A train overspeed protection method and system based on line curve radius identification

Abstract

The present invention relates to a train overspeed protection method and system based on line curve radius identification. The method is for a train in the second stage and includes: obtaining a curved tunnel image and a laser signal, and generating a distance feature; inputting the distance feature into a constructed line radius identification model to obtain the curve radius of the curved tunnel wall ahead; inputting the curve radius into a constructed overspeed protection model to obtain a deceleration mileage; determining a train speed limit plan, and communicating the train speed limit plan to the operator before the train reaches the deceleration mileage to complete the maximum braking operation. The system includes: an acquisition module, a distance feature generation module, a line radius identification module, a speed limit mileage calculation module, and a real-time protection module. The system generates a distance feature using the obtained curved tunnel image and laser signal, obtains the curve radius of the curved tunnel wall ahead, calculates the deceleration mileage, and communicates the train speed limit plan to the operator before the train reaches the deceleration mileage.

Description

Technical Field

[0001] The present invention relates to the field of rail transportation technology, and in particular to a train overspeed protection method and system based on line curve radius identification. Background Art

[0002] Urban rail transit is typically designed for a speed of 160 km / h, double the speed of a typical subway. This poses greater challenges to train stability and places higher demands on the line. To verify the line's safety margin and ensure safe operation under adverse conditions, progressive speed increases are required. Relevant regulations stipulate that the maximum operating speed must reach 110% of the design speed (176 km / h). Testing at this extreme state presents certain safety risks.

[0003] During normal train operation, the train has an overspeed protection system (ATP). This system monitors the train's speed in real time by receiving track circuit information and line data provided by the balise. When the train exceeds the preset safe speed, the braking system is automatically activated to prevent overspeeding. This ensures that the train maintains a safe speed range under all operating conditions, preventing accidents. However, before operation begins, the ATP system is in the commissioning phase, and when testing at speeds exceeding the design speed, the ATP system must be isolated.

[0004] When ATP is isolated, manual driving mode is used. The cab interface cannot display information such as train speed, permitted speed, target speed, and target distance. Speed control is based solely on reading mileposts within the tunnel and combining them with line information. Mileposts require dedicated personnel to read and raise, and tunnels are characterized by relatively low light levels and the presence of dust and moisture, making them difficult to discern at high speeds. If speeding is not preemptively reduced when entering speed-restricted sections, excessive train speeding can easily lead to safety risks such as derailment on tight curves.

[0005] CN106627670A relates to a train protection system and method based on laser detection. The train protection system includes: a laser sensor module for emitting laser pulses in front of the current train and receiving echo pulse information; an electronic map positioning module for obtaining the train's current position, whether it is on a curve, and information about the curvature of the curve when it is on a curve; an information acquisition and storage module for real-time acquisition and storage of the echo pulse information; and a data processing and analysis module for obtaining ranging data and the current train's current position and / or curve curvature information to adjust the current train's running speed. This invention only requires the installation of corresponding equipment at the front of the train, eliminating the need to install ground equipment at each station and relying on other trains being in normal working order. It occupies little onboard space and is simple and convenient to install.

[0006] During the step-by-step speed test, the 110% design speed condition requires isolation of the ATP. The only way to calculate mileage is by integrating the vehicle speed using the onboard speed sensor. This conversion process inevitably introduces errors. Over long distances, this error accumulates, leading to significant discrepancies between the calculated and actual mileage, making it difficult to accurately determine the train's current location. Existing train protection systems and methods struggle to address this technical issue, significantly impacting the safety and accuracy of the test.

[0007] In addition, on the one hand, there are differences in understanding among those skilled in the art; on the other hand, the applicant studied a large number of documents and patents when making the present invention, but due to space limitations, not all details and contents are listed in detail. However, this does not mean that the present invention does not have the characteristics of these prior arts. On the contrary, the present invention already has all the characteristics of the prior art, and the applicant reserves the right to add relevant prior art to the background technology. Summary of the Invention

[0008] In response to the shortcomings of the existing technology, the present invention provides a train overspeed protection method and system based on line curve radius identification, which is particularly suitable for the period of gradual speed increase of newly built lines before operation, allowing trains (especially in tunnels) to release vehicle speed limits through the isolation ATP system, thereby conducting operation tests at a speed of 110% of the design speed.

[0009] The present invention discloses a train overspeed protection method based on line curve radius identification, which is aimed at a train in the second stage and includes the following steps:

[0010] Obtain curved tunnel images and laser signals, and generate distance features;

[0011] Input the distance feature into the constructed line radius recognition model to obtain the curve radius of the curved tunnel wall ahead;

[0012] The curve radius is input into the constructed overspeed protection model to obtain the deceleration mileage;

[0013] Determine the train speed limit plan and communicate it to the operator before the train reaches the deceleration mileage to complete the maximum braking operation.

[0014] Preferably, the second stage is the deceleration protection stage after reaching 110% of the design speed and before entering the speed limit section in the tunnel. By comparing the speed of each passing time node of the train with the target speed, the accuracy of the current overspeed protection model in protecting the train in the second stage is evaluated.

[0015] In the second phase, the train overspeed protection method of the present invention introduces images and laser signals to generate distance features, which are then input into a line radius recognition model to determine the curve radius of the curved tunnel wall ahead. This process enables the system to accurately determine the train's position and the geometric characteristics of the line ahead. After obtaining the curve radius, it is input into the constructed overspeed protection model to calculate the deceleration distance, ensuring that the train can safely decelerate before entering the speed-restricted section. Finally, by communicating the train speed limit plan to the operator in advance, maximum braking operation can be achieved. This method not only addresses the limitations and safety hazards of traditional reliance on manual reading of mileage markers, but also improves train safety through intelligent means. In addition, by comparing the speed at each passing time node with the target speed, the accuracy of the overspeed protection model can be continuously evaluated, thus ensuring the long-term effectiveness of the system. This evaluation mechanism based on real-time data feedback helps to promptly identify and correct potential problems, further enhancing the safety of train operations.

[0016] According to a preferred embodiment, the distance feature can be determined by the following sub-steps:

[0017] Obtain image information and laser information of the track in front of the train cab;

[0018] Based on the image information, determine the position of the left and right rails in the curved tunnel ahead;

[0019] Based on the laser information, determine the exact distance between the left and right rails in the curved tunnel ahead;

[0020] Calculate the straight-line distances of the left and right rails from the starting position of the speed limit section to the tunnel wall as the corresponding distance features.

[0021] To more accurately determine distance signatures, the present invention progressively refines the steps involved in determining distance signatures. This process ensures the accuracy of the resulting distance signatures. The combined use of image and laser information provides multi-dimensional data support for the system, making distance signature generation more reliable. Specifically, image information is used to locate rail positions, while laser information provides high-precision distance measurements. The two complement each other, collectively enhancing the reliability of distance signatures. Each step in this process is closely linked, ultimately achieving precise perception of the environment ahead of the train, providing a solid foundation for subsequent overspeed protection measures.

[0022] According to a preferred embodiment, when determining the position of the left and right rails in the curved tunnel ahead based on image information, a convolutional neural network can be used to perform feature extraction on the image information acquired by the camera to identify the intersection position of the left and right rails in the tunnel where the left and right rails connect to the gentle curve section ahead along the extension direction of the straight section, and use the intersection position as the starting point of the speed limit section to calculate the transition distance between the train and the intersection position when the camera acquires the image information.

[0023] When determining the position of the left and right rails in the curved tunnel ahead based on the image information, the present invention uses a convolutional neural network (CNN) to perform feature extraction on the image information acquired by the camera. Through CNN, the intersection position of the left and right rails in the tunnel where the left and right rails connect to the gentle curve section ahead along the extension direction of the straight section can be effectively identified from complex image data, and the intersection position is used as the starting point of the speed limit section. This approach not only improves the accuracy of intersection position identification, but also greatly shortens the processing time. The powerful feature extraction capability of CNN enables it to work stably even under complex and changing lighting conditions, thereby enhancing the robustness of the system. Moreover, by using the intersection position as the starting point of the speed limit section, the transition distance between the train and the position can be accurately calculated, providing key parameters for the subsequent distance feature generation. This step greatly improves the adaptability and response speed of the system, enabling it to better cope with various challenges that may arise in the rail transit environment, while making important contributions to improving the safety and efficiency of train operation.

[0024] According to a preferred embodiment, after obtaining the distance feature, the curve radius of the curved tunnel wall ahead is obtained based on geometric calculation and support vector machine algorithm, and the curve radius of the curved tunnel wall ahead is preliminarily calculated using the first extension distance, the second extension distance and the layout spacing between the first laser sensor and the second laser sensor when obtaining the curve radius, wherein the first laser sensor and the second laser sensor are respectively arranged on both sides in front of the train cab and are respectively located above the right and left rails, and the distance from the first laser sensor and the second laser sensor to the curved tunnel wall ahead along the extension direction of their respective corresponding rails minus the transition distance to obtain the first extension distance and the second extension distance respectively.

[0025] After obtaining the distance features, the present invention combines geometric calculations with a support vector machine (SVM) algorithm to determine the curve radius of the upcoming curved tunnel wall. This method leverages the advantages of geometric principles and machine learning, ensuring rapid response for the initial estimate while improving the accuracy of the final prediction. Specifically, a first laser sensor and a second laser sensor, positioned on either side of the train cab, above the left and right rail tracks, accurately measure the distances from the respective rail extensions to the upcoming curved tunnel wall—the first extension distance and the second extension distance. Subtracting the transition distance from these two distances provides the key data for calculating the curve radius of the upcoming curved tunnel wall. Geometric calculations serve as a bridge in this process, converting the measured distances into the parameters required for the curve radius. The SVM algorithm, with its powerful classification capabilities and insensitivity to noise and outliers, maintains stable performance in complex and changing environments. The optimized SVM model can quickly respond to real-time data, meeting the requirements for instant identification under high-speed train conditions. Therefore, by combining these two methods, the present invention achieves precise identification of railway line curve parameters, significantly improving the reliability and safety of the train overspeed protection system.

[0026] According to a preferred embodiment, the deceleration mileage can be determined by the following sub-steps:

[0027] Obtain the starting mileage and target speed limit of the speed limit section through the design document;

[0028] Determine whether speed limit protection is needed by comparing the current speed with the target speed limit and calculate the braking distance if necessary;

[0029] Calculate safe deceleration distance based on preset safety factor;

[0030] The deceleration mileage is calculated based on the obtained starting mileage of the speed limit section and the safe deceleration distance, wherein the train's travel direction is taken into consideration when calculating the deceleration mileage.

[0031] In the process of determining the deceleration mileage, the present invention first obtains the starting mileage and the established target speed limit of each speed limit section through the design document, laying the foundation for subsequent calculations; secondly, by comparing the current speed with the target speed limit, it determines whether speed limit protection is needed, ensuring that protective measures are only activated when necessary, avoiding unnecessary deceleration operations; next, the braking distance is calculated according to the actual situation, taking into account the current speed of the train and the expected target speed limit, ensuring the safety of the deceleration process; finally, the safe deceleration distance is calculated based on the preset safety factor, and the deceleration mileage is calculated accordingly. Taking into account the direction of travel of the train, this meticulous calculation method ensures that whether the train is traveling in the direction of increasing or decreasing mileage, the best time to start deceleration can be accurately found. The entire process reflects the refined management of the train's driving status, which not only improves driving safety, but also optimizes train operation efficiency, reduces unnecessary stops and passenger waiting time, and thus improves the overall service quality.

[0032] According to a preferred embodiment, the operator learns about the train speed limit plan by receiving a braking prompt, wherein the braking prompt is conveyed to the operator via voice so that the operator who receives the braking prompt can complete the maximum braking operation before the train reaches the deceleration mileage.

[0033] The present invention informs the operator of the train speed limit plan by receiving braking prompts, especially in the form of voice communication to the operator. This method ensures that the operator can receive the necessary instructions before the train reaches the deceleration mileage, so as to complete the maximum braking operation. Compared with traditional visual prompts, voice prompts are more immediate and understandable. Especially in emergency situations or when the operator cannot immediately view the screen, voice prompts can quickly attract attention and reduce reaction time. In addition, because voice prompts can directly convey complex information, such as the current line status, the radius of the curve ahead, etc., it can also help operators make more informed decisions. This interactive mode not only enhances the effect of human-machine collaboration, but also improves the user experience of the system. More importantly, in this way, it can be ensured that under any circumstances, the train can be adjusted according to the predetermined speed limit plan, thereby effectively preventing the risk of speeding and maintaining the safety of the operator.

[0034] According to a preferred embodiment, for a train in the first stage, whether to perform an overspeeding test is determined by calculating the acceleration distance and judging the relationship between the acceleration distance and the length of each straight section, wherein the acceleration mileage is calculated based on the obtained starting mileage of the straight section and the acceleration distance, and the train's travel direction is taken into account when calculating the acceleration mileage, so that the operator can receive an overspeeding prompt before the train reaches the acceleration mileage.

[0035] For trains in the first stage, the present invention determines whether to conduct an overspeed test by calculating the acceleration distance and determining the relationship between the acceleration distance and the length of each straight section. This strategy fully considers the actual operating conditions of the train and the characteristics of the line, ensuring the feasibility and safety of the overspeed test. Specifically, based on the relationship between the length of the straight section and the calculated acceleration distance, it can be scientifically determined whether an overspeed test at 110% of the design speed is appropriate. If the acceleration distance is less than half the length of the straight section, the test is allowed; otherwise, it is not conducted. This logical judgment not only avoids dangerous situations caused by insufficient space but also ensures the effectiveness of the test. In addition, the acceleration distance is calculated based on the starting distance of the straight section and the acceleration distance, and the train's direction of travel is taken into account, ensuring that the operator receives an overspeed warning before the train reaches the acceleration distance. The addition of this step makes the entire acceleration process more controllable and reduces the impact of uncertain factors.

[0036] According to a preferred embodiment, the accuracy calculated at all time nodes is weighted averaged to obtain the accuracy during the entire driving process, wherein if the evaluation result is lower than the set accuracy threshold, the model needs to be adjusted or recalibrated.

[0037] By taking a weighted average of the accuracies calculated at all time nodes, the present invention provides a method for comprehensively evaluating the accuracy of the overspeed protection model during the entire driving process. This method not only reflects the performance at a single time node, but also comprehensively considers the dynamic changes during the entire driving cycle, thereby obtaining a more objective and comprehensive evaluation result. If the evaluation result is lower than the set accuracy threshold, it indicates that the model needs to be adjusted or recalibrated. The importance of this step lies in that it establishes a closed-loop feedback mechanism to ensure that the overspeed protection model can continuously optimize its own performance according to actual conditions. Through continuous monitoring and timely adjustment, the system can always maintain the best working state, effectively reducing the risks caused by error accumulation or changes in other external factors. In addition, this evaluation method also provides data support for the iterative upgrade of the model, promoting the continuous advancement of technology. Overall, the introduction of accuracy evaluation and real-time correction mechanism has greatly improved the reliability and stability of the train overspeed protection system, providing a strong guarantee for the safety testing of rail transit.

[0038] The present invention also discloses a train overspeeding protection system based on line curve radius identification, which includes: an acquisition module for obtaining curved tunnel images and laser signals; a distance feature generation module for generating distance features based on data information obtained by the acquisition module; a line radius identification module for inputting the distance features into a constructed line radius identification model to obtain the curve radius of the curved tunnel wall ahead; a speed limit mileage calculation module for inputting the curve radius into a constructed overspeeding protection model to obtain a deceleration mileage; and a real-time protection module for conveying the train speed limit plan to the operator before the train reaches the deceleration mileage.

[0039] According to a preferred embodiment, the train overspeed protection system also includes: an acceleration mileage calculation module, which is used to determine whether to perform an overspeeding test and determine the acceleration mileage when the overspeeding test is performed, so that when the acceleration mileage calculation module determines that the train is performing an overspeeding test, the real-time protection module can convey the train acceleration plan to the operator before the train reaches the acceleration mileage. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 It is a flow chart of the steps of the train overspeed protection method provided by the present invention;

[0041] Figure 2 This is a flow chart of sub-steps of step S110 of the train overspeed protection method provided by the present invention;

[0042] Figure 3 This is a schematic diagram of the layout of the composite sensor device provided by the present invention in the cab;

[0043] Figure 4 It is a schematic diagram of the line radius identification model provided by the present invention performing a preliminary estimation of the curve radius of the curved tunnel wall ahead;

[0044] Figure 5 This is a sub-step flow chart of step S130 of the train overspeed protection method provided by the present invention;

[0045] Figure 6 It is a schematic diagram of a design file of a certain circuit provided by the present invention;

[0046] Figure 7 This is a hardware connection diagram of the train overspeed protection system provided by the present invention.

[0047] Reference Signs List

[0048] 100: acquisition module; 110: camera; 120: laser sensor; 121: first laser sensor; 122: second laser sensor; 200: distance feature generation module; 300: route radius recognition module; 400: speed limit mileage calculation module; 500: real-time protection module. DETAILED DESCRIPTION

[0049] The following is a detailed description with reference to the accompanying drawings.

[0050] Example 1

[0051] The present invention discloses a train overspeed protection method based on line curve radius identification. This method can be applied to two phases of a newly constructed line during the pre-operational, step-by-step speed increase. The first phase is the acceleration to overspeed, during acceleration to 110% of the design speed. The second phase is the deceleration protection phase after reaching 110% of the design speed before entering a speed-restricted section, particularly in tunnels. Preferably, the entire line can be considered to consist of a number of straight sections and speed-restricted sections (such as curved sections) alternating end-to-end. During the pre-operational, step-by-step speed increase, trains on newly constructed lines will switch back and forth between the two phases. Furthermore, when a train is neither in the first nor the second phase, the operator can steer the train based on their experience and established travel plans. Preferably, the "curve radius" described in the present invention, also referred to as the minimum curve radius, is a commonly used technical standard in railways. Its meaning is equivalent to the geometric curve radius, reflecting the degree of curvature of a curve and being the inverse of the curvature. In the railway field, this parameter typically refers to curves on a horizontal plane.

[0052] During the step-by-step speed test, the ATP (Train Overspeed Protection System) must be isolated at 110% of the design speed. Mileage can only be calculated by integrating the vehicle speed using the onboard speed sensor. This conversion process inevitably introduces errors. Over long distances, these errors accumulate, leading to significant discrepancies between the calculated and actual mileage.

[0053] In practice, mileage calibration is performed by having dedicated drivers read mileage markers and compare them with calculated mileage. This approach relies heavily on driver experience and concentration. Furthermore, tunnels are characterized by low light levels, high levels of dust and moisture, making it difficult for the human eye to discern mileage markers at high speeds, posing a safety risk.

[0054] Based on this, the present invention uses a camera 110 and a laser sensor 120 to identify curve features in real time each time the train passes a curve. Through the support vector machine (SVM) algorithm, the identified curve features are matched with the preset line information, thereby clarifying the current location of the train and performing overspeed protection according to the preset speed of the line.

[0055] Preferably, if Figure 1 As shown, for a train in the second stage, the method of the present invention may include the following steps:

[0056] S110 obtains a curved tunnel image and laser signal, and generates a distance feature;

[0057] S120. Input the distance feature into the constructed line radius recognition model to obtain the curve radius of the curved tunnel wall ahead;

[0058] S130. Input the curve radius into the constructed speeding protection model to obtain the deceleration mileage;

[0059] S140. Determine a train speed limit plan and communicate the train speed limit plan to the operator before the train reaches the deceleration mileage to complete the maximum braking operation.

[0060] Preferably, in step S110, an artificial intelligence visual ranging algorithm is used to achieve high-precision recognition and distance measurement of the tunnel environment in front of the train by combining deep learning, computer vision and machine learning technologies. The present invention combines multiple technologies to achieve real-time, accurate and adaptable distance feature acquisition, ensuring that image processing and distance calculation are completed in a very short time, ensuring timely response under high-speed driving conditions, and even under complex and unstable visual conditions, the algorithm can maintain high-precision distance measurement, can adapt to different tunnel environments and lighting conditions, and does not require frequent manual intervention. Preferably, as Figure 2 As shown, step S110 may include the following sub-steps:

[0061] S111 obtains image information and laser information of the line in front of the train cab;

[0062] S112. Determine the position of the left and right rails in the curved tunnel ahead based on the image information;

[0063] S113. Based on the laser information, determine the exact distance between the left and right rails in the curved tunnel ahead;

[0064] S114. Calculate the straight-line distances (i.e., extension distances) of the left and right rails from the starting position of the speed-limited section to the tunnel wall, respectively, as corresponding distance features.

[0065] Preferably, for step S111, Figure 3As shown, a composite sensor device positioned in front of a train cab can be used to achieve rapid and efficient data acquisition. The composite sensor device can include a high-precision camera 110 and a laser sensor 120, and can be detachably positioned in front of the train cab. Furthermore, the camera 110 is responsible for capturing image information (or simply, images) of the tunnel interior, while the laser sensor 120 measures the distance between the target object and the train. Both the camera 110 and the laser sensor 120 are positioned toward the train's forward direction. Preferably, the composite sensor device can include two laser sensors 120, which can be positioned on either side of the train cab, with their vertical projections overlapping the left and right rails, respectively. That is, the two laser sensors 120 can be positioned above the left and right rails, forming a second laser sensor 122 and a first laser sensor 121, respectively.

[0066] Preferably, camera 110 should have sufficient resolution (e.g., at least 1920×1080 pixels) to ensure clear image capture; laser sensor 120 should have high-frequency transmission and reception capabilities to obtain accurate distance data in a short period of time. Preferably, camera 110 and laser sensor 120 should operate synchronously to ensure real-time correspondence between image and distance information, providing a reliable foundation for subsequent data processing.

[0067] Preferably, the laser sensor 120 periodically emits laser pulses through a laser transmitter. When the laser pulse encounters a tunnel wall or other object, the laser signal is reflected back to the receiver, and the receiver records the time required for the laser signal to be emitted and received (echo time). Preferably, the laser sensor 120 can be equipped with a high-precision clock module to accurately measure the emission and reception time of the laser signal. The accuracy of the clock directly affects the accuracy of the ranging. Therefore, the laser sensor 120 of the present invention can use a clock that can reach the microsecond level. Preferably, the laser transmitter can emit laser signals at a certain frequency (for example, thousands of laser pulses per second) to ensure real-time monitoring of obstacles in the tunnel, wherein the trigger mechanism can be implemented by an efficient microcontroller or FPGA to ensure a fast response.

[0068] Preferably, in step S112, the position of the left and right rails in the forward curved tunnel can be determined based on the image information captured by camera 110. The forward curved tunnel position of the left and right rails refers to the intersection of the two rails where they connect to the forward transition curve section along the direction in which the straight section extends, and this intersection is located within the tunnel. In other words, this intersection is the starting point of the speed limit section. Furthermore, in step S112, a convolutional neural network (CNN) can be used to perform feature extraction on the image information captured by camera 110 to identify the aforementioned intersection, and the transition distance between the train and the intersection at the time the camera 110 captured the image information can be calculated.

[0069] Preferably, CNN is able to identify key visual elements such as lines, shapes and textures in the image, which are crucial for determining the relative position between the train and objects in the tunnel. Preferably, CNN processes the image through multiple convolutional layers and pooling layers to extract feature information of the rails and the surrounding environment. At the same time, data enhancement technology is used to improve the generalization ability of the model so that it can work stably under different lighting and complex conditions. Preferably, image processing operations and data enhancement operations may be included when performing data preprocessing. When performing image processing operations, image denoising algorithms (such as Gaussian blur) may be used to reduce the impact of environmental noise on image quality, and techniques such as histogram equalization may be applied to enhance image contrast and improve feature recognizability. When performing data enhancement operations, affine transformations such as rotation, scaling and translation may be performed on the image to generate diverse training samples to enhance the robustness of the model, and images under different lighting conditions may be simulated to improve the adaptability of the model in complex environments. Preferably, a CNN can include multiple convolutional layers, activation functions (such as ReLU), pooling layers, and fully connected layers. Convolution kernels are used to convolve the input image to extract low-level features (such as edges and textures). Nonlinear activation functions (such as ReLU) are used to introduce nonlinearity, enabling the model to better learn complex features. Max pooling or average pooling is used to reduce the spatial dimension of the feature map, reducing computational effort and extracting important features. Preferably, a dataset of annotated tunnel images can be used as the training dataset, where the annotations include the position and curve characteristics of the left and right rails. Preferably, data augmentation techniques (such as rotation, cropping, and flipping) can be used to improve the model's generalization capabilities and ensure that the model can adapt to various lighting and environmental conditions. Preferably, a cross-entropy loss function and an optimization algorithm (such as Adam) are used for model training, enabling it to learn effective feature representations through backpropagation. Preferably, in practical applications, real-time tunnel images captured in real time can be input into a trained CNN model, so that the model can output a feature map containing important features, further supporting subsequent positioning processing.

[0070] Preferably, for step S112, a semantic segmentation network (such as U-Net, SegNet, DeepLab, etc.) can be used to classify the image at the pixel level. Preferably, features can be gradually extracted and the spatial dimension can be reduced through a series of convolutional layers and pooling layers, and then the feature map can be restored to the original image size through transposed convolution (upsampling) to generate pixel-level segmentation output. Preferably, a labeled data set can be used as training data, wherein the labeling can include objects of different categories such as rails and tunnel walls, and cross-entropy loss can be used to optimize the accuracy of pixel-level classification to train the model by calculating the difference between the predicted value of each pixel and the true label. Preferably, the model will assign a category label to each pixel and output a segmentation map of the same size, where the value of each pixel represents the category of the position (such as left rail, right rail, tunnel wall, etc.).

[0071] Preferably, in step S112, the high-level features extracted by CNN and the pixel-level information of semantic segmentation are used to combine the advantages of both to improve the accuracy and robustness of recognition, wherein CNN can serve as a feature extraction module, and semantic segmentation is responsible for identifying and locating specific objects (such as rails). Preferably, the positions of the left and right rails can be efficiently identified using the results of semantic segmentation. Preferably, the outputs of CNN and semantic segmentation are combined to form a complete model output to determine the relative positions of the left and right rails and their curve characteristics in the tunnel.

[0072] Preferably, in step S113, the distance from the laser sensor 120 to the object can be calculated according to the laser information (such as echo time) obtained in step S111 and the known laser speed using the following formula:

[0073]

[0074] Wherein, A is the distance from the laser sensor 120 to the object, c is the speed of light, and t is the echo time.

[0075] Preferably, during each laser measurement, the distance from the first laser sensor 121 along the right rail extension direction to the front curved tunnel wall and the distance from the second laser sensor 122 along the left rail extension direction to the front curved tunnel wall are recorded respectively, that is, the first laser distance A1 and the second laser distance A2.

[0076] Preferably, the collected data can be stored in a data buffer in real time and preliminarily processed to exclude abnormal values (such as values outside the normal measurement range), wherein the distance data can be further smoothed by a data filtering algorithm (such as Kalman filtering) to reduce the impact of measurement noise.

[0077] Preferably, in step S114, the two laser distances obtained in step S113 may be subtracted from the transition distance obtained in step S112 to obtain a first extension distance S1 and a second extension distance S2, respectively, and these two extension distances may be used as distance features.

[0078] Preferably, the sizes of the first extension distance and the second extension distance are related to the direction of the curved tunnel wall ahead, wherein, if the curved tunnel wall ahead turns to the left, the second extension distance is greater than the first extension distance; conversely, if the curved tunnel wall ahead turns to the right, the second extension distance is less than the first extension distance.

[0079] Preferably, the artificial intelligence visual ranging algorithm used in step S110 can also be combined with a recursive neural network (RNN) for processing time series data, so as to maintain continuous recognition and prediction of distance features during the movement of the train; and / or combined with a generative adversarial network (GAN) for generating high-quality image data during training to enhance the algorithm's recognition ability in complex environments such as dim light, dust and moisture; and / or combined with deep reinforcement learning (DRL) that can achieve self-learning and optimize ranging strategies by simulating the train driving environment, so as to improve the ranging accuracy and robustness under different environmental conditions.

[0080] Preferably, in step S120, the powerful classification capabilities of the Support Vector Machine (SVM) are utilized to analyze sensor data during train travel, combined with distance features, to obtain real-time information about the route curve ahead, thereby accurately identifying the railway route curve parameters. This invention combines the advantages of geometric calculations and machine learning, ensuring both rapid response of preliminary estimates and improved accuracy of final predictions. The SVM is insensitive to noise and outliers, maintaining stable performance in complex and changing environments. The optimized SVM model can quickly respond to real-time data, meeting the needs of instant identification under high-speed train travel.

[0081] Preferably, in addition to the first extension distance S1, the second extension distance S2, and the spacing D between the first laser sensor 121 and the second laser sensor 122, other features relevant to curve radius identification can be extracted from sensor data such as speed, acceleration, and steering angle collected during train travel. Furthermore, noise and outliers can be removed to ensure data quality and provide accurate basic data for subsequent steps.

[0082] Preferably, if Figure 4As shown in the figure, based on the collected data, the curve radius of the curved tunnel wall ahead can be preliminarily calculated. This preliminary calculation enables a rapid estimation of the current curve radius, which can be used as an additional feature in SVM model training or directly used for preliminary judgment in real-time applications. Furthermore, the curve radius R of the curved tunnel wall ahead can be quickly calculated by simultaneously solving the following formula:

[0083] S2=R·sina

[0084] S1=R·sinb

[0085] R·cosa+D=R·cosb

[0086] Wherein, R is the curve radius, S1 is the first extension distance, S2 is the second extension distance, a is the second angle, b is the first angle, and D is the layout spacing between the two laser sensors 120.

[0087] Furthermore, the first angle b is the angle between the lines connecting the two end points of the first extension distance to the virtual center of the curved wall; the second angle a is the angle between the lines connecting the two end points of the second extension distance to the virtual center of the curved wall.

[0088] Preferably, an appropriate kernel function (e.g., linear kernel, polynomial kernel, radial basis function (RBF), etc.) is selected based on the data characteristics to effectively map the input data into a high-dimensional feature space. The SVM model is trained using a preprocessed dataset containing the distance features measured by the laser sensor 120 and the curve radius R preliminarily calculated using the above formula. The support vector is determined by optimizing the objective function (e.g., the hinge loss function) and the classification boundary is constructed.

[0089] Preferably, as a train approaches a curve, the latest train operation data is collected in real time and converted into a feature vector suitable for SVM model analysis. This feature vector includes the original distance feature, the initially calculated curve radius, and other relevant features. The SVM model outputs a final curve radius prediction based on the input feature vector, thereby accurately identifying the curve radius at the train's current location.

[0090] Preferably, cross-validation is used to evaluate the generalization ability of the SVM model to ensure its accuracy on unknown data. Based on the validation results, the model parameters (such as the penalty parameter C, kernel function parameters, etc.) are adjusted to obtain the best recognition effect.

[0091] Preferably, if Figure 5 As shown, for step S130, it constructs an overspeed protection model through the following sub-steps and calculates the deceleration mileage based on it:

[0092] S131. Get the starting mileage and target speed limit of the speed limit section;

[0093] S132. Determine whether speed limit protection is required and calculate the braking distance if necessary;

[0094] S133. Calculate safe deceleration distance;

[0095] S134. Calculate the deceleration mileage.

[0096] Preferably, in step S131, the starting mileage and the established target speed limit of each speed limit section can be obtained through the design document, wherein the conventional speed limit section may include line curves, switches, stations and / or tunnels, etc., and in the present invention may mainly refer to the section of the line curve in the tunnel. Figure 6 This is a schematic diagram of a line design document. Preferably, after a train passes through any speed-limited section in its forward direction, the starting mileage of the next speed-limited section along the line can be extracted for backup. Furthermore, when setting the target speed limit in advance, consideration should be given to its compatibility with the corresponding curve radius.

[0097] Preferably, before implementing speed limit protection, the operator has at least partially performed a braking operation, thereby reducing the train's current speed. However, since it is difficult to determine the train's actual position in a tunnel, it is difficult to accurately determine the established target speed limit, and thus it is impossible to determine whether speed limit protection is necessary. In step S132, the current speed is compared with the target speed limit. If the current speed is greater than the target speed limit, it is determined that speed limit protection is necessary. Conversely, if the current speed is less than the target speed limit, travel may continue at the current speed or at a speed not exceeding the target speed limit.

[0098] Preferably, when speed limit protection is required, the braking distance L2 can be calculated using the following formula:

[0099]

[0100] Where L2 is the braking distance, V1 is the current speed, V3 is the target speed limit, and a2 is the maximum common braking average deceleration.

[0101] Preferably, in step S133, based on the obtained braking distance L2, the safe deceleration distance L3 can be calculated by the following formula:

[0102] L3=σ×L2

[0103] Among them, σ is the safety factor and L2 is the braking distance.

[0104] Preferably, the safety factor σ can be adjusted according to actual conditions, wherein a value of 2 can be taken.

[0105] Preferably, in step S134, the deceleration mileage can be calculated based on the obtained starting mileage of the speed limit section and the safe deceleration distance. The direction of the train's travel needs to be considered when calculating the deceleration mileage. When the train travels in the direction of increasing mileage, the deceleration mileage is obtained by subtracting the safe deceleration distance from the starting mileage of the speed limit section; when the train travels in the direction of decreasing mileage, the deceleration mileage is obtained by adding the safe deceleration distance to the starting mileage of the speed limit section.

[0106] Preferably, after the deceleration mileage is calculated in step S130, in step S140, the operator can be informed of the train speed limit plan before the train reaches the deceleration mileage and perform a braking operation accordingly, wherein the train speed limit plan may include current line information (such as the curve radius of the curved tunnel wall ahead, the starting mileage of the speed limit section, etc.) and the target speed limit. Furthermore, the operator can be informed of the train speed limit plan by receiving a braking prompt, wherein the braking prompt can be conveyed to the operator by various means such as voice. Preferably, the current mileage information of the train can be obtained based on the transition distance obtained by the camera 110, so that the train speed limit plan can be conveyed to the operator by issuing a braking prompt before the train reaches the deceleration mileage, so that the operator can complete the maximum braking operation, wherein the maximum braking operation refers to braking the train at the maximum common braking average deceleration.

[0107] Preferably, if the current speed of the train reaches the maximum design speed (such as 160km / h), the train is in the first stage and needs to accelerate to the target overspeed to perform an overspeeding test, wherein the target overspeed is set to 1.1 times the maximum design speed.

[0108] Preferably, the acceleration distance L1 can be calculated by the following formula:

[0109]

[0110] Where L1 is the acceleration distance, V1 is the current speed, V2 is the target speeding, and a1 is the average acceleration.

[0111] Furthermore, by judging the acceleration distance L1 and the length L of each straight line segment 直 The relationship between L1 and L2 determines whether to conduct an overspeed test. 直 / 2, no overspeed test at 110% design speed will be conducted; if L1 <L 直 / 2, an overspeed test of 110% of the design speed is conducted.

[0112] Preferably, the acceleration mileage can be calculated based on the obtained starting mileage of the straight section and the acceleration distance. The direction of the train's travel needs to be considered when calculating the acceleration mileage. When the train travels in the direction of increasing mileage, the acceleration mileage is obtained by subtracting the acceleration distance from the small mileage at the starting point of the overspeeding interval; when the train travels in the direction of decreasing mileage, the acceleration mileage is obtained by adding the large mileage from the starting point of the overspeeding interval to the acceleration distance.

[0113] Preferably, the operator can be informed of the train's acceleration plan before the train reaches the acceleration distance and use it to perform braking operations. The train acceleration plan may include current route information and target overspeed. Furthermore, the operator can be informed of the train's acceleration plan by receiving an overspeed notification, which can be communicated to the operator through various means such as voice.

[0114] The overspeed protection model is designed to ensure that trains maintain safe speeds during acceleration and deceleration by monitoring the train's running status in real time. Before a train enters an acceleration phase, it notifies the operator (e.g., the train driver) through various means, including voice notification, that the train is currently accelerating and provides real-time route information and target speed. This information is calculated and updated in real time based on the train's current position, the track's slope, curve radius, and other factors that may affect train acceleration, ensuring that the driver can control the train's acceleration within a safe range. Before a train enters a deceleration phase, it uses various means, including voice notification, to remind the driver to decelerate and provide them with the current route information and target deceleration speed. At this point, the overspeed protection model calculates the required deceleration value based on the train's current speed, the preset safe speed, and track restrictions. It then issues braking prompts to the driver, including voice notification, to ensure that the train does not exceed the speed limit when reaching the next target point.

[0115] This invention uses an integrated artificial intelligence visual odometry algorithm to monitor train travel status in real time. Combined with a curve radius identification method based on a support vector machine (SVM) algorithm, this method ensures accurate identification of the train's trajectory. Each time a train passes through a curve, the curve radius is measured in real time using image recognition technology. This technology automatically adjusts the parameters of the overspeed protection model based on parameters such as train speed and acceleration, ensuring the model remains consistently adapted to actual line conditions.

[0116] This method accurately calculates the radius of each curve during train travel and optimizes the model using historical data. By accurately identifying the train's line in real time, the system can dynamically adjust overspeed protection strategies to mitigate the risk of overspeeding due to changing line characteristics.

[0117] This paper also proposes a time-based overspeed protection accuracy assessment method, which compares the actual speed of the train at each time point with the target speed to evaluate the accuracy of the current overspeed protection model. The specific assessment process is as follows:

[0118] The train is at time node t i The target speed is v target (t i ), the actual driving speed is v actual (t i The accuracy of the overspeed protection model can be calculated using the following formula:

[0119]

[0120] Among them, Precision(t i ) represents the time node t i Overspeed protection accuracy, v actual (t i ) is the actual speed of the train, v target (t i ) is the target speed preset by the overspeed protection model. This formula reflects the deviation between the actual train speed and the target speed. The closer the value is to 1, the more accurate the protection model is.

[0121] Furthermore, the overall overspeed protection accuracy evaluation can be obtained by taking a weighted average of the accuracy of all time nodes:

[0122]

[0123] Where N is the total number of time nodes.

[0124] Preferably, the value calculated using the above formula can fully reflect the accuracy of the speed protection model throughout the entire driving process. If the evaluation result is lower than the set accuracy threshold, the model needs to be adjusted or recalibrated to ensure the safety and effectiveness of speed protection.

[0125] In addition to accuracy assessment, this invention also proposes a real-time correction mechanism for the overspeed protection model. When the integrated artificial intelligence visual odometry and support vector machine (SVM) algorithms detect changes in the train's current position, the model automatically applies corrections, ensuring that the protection strategy can readily adapt to changes in the train's actual operating environment. This correction mechanism effectively reduces the risk of overspeeding caused by errors, changes in external factors, or changes in line conditions, thereby improving the reliability and stability of the entire protection system.

[0126] Example 2

[0127] This embodiment is a further improvement of embodiment 1, and repeated contents will not be repeated here.

[0128] The present invention also discloses a train overspeed protection system based on line curve radius identification, which can execute the train overspeed protection method described in Example 1.

[0129] Preferably, if Figure 7 As shown, the train overspeed protection system may include: an acquisition module 100, used to obtain curved tunnel images and laser signals; a distance feature generation module 200, used to generate distance features based on the data information obtained by the acquisition module 100; a line radius identification module 300, used to input the distance features into a constructed line radius identification model to obtain the curve radius of the curved tunnel wall ahead; a speed limit mileage calculation module 400, used to input the curve radius into the constructed overspeed protection model to obtain the deceleration mileage; and a real-time protection module 500, used to convey the train speed limit plan to the operator before the train reaches the deceleration mileage.

[0130] Preferably, the train overspeed protection system may also include an acceleration mileage calculation module for determining whether to conduct an overspeed test and determining the acceleration mileage required for the overspeed test. Furthermore, if the train is eligible for an overspeed test, the real-time protection module 500 may also be configured to communicate the train acceleration plan to the operator before the train reaches the acceleration mileage.

[0131] It should be noted that the above-mentioned specific embodiments are exemplary, and those skilled in the art can come up with various solutions inspired by the disclosure of the present invention, and these solutions also belong to the disclosure scope of the present invention and fall within the protection scope of the present invention. Those skilled in the art should understand that the present invention specification and its drawings are illustrative and do not constitute a limitation on the claims. The scope of protection of the present invention is defined by the claims and their equivalents. The present invention specification contains multiple inventive concepts, such as "preferably" or "according to a preferred embodiment", which means that the corresponding paragraph discloses an independent concept, and the applicant reserves the right to file a divisional application based on each inventive concept. Throughout the text, the features guided by "preferably" are only an optional method and should not be understood as having to be set. Therefore, the applicant reserves the right to abandon or delete the relevant preferred features at any time.

Claims

1. A train overspeed protection method based on line curve radius identification, characterized in that: The second stage is the deceleration protection stage after reaching 110% of the design speed and before entering the speed-restricted section in the tunnel. For trains in the second stage, the following steps are included: Obtaining a curved tunnel image and a laser signal to obtain image information and laser information of the line in front of the train cab, using a convolutional neural network to extract features from the image information obtained by the camera (110), identifying the intersection position of the left and right rails in the tunnel where the left and right rails are connected to the transition curve section in the extension direction of the straight section, and using the intersection position as the starting point of the speed limit section to calculate the transition distance between the train and the intersection position when the camera (110) obtains the image information, thereby determining the position of the left and right rails in the curved tunnel in front, determining the accurate distance of the left and right rails in the curved tunnel in front based on the laser information, and calculating the straight-line distance of the left and right rails from the starting position of the speed limit section to the tunnel wall to generate distance features; The distance feature is input into the constructed line radius recognition model, and the curve radius of the front curved tunnel wall is preliminarily calculated based on geometric calculation and support vector machine algorithm using the first extension distance, the second extension distance and the layout spacing between the first laser sensor (121) and the second laser sensor (122), wherein the first laser sensor (121) and the second laser sensor (122) are respectively arranged on both sides in front of the train cab and are respectively located above the left and right rails, and the distance from the first laser sensor (121) and the second laser sensor (122) along the respective corresponding rail extension direction to the front curved tunnel wall minus the transition distance to obtain the first extension distance and the second extension distance respectively; The curve radius is input into the constructed overspeed protection model to obtain the deceleration mileage; Determine the train speed limit plan and communicate it to the operator before the train reaches the deceleration mileage to complete the maximum braking operation, including: By comparing the speed of the train at each passing time node with the target speed, the accuracy of the current overspeed protection model in protecting trains in the second stage is evaluated.

2. The train overspeed protection method according to claim 1, characterized in that: Determine the deceleration mileage through the following sub-steps: Obtain the starting mileage and target speed limit of the speed limit section from the design document; Determine whether speed limit protection is needed by comparing the current speed with the target speed limit and calculate the braking distance if necessary; Calculate safe deceleration distance based on preset safety factor; The deceleration mileage is calculated based on the obtained starting mileage of the speed limit section and the safe deceleration distance, wherein the travel direction of the train is taken into consideration when calculating the deceleration mileage.

3. The train overspeed protection method according to claim 1, characterized in that: The operator learns about the train speed limit plan by receiving a braking prompt, wherein the braking prompt is conveyed to the operator through voice so that the operator who receives the braking prompt can complete the maximum braking operation before the train reaches the deceleration mileage.

4. The train overspeed protection method according to claim 1, characterized in that: For trains in the first stage, whether to conduct an overspeeding test is determined by calculating the acceleration distance and judging the relationship between the acceleration distance and the length of each straight section. The acceleration mileage is calculated based on the obtained starting mileage of the straight section and the acceleration distance. The train's direction of travel is taken into account when calculating the acceleration mileage so that the operator can receive an overspeeding prompt before the train reaches the acceleration mileage.

5. The train overspeed protection method according to claim 1, characterized in that: The accuracy calculated at all time nodes is weighted averaged to obtain the accuracy of the entire driving process. If the evaluation result is lower than the set accuracy threshold, the model needs to be adjusted or recalibrated.

6. A train overspeed protection system based on line curve radius identification, characterized in that: It can execute the train overspeed protection method according to any one of claims 1 to 5, which comprises: An acquisition module (100) is used to obtain curved tunnel images and laser signals; A distance feature generation module (200) is used to generate a distance feature based on the data information acquired by the acquisition module (100); A line radius identification module (300) is used to input the distance feature into the constructed line radius identification model to obtain the curve radius of the front curved tunnel wall; A speed limit mileage calculation module (400) is used to input the curve radius into the constructed overspeed protection model to obtain the deceleration mileage; The real-time protection module (500) is used to convey the train speed limit plan to the operator before the train reaches the deceleration mileage.

7. The train overspeed protection system according to claim 6, characterized in that: It includes: The acceleration mileage calculation module is used to determine whether to conduct an overspeeding test and to determine the acceleration mileage when conducting the overspeeding test, so that when the acceleration mileage calculation module determines that the train is conducting an overspeeding test, the real-time protection module (500) can convey the train acceleration plan to the operator before the train reaches the acceleration mileage.

Images