Intelligent monitoring method for insulation state of railway signal equipment

Abstract

The application relates to an intelligent monitoring method for an insulation state of a railway signal device, and comprises the following steps: S1, a multi-parameter sensing unit is arranged and initialized at an insulation monitoring point; S2, multi-dimensional insulation state data is collected and preprocessed in real time; S3, an AI algorithm is used for intelligent diagnosis and hierarchical evaluation of the insulation state; S4, a fault point is accurately positioned and visually displayed; S5, hierarchical early warning and operation and maintenance instruction pushing are performed; and S6, data archiving and model iteration optimization are performed. The method disclosed by the application adopts multi-dimensional parameter fusion collection and AI intelligent diagnosis, combines multiple parameters such as insulation resistance, leakage current, temperature and humidity and partial discharge, eliminates the limitations of single parameter monitoring, intelligently distinguishes normal fluctuation from fault deterioration through an algorithm model, and greatly reduces the misjudgment rate through an environmental compensation algorithm.

Description

Technical Field

[0001] This application relates to the field of railway signaling equipment operation and maintenance monitoring technology, specifically to an intelligent monitoring method for the insulation status of railway signaling equipment. Technical Field

[0002] Railway signaling systems are core equipment for ensuring railway traffic safety and improving transportation efficiency. Their insulation performance directly determines whether signaling equipment can function properly and is a key indicator for preventing signal malfunctions and traffic accidents. Insulation faults in signaling equipment mainly include decreased insulation between core wires, damaged insulation to ground, aging of insulation joints, and moisture contamination. These faults can easily lead to serious problems such as red light bands on track circuits, abnormal signal displays, and switch machine malfunctions, directly interfering with normal railway operations and even threatening traffic safety.

[0003] Currently, insulation status monitoring of railway signaling equipment mainly relies on traditional manual inspections and offline testing methods, which have several technical shortcomings: First, conventional monitoring requires power outages, shutdowns, and load removal before point-by-point testing with a megohmmeter, making online real-time monitoring impossible. This interrupts normal operation of the signaling equipment, impacting transportation efficiency and making it difficult to capture dynamic insulation degradation trends. Second, the testing coverage is limited. For the massive number of signal cable cores and distributed insulation joints within stations, manual point-by-point testing is labor-intensive and time-consuming, easily leading to missed or false detections. Third, existing monitoring methods can only obtain single insulation resistance values, failing to simultaneously collect related parameters such as temperature, humidity, leakage current, and partial discharge. This lack of multi-dimensional data support means fault diagnosis relies heavily on human experience, resulting in a high rate of misjudgment. Fourth, it cannot predict insulation degradation trends or accurately locate fault points. After a fault occurs, only insulation abnormalities are known, making it difficult to quickly pinpoint the fault location and type, leading to low maintenance and repair efficiency and prolonged fault handling time.

[0004] In existing technologies, some online insulation monitoring solutions only achieve single-parameter acquisition and simple threshold judgment, without being specifically optimized for the operating scenarios of railway signaling equipment. They have weak anti-interference capabilities, are easily affected by on-site electromagnetic interference such as track circuit signals and traction current, and suffer from insufficient accuracy of monitoring data. Furthermore, lacking intelligent algorithms, they cannot distinguish between normal fluctuations and fault degradation, resulting in delayed early warnings and failing to meet the high safety and reliability requirements of railway signaling equipment operation and maintenance. Therefore, developing a method for monitoring the insulation status of railway signaling equipment that requires no power outage, is online and real-time, integrates multiple parameters, provides intelligent diagnosis, and accurately locates the problem is a pressing technical issue that needs to be addressed in the current railway signaling operation and maintenance field. Summary of the Invention

[0005] To address the aforementioned shortcomings in the existing technology, this invention provides an intelligent monitoring method for the insulation status of railway signaling equipment, comprising: An intelligent monitoring method for the insulation status of railway signaling equipment, improved in that the intelligent monitoring method is based on the collaborative implementation of an edge computing terminal, a distributed multi-parameter sensing unit, and a cloud-based intelligent analysis platform, and includes the following steps: Step S1: Set and initialize the multi-parameter sensing unit at the insulation monitoring point; The insulation monitoring points include: signal distribution panel, track circuit insulation joint, signal power input terminal and switch machine control circuit; Each of the multi-parameter sensing units integrates an insulation resistance acquisition module, a leakage current detection module, a temperature and humidity acquisition module, a partial discharge detection module, and an electromagnetic interference suppression module; Step S2 involves real-time acquisition and preprocessing of multi-dimensional insulation status data, including: edge computing terminal control sensing unit acquiring raw data on insulation resistance, leakage current, temperature and humidity, and partial discharge; completing outlier removal, noise filtering, and timestamp synchronization preprocessing to form a standardized dataset for uploading to the cloud platform; Step S3: Intelligent diagnosis and grading assessment of insulation status using AI algorithms: The cloud platform analyzes the dataset through a pre-trained convolutional neural network and long short-term memory network fusion diagnostic model, judges the insulation status, identifies fault types, and predicts the deterioration trend. Combined with environmental parameter compensation and correction, the insulation status level is classified. Step S4: Accurate fault location and visualization. If the diagnosis is deterioration or fault, the fault location is accurately located based on the point distribution and parameter gradient, and a visualization interface and fault report are generated. Step S5: Tiered early warning and operation and maintenance instruction push. Based on the insulation fault level, a differentiated early warning mechanism is triggered to push early warning information and operation and maintenance instructions to the corresponding terminals. Step S6: Data archiving and model iterative optimization.

[0006] Preferably, step S1, initializing the multi-parameter sensing unit, includes: Step S1-1, Address Allocation and Point Binding: According to the preset monitoring point numbering rules, a unique device address is assigned to each multi-parameter sensing unit, and a one-to-one correspondence is formed with the corresponding railway signal equipment number, monitoring location, and loop number. Step S1-2, Acquisition Channel Calibration and Zero Point Correction: Perform no-load calibration and zero point correction on the acquisition channels for insulation resistance, leakage current, and partial discharge.

[0007] Preferably, step S2, real-time acquisition and preprocessing of multi-dimensional insulation status data, includes: Step S2-1: The edge computing terminal issues a data acquisition command to control the sensing units at each point to synchronously acquire insulation status data; the acquisition parameters include: insulation resistance between signal core wires, insulation resistance between core wires and ground, circuit leakage current, ambient temperature and humidity, and partial discharge pulse signal. In step S2-2, the sensing unit transmits the raw collected data to the edge computing terminal. The edge computing terminal preprocesses the data, including outlier removal, noise filtering, data normalization, and timestamp synchronization, removing invalid data caused by electromagnetic interference and environmental fluctuations, and forming a standardized insulation state dataset.

[0008] Preferably, step S3, which uses AI algorithms for intelligent diagnosis and grading assessment of insulation status, includes: Step S3-1: Perform multi-source feature extraction and standardization preprocessing; Step S3-2: Construct and train the hybrid diagnostic model architecture; Step S3-3: Determine the insulation status classification; Step S3-4, Fault type identification; Step S3-5: Model inference and iterative optimization.

[0009] Preferably, step S4: the fault location algorithm includes: Step S4-1, Location of the faulty equipment and circuit: The cloud platform retrieves the pre-entered railway signal equipment insulation monitoring topology database, and through the sensing unit ID corresponding to the abnormal data, it locks the type of equipment and power supply-control circuit to which the fault belongs, completes the preliminary division of the fault area, and eliminates circuits without abnormalities; Step S4-2: For insulation faults in long-distance signal cables along railway lines, the gradient difference algorithm of adjacent sensing node parameters is used to accurately locate the fault section.

[0010] Preferably, step S3-1, the multi-source feature extraction and standardization preprocessing method is as follows: The standardized dataset received from the cloud platform is processed to construct a multi-dimensional feature vector, which serves as the input data for the algorithm. The feature dimensions include: Static feature acquisition: Obtain a real-time 7-dimensional static feature vector for a single location; Dynamic feature acquisition: acquire insulation resistance change rate ΔR over 5 minutes, leakage current fluctuation coefficient ΔI, temperature and humidity coupling coefficient K, discharge signal distortion rate D, and equipment running time L; The above features are normalized using a min-max method, as shown in the following equation: In the formula, Here, x represents the normalized eigenvalues, and x represents the original eigenvalues. min x maxThese are the minimum and maximum values ​​of the historical samples for this feature, respectively.

[0011] Preferably, step S3-2, constructing and training the hybrid diagnostic model architecture, includes the following steps: Step S3-2-1, establish a CNN static feature extraction layer: adopt a lightweight CNN structure with 4 convolutional layers + 2 pooling layers, with a convolutional kernel size of 3×3 and the ReLU activation function to extract the latent features of insulation faults, output a static fault feature matrix, and identify steady-state fault types such as dampness, damage, and aging. Step S3-2-2, establish an LSTM time series trend prediction layer: a bidirectional LSTM network is used with 64 hidden layer nodes and an input sequence length of 20. This process handles the time series variation data of insulation parameters, learns the gradual pattern of insulation degradation, predicts the trend of insulation parameter changes over the next 10 periods, and calculates the degradation rate v, as shown in the following formula: ; In the formula: The insulation resistance at the current moment. To predict the insulation resistance at a given time, Where n is the acquisition period and n is the prediction period number; Step S3-2-3, Establish an expert rule correction layer: Based on the railway signaling equipment insulation operation and maintenance specifications and fault tree analysis logic, construct an expert rule base, and perform weighted fusion and rule verification on the output results of CNN and LSTM. The weight allocation adopts adaptive dynamic weights, with static feature weights set as ω1 and temporal feature weights as ω2, satisfying ω1+ω2=1; at the same time, normal parameter fluctuations caused by environmental temperature and humidity are removed, and the correction formula is as follows: R true =R meas ×α(T)×β(H); In the formula, R true To correct for the true insulation resistance, R meas These are measured values, α(T) is the temperature compensation coefficient, and β(H) is the humidity compensation coefficient.

[0012] Preferably, step S4-2 includes: Step S4-2-1: Retrieve real-time monitoring data from all distributed sensing units along the fault circuit, select the core judgment parameters: ground insulation resistance R2, leakage current I, and calculate the parameter gradient difference between adjacent sensing nodes, as shown in the following formula: ; ; In the formula: and These are the ground insulation resistance values ​​of the nth and n+1th sensing units, respectively. and These are the leakage current values ​​of the nth and n+1th sensing units, respectively. Step S4-2-2, Fault Zone Locking: Set a gradient difference judgment threshold; when adjacent nodes... , When the faulty section is identified, the amplitude and pulse frequency of the partial discharge signal are compared with those of each node to further verify the faulty section, eliminate the interference of abnormal data from single-point sensing units, and reduce the positioning accuracy to the distance between adjacent sensing units with an error of no more than 5m.

[0013] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. Enables real-time online monitoring without power outages, load removal, or disruption of railway signal equipment operation. This completely solves the problem of traditional offline monitoring affecting transportation efficiency, achieving 24-hour uninterrupted insulation status monitoring and real-time capture of insulation degradation dynamics.

[0014] 2. Employing multi-dimensional parameter fusion acquisition and AI intelligent diagnosis, combining multiple parameters such as insulation resistance, leakage current, temperature and humidity, and partial discharge, it eliminates the limitations of single parameter monitoring. Through algorithm models, it intelligently distinguishes between normal fluctuations and fault deterioration, and combined with environmental compensation algorithms, it significantly reduces the false judgment rate.

[0015] 3. It has the functions of accurate fault location and degradation trend prediction, which can quickly locate the fault point and predict the insulation degradation trend in advance, transforming post-event repair into pre-event prevention and in-event control, greatly shortening the fault handling time, improving operation and maintenance efficiency, and reducing operation and maintenance costs.

[0016] 4. It has strong anti-interference capabilities and is suitable for the complex electromagnetic environment and harsh weather conditions at railway sites. The hardware adopts an isolated design and anti-interference module, and the algorithm incorporates interference filtering logic to ensure stable and reliable monitoring data in complex scenarios.

[0017] 5. It enables cloud-edge collaborative management and control, with traceable and iteratively optimized data. The hierarchical early warning mechanism aligns with railway operation and maintenance processes. At the same time, the model has self-learning capabilities, continuously adapting to the monitoring needs of different lines and equipment. It is highly versatile and can be widely applied to insulation monitoring scenarios of various signal equipment in conventional and high-speed railways. Attached Figure Description

[0018] Figure 1 This application relates to a flowchart of a method for intelligent monitoring of the insulation status of railway signaling equipment. Detailed Implementation

[0019] To better understand the present invention, the following description, in conjunction with the accompanying drawings and examples, will further illustrate the content of the present invention.

[0020] An intelligent monitoring method for the insulation status of railway signaling equipment is improved in that the monitoring method is based on the collaborative implementation of an edge computing terminal, a distributed multi-parameter sensing unit, and a cloud-based intelligent analysis platform, such as... Figure 1 As shown, it includes the following steps: Step S1: Set up and initialize multi-parameter sensing units at insulation monitoring points; set up multi-parameter sensing units at key insulation monitoring points such as signal distribution panels, track circuit insulation joints, signal machines, and switch machines to integrate insulation resistance acquisition, leakage current detection, temperature and humidity acquisition, partial discharge detection, and electromagnetic interference detection, establish a full-coverage insulation monitoring and sensing network, and realize live online monitoring.

[0021] Specifically, step S1: Set up and initialize the multi-parameter sensing unit at the monitoring point: Based on the layout of railway signaling equipment, multi-parameter sensing units are distributed and installed at key insulation monitoring points such as signal distribution panels, both ends of track circuit insulation joints, signal power input terminals, and switch machine control circuits within stations. Each sensing unit integrates an insulation resistance acquisition module, a leakage current detection module, a temperature and humidity acquisition module, a partial discharge detection module, and an electromagnetic interference suppression module. Hardware initialization and point numbering and address binding are completed to establish a comprehensive, blind-spot-free insulation monitoring and sensing network, ensuring that the insulation status of all core signaling equipment is included in the monitoring scope.

[0022] The sensing unit adopts an isolated power supply design, which eliminates the need to disconnect the original power supply circuit of the signal equipment, enabling live online monitoring and avoiding impact on the normal operation of the signal equipment. At the same time, an electromagnetic interference suppression module is used to filter out electromagnetic interference such as traction current and track circuit signals on site, ensuring the stability and accuracy of the collected data. The acquisition frequency is set to be dynamically adjustable. During normal operation, periodic low-frequency acquisition is used, and when the insulation parameters approach the threshold, it automatically switches to high-frequency continuous acquisition to accurately capture the details of insulation status changes.

[0023] The initialization of the multi-parameter sensing unit includes: Step S1-1, Address Allocation and Point Binding: According to the preset monitoring point numbering rules, a unique device address is assigned to each multi-parameter sensing unit, and it is bound to the corresponding railway signal equipment number, monitoring location, and loop number to form a one-to-one correspondence between "address - device - location" to ensure that subsequent data can be accurately traced and faults can be accurately located.

[0024] Step S1-2, Acquisition Channel Calibration and Zero-Point Correction: Perform no-load calibration and zero-point correction on the acquisition channels for insulation resistance, leakage current, partial discharge, etc., to eliminate initial errors caused by hardware drift and environmental interference, and ensure that the acquired data is true and reliable.

[0025] Step S2 involves real-time acquisition and preprocessing of multi-dimensional insulation status data, including: edge computing terminal control sensing unit acquiring raw data on insulation resistance, leakage current, temperature and humidity, and partial discharge; completing outlier removal, noise filtering, and timestamp synchronization preprocessing; and forming a standardized dataset which is then uploaded to the cloud platform.

[0026] Specifically, step S2, real-time acquisition and preprocessing of multi-dimensional insulation status data, includes: Step S2-1: The edge computing terminal issues a data acquisition command to control the sensing units at each location to synchronously acquire insulation status data. The acquired parameters include: insulation resistance between signal core wires, insulation resistance between core wires and ground, circuit leakage current, ambient temperature and humidity, and partial discharge pulse signal.

[0027] The edge computing terminal includes embedded computing hardware devices with functions such as data acquisition, protocol parsing, real-time computing, local caching, edge inference, anti-interference communication, and local control. It is used to complete the preprocessing, noise filtering, outlier removal, timestamp synchronization, and preliminary judgment of the original insulation status data on the field side.

[0028] In step S2-2, the sensing unit transmits the raw collected data to the edge computing terminal. The terminal preprocesses the data, including outlier removal, noise filtering, data normalization, and timestamp synchronization, removing invalid data caused by electromagnetic interference and environmental fluctuations, and forming a standardized insulation state dataset.

[0029] Specifically, in the preprocessing step S2-2, the raw data is initially screened based on the insulation standard threshold of railway signaling equipment, retaining valid data and removing obviously distorted data. At the same time, the data of each point is bound with the equipment number, geographical location, and collection time to form a traceable dataset. The preprocessed data is uploaded to the cloud intelligent analysis platform through the railway dedicated communication network (industrial Ethernet, LORA wireless communication). The edge computing terminal synchronously caches the local data to ensure data integrity in the event of data transmission interruption.

[0030] Step S3: AI algorithm is used for intelligent diagnosis and graded assessment of insulation status. The cloud platform analyzes the dataset through a pre-trained convolutional neural network and long short-term memory network fusion diagnostic model to achieve qualitative judgment of insulation status, fault type identification, and prediction of degradation trend. Combined with environmental parameter compensation and correction, the insulation status level is classified.

[0031] Specifically, intelligent diagnosis and grading assessment of insulation status based on AI fusion algorithms includes: This step employs a hybrid intelligent diagnostic algorithm that integrates Convolutional Neural Networks (CNN), Long Short-Term Memory Networks (LSTM), and fault tree expert rules. This algorithm balances static multi-feature classification with dynamic time-series trend prediction capabilities, and is specifically adapted to scenarios involving multi-source heterogeneous data, strong electromagnetic interference, and imbalanced fault samples in railway signaling equipment insulation monitoring. It fully couples data-driven approaches with expert experience, eliminating misjudgments from single algorithms and improving diagnostic accuracy under complex operating conditions. The cloud-based intelligent analysis platform pre-completes offline model training and lightweight deployment, enabling low-latency, rapid judgment at the edge during the inference phase. The specific steps are as follows: Step S3-1, the multi-source feature extraction and standardization preprocessing method is as follows: The standardized dataset received from the cloud platform is processed to construct a multi-dimensional feature vector, which serves as the input data for the algorithm. The feature vector has a total of 12 dimensions, specifically including: Static feature acquisition: Obtain a real-time 7-dimensional static feature vector for a single point, as shown in the following formula: ; Among them, the insulation resistance between core wires is R1, the insulation resistance to ground is R2, the leakage current is I, the ambient temperature is T, the ambient humidity is H, the partial discharge amplitude is P, and the partial discharge pulse frequency is F; Dynamic feature acquisition: insulation resistance change rate ΔR over 5 minutes, leakage current fluctuation coefficient ΔI, temperature and humidity coupling coefficient K, discharge signal distortion rate D, and equipment running time L.

[0032] The above features are subjected to min-max normalization to eliminate dimensional differences. The normalization formula is as follows: In the formula, Here, x represents the normalized eigenvalues, and x represents the original eigenvalues. min x max These represent the minimum and maximum values ​​of the historical samples for this feature, respectively. For abrupt changes such as partial discharge and leakage current, a wavelet threshold denoising secondary filter is used to remove pulse interference from traction current and track signals, retaining the fault feature components. After normalization, a standardized feature vector is obtained, as shown in the following equation:

[0033] Step S3-2, constructing the hybrid diagnostic model architecture and training methods, includes: The hybrid model comprises a three-layer architecture, achieving a closed loop throughout the entire process of feature extraction, trend analysis, and rule correction. Step S3-2-1, establish a CNN static feature extraction layer: adopt a lightweight CNN structure with 4 convolutional layers + 2 pooling layers, with a convolutional kernel size of 3×3 and the ReLU activation function. It is responsible for mining the correlation between static features, extracting the latent features of insulation faults, and outputting a static fault feature matrix. It focuses on identifying steady-state fault types such as dampness, damage, and aging, and solves the one-sided problem of judging by a single resistance threshold.

[0034] Specifically, the basic size of the convolution kernel is 3×1 (to adapt to a 7×1 one-dimensional feature matrix, to fit the operation of isolated feature sequences, discarding the conventional 3×3 size of image convolution to ensure the targeted nature of feature extraction); the activation function is ReLU activation function f(x)=max(0,x) throughout the process to complete nonlinear feature mapping and association mining; the padding method is Same padding to avoid loss of edge features; the stride is 1 to ensure the integrity of feature extraction.

[0035] The four-layer convolutional architecture includes: Conv1 (first convolutional layer): 16 convolutional kernels, performs primary local feature extraction on 7 original static features, captures the basic change pattern of individual features, and outputs 16 7×1 primary feature maps; Conv2 (second convolutional layer): 32 convolutional kernels, perform deep extraction of primary features from the previous layer, explore the coupling relationship between pairs of features, and output 32 7×1 intermediate feature maps; Conv3 (third convolutional layer): 64 convolutional kernels, focusing on the cross-correlation between environmental parameters, insulation electrical parameters, and partial discharge parameters, outputting 64 7×1 high-level feature maps; Conv4 (fourth convolutional layer): 32 convolutional kernels, compressing redundant features, extracting core fault representations, and outputting 32 7×1 depth feature maps; The two-layer pooling layer is constructed, including: All pooling layers use max pooling with a 2×1 kernel size. These kernels are inserted between Conv2 and Conv3 and after Conv4, respectively, to downsample the convolutional features, remove redundant information, retain core fault features, and reduce the computational load of the network, thus adapting to the low-latency requirements of railway online monitoring. After pooling, 32 final 3×1 convolutional feature maps are output, which are then flattened into one-dimensional feature vectors and fed into subsequent networks.

[0036] Step S3-2-2, establish an LSTM time series trend prediction layer: a bidirectional LSTM network is used with 64 hidden layer nodes and an input sequence length of 20 (corresponding to the last 20 sets of time series data). This layer is specifically designed to process the time series variation data of insulation parameters, learn the gradual law of insulation degradation, predict the trend of insulation parameter changes over the next 10 periods, and calculate the degradation rate v, as shown in the following formula: ; In the formula: The insulation resistance at the current moment. To predict the insulation resistance at a given time, The data acquisition period is n, and the prediction period number is n. The remaining time for the insulation parameters to drop to the fault threshold is predicted by the degradation rate, thus achieving early warning.

[0037] Step S3-2-3: Establish an expert rule correction layer: Based on the railway signaling equipment insulation operation and maintenance specifications and fault tree analysis logic, an expert rule base is constructed. The output results of CNN and LSTM are weighted and fused, and rules are verified. The weight allocation adopts adaptive dynamic weights. Let the static feature weight be ω1 and the temporal feature weight be ω2, and satisfy ω1+ω2=1. Under normal operating conditions, ω1=0.4 and ω2=0.6; under the condition of parameter mutation, it is automatically adjusted to ω1=0.6 and ω2=0.4; at the same time, the normal fluctuation of parameters caused by environmental temperature and humidity is eliminated. The correction formula is as follows: R true =R meas ×α(T)×β(H); In the formula, R true To correct for the true insulation resistance, R meas These are measured values. α(T) is the temperature compensation coefficient, and β(H) is the humidity compensation coefficient. The coefficients are calibrated through numerous field tests to adapt to the high and low temperature and humid environment of railway outdoor environments.

[0038] Step S3-3, Insulation Status Grading Determination: Based on the "Electrical Characteristics Standard for Railway Signaling Equipment", the insulation status is divided into four levels: normal, slightly deteriorated, moderately deteriorated, and severely faulty. The grading is completed by combining the fault probability value and deterioration rate output by the model with the corrected insulation resistance threshold. The specific determination criteria are as follows: S3-3-1 Normal Level Judgment: After correction, the insulation resistance to ground R ≥ 10MΩ, the failure probability P < 0.1, the degradation rate is slow, there are no abnormal partial discharge signals, the equipment insulation performance meets the standards, and there are no safety hazards. S3-3-2 Mild Deterioration Level Judgment: 5MΩ≤R<10MΩ, 0.1≤P<0.3, slow deterioration rate, weak partial discharge signal, belongs to early aging or slight moisture, no immediate risk; S3-3-3 Moderate Deterioration Level Judgment: 1MΩ≤R<5MΩ, 0.3≤P<0.7, the deterioration rate is relatively fast, the partial discharge signal is obvious, the leakage current exceeds the standard, which is a clear insulation deterioration, which is easy to develop into a fault and needs to be investigated within a time limit. S3-3-4 Severe Fault Level Judgment: R < 1MΩ, P ≥ 0.7, the degradation rate is rapidly accelerating, the partial discharge signal is strong, and the leakage current exceeds the standard significantly. This indicates insulation damage or a serious short circuit, which directly threatens driving safety and requires immediate action.

[0039] Step S3-4, Accurate Fault Type Identification: The model identifies 5 core fault types through feature matching, outputting fault type labels and confidence scores, including: Step S3-4-1 Insulation aging identification includes: slow decrease in resistance, no sudden change, and minimal impact from temperature and humidity; Step S3-4-2 Moisture and contamination identification includes: increased humidity, synchronous decrease in resistance, and stable discharge signal; Step 3-4-3 Mechanical damage identification includes: sudden change in resistance, violent discharge signal, and sudden increase in leakage current; Step 3-4-4 Indirect short circuit identification includes: large resistance fluctuations and abnormal synchronization of multi-core wire parameters; Steps 3-4-5 involve electromagnetic interference false alarm identification, including: single parameter abrupt changes, lack of temporal correlation, and rule-based elimination. To address the imbalanced fault sample problem, the SMOTE oversampling algorithm is used to expand the minority class of fault samples, improving the accuracy of identifying low-probability faults. The overall diagnostic accuracy of the model is ≥98.5%.

[0040] Step S3-5 Model Inference and Iterative Optimization: The inference phase achieves millisecond-level rapid judgment, and the results are transmitted back to the edge terminal in real time. The model adopts an incremental learning mechanism. After each fault handling and verification is completed, the labeled fault data is automatically included in the training set. The model parameters are fine-tuned regularly, and the feature weights and compensation coefficients are optimized to gradually adapt to the monitoring needs of different railway lines, different equipment models, and different regional climates, so as to realize the self-iterative upgrade of the algorithm.

[0041] Step S4: Accurate fault location and visualization. If the diagnosis is deterioration or fault, the fault location is accurately located based on the point distribution and parameter gradient, and a visualization interface and fault report are generated.

[0042] Step S4: Accurate location and visualization of the fault point.

[0043] Specifically, if the intelligent diagnostic result determines that the insulation is deteriorated or faulty, the cloud platform activates the fault location algorithm. Based on the distribution of sensing unit points, data transmission delay, parameter change gradient, and combined with the topology diagram of railway signal equipment, the fault point is accurately located: for signal cables, it is located to the specific core wire and section; for track insulation joints, it is located to the specific insulation node; for indoor and outdoor signal equipment, it is located to the specific equipment and circuit.

[0044] The location results are simultaneously pushed to the operation and maintenance terminal, generating a visual monitoring interface that intuitively displays the insulation status, fault level, fault type, precise location, and degradation trend curve of each point. At the same time, a fault detail report is generated, which includes the fault location, parameter data, occurrence time, and suggested handling measures, providing operation and maintenance personnel with a clear basis for fault handling.

[0045] Specifically, the fault location algorithm includes: Step S4-1, Location of the faulty equipment and circuit: The cloud platform retrieves the pre-entered railway signal equipment insulation monitoring topology database, which contains all sensing unit IDs, monitoring point names, circuits, equipment numbers, physical locations, and line topology connections. By using the sensing unit IDs corresponding to the abnormal data, the type of equipment to which the fault belongs (such as signal cables, track insulation joints, switch machines, and signal machines) and the power supply / control circuit are quickly identified, completing the initial division of a large-scale fault area, excluding circuits without abnormalities, and narrowing down the location range.

[0046] Step S4-2, for insulation faults in long-distance signal cables along railway lines, employs an adjacent sensing node parameter gradient difference algorithm to accurately pinpoint the fault section, including: Step S4-2-1: Retrieve real-time monitoring data from all distributed sensing units along the fault circuit, select the core judgment parameters: ground insulation resistance R2, leakage current I, and calculate the parameter gradient difference between adjacent sensing nodes, as shown in the following formula: ; ; In the formula: and These are the ground insulation resistance values ​​of the nth and n+1th sensing units, respectively. and These are the leakage current values ​​of the nth and n+1th sensing units, respectively; the larger the gradient difference, the more likely the fault point is located in the cable section between these two adjacent sensing units.

[0047] Step S4-2-2, Fault Zone Locking: Set a gradient difference judgment threshold; when adjacent nodes... , When the faulty section is identified, the amplitude and pulse frequency of the partial discharge signal are compared with those of each node to further verify the faulty section, eliminate the interference of abnormal data from single-point sensing units, and reduce the positioning accuracy to the distance between adjacent sensing units with an error of no more than 5m.

[0048] Step S4-3, precise location of the fault point (for insulation joints and individual devices).

[0049] For individual pieces of equipment such as track circuit insulation joints, indoor signaling equipment, and switch machines, an impedance distortion positioning + unique point mapping method is used to directly lock the precise location: Step S4-3-1, track circuit insulation joint fault: retrieve the bidirectional insulation resistance and leakage current data of the sensing units at both ends of the insulation joint, determine whether the fault is located in the rail end insulation joint, the slot-type insulation joint or the splice insulation, and simultaneously match the track circuit mileage station number and section number to directly locate the individual insulation node. Step S4-3-2, Individual signal equipment failure: By using the unique binding ID between the sensing unit and the equipment, the corresponding signal, switch machine, and indoor distribution panel core wire can be directly located to determine whether the fault is located in the internal insulation of the equipment or the external wiring insulation. Step S4-3-3, Core wire fault location: For multi-core signal cables, the faulty core wire number is identified by comparing the insulation resistance of each core wire, and the corresponding terminal or connection point is located by combining the topology path.

[0050] Step S4-4, Basic Fault Visualization: On the railway signal topology electronic map, the insulation status of each point is marked with different colors: green (normal), yellow (slight degradation), orange (moderate degradation), and red (severe fault). The fault point is highlighted and flashed to intuitively display the overall insulation status.

[0051] Step S5: Tiered early warning and operation and maintenance instruction push. Based on the insulation fault level, a differentiated early warning mechanism is triggered to push early warning information and operation and maintenance instructions to the corresponding terminals, forming a closed-loop management and control. Specifically, step S5: Tiered early warning and maintenance instruction push. Based on the insulation fault grading results, the cloud platform triggers a tiered early warning mechanism, with different warning methods used for different fault levels: (1) Mild degradation warning: Generate a warning message, push it to the operation and maintenance management backend, incorporate it into the daily operation and maintenance plan, and do not require emergency handling; (2) Moderate degradation warning: Triggering an audible and visual warning, pushing it to the terminal of the area maintenance personnel, requiring them to conduct inspections and investigations within a specified period; (3) Severe fault warning: The emergency warning is triggered immediately and pushed to the terminal of the operation and maintenance manager and the on-site operation and maintenance personnel. Multiple reminders are sent through SMS and APP pop-up windows, requiring them to rush to the site immediately to handle the situation and prevent the fault from escalating.

[0052] Meanwhile, the cloud platform synchronously feeds back diagnostic results, location information, and early warning information to the edge computing terminal, which stores and updates the monitoring data locally, forming a closed-loop management and control mechanism that combines cloud and edge collaboration. After the maintenance personnel complete the handling, they feed back the handling results to the cloud platform, and the model automatically updates the fault cases, completes self-learning optimization, and continuously improves diagnostic accuracy.

[0053] Step 6: Data archiving and model iteration optimization. Archive all data in the cloud and regularly optimize the diagnostic model based on new data to improve monitoring accuracy.

[0054] Specifically, step 6: data archiving and model iterative optimization.

[0055] The cloud platform archives all monitoring data, diagnostic results, fault cases, and operation and maintenance records for a long period of time, establishing a database of insulation status for railway signaling equipment. Based on newly added fault data and normal operation data, it iteratively trains the insulation status diagnostic model regularly, optimizes model parameters, adapts to the insulation monitoring needs of signaling equipment in different lines and environments, improves the model's generalization ability and diagnostic accuracy, and achieves continuous optimization and upgrading of monitoring technology.

[0056] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0057] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0058] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0059] These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable apparatus for implementing the process. Figure 1 One or more processes and / or boxes Figure 1The steps of the function specified in one or more boxes.

[0060] The above are merely embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention are included within the scope of the claims of the present invention pending approval.

Claims

1. A method for intelligent monitoring of the insulation status of railway signaling equipment, characterized in that, The intelligent monitoring method is based on the collaborative implementation of edge computing terminals, distributed multi-parameter sensing units, and cloud-based intelligent analysis platforms, and includes the following steps: Step S1: Set and initialize the multi-parameter sensing unit at the insulation monitoring point; The insulation monitoring points include: signal distribution panel, track circuit insulation joint, signal power input terminal and switch machine control circuit; Each of the multi-parameter sensing units integrates an insulation resistance acquisition module, a leakage current detection module, a temperature and humidity acquisition module, a partial discharge detection module, and an electromagnetic interference suppression module; Step S2 involves real-time acquisition and preprocessing of multi-dimensional insulation status data, including: edge computing terminal control sensing unit acquiring raw data on insulation resistance, leakage current, temperature and humidity, and partial discharge; completing outlier removal, noise filtering, and timestamp synchronization preprocessing to form a standardized dataset for uploading to the cloud platform; Step S3: Intelligent diagnosis and grading assessment of insulation status using AI algorithms: The cloud platform analyzes the dataset through a pre-trained convolutional neural network and long short-term memory network fusion diagnostic model, judges the insulation status, identifies fault types, and predicts the deterioration trend. Combined with environmental parameter compensation and correction, the insulation status level is classified. Step S4: Accurate fault location and visualization. If the diagnosis is deterioration or fault, the fault location is accurately located based on the point distribution and parameter gradient, and a visualization interface and fault report are generated. Step S5: Tiered early warning and operation and maintenance instruction push. Based on the insulation fault level, a differentiated early warning mechanism is triggered to push early warning information and operation and maintenance instructions to the corresponding terminals. Step S6: Data archiving and model iterative optimization.

2. The intelligent monitoring method for insulation status of railway signaling equipment as described in claim 1, characterized in that, Step S1, initializing the multi-parameter sensing unit, includes: Step S1-1, Address Allocation and Point Binding: According to the preset monitoring point numbering rules, a unique device address is assigned to each multi-parameter sensing unit, and a one-to-one correspondence is formed with the corresponding railway signal equipment number, monitoring location, and loop number. Step S1-2, Acquisition Channel Calibration and Zero Point Correction: Perform no-load calibration and zero point correction on the acquisition channels for insulation resistance, leakage current, and partial discharge.

3. The intelligent monitoring method for insulation status of railway signaling equipment as described in claim 1, characterized in that, Step S2, real-time acquisition and preprocessing of multi-dimensional insulation status data, includes: Step S2-1: The edge computing terminal issues a data acquisition command to control the sensing units at each point to synchronously acquire insulation status data; the acquisition parameters include: insulation resistance between signal core wires, insulation resistance between core wires and ground, circuit leakage current, ambient temperature and humidity, and partial discharge pulse signal. In step S2-2, the sensing unit transmits the raw collected data to the edge computing terminal. The edge computing terminal preprocesses the data, including outlier removal, noise filtering, data normalization, and timestamp synchronization, removing invalid data caused by electromagnetic interference and environmental fluctuations, and forming a standardized insulation state dataset.

4. The intelligent monitoring method for insulation status of railway signaling equipment as described in claim 1, characterized in that, Step S3, using AI algorithms for intelligent diagnosis and grading assessment of insulation status, includes: Step S3-1: Perform multi-source feature extraction and standardization preprocessing; Step S3-2: Construct and train the hybrid diagnostic model architecture; Step S3-3: Determine the insulation status classification; Step S3-4, Fault type identification; Step S3-5: Model inference and iterative optimization.

5. The intelligent monitoring method for insulation status of railway signaling equipment as described in claim 1, characterized in that, Step S4: The fault location algorithm includes: Step S4-1, Location of the faulty equipment and circuit: The cloud platform retrieves the pre-entered railway signal equipment insulation monitoring topology database, and through the sensing unit ID corresponding to the abnormal data, it locks the type of equipment and power supply-control circuit to which the fault belongs, completes the preliminary division of the fault area, and eliminates circuits without abnormalities; Step S4-2: For insulation faults in long-distance signal cables along railway lines, the gradient difference algorithm of adjacent sensing node parameters is used to accurately locate the fault section.

6. The intelligent monitoring method for insulation status of railway signaling equipment as described in claim 4, characterized in that, Step S3-1, the multi-source feature extraction and standardization preprocessing method is as follows: The standardized dataset received from the cloud platform is processed to construct a multi-dimensional feature vector, which serves as the input data for the algorithm. The feature dimensions include: Static feature acquisition: Obtain a real-time 7-dimensional static feature vector for a single location; Dynamic feature acquisition: acquire insulation resistance change rate ΔR over 5 minutes, leakage current fluctuation coefficient ΔI, temperature and humidity coupling coefficient K, discharge signal distortion rate D, and equipment running time L; The above features are normalized using a min-max method, as shown in the following equation: In the formula, Here, x represents the normalized eigenvalues, and x represents the original eigenvalues. min x max These are the minimum and maximum values ​​of the historical samples for this feature, respectively.

7. The intelligent monitoring method for insulation status of railway signaling equipment as described in claim 4, characterized in that, Step S3-2, constructing and training the hybrid diagnostic model architecture includes the following steps: Step S3-2-1, establish a CNN static feature extraction layer: adopt a lightweight CNN structure with 4 convolutional layers + 2 pooling layers, with a convolutional kernel size of 3×3 and the ReLU activation function to extract the latent features of insulation faults, output a static fault feature matrix, and identify steady-state fault types such as dampness, damage, and aging. Step S3-2-2, establish an LSTM time series trend prediction layer: a bidirectional LSTM network is used with 64 hidden layer nodes and an input sequence length of 20. This process handles the time series variation data of insulation parameters, learns the gradual pattern of insulation degradation, predicts the trend of insulation parameter changes over the next 10 periods, and calculates the degradation rate v, as shown in the following formula: ； In the formula: The insulation resistance at the current moment. To predict the insulation resistance at a given time, Where n is the acquisition period and n is the prediction period number; Step S3-2-3, Establish an expert rule correction layer: Based on the railway signaling equipment insulation operation and maintenance specifications and fault tree analysis logic, construct an expert rule base, and perform weighted fusion and rule verification on the output results of CNN and LSTM. The weight allocation adopts adaptive dynamic weights, with static feature weights set as ω1 and temporal feature weights as ω2, satisfying ω1+ω2=1; at the same time, normal parameter fluctuations caused by environmental temperature and humidity are removed, and the correction formula is as follows: R true =R meas ×α(T)×β(H); In the formula, R true To correct for the true insulation resistance, R meas These are measured values, α(T) is the temperature compensation coefficient, and β(H) is the humidity compensation coefficient.

8. The intelligent monitoring method for insulation status of railway signaling equipment as described in claim 5, characterized in that, Step S4-2 includes: Step S4-2-1: Retrieve real-time monitoring data from all distributed sensing units along the fault circuit, select the core judgment parameters: ground insulation resistance R2, leakage current I, and calculate the parameter gradient difference between adjacent sensing nodes, as shown in the following formula: ； ； In the formula: and These are the ground insulation resistance values ​​of the nth and n+1th sensing units, respectively. and These are the leakage current values ​​of the nth and n+1th sensing units, respectively. Step S4-2-2, Fault Zone Locking: Set a gradient difference judgment threshold; when adjacent nodes... , When the faulty section is identified, the amplitude and pulse frequency of the partial discharge signal are compared with those of each node to further verify the faulty section, eliminate the interference of abnormal data from single-point sensing units, and reduce the positioning accuracy to the distance between adjacent sensing units with an error of no more than 5m.

Images