Railway hidden danger road section early warning system and method based on edge calculation

Abstract

The invention discloses a railway hidden danger road section early warning system and method based on edge calculation, and relates to the technical field of railway safety, and the method comprises the steps: collecting railway section information of a railway road section monitored in real time; importing the railway section information into a pre-constructed early warning identification model, and identifying a plurality of early warning type events; wherein the early warning identification model comprises a first analysis unit for processing railway section images, a second analysis unit for processing railway section environment data, and a third analysis unit for processing train operation data and determining risk grade types; a pre-configured judgment and analysis mechanism triggers corresponding color early warning after receiving the risk level type; according to the technical key points, a plurality of prompt signals are generated by accurately capturing hidden danger feature groups in railway section images, and a curve associated with a risk coefficient is also established, so that quantitative evaluation of floating objects is realized, subsequent risk prediction is facilitated, and active early warning and accurate prevention and control of hidden dangers are realized.

Description

Technical Field

[0001] The present invention relates to the field of railway safety technology, and specifically to an edge computing-based railway hazard section early warning system and method. Background Art

[0002] As train speeds and density continue to increase, timely detection and accurate handling of safety hazards along railway lines have become increasingly crucial. While railway safety monitoring technology has made some progress, it still has many shortcomings and is unable to meet the growing safety needs: On the one hand, different types of monitoring data (such as railway section images, environmental data, and train operation data) are independent of each other and lack an effective integrated analysis mechanism, making it difficult to comprehensively and accurately identify railway hazard types and determine risk levels. Existing hazard analysis methods rely on single data thresholds, and their ability to analyze trends in hazards such as track gauge changes and roadbed settlement is weak. Traditional systems only use fixed thresholds to determine whether to issue an alarm and are unable to predict hazard development trends through comprehensive analysis of historical and real-time data. This leads to delayed hazard handling and makes it difficult to meet the needs of efficient railway operation and maintenance. On the other hand, the analysis of environmental data along the railway is not comprehensive and in-depth enough. For example, in farmland and areas near cities, the number of floating objects around the railway will increase significantly. These floating objects can easily be blown up by strong winds and invade the railway safety limits, covering the contact network, and then causing the power supply equipment to short-circuit and trip, seriously threatening the safety of railway operations. Traditional systems fail to fully explore the potential correlations between data during analysis, and are unable to establish a corresponding risk assessment system, resulting in a high false alarm rate and difficulty in achieving active early warning and precise prevention and control of corresponding hidden dangers. Summary of the Invention

[0003] (1) Technical problems solved In response to the shortcomings of the existing technology, the present invention provides a railway hazard section early warning system and method based on edge computing. It adopts a deep convolutional neural network to accurately capture the hazard feature group in the railway section image to generate a number of prompt signals. It also establishes a curve associated with the risk coefficient to achieve quantitative assessment of floating objects, facilitate subsequent risk prediction, and realize active early warning and precise prevention and control of hidden dangers, solving the problems raised in the background technology.

[0004] (2) Technical solution To achieve the above objectives, the present invention is implemented through the following technical solutions: In a first aspect, the present application provides an edge computing-based railway hazard section early warning method, the method comprising: Collecting railway section information for real-time monitoring of railway sections; wherein the railway section information includes railway section images, railway section environmental data, and train operation data; Importing railway segment information into a pre-built early warning recognition model to identify several early warning events; wherein the early warning recognition model includes a first analysis unit for processing railway segment images, a second analysis unit for processing railway segment environmental data, and a third analysis unit for processing train operation data and determining risk level types; The pre-configured judgment analysis mechanism receives the risk level type and triggers the corresponding color warning.

[0005] Furthermore, the pre-built early warning identification model includes: Based on the first analysis unit processing and analyzing the railway segment image, a plurality of prompt signals are generated; Upon receiving a prompt signal, the second analysis unit is triggered to perform inference optimization on the railway section environmental data and output a number of warning type events; the inference optimization step includes: establishing a dynamic feedback loop, executing the data set incremental injection mechanism in proportion under each new loop, and reusing the historical data generated by the previous iteration in each iterative training; through real-time monitoring of the performance indicators on the training set and the validation set, the training strategy parameters are dynamically implemented based on the performance indicators; the step of adding the data set in proportion includes: constructing a function for the train operation data based on the third analysis unit and outputting the risk level type.

[0006] Furthermore, the railway segment image includes a track structure image, a roadbed state image, and a surrounding environment image. The first analysis unit for processing the railway segment image, and the first analysis unit has a built-in deep convolutional neural network architecture, includes: Preprocessing: Preprocessing of railway segment images, including at least filtering denoising, histogram equalization and data enhancement; Feature extraction: The processed data is processed using a residual module and dilated convolution, and an attention mechanism is introduced to determine feature points based on channel attention and spatial attention. The first key feature and the second key feature are extracted, and a hidden danger feature group is formed based on the first key feature. The first key feature includes at least: gauge parameters, settlement parameters, inclination parameters, regional accumulation parameters, and recognition time; the second key feature includes at least: floating object appearance parameters, including at least area and weight; Feature analysis: Based on the hidden danger feature group, a rectangular coordinate system is established to draw several hidden danger change curves, including at least a gauge change curve, a settlement change curve, an inclination change curve, and a regional expansion change curve. At the same time, standard threshold curves corresponding to the several hidden danger change curves are drawn on the rectangular coordinate system to obtain the situation where the hidden danger change curve is above the standard threshold curve and the distance between two points; Signal classification: Analyze the status of the hidden danger change curve, obtain the corresponding status cycle, and issue several early warning signals based on the status cycle; among them, the status cycle includes the initial period, development period and decline period.

[0007] Furthermore, the railway section environmental data includes wind speed, mountain height, and air pressure. Upon receiving the prompt signal, the second analysis unit is triggered, including: Obtain the second key feature and generate a risk coefficient based on the second key feature; collect and initialize wind speed, mountain height, and air pressure, perform dimensionless processing, and obtain processed indicators; The second key feature and the processed indicator are used as input information for the second analysis unit and initially trained. During the training process, a correlation curve between the risk coefficient and each indicator is constructed, and the trend change feature is extracted, and a data set corresponding to the trend change feature is obtained; based on the data set, a rule engine is built, and the data set is divided proportionally; the divided training set is used to iteratively train the second analysis unit again, and a number of warning type events are output; among which, the trend change feature represents the trend change rate.

[0008] Furthermore, the steps of building a rule engine to divide the data set into proportions include: Obtain the corresponding trend change rate based on the risk coefficient-various indicator correlation curve; Compare and analyze each trend change rate with the preset first trend threshold U1 and second trend threshold U2 to obtain the corresponding dominant conditions; based on the dominant conditions, call the third analysis unit to predict the risk level; The corresponding warning type events under each dominant condition are collected, and the data set corresponding to the warning type events includes the number of ascending sequences, the proportion of risk level types and the frequency of occurrence of warning type events. The number of ascending sequences, the proportion of risk level types and the frequency of occurrence of warning type events are combined to obtain the proportion division index.

[0009] Furthermore, the third analysis unit is called to perform risk level prediction, including: Under the condition that wind speed dominates floating objects, R (speed, pressure) is used to predict the risk level; Under the condition that the mountain undulation dominates the floating objects, the risk level is predicted using R (speed, load); Under the condition that air pressure dominates floating objects, R(load) is used to predict the risk level.

[0010] Furthermore, the third analysis unit includes: The train operation data includes at least train speed, brake pressure and carriage load, and is dimensionlessly processed. The processing results are marked as speed, pressure and load respectively, and the relationship functions of R (speed, pressure), R (speed, load) and R (load) are constructed.

[0011] Furthermore, the proportion division index is obtained, including: Set a first interval and a second interval of the proportion partitioning index, and compare the proportion partitioning index with the first interval and the second interval: when the proportion partitioning index falls into the first interval, select the low proportion of the training set for partitioning; when the proportion partitioning index falls into the second interval, select the high proportion of the training set for partitioning; wherein, the first interval is smaller than the second interval.

[0012] Furthermore, the risk level types include level 1 risk type, level 2 risk type, and level 3 risk type. The pre-configured judgment and analysis mechanism triggers corresponding color warnings after receiving the risk level type, including: If the risk level type is level three, a red alert is triggered; If the risk level type is level 2 risk type or level 1 risk type, a yellow warning will be triggered.

[0013] In a second aspect, the present application provides a railway hazard section early warning system based on edge computing, the system comprising: Information collection module: collects railway section information for real-time monitoring of railway sections; the railway section information includes railway section images, railway section environmental data, and train operation data; Event recognition module: This module imports railway segment information into a pre-built warning recognition model to identify several warning-type events. The warning recognition model includes a first analysis unit that processes railway segment images, a second analysis unit that processes railway segment environmental data, and a third analysis unit that processes train operation data and determines risk level types. The early warning prompt module uses a pre-configured judgment and analysis mechanism to trigger a corresponding color warning after receiving the risk level type.

[0014] (3) Beneficial effects The present invention provides a railway hazard section early warning system and method based on edge computing, which has the following beneficial effects: 1. The present invention processes railway segment image data through a first analysis unit, typically using a deep convolutional neural network (CNN) architecture. The CNN's multi-scale feature extraction module can extract image features at different resolutions, capturing details from the macro to the micro level. An attention mechanism module is introduced to enhance the feature response to sections with potential hazards, allowing the model to focus more closely on areas where potential hazards may exist. 2. The present invention processes railway section environmental data through a second analysis unit and analyzes wind speed, mountain height, and air pressure to achieve a risk assessment of floating objects, especially light floating objects, in the railway's external environment. In this process, the area and weight of the floating objects are combined to generate a risk coefficient. Correlation curves are then established between wind speed, mountain height, and air pressure and the risk coefficient. The dominant position is determined based on the trend changes of each correlation curve for subsequent risk prediction. This supplements the risk assessment method for floating objects in the railway's external environment and provides technical support for further improving the external environmental safety risk investigation work corresponding to railway sections. 3. The present invention processes train operation data through the third analysis unit, further refining the second analysis unit and assisting in model training. The early warning recognition model implements a closed loop of image details, environmental drivers, and operational responses, upgrading railway hazard monitoring from passive alarms to active prediction and precise disposal. This is expected to reduce train delays caused by missed hazard detection. 4. After two curve judgment analyses, the present invention can obtain the corresponding change trends of the gauge change curve, settlement change curve, inclination change curve and area expansion change curve, realize the preliminary screening of hidden dangers and trend capture, generate several prompt signals, and significantly improve the timeliness, accuracy and operation and maintenance efficiency of monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 1 is a schematic diagram showing the steps of a method for early warning of railway hidden danger sections according to an exemplary embodiment; Figure 2 The figure is a module diagram of a railway hidden danger section early warning system according to an exemplary embodiment. DETAILED DESCRIPTION

[0016] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0017] Example 1: The embodiment of the present invention provides an early warning method for railway hidden danger sections based on edge computing; Figure 1 is a flow chart showing a method for warning railway hidden danger sections according to an exemplary embodiment; Figure 1 , the method comprises the following steps: Deploy edge computing nodes in railway section hidden danger monitoring scenarios, and integrate sensor components in the edge computing nodes to collect railway section information for real-time monitoring of the railway section. The railway section information includes railway section images, railway section environmental data, and train operation data. For example: an edge computing node is set up every certain number of kilometers (for example: 5-10 kilometers) of railway sections, installed in substations or communication base stations along the line, and connected to sensor components, including but not limited to laser displacement sensors, inclination sensors, stress sensors, visual monitoring equipment, wind speed sensors, and three-dimensional laser scanners; railway section images, railway section environmental data, and train operation data are collected. Since the collected data is multi-source and heterogeneous, it needs to be standardized and aligned in time and space; standardization unifies data of different formats and units so that the model can effectively process them; time and space alignment ensures that data collected at different times and spaces are consistent in time and space dimensions, providing an accurate data basis for subsequent analysis.

[0018] Importing railway segment information into a pre-built early warning recognition model to identify a number of early warning type events, wherein the number of early warning type events includes at least a risk level type; wherein the early warning recognition model includes a first analysis unit for processing railway segment images, a second analysis unit for processing railway segment environmental data, and a third analysis unit for processing train operation data and determining the risk level type; The pre-configured decision analysis mechanism receives the risk level type and triggers the execution of the following actions: If the risk level is level 3, the edge computing node triggers a red alert; If the risk level is level 2 or level 1, the edge computing node triggers a yellow warning. Among them, the pre-built early warning identification model includes: Based on the first analysis unit processing and analyzing the railway segment image, a plurality of prompt signals are generated; Upon receiving a prompt signal, the second analysis unit is triggered to perform inference optimization on the railway section environmental data and output a number of warning type events. The inference optimization step includes: establishing a dynamic feedback loop, proportionally executing a data set incremental injection mechanism under each new loop, and reusing historical data generated by the previous iteration in each iterative training; dynamically implementing training strategy parameter adjustment based on the performance indicators by monitoring the performance indicators on the training set and validation set in real time; and the step of proportionally adding data sets includes: constructing a function based on the train operation data based on the third analysis unit and outputting the risk level type. The railway segment image includes a track structure image, a roadbed state image, and a surrounding environment image. The first analysis unit for processing the railway segment image has a built-in deep convolutional neural network architecture, including: Preprocessing: Preprocessing of railway segment images, including at least filtering denoising, histogram equalization, contrast enhancement and data enhancement; Specifically, the pixel values of the railway segment image are normalized to the range [0, 255]. Filter denoising can remove image noise by sequentially applying a Gaussian filter with a kernel size of 5×5 and a median filter with a kernel size of 3×3. Histogram equalization can be adaptively performed using the CLAHE algorithm to compensate for uneven lighting in scenes such as tunnels and strong light. Track lines are detected using the Hough transform, and the image is corrected to an orthographic perspective using a perspective transformation matrix to enhance contrast. Data augmentation includes random rotation, scaling, translation, and brightness adjustment. Feature extraction: The processed data is processed using residual modules and dilated convolutions, and the CBAM attention mechanism is introduced. Feature points are determined based on channel attention and spatial attention. A multi-scale feature fusion strategy is used to integrate shallow texture features and deep semantic features, and extract the first key feature and the second key feature. A hidden danger feature group is formed based on the first key feature. The first key feature includes at least: gauge parameters, settlement parameters, inclination parameters, regional accumulation parameters and identification time; the second key feature includes at least: appearance parameters of floating objects, including at least area and weight; Feature analysis: Based on the hidden danger feature group, a rectangular coordinate system is drawn with the measurement period as the horizontal axis and the value of the current data item as the vertical axis. The hidden danger change curve of the corresponding data item is drawn, including at least the gauge change curve, settlement change curve, inclination change curve, and regional expansion change curve. At the same time, the corresponding standard threshold curve is drawn on the rectangular coordinate system to obtain the situation where the hidden danger change curve is above the standard threshold curve and the distance between two points; Signal classification: Analyze the status of the hidden danger change curve, obtain the corresponding status cycle, and issue several warning signals based on the status cycle. The status cycle includes the initial stage, development stage, and decline stage. In addition, if the hidden danger change curve is below the standard threshold curve, it indicates no risk and no warning signal is issued. If the distance is less than the distance threshold, the hidden danger change curve status analysis includes: Primary curve analysis: Obtain the slope of all points on each hidden danger change curve. For any hidden danger change curve, establish a peak-valley value set, obtain all points on the curve with a slope of 0, and include them in the peak-valley value set for judgment analysis, including: obtaining the next closest point of the current point f0 on the curve and marking it as point f1, marking the slope of point f1 as kf1. If the slope kf1 is greater than 0, it means that the subsequent curve of the peak-valley value point is on an upward trend, and the marked point f0 is a valley point; if the slope kf1 is less than 0, it means that the subsequent curve of the peak-valley value point is on a downward trend, and the marked point f0 is a peak point; Quadratic curve analysis: After judgment, the elements in the peak-valley value set are split into the peak point set and the valley point set, any point in the peak point set is marked as Df, and any point in the valley point set is marked as Dg, and the ascending function set and the descending function set are set and obtained, including: if the slope between the current peak point Df and the next nearest valley point Dg is less than 0, it means that the curve segment is in a downward trend, and the curve segment corresponding to the curve is included in the descending function set; if the slope between the valley point Dg and the next nearest peak point Df is greater than 0, it means that the curve segment is in an upward trend, and the curve segment corresponding to the curve is included in the ascending function set; if the slope between the current peak point Df and the next nearest valley point Dg is less than or equal to 0, it means that the curve is in a stable trend, and the curve segment corresponding to the curve is included in the stable function set; After two curve determination analyses, the corresponding change trends of the gauge change curve, settlement change curve, inclination change curve, and regional expansion change curve can be obtained. In this embodiment, all hidden danger change curves are defined as a set of ascending functions. The change trends are used to define the state cycle of hidden dangers (such as gauge change, settlement change, inclination change, and regional expansion change) in the railway hidden danger section, including the initial period, development period, and decline period. If it is detected that a curve has entered a recession period, a third prompt signal is triggered; If two curves are detected to be in the development stage, the second prompt signal is triggered; If it is monitored that a curve enters the development stage, or more than two curves are in the initial stage and the slope continues to increase, the first warning signal is issued; In addition, edge computing nodes have built-in deep learning-based object detection models (such as YOLOv8 and Faster R-CNN). Track structure images are used to monitor track gauge changes. Visual monitoring equipment can be installed at key track locations (such as switch areas and bridge junctions) to collect track structure images, identify rail profiles and fastener positions in real time, and calculate track gauge changes through pixel-level comparison. Roadbed status images are used to monitor settlement and inclination changes, identify settlement cracks, and slope stability. Visual monitoring equipment (such as dual-view cameras and IMU sensors) can be deployed at key locations such as roadbed slopes and culverts to collect roadbed surface images. Edge computing nodes use semantic segmentation algorithms (such as DeepLabv3+) to identify anomalies such as settlement cracks and soil displacement. Surrounding environment images are used to monitor regional expansion changes and identify the presence of foreign object accumulation. Panoramic cameras and millimeter-wave radars are deployed within 200 meters along the railway line to monitor foreign object intrusion and accumulation in real time. Edge computing nodes use instance segmentation algorithms (such as Mask R-CNN) to identify foreign objects such as plastic bags, branches, and construction materials. The standard threshold curve is calculated and simulated based on historical data under no-warning conditions. By obtaining all the numerical points corresponding to the hidden danger change curve under any no-warning condition, the average value and standard deviation of the numerical points under no-warning conditions are determined. The average value plus 2 times the standard deviation is used as the standard threshold. By fitting the standard threshold, a standard threshold change curve similar to a straight line is formed. The above content mainly processes railway section image data through the first analysis unit, usually using a deep convolutional neural network (CNN) architecture. The CNN's multi-scale feature extraction module can extract image features at different resolutions, capturing details in the image from macro to micro. The introduction of the attention mechanism module enhances the feature response to sections with hidden dangers, allowing the model to focus more on areas where hidden dangers may exist.

[0019] The railway section environmental data includes wind speed, mountain height, and air pressure. When a prompt signal is received, the second analysis unit is triggered, including: Training preparation: Obtain the second key feature and generate a risk coefficient based on the second key feature, including normalizing the area and weight data of floating objects (e.g., mapping them to a range of 0-1) to eliminate dimensionality effects; Set the formula to obtain the risk factor: ; In the formula, fx represents the risk coefficient, b1 is the standardized area, b2 represents the standardized weight, ω1, ω2 and ω3 are all weight coefficients, and ω1, ω2 and ω3 are all greater than 0; The weight coefficients involved in the above process come from the following sources: the weight coefficients are determined using the coefficient of variation method, which assigns weights to each indicator based on the degree of variation between the current value of each evaluation indicator and the target value. If the numerical difference of a certain indicator is large, it can clearly distinguish the evaluated objects, indicating that the indicator has rich discrimination information, and thus the indicator should be given a larger weight. Conversely, if the numerical difference of each evaluated object on a certain indicator is small, then the ability of this indicator to distinguish the evaluated objects is weak, and thus the indicator should be given a smaller weight. This method directly uses the information contained in each indicator to obtain the weight of the indicator through calculation, so it is objective and not artificially assigned. In practical applications, these weights can be set equal as needed (e.g., ω1 = ω2 = ω3 = 1 / k, where k is the total number of weight coefficients, k = 3 here) to simulate the generation of risk coefficients while still retaining the interaction terms. Similarly, a data-driven approach can be considered to determine the weights, but here we still adopt the direct weight setting approach and emphasize the importance of nonlinear interactions. To simplify the explanation, we ultimately obtain a simplified, but still somewhat complex, non-weighted (or equal-weighted) evaluation formula, which can be expressed as: risk coefficient = 1 / 3 × normalized area + 1 / 3 × normalized weight + 1 / 3 × (normalized area × normalized weight). This formula combines the standardized area, standardized weight, and their interaction terms in an equally weighted manner to assess the risk level of floating debris to a railway section. It should be noted that the weight coefficient (1 / 3) here reflects equal weighting, but in practice it may need to be adjusted based on specific circumstances. It should be noted that area and weight are important factors in risk analysis. Large floating objects (such as advertising cloth or tarpaulins over 1 square meter) have a large windward area and are easily affected by wind, which can significantly increase the probability of intrusion into railway clearances. Once they come into contact with key equipment such as the contact network and pantographs, they may cause systemic failures such as circuit short circuits and equipment jams. Weight determines the object's inertia and destructive potential. Light objects (such as foam boards or plastic bags weighing less than 0.1 kilograms) are highly maneuverable in air currents and can migrate long distances with the help of rising air currents. Heavy objects (metal components or stones weighing over 1 kilogram) have limited mobility, but the kinetic energy impact generated when they fall may directly damage track structures or signal equipment. At the same time, wind speed, mountain height and air pressure are collected and initialized without dimension processing to obtain processed indicators. The significance of the analysis is that wind speed determines the speed of floating objects, which indirectly affects people's recognition and processing time. The smaller the wind speed, the smaller the risk. The undulation and height of the mountain have an important impact on the movement of floating objects, the movement path and final position of floating objects (such as: affecting the distance and path of floating objects). The more mountains there are, the smaller the risk, which is a negative indicator. Air pressure determines the movement height of floating objects. The higher the air pressure, the easier it is for floating objects to be identified. Blowing out a guardrail increases the risk, which is a positive indicator. Furthermore, human activities such as construction, waste disposal, and agricultural operations are also significant sources of floating debris. The number of construction sites affects the amount of lightweight construction waste like dust screens generated. Fewer construction sites reduce the risk, which is a positive indicator. The number of landfills affects the amount of household waste like plastic bags. Fewer landfills reduce the risk, which is a positive indicator. Plastic film from greenhouses also becomes part of floating debris when damaged or blown away. Fewer greenhouses reduce the risk, which is a positive indicator. Training phase: The second key feature and the processed indicator are used as input information for the second analysis unit, and initial training is performed. During the training process, a correlation curve between the risk coefficient and each indicator is constructed, and trend change characteristics are extracted, and a data set corresponding to the trend change characteristics is obtained. Based on the data set, a rule engine is built, and the data set is divided proportionally into a training set, a validation set, and a test set. The second analysis unit is iteratively trained again using the training set, and a number of warning type events are output. Among them, the trend change characteristic represents the trend change rate. The steps of building a rule engine and dividing the data set into proportions include: Based on the correlation curves between risk factor and various indicators, the corresponding trend change rates are obtained. The trend change rate corresponding to the correlation curve between risk factor and wind speed indicator is marked as zd1, the trend change rate corresponding to the correlation curve between risk factor and mountain relief indicator is marked as zd2, and the trend change rate corresponding to the correlation curve between risk factor and air pressure indicator is marked as zd3. Compare and analyze each trend change rate with the preset first trend threshold U1 and second trend threshold U2: When zd1>U2 or zd2≤U1 and zd3≤U1, it means that the wind speed dominates the floating objects; When zd2>U2 or zd1≤U1 and zd3≤U1, it means that the mountain undulation dominates the floating objects; When zd3>U2 or zd1≤U1 and zd2≤U1, it means that the air pressure dominates the floating objects; Among them, U1 is smaller than U2; Under different dominant conditions, the third analysis unit is called to predict the risk level; Under the condition that wind speed dominates floating objects, R (speed, pressure) is used to predict the risk level; Under the condition that the mountain undulation dominates the floating objects, the risk level is predicted using R (speed, load); Under the condition that air pressure dominates floating objects, R(load) is used to predict the risk level; At the same time, the result of risk level prediction is a probability value, and the output risk level is compared with the preset risk level interval [QJ min , QJ max Comparative Analysis: When the risk level is less than or equal to QJ min , mark the risk level as 1, edit it as a first-level character, and combine 1 and the first-level character to generate a first-level risk type; When the risk level is greater than QJ min and is smaller than QJ max , mark the risk level as 2, edit it as a secondary character, and combine 2 and the secondary character to generate a secondary risk type; When the risk level is greater than or equal to QJ max , mark the risk level as 3, edit it as level 3 characters, and combine 3 and level 3 characters to generate a level 3 risk type; By analyzing wind speed, mountain height, and air pressure, we can assess the risk of floating objects, especially light floating objects, in the railway's external environment. In this process, we combine the area and weight of the floating objects to generate a risk coefficient, and establish correlation curves between wind speed, mountain height, and air pressure and the risk coefficient. By analyzing the trend changes of each correlation curve, we can determine the dominant position for subsequent risk prediction. This supplements the risk assessment method for floating objects in the railway's external environment and provides technical support for further improving the external environmental safety risk investigation work corresponding to railway sections. The third analysis unit includes: performing dimensionless processing on the train operation data including at least train speed, brake pressure, and carriage load, and marking the processed results as speed, pressure, and load in sequence, and constructing relationship functions of R (speed, pressure), R (speed, load), and R (load); It should be noted that a mapping relationship table is set between the data. By receiving the railway section information in the current time period and predicting the risk level type of the next time period, the mapping relationship between the railway section information of the next time period and the risk level type of the next section is saved in the mapping relationship table. The level inversion is performed through the relevant function to further improve the accuracy of the prediction; Collect the corresponding warning type events under each dominant condition. The data set corresponding to the warning type events includes the number of ascending sequences, the proportion of risk level types, and the frequency of warning type events. The number of ascending sequences, the proportion of risk level types, and the frequency of warning type events are combined to obtain the proportion division index. Steps to obtain the proportional partitioning index: Normalize the number of ascending sequences, the proportion of risk level types, and the frequency of warning type events, and map the original parameters to [0, 1]. Assign the first weight to the number of ascending sequences, the second weight to the proportion of risk level types, and the third weight to the frequency of warning type events. At the same time, the entropy value of the risk level type ratio is calculated: Where E represents the entropy value, Pi represents the proportion of each risk level type, and n represents the number of risk level types. A higher entropy value indicates a more uniform distribution of risk levels and a more comprehensive coverage of the corresponding dataset. The first result is obtained by multiplying the normalized ascending sequence number by the first weight; the entropy value is divided by the logarithm of the number of risk level types. 10 (n), and multiply it by the second weight to obtain a second result; multiply the normalized frequency of warning type events by the third weight to obtain a third result; and then sum the first result, the second result, and the third result to obtain a proportion division index; Specifically, the sources of the first weight, the second weight, and the third weight include: The first weight is derived from the ability of the number of ascending sequences to depict the continuity of risk trends, which is directly related to the coverage of the risk evolution pattern in the training set. The number of ascending sequences is the number of sequences in which the quantitative hidden danger indicators show a monotonically increasing trend, reflecting the proportion of samples in the training set with continuously worsening risks. The more ascending sequences there are, the more complete the risk evolution process contained in the training set (such as the ascending sequence of 10 consecutive days of settlement data), and the more complete the model can learn to express it. Conversely, if the training set lacks ascending sequences, the model may miss trend risks. The second weight is derived from the impact of the risk level distribution on the comprehensive coverage of the training set scenarios, which determines the basis of the model's generalization ability. If the proportion of a certain risk level in the training set is unbalanced (for example, the proportion of warning-type events corresponding to the third-level risk type is very small), the model may overfit or underfit this type of risk, indicating that the training set scenario coverage is more incomplete. The third weight is derived from the characterization of the risk density of the training set by the warning frequency, which directly affects the response speed of the model to high-frequency risk scenarios. Among them, the frequency of warning type events quantifies the number of times warning type events are triggered per unit time, reflecting the proportion of high-risk density samples in the training set. The higher the frequency of warning type events, the more high-frequency risk warning samples the training set needs to contain to avoid the model from misjudging high-frequency events in actual applications.

[0020] The calculation process can be in the form of a piecewise linear function. The greater the number of ascending sequences obtained, the greater the corresponding first weight, and the smaller the number of ascending sequences, the smaller the corresponding first weight. The entropy value is used to measure the distribution balance. The lower the entropy value (the more uneven the distribution), the smaller the second weight. Conversely, the higher the entropy value, the more uniform the risk level distribution, the more comprehensive the corresponding data set coverage scenario, and the greater the second weight. The higher the frequency of the warning type event obtained, the greater the corresponding third weight, and conversely, the smaller the corresponding third weight. Set a first interval and a second interval of the proportion division index, and compare the proportion division index with the first interval and the second interval: when the proportion division index falls into the first interval, select a low proportion of the training set for division, for example, the training set, validation set, and test set are divided according to a ratio of 3:1:1; when the proportion division index falls into the second interval, select a high proportion of the training set for division, for example, the training set, validation set, and test set are divided according to a ratio of 5:1:1; wherein the first interval is smaller than the second interval; Specifically, a higher proportion division index indicates that the corresponding data set covers more comprehensive scenarios, and the data is relatively better trained. When the proportion of the training set is higher, the model is exposed to more data and can theoretically learn more detailed rules. In this embodiment, the amount of data is sufficient to support the representativeness of the test set. Although the proportion of the validation set and the test set is relatively small at this time, the above-mentioned different proportion divisions can balance the number of training samples in each risk level group to solve the problem of insufficient model training.

[0021] Example 2: The embodiment of the present invention provides a railway hidden danger section early warning system based on edge computing; Figure 2 is a schematic diagram of a module of a railway hidden danger section early warning system according to an exemplary embodiment; Figure 2 The system includes: an information collection module, an event identification module and a first-level early warning prompt module, and the information collection module, the event identification module and the first-level early warning prompt module are communicatively connected; Information collection module: collects railway section information for real-time monitoring of railway sections; the railway section information includes railway section images, railway section environmental data, and train operation data; Event recognition module: This module imports railway segment information into a pre-built warning recognition model to identify several warning-type events. The warning recognition model includes a first analysis unit that processes railway segment images, a second analysis unit that processes railway segment environmental data, and a third analysis unit that processes train operation data and determines risk level types. The early warning prompt module uses a pre-configured judgment and analysis mechanism to trigger a corresponding color warning after receiving the risk level type.

[0022] In the application, the several formulas involved are all calculated by taking their numerical values after removing the dimensions, and the formula is a formula of the most recent real situation obtained by collecting a large amount of data and performing software simulation. The formula is set by technical personnel in this field according to actual conditions.

[0023] The above embodiments can be implemented in whole or in part by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. Those skilled in the art will appreciate that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution.

[0024] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, and may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment as needed.

[0025] The above is only a specific implementation method of the present application, but the scope of protection of the present application is not limited thereto. Any technician familiar with this technical field can easily think of changes or replacements within the technical scope disclosed in this application, which should be covered by the scope of protection of the present application.

Claims

1. The railway hidden danger section early warning method based on edge computing is characterized by: The method comprises: Collecting railway section information for real-time monitoring of railway sections; wherein the railway section information includes railway section images, railway section environmental data, and train operation data; Importing railway segment information into a pre-built early warning recognition model to identify several early warning events; wherein the early warning recognition model includes a first analysis unit for processing railway segment images, a second analysis unit for processing railway segment environmental data, and a third analysis unit for processing train operation data and determining risk level types; The pre-configured judgment analysis mechanism receives the risk level type and triggers the corresponding color warning.

2. The railway hazard section early warning method based on edge computing according to claim 1 is characterized in that: The pre-built early warning identification model includes: Based on the first analysis unit processing and analyzing the railway segment image, a plurality of prompt signals are generated; Upon receiving a prompt signal, the second analysis unit is triggered to perform inference optimization on the railway section environmental data and output a number of warning type events; the inference optimization step includes: establishing a dynamic feedback loop, executing the data set incremental injection mechanism in proportion under each new loop, and reusing the historical data generated by the previous iteration in each iterative training; through real-time monitoring of the performance indicators on the training set and the validation set, the training strategy parameters are dynamically implemented based on the performance indicators; the step of adding the data set in proportion includes: constructing a function for the train operation data based on the third analysis unit and outputting the risk level type.

3. The railway hazard section early warning method based on edge computing according to claim 1 is characterized in that: The railway segment image includes a track structure image, a roadbed state image, and a surrounding environment image. The first analysis unit for processing the railway segment image has a built-in deep convolutional neural network architecture, including: Preprocessing: Preprocessing of railway segment images, including at least filtering denoising, histogram equalization and data enhancement; Feature extraction: The processed data is processed using a residual module and dilated convolution, and an attention mechanism is introduced to determine feature points based on channel attention and spatial attention. The first key feature and the second key feature are extracted, and a hidden danger feature group is formed based on the first key feature. The first key feature includes at least: gauge parameters, settlement parameters, inclination parameters, regional accumulation parameters, and recognition time; the second key feature includes at least: floating object appearance parameters, including at least area and weight; Feature analysis: Based on the hidden danger feature group, a rectangular coordinate system is established to draw several hidden danger change curves, including at least a gauge change curve, a settlement change curve, an inclination change curve, and a regional expansion change curve. At the same time, standard threshold curves corresponding to the several hidden danger change curves are drawn on the rectangular coordinate system to obtain the situation where the hidden danger change curve is above the standard threshold curve and the distance between two points; Signal classification: Analyze the status of the hidden danger change curve, obtain the corresponding status cycle, and issue several early warning signals based on the status cycle; among them, the status cycle includes the initial period, development period and decline period.

4. The railway hazard section early warning method based on edge computing according to claim 1 is characterized in that: The railway section environmental data includes wind speed, mountain height, and air pressure. Upon receiving the prompt signal, the second analysis unit is triggered, including: Obtain the second key feature and generate a risk coefficient based on the second key feature; collect and initialize wind speed, mountain height, and air pressure, perform dimensionless processing, and obtain processed indicators; The second key feature and the processed indicator are used as input information for the second analysis unit and initially trained. During the training process, a correlation curve between the risk coefficient and each indicator is constructed, and the trend change feature is extracted, and a data set corresponding to the trend change feature is obtained; based on the data set, a rule engine is built, and the data set is divided proportionally; the divided training set is used to iteratively train the second analysis unit again, and a number of warning type events are output; among which, the trend change feature represents the trend change rate.

5. The railway hidden danger section early warning method based on edge computing according to claim 4 is characterized in that: The step of building a rule engine and dividing the data set into proportions includes: Obtain the corresponding trend change rate based on the risk coefficient-various indicator correlation curve; Compare and analyze each trend change rate with the preset first trend threshold U1 and second trend threshold U2 to obtain the corresponding dominant conditions; based on the dominant conditions, call the third analysis unit to predict the risk level; The corresponding warning type events under each dominant condition are collected, and the data set corresponding to the warning type events includes the number of ascending sequences, the proportion of risk level types and the frequency of occurrence of warning type events. The number of ascending sequences, the proportion of risk level types and the frequency of occurrence of warning type events are combined to obtain the proportion division index.

6. The railway hidden danger section early warning method based on edge computing according to claim 5 is characterized in that: The calling of the third analysis unit to perform risk level prediction includes: Under the condition that wind speed dominates floating objects, R (speed, pressure) is used to predict the risk level; Under the condition that the mountain undulation dominates the floating objects, the risk level is predicted using R (speed, load); Under the condition that air pressure dominates floating objects, R(load) is used to predict the risk level.

7. The railway hazard section early warning method based on edge computing according to claim 6 is characterized in that: The third analysis unit comprises: The train operation data includes at least train speed, brake pressure and carriage load, and is dimensionlessly processed. The processing results are marked as speed, pressure and load respectively, and the relationship functions of R (speed, pressure), R (speed, load) and R (load) are constructed.

8. The railway hazard section early warning method based on edge computing according to claim 5 is characterized in that: The obtaining of the proportional division indicator includes: Set a first interval and a second interval of the proportion partitioning index, and compare the proportion partitioning index with the first interval and the second interval: when the proportion partitioning index falls into the first interval, select the low proportion of the training set for partitioning; when the proportion partitioning index falls into the second interval, select the high proportion of the training set for partitioning; wherein, the first interval is smaller than the second interval.

9. The railway hazard section early warning method based on edge computing according to claim 1 is characterized in that: The risk level types include level 1 risk type, level 2 risk type, and level 3 risk type. The pre-configured judgment and analysis mechanism triggers the corresponding color warning after receiving the risk level type, including: If the risk level type is level three, a red alert is triggered; If the risk level type is level 2 risk type or level 1 risk type, a yellow warning will be triggered.

10. The railway hidden danger section early warning system based on edge computing is characterized by: include: Information collection module: collects railway section information for real-time monitoring of railway sections; the railway section information includes railway section images, railway section environmental data, and train operation data; Event recognition module: This module imports railway segment information into a pre-built warning recognition model to identify several warning-type events. The warning recognition model includes a first analysis unit that processes railway segment images, a second analysis unit that processes railway segment environmental data, and a third analysis unit that processes train operation data and determines risk level types. The early warning prompt module uses a pre-configured judgment and analysis mechanism to trigger a corresponding color warning after receiving the risk level type.

Images