Operation monitoring method and system for intelligent rail inspection operation robot of high-speed railway

Abstract

The invention relates to the technical field of railway track inspection, in particular to an operation monitoring method and system for a high-speed railway intelligent track inspection operation robot. Track turnout state data and earthquake influence data are combined, abnormal conditions of track passing detection conditions are analyzed, and according to the analysis result, the track turnout state data and the earthquake influence data are analyzed; analyzing the execution environment threat degree of the post-earthquake track inspection task of the robot; in combination with the body operation data of the robot and the track turnout state data, the body operation control abnormal condition of the post-earthquake track inspection operation task of the robot is analyzed; based on the execution environment threat degree of the post-earthquake track inspection operation task of the robot and the body operation control abnormal condition, the track inspection operation risk of the post-earthquake track inspection operation task of the robot is evaluated; the robot emergency braking early warning is realized according to the evaluation result, so that the operation safety of the inspection robot in an earthquake destructive environment is guaranteed, and the intelligent level of the high-speed railway post-earthquake rail inspection operation is remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of railway track inspection technology, and in particular to a method and system for monitoring the operation of intelligent track inspection robots for high-speed railways. Background Technology

[0002] High-speed railways, as the arteries of modern transportation, rely on their track systems, especially the complex turnout areas, as key infrastructure for ensuring the safe, smooth, and efficient operation of trains. During long-term service, high-speed railway tracks not only experience wear and deformation due to train cyclical loads and the influence of the natural environment, but their structural integrity faces extreme threats in high-risk areas located in earthquake zones. Strong earthquakes and frequent aftershocks can cause abrupt deterioration in track geometry, jamming or failure of core turnout locking components, and may trigger rockfalls from slopes into the clearance gauge. If a rapid and accurate safety status assessment and traffic condition determination of the line, especially critical turnout areas, cannot be conducted after a disaster, it may lead to damage to inspection equipment and maintenance delays. To address the challenges of track inspection, using robots to replace traditional manual inspections has become a clear trend. In recent years, the development of intelligent inspection robots and intelligent track diagnostic systems has successfully achieved the capture of millimeter-level changes in track geometry and the automated identification of some faults, significantly improving inspection efficiency and objectivity.

[0003] However, existing technologies still have significant shortcomings when applied to track inspection in high-risk post-earthquake scenarios, particularly in turnout areas. First, current monitoring methods for track inspection robots are typically designed for conventional environments and do not deeply correlate or comprehensively analyze the geometric state data of the track turnouts collected by the robot with real-time seismic impact data. For example, when lateral track displacement occurs in the turnout area, if the cause of this displacement—whether it stems from long-term creep or a recent strong aftershock—is not considered, the instantaneous stability and subsequent risks of the structure at that location cannot be accurately assessed, leading to misjudgments of abnormal traffic conditions. Second, existing technologies also lack real-time, quantitative risk assessment of the robot's stability on post-earthquake distorted tracks. Large local deformations and obstacles on the track after an earthquake can severely affect the robot's passability, potentially causing it to slip, overturn, or collide. Traditional methods often only focus on the track state itself, neglecting the operational risks faced by the robot performing the task. This results in an inability to accurately assess the overall risk of post-earthquake track inspection operations, limiting its reliable application in truly high-risk and complex emergency scenarios. Summary of the Invention

[0004] To overcome the defects and shortcomings of existing technologies, this invention provides a method and system for monitoring the operation of an intelligent track inspection robot for high-speed railways. By simultaneously collecting track, earthquake, and robot body data during post-earthquake operations, and by constructing and applying machine learning models such as gradient boosting trees, the system quantitatively analyzes the abnormal track conditions, the degree of threat to the execution environment, and the abnormal operation of the robot body at the current work location. This allows for the assessment of the final track inspection operation risk and the implementation of emergency braking warnings, significantly improving the intelligence level and overall safety redundancy of post-earthquake track inspection operations on high-speed railways.

[0005] To achieve the above objectives, the present invention adopts the following technical solution:

[0006] In a first aspect, embodiments of the present invention provide a method for monitoring the operation of an intelligent track inspection robot for high-speed railways, comprising the following steps:

[0007] S1. During the robot's post-earthquake track inspection operation, the robot's own operation data, track switch status data, and earthquake impact data are collected simultaneously.

[0008] S2. Combining track turnout status data and earthquake impact data, analyze abnormal track access detection conditions, and based on the analysis results of abnormal track access detection conditions, analyze the degree of environmental threat to the robot's post-earthquake track inspection operation tasks.

[0009] S3. Combining the robot's own operation data and track switch status data, analyze the abnormal operation control of the robot's own operation in the post-earthquake track inspection task.

[0010] S4. Based on the analysis results of the threat level of the execution environment of the robot's post-earthquake track inspection task and the analysis results of abnormal operation control of the robot, the track inspection risk of the robot's post-earthquake track inspection task is assessed.

[0011] S5. Based on the risk assessment results of the robot's post-earthquake track inspection operation, conduct an emergency braking warning for the robot.

[0012] According to the above scheme, step S2 combines track turnout status data and seismic impact data to analyze abnormal track operation detection conditions; including the following specific steps:

[0013] S21. Extract track state parameters from the track turnout status data of the robot's operating position recorded in multiple historical post-earthquake track inspection tasks, wherein the track state parameters include: track lateral displacement value, track vertical settlement value, and turnout tightness value; extract the peak ground acceleration value and aftershock frequency of the robot's operating position in the corresponding historical track inspection cycle recorded in multiple historical post-earthquake track inspection tasks from the earthquake impact data; simultaneously, extract the traffic condition anomaly indicators of the robot's operating position recorded in the corresponding historical track inspection cycle after the completion of multiple historical post-earthquake track inspection tasks; construct a traffic condition anomaly analysis dataset based on the track lateral displacement value, track vertical settlement value, and turnout tightness value of the robot's operating position recorded in multiple historical post-earthquake track inspection tasks, as well as the peak ground acceleration value, aftershock frequency, and traffic condition anomaly indicators of the robot's operating position in the corresponding historical track inspection cycle.

[0014] S22. Statistically determine the percentage of records where each track state parameter exceeds the corresponding preset safety threshold within different peak ground acceleration (PGA) value ranges, relative to the total number of records within the current PGA value range. Define the corresponding percentages as the baseline anomaly probabilities of the corresponding track state parameters under different ground motion levels.

[0015] S23. Construct and train a traffic condition anomaly analysis model based on a logistic regression model; take the peak ground acceleration value, aftershock frequency, and the product of the track lateral displacement value and the corresponding baseline anomaly probability, the product of the track vertical settlement value and the corresponding baseline anomaly probability, and the product of the turnout tightness value and the corresponding baseline anomaly probability of each historical post-earthquake track inspection task record in the traffic condition anomaly analysis dataset as the input features of the model, and take the traffic condition anomaly indicator as the training target of the model to train the model and obtain the trained traffic condition anomaly analysis model;

[0016] S24. Real-time acquisition of peak ground acceleration (PGA), aftershock frequency, track lateral displacement, track vertical settlement, and turnout tightness at the robot's current working position; based on the PGA interval of the robot's current working position, query the baseline anomaly probability corresponding to each track state parameter; input the PGA, aftershock frequency, and the products of the corresponding track lateral displacement and the corresponding baseline anomaly probability, the corresponding track vertical settlement and the corresponding baseline anomaly probability, and the corresponding turnout tightness and the corresponding baseline anomaly probability at the robot's current working position into the trained traffic condition anomaly analysis model for analysis; use the probability values ​​output by the traffic condition anomaly analysis model as the track traffic detection anomaly status at the robot's current working position in the robot's post-earthquake track inspection task.

[0017] According to the above scheme, step S2 analyzes the threat level of the execution environment for the robot's post-earthquake track inspection task based on the analysis results of abnormal track access detection conditions; this includes the following specific steps:

[0018] S25. Extract the projected area of ​​track obstacles and the status code of the turnout locking device in the section where the robot operates from the track turnout status data of multiple historical post-earthquake track inspection task records, and correlate and extract the abnormal track passage detection conditions of the corresponding robot operating positions; at the same time, extract the environmental threat indicators of the robot operating positions from the multiple historical post-earthquake track inspection task records; and use the projected area of ​​track obstacles and the status code of the turnout locking device in the section where the robot operates from the multiple historical post-earthquake track inspection task records, the abnormal track passage detection conditions of the corresponding robot operating positions, and the environmental threat indicators as an environmental threat level analysis dataset.

[0019] S26. Construct and train an environmental threat level analysis model based on a gradient boosting decision tree model; take the abnormal track passage detection conditions of the robot operation position, the corresponding line obstacle projection area, and the turnout locking device status code of the robot operation position recorded in the environmental threat level analysis dataset as input features, and take the environmental threat sign as the output target to train the model and obtain the trained environmental threat level analysis model.

[0020] S27. Real-time acquisition of abnormal track passage detection conditions at the robot's current working position, as well as the projected area of ​​track obstacles and the status code of turnout locking devices in the section where the robot's current working position is located. Input these into the trained environmental threat level analysis model for analysis, and use the probability value output by the environmental threat level analysis model as the execution environmental threat level of the robot's current working position in the post-earthquake track inspection task.

[0021] According to the above scheme, step S3 combines the robot's own operational data and track switch status data to analyze the abnormal operation control of the robot's post-earthquake track inspection task; including the following specific steps:

[0022] S31. Extract the three-dimensional coordinates and running direction of the robot's working position from the robot's body operation data, and extract the nearest track switch position that has been passed from the robot's working position based on the three-dimensional coordinates and running direction of the robot's working position; based on the robot's working position and the nearest track switch position, collect in real time the robot's lateral displacement deviation value sequence, robot pitch attitude angle sequence, robot tilt attitude angle sequence, and drive motor current fluctuation value sequence on the track segment between the robot's working position and the nearest track switch position during the robot's operation; at the same time, obtain the track curvature and track slope of the track segment between the robot's working position and the nearest track switch position from the track switch status data;

[0023] S32. Based on the robot's lateral displacement deviation value sequence, robot pitch attitude angle sequence, robot roll attitude angle sequence, and drive motor current fluctuation value sequence, calculate the degree of variation of robot's lateral displacement, robot pitch attitude angle, robot roll attitude angle, and drive motor current fluctuation on the track segment between the robot's working position and the nearest past track switch position.

[0024] S33. In multiple historical post-earthquake track inspection tasks, obtain the track curvature and track slope of the track segment between any work position and the corresponding nearest passed track switch position for each time the robot passes through any work position, as well as the robot's lateral displacement variation, robot pitch angle variation, robot tilt angle variation, and robot drive motor current fluctuation variation obtained statistically on the corresponding track segment; at the same time, obtain the body operation abnormality flags of the robot when passing through the corresponding track segment in the corresponding historical post-earthquake track inspection tasks.

[0025] S34. The track curvature, track slope, robot lateral displacement variation, robot pitch angle variation, robot tilt angle variation, robot drive motor current fluctuation variation, and robot body operation anomaly flags obtained from the historical post-earthquake track inspection tasks on the corresponding track segments are used as anomaly prediction datasets. The anomaly prediction datasets are then randomly divided into anomaly prediction training set and anomaly prediction validation set.

[0026] According to the above scheme, step S3 combines the robot's own operational data and track switch status data to analyze the abnormal operation control of the robot's post-earthquake track inspection task; it also includes the following specific steps:

[0027] S35. Construct a gradient boosting tree model. Based on the anomaly prediction training set, use the track curvature, track slope, robot lateral displacement variation, robot pitch angle variation, robot roll angle variation, and robot drive motor current fluctuation variation obtained from historical post-earthquake track inspection tasks on the corresponding track segments as input features of the gradient boosting tree model. Use the robot's body operation anomaly flags through the corresponding track segments as the model's output target. Train the gradient boosting tree model to obtain the initial anomaly prediction model. Validate the initial anomaly prediction model using the anomaly prediction validation set. Use the initial anomaly prediction model whose accuracy on the anomaly prediction validation set reaches the preset model accuracy threshold as the body operation anomaly analysis model.

[0028] S36. In real time, acquire the track curvature and track slope on the track segment between the robot's current working position and the corresponding nearest passed track switch position, as well as the robot's lateral displacement variation, robot pitch angle variation, robot roll angle variation, and robot drive motor current fluctuation variation obtained statistically on the corresponding track segment. Input these values ​​into the body operation anomaly analysis model and output the predicted probability value as the body operation control anomaly of the robot's current working position in the robot's post-earthquake track inspection task.

[0029] According to the above scheme, step S4 assesses the track inspection risk of the robot's post-earthquake track inspection task based on the analysis results of the threat level of the execution environment and the analysis results of abnormal operation control of the robot itself; including the following specific steps:

[0030] S41. Extract the threat level of the execution environment and abnormal operation control of the robot at its current working position during the robot's post-earthquake track inspection task;

[0031] S42. The weighted sum of the threat level of the execution environment and the abnormal situation of the robot's operation and control at the current working position in the post-earthquake track inspection task is used to obtain the track inspection operation risk of the robot at the current working position in the post-earthquake track inspection task.

[0032] According to the above scheme, step S5 involves issuing an emergency braking warning for the robot based on the risk assessment results of the robot's post-earthquake track inspection tasks; specifically including:

[0033] S51. Obtain the risk assessment results of the robot's current working position in the post-earthquake track inspection task;

[0034] S52. Preset trajectory operation risk threshold. When the risk assessment result of the robot's current working position in the post-earthquake track inspection task is greater than the trajectory operation risk threshold, the robot will perform an emergency braking warning. When the risk assessment result of the robot's current working position in the post-earthquake track inspection task is less than or equal to the trajectory operation risk threshold, the high-speed railway track will continue to be inspected.

[0035] Secondly, embodiments of the present invention also provide a high-speed railway intelligent track inspection robot operation monitoring system, including:

[0036] The data acquisition module is used to simultaneously collect the robot's own operation data, track switch status data, and earthquake impact data during the robot's post-earthquake track inspection operation.

[0037] The execution environment monitoring module is used to analyze abnormal conditions of track access detection by combining track turnout status data and earthquake impact data, and to analyze the degree of environmental threat to the robot's post-earthquake track inspection operation based on the analysis results of abnormal conditions of track access detection.

[0038] The control anomaly detection module is used to analyze abnormal control situations of the robot's body operation during post-earthquake track inspection tasks by combining the robot's body operation data and track switch status data.

[0039] The track inspection operation monitoring module is used to assess the track inspection operation risk of the robot after the earthquake based on the analysis results of the threat level of the execution environment of the robot's track inspection operation task and the analysis results of abnormal operation control of the robot itself.

[0040] The braking warning module is used to provide emergency braking warnings for the robot based on the risk assessment results of the robot's post-earthquake track inspection tasks.

[0041] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0042] 1. This invention achieves probabilistic and quantitative assessment of abnormal conditions in traffic detection by correlating and analyzing post-earthquake track turnout status data with earthquake impact data, thereby improving the scientificity and accuracy of initial safety assessments for emergency inspections.

[0043] 2. This invention constructs an anomaly analysis model for robot operation control by combining robot body operation data and track status data. It can predict the robot's own slippage, instability and other risks in real time, ensuring the safety of the inspection robot and the integrity of the equipment in earthquake-damaged environments.

[0044] 3. This invention integrates the degree of threat in the execution environment with abnormal operation conditions of the track inspection task to conduct a comprehensive risk assessment, which significantly improves the intelligence level and overall safety redundancy of post-earthquake track inspection operations on high-speed railways. Attached Figure Description

[0045] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0046] Figure 1 This is a schematic diagram of the overall process of the operation monitoring method for the intelligent track inspection robot of high-speed railway according to the present invention;

[0047] Figure 2 This is a flowchart of step S2 in the high-speed railway intelligent track inspection robot operation monitoring method of the present invention;

[0048] Figure 3 This is a schematic diagram of the operation monitoring system for the intelligent track inspection robot of high-speed railway according to the present invention;

[0049] Figure 4 This is a flowchart of step S3 in the high-speed railway intelligent track inspection robot operation monitoring method of the present invention. Detailed Implementation

[0050] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments of the present invention and the specific features in the embodiments are detailed descriptions of the technical solution of the present invention, rather than limitations thereof. In the absence of conflict, the embodiments of the present invention and the technical features in the embodiments can be combined with each other.

[0051] Example 1

[0052] like Figure 1 As shown, this embodiment provides a method for monitoring the operation of an intelligent track inspection robot for high-speed railways, which specifically includes the following steps:

[0053] S1. During the robot's post-earthquake track inspection operation, the robot's own operation data, track switch status data, and earthquake impact data are collected simultaneously.

[0054] S2. Combining track turnout status data and earthquake impact data, analyze abnormal track access detection conditions, and based on the analysis results of abnormal track access detection conditions, analyze the degree of environmental threat to the robot's post-earthquake track inspection operation tasks.

[0055] S3. Combining the robot's own operation data and track switch status data, analyze the abnormal operation control of the robot's own operation in the post-earthquake track inspection task.

[0056] S4. Based on the analysis results of the threat level of the execution environment of the robot's post-earthquake track inspection task and the analysis results of abnormal operation control of the robot, the track inspection risk of the robot's post-earthquake track inspection task is assessed.

[0057] S5. Based on the risk assessment results of the robot's post-earthquake track inspection operation, conduct an emergency braking warning for the robot.

[0058] In this embodiment, as Figure 2 As shown, step S2 combines track turnout status data and seismic impact data to analyze abnormal track access detection conditions; this includes the following specific steps:

[0059] S21. Extract track state parameters from the track turnout status data of the robot's operating position recorded in multiple historical post-earthquake track inspection tasks, wherein the track state parameters include: track lateral displacement value, track vertical settlement value, and turnout tightness value; extract the peak ground acceleration value and aftershock frequency of the robot's operating position within the corresponding historical track inspection cycle recorded in the earthquake impact data of multiple historical post-earthquake track inspection tasks; simultaneously, extract the traffic condition anomaly indicators of the robot's operating position within the corresponding historical track inspection cycle recorded after the completion of multiple historical post-earthquake track inspection tasks; wherein the traffic condition anomaly indicators are based on the corresponding historical track inspection cycle. The safety status of a high-speed railway is determined by whether the lateral displacement, vertical settlement, and turnout gap values ​​exceed their respective preset safety thresholds. If any track state parameter exceeds the threshold, it is marked as 1; otherwise, it is marked as 0. Based on the lateral displacement, vertical settlement, and turnout gap values ​​recorded from multiple historical post-earthquake track inspection tasks at the robot's operating location, along with the peak ground acceleration, aftershock frequency, and abnormal traffic condition indicators at the corresponding robot operating location within the historical inspection cycle, a traffic condition anomaly analysis dataset is constructed. It should be noted that the core of this embodiment is to construct an analysis dataset that can accurately reflect the abnormal traffic conditions in the track turnout area after an earthquake. Unlike ordinary railway track section inspections, which typically only focus on the long-term, slow changes in track geometry, the safety status of the high-speed railway turnout area after an earthquake is a dynamic, high-risk variable continuously affected by seismic activity. To quantify this risk, this embodiment integrates multi-source heterogeneous data. Specifically, from the massive track turnout status data of historical track inspection tasks, the lateral displacement, vertical settlement, and turnout gap values ​​recorded at each positioning point for each task are extracted. These parameters are direct physical representations of track structure integrity and turnout functionality. Their measurement relies on high-precision sensors installed on robots or fixed monitoring points, such as total stations, hydrostatic levels, and laser rangefinders, to acquire millimeter-level change data in real time and objectively. Simultaneously, peak ground acceleration (PGA) values ​​and aftershock frequencies at the corresponding times and locations are extracted from associated seismic impact data. PGA data originates from a network of high-sensitivity seismic accelerometers deployed along the railway line, while aftershock frequencies are obtained by accessing real-time data streams from national or regional seismic monitoring stations.In this embodiment, the most crucial step is generating anomaly indicators for traffic conditions. These indicators are not based on subjective judgment but rather on strict engineering safety regulations: after each historical task, the system automatically backtracks the process to determine whether the three track state parameters at each location point exceed the preset safety thresholds defined by railway engineering regulations. In this embodiment, the preset safety thresholds are differentiated based on the different design standards of high-speed railways and conventional railways, as well as the higher sensitivity of turnout areas compared to conventional track sections. For example, for high-speed railway turnouts, the threshold for tight clearance is much smaller than that for conventional tracks. This embodiment further aligns and correlates the track state parameters, corresponding seismic motion parameters, and automatically determined anomaly indicators of all locations in all historical tasks in a spatiotemporal manner, forming a traffic condition anomaly analysis dataset required for machine learning. This correlates sporadic, high-intensity seismic disturbances with the microscopic response of the track structure, laying a solid and traceable data foundation for subsequent probabilistic risk prediction.

[0060] S22. Statistically calculate the percentage of records where each track state parameter exceeds its corresponding preset safety threshold within different peak ground acceleration (PGA) intervals, relative to the total number of records within the current PGA interval. Define these percentages as the baseline anomaly probabilities for the corresponding track state parameters under different seismic ground motion levels. This embodiment is used to quantify the prior probability, i.e., the baseline anomaly probability, of anomalies occurring for each track state parameter under different earthquake intensities. In conventional railway track inspection, parameter exceeding limits is often treated as an independent event for analysis. However, in post-earthquake scenarios, seismic ground motion intensity is the core dominant factor driving track state deterioration, and its impact is non-linear. In order to objectively reveal this correlation, this embodiment performs a refined statistical analysis on the dataset constructed in S21. In the specific operation of this embodiment, the range of peak ground acceleration values ​​in the entire dataset is first divided into several continuous intervals (e.g., 0-0.05g, 0.05g-0.1g, 0.1g-0.2g, etc.). Then, for each interval, all data records falling within this interval are statistically analyzed. For each record, it is checked whether its track lateral displacement value, track vertical settlement value, and turnout tightness value exceed their respective preset safety thresholds. Then, the percentage of records showing an excessive value for a specific parameter (such as track lateral displacement) within a specific peak ground acceleration (PGA) range is calculated as the baseline anomaly probability for that parameter at that specific PGA level. In this embodiment, the mathematical meaning of the baseline anomaly probability is the empirical probability of a specific type of track anomaly occurring when encountering a certain intensity of ground motion in historical data. For example, this embodiment shows that when the PGA value is in the range of 0.1g-0.2g, the baseline probability of the turnout tightness exceeding the limit is as high as 40%, while the probability of the lateral displacement exceeding the limit is only 15%. Based on the above, the function of this embodiment is as follows:

[0061] First, this embodiment realizes the transformation of risk assessment from a binary judgment of whether or not a threshold is exceeded to a continuous probability thinking of what probability of exceeding the threshold, which is more in line with the uncertain nature of post-earthquake risks.

[0062] Second, this embodiment assigns a dynamic weight to each track state parameter. This weight (i.e., the baseline anomaly probability) changes in real time with the intensity of the ground motion, enabling the model to understand that under strong earthquakes, the failure of the close-fitting gap is a risk that needs more attention than conventional settlement, while under weak earthquakes, the opposite is true.

[0063] S23. Construct and train a traffic condition anomaly analysis model based on a logistic regression model; use the peak ground acceleration value, aftershock frequency, and the product of the track lateral displacement value and the corresponding baseline anomaly probability, the product of the track vertical settlement value and the corresponding baseline anomaly probability, and the product of the turnout tightness value and the corresponding baseline anomaly probability of each historical post-earthquake track inspection task record in the traffic condition anomaly analysis dataset as the input features of the model, and use the traffic condition anomaly indicator as the training target of the model for training, to obtain a trained traffic condition anomaly analysis model; the core of this embodiment is to construct and train a traffic condition anomaly analysis model that can integrate the seismic environment and track status and output anomaly probabilities; considering the complex nonlinear relationship between input features and output targets, and the need to output easily interpretable probability values, this embodiment selects a logistic regression model as the basis, and enhances its input through feature engineering; the specific process of model construction is as follows:

[0064] From the dataset constructed from S21, the peak ground acceleration (PGA) value (continuous value) and aftershock frequency (continuous value) of each data record are used as basic environmental characteristics;

[0065] The measured lateral displacement, vertical settlement, and turnout gap values ​​in each record are multiplied by the baseline anomaly probability of each parameter corresponding to the interval of the peak ground acceleration value of that record, calculated in step S22. The significance of using multiplication in this embodiment is that it uses a risk-weighted value. For example, a smaller displacement measured under a strong earthquake (high peak ground acceleration value) may have a larger weighted characteristic value than a larger displacement measured under a weak earthquake (low peak ground acceleration value) because the baseline probability of displacement deterioration is higher under a strong earthquake. This is equivalent to allowing the model to have a built-in understanding of parameter risk weights under different seismic environments during learning.

[0066] These three weighted features, together with the two basic environmental features, form a five-dimensional input feature vector; the training target of the model is the passage condition anomaly flag (0 or 1) defined in S21.

[0067] Using 70% of the dataset as the training set, the logistic regression model is trained. In this embodiment, the logistic regression model maps the linear combination of input features to a continuous value between 0 and 1 using a log-odds function. This value represents the predicted probability that a sample is judged as an anomaly (the anomaly flag for passage conditions is 1). The model learns the probability mapping relationship between input features and passage condition anomaly flags through algorithms such as maximum likelihood estimation, and finally obtains a set of optimal model coefficients so that the predicted probability output by the model is as close as possible to the true binary flag distribution.

[0068] After training, the model performance is evaluated using the remaining 30% of the validation set, its prediction accuracy is calculated, and the optimal probability classification threshold is determined using the ROC curve. Model training is complete when the model's prediction accuracy on the validation set reaches a preset standard (90% by default in this embodiment).

[0069] S24. Real-time acquisition of peak ground acceleration (PGA), aftershock frequency, track lateral displacement, track vertical settlement, and turnout tightness at the robot's current working position; based on the PGA interval of the robot's current working position, query the baseline anomaly probability corresponding to each track state parameter; input the PGA, aftershock frequency, and the products of the corresponding track lateral displacement and the corresponding baseline anomaly probability, the corresponding track vertical settlement and the corresponding baseline anomaly probability, and the corresponding turnout tightness and the corresponding baseline anomaly probability at the robot's current working position into the trained traffic condition anomaly analysis model for analysis; use the probability values ​​output by the traffic condition anomaly analysis model as the track traffic detection anomaly status at the robot's current working position in the robot's post-earthquake track inspection task.

[0070] In this embodiment, step S2 analyzes the threat level of the execution environment for the robot's post-earthquake track inspection task based on the analysis results of abnormal track passage detection conditions; this includes the following specific steps:

[0071] S25. Extract the projected area of ​​track obstacles and the status code of the turnout locking device in the section where the robot operates from multiple historical post-earthquake track inspection task records from the track turnout status data, and correlate them to extract the abnormal track passage detection conditions at the corresponding robot operating positions; simultaneously, extract the environmental threat flags for the robot operating positions from multiple historical post-earthquake track inspection task records, wherein the environmental threat flags are determined based on whether the robot's operation was interrupted or canceled at the corresponding operating position due to environmental problems in the corresponding historical post-earthquake track inspection task records; if so, the environmental threat flag is 1, otherwise the environmental threat flag is 0; and then extract the environmental threat flags from multiple historical post-earthquake track inspection task records. The data collected from post-earthquake track inspection tasks, including the projected area of ​​track obstacles in the section where the robot was operating, the status codes of the turnout locking devices, abnormal track access conditions at the corresponding robot operating locations, and environmental threat markers, forms the dataset for analyzing the degree of environmental threat. It should be noted that for post-earthquake high-speed railway turnout areas, abnormal access conditions (track-related issues) are only one part of the environmental threat. These abnormal conditions may, along with other environmental factors such as falling rocks from slopes, foreign objects intruding into the clearance gauge, and the failure of the locking status of critical turnout equipment, constitute a multi-dimensional and complex threat environment. This differs from the situation observed during ordinary track section inspections. The focus is often significantly different between track geometry and a few obstacles. Specifically, from the track switch status data recorded in historical track inspection tasks, the projected area of ​​track obstacles and the status code of the switch locking device associated with each work position for each task are extracted. The obstacle projection area is calculated by scanning point cloud data with a LiDAR or binocular vision sensor on the robot, transforming the coordinates, and projecting it onto the track plane. It is an objective, continuous value. The calculation method for the obstacle projection area is existing technology and will not be elaborated here. The status code of the switch locking device comes from sensor monitoring of equipment such as switch machines and contactless inspection devices (e.g., current, voltage, etc.). The records represent rod displacement, and are output as discrete codes after passing through a state diagnosis algorithm (e.g., 00-normal, 01-alarm, 10-failure). Further, this embodiment precisely correlates these records with the abnormal track passage detection conditions calculated in step S24 for the same historical task and the same work location. Each record is also marked with an environmental threat flag, determined based on objective operation logs: whether, in this historical task, the robot was forced to interrupt the preset inspection process or cancel subsequent tasks at this work location due to environmental issues (such as encountering large obstacles that cannot be overcome, or a switch lock failure causing the path to be locked). If so, the flag is "1" (environmental threat exists); otherwise, it is "0". Based on the above, this embodiment achieves the fusion of multi-level environmental information.It links the probability of anomalies reflecting the basic state of the track, the area of ​​obstacles reflecting external risks, the lockout status code reflecting the functional state of the equipment, and the resulting consequences of work interruption. This allows subsequent models to learn that not all track anomalies will lead to work interruption (for example, slight subsidence may be overcome by the robot), but when track anomalies are combined with large obstacles or equipment failure, the threat level will increase dramatically.

[0072] S26. Construct and train an environmental threat level analysis model based on a gradient boosting decision tree model; using the abnormal track access detection conditions, corresponding track obstacle projection areas, and turnout locking device status codes of the robot operation positions recorded in multiple historical post-earthquake track inspection tasks in the environmental threat level analysis dataset as input features, and using environmental threat indicators as output targets for model training, a trained environmental threat level analysis model is obtained; this embodiment uses a gradient boosting decision tree model, which, by integrating multiple weak decision trees, can automatically capture high-order interactions and nonlinear relationships between features, and is insensitive to feature dimensions, making it very suitable for processing mixed data in this step; the specific training process of the environmental threat level analysis model in this embodiment can be set as follows:

[0073] A1. Obtain the environmental threat level analysis set. This set consists of three-dimensional data points composed of feature values ​​from multiple historical post-earthquake track inspection operation records for each operation location, corresponding to abnormal traffic conditions, projected area of ​​track obstacles, and status codes of turnout locking devices. It also consists of environmental threat flags (1 indicates a threat causing operation interruption, 0 indicates normal) for each data point, determined based on historical task logs. The environmental threat level analysis set is randomly divided into an environmental threat level training set and an environmental threat level verification set.

[0074] A2. This embodiment sets the three core construction parameters of the gradient boosting decision tree model; among them, the total number of decision trees is 100 by default, the maximum growth depth limit of a single decision tree is 5 by default, and the learning rate (shrinkage step size) of the model is 0.1 by default.

[0075] A3. For each iteration of the model to build a new decision tree, perform the following operations:

[0076] Calculate the predicted value of the current ensemble model for each sample in the training set for analyzing the degree of environmental threat (the initial round uses a basic constant value), and calculate its pseudo residual (i.e., the difference between the true label and the current predicted probability) based on the log loss function; use this pseudo residual as the new target to be fitted in the current round;

[0077] From all samples in the training set for analyzing the degree of environmental threat, a subset is randomly selected to construct the current tree. During the tree construction process, for any node that needs to be split, its current node data is the sample subset assigned to that node, consisting of the aforementioned three-dimensional feature vectors and corresponding pseudo-residual target values. From the three feature dimensions, by traversing all possible split points, the feature and splitting value that minimizes the sum of squared pseudo-residuals of samples in the two child nodes after splitting are selected. Based on this splitting value, the current node data is divided into two subsets and assigned to two child nodes respectively. The process of selecting the best feature and splitting value is recursively executed on the generated child nodes to continue splitting until any of the following conditions are met, at which point the splitting stops and leaf nodes are formed: the current node depth reaches 5, the number of data points in the current node is less than a preset minimum, or the feature value splitting benefit of all samples in the current node data is lower than a certain threshold.

[0078] A4. Repeat step A3 until a gradient boosting decision tree model is generated by integrating 100 decision trees through an additive model, which serves as the trained environmental threat level analysis model. This model receives a new 3D data point input, traverses each decision tree, moves from the root node to a leaf node according to the judgment rules of each tree, and sums the weight values ​​of the corresponding leaf nodes of all trees. Finally, it is converted into a probability value between 0 and 1 through the Sigmoid function as the execution environmental threat level of the job location.

[0079] S27. Real-time acquisition of abnormal track passage detection conditions at the robot's current working position, as well as the projected area of ​​track obstacles and the status code of turnout locking devices in the section where the robot's current working position is located. Input these into the trained environmental threat level analysis model for analysis, and use the probability value output by the environmental threat level analysis model as the execution environmental threat level of the robot's current working position in the post-earthquake track inspection task.

[0080] In this embodiment, as Figure 4 As shown, step S3 combines the robot's own operational data and track switch status data to analyze abnormal situations in the robot's post-earthquake track inspection task; this includes the following specific steps:

[0081] S31. Extract the three-dimensional coordinates and running direction of the robot's working position from the robot's body operation data, and extract the nearest track switch position that has been passed from the robot's working position based on the three-dimensional coordinates and running direction of the robot's working position; based on the robot's working position and the nearest track switch position, collect in real time the robot's lateral displacement deviation value sequence, robot pitch attitude angle sequence, robot tilt attitude angle sequence, and drive motor current fluctuation value sequence on the track segment between the robot's working position and the nearest track switch position during the robot's operation; at the same time, obtain the track curvature and track slope of the track segment between the robot's working position and the nearest track switch position from the track switch status data;

[0082] S32. Based on the robot's lateral displacement deviation sequence, robot pitch angle sequence, robot roll angle sequence, and drive motor current fluctuation sequence, calculate the variability of the robot's lateral displacement, pitch angle, roll angle, and drive motor current on the track segment between the robot's working position and the nearest passed switch position. The core of this embodiment is to perform feature engineering on the original time series data to extract statistical features that characterize the robot's operational stability on a specific track segment. On irregular tracks after an earthquake, the variability (variability) of the robot's operating state is more revealing of potential risks than the average value. For example, on a normal smooth track, the lateral displacement deviation may fluctuate slightly around a small mean; however, in the damaged switch area, even if the average displacement is small, there may be severe, intermittent large shaking, which is a precursor to derailment. Therefore, this method abandons simple mean analysis and introduces the degree of variability as the core feature; specifically including:

[0083] Calculate the mean and standard deviation of the robot's lateral displacement deviation value sequence, and use the ratio of the standard deviation to the mean of the robot's lateral displacement deviation value sequence as the degree of variation of the robot's lateral displacement on the track segment between the robot's working position and the nearest past track switch position.

[0084] Calculate the mean and standard deviation of the robot pitch attitude angle sequence, and use the ratio of the standard deviation to the mean of the robot pitch attitude angle sequence as the degree of variation of the robot pitch attitude angle on the track segment between the robot's working position and the nearest past track switch position.

[0085] Calculate the mean and standard deviation of the robot tilt attitude angle sequence, and use the ratio of the standard deviation to the mean of the robot tilt attitude angle sequence as the degree of variation of the robot tilt attitude angle on the track segment between the robot's working position and the nearest past track switch position;

[0086] Calculate the mean and standard deviation of the drive motor current fluctuation value sequence, and use the ratio of the standard deviation to the mean of the drive motor current fluctuation value sequence as the degree of variation of the robot drive motor current fluctuation on the track segment between the robot's working position and the nearest past track switch position.

[0087] S33. Obtain the track curvature and slope of the track segment between any work position and the nearest passed track switch position for each historical post-earthquake track inspection task, as well as the lateral displacement variation, pitch angle variation, roll angle variation, and drive motor current fluctuation variation of the robot statistically obtained on the corresponding track segment; simultaneously, obtain the body operation anomaly flags of the robot when passing through the corresponding track segment in the corresponding historical post-earthquake track inspection task; wherein, the body operation anomaly flag indicates whether the robot experiences drive slippage, attitude loss, or positioning anomaly when passing through the corresponding track segment; if so, the body operation anomaly flag is 1, otherwise the body operation anomaly flag is 0; specifically In this case, the determination of the abnormal operation flag of the robot body relies on the objective system logs and fault codes recorded during the task. In this embodiment, the system queries historical data to see if the robot's status monitoring system immediately generated a clear fault alarm after the robot passed through this specific track segment during historical post-earthquake track inspection tasks, such as "drive wheel speed difference exceeds limit" (drive slippage), "attitude angle exceeds safety threshold" (attitude loss) or "positioning information jump / loss" (positioning anomaly). These alarms are generated by the robot's underlying control system based on real-time logical judgment of sensor data and are a digital record of objective facts. If any of the above-mentioned abnormal faults occur, the abnormal operation flag of the robot body in this historical record will be recorded as "1" (abnormal), otherwise it will be recorded as "0" (normal).

[0088] S34. The track curvature, track slope, robot lateral displacement variation, robot pitch angle variation, robot tilt angle variation, robot drive motor current fluctuation variation, and robot body operation anomaly flags obtained from the historical post-earthquake track inspection tasks on the corresponding track segments are used as anomaly prediction datasets. The anomaly prediction datasets are randomly divided into anomaly prediction training set and anomaly prediction verification set.

[0089] In this embodiment, step S3 combines the robot's body operation data and track switch status data to analyze abnormal situations in the robot's body operation control during post-earthquake track inspection tasks; it also includes the following specific steps:

[0090] S35. Construct a gradient boosting tree model. Based on the anomaly prediction training set, use the track curvature, track slope, robot lateral displacement variation, robot pitch angle variation, robot roll angle variation, and robot drive motor current fluctuation variation obtained statistically from historical post-earthquake track inspection tasks on the corresponding track segments as input features of the gradient boosting tree model. Use the robot's body operation anomaly flags on the corresponding track segments as the model's output target to train the gradient boosting tree model and obtain the initial anomaly prediction model. Validate the initial anomaly prediction model using the anomaly prediction validation set. The initial anomaly prediction model whose accuracy on the anomaly prediction validation set reaches the preset model accuracy threshold is used as the body operation anomaly analysis model. In this embodiment, the specific training process of the body operation anomaly analysis model can be set as follows:

[0091] B1. Obtain the anomaly prediction training set. This set consists of six-dimensional data points composed of the feature values ​​of six characteristic dimensions obtained from the corresponding track segments in historical post-earthquake inspection tasks: track curvature, track slope, robot lateral displacement variation, robot pitch attitude angle variation, robot roll attitude angle variation, and drive motor current fluctuation variation. It also consists of the body operation anomaly flag (1 indicates anomaly, 0 indicates normal) corresponding to each data point, which is determined based on whether there are objective fault logs such as drive slippage, attitude loss of control, or positioning anomaly recorded after passing through the segment.

[0092] B2. In this embodiment, the three core construction parameters of the gradient boosting tree model are set; among them, the total number of decision trees (i.e., the number of iterations) is 100 by default, the maximum growth depth limit of a single decision tree is 6 by default, and the learning rate of the model is 0.1 by default.

[0093] B3. For each iteration of the model to build a new decision tree, perform the following operations:

[0094] Calculate the predicted value of the current ensemble model for each sample in the anomaly prediction training set (in the initial round, the predicted value of all samples is initialized to the log odds of the proportion of negative samples in the dataset), and calculate its pseudo residual (i.e. the difference between the true label and the predicted probability of the current model) based on the log loss function; use this pseudo residual as the new target to be fitted in the current round.

[0095] Using all or a random portion of the training samples, during the tree construction process, for any node that needs to be split, from the six feature dimensions, by traversing all possible split points, the feature and specific split value that maximizes the reduction of the sum of squared pseudo-residuals of samples in the two child nodes after the split are selected. Based on this split value, the data of the current node is divided into two subsets and assigned to the two child nodes respectively. The above process of selecting the best feature and split value is recursively executed on the generated child nodes to continue splitting until any of the following conditions are met, at which point the splitting stops and leaf nodes are formed: the depth of the current node reaches 6, the number of data points in the current node is less than 10, or the pseudo-residuals of all samples in the data of the current node are completely identical.

[0096] B4. Repeat step B3 until a gradient boosting tree model is generated by integrating 100 decision trees through a forward step-by-step addition model. This model serves as the trained ontology operation anomaly analysis model. It is used to receive a new six-dimensional data point input, and by having the data point traverse each decision tree in the forest, it moves from the root node to a leaf node according to the judgment rules of each tree. The weight contribution values ​​of the corresponding leaf nodes of all trees are accumulated and summed. Finally, the Sigmoid function is used to convert it into a probability value between 0 and 1 as an ontology operation control anomaly.

[0097] S36. In real time, acquire the track curvature and track slope on the track segment between the robot's current working position and the corresponding nearest passed track switch position, as well as the robot's lateral displacement variation, robot pitch angle variation, robot roll angle variation, and robot drive motor current fluctuation variation obtained statistically on the corresponding track segment. Input these values ​​into the body operation anomaly analysis model and output the predicted probability value as the body operation control anomaly of the robot's current working position in the robot's post-earthquake track inspection task.

[0098] In this embodiment, step S4 assesses the track inspection risk of the robot's post-earthquake track inspection task based on the analysis results of the threat level of the execution environment and the analysis results of abnormal operation control of the robot itself; it includes the following specific steps:

[0099] S41. Extract the threat level of the execution environment and abnormal operation control of the robot at its current working position during the robot's post-earthquake track inspection task;

[0100] S42. The weighted sum of the threat level of the execution environment and the abnormal situation of the robot's operation and control at the current working position in the post-earthquake track inspection task is used to obtain the track inspection operation risk of the robot at the current working position in the post-earthquake track inspection task.

[0101] In this embodiment, step S5 involves issuing an emergency braking warning for the robot based on the risk assessment results of the robot's post-earthquake track inspection task; specifically, it includes:

[0102] S51. Obtain the risk assessment results of the robot's current working position in the post-earthquake track inspection task;

[0103] S52. A preset trajectory operation risk threshold is established. When the risk assessment result of the robot's current working position in the post-earthquake track inspection task is greater than the trajectory operation risk threshold, an emergency braking warning is issued for the robot. When the risk assessment result of the robot's current working position in the post-earthquake track inspection task is less than or equal to the trajectory operation risk threshold, the inspection of the high-speed railway track continues. It should be noted that the weights and thresholds in this embodiment are determined as follows: The robot's body operation data, track turnout status data, and earthquake impact data are obtained from 3000 sets of robot post-earthquake track inspection tasks to calculate the track inspection risk of the robot's post-earthquake track inspection task. Simultaneously, the judgment results of whether the 3000 sets of robot post-earthquake track inspection tasks have been completed are obtained. The track inspection risk of the 3000 sets of robot post-earthquake track inspection tasks and the corresponding judgment results of whether the robot post-earthquake track inspection tasks have been completed are imported into fitting software for fitting, and the corresponding weights and thresholds that meet the highest coefficient of determination are output.

[0104] Example 2

[0105] like Figure 3 As shown, this embodiment provides an intelligent track inspection robot operation monitoring system for high-speed railways, including:

[0106] The data acquisition module is used to simultaneously collect the robot's own operation data, track switch status data, and earthquake impact data during the robot's post-earthquake track inspection operation.

[0107] The execution environment monitoring module is used to analyze abnormal conditions of track access detection by combining track turnout status data and earthquake impact data, and to analyze the degree of environmental threat to the robot's post-earthquake track inspection operation based on the analysis results of abnormal conditions of track access detection.

[0108] The control anomaly detection module is used to analyze abnormal control situations of the robot's body operation during post-earthquake track inspection tasks by combining the robot's body operation data and track switch status data.

[0109] The track inspection operation monitoring module is used to assess the track inspection operation risk of the robot after the earthquake based on the analysis results of the threat level of the execution environment of the robot's track inspection operation task and the analysis results of abnormal operation control of the robot itself.

[0110] The braking warning module is used to provide emergency braking warnings for the robot based on the risk assessment results of the robot's post-earthquake track inspection tasks.

[0111] The parameters and steps for implementing the corresponding functions of each unit module in the high-speed railway intelligent track inspection robot operation monitoring system of the present invention can be referred to the parameters and steps in the embodiments of the high-speed railway intelligent track inspection robot operation monitoring method above, and will not be repeated here.

[0112] The various embodiments in this invention are described in a progressive manner. Similar or identical parts between embodiments can be referred to interchangeably. Each embodiment focuses on its differences from other embodiments. In particular, the embodiments for IoT devices and media are relatively simple in description because they are fundamentally similar to the method embodiments; relevant parts can be referred to the descriptions in the method embodiments.

[0113] The systems, media, and methods provided in the embodiments of the present invention are in one-to-one correspondence. Therefore, the systems and media also have similar beneficial technical effects as their corresponding methods. Since the beneficial technical effects of the methods have been described in detail above, the beneficial technical effects of the systems and media will not be repeated here.

[0114] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention 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.

[0115] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, as well as 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 illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0116] 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.

[0117] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0118] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0119] Computer-readable media include both permanent and non-permanent, removable and non-removable media that can store information by any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0120] It should also be noted that the terms include, encompass, or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Without further limitations, the inclusion of an element by a statement that includes a… does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.

[0121] The above are merely embodiments of the present invention and are not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of the present invention should be included within the scope of the claims of the present invention.

Claims

1. A method for monitoring the operation of an intelligent track inspection robot for high-speed railways, characterized in that, Includes the following steps: S1. During the robot's post-earthquake track inspection operation, the robot's own operation data, track switch status data, and earthquake impact data are collected simultaneously. S2. Combining track turnout status data and earthquake impact data, analyze abnormal track access detection conditions, and based on the analysis results of abnormal track access detection conditions, analyze the degree of environmental threat to the robot's post-earthquake track inspection tasks. S3. Combining the robot's own operation data and track switch status data, analyze the abnormal operation control of the robot's own operation in the post-earthquake track inspection task. S4. Based on the analysis results of the threat level of the execution environment of the robot's post-earthquake track inspection task and the analysis results of abnormal operation control of the robot, the track inspection risk of the robot's post-earthquake track inspection task is assessed. S5. Based on the risk assessment results of the robot's post-earthquake track inspection operation, conduct an emergency braking warning for the robot.

2. The method for monitoring the operation of a high-speed railway intelligent track inspection robot according to claim 1, characterized in that, In step S2, the abnormal conditions of track passage detection are analyzed by combining track turnout status data and earthquake impact data. The specific steps include the following: S21. Extract track state parameters from the track turnout status data of the robot's operating position recorded in multiple historical post-earthquake track inspection tasks, wherein the track state parameters include: track lateral displacement value, track vertical settlement value, and turnout tightness value; extract the peak ground acceleration value and aftershock frequency of the robot's operating position in the corresponding historical track inspection cycle recorded in multiple historical post-earthquake track inspection tasks from the earthquake impact data; simultaneously, extract the traffic condition anomaly indicators of the robot's operating position recorded in the corresponding historical track inspection cycle after the completion of multiple historical post-earthquake track inspection tasks; construct a traffic condition anomaly analysis dataset based on the track lateral displacement value, track vertical settlement value, and turnout tightness value of the robot's operating position recorded in multiple historical post-earthquake track inspection tasks, as well as the peak ground acceleration value, aftershock frequency, and traffic condition anomaly indicators of the robot's operating position in the corresponding historical track inspection cycle. S22. Statistically determine the percentage of records where each track state parameter exceeds the corresponding preset safety threshold within different peak ground acceleration (PGA) value ranges, relative to the total number of records within the current PGA value range. Define the corresponding percentages as the baseline anomaly probabilities of the corresponding track state parameters under different ground motion levels. S23. Construct and train a traffic condition anomaly analysis model based on a logistic regression model; take the peak ground acceleration value, aftershock frequency, and the product of the track lateral displacement value and the corresponding baseline anomaly probability, the product of the track vertical settlement value and the corresponding baseline anomaly probability, and the product of the turnout tightness value and the corresponding baseline anomaly probability of each historical post-earthquake track inspection task record in the traffic condition anomaly analysis dataset as the input features of the model, and take the traffic condition anomaly indicator as the training target of the model to train the model and obtain the trained traffic condition anomaly analysis model; S24. Real-time acquisition of peak ground acceleration (PGA), aftershock frequency, track lateral displacement, track vertical settlement, and turnout tightness at the robot's current working position; based on the PGA interval of the robot's current working position, query the baseline anomaly probability corresponding to each track state parameter; input the PGA, aftershock frequency, and the products of the corresponding track lateral displacement and the corresponding baseline anomaly probability, the corresponding track vertical settlement and the corresponding baseline anomaly probability, and the corresponding turnout tightness and the corresponding baseline anomaly probability at the robot's current working position into the trained traffic condition anomaly analysis model for analysis; use the probability values ​​output by the traffic condition anomaly analysis model as the track traffic detection anomaly status at the robot's current working position in the robot's post-earthquake track inspection task.

3. The method for monitoring the operation of a high-speed railway intelligent track inspection robot according to claim 2, characterized in that, In step S2, the degree of threat to the execution environment of the robot's post-earthquake track inspection task is analyzed based on the analysis results of abnormal track passage detection conditions. The specific steps include the following: S25. Extract the projected area of ​​track obstacles and the status code of the turnout locking device in the section where the robot operates from the track turnout status data of multiple historical post-earthquake track inspection task records, and correlate and extract the abnormal track passage detection conditions of the corresponding robot operating positions; at the same time, extract the environmental threat indicators of the robot operating positions from the multiple historical post-earthquake track inspection task records; and use the projected area of ​​track obstacles and the status code of the turnout locking device in the section where the robot operates from the multiple historical post-earthquake track inspection task records, the abnormal track passage detection conditions of the corresponding robot operating positions, and the environmental threat indicators as an environmental threat level analysis dataset. S26. Construct and train an environmental threat level analysis model based on a gradient boosting decision tree model; take the abnormal track passage detection conditions of the robot operation position, the corresponding line obstacle projection area, and the turnout locking device status code of the robot operation position recorded in the environmental threat level analysis dataset as input features, and take the environmental threat sign as the output target to train the model and obtain the trained environmental threat level analysis model. S27. Real-time acquisition of abnormal track passage detection conditions at the robot's current working position, as well as the projected area of ​​track obstacles and the status code of turnout locking devices in the section where the robot's current working position is located. Input these into the trained environmental threat level analysis model for analysis, and use the probability value output by the environmental threat level analysis model as the execution environmental threat level of the robot's current working position in the post-earthquake track inspection task.

4. The method for monitoring the operation of a high-speed railway intelligent track inspection robot according to claim 3, characterized in that, In step S3, the abnormal operation control of the robot body is analyzed by combining the robot's body operation data and the track turnout status data. The specific steps include the following: S31. Extract the three-dimensional coordinates and running direction of the robot's working position from the robot's body operation data, and extract the nearest track switch position that has been passed from the robot's working position based on the three-dimensional coordinates and running direction of the robot's working position; based on the robot's working position and the nearest track switch position, collect in real time the robot's lateral displacement deviation value sequence, robot pitch attitude angle sequence, robot tilt attitude angle sequence, and drive motor current fluctuation value sequence on the track segment between the robot's working position and the nearest track switch position during the robot's operation; at the same time, obtain the track curvature and track slope of the track segment between the robot's working position and the nearest track switch position from the track switch status data; S32. Based on the robot's lateral displacement deviation value sequence, robot pitch attitude angle sequence, robot roll attitude angle sequence, and drive motor current fluctuation value sequence, calculate the degree of variation of robot's lateral displacement, robot pitch attitude angle, robot roll attitude angle, and drive motor current fluctuation on the track segment between the robot's working position and the nearest past track switch position. S33. In multiple historical post-earthquake track inspection tasks, obtain the track curvature and track slope of the track segment between any work position and the corresponding nearest passed track switch position for each time the robot passes through any work position, as well as the robot's lateral displacement variation, robot pitch angle variation, robot tilt angle variation, and robot drive motor current fluctuation variation obtained statistically on the corresponding track segment; at the same time, obtain the body operation abnormality flags of the robot when passing through the corresponding track segment in the corresponding historical post-earthquake track inspection tasks. S34. The track curvature, track slope, robot lateral displacement variation, robot pitch angle variation, robot tilt angle variation, robot drive motor current fluctuation variation, and robot body operation anomaly flags obtained from the historical post-earthquake track inspection tasks on the corresponding track segments are used as anomaly prediction datasets. The anomaly prediction datasets are then randomly divided into anomaly prediction training set and anomaly prediction validation set.

5. The method for monitoring the operation of a high-speed railway intelligent track inspection robot according to claim 4, characterized in that, Step S3, which combines the robot's own operational data and track switch status data, analyzes abnormalities in the robot's operation and control during post-earthquake track inspection tasks; it also includes the following specific steps: S35. Construct a gradient boosting tree model. Based on the anomaly prediction training set, use the track curvature, track slope, robot lateral displacement variation, robot pitch angle variation, robot roll angle variation, and robot drive motor current fluctuation variation obtained from historical post-earthquake track inspection tasks on the corresponding track segments as input features of the gradient boosting tree model. Use the robot's body operation anomaly flags through the corresponding track segments as the model's output target. Train the gradient boosting tree model to obtain the initial anomaly prediction model. Validate the initial anomaly prediction model using the anomaly prediction validation set. Use the initial anomaly prediction model whose accuracy on the anomaly prediction validation set reaches the preset model accuracy threshold as the body operation anomaly analysis model. S36. In real time, acquire the track curvature and track slope on the track segment between the robot's current working position and the corresponding nearest passed track switch position, as well as the robot's lateral displacement variation, robot pitch angle variation, robot roll angle variation, and robot drive motor current fluctuation variation obtained statistically on the corresponding track segment. Input these values ​​into the body operation anomaly analysis model and output the predicted probability value as the body operation control anomaly of the robot's current working position in the robot's post-earthquake track inspection task.

6. The method for monitoring the operation of a high-speed railway intelligent track inspection robot according to claim 5, characterized in that, In step S4, the risk of the robot's post-earthquake track inspection task is assessed based on the analysis results of the threat level of the execution environment of the robot's post-earthquake track inspection task and the analysis results of abnormal operation control of the robot itself. The specific steps include the following: S41. Extract the threat level of the execution environment and abnormal operation control of the robot at its current working position during the robot's post-earthquake track inspection task; S42. The weighted sum of the threat level of the execution environment and the abnormal situation of the robot's operation and control at the current working position in the post-earthquake track inspection task is used to obtain the track inspection operation risk of the robot at the current working position in the post-earthquake track inspection task.

7. The method for monitoring the operation of a high-speed railway intelligent track inspection robot according to claim 4, characterized in that, In step S5, an emergency braking warning for the robot is issued based on the risk assessment results of the robot's post-earthquake track inspection task. Specifically, it includes: S51. Obtain the risk assessment results of the robot's current working position in the post-earthquake track inspection task; S52. Preset trajectory operation risk threshold. When the risk assessment result of the robot's current working position in the post-earthquake track inspection task is greater than the trajectory operation risk threshold, the robot will perform an emergency braking warning. When the risk assessment result of the robot's current working position in the post-earthquake track inspection task is less than or equal to the trajectory operation risk threshold, the high-speed railway track will continue to be inspected.

8. A high-speed railway intelligent track inspection robot operation monitoring system, implemented based on the high-speed railway intelligent track inspection robot operation monitoring method according to any one of claims 1-7, characterized in that, The system includes: The data acquisition module is used to simultaneously collect the robot's own operation data, track switch status data, and earthquake impact data during the robot's post-earthquake track inspection operation. The execution environment monitoring module is used to analyze abnormal conditions of track access detection by combining track turnout status data and earthquake impact data, and to analyze the degree of environmental threat to the robot's post-earthquake track inspection operation based on the analysis results of abnormal conditions of track access detection. The control anomaly detection module is used to analyze abnormal control situations of the robot's body operation during post-earthquake track inspection tasks by combining the robot's body operation data and track switch status data. The track inspection operation monitoring module is used to assess the track inspection operation risk of the robot after the earthquake based on the analysis results of the threat level of the execution environment of the robot's track inspection operation task and the analysis results of abnormal operation control of the robot itself. The braking warning module is used to provide emergency braking warnings for the robot based on the risk assessment results of the robot's post-earthquake track inspection tasks.

Images