Train positioning anti-interference method and device, electronic equipment and storage medium

Abstract

The invention relates to the technical field of train positioning, discloses a train positioning anti-interference method and device, electronic equipment and a storage medium, and aims to solve the problem of insufficient train positioning precision and reliability in a complex electromagnetic environment. The method comprises the following steps: identifying trackside marks through a deep learning visual detection means, and correcting the output of an original positioning system to obtain optimized position and speed information; predicting a train running track by using a long short-term memory network in combination with the historical running data and the current driving information; and the electromagnetic interference state is judged by comparing the deviation between the optimized information and the predicted trajectory, and then the positioning output is subjected to differential constraint correction. A closed-loop anti-interference mechanism is constructed, the influence of electromagnetic interference and accumulative errors is effectively offset, the positioning precision and the anti-interference robustness are remarkably improved, existing hardware does not need to be transformed on a large scale, a complex line scene does not need to be adapted, and reliable position support is provided for safe and efficient operation of a train.

Description

Technical Field

[0001] This invention relates to the field of train positioning technology, specifically to a train positioning anti-interference method, device, electronic equipment, and storage medium. Background Technology

[0002] With the development of railway transportation towards high speed and intelligence, the accuracy and anti-interference capability of train positioning systems have become the core support for ensuring the safe operation of train control systems. Traditional train positioning methods mostly rely on a single sensor or information source, which has significant limitations: track circuit positioning accuracy is low and it is difficult to meet the requirements of advanced train control; transponder positioning is easily affected by weather and terrain, and large-scale deployment and maintenance costs are high; onboard ground radar and Global Navigation Satellite System (GNSS) are easily interfered with by complex electromagnetic environments, and signal limitation or loss of lock often occurs in mountainous areas, tunnels and other scenarios; odometer positioning suffers from cumulative errors due to wheel slippage, wheel spin, and geometric changes, affecting long-term positioning reliability.

[0003] To compensate for the shortcomings of single-source positioning technologies, multi-source information fusion positioning technologies are being gradually promoted, especially the combination of the BeiDou satellite navigation system with other positioning methods, which has gained widespread application due to its all-weather operation advantage. However, such systems generally include air interface equipment such as vehicle-mounted antennas, which are highly susceptible to electromagnetic coupling from space electromagnetic fields, leading to a decrease in positioning accuracy. In complex and challenging areas, the combination of harsh electromagnetic environments and complex terrain further exacerbates the reliability risks of positioning systems, directly threatening train operation safety and transportation efficiency.

[0004] While existing combined positioning technologies can partially compensate for the shortcomings of single positioning, they fail to address the core issues of electromagnetic coupling interference and cumulative errors at the system level, and lack adaptive anti-interference mechanisms for complex electromagnetic environments. Therefore, there is an urgent need for a train positioning anti-interference solution that balances accuracy, robustness, and economy. This solution should improve positioning reliability in complex scenarios by optimizing the electromagnetic compatibility of the positioning system, introducing reliable auxiliary positioning methods, and implementing dynamic constraint mechanisms, thereby ensuring the safe and efficient operation of trains. Summary of the Invention

[0005] To address the technical problems of traditional train positioning systems being susceptible to electromagnetic interference and accumulating errors in complex electromagnetic environments, and lacking adaptive anti-interference mechanisms, resulting in insufficient positioning accuracy and reliability and failing to meet the requirements for safe and efficient train operation, this invention provides a train positioning anti-interference method, device, electronic device, and storage medium. It can achieve assisted positioning by constraining train operation based on trackside sign recognition, historical train operation data, and existing train operation information, thereby improving the anti-interference capability of the train positioning system.

[0006] In a first aspect, the present invention provides a train positioning anti-interference method, comprising: The trackside signs are detected and identified using deep learning-based visual detection methods, and the original train positioning system output is corrected based on the detection and identification results to obtain optimized train position and speed information. Based on historical train operation data and current train operation information, the long short-term memory network is used to predict the train's trajectory. The optimized train position and speed information is compared with the predicted running trajectory, and the electromagnetic interference status is determined based on the deviation of the comparison result. Based on the judgment results of the electromagnetic interference state, the output differences of the train positioning system are constrained and corrected.

[0007] The train positioning anti-interference method provided in this invention addresses the pain points of train positioning in complex electromagnetic environments through a closed-loop process of visual correction of trackside signs, LSTM trajectory prediction, deviation judgment of interference, and differentiated constraint correction. Compared with traditional single positioning methods, it relies on deep learning to achieve accurate identification and positioning correction of trackside signs, constructs trajectory redundancy through LSTM prediction, and dynamically adjusts the correction strategy according to the interference state, effectively avoiding positioning inaccuracies caused by electromagnetic interference to GNSS and cumulative errors in odometers. Simultaneously, it requires no large-scale modification of existing hardware, balances positioning accuracy and system compatibility, significantly improves the anti-interference capability of train positioning, provides reliable position support for the safe and efficient operation of trains, and is suitable for complex scenarios such as mountainous areas and tunnels.

[0008] In one optional implementation, the step of detecting and recognizing trackside signs using a deep learning-based visual detection method, and correcting the original train positioning system output based on the detection and recognition results to obtain corrected train position and speed information, includes: By using a high-speed camera installed at the front of the train, images of trackside signs around the train's route are captured in real time, and learning samples and real-time detection images are output. A deep convolutional generative adversarial network is used to extract features from the trackside sign learning samples and train them to generate fake trackside sign samples with high similarity to real samples. The fake samples are then embedded in appropriate positions in the railway track background image to expand the dataset. An improved single-anchor-point multi-border detector is used to detect and locate trackside signs. The trackside signs are segmented by vertical projection and the signs are identified by template matching. The Kalman filter algorithm is used to integrate the position and speed information of the original train positioning system with the trackside sign detection and recognition results to output optimized train position and speed information. The process noise covariance matrix and observation noise covariance matrix in the Kalman filter algorithm are dynamically adjusted according to the real-time motion state of the train and the intensity of electromagnetic interference.

[0009] This invention addresses the low recognition accuracy issue caused by insufficient trackside sign samples in complex scenarios (dim tunnels, blurry rainy conditions) by real-time acquisition of trackside signs using a high-speed camera and supplementing the dataset with DCGAN. Improved SSD detectors and template matching methods ensure accurate sign detection, positioning, and recognition. Kalman filtering fuses multi-source data and dynamically adjusts the noise covariance matrix, adapting in real-time to train motion and electromagnetic interference intensity, effectively offsetting random biases and cumulative errors in the original positioning system. Compared to traditional fixed-parameter fusion methods, positioning accuracy is significantly improved, meeting the stringent requirements of advanced train control systems.

[0010] In one optional implementation, the improved single-anchor-point multi-boundary detector network includes at least one of the following improvements: the Z-shaped network and cascaded residual meshes in the original network structure are removed; the number of preset anchor boxes in the model is reduced from 4 to 2; the backbone network includes an optimized VGG-16 base network and additional convolutional layers. The VGG-16 base network replaces the last two fully connected layers with convolutional layers. The additional convolutional layers are used to increase the receptive field, reduce the feature layer resolution by a factor of 2, and are named C6. Together with the C3, C4, and C5 layers of VGG-16, they constitute multi-scale features to achieve multi-scale detection.

[0011] This invention improves the single-anchor-point multi-frame detector by reducing complexity, increasing efficiency, and maintaining accuracy in both network structure and parameters. The Z-shaped network and cascaded residual mesh are removed to reduce redundant computation and avoid overfitting. The number of anchor frames is reduced from four to two, significantly increasing detection speed while maintaining accuracy, meeting the real-time requirements of high-speed train operation. The VGG-16 algorithm is optimized and convolutional layers are added to create multi-scale features, enabling accurate capture of trackside signs of different sizes and distances. This addresses the insufficient multi-scale detection capability of traditional detectors in complex scenarios, improving sign recognition accuracy and providing a reliable data foundation for positioning correction.

[0012] In one optional implementation, the parameter setting and dynamic adjustment rules of the Kalman filter algorithm include: The state vector is set to the train position, speed, and acceleration, and the observation vector is set to the original positioning system observation position, trackside marker detection position, and original positioning system observation speed. Process noise covariance matrix Q k It is a 3×3 matrix, with diagonal elements corresponding to the process noise variance of position, velocity, and acceleration, and off-diagonal elements being 0. It is initialized based on train design parameters and line characteristics, and dynamically adjusted according to train acceleration fluctuations, line gradient changes and operation stages, maintaining the minimum value during the uniform speed stage. Observation noise covariance matrix R kThis is a 3×3 matrix. The diagonal elements correspond to the observation noise variance of the original positioning system position, trackside marker position, and original positioning system velocity. The off-diagonal elements are 0. It is initialized based on sensor accuracy indicators and dynamically adjusted according to electromagnetic interference intensity, trackside marker clarity, and observation data consistency. The data source is R. k The more reliable it is, the smaller it is. This invention overcomes the problem of insufficient filtering accuracy caused by traditional fixed noise covariance matrices by setting and dynamically adjusting Kalman filter parameters. The state vector and observation vector settings comprehensively cover key train motion parameters, ensuring the integrity of data fusion. The process noise covariance matrix is ​​dynamically adjusted based on train design parameters and track characteristics, adapting to different operating scenarios such as constant speed / acceleration / deceleration, straight sections / slopes. The observation noise covariance matrix dynamically changes with electromagnetic interference intensity and marker clarity, prioritizing reliable data sources. Compared to fixed-parameter filtering, this rule makes the filtering results more closely match the actual operating conditions, improving the stability of positioning data and effectively avoiding positioning jumps caused by interference.

[0013] In one optional implementation, the step of comparing the corrected train positioning information with the predicted trajectory includes: calculating the position deviation between the corrected position and the predicted position, and the speed deviation between the corrected speed and the predicted speed; and dynamically setting a position deviation threshold based on the recognition confidence of the trackside markers. The speed deviation threshold is dynamically set based on the speed uncertainty predicted by the Long Short-Term Memory network.

[0014] This invention addresses the problem of misjudging interference states using traditional fixed threshold methods by calculating position and velocity deviations and dynamically setting thresholds based on trackside marker recognition confidence and LSTM velocity prediction uncertainty. When trackside marker recognition is clear (high confidence), the position deviation threshold is more stringent to ensure positioning accuracy; when LSTM prediction is reliable (low uncertainty), the velocity deviation threshold is more stringent to avoid trajectory deviation. This method adapts to data reliability in real time, accurately identifies the degree of electromagnetic interference, provides a scientific basis for subsequent differentiated constraint corrections, and reduces efficiency losses caused by "misjudged interference" or safety risks caused by "missed interference." In one optional implementation, based on the judgment result of the electromagnetic interference state, the output differentiation of the train positioning system is constrained and corrected, including: If both the position deviation and speed deviation are less than their corresponding dynamic thresholds, the corrected train positioning information and the predicted trajectory are weighted and fused, and the fusion result is output. If only one of the position deviation or speed deviation exceeds its corresponding threshold, the dimension data that exceeds the threshold is corrected based on the dimension data that does not exceed the threshold. If both the position deviation and speed deviation exceed their corresponding thresholds, the preferred data source is selected based on the recognition confidence of the trackside sign and the uncertainty of the prediction by the long short-term memory network, or emergency constraint logic is triggered.

[0015] This invention achieves a positioning output logic of "precise correction, efficiency priority, and safety fallback" through a differentiated constraint correction strategy. When there is no deviation exceeding limits, weighted fusion improves accuracy and avoids the impact of random errors in single data points. When a single dimension exceeds limits, reliable data is used as a benchmark for correction, ensuring positioning reliability without speed reduction and minimizing efficiency loss. When two dimensions exceed limits, reliable data sources are prioritized or emergency constraints are triggered to avoid safety risks caused by inaccurate positioning under strong interference. Compared to the traditional "one-size-fits-all" correction, this strategy maximizes both positioning accuracy and operational efficiency while providing a safety fallback in extreme interference scenarios, balancing the conflict between safety and efficiency.

[0016] In one optional implementation, the historical train operation data includes one or more of historical location, historical speed, historical acceleration, track gradient, and curve radius; the current train operation information includes one or more of train timetable, timetable, and planned speed.

[0017] This invention provides comprehensive and accurate input features for LSTM trajectory prediction by combining historical train operation data with current train operation information. Historical data covers key parameters of train motion and the line, helping the LSTM learn the time-series patterns of train operation; current train operation information includes data related to the operation plan, ensuring the predicted trajectory aligns with actual operational needs. Compared to prediction methods that rely on only partial data, this data selection allows the LSTM to more accurately capture train operation trends. Even when trackside marker detection fails, it can still output a continuous and smooth predicted trajectory, avoiding positioning interruptions, improving the robustness of the positioning system against interference, and adapting to different lines and operational scenarios.

[0018] In a second aspect, the present invention provides a train positioning anti-interference device, the device comprising: The equation system construction module is used to construct a nonlinear equation system with node pressure as the unknown quantity based on the topology and operating parameters of the gas pipeline network, and to set the global convergence tolerance error. The equation solving module is used to iteratively solve the equation system. During the iteration process, the following friction model switching strategy is executed, including: The initialization unit is used to calculate the friction coefficient of each pipe using a preset high-speed model during the initialization phase. The flow residual calculation unit is used to calculate the infinite norm of the node flow residual in the current iteration step; The first model switching detection unit is used to continue using the preset high-speed model to calculate the pipeline friction coefficient in the next iteration step when the infinity norm is greater than the first threshold. The second model switching detection unit is used to switch to using a preset high-precision model to calculate the friction coefficient when the infinite norm is less than or equal to the first threshold for the first time, and to continue using the preset high-speed model in subsequent iterations without switching back. The simulation result output unit is used to determine the calculation convergence when the infinity norm is less than the global convergence tolerance, and output the pressure and flow distribution results of the pipeline nodes and the pipeline segments.

[0019] Thirdly, the present invention provides an electronic device, comprising: a memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the computer instructions to perform the train positioning anti-interference method of the first aspect or any corresponding embodiment described above.

[0020] Fourthly, the present invention provides a computer-readable storage medium storing computer instructions for causing a computer to execute the train positioning anti-interference method of the first aspect or any corresponding embodiment described above.

[0021] Fifthly, the present invention provides a computer program product, including computer instructions, which are used to cause a computer to execute the train positioning anti-interference method of the first aspect or any corresponding embodiment described above. Attached Figure Description

[0022] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0023] Figure 1 This is a flowchart illustrating a train positioning anti-interference method according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the learning and training of the DCGAN model according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the trackside sign detection and recognition process according to an embodiment of the present invention; Figure 4 This is a schematic diagram of a trackside marker detection and positioning model based on the improved SSD algorithm according to an embodiment of the present invention. Figure 5 This is a structural block diagram of a train positioning anti-interference device according to an embodiment of the present invention; Figure 6 This is a schematic diagram of the hardware structure of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] It is understood that before using the technical solutions disclosed in the various embodiments of the present invention, users should be informed of the types, scope of use, and usage scenarios of the personal information involved in the present invention and their authorization should be obtained in accordance with relevant laws and regulations through appropriate means.

[0026] The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0027] Current train positioning methods often suffer from inherent flaws when facing strong electromagnetic interference due to their reliance on single or homogeneous data sources. Solutions based on Global Navigation Satellite Systems (GNSS) or trackside electronic tags are susceptible to direct interference, leading to data distortion or even loss of lock-on. Methods relying solely on inertial estimation or historical data prediction suffer from error accumulation and lack effective real-time correction mechanisms. Existing technologies lack a mechanism for real-time, quantitative, and adaptive cross-validation and decision-making among positioning data sources based on different principles. This results in the system's inability to reliably identify abnormal data in complex electromagnetic environments, making it difficult to guarantee the robustness and reliability of positioning results.

[0028] This invention provides an embodiment of a train positioning anti-interference method, which can achieve assisted positioning by constraining train operation based on trackside sign recognition, historical train operation data, and existing train operation information, thereby improving the anti-interference capability of the train positioning system. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0029] Figure 1 This is a flowchart of a train positioning anti-interference method according to an embodiment of the present invention, such as... Figure 1 As shown, the process includes the following steps: Step S1: The trackside signs are detected and identified using a deep learning-based visual detection method. Based on the detection and identification results, the original positioning system output of the train is corrected to obtain optimized train position and speed information.

[0030] Specifically, trackside signs possess known absolute position attributes. This invention can quickly and accurately identify trackside signs using deep learning-based visual detection methods, providing a reliable absolute position benchmark for positioning correction. By fusing trackside sign detection results with existing positioning system data, it can effectively offset the random deviations of GNSS due to electromagnetic interference and the cumulative errors of odometers, providing data support for precise train control.

[0031] Step S2: Based on historical train operation data and current train operation information, predict the train's trajectory using a long short-term memory network.

[0032] Specifically, historical train operation data includes one or more of historical position, historical speed, historical acceleration, track gradient, and curve radius; current train operation information includes one or more of train timetable, timetable, and planned speed. Train operation trajectories have strong time series dependencies (such as constant speed on straight sections, acceleration and deceleration on gradients, and speed adjustments according to timetables). Traditional recurrent neural networks (RNNs) are prone to failing to capture long-term correlations due to "gradient vanishing / gradient exploding." LSTM, through its gating mechanism (forget gate, input gate, and output gate), can effectively solve this problem: the LSTM's forget gate can autonomously filter core features from historical operation data; the input gate can integrate current train operation information (track gradient, curve radius, remaining arrival time, and planned speed) to dynamically adjust the prediction logic; and the output gate predicts a trajectory with gradually decreasing speed, rather than maintaining a constant speed prediction, making the deviation between the predicted position and the actual position extremely small.

[0033] Step S3: Compare the optimized train position and speed information with the predicted running trajectory, and determine the electromagnetic interference status based on the deviation of the comparison result.

[0034] Specifically, LSTM prediction relies solely on historical operational data and current train information (such as line parameters and timetables), is unaffected by electromagnetic coupling from spatial electromagnetic fields, and does not depend on hardware devices such as cameras. Even if trackside marker detection fails or GNSS lock-on is lost, it can still output a continuous and smooth operating trajectory. For example, on mountain slopes, LSTM prioritizes associating historical patterns of "gradient-acceleration-velocity" to predict a speed change trajectory that matches the slope. In tunnel scenarios, even without GNSS data, LSTM can still predict a continuous trajectory based on historical tunnel operational data (such as tunnel length and speed limits), without the need to develop additional tunnel-specific models. Therefore, its output predicted trajectory not only serves as a benchmark for interference assessment but can also assist in determining the train's electromagnetic interference status.

[0035] Step S4: Based on the judgment result of the electromagnetic interference state, constrain and correct the output difference of the train positioning system.

[0036] In this embodiment of the invention, different interference states are obtained based on the judgment results of electromagnetic interference state, thereby performing differentiated corrections on the output of the train positioning system in a targeted manner, eliminating abnormal data output from the root cause, improving the reliability of positioning data to over 99%, and meeting the zero-abnormality requirement of the train control system for data reliability.

[0037] The method provided in this invention, through a closed-loop process of "trackside sign visual detection and correction - LSTM trajectory prediction - deviation judgment and interference - dynamic constraint correction," has multiple benefits: Deep learning-based visual detection can accurately identify trackside signs, and combined with existing positioning data, effectively offset GNSS electromagnetic interference deviations and odometer cumulative errors; the LSTM network predicts continuous trajectories based on historical operating data and current train operation information, forming anti-interference redundancy and preventing positioning interruptions due to the failure of a single data source; by quantifying deviations to determine the electromagnetic interference state, targeted constraint strategies are implemented, filtering abnormal data to ensure reliable positioning while avoiding "one-size-fits-all" speed reductions, balancing operational safety and efficiency; no large-scale modification of existing hardware is required, adapting to different line scenarios, significantly reducing engineering and maintenance costs, and providing an efficient solution for precise train positioning in complex electromagnetic environments.

[0038] Specifically, step S1 above includes the following steps: Step S11: Using a high-speed camera installed at the front of the train, capture real-time images of trackside signs around the train's running line and output learning samples and real-time detection images. Step S12: A deep convolutional generative adversarial network is used to extract features from the trackside sign learning samples and train them to generate fake trackside sign samples that are highly similar to real samples. The fake samples are then embedded in appropriate positions in the railway track background image to expand the dataset.

[0039] In this embodiment of the invention, a small amount of actual trackside sign data is first obtained, and then the trackside signs are extracted from this actual data. The extracted trackside signs are used as learning samples, such as... Figure 2 As shown, a Deep Convolutional Generative Adversarial Network (DCGAN) algorithm is used for training to generate trackside sign samples. These samples are then embedded into appropriate locations within a railway track background image to obtain newly constructed trackside sign data. The discriminator primarily determines the authenticity of the input data, while the generator generates fake data that closely resembles the real sample data. During network training, both real and fake data are input into the discriminator, and the parameters of both the generator and discriminator are continuously optimized based on the discriminator's judgment. When the discriminator makes a correct judgment, the generator parameters are further optimized to generate fake data that more closely approximates the real data; conversely, when an incorrect judgment is obtained, the discriminator parameters are optimized to provide a more accurate judgment. Through this continuous, balanced, and adversarial optimization, the data generated by the generator ultimately becomes very close to the real sample data.

[0040] This invention addresses the problem of low recognition accuracy caused by insufficient trackside sign samples in complex scenarios (dim tunnels, blurry rainy days) by using a high-speed camera to collect trackside signs in real time and combining this with a DCGAN-enhanced dataset. Improved SSD detectors and template matching methods ensure the accuracy of sign detection, positioning, and recognition. Kalman filtering fuses multi-source data and dynamically adjusts the noise covariance matrix, enabling real-time adaptation to train motion and electromagnetic interference intensity, effectively offsetting the random bias and cumulative error of the original positioning system.

[0041] Step S13: An improved single-anchor-point multi-border detector is used to detect and locate the trackside signs. The trackside signs are segmented by the vertical projection method and the signs are identified by the template matching method.

[0042] Specifically, because the detection target is singular—a trackside marker—and there is virtually no overlap between detection targets, to avoid overfitting that might occur with complex network structures, this embodiment of the invention removes the Z-shaped network and cascaded residual meshes from the original network, extracting only the backbone network. To further improve detection efficiency, the number of anchor boxes in the model is reduced from 4 to 2. Experimental analysis shows that this does not reduce detection accuracy in application scenarios and can even improve efficiency. The structure of the trackside marker detection and localization model based on the improved SSD algorithm is as follows: Figure 3 As shown.

[0043] The detection model mainly consists of a backbone network (BBN) and a backbone detection network (BDN). The BBN primarily comprises an optimized VGG-16 base network and additional convolutional layers. The optimization of the VGG-16 base network mainly involves replacing the last two fully connected layers with convolutional layers of different settings. This not only reduces the number of parameters and model runtime but also enhances the model's discriminative ability. The resolution of the feature maps in layers C3, C4, and C5 is reduced by 8, 16, and 32 times respectively compared to the original image. Furthermore, to further increase the receptive field, capture richer contextual information, and perform effective multi-scale detection, an additional convolutional layer is added to deepen the network and further reduce the resolution of the feature layers by a factor of 2, named C6, which, together with C3, C4, and C5, constitutes a multi-scale feature network.

[0044] By improving the single-anchor-point multi-frame detector, we achieve "reduced complexity, increased efficiency, and guaranteed accuracy" in both network structure and parameters. The Z-shaped network and cascaded residual mesh are removed to reduce redundant computation and avoid overfitting; the number of anchor frames is reduced from 4 to 2, significantly improving detection speed while maintaining accuracy, meeting the real-time requirements of high-speed train operation; the VGG-16 algorithm is optimized and convolutional layers are added to create multi-scale features, enabling accurate capture of trackside signs of different sizes and distances, solving the problem of insufficient multi-scale detection capability of traditional detectors in complex scenarios, achieving a sign recognition accuracy of over 95%, and providing a reliable data foundation for positioning correction. Figure 4 The image shown is a segmented trackside sign icon obtained by projecting a grayscale image using the color and texture features of the trackside sign. By employing vertical projection, the position of each character can be determined through grayscale projection of the target image, thus achieving effective segmentation of the sign.

[0045] Step S14: The Kalman filter algorithm is used to fuse the position and speed information of the original train positioning system with the trackside sign detection and recognition results, and output the optimized train position and speed information. The process noise covariance matrix and observation noise covariance matrix in the Kalman filter algorithm are dynamically adjusted according to the real-time motion state of the train and the intensity of electromagnetic interference. In this embodiment of the invention, Kalman filtering is used to fuse the position and speed information of the original train positioning system with the position and speed information detected and identified by the trackside markers, so as to obtain more accurate position and speed information under electromagnetic interference.

[0046] The specific parameter settings and dynamic adjustment rules for the Kalman filter algorithm are illustrated below, including: State vector (X) k ):X k =Train position, v k =Train speed, a k =Train acceleration; Observation vector (Z) k ):z 1,k =Original positioning system observation position, z 2,k = Trackside marker detection location, z 3,k =The speed observed by the original positioning system; State transition matrix (F) k ): Describes the transition relationship from the state at time k-1 to the state at time k, derived based on kinematic formulas; Observation matrix (H) k ): Describes the mapping relationship between the state vector and the observation vector at time k; Process noise (W) k State errors caused by uncertainties in train motion (such as wheelset slippage and changes in track gradient); Observation noise (V) k ): Observation errors caused by sensor accuracy (such as position deviation of GNSS due to electromagnetic interference, pixel errors in trackside mark recognition); The mathematical expression for the state equation is:

[0047] ; The state transition matrix is:

[0048] The mathematical expression for the observation equation is:

[0049] The observation matrix is:

[0050] The initialization and update rules for the noise covariance matrix in this embodiment of the invention include: 1. Process noise covariance matrix Q k The dimension is the same as the state vector (3×3), the diagonal elements correspond to the process noise variance of "position, velocity, acceleration", and the off-diagonal elements are 0 (assuming that the process noise of each state is uncorrelated).

[0051] Initialization rules: Initial values ​​for train motion characteristics and track parameters must be set according to train design parameters (such as maximum acceleration and wheelset slip rate) and track characteristics (such as gradient and curve radius), as shown in the following example:

[0052] Dynamic update rules: Adaptively adjust based on the current motion state.

[0053] Q k The filtering accuracy needs to be dynamically adjusted according to the train's motion (e.g., constant speed / acceleration / deceleration, straight section / gradient) to avoid a decrease in filtering accuracy caused by a fixed value. Rule 1: The greater the acceleration fluctuation, The larger; Rule 2: The greater the gradient of the route, The larger; Rule 3: During the constant velocity phase, Q k Keep the minimum value.

[0054] 2. The initialization and update rules for the observation noise covariance matrix in this embodiment of the invention include: Observation noise covariance matrix R k The dimension is the same as the state vector (3×3). The diagonal elements correspond to the observation noise variance of "original positioning system position, trackside marker position, and original positioning system velocity", while the off-diagonal elements are 0 (assuming that the noise of each observation is uncorrelated).

[0055] Initialization rules: Based on the sensor's accuracy specifications, the initial values ​​need to be set according to the factory accuracy or field calibration results of the two types of data sources, as shown in the example below:

[0056] Dynamic update rules: Based on the consistency between electromagnetic interference status and observations, R k The adjustment needs to be dynamic based on the intensity of electromagnetic interference and the consistency of observation data. The core logic is that the more reliable the data source, the better the R... k The smaller.

[0057] Rule 1: When GNSS is subjected to electromagnetic interference, increase... and ; Rule 2: When trackside markings are blurry, increase the size. ; Rule 3: When there is no interference, R k Keep the initial values ​​and prioritize trusting the trackside flags.

[0058] In one embodiment, step S2, which predicts the train's trajectory using a long short-term memory network based on historical train operation data and current train operation information, includes the following steps: 1. Data Input: Multi-dimensional time series dataset Suppose we extract historical data for a certain line (such as a section of the Beijing-Shanghai High-Speed ​​Railway) from the "historical train operation database" and construct a time series dataset.

[0059] (1) Time step: 1 second / step, a single sample contains 300 consecutive time steps (5 minutes of running data). (2) Input features (dimensions at each time step): (3) Historical trajectory characteristics: location (kilometer marker, such as K100+500), speed (km / h, such as 80), acceleration (m / s², such as 0.2); (4) Line constraint characteristics: line gradient (e.g., 0‰, straight section), curve radius (e.g., ∞, straight line); (5) Timetable characteristics: planned arrival time at the next station (e.g., 10 minutes remaining), planned speed (e.g., 70 km / h); (6) Output labels: the position (e.g., K100+522.22) and speed (e.g., 80.5km / h) of the next time step (301st second).

[0060] 2. LSTM Network Training and Prediction Process (1) Data set partitioning: The historical data is divided into a training set (70%, used for model parameter learning), a validation set (20%, used to adjust hyperparameters), and a test set (10%, used to verify accuracy) in a ratio of 7:2:1. (2) Network structure design: Input layer: Receives a 300×5 feature matrix (300 time steps, 5 features per time step); LSTM layer: 2 stacked layers, each with 64 hidden units (capturing time dependencies at different granularities, such as short-term velocity fluctuations and long-term position trends); Fully connected layer: maps the hidden state output by LSTM to the prediction result of "position + velocity"; Training process: The loss function (such as mean squared error MSE, minimizing the deviation between the predicted and actual positions) is optimized through backpropagation. The gating mechanism of LSTM learns autonomously: the forget gate retains key historical information such as "kilometer marker K100" and "slope of straight section", and discards accidental speed fluctuations (such as the instantaneous speed drop to 75km / h due to wheel slippage); the input gate combines the current timetable "10 minutes remaining to the next station" to update the speed prediction (gradually adjusting from 80km / h to 75km / h). 3. Real-time prediction: Input the data in the current driving information database (such as current position K100+500, speed 80km / h, 10 minutes remaining to the next station) into the trained LSTM, and output the running trajectory in the next 10 minutes (such as position and speed every 1 second).

[0061] The core logic of this invention, which compares the corrected train positioning information with the predicted trajectory, is to calculate the quantization deviation of the two types of input data in the dimensions of position and speed, and dynamically set an error threshold based on the data confidence level / prediction uncertainty to determine whether the current positioning data is reliable.

[0062] In one example, step S4, the process of constraining and correcting the output differences of the train positioning system based on the judgment result of the electromagnetic interference state, includes the following steps: Step S41: Calculate the positional deviation between the corrected position and the predicted position, and the velocity deviation between the corrected velocity and the predicted velocity.

[0063] Specifically, for the positioning data at each time step, the position deviation ΔP and velocity deviation ΔV are calculated separately, using the following formulas: ΔP = P1 - P2, ΔV = V1 - V2, Where P1 represents the calculated position of the train position and velocity, and P2 represents the predicted position of the LSTM; V1 represents the output velocity of the train position and velocity calculation module, and V2 represents the predicted velocity of the LSTM. Step S42: Dynamically set the position deviation threshold based on the recognition confidence level of the trackside marker.

[0064] Specifically, the error threshold in this embodiment of the invention is not a fixed value. It needs to be dynamically adjusted in combination with the trackside sign recognition confidence (C1) and LSTM prediction uncertainty (σ). The higher the data reliability, the stricter the threshold; the lower the reliability, the more lenient the threshold, in order to avoid misjudgment.

[0065] Specifically, the position deviation threshold is strongly correlated with the trackside sign recognition confidence level (C1) (the higher the C1, the more accurate the sign recognition, and the stricter the threshold). See the formula below:

[0066] Among them, T p0 The base position threshold is represented by C1 (based on the accuracy of the trackside sign, such as 1m under standard operating conditions, meaning the allowable deviation is ≤1m when the sign is clearly identifiable); C1 represents the trackside sign recognition confidence level (0-1, output by the train position and speed calculation module, such as C1=0.9 when the sign is clear and C1=0.3 when it is blurry). Step S43: Dynamically set the speed deviation threshold based on the speed uncertainty predicted by the long short-term memory network.

[0067] Specifically, the velocity deviation threshold and the LSTM prediction uncertainty (σ) (Speed ​​prediction variance) correlation (σ) The smaller the threshold, the more reliable the LSTM prediction (and the stricter the threshold). (Formula reference:)

[0068] Where, TV0: base speed threshold (set based on train running stability, such as 1 km / h under standard operating conditions, meaning that when LSTM prediction is reliable, the allowable deviation is ≤1 km / h); σ LSTM speed prediction variance (0-0.5, the smaller the variance, the more accurate the prediction; for example, on straight roads, σ) =0.1, complex slope σ =0.4).

[0069] In this embodiment of the invention, based on whether the quantization deviation is less than a dynamic threshold, three types of judgment results are output as the basis for subsequent correction strategies, as shown in Table 1 below: Table 1

[0070] Correction strategies need to combine judgment results and data reliability, prioritizing more reliable data sources or reducing bias through fusion algorithms. Specifically, this can be divided into three scenarios: Scenario 1: Normal matching (no interference) - Weighted fusion output When the two types of data are identical, no correction is needed. Weighted fusion is used to further improve positioning accuracy. The fusion formula is as follows: Fusion location: P_fusion = W1×P1+W2×P2 Fusion speed: V_fusion = W1×V1 + W2×V2 Where W1+W2=1 (W1 is the weight of trackside fusion data, and W2 is the weight of LSTM prediction).

[0071] Scenario 2: Slight Deviation (Slight Interference) - Single-Dimensional Correction When only one dimension (position / velocity) exceeds the limit, the deviation dimension is corrected based on reliable data from the other dimension: Speed ​​deviation (ΔV ≥ T) Location dimension reliability (ΔP) <T p Using P1 as the absolute position reference, the velocity is recalculated using the "rate of change of position" to replace the original V1 or V2.

[0072] Position deviation (ΔP ≥ T) p ): Reliability in the velocity dimension (ΔV) <T Using V2 (LSTM predicted velocity, continuously smoothed) as a baseline, position is predicted by velocity integration, replacing the original P1.

[0073] Scenario 3: Severe Deviation (Strong Interference) – Switch Data Source or Emergency Constraints When the two types of data are completely inconsistent, it indicates that the current positioning system is under strong interference (such as GNSS lock-up or camera inability to recognize markers). It is necessary to switch to a more reliable data source first, or trigger emergency constraint logic, as shown in Table 2: Table 2

[0074] This invention enables real-time acquisition of trackside markings and processing of historical operating data and current train operation information. It achieves assisted positioning based on deep learning and train operation information constraints, solves the problem of existing train positioning systems being susceptible to electromagnetic interference, optimizes the anti-interference capability of train positioning systems, and has profound practical significance for the safe operation of railway trains.

[0075] This embodiment also provides a train positioning anti-interference device, which is used to implement the above embodiments and preferred embodiments; details already described will not be repeated. As used below, the term "module" can refer to a combination of software and / or hardware that implements a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0076] This embodiment provides a train positioning anti-interference device, such as... Figure 5 As shown, it includes: The sign recognition and correction train positioning module 51 is used to detect and recognize the acquired trackside signs through deep learning-based visual detection methods, and correct the original train positioning system output based on the detection and recognition results to obtain optimized train position and speed information. The train trajectory prediction module 52 is used to predict the train's trajectory based on historical train operation data and current train operation information using a long short-term memory network. The electromagnetic interference judgment module 53 is used to compare the optimized train position and speed information with the predicted running trajectory, and judge the electromagnetic interference status based on the deviation of the comparison result. The positioning correction module 54 is used to constrain and correct the output differences of the train positioning system based on the judgment result of the electromagnetic interference state.

[0077] The train positioning anti-interference device provided in this embodiment of the invention can execute the train positioning anti-interference method provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects for executing the method. Further functional descriptions of the various modules and units described above are the same as in the corresponding embodiments described above, and will not be repeated here.

[0078] Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention.

[0079] The following is a detailed reference. Figure 6 This diagram illustrates a suitable structural design for implementing an electronic device according to embodiments of the present invention. The electronic device may include a processor (e.g., a central processing unit, graphics processor, etc.) 601, which can perform various appropriate actions and processes based on a program stored in read-only memory (ROM) 602 or a program loaded from memory 608 into random access memory (RAM) 603. RAM 603 also stores various programs and data required for the operation of the electronic device. The processor 601, ROM 602, and RAM 603 are interconnected via a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.

[0080] Typically, the following devices can be connected to I / O interface 605: input devices 606 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 607 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; memory devices 608 including, for example, magnetic tapes, hard disks, etc.; and communication devices 609. Communication device 609 allows electronic devices to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 6 Electronic devices with various devices are shown, but it should be understood that it is not required to implement or have all of the devices shown, and more or fewer devices may be implemented or have instead.

[0081] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device 609, or installed from a memory 608, or installed from a ROM 602. When the computer program is executed by the processor 601, it performs the functions defined in the train positioning anti-interference method of the embodiments of the present invention.

[0082] Figure 6 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments of the present invention.

[0083] This invention also provides a computer-readable storage medium. The methods described above according to embodiments of the invention can be implemented in hardware or firmware, or implemented as computer code that can be recorded on a storage medium, or implemented as computer code downloaded via a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and then stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code. When the software or computer code is accessed and executed by the computer, processor, or hardware, the train positioning anti-interference method shown in the above embodiments is implemented.

[0084] A portion of this invention can be applied as a computer program product, such as computer program instructions, which, when executed by a computer, can invoke or provide the methods and / or technical solutions according to the invention through the operation of the computer. Those skilled in the art will understand that the forms in which computer program instructions exist in a computer-readable medium include, but are not limited to, source files, executable files, installation package files, etc. Correspondingly, the ways in which computer program instructions are executed by a computer include, but are not limited to: the computer directly executing the instructions, or the computer compiling the instructions and then executing the corresponding compiled program, or the computer reading and executing the instructions, or the computer reading and installing the instructions and then executing the corresponding installed program. Here, the computer-readable medium can be any available computer-readable storage medium or communication medium accessible to a computer.

[0085] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A train positioning anti-interference method, characterized in that, include: The trackside signs are detected and identified using deep learning-based visual detection methods, and the original train positioning system output is corrected based on the detection and identification results to obtain optimized train position and speed information. Based on historical train operation data and current train operation information, the long short-term memory network is used to predict the train's trajectory. The optimized train position and speed information is compared with the predicted running trajectory, and the electromagnetic interference status is determined based on the deviation of the comparison result. Based on the judgment results of the electromagnetic interference state, the output differences of the train positioning system are constrained and corrected.

2. The method according to claim 1, characterized in that, The process involves detecting and recognizing trackside signs using deep learning-based visual detection methods, and then correcting the original train positioning system output based on the detection and recognition results to obtain corrected train position and speed information. This includes: By using a high-speed camera installed at the front of the train, images of trackside signs around the train's route are captured in real time, and learning samples and real-time detection images are output. A deep convolutional generative adversarial network is used to extract features from the trackside sign learning samples and train them to generate fake trackside sign samples with high similarity to real samples. The fake samples are then embedded in appropriate positions in the railway track background image to expand the dataset. An improved single-anchor-point multi-border detector is used to detect and locate trackside signs. The trackside signs are segmented by vertical projection and the signs are identified by template matching. The Kalman filter algorithm is used to integrate the position and speed information of the original train positioning system with the trackside sign detection and recognition results to output optimized train position and speed information. The process noise covariance matrix and observation noise covariance matrix in the Kalman filter algorithm are dynamically adjusted according to the real-time motion state of the train and the intensity of electromagnetic interference.

3. The method according to claim 2, characterized in that, The improved single-anchor-point multi-boundary detector network includes at least one of the following improvements: the Z-shaped network and cascaded residual meshes in the original network structure are removed; the number of preset anchor boxes in the model is reduced from 4 to 2; the backbone network includes an optimized VGG-16 base network and additional convolutional layers. The VGG-16 base network replaces the last two fully connected layers with convolutional layers. The additional convolutional layers are used to increase the receptive field, reduce the feature layer resolution by a factor of 2, and are named C6. Together with the C3, C4, and C5 layers of VGG-16, they constitute multi-scale features to achieve multi-scale detection.

4. The method according to claim 2, characterized in that, The parameter setting and dynamic adjustment rules of the Kalman filter algorithm include: The state vector is set to the train position, speed, and acceleration, and the observation vector is set to the original positioning system observation position, trackside marker detection position, and original positioning system observation speed. Process noise covariance matrix Q k It is a 3×3 matrix, with diagonal elements corresponding to the process noise variance of position, velocity, and acceleration, and off-diagonal elements being 0. It is initialized based on train design parameters and line characteristics, and dynamically adjusted according to train acceleration fluctuations, line gradient changes and operation stages, maintaining the minimum value during the uniform speed stage. Observation noise covariance matrix R k This is a 3×3 matrix. The diagonal elements correspond to the observation noise variance of the original positioning system position, trackside marker position, and original positioning system velocity. The off-diagonal elements are 0. It is initialized based on sensor accuracy indicators and dynamically adjusted according to electromagnetic interference intensity, trackside marker clarity, and observation data consistency. The data source is R. k The more reliable it is, the smaller it is.

5. The method according to claim 1, characterized in that, The step of comparing the corrected train positioning information with the predicted running trajectory includes: calculating the position deviation between the corrected position and the predicted position, and the speed deviation between the corrected speed and the predicted speed; dynamically setting a position deviation threshold based on the recognition confidence of the trackside markers; and dynamically setting a speed deviation threshold based on the speed uncertainty predicted by the long short-term memory network.

6. The method according to claim 5, characterized in that, Based on the judgment result of the electromagnetic interference state, the output difference of the train positioning system is constrained and corrected, including: If both the position deviation and speed deviation are less than their corresponding dynamic thresholds, the corrected train positioning information and the predicted trajectory are weighted and fused, and the fusion result is output. If only one of the position deviation or speed deviation exceeds its corresponding threshold, the dimension data that exceeds the threshold is corrected based on the dimension data that does not exceed the threshold. If both the position deviation and speed deviation exceed their corresponding thresholds, the preferred data source is selected based on the recognition confidence of the trackside sign and the uncertainty of the prediction by the long short-term memory network, or emergency constraint logic is triggered.

7. The method according to claim 1, characterized in that, The historical train operation data includes one or more of the following: historical location, historical speed, historical acceleration, track gradient, and curve radius; the current train operation information includes one or more of the following: train timetable, timetable, and planned speed.

8. A train positioning anti-interference device, characterized in that, The device includes: The sign recognition and correction train positioning module is used to detect and recognize trackside signs using deep learning-based visual detection methods, and correct the original train positioning system output based on the detection and recognition results to obtain optimized train position and speed information. The train trajectory prediction module is used to predict the train's trajectory based on historical train operation data and current train operation information, using a long short-term memory network. The electromagnetic interference judgment module is used to compare the optimized train position and speed information with the predicted running trajectory, and judge the electromagnetic interference status based on the deviation of the comparison result. The positioning correction module is used to constrain and correct the output differences of the train positioning system based on the judgment result of the electromagnetic interference state.

9. An electronic device, characterized in that, include: A memory and a processor are communicatively connected, the memory stores computer instructions, and the processor executes the computer instructions to perform the train positioning anti-interference method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing the computer to execute the train positioning anti-interference method according to any one of claims 1 to 7.

Images