Track infrastructure safety guarantee method, system and equipment and storage medium

Abstract

The invention discloses a rail infrastructure safety guarantee method, system and device and a storage medium, and relates to the technical field of rail traffic intelligent safety monitoring and control, and the method comprises the steps: collecting original rail data through a multi-source sensing module; processing the original orbit data by using a bilateral filter and a region growing method to obtain orbit data containing semantic tags; performing feature extraction on the orbit data through a data fusion center, and constructing a space-time correlation feature matrix; based on the track data, thermal-mechanical coupling simulation is carried out through a digital twinborn decision-making platform, and a simulation result is obtained; and based on the space-time correlation characteristic matrix and the simulation result, through an adaptive control module, generating an optimization control instruction and alarm information. According to the invention, the problems of fragmentization of a perception system, insufficient diagnosis decision intelligence and risk disposal passivity in related technologies are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent safety monitoring and control of rail transit, and particularly relates to a method, system, device and storage medium for ensuring the safety of rail infrastructure. Background Art

[0002] With the gradual shortening of the train tracking interval and the development of rail transit unmanned technology, the demand for perceiving the train operation environment has become increasingly prominent. Especially the construction of ultra-large-scale railway networks has made the railway network face complex geological, topographical and meteorological conditions. Due to the influence of these complex factors, the types of safety risks in train operation are numerous, with high dimensions, wide ranges, strong suddenness and difficult identification. Common safety hazards include infrastructure failures, train failures, foreign object intrusion, natural environment changes and disasters, etc., and these factors may directly threaten the operation safety of trains.

[0003] The current rail infrastructure safety assurance technology faces the following key bottlenecks: (I) Fragmentation problem of the perception system: Most existing monitoring systems adopt a single sensor network (such as only deploying strain gauges or cameras), resulting in: Lack of data dimensions: Traditional fiber optic sensors can only collect axial strain and cannot synchronously obtain the track temperature gradient (affecting the calculation accuracy of rail expansion joints by 5 - 8%); Difficulty in spatio-temporal alignment: There is a millisecond-level time delay deviation between vibration signals (sampled at 10 kHz) and video data (30 fps), resulting in a false alarm rate of up to 22% for switch jamming failures; Poor environmental adaptability: The recognition accuracy of conventional cameras drops to 61% in rainy and foggy weather, unable to meet the all-weather monitoring requirements.

[0004] (II) Insufficient intelligence in diagnosis and decision-making: Mainstream solutions rely on threshold alarms and manual experience judgment, and there are: Serious false alarms and missed alarms: In the scenario of dynamic changes in the wheel-rail adhesion coefficient, the false trigger rate of the alarm system based on fixed thresholds exceeds 35%; Significant response delay: The average time from abnormal detection to the issuance of a dispatching instruction is 8.7 minutes, unable to meet the 5-minute-level train control response requirements for heavy-haul trains; Lack of root cause location: Existing systems can only report "track geometry overrun", but cannot distinguish whether it is caused by loose fasteners (accounting for 62%) or subgrade settlement (accounting for 29%).

[0005] (III) Dilemma of passive risk handling: Traditional methods adopt a passive mode of "fault occurrence - manual inspection - planned maintenance". Insufficient preventive maintenance results in 78% of sudden failures, and the loss of each unplanned outage exceeds 2 million yuan per hour, leading to high maintenance costs.

[0006] Therefore, there is an urgent need for a technical solution that can not only improve the holographic, intelligent, and autonomous levels of line perception but also effectively enhance the protection ability of train operation safety, ultimately achieving more accurate and efficient fault prediction and risk response to ensure the safe and stable operation of trains. Summary of the Invention

[0007] To solve the above technical problems, the invention provides a method, system, device, and storage medium for ensuring the safety of track infrastructure.

[0008] To achieve the above object, the present invention provides the following technical solutions: In a first aspect, an embodiment of the present invention provides a method for ensuring the safety of track infrastructure, the method comprising: Collecting original track data through a multi-source perception module; Processing the original track data using a bilateral filter and region growing method to obtain track data containing semantic tags; Extracting features from the track data through a data fusion center to construct a spatio-temporal correlation feature matrix; Based on the track data, performing a thermal-mechanical coupling simulation through a digital twin decision platform to obtain a simulation result; Based on the spatio-temporal correlation feature matrix and the simulation result, generating an optimized control instruction and an alarm message through an adaptive control module.

[0009] In a second aspect, an embodiment of the present invention provides a system for ensuring the safety of track infrastructure, the system comprising: a multi-source perception module, a data fusion center, a digital twin platform, and an adaptive control module, The multi-source perception module is configured to collect original track data; The data fusion center is configured to process the original track data using a bilateral filter and region growing method to obtain track data containing semantic tags; extracting features from the track data to construct a spatio-temporal correlation feature matrix; The digital twin decision platform is configured to perform a thermal-mechanical coupling simulation based on the track data to obtain a simulation result, wherein the simulation result includes the track thermal expansion coefficient and the current track stress field distribution data; The adaptive control module is configured to generate an optimized control instruction and an alarm message based on the spatio-temporal correlation feature matrix and the simulation result.

[0010] In a third aspect, an embodiment of the present invention further provides an electronic device, including: a memory and a processor; the processor is configured to read and execute a computer program stored in the memory to implement the steps of the foregoing method for ensuring the safety of track infrastructure.

[0011] In a fourth aspect, an embodiment of the present invention further provides a computer storage medium, in which computer-executable instructions are stored, and when the computer-executable instructions are executed, the steps of the foregoing method for ensuring the safety of track infrastructure are implemented.

[0012] The beneficial effects brought by the technical solutions provided by the embodiments of the present invention include: Collect original track data through a multi-source perception module; process the original track data by using a bilateral filter and a region growing method to obtain track data containing semantic labels; extract features from the track data through a data fusion center to construct a spatio-temporal correlation feature matrix; based on the track data, perform a thermal coupling simulation through a digital twin decision platform to obtain a simulation result; based on the spatio-temporal correlation feature matrix and the simulation result, generate an optimized control instruction and an alarm message through an adaptive control module. Through the present invention, it integrates high-precision multi-source perception, intelligent decision-making analysis, and digital twin simulation. By deeply integrating BIM modeling, 3D physical simulation, real-time data monitoring, and predictive maintenance technologies, it realizes all-element, all-time, and all-dimensional three-dimensional perception and dynamic risk control of the train operation environment, significantly improves the reliability of railway infrastructure and the safety of train operation, and solves the problems of fragmented perception systems, insufficient intelligent diagnosis and decision-making, and passive risk handling dilemmas in related technologies. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0014] Figure 1 It is a schematic flowchart of an embodiment of the method for ensuring the safety of track infrastructure of the present invention; Figure 2 It is a schematic diagram of the functional modules of an embodiment of the system for ensuring the safety of track infrastructure of the present invention; Figure 3 It is a schematic structural diagram of an electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0015] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0016] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be described in detail below in conjunction with the accompanying drawings.

[0017] In a first aspect, an embodiment of the present invention provides a method for ensuring the safety of track infrastructure.

[0018] In one embodiment, referring to Figure 1 , Figure 1 is a schematic flowchart of an embodiment of the method for ensuring the safety of track infrastructure of the present invention. As Figure 1 shown, the method for ensuring the safety of track infrastructure includes: Step S10, collecting original track data through a multi-source perception module; In this embodiment, the multi-source perception module includes fiber Bragg grating sensors, lidar, interferometric synthetic aperture radar, multispectral cameras, and infrared thermal imagers.

[0019] Deploy high-precision fiber Bragg grating sensors (FBGs) along the track for real-time monitoring of the stress, temperature, and deformation data of the track. Among them, the strain measurement accuracy of the fiber Bragg grating sensor (FBG) can reach ±0.5 με (microstrain, ), the temperature measurement accuracy is ±0.5 °C, and the sampling frequency is 10 kHz.

[0020] Adopt LiDAR (light detection and ranging) and InSAR (interferometric synthetic aperture radar) technologies for large-scale terrain scanning, combined with three-dimensional point cloud modeling, to achieve millimeter-level accuracy monitoring of the track surrounding environment and obtain terrain data.

[0021] Deploy multispectral cameras and infrared thermal imagers, and use image recognition algorithms (such as YOLOv5) to detect anomalies such as foreign object intrusion and equipment overheating on the track in real time, and obtain environmental image data of the track.

[0022] That is, the stress data, deformation data, temperature data, terrain data, and environmental image data of the track can be collected through the multi-source perception module.

[0023] Through this embodiment, the fiber Bragg grating sensor can be used to synchronously obtain the track temperature gradient, solving the problem of missing data dimensions in the prior art. Moreover, the multi-spectral camera and the infrared thermal imager are not affected by rainy and foggy weather and can detect the track, solving the problem that the devices used in the prior art have poor environmental adaptability and cannot meet the all-weather monitoring requirements.

[0024] The multi-source perception module realizes the spatio-temporal alignment of vibration signals and video data through the GPS clock synchronization unit (nanosecond-level accuracy). Among them, the timestamps of the vibration signals (sampled at 10 kHz) and video frames (30 fps) are synchronized through the PTP precise time protocol; the data fusion center uses the cubic spline interpolation algorithm to align the non-uniformly sampled data, and the maximum delay error ≤ 0.1 ms. All sensors are integrated with a GPS module (such as u-blox ZED-F9T), outputting a 1PPS (pulses per second) signal, and the synchronization error ≤ 20 ns; the video camera is built with a hardware encoder to write the frame timestamp into the video stream metadata. This solves the problem of difficult spatio-temporal alignment in the prior art: there is a millisecond-level time delay deviation between the vibration signals (sampled at 10 kHz) and video data (30 fps), resulting in a high false alarm rate for switch jamming failures.

[0025] Among them, .

[0026] Step S20: Process the original track data by using a bilateral filter and a region growing method to obtain track data containing semantic labels; In this embodiment, after the multi-source perception module obtains the stress data, deformation data, temperature data, terrain data, and environmental image data of the track, since the terrain data obtained during terrain scanning by LiDAR (light detection and ranging) and InSAR (interferometric synthetic aperture radar) is point cloud data, the bilateral filter and the region growing method are used to denoise and segment the terrain data in the original track data, and then the track data containing semantic labels such as tracks, fasteners, and ballast can be obtained.

[0027] Among them, the bilateral filter is a non-linear filter that combines the similarity in the spatial domain and the pixel value domain. It performs filtering by considering the spatial distance and gray value difference between pixels. The bilateral filter can retain edge details and effectively suppress noise.

[0028] The region growing method refers to the process of developing groups of pixels or regions into larger regions. Starting from a set of seed points, the region growth from these points is achieved by merging adjacent pixels with similar attributes such as intensity, gray level, texture color, etc. into this region.

[0029] The code for processing the original track data using a bilateral filter and region growing method is as follows.

[0030] rawpoints = lidar.scan() # Raw point cloud acquisition denoised = bilateral filter(raw points) # Bilateral filtering for noise reduction segmented = region_growing(denoised) # Segmenting track components using region growing method Step S30: Extract features from the track data through the data fusion center and construct a spatio-temporal correlation feature matrix; In some specific embodiments, step S30 includes: Step S301: Use wavelet transform and Kalman filtering algorithm to denoise and smooth the track data to obtain the first track data; In this embodiment, wavelet transform is used to eliminate noise from the track data containing semantic labels, and Kalman filtering algorithm is combined for data smoothing to obtain the first track data.

[0031] Among them, wavelet transform (WT) is a new transform analysis method that can provide a "time-frequency" window that changes with frequency and is an ideal tool for time-frequency analysis and processing of signals. Its main feature is that through the transform, it can fully highlight the characteristics of certain aspects of the problem, can perform local analysis of time (space) frequency, gradually refine the signal (function) through stretching and translation operations at multiple scales, finally achieving fine time division at high frequencies and fine frequency division at low frequencies, and can automatically adapt to the requirements of time-frequency signal analysis, thus focusing on any details of the signal.

[0032] Kalman filtering is an algorithm that uses a linear system state equation to optimally estimate the system state through system input and output observation data. Since the observation data includes the influence of noise and interference in the system, the optimal estimate can also be regarded as a filtering process. Data filtering is a data processing technology for removing noise and restoring real data.

[0033] Step S302: Use principal component analysis and t-distributed stochastic neighbor embedding algorithm to reduce the dimension of the first track data to obtain the second track data; In this embodiment, principal component analysis (PCA) is a commonly used data dimension reduction technology used to convert a high-dimensional data set into a low-dimensional representation while retaining the main features of the data.

[0034] The t-distributed Stochastic Neighbor Embedding algorithm (t-SNE for short) is a non-linear technique for dimensionality reduction of high-dimensional data. This algorithm is particularly suitable for mapping high-dimensional data into two-dimensional or three-dimensional space for visualization purposes.

[0035] The first orbit data is reduced in dimension using the principal component analysis method and the t-distributed Stochastic Neighbor Embedding algorithm to obtain the second orbit data, so as to remove redundant features in the data and reduce the computational complexity of high-dimensional data.

[0036] Step S303: Extract key features from the second orbit data through a spatio-temporal graph convolutional network to obtain feature orbit data; In this embodiment, the network structure: A 3-layer spatio-temporal graph convolutional network (ST-GCN) is adopted, and each layer contains: Spatio-temporal convolutional module: In the spatial dimension, a graph convolutional kernel (kernel size K = 3) is used, and in the temporal dimension, a one-dimensional convolutional layer (kernel size T = 5, stride 1) is used; Activation function: ReLU; Normalization layer: Batch normalization (BatchNorm); Input data: The second orbit data, in the format X ∈ RN×T×D, where R is a real number, N is the number of nodes, T is the number of time steps, and D is the feature dimension; Output data: Feature orbit data, that is, the spatio-temporal correlation feature matrix M ∈ RN×T×128.

[0037] The execution code of the spatio-temporal graph convolutional network (ST-GCN) is as follows: # Spatio-temporal alignment pseudocode def time_align( vibration signal,# Vibration signal (sampled at 10 kHz) video frame# Video frame (30 fps) ): timestamp = get precise timestamp()# Synchronize to GPS clock (nanosecond-level precision) return aligned_data# Feature orbit data after key feature extraction Step S304: Based on the feature orbit data, construct a spatio-temporal graph through a graph neural network library; In this embodiment, the graph structure update rule: When the deformation difference between nodes > 3 mm or the temperature gradient difference > 2 °C / m, edges are dynamically added / deleted.

[0038] Construction Tools and Algorithms: PyTorch Geometric: For efficiently constructing and processing spatio-temporal graphs, supporting GPU acceleration; DTW (Dynamic Time Warping) Algorithm: Calculates the similarity of node time series data to solve the problem of asynchronous sampling.

[0039] Based on the feature track data, a spatio-temporal graph G=(V, E) is constructed through a graph neural network library, such as the PyTorch Geometric library.

[0040] 1. Node Definition: The track is divided into nodes v1, v2,..., v at 50m intervals N , and the feature track data of each node includes: Vibration Spectrum: The first 5 main frequency amplitudes (0 - 500Hz) are extracted through FFT; Temperature Gradient: (Lateral Temperature Change Rate), (Time Temperature Change Rate); LiDAR (LightLaser Detection and Ranging) Deformation Vector: Three-dimensional deformation amounts (Δx, Δy, Δz), with an accuracy of ±1mm; 2. Edge Connection Rule: If the distance between two nodes ≤ 100m and the deformation correlation coefficient ρ ≥ 0.7, then an edge e is established ij ; Edge e ij ∈E: Calculate the spatio-temporal correlation between nodes based on the Dynamic Time Warping (DTW) algorithm, and the edge weight w ij =1 + DTW(v i , v j ), and the update period is 1 minute.

[0041] 3. Spatio-temporal Graph Construction: Use the Python library NetworkX to generate the spatio-temporal graph G=(V, E), with the number of nodes N ≤ 2000 (covering a 100km track); update the edge weights every 5 minutes to respond to the propagation of deformation.

[0042] Step S305, generate a spatio-temporal correlation feature matrix based on the feature track data and the spatio-temporal graph.

[0043] In this embodiment, first, perform spatio-temporal graph convolution operations to obtain the time series features of each node; then perform feature aggregation on the time series features of each node to obtain node-level feature vectors; finally, based on the node-level feature vectors, construct a spatio-temporal correlation feature matrix.

[0044] Spatio-temporal graph convolution operation: The feature track data is processed through a 3-layer spatio-temporal graph convolutional network (ST-GCN) to obtain the time series features of each node. Among them, each layer of the spatio-temporal graph convolutional network (ST-GCN) performs the following steps: Spatial convolution: Graph convolution based on the adjacency matrix A, with a kernel size of 3, and the weight matrix W ∈ R^D×D′ (R is a real number, D = 64, D′ = 128); Temporal convolution: A one-dimensional convolutional kernel (size 5, stride 1) to extract features in the time dimension; Activation and normalization: ReLU activation function and batch normalization (BatchNorm); Feature aggregation: Adaptive average pooling (AdaptiveAvgPool1d) is performed on the time series features of each node, and the node-level feature vector h is output i ∈ R^128; Construct a spatio-temporal correlation feature matrix: Sort all node-level feature vectors according to their spatial positions, extract high-order features through the spatio-temporal graph convolutional network (ST-GCN), and construct a spatio-temporal correlation feature matrix M ∈ R^N×T×128, where N is the number of nodes, and T = 6 represents 6 time steps (30-minute window).

[0045] Step S40: Based on the said track data, perform a thermal-mechanical coupling simulation through the digital twin decision platform to obtain the simulation results; In some specific embodiments, step S40 includes: Step S401: Use a semantic segmentation network to identify and classify the said track data to determine the faulty components; Step S402: Adjust the parameters of the faulty components through the building information model in the digital twin platform; Step S403: Based on the adjusted parameters, perform a thermal-mechanical coupling simulation to obtain the track thermal expansion coefficient and the current track stress field distribution data.

[0046] In this embodiment, the digital twin decision platform is constructed based on the building information model (BIM model) and the physics engine (such as NVIDIA PhysX). The digital twin decision platform includes a geometric model, a behavior model, and a physical model. Among them, the geometric model: Parametric modeling of devices such as signal lights and turnouts is realized through Revit (Revit is the name of a series of software), with an LOD400 accuracy. The behavior model: Use Petri nets (Petri nets are a mathematical representation of discrete parallel systems) to construct an interlocking logic simulation, supporting the dynamic deduction of route arrangement and signal light state switching. The physical model: Perform a coupling simulation of the track stress field and the thermodynamic field through ANSYS software (a large general finite element analysis (FEA) software), with a calculation accuracy of 0.01 mm level.

[0047] Use a semantic segmentation network (U-Net++) to identify and classify the point cloud data scanned by LiDAR (Light Detection and Ranging) to determine the faulty components. Exemplarily, when LiDAR scans a 2.3-mm lateral displacement of the sleeper, the U-Net++ classification module classifies it as "fastener loose".

[0048] The structural parameters of the semantic segmentation network (U-Net++) are as follows.

[0049] model = UnetPlusPlus( backbone='ResNet34' input_shape=(1024, 1024, 3) classes=6 # track / catenary / switch / fastener / ballast / others ) Adjust the fastener friction coefficient through the Building Information Model (BIM model) in the digital twin platform. Based on the adjusted fastener friction coefficient, perform thermo-mechanical coupling simulation through the physical model in the digital twin platform to obtain the track thermal expansion coefficient and the current track stress field distribution data.

[0050] The thermo-mechanical coupling simulation code is as follows: graph TB A[Input state vector]-->B(ANSYS solver) B-->C{Deformation prediction} -->>3mm| D[Trigger MPC control] C-->≤3mm|E[Generate monitoring report] Step S50, based on the spatio-temporal correlation feature matrix and the simulation results, generate an optimized control instruction and an alarm message through the adaptive control module.

[0051] In some specific embodiments, step S50 includes: Step S501, obtain historical track data within a preset time period; Step S502, based on the historical track data and the simulation results, obtain the track deformation prediction value and the confidence interval of the track deformation prediction value through a long-sequence prediction model; Step S503, when the track deformation prediction value is not within the confidence interval, optimize and solve the spatio-temporal correlation feature matrix and the simulation results through the target optimization function included in the adaptive control module to obtain an optimized control instruction; Step S504: Query the correspondence table between the track deformation value and the deformation level based on the predicted track deformation value, determine the track deformation level, and obtain the corresponding warning information based on the track deformation level.

[0052] In this embodiment, taking the preset duration of 72 hours as an example, the simulation results (track thermal expansion coefficient and current track stress field distribution data) and the historical track data of 72 hours are input into the long-sequence prediction model to obtain the predicted track deformation value and the confidence interval of the predicted track deformation value. Among them, the long-sequence prediction model (LSTM+Attention) is established using the Transformer model architecture.

[0053] The data structure code of the input to the long-sequence prediction model is as follows.

[0054] input_data ={ 'temp_history':np.array([-15,-18,...,20]), # Track data sequence in the past 72 hours 'Stress_distribution':(256,256)tensor, # Current track stress field distribution 'Material_params':{'thermal_expansion':11.7e-6} # Thermal expansion coefficient of the rail } When the predicted track deformation value is not within the confidence interval of the predicted track deformation value, through the objective optimization function included in the adaptive control module, the spatio-temporal correlation feature matrix and the simulation results are optimized and solved to obtain the optimized control instruction. The objective optimization function is:

[0055] In the formula, u represents the set of control variables, including the train speed v, the braking force B, and the turnout compensation angle θ, J represents the total cost function, which measures the weighted sum of safety, equipment wear and energy consumption, represents the minimum total cost corresponding to the set of control variables, represents the longitudinal impulse force when the rolling optimization time domain step of the model predictive control is k, represents the wheel-rail wear index when the rolling optimization time domain step of the model predictive control is k, represents the energy consumption when the rolling optimization time domain step of the model predictive control is k, , and represent the weight coefficients (α+β+γ=1), k represents the rolling optimization time domain step of the model predictive control. Preferably, α=0.6, β=0.3, γ=0.1.

[0056] Represents the longitudinal impact force when the rolling optimization time domain step of the model predictive control is k, and the calculation formula , where m is the mass of the train, a is the acceleration, c is the air resistance coefficient, is the train speed; Represents the wheel-rail wear index when the rolling optimization time domain step of the model predictive control is k, calculated based on the Archard model , Fn is the normal force, s is the sliding distance, and H is the material hardness; Represents the energy consumption when the rolling optimization time domain step of the model predictive control is k, , where η is the braking energy recovery efficiency, B is the braking force, and v is the train speed.

[0057] Among them, the constraint conditions of the spatio-temporal correlation feature matrix M: ; is the correlation feature matrix at the k-th time step, is the correlation feature matrix at the (k-1)-th time step.

[0058] Simulation result constraint conditions: , the thermal stress of the track does not exceed the yield strength.

[0059] Represents the threshold of the spatio-temporal feature matrix change rate , determined by the fatigue characteristics of the track material; Represents the thermal stress of the track output by the thermo-mechanical coupling simulation, is the yield strength of the rail. For example, for Q345 steel, the yield strength is 345 MPa.

[0060] The input data stream flows to the target optimization function, and the input data stream includes: 1. Multi-source perception module: Real-time collection of data such as vibration, temperature, and deformation (frequency 10 kHz to 30 Hz).

[0061] 2. Model prediction: Based on LSTM or Kalman filter, generate the predicted deformation values Dpred and confidence intervals [D low , D high for the next 6 steps of deformation.

[0062] 3. Thermo-mechanical coupling simulation: Real-time calculation of the thermal stress of the track through the solver of the finite element analysis software (ANSYS) .

[0063] Align the input data stream in time and perform preprocessing: 1. All input data streams are synchronized by GPS clock (error ≤ 0.1ms) to ensure alignment within the optimization period (≤ 100ms).

[0064] 2. Data preprocessing (filtering, dimensionality reduction) compresses the original data into a 128 - dimensional state variable x ∈ R128, which contains key features (vibration spectrum, temperature gradient, deformation vector).

[0065] Real - time guarantee for the rolling optimization period (≤ 100ms): Calculation process: 1. Data acquisition and prediction (20ms): Multi - source sensor data acquisition (≤ 5ms); Model prediction to generate future 6 - step deformations and confidence intervals (≤ 10ms); Thermomechanical coupling simulation to calculate the current stress field (≤ 5ms).

[0066] 2. Optimization problem construction (10ms): Define the objective function J(u) and constraints (deformation threshold, stress limit); Take the 128 - dimensional state variable x as the input of the optimization model.

[0067] 3. Optimization solution (60ms): Use the SQP algorithm to solve the constrained optimization problem and generate future 6 - step control instructions u1, u2, ……, u6.

[0068] 4. Instruction issuance (10ms): Only apply the first - step instruction u1 (such as adjusting the braking force or switch angle); The remaining instructions u2, ……, u6 are used as initial guesses for the next cycle.

[0069] Key design: Model lightweighting: The prediction model and simulator adopt simplified versions (such as reduced - order models) to ensure the calculation speed.

[0070] Hardware acceleration: Through parallel computing of FPGA or GPU, compress the time of thermomechanical simulation and optimization solution.

[0071] Dynamic adjustment of the control time domain (6 steps): Time - domain division: Each step corresponds to the optimization period (100ms), and the total duration of the control time domain T = 6×100ms = 600ms.

[0072] In each optimization, update the prediction and constraints for the future 6 steps based on the latest data.

[0073] Rolling strategy: Initial optimization window: t0 to t0 + 600 ms; Next cycle window: t0 + 100 ms to t0 + 700 ms, and so on.

[0074] Adaptive adjustment: If the confidence interval width Dhigh - Dlow > 3 mm, automatically expand the control time domain to 8 steps (increase fault tolerance).

[0075] For the optimization control instruction, specifically, 1. Input data preprocessing: The spatio-temporal correlation feature matrix M ∈ RN×6×128 is unfolded in time steps to M′ ∈ R6N×128; Extract key parameters from the thermo-mechanical coupling simulation results: coefficient of thermal expansion α thermal , maximum stress σ max , temperature field distribution T(x, y); 2. Optimization variable definition: Control variable u = [v, B, θ], where: v: target speed of the train (range 0 ≤ v ≤ 120 km / h); B: braking torque (range 0 ≤ B ≤ 500 kN); θ: turnout compensation angle (-5° ≤ θ ≤ +5°); 3. Objective function construction: .

[0076] 4 Constraint condition loading: Spatio-temporal correlation constraint: (Track deformation rate limit); Thermal stress constraint: ≤ 345 MPa (yield strength of Q345 steel); Dynamic constraint: ∣a∣ ≤ 0.3g (acceleration limit, g = 9.8 m / s 2 ); 5. Optimization solution: Use the sequential quadratic programming (SQP) algorithm to solve the constrained non-linear optimization problem, and the initial guess value u0 comes from the historical optimal instruction library; Output the optimization variables , and send them as optimization control instructions to the train control system and the turnout actuator.

[0077] Furthermore, after obtaining the control instructions, the control instructions (such as route arrangement, signal status change instructions) need to be authenticated by the deployed blockchain (Hyperledger Fabric) node consensus verification to ensure that the control instructions cannot be tampered with.

[0078] The code for the blockchain verification process is as follows: Function executecommand(bytes memory cmd)public { require(validateHash(cmd.hash),"Invalid hash"); require(checksignatures(cmd.sigs,2),"Insufficient signatures") emit CommandExecuted(cmd); / / Command execution event } cmd.hash is the simulation hash value (SHA-256), and cmd.sigs is the edge node signature (greater than or equal to 2).

[0079] After obtaining the predicted value of track deformation, it is also necessary to query the correspondence table between the track deformation value and the deformation level based on the predicted value of track deformation, determine the track deformation level, and obtain the corresponding warning information based on the track deformation level.

[0080] The correspondence table between the track deformation value and the deformation level is shown in Table 1.

[0081] Table 1

[0082] If the track deformation level is level one, the indicator light turns on. At this time, the trend curve is displayed on the Web interface, and the AR model is highlighted for prompt. Log records are made to prompt the start of the manual inspection plan.

[0083] If the track deformation level is level two, the warning light turns on. At this time, a red area HUD pop-up window reminder appears in the three-dimensional situation map. The system performs self-check and starts the standby control strategy. HUD (Head-Up Display) is a display technology that projects information in front of the user's line of sight so that the user can view the information without having to lower their head.

[0084] If the track deformation level is level three, the alarm light turns on. At this time, the full screen flashes and an alarm voice is broadcast, the speed is automatically limited, and the control strategy is optimized through the Model Predictive Control (MPC) algorithm.

[0085] If the track deformation level is level four, the emergency light turns on. At this time, the power supply of non-critical equipment is cut off, and the emergency brake of the EMU is triggered.

[0086] In another embodiment, based on the predicted value of track deformation and its confidence interval, and combined with real-time environmental parameters (including the track temperature gradient ΔT, stress field distribution σ maxBased on the main frequency offset Δf of the vibration spectrum, determine the level of track deformation and generate corresponding alarm information.

[0087] Exemplarily, the alarm light gives a blue hint, the track deformation is 12 mm, the temperature gradient is 35 °C / m, and the trend curve of the AR model is highlighted and displayed on the Web interface.

[0088] The alarm light gives a yellow early warning, the track deformation is 23 mm, the vibration energy suddenly increases by 20%, and a HUD pop-up window reminder appears in the red area of the 3D situation icon; The alarm light gives an orange alarm, the deformation is 35 mm, the stress > 250 MPa, and the full screen flashes with alarm voice broadcast; The alarm light gives a red emergency alarm, the deformation > 5 mm, the communication interruption > 5 s, cut off the power supply of non-critical equipment and trigger the emergency braking of the EMU.

[0089] In this embodiment, the original track data is collected by the multi-source perception module; the bilateral filter and region growing method are used to process the original track data to obtain track data containing semantic tags; the data fusion center extracts features from the track data to construct a spatio-temporal correlation feature matrix; based on the track data, a thermal-mechanical coupling simulation is carried out through the digital twin decision platform to obtain a simulation result; based on the spatio-temporal correlation feature matrix and the simulation result, an optimization control instruction and an alarm information are generated through the adaptive control module. Through this embodiment, it integrates high-precision multi-source perception, intelligent decision-making analysis, and digital twin simulation. By deeply integrating BIM modeling, 3D physical simulation, real-time data monitoring, and predictive maintenance technologies, it realizes the three-dimensional perception and dynamic risk control of all elements, all time periods, and all dimensions of the train operation environment, significantly improves the reliability of railway infrastructure and the safety of train operation, and solves the problems of fragmented perception systems, insufficient intelligent diagnosis and decision-making, and passive risk handling dilemmas in related technologies.

[0090] Optionally, in one embodiment, after step S50, it includes: Obtain historical track data within a preset time period; Based on the historical track data and the simulation result, through a long sequence prediction model, obtain the track deformation prediction value and the confidence interval of the track deformation prediction value; Construct a distributed diagnosis model based on the federated learning framework, and aggregate the distributed diagnosis model through the central server to obtain a global model; When the difference between the vibration frequencies and amplitudes of multiple track components is detected to be less than a preset value based on the global model, use the knowledge graph and root cause analysis method to infer the root cause probability of the failures of multiple track components; Based on the track deformation prediction value, the confidence interval of the track deformation prediction value, and the root cause probability, obtain the risk probability distribution of the track.

[0091] In this embodiment, taking the preset duration as 72 hours as an example, the simulation results (orbital thermal expansion coefficient and current orbital stress field distribution data) and the historical orbital data of 72 hours are input into the long sequence prediction model to obtain the predicted orbital deformation value and the confidence interval of the predicted orbital deformation value.

[0092] Based on the federated learning framework (FATE platform), a distributed diagnostic model is constructed, and the distributed diagnostic models are aggregated through a central server to obtain a global model.

[0093] Specifically, the central server distributes the distributed diagnostic model parameters to each client. The client independently trains the model using local data and updates the parameters (such as gradient descent). The server collects the parameters updated by the client and generates an initial global model after weighted averaging according to the data volume. Repeat the above steps until the model converges to obtain the final global model.

[0094] When the difference between the vibration frequencies and amplitudes of multiple track components is detected to be less than the preset value based on the obtained final global model, the root cause probabilities of the failures of multiple track components are inferred using the root cause analysis method and the knowledge graph. Exemplarily, when the final global model detects that multiple track components have similar vibration characteristics, that is, the difference between the vibration frequencies and amplitudes of multiple track components is less than the preset value, the historical case library is retrieved through the knowledge graph, and combined with the root cause analysis method and environmental parameters such as the current temperature of -4°C and humidity of 85%, the probability of frozen heave causing fastener loosening is inferred to be 82%. Among them, the knowledge graph (KG) is an important part of artificial intelligence technology. It is a structured and semantic knowledge representation method that can help computers understand and process human language. The root cause analysis method (RCA) is a structured problem-solving method used to gradually find the root cause of a problem and solve it.

[0095] Based on the predicted orbital deformation value, the confidence interval of the predicted orbital deformation value, and the root cause probability, the calculation process of the risk probability distribution of the orbit is as follows: 1. Definition of input parameters: Predicted orbital deformation value (Dpred): The predicted value of the orbital deformation in a future period (unit: millimeter), such as Dpred = 5.2mm.

[0096] Confidence interval (CI): The statistical range of the predicted deformation value, assuming a normal distribution. For example, 95%CI = [4.0, 6.4]mm (mean μ = 5.2, standard deviation σ = 0.6).

[0097] Root cause probability (P(Ri)): A set of fault root cause probabilities inferred through a Bayesian network. For example, P(R1 = fastener loosening) = 0.62, P(R2 = ballast settlement) = 0.28, P(R3 = others) = 0.10.

[0098] 2. Risk threshold setting: Safety threshold (D th ): The maximum allowable value of track deformation, set according to the track type (e.g., D th = 6.0mm). Risk level classification: Low risk: Dpred < 4.0mm; Medium risk: 4.0 ≤ Dpred < 6.0mm; High risk: Dpred ≥ 6.0mm.

[0099] 3. Calculate the basic risk probability: Based on the normal distribution of the predicted track deformation value D ∼ N(μ = 5.2, σ2 = 0.62), calculate the probability P that the deformation exceeds the threshold D th : .

[0100] Where Φ is the cumulative function of the standard normal distribution.

[0101] 4. Root cause conditional risk correction: The influence of different root causes on track deformation is different, and the risk probability needs to be adjusted: Fastener loosening (R1): Prone to cause rapid deformation, risk coefficient ; Ballast settlement (R2): The deformation develops slowly, risk coefficient ; Others (R3): Risk coefficient .

[0102] Conditional risk probability: .

[0103] 6. Output of risk probability distribution: The final risk probability distribution after root cause correction is classified by level: ; ; .

[0104] Optionally, in one embodiment, after step S50, it further includes: When receiving an alarm message, query the location coordinates of the faulty track component through the digital twin platform based on the alarm message, and generate a maintenance work order for the faulty track component through the equipment remaining service life prediction model; Send the position coordinates and the maintenance work order to a preset terminal.

[0105] In this embodiment, a geographic information system is integrated with a building information model (BIM model) for two-way association query between equipment position coordinates (WGS84) and attribute data. Therefore, when an alarm message is received, the position coordinates of the faulty track component can be queried based on the predicted value of track deformation. Then, through the equipment remaining service life prediction model, a maintenance work order for the faulty track component is generated, and the position coordinates and the maintenance work order of the faulty track component are sent to a preset terminal for maintenance personnel to maintain the faulty track component in a timely manner.

[0106] In a second aspect, an embodiment of the present invention further provides a track infrastructure security guarantee system.

[0107] In one embodiment, refer to Figure 2 , Figure 2 which is a schematic diagram of the component modules of an embodiment of the track infrastructure security guarantee system of the present invention. As Figure 2 shown, the track infrastructure security guarantee system includes: a multi-source perception module, a data fusion center, a digital twin platform, and an adaptive control module. The multi-source perception module is configured to collect original track data. The data fusion center is configured to process the original track data by using a bilateral filter and a region growing method to obtain track data containing semantic labels; extract features from the track data to construct a spatio-temporal correlation feature matrix. The digital twin decision platform is configured to perform thermal-mechanical coupling simulation based on the track data to obtain a simulation result, where the simulation result includes the track thermal expansion coefficient and the current track stress field distribution data. The adaptive control module is configured to generate an optimized control instruction and an alarm message based on the spatio-temporal correlation feature matrix and the simulation result.

[0108] Optionally, in one embodiment, the multi-source perception module includes a fiber Bragg grating sensor, a lidar, an interferometric synthetic aperture radar, a multispectral camera, and an infrared thermal imager. The original track data includes: the stress, deformation, temperature, terrain, and environmental image data of the track.

[0109] Optionally, in one embodiment, the data fusion center is specifically configured to: Use wavelet transform and Kalman filter algorithm to denoise and smooth the data of the track to obtain first track data. Use principal component analysis method and t-distributed stochastic neighbor embedding algorithm to reduce the dimension of the first track data to obtain second track data. Extract key features from the second track data through a spatio-temporal graph convolutional network to obtain feature track data; Based on the feature track data, construct a spatio-temporal graph through a graph neural network library; Generate a spatio-temporal correlation feature matrix based on the feature track data and the spatio-temporal graph.

[0110] Optionally, in one embodiment, the digital twin decision-making platform is specifically configured to: Use a semantic segmentation network to identify and classify the track data to determine the faulty components; Adjust the parameters of the faulty components through the building information model in the digital twin platform; Perform thermo-mechanical coupling simulation based on the adjusted parameters to obtain the track thermal expansion coefficient and the current track stress field distribution data.

[0111] Optionally, in one embodiment, the simulation results include the track thermal expansion coefficient and the current track stress field distribution data. The adaptive control module is specifically configured to: Obtain historical track data within a preset time period; Based on the historical track data and the simulation results, use a long sequence prediction model to obtain the track deformation prediction value and the confidence interval of the track deformation prediction value; When the track deformation prediction value is not within the confidence interval, use the objective optimization function included in the adaptive control module to optimize and solve the spatio-temporal correlation feature matrix and the simulation results to obtain an optimized control instruction; Query the correspondence table between the track deformation value and the deformation level based on the track deformation prediction value to determine the track deformation level, and obtain the corresponding warning information based on the track deformation level.

[0112] Optionally, in one embodiment, the objective optimization function is:

[0113] In the formula, \(u\) represents the set of control variables, \(J\) represents the total cost function, represents the minimum total cost corresponding to the set of control variables, represents the longitudinal impulse force when the rolling optimization time domain step of the model predictive control is \(k\), represents the wheel-rail wear index when the rolling optimization time domain step of the model predictive control is \(k\), represents the energy consumption when the rolling optimization time domain step of the model predictive control is \(k\), 、 and represent the weight coefficients, and \(k\) represents the rolling optimization time domain step of the model predictive control.

[0114] Optionally, in one embodiment, the rail infrastructure security assurance system further includes a risk probability distribution calculation module, which is configured to: Obtain historical rail data within a preset time period; Based on the historical rail data and the simulation results, through a long sequence prediction model, obtain the predicted rail deformation value and the confidence interval of the predicted rail deformation value; Construct a distributed diagnostic model based on the federated learning framework, and aggregate the distributed diagnostic model through a central server to obtain a global model; When it is detected based on the global model that the difference between the vibration frequencies and amplitudes of multiple rail components is less than a preset value, use a knowledge graph and root cause analysis method to infer the root cause probability of the failures of the multiple rail components; Based on the predicted rail deformation value, the confidence interval of the predicted rail deformation value, and the root cause probability, obtain the risk probability distribution of the rail.

[0115] Optionally, in one embodiment, the rail infrastructure security assurance system further includes an alarm module, which is configured to: When receiving an alarm message, query the position coordinates of the faulty rail component through the digital twin platform based on the alarm message, and generate a maintenance work order for the faulty rail component through the equipment remaining service life prediction model; Send the position coordinates and the maintenance work order to a preset terminal.

[0116] Wherein, the function implementation of each module in the above rail infrastructure security assurance system corresponds to each step in the above rail infrastructure security assurance method embodiment, and its function and implementation process will not be elaborated here one by one.

[0117] In a third aspect, an embodiment of the present invention further provides an electronic device, the structure of which is as Figure 3 shown, including: a memory, a processor, and the processor is used to read and execute the computer program stored in the memory to implement the foregoing rail infrastructure security assurance method.

[0118] In a fourth aspect, an embodiment of the present invention further provides a computer storage medium, in which computer executable instructions are stored, and when the computer executable instructions are executed, the foregoing rail infrastructure security assurance method is implemented.

[0119] Finally, it should be noted that in some processes described in the embodiments of the present invention, multiple operations or steps appear in a specific order. However, it should be understood that these operations or steps may not be executed in the order in which they appear in the embodiments of the present invention or may be executed in parallel. The serial numbers of the operations are only used to distinguish different operations, and the serial numbers themselves do not represent any execution order. In addition, these processes may include more or fewer operations, and these operations or steps may be executed in sequence or in parallel, and these operations or steps may be combined.

[0120] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or perform equivalent replacements for some of the technical features. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for ensuring the safety of track infrastructure, characterized in that, The method includes: Collecting original track data through a multi-source perception module; Processing the original track data using a bilateral filter and region growing method to obtain track data containing semantic labels; Extracting features from the track data through a data fusion center to construct a spatio-temporal correlation feature matrix; Based on the track data, performing a thermal-mechanical coupling simulation through a digital twin decision platform to obtain a simulation result; Based on the spatio-temporal correlation feature matrix and the simulation result, generating an optimized control instruction and an alarm message through an adaptive control module.

2. The method for ensuring the safety of track infrastructure according to claim 1, characterized in that, The multi-source perception module includes a fiber Bragg grating sensor, a lidar, an interferometric synthetic aperture radar, a multispectral camera, and an infrared thermal imager; The original track data includes: stress, deformation, temperature, terrain, and environmental image data of the track.

3. The method for ensuring the safety of track infrastructure according to claim 1, characterized in that The extracting features from the track data through a data fusion center to construct a spatio-temporal correlation feature matrix includes: Using wavelet transform and Kalman filter algorithm to denoise and smooth the track data to obtain first-track data; Using principal component analysis and t-distributed stochastic neighbor embedding algorithm to reduce the dimension of the first-track data to obtain second-track data; Performing key feature extraction on the second-track data through a spatio-temporal graph convolutional network to obtain feature track data; Based on the feature track data, constructing a spatio-temporal graph through a graph neural network library; Based on the feature track data and the spatio-temporal graph, generating a spatio-temporal correlation feature matrix.

4. The method for ensuring the safety of track infrastructure according to claim 1, characterized in that, The performing a thermal-mechanical coupling simulation through a digital twin decision platform based on the track data to obtain a simulation result includes: Using a semantic segmentation network to identify and classify the track data to determine faulty components; Adjusting the parameters of the faulty components through the building information model in the digital twin platform; Performing a thermal-mechanical coupling simulation based on the adjusted parameters to obtain the track thermal expansion coefficient and the current track stress field distribution data.

5. The method for ensuring the safety of track infrastructure according to claim 1, characterized in that, The simulation result includes the track thermal expansion coefficient and the current track stress field distribution data. The generating an optimized control instruction and an alarm message through an adaptive control module based on the spatio-temporal correlation feature matrix and the simulation result includes: Obtaining historical track data within a preset time period; Based on the historical track data and the simulation result, obtaining a track deformation prediction value and a confidence interval of the track deformation prediction value through a long sequence prediction model; When the track deformation prediction value is not within the confidence interval, optimizing and solving the spatio-temporal correlation feature matrix and the simulation result through the target optimization function included in the adaptive control module to obtain an optimized control instruction; Querying a correspondence table between track deformation values and deformation levels based on the track deformation prediction value to determine the track deformation level, and obtaining a corresponding alarm message based on the track deformation level.

6. The method for ensuring the safety of track infrastructure according to claim 5, wherein, The target optimization function is: where \(u\) represents the set of control variables, and \(J\) represents the total cost function. represents the minimum total cost corresponding to the set of control variables. represents the longitudinal impulse force when the rolling optimization time domain step of the model predictive control is \(k\). represents the wheel-rail wear index when the rolling optimization time domain step of the model predictive control is \(k\). represents the energy consumption when the rolling optimization time domain step of the model predictive control is \(k\). 、 and represent the weight coefficients, and \(k\) represents the rolling optimization time domain step of the model predictive control.

7. The method for ensuring the safety of track infrastructure according to claim 1, wherein After performing a thermal-mechanical coupling simulation through a digital twin decision platform based on the track data to obtain a simulation result, it includes: Obtaining historical track data within a preset time period; Based on the historical track data and the simulation results, obtain the predicted values of track deformation and the confidence intervals of the predicted values of track deformation through a long sequence prediction model; Construct a distributed diagnostic model based on the federated learning framework, and aggregate the distributed diagnostic model through a central server to obtain a global model; When the difference between the vibration frequencies and amplitudes of multiple track components is detected to be less than a preset value based on the global model, infer the root cause probabilities of the failures of the multiple track components using a knowledge graph and a root cause analysis method; Based on the predicted values of track deformation, the confidence intervals of the predicted values of track deformation, and the root cause probabilities, obtain the risk probability distribution of the track.

8. The method for ensuring the safety of track infrastructure according to claim 1, wherein, After generating the optimized control instructions and alarm information through the adaptive control module based on the spatio-temporal correlation feature matrix and the simulation results, it further includes: When receiving the alarm information, query the position coordinates of the faulty track components through the digital twin platform based on the alarm information, and generate a maintenance work order for the faulty track components through the equipment remaining useful life prediction model; Send the position coordinates and the maintenance work order to a preset terminal.

9. An orbit infrastructure safety guarantee system, characterized in that, The system includes: a multi-source perception module, a data fusion center, a digital twin platform, and an adaptive control module, The multi-source perception module is configured to collect original track data; The data fusion center is configured to process the original track data using a bilateral filter and a region growing method to obtain track data containing semantic labels; extract features from the track data and construct a spatio-temporal correlation feature matrix; The digital twin decision platform is configured to perform a thermal-mechanical coupling simulation based on the track data to obtain simulation results, where the simulation results include the track thermal expansion coefficient and the current track stress field distribution data; The adaptive control module is configured to generate optimized control instructions and alarm information based on the spatio-temporal correlation feature matrix and the simulation results.

10. The system according to claim 9, characterized in that, The data fusion center is specifically configured to: Use wavelet transform and Kalman filtering algorithm to denoise and smooth the data of the track to obtain first track data; Use principal component analysis and t-distributed stochastic neighbor embedding algorithm to reduce the dimension of the first track data to obtain second track data; Extract key features from the second track data through a spatio-temporal graph convolutional network to obtain feature track data; Based on the feature track data, construct a spatio-temporal graph through a graph neural network library; Generate a spatio-temporal correlation feature matrix based on the feature track data and the spatio-temporal graph.

11. The system according to claim 9, wherein The digital twin decision platform is specifically configured to: Use a semantic segmentation network to identify and classify the track data to determine faulty components; Adjust the parameters of the faulty components through the building information model in the digital twin platform; And perform a thermal-mechanical coupling simulation based on the adjusted parameters to obtain the track thermal expansion coefficient and the current track stress field distribution data.

12. The system according to claim 9, characterized in that The adaptive control module is specifically configured to: Obtain historical track data within a preset time period; Based on the historical track data and the simulation results, obtain the predicted values of track deformation and the confidence intervals of the predicted values of track deformation through a long sequence prediction model; When the predicted value of the track deformation is not within the confidence interval, an optimized control instruction is obtained by optimizing and solving the spatio-temporal correlation feature matrix and the simulation result through the target optimization function included in the adaptive control module; Based on the predicted value of the track deformation, query the corresponding relationship table between the track deformation value and the deformation level, determine the track deformation level, and obtain the corresponding warning information based on the track deformation level.

13. An electronic device, characterized in that, Including: A memory and a processor; The processor is configured to read and execute the computer program stored in the memory to implement the steps of the track infrastructure security guarantee method as described in any one of claims 1-8.

14. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, and when the computer-executable instructions are executed, the steps of the track infrastructure security guarantee method as described in any one of claims 1-8 are implemented.

Images