC voltage amplitude (0.25), waveform distortion (0.2), phase offset angle (0.15), and ground resistance value (0.1).

[0079] like Figure 1 As shown, in step S104, by combining the fault type with the subway line topology, the potential gradient distribution and stray current flow characteristics are analyzed to determine the physical location interval of the fault point.

[0080] Specifically, in step S104, based on the subway line topology diagram, the line connection relationship and segment division information are extracted, and the line section is spatially modeled using a regular hexagonal grid division method. The side length of each grid is set to 0.5 meters, covering the entire length of the line. For example, a 25-kilometer line is divided into approximately 50,000 basic grid cells.

[0081] Combined with the three-dimensional coordinates of the grounding point, for example, a grounding point measured by the global positioning system is 116.123 degrees east longitude, 39.456 degrees north latitude, and 45 meters above sea level, to establish the initial section potential distribution data, laying the foundation for subsequent potential field analysis;

[0082] Based on the grid modeling, the initial section potential distribution data was refined, and triangular cells were used to refine the grid for complex sections of stations and tunnels. For example, the number of grid cells in a station area was increased from the basic 50,000 to 80,000.

[0083] The potential field distribution equation was constructed using the finite element method. Considering the material properties of the rail conductivity of 5.8×107 Siemens / m and the concrete conductivity of 1×10-6 Siemens / m, a convergence threshold of 0.001 volt and a maximum number of iterations of 1000 were set. The potential value of each grid node was iteratively calculated to generate the segment potential field distribution data.

[0084] For example, the potential of a rail node is measured to be 850 millivolts, while that of the adjacent concrete node is 50 millivolts, reflecting a significant potential difference.

[0085] Based on the segment potential field distribution data, the central difference method is used to calculate the potential gradient between adjacent grid nodes. For example, the potential gradient between rails and concrete can reach 1.6 V / m.

[0086] To extract spatial features, an 8-layer convolutional neural network was used to process the potential gradient distribution map. The input size was 120×120 pixels, and the convolution kernel size was 3×3. Multi-layer convolution and pooling operations were used to capture the local gradient variation pattern and generate a segment potential gradient distribution feature map.

[0087] Compared with traditional differential calculation, this method can more comprehensively characterize the spatial distribution characteristics of the potential field and provide high-dimensional feature support for fault location;

[0088] In the current analysis, the current continuity equation is used to calculate the current density between grid nodes. The current value is corrected based on the rail cross-sectional area of 7700 square millimeters and the material conductivity. For example, the current density at a certain location is calculated to be 0.12 amperes per square centimeter, pointing towards the concrete structure.

[0089] Spatial clustering of current flow direction vectors was performed to identify current inflow and outflow areas. For example, the current inflow in the station platform area accumulated 15 amperes, while the outflow in the adjacent track section was 12 amperes, revealing the unbalanced distribution of stray current.

[0090] The clustering results are used to quantify the integrated value of the current density in each region and generate the current inflow and outflow distribution characteristics of the structure.

[0091] The potential gradient distribution characteristics and current flow direction characteristics are compared with the pre-calibrated fault feature library, and the spatial coordinates of the fault point are predicted using a deep regression tree;

[0092] The training sample contains 50 sets of typical fault data, each of which includes potential gradient, current density, and flow direction vector. The prediction error is controlled within 1.5 meters.

[0093] For example, in one analysis, the fault point was predicted to be at K15+750 on Line 3. On-site verification showed that the potential of the concrete structure there was abnormally low, at 300 millivolts, consistent with the model output, indicating that the fault point was closely related to stray current leakage.

[0094] Spatial grid modeling transforms complex lines into computable units. The finite element method accurately describes the potential field distribution through numerical solution, and the convolutional neural network enhances the depth and breadth of feature extraction. Compared with traditional manual analysis, this method significantly improves positioning efficiency, especially in complex structural areas such as stations, and can effectively reveal the convergence patterns of stray currents.

[0095] In practical applications, if the line section involves the intersection of multiple materials, the grid density can be dynamically adjusted. For example, the grid edge length at the intersection of rails and concrete can be reduced to 0.2 meters.

[0096] Temperature correction factors can be introduced into current density calculations to adapt to environmental changes;

[0097] The deep regression tree model can also regularly update training samples to incorporate new failure modes, ensuring that prediction accuracy is optimized over time;

[0098] Through multi-level analysis, this method not only locates the fault point but also reveals the extent to which stray currents affect surrounding structures. Station areas, due to complex grounding conditions, often become current hotspots, requiring enhanced monitoring to reduce corrosion risks. In actual operation, the grid division and algorithm parameters can be flexibly adjusted according to line characteristics to improve the applicability of the method.

[0099] In practical applications, the method for obtaining the stray current flow direction characteristics is: calculate the current density between nodes through the current continuity equation, and determine the stray current flow direction vector based on the potential gradient direction.

[0100] In one embodiment, the potential distribution data is derived from initial segment potential distribution data, which is generated iteratively using the finite element method based on actual measurements of track potential sensors. In one embodiment, the initial segments can be defined based on natural segmentation within the subway line topology, based on the distance between stations.

[0101] In one embodiment, the finite element iteration parameters (convergence threshold 0.001V, maximum iteration 1000 times) are: based on the rail conductivity of 5.8×10 7 The convergence test results under S / m conditions show that the positioning error is less than 0.1% when the threshold is less than 0.001V;

[0102] Dynamic adjustment formula for high and low frequency thresholds:

[0103] High frequency threshold = 0.2 × (electromagnetic interference intensity / 100μT) + 0.3V;

[0104] Low frequency threshold = 0.3 × (band separation) + 0.5V.

[0105] like Figure 1 As shown, in step S105, if step S104 determines that there is a physical location interval of the fault point, step S105 is executed to determine the grounding loop path by measuring the resistance between the pipeline and the equipment casing and the preset grounding point, and extract corrosion distribution data in combination with the potential gradient distribution and the stray current flow direction characteristics, and analyze the correlation characteristics between the corrosion of the pipeline and the equipment casing and the fault point to reveal the fault influencing mechanism.

[0106] Specifically, in step S105, measurement electrodes are arranged based on the preset grounding point coordinates, a 1 ampere test current is applied using the four-terminal method, and the loop resistance is calculated by detecting the voltage drop between the pipeline and the grounding point using the voltage electrodes. For example, if the voltage drop at a certain point is 0.15 volts, the resistance value is 0.15 ohms.

[0107] In order to eliminate the influence of temperature, a temperature compensation coefficient is introduced. For example, when the resistance is measured at 35 degrees Celsius and is 0.156 ohms, it is multiplied by 0.96 to correct it to 0.15 ohms.

[0108] The loop resistance distribution data is constructed through multi-point measurement to reflect the electrical characteristics of the ground loop and provide basic parameters for subsequent corrosion analysis.

[0109] During the corrosion inspection, a 5 MHz ultrasonic flaw detector was used to perform a grid scan of the pipeline surface with a scanning interval of 50 mm, covering 200 measurement points. Multiple corrosion pits were found in a certain section of the pipeline, with depths ranging from 1 to 2.8 mm and diameters ranging from 15 to 40 mm.

[0110] At the same time, an infrared thermal imager was used to collect the surface temperature distribution of the equipment shell. It was detected that the temperature of the corrosion area was 0.8 to 1.2 degrees Celsius lower than the surrounding area. The area of the abnormal area was about 400 square centimeters.

[0111] These initial data were divided into spatial grids and cubic spline interpolation was used with an interpolation interval of 10 mm to generate a continuous corrosion depth distribution surface. Then, the least squares method was used to fit the elliptical corrosion boundary with a major axis of 80 cm and a minor axis of 50 cm to accurately characterize the spatial morphology of the corrosion.

[0112] Combined with the potential gradient distribution characteristics, a potential gradient threshold of 0.5 V / m was set to define the potential abnormal area. When the potential gradient exceeded 0.5 V / m, it was classified as an abnormal area. For example, the area of an abnormal area reached 600 square centimeters, and the overlap with the corrosion distribution data reached 75%, indicating that there was a significant spatial correspondence between corrosion and potential anomaly.

[0113] A deep neural network is used to extract the corrosion location feature vector. The input includes the corrosion pit depth, area, and temperature anomaly point coordinates. Multi-layer convolution processing is used to generate a high-dimensional feature representation for correlation analysis.

[0114] In the corrosion severity assessment, a three-level judgment standard is established: mild corrosion (average depth <1mm, area ratio <5%), moderate corrosion (average depth 1-2mm, area ratio 5-10%), and severe corrosion (average depth >2mm, area ratio >10%). In this case, the average corrosion depth is 1.5mm, the maximum depth is 2.8mm, and the corrosion area ratio is 12%, which is considered severe corrosion;

[0115] The corresponding electrical parameter characteristics include loop resistance of 0.15 ohms, potential difference of 350 millivolts, and current density of 0.08 amperes per square centimeter;

[0116] Support vector regression was used to construct a mapping model between corrosion location feature vectors and electrical parameter characteristics. A radial basis kernel function was used in the training process. The results showed that when the potential difference exceeded 300 millivolts and the current density was greater than 0.05 amperes per square centimeter, the corrosion depth growth rate accelerated significantly, revealing the driving role of electrical anomalies in corrosion.

[0117] Based on the coordinates of the center of the corrosion area, for example, at K15+800 on the subway line, 15 meters away from the fault point, and combined with the characteristics of the stray current flow, the ground loop path was analyzed. It was found that the stray current at the fault point was conducted to the pipe section through the concrete return channel, causing localized corrosion to worsen.

[0118] In one embodiment, support vector regression analysis was used to analyze the mapping relationship between quantitative indicators and electrical parameter characteristics, and a correlation coefficient of 0.85 was calculated, indicating a strong correlation between the degree of corrosion and changes in electrical parameters. Corrosion distribution data and fault point correlation characteristic data were generated, reflecting the dual characteristics of location proximity and coupling of electrical parameter characteristics, providing a basis for tracing the cause of the fault.

[0119] In actual applications, if the pipe material or ambient humidity changes, the ultrasonic probe frequency can be adjusted to 7 MHz to improve detection accuracy;

[0120] The support vector regression model can introduce more features, such as pipe material resistivity and humidity factor, to further optimize mapping accuracy;

[0121] Through long-term monitoring, it was found that under the continuous action of stray current, the boundary of the corrosion area may expand over time, and the measurement data needs to be updated regularly to track dynamic changes;

[0122] Through multi-dimensional detection methods and data fusion analysis, this method not only accurately locates the grounding loop, but also deeply reveals the causal relationship between corrosion and fault points. In actual operation, the aggravation of pipeline corrosion is often highly correlated with the stray current leakage path, and protection needs to be strengthened in combination with line maintenance strategies.

[0123] like Figure 1 As shown, in step S106, the material resistivity of the pipeline and equipment casing is obtained and combined with the correlation characteristics of corrosion and fault points, the current density distribution and potential gradient changes of stray current in different materials and corrosion areas are analyzed, the law of corrosion distribution data is derived and its impact on the electrical characteristics of the grounding path is evaluated, including local resistance increase, impedance discontinuity and increased signal attenuation.

[0124] Specifically, in step S106, the resistivity of the pipe and the shell material is measured using the quadrupole method. The electrode spacing is set to 10 cm. The initial resistivity is calculated by injecting a 1 ampere measuring current and detecting the voltage drop. For example, the carbon steel pipe is measured to be 1.7×10 -7 Ohm meter;

[0125] In view of the resistivity change caused by metal loss in the pitting area, a temperature compensation coefficient and a corrosion depth correction factor are introduced. For example, when the pitting depth is 2 mm, the local resistivity rises to 2.8×10 -7 Ohm-meter, with a correction factor of 1.65;

[0126] Obtain material conductivity data through multi-point measurement to reflect the impact of corrosion on conductivity and provide a basis for subsequent analysis;

[0127] In the corrosion area analysis, the area was divided into a 5 cm × 5 cm square grid, generating a total of 400 cells. The equivalent resistance of each cell was calculated based on the corrosion depth. For example, the resistance increased by about 20% at a depth of 2 mm;

[0128] The finite element method is used to solve the stray current distribution, and the boundary condition is set as the pipeline conductivity of 5.8×10 -7 Siemens / meter, iterative calculations show that the current density is concentrated near the pitting pit, with a maximum value of 0.15 amperes / square centimeter, which is three times higher than that of the uncorroded area;

[0129] This method is used to obtain the spatial distribution data of current density, revealing the flow characteristics of stray current in the corrosion area;

[0130] A deep neural network is used to process the spatial distribution data of current density. The network structure includes two convolutional layers and three fully connected layers. The input features are the grid cell coordinates and the grid cell equivalent resistance value, and the output is a continuous current density vector.

[0131] The back propagation algorithm is used to iterate 8000 times to optimize the parameters and generate a current density distribution function, which can better capture local abnormal changes than traditional interpolation methods;

[0132] At the same time, a bipolar potential gradient detector was used to scan along the corrosion boundary, with a probe spacing of 1 cm and data recorded every 5 cm. For example, the potential gradient in the severely pitted area reached 2 V / m, while that in the normal area was only 0.3 V / m. A potential gradient spatial distribution map was generated through 3D surface fitting, showing that the edge change rate exceeded 1 V / m / cm.

[0133] Impedance variation parameters are calculated based on the potential gradient spatial distribution map, including impedance mean, standard deviation, and maximum change rate. For example, the impedance mean in the corrosion area is 35% higher than that in the normal area, with a standard deviation of 0.08 ohms and a change rate of 50% / meter.

[0134] The impedance data spectrum was analyzed by 1024-point Fourier transform, and it was found that the high-frequency components were prominent, indicating that the impedance distribution was discontinuous;

[0135] A support vector machine (SVM) with a radial basis kernel function was used to establish a mapping model between impedance characteristics and signal attenuation. Training on 50 sets of corrosion samples revealed that when the local resistance rise index exceeded 1.5 and the impedance discontinuity index exceeded 2.0, the attenuation of a 1 kHz signal over 100 meters reached 45 dB, far exceeding the normal 25 dB. This quantified the impact of corrosion on electrical properties.

[0136] In practical applications, if the pipe materials are diverse, the grid size can be adjusted according to the material characteristics, for example, the stainless steel area can be reduced to 3 cm to improve the resolution;

[0137] Potential gradient detection can introduce a dynamic scanning path to encrypt measurement points along the high current density area;

[0138] The support vector machine model can also regularly update the sample library to incorporate new corrosion patterns, ensuring that the assessment accuracy is optimized as operating conditions change;

[0139] Combining material resistivity and current density distribution analysis, we derived patterns in the corrosion distribution data and found that current concentration around pits significantly increases resistance, leading to impedance discontinuity and affecting ground loop stability.

[0140] A comprehensive assessment revealed that this change not only increases local resistance but also exacerbates high-frequency signal reflection and attenuation, necessitating targeted anti-corrosion measures to safeguard electrical performance.

[0141] In practical applications, a 20% increase in resistance is calculated relative to the conductivity of the uncorroded area.

[0142] In one embodiment, the corrosion resistance increase benchmark is a 20% resistance increase, which means that the resistance value of the corroded area increases by 20% relative to the benchmark resistance value of the uncorroded area of the same material. This benchmark helps analyze the pattern of corrosion distribution data.

[0143] In one embodiment, the corrosion distribution data shows a pattern: when the pitting depth is greater than 1 mm, the local resistivity increases by greater than 20%, the current density is concentrated at the corrosion edge, and the impedance discontinuity index is greater than 2.0. The local resistivity increase of greater than 20% is relative to the corrosion resistance increase benchmark.

[0144] like Figure 1 As shown, in step S107, a pulse signal is sent to the traction network and signal system, and the fault distance is calculated by analyzing the propagation time and amplitude attenuation data of the reflected waveform. At the same time, based on the evaluation results of the impact of corrosion and pitting on the electrical characteristics of the grounding path, the waveform interference offset is corrected and the fault point position is calibrated.

[0145] Specifically, in step S107, a pulse signal generator is used to send a microsecond narrow pulse test signal to the traction network and signal line, for example, the pulse width is set to 1 microsecond and the amplitude is 5 volts, and the reflected waveform is captured by a high-precision oscilloscope;

[0146] In order to remove the influence of environmental noise, a bandpass filter is used to set the passband range from 10 Hz to 10 kHz to filter out high-frequency interference and generate initial reflection waveform data;

[0147] In the noise reduction process, an adaptive threshold method is applied to the initial reflection waveform data, and the threshold is dynamically adjusted according to the real-time signal-to-noise ratio. For example, when the signal-to-noise ratio is lower than 10 decibels, the threshold is set to 1.5 times the signal mean;

[0148] Combined with a median filter, the window length is set to 5 sampling points based on the noise characteristics to smooth out pulse interference, such as eliminating noise peaks with amplitude changes exceeding 2 volts, and generate a noise-reduced reflected waveform;

[0149] This method effectively preserves the main features of the waveform while reducing the impact of interference on positioning accuracy;

[0150] A transmission medium loss model was constructed based on corrosion depth and pitting distribution data, defining an attenuation coefficient of 0.02 decibels per meter, a dispersion coefficient of 0.01 microseconds per meter, and a reflection coefficient of 0.3. A three-layer neural network was used to calculate waveform attenuation compensation. The input layer included corrosion depth and fault point distance parameters, where the distance parameter refers to the straight-line distance from the corrosion location to the measurement starting point. The hidden layer had 10 nodes, and the output layer was the corrected propagation delay.

[0151] For example, when the corrosion depth of a certain path is 2 mm, the propagation delay increases by 0.05 microseconds. After compensation, it is corrected to the true value, improving the distance calculation accuracy;

[0152] Perform wavelet decomposition on the reflected waveform after noise reduction, select the db4 wavelet basis function to perform 4-layer decomposition, and extract the interference offset characteristics;

[0153] Identify waveform distortion components through multi-scale analysis. For example, the first-layer high-frequency components reveal residual noise, and the fourth-layer low-frequency components reflect changes in the propagation path. Construct a time-frequency feature vector containing time-domain and frequency-domain features.

[0154] An adaptive compensation algorithm is then used to calculate the phase offset angle and amplitude attenuation correction parameters based on the time-frequency eigenvector, such as a phase correction of 0.1 radian and an amplitude gain of 1.2 times. This generates a compensated reflection waveform, ensuring that the waveform accurately reflects the characteristics of the fault point.

[0155] In distance calculation, a cross-correlation algorithm is used to analyze the time offset between the reflected waveform and the transmitted signal after compensation. For example, when the peak correlation coefficient is 0.9, the propagation delay is 50 microseconds;

[0156] Combined with the medium transmission speed, for example, 3×10^8 m / s in rails, the initial fault distance is calculated to be 15 m;

[0157] For further calibration, a positioning error compensation model is constructed based on support vector regression. The input features include waveform attenuation rate, distortion degree and propagation delay, and the output is the position calibration value.

[0158] The model was trained with 500 sets of samples and iteratively optimized to a mean square error of less than 0.01 meters. For example, if the calibration distance was 0.3 meters, the coordinates of the fault point were ultimately determined to be 15.3 meters.

[0159] In practical applications, if the corrosion area is complex, the number of wavelet decomposition layers can be increased to 6 to capture more subtle distortions;

[0160] The support vector regression model can introduce environmental temperature and humidity characteristics to further optimize error compensation. Through this method, positioning accuracy can be controlled within 1 meter, significantly improving the efficiency of troubleshooting in complex environments. Combined with multi-level signal processing and machine learning, it not only accurately calculates the fault distance but also effectively corrects interference offsets through corrosion impact assessment. In actual operation, this method can dynamically adjust parameters according to the line material and environment to ensure the reliability and practicality of the positioning results.

[0161] In one embodiment, data on potential changes between the track and the grounding grid in a subway traction DC power supply system, as well as frequency characteristic data on the voltage signal in the AC power supply system, are collected to determine whether the fault point is within the physical location interval output in step S104. A pulse signal is then sent to the traction network and signal line. By receiving the reflected waveform from the fault point, propagation time and amplitude attenuation data are extracted to calculate the fault distance. The potential change data refers to the actual potential fluctuation data collected between the track foundation and the grounding grid nodes. Frequency characteristic data refers to data characterizing the spectral characteristics of the voltage signal in the AC power supply system. The reflected waveform is the waveform formed when a portion of the pulse signal is reflected back from impedance discontinuities (e.g., the fault point) in the path of a pulse signal sent to the subway traction network. Analyzing the reflected waveform can help locate the fault point.

[0162] In one embodiment, the pitting distribution data is generated by fusing the pitting distribution data with ultrasonic grid scanning and infrared thermal imaging.

[0163] In one embodiment, neural network waveform compensation: a 3-layer fully connected network (input layer 3 nodes: corrosion depth, distance, temperature; hidden layer 10 nodes; output layer 1 node: delay compensation value), the optimizer is Adam, and the loss function is mean square error.

[0164] The present invention provides an AC / DC integrated ground fault location system, which mainly includes:

[0165] The potential acquisition module is used to collect potential changes in the subway traction DC power supply system and frequency characteristics in the AC power supply system to determine the electrical signal characteristics of stray current corrosion;

[0166] The signal decomposition module is used to perform 4-layer wavelet decomposition on the electrical signal using the db4 wavelet basis function in a strong electromagnetic interference environment, extracting the AC component eigenvalues in the high-frequency component and the DC component eigenvalues in the low-frequency component;

[0167] A fault determination module is used to calculate the amplitude of the DC component characteristic value. If it exceeds a preset DC amplitude threshold, it is determined to be a DC ground fault. The amplitude, waveform distortion, and phase offset angle of the AC component characteristic value are calculated. If the amplitude exceeds the preset AC amplitude threshold and the waveform distortion is greater than 3% or the phase offset angle is greater than 30°, it is determined to be an AC ground fault.

[0168] The fault location module is used to iteratively calculate the potential values of grid nodes based on the finite element method according to the subway line topology, generate a potential gradient distribution map, calculate the current density between nodes through the current continuity equation, and determine the stray current flow direction vector based on the potential gradient direction;

[0169] The corrosion analysis module is used to determine the physical location interval of the fault point based on the potential gradient distribution map and the stray current flow direction vector. If the fault point physical location interval is determined to exist, the module performs the following operations: measuring the resistance between the pipeline or equipment casing and the preset grounding point, extracting corrosion distribution data based on the potential gradient distribution and the stray current flow direction vector, obtaining the material resistivity, analyzing the current density distribution and potential gradient changes in the corrosion area, and evaluating the degree of local resistance increase, impedance discontinuity, and signal attenuation;

[0170] The fault calibration module is used to send narrow pulse signals with a pulse width of ≤1μs to the traction network and signal system, calculate the initial fault distance based on the propagation delay and amplitude attenuation data of the reflected waveform, correct the waveform interference offset based on the evaluation results of local resistance increase, impedance discontinuity and signal attenuation, and output the coordinates of the calibrated fault point.

[0171] Those skilled in the art can make various other corresponding changes and deformations based on the technical solutions and concepts described above, and all of these changes and deformations should fall within the scope of protection of the claims of this application.

Claims

1. An AC / DC integrated ground fault location method, characterized in that: The following steps are involved: Collect potential change data of the subway traction DC power supply system and frequency characteristic data of the AC power supply system to determine the electrical signal characteristics of stray current corrosion; In a strong electromagnetic interference environment, the db4 wavelet basis function is used to perform a 4-layer wavelet decomposition on the electrical signal to extract the AC component eigenvalues in the high-frequency component and the DC component eigenvalues in the low-frequency component. The amplitude of the DC component characteristic value is calculated. If it exceeds the preset DC amplitude threshold, a DC ground fault is determined. The amplitude, waveform distortion, and phase offset angle of the AC component characteristic value are calculated. If the amplitude exceeds the preset AC amplitude threshold and the waveform distortion is greater than 3% or the phase offset angle is greater than 30°, an AC ground fault is determined. Based on the subway line topology, the line connection relationship and section division information are extracted. The line section is spatially modeled using a regular hexagonal grid division method. The grid node potential values are iteratively calculated using the finite element method to generate a potential gradient distribution map. The current density between nodes is calculated using the current continuity equation, and the stray current flow vector is determined based on the potential gradient direction. Based on the potential gradient distribution and stray current flow direction vector, the physical location interval of the fault point is determined. If the physical location interval of the fault point is determined to exist, the following operations are performed: Measure the resistance between the pipeline or equipment casing and the preset grounding point. Combine the potential gradient distribution and stray current flow direction vector to extract corrosion distribution data, obtain material resistivity, analyze the current density distribution and potential gradient changes in the corrosion area, and evaluate the degree of local resistance increase, impedance discontinuity, and signal attenuation; The potential change data between the track and the grounding grid in the subway traction DC power supply system and the frequency characteristic data of the voltage signal in the AC power supply system are collected to determine whether the fault point is located within the physical location range. A narrow pulse signal with a pulse width of ≤1μs is sent to the traction network and signal system. The initial fault distance is calculated based on the propagation delay and amplitude attenuation data of the reflected waveform. The waveform interference offset is corrected based on the evaluation results of local resistance increase, impedance discontinuity and signal attenuation, and the coordinates of the calibrated fault point are output.

2. The AC / DC integrated ground fault location method according to claim 1, characterized in that: The extracting of the AC component characteristic value in the high-frequency component and the DC component characteristic value in the low-frequency component includes: The frequency band of the high-frequency component is 250Hz-5kHz, which is used to extract the characteristic value of the AC component; The frequency band of the low-frequency component is below 250 Hz and is used to extract the DC component characteristic value.

3. The AC / DC integrated ground fault location method according to claim 1, characterized in that: The waveform distortion is calculated by extracting the fundamental wave and harmonic components through fast Fourier transform, and the phase shift angle is obtained by comparing the reference waveform through a cross-correlation algorithm.

4. The AC / DC integrated ground fault location method according to claim 1, characterized in that: The convergence threshold of the iterative calculation of the grid node potential value is 0.001V, the maximum number of iterations is 1000, and the spatial clustering of the stray current flow direction vector uses the K-means algorithm to identify the current inflow and outflow areas.

5. The AC / DC integrated ground fault location method according to claim 1, characterized in that: Waveform interference offset correction includes: performing db4 wavelet 4-layer decomposition on the reflected waveform to extract the distortion component, and correcting the phase and amplitude through an adaptive compensation algorithm.

6. An AC / DC integrated ground fault location system, used to execute the method according to any one of claims 1 to 5, characterized in that: include: The potential acquisition module is used to collect potential changes in the subway traction DC power supply system and frequency characteristics in the AC power supply system to determine the electrical signal characteristics of stray current corrosion; The signal decomposition module is used to perform 4-layer wavelet decomposition on the electrical signal using the db4 wavelet basis function in a strong electromagnetic interference environment, extracting the AC component eigenvalues in the high-frequency component and the DC component eigenvalues in the low-frequency component; A fault determination module is used to calculate the amplitude of the DC component characteristic value. If it exceeds a preset DC amplitude threshold, it is determined to be a DC ground fault. The amplitude, waveform distortion, and phase offset angle of the AC component characteristic value are calculated. If the amplitude exceeds the preset AC amplitude threshold and the waveform distortion is greater than 3% or the phase offset angle is greater than 30°, it is determined to be an AC ground fault. The fault location module is used to iteratively calculate the potential values of grid nodes based on the finite element method according to the subway line topology, generate a potential gradient distribution map, calculate the current density between nodes through the current continuity equation, and determine the stray current flow direction vector based on the potential gradient direction; The corrosion analysis module is used to determine the physical location interval of the fault point based on the potential gradient distribution map and the stray current flow direction vector. If the fault point physical location interval is determined to exist, the module performs the following operations: measuring the resistance between the pipeline or equipment casing and the preset grounding point, extracting corrosion distribution data based on the potential gradient distribution and the stray current flow direction vector, obtaining the material resistivity, analyzing the current density distribution and potential gradient changes in the corrosion area, and evaluating the degree of local resistance increase, impedance discontinuity, and signal attenuation; The fault calibration module is used to send narrow pulse signals with a pulse width of ≤1μs to the traction network and signal system, calculate the initial fault distance based on the propagation delay and amplitude attenuation data of the reflected waveform, correct the waveform interference offset based on the evaluation results of local resistance increase, impedance discontinuity and signal attenuation, and output the coordinates of the calibrated fault point.

Images