Intelligent dispatching operation management and control method and system based on railway small operation scheduling

Abstract

The invention belongs to the technical field of small railway operation, provides an intelligent dispatching operation management and control method and system based on small railway operation scheduling, and aims to solve the technical problems that small operation track foreign matter invasion detection is lagged in transmission, and dispatching response is not timely. Comprising the steps that multiple sensors are deployed to collect foreign matter invasion data, a track circuit is linked to dynamically regulate and control the sampling frequency, a low-delay transmission channel is constructed through 5G-A + TSN, a priority mechanism and edge-cloud collaborative storage are combined, double coordinates are constructed based on a lightweight LSTM model and detection data, prediction precision is optimized, and the conflict relation between a train and foreign matter is analyzed. According to the method, the data timeliness and the prediction accuracy are improved, rapid response and safe disposal of foreign matter invasion are achieved, and safe and efficient operation of the small running railway is guaranteed.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of railway small operation, and in particular to an intelligent dispatching operation management and control method and system based on railway small operation scheduling. BACKGROUND

[0002] Track intruding objects (such as falling rocks, scattered goods, equipment parts falling off, and personnel entering by mistake, etc.) are major hidden dangers that threaten the safety of railway transportation, especially in the small operation scenario. Small operation trains are mainly responsible for short-distance transportation between stations in the hub, technical stations and intermediate stations. The operation environment has the characteristics of short section, dense track, frequent shunting, complex terrain (such as mountainous areas, tunnel groups, and shunting yard blind areas), etc. Once an intruding object event occurs, the disposal window left for the dispatching system is extremely short (usually only a few seconds to tens of seconds), and it needs to rely on real-time and accurate data to support rapid decision-making.

[0003] In the existing track intruding object detection and dispatching technology, data collection relies on a single type of sensor (such as trackside cameras, millimeter wave radars), and data processing mainly uses remote center end centralized analysis. However, this technical architecture exposes fatal defects in the small operation scenario: on the one hand, there are problems such as signal blind area in mountainous areas, electromagnetic interference in shunting yards, etc. in the small operation section, resulting in unstable intruding object data transmission link and transmission delay up to several seconds or even tens of seconds; on the other hand, the center end needs to process a large amount of raw data (such as high-definition video stream, radar point cloud), the processing flow is complex and time-consuming, and it is difficult to match the characteristics of high-frequency dynamic changes in the small operation environment.

[0004] The above problems directly lead to a core technical dilemma: when the track has intruding objects, the detection data often fails to match the transmission delay and processing speed, and by the time it reaches the dispatching end, it has completely lost its timeliness - that is, the data reflects the position and state of the intruding object, which has seriously deviated from the actual track environment (for example, the data shows that the intruding object is at position A, but the actual intruding object has rolled to position B due to train vibration, or the train has approached the area where the intruding object is located).

[0005] This data timeliness failure has become a prominent technical problem in the railway industry, especially in the field of small operation dispatching: the decisions (such as braking instructions, track avoidance schemes) generated by the dispatching system based on outdated data either cannot be executed due to lag or cause secondary risks (such as insufficient emergency braking distance due to misjudgment of the position of the intruding object, and collision due to incorrect occupation of the standby track) due to inconsistency with the actual environment, ultimately leading to dispatching lag, causing operation interruption and efficiency drop, or even causing train collision, derailment and other safety accidents.

[0006] At present, the existing technology has not effectively solved this problem, and there is an urgent need for a technical solution that can break through the data transmission and processing timeliness bottleneck and still achieve accurate dispatching in the extreme scenario where intruding object data may fail.

[0007] Therefore, this invention provides an intelligent scheduling and operation management method and system based on railway short-distance shift scheduling. Summary of the Invention

[0008] In order to overcome the shortcomings of the prior art, at least one technical problem raised in the background art is solved.

[0009] The technical solution adopted by this invention to solve its technical problem is: One of the objectives of this invention is to provide an intelligent scheduling and operation management method based on railway short-haul scheduling, comprising: S10: Deploy detection equipment at key nodes of the short-running track to identify and upload foreign object intrusion data, link the detection equipment with the track circuit signals, and dynamically control the sampling frequency of the detection equipment. S20: Based on the 5G-A private network and the Time Sensitive Network (TSN), a fast channel for uploading foreign object intrusion data is constructed, and a priority preemption mechanism for foreign object intrusion data is established. Foreign object intrusion data and non-foreign object intrusion data are prioritized and the highest priority foreign object intrusion data is uploaded first. At key nodes of the small operation track, edge cache nodes are deployed to establish a collaborative mechanism for local storage and cloud upload of foreign object intrusion data. S30: By using a historical foreign object movement database, a lightweight LSTM model is constructed to predict foreign object intrusion parameters. Combined with the foreign object intrusion data identified by the detection equipment, a dual coordinate system is constructed. Based on the dual coordinate system, the lightweight LSTM model is optimized to ensure that the deviation between the predicted trajectory and the actual trajectory is within a preset range. S40: Based on the dual coordinates and the arrival time of the shuttle train, analyze whether there is a conflict between the shuttle train and the foreign object intrusion, and adjust the shuttle train according to the analysis results.

[0010] As a further improvement of the present invention, the specific process of deploying detection equipment at key nodes of the small-scale operation track is as follows: Millimeter-wave radar, high-definition cameras, and vibration sensors are deployed at key nodes of the small-scale operation track to collect and detect data on foreign object intrusion. All detection equipment includes: millimeter-wave radar, high-definition camera, and vibration sensor directly connected to the track edge sensing node. The millimeter-wave radar collects the reflective cross-sectional area (RCS) and three-dimensional position coordinates of the foreign object; the high-definition camera collects the color, texture, and shape visual features of the foreign object; the vibration sensor collects the vibration waveform energy and main frequency distribution dynamic characteristics of the rail. The foreign object intrusion data includes: the real-time specific coordinates of the foreign object.

[0011] As a further improvement of the present invention, the specific process of dynamically adjusting the sampling frequency of the detection device is as follows: The detection equipment is connected to the track circuit signal system at the hardware level to obtain the track circuit occupancy status and real-time train speed data in real time. When the track circuit signal shows that it is occupied and the speed of the small train is <15km / h, it is determined to be a high-frequency detection demand field; when the track circuit signal shows that it is not occupied and the speed of the small train is greater than 15km / h, it is determined to be a low-frequency detection demand field. In high-frequency scenarios: The frame rate of the millimeter-wave radar has been dynamically increased from the conventional 30 frames per second to 100 frames per second, and the angular resolution has been simultaneously optimized to 0.5°. The frame rate of the high-definition camera has been increased from the usual 25 frames per second to 100 frames per second to avoid image ghosting caused by camera movement; in normal scenes, it will drop back to 25 frames per second. The sampling rate of the vibration sensor has been increased from the conventional 200Hz to 500Hz, and the low-frequency filtering channel has been turned off to fully capture the subtle vibration waveforms generated by the rolling of foreign objects; in normal scenarios, it has fallen back to 200Hz.

[0012] As a further improvement of the present invention, the specific process of constructing a fast data upload channel for foreign object intrusion based on 5G-A private network and Time Sensitive Network (TSN) is as follows: A fast data upload channel for foreign object intrusion is constructed through a 5G-A private network and a Time-Sensitive Network (TSN), wherein: Time-Sensitive Networking (TSN): Adopting an industrial-grade Ethernet architecture, TSN switches are deployed in communication cabinets along the track. Through a preset time-triggered mechanism, the deterministic latency of data transmission is guaranteed to be ≤5 milliseconds.

[0013] As a further improvement of the present invention, the specific process of prioritizing foreign object intrusion data and non-foreign object intrusion data, and uploading the highest priority foreign object intrusion data first, is as follows: All data collected and transmitted by the small-scale track inspection equipment is prioritized, with foreign object intrusion data marked as the highest priority and assigned a dedicated transmission channel: Set data transmission scheduling rules at communication nodes along the track: When foreign object intrusion data and non-foreign object intrusion data exist simultaneously, the foreign object intrusion data is marked as the highest priority and the non-foreign object intrusion data is marked as the lowest priority. The transmission equipment will prioritize scheduling the foreign object intrusion data, directly occupy the communication bandwidth, and bypass the transmission queue of non-urgent data.

[0014] As a further improvement of the present invention, the specific process of constructing the dual coordinates is as follows: A lightweight LSTM model is constructed, with 2 LSTM layers, 1 fully connected layer, and 3 neurons in the output layer, corresponding to the longitudinal and lateral displacements of the foreign object within a preset time period, thereby compressing the model volume to a preset range. The historical foreign object movement database was divided into a training set, a validation set, and a test set in an 8:1:1 ratio. The training set was used for model parameter learning, and the validation set was used for hyperparameter tuning. The training parameters were set as follows: the optimizer was Adam, the initial learning rate was 0.001, the batch size was set to 32, and the number of iterations was preset to 200 rounds to complete the model training. After the detection device uploads the foreign object intrusion data, the edge computing node calls the lightweight LSTM model to output the longitudinal and lateral displacements of the foreign object within a preset time period in the future. Based on the real-time specific coordinates of the foreign object, combined with the longitudinal and lateral displacements of the foreign object within a preset future time, the predicted coordinates of the foreign object at the preset future time are calculated. Finally, a dual coordinate system is constructed, which includes: the predicted coordinates at the preset future time and the real-time specific coordinates of the foreign object.

[0015] As a further improvement of the present invention, the specific process of ensuring that the deviation between the predicted trajectory and the actual trajectory is within a preset range is as follows: In the input feature vector set of the lightweight LSTM model, new feature vectors are added for the curvature radius of the curve and the angle between the initial velocity direction of the foreign object and the tangent of the curve. When the foreign object enters the curve region with a curvature radius ≤ 300m, the centrifugal force influence factor built into the lightweight LSTM model is increased from the baseline value of 0.2 to 0.4. The weights of the curve-related feature vectors in the input gate and forget gate of the LSTM unit are strengthened. The prediction time window of the lightweight LSTM model is fixed at 3 seconds, and the time step is optimized to 3 steps. When the remaining track distance is less than 50 meters, the lightweight LSTM model automatically triggers the longitudinal displacement prediction amplitude attenuation mechanism, dynamically reducing the longitudinal displacement prediction value according to the proportion of the remaining distance, ensuring that the deviation between the predicted trajectory and the actual trajectory is within the preset range.

[0016] As a further improvement of the present invention, the specific process of analyzing whether there is a conflict between the small-scale train and the foreign object intrusion is as follows: Foreign object movement time refers to the duration during which a foreign object moves from its real-time specific coordinates to a predicted coordinate at a future preset time. Arrival time refers to the total time required for the miniature train to travel from its current location to the track section corresponding to the predicted future time coordinates of the object.

[0017] If the train's arrival time is greater than the foreign object's movement time, it is determined that the short-distance train and the foreign object's intrusion into the limit do not have a conflict relationship. If the train's arrival time is less than or equal to the time the foreign object moves, it is determined that there is a conflict between the short-distance train and the foreign object's intrusion.

[0018] As a further improvement of the present invention, the specific process of adjusting the small-scale train based on the analysis results is as follows: If the analysis results indicate a conflict between the short-distance train and the foreign object encroaching on the limit, then the train control system will apply emergency braking to the short-distance train.

[0019] The second objective of this invention is to provide an intelligent dispatching and operation management system based on railway short-haul scheduling, comprising: Data acquisition module: Detection equipment is deployed at key nodes of the small-scale operation track to identify and upload data on foreign object intrusion. The detection equipment is linked with the track circuit signals for adjustment, and the sampling frequency of the detection equipment is dynamically controlled. Delayed transmission module: Based on 5G-A private network and time-sensitive network (TSN), a fast channel for uploading foreign object intrusion data is constructed, and a priority preemption mechanism for foreign object intrusion data is established. Foreign object intrusion data and non-foreign object intrusion data are prioritized and uploaded first. Edge cache nodes are deployed at key nodes of the small operation track to establish a collaborative mechanism for local storage and cloud upload of foreign object intrusion data. Dual coordinate generation module: By using a historical foreign object movement database, a lightweight LSTM model is constructed to predict foreign object intrusion parameters and combined with the foreign object intrusion data identified by the detection device to construct dual coordinates. Based on the dual coordinates, the lightweight LSTM model is optimized to ensure that the deviation between the predicted trajectory and the actual trajectory is within a preset range. Mini-train control module: Based on dual coordinates and the arrival time of the mini-train, analyze whether there is a conflict between the mini-train and the foreign object intrusion limit, and control the mini-train according to the analysis results.

[0020] The beneficial effects of this invention are as follows: The combination of multi-sensor collaborative acquisition and dynamic sampling control accurately captures core information such as the location, material, and motion state of foreign objects while avoiding redundant data, balancing detection accuracy and equipment power consumption. A 5G-A+TSN dual-mode transmission channel and data priority mechanism ensure low-latency, uninterrupted transmission of foreign object intrusion data, overcoming signal transmission bottlenecks in complex track topologies. Lightweight LSTM model optimization and dual coordinate construction control the deviation of foreign object trajectory prediction within a preset range, providing accurate data support for conflict analysis and adapting to real-time computing needs at the edge. Quantitative analysis of conflict relationships and linkage with emergency braking control enable rapid handling of foreign object intrusion risks, solving the problem of traditional scheduling lag, significantly reducing the incidence of safety accidents such as train collisions and derailments, and improving the safety and efficiency of short-haul railway scheduling and operation. Attached Figure Description

[0021] The invention will now be further described with reference to the accompanying drawings.

[0022] Figure 1 This is a flowchart of the steps of the intelligent scheduling, operation and management method based on railway short-distance shift scheduling of the present invention; Figure 2 This is a system module diagram of the intelligent dispatching, operation and management system based on railway short-distance shift scheduling of the present invention. Detailed Implementation

[0023] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0024] Example 1

[0025] like Figure 1 As shown in the embodiment of the present invention, the intelligent scheduling and operation management method based on railway short-haul scheduling includes: S10: Deploy detection equipment at key nodes of the short-running track to identify and upload foreign object intrusion data, link the detection equipment with the track circuit signals, and dynamically control the sampling frequency of the detection equipment. In S10, detection equipment is deployed at key nodes of the small-scale operation track to identify and upload data on foreign object intrusion. Millimeter-wave radar, high-definition camera and vibration sensor triple detection equipment are deployed at key nodes of the short-distance track (including but not limited to: entrance and exit of dedicated line, throat area of ​​marshalling yard, inside of curve, etc.) to collect and detect foreign object intrusion data; All detection equipment, including millimeter-wave radar, high-definition cameras, and vibration sensors, is directly connected to the track edge sensing node. The edge sensing node is used for localized processing of foreign object intrusion, uploading only the foreign object intrusion data to reduce the amount of data uploaded. The millimeter-wave radar collects the reflective cross-section (RCS) and three-dimensional position coordinates of the foreign object (based on the track coordinate system); the high-definition camera collects the color, texture, and shape visual features of the foreign object; the vibration sensor collects the vibration waveform energy and main frequency distribution dynamic characteristics of the rail. The foreign object intrusion data includes the real-time specific coordinates of the foreign object. The process of identifying and uploading foreign object intrusion data is illustrated below: A multi-feature fusion convolutional neural network (CNN) classification model can be constructed. The above-mentioned millimeter-wave radar features, high-definition camera visual features, and vibration sensor dynamic features are used as inputs, and the foreign object intrusion / non-intrusion is used as the output label for training, so that the model has the ability to distinguish whether a foreign object has intruded. The multi-feature fusion convolutional neural network (CNN) classification model is trained by inputting real-time data collected from millimeter-wave radar, high-definition cameras, and vibration sensors. If the multi-feature fusion convolutional neural network (CNN) classification model determines that there is a foreign object intrusion, then the core parameters of the foreign object (including but not limited to: the specific coordinates of the foreign object, the material type of the foreign object (metal / non-metal), and the motion state of the foreign object (stationary / rolling)) are extracted, and uploaded to the scheduling system after lightweight encoding; If the multi-feature fusion convolutional neural network (CNN) classification model determines that there is no foreign object intrusion, the data upload process will be terminated, the original data will be retained locally (the retention time is configurable) and then automatically discarded to reduce the amount of invalid data transmission. In S10, the specific process of dynamically adjusting the sampling frequency of the detection equipment by linking the detection equipment with the track circuit signal is as follows: In response to the characteristics of frequent starts and stops and intensive shunting operations of short-haul trains, the sampling frequency of the detection equipment (millimeter-wave radar, high-definition camera, vibration sensor) is dynamically adjusted by linking it with track circuit signals. The detection equipment (millimeter-wave radar, high-definition camera, vibration sensor) is connected to the track circuit signal system at the hardware level to obtain the occupancy status (occupied / unoccupied) of the track circuit and the real-time speed data of the train (synchronously output by the track circuit or on-board speed measuring device). The detection equipment is connected to the track circuit signal. When the track circuit signal shows that it is occupied and the speed of the small-scale train is <15km / h (shunting locomotive start-up and shutdown, small-scale train low-speed operation scenario), it is determined to be a high-frequency detection demand field; when the track circuit signal shows that it is not occupied and the speed of the small-scale train is greater than 15km / h (shunting locomotive start-up and shutdown, small-scale train low-speed operation scenario), it is determined to be a low-frequency detection demand field. In high-frequency scenarios, the frame rate of the millimeter-wave radar is dynamically increased from the conventional 30 frames / second to 100 frames / second, and the angular resolution is simultaneously optimized to 0.5° to ensure real-time tracking of minute displacements of foreign objects (such as a rolling speed of 0.1 m / s); in conventional scenarios, it drops back to 30 frames / second. In high-frequency scenarios, the frame rate of the high-definition camera is increased from the usual 25 frames per second to 100 frames per second to avoid image ghosting caused by camera adjustment movements; in normal scenarios, it drops back to 25 frames per second. In high-frequency scenarios, the sampling rate of the vibration sensor is increased from the conventional 200Hz to 500Hz, the low-frequency filtering channel is turned off, and the subtle vibration waveforms generated by the rolling of foreign objects are fully captured; in conventional scenarios, the sampling rate drops back to 200Hz, and the 50Hz power frequency filter is turned on to suppress electromagnetic interference. S20: Based on the 5G-A private network and the Time Sensitive Network (TSN), a fast channel for uploading foreign object intrusion data is constructed, and a priority preemption mechanism for foreign object intrusion data is established. Foreign object intrusion data and non-foreign object intrusion data are prioritized and the highest priority foreign object intrusion data is uploaded first. At key nodes of the small operation track, edge cache nodes are deployed to establish a collaborative mechanism for local storage and cloud upload of foreign object intrusion data. In S20, the specific process of constructing a fast channel for uploading foreign object intrusion data based on the 5G-A private network and Time Sensitive Network (TSN) is as follows: To address the complex topology of short-haul railway tracks, including tunnels and blind spots on dedicated lines, a rapid data upload channel for foreign object intrusion is constructed. The specific construction process is as follows: A fast data upload channel for foreign object intrusion is constructed through a 5G-A private network and a Time-Sensitive Network (TSN), wherein: Time-Sensitive Networking (TSN): Adopting an industrial-grade Ethernet architecture, Time-Sensitive Networking (TSN) switches are deployed in communication cabinets along the track. Through a preset time triggering mechanism, the deterministic latency of data transmission is strictly guaranteed to be ≤5 milliseconds. For example, in sections with high incidence of foreign object intrusion and complex track topology, such as curves and turnouts, Time-Sensitive Networking (TSN) can precisely control the forwarding sequence of data at each communication node through a preset time-triggered mechanism: When the detection device captures foreign object intrusion data in this section, the TSN switch prioritizes allocating forwarding resources for this data according to the pre-configured time slots, ensuring that the transmission delay from the edge sensing node to the edge decision node is strictly ≤5 milliseconds, completely avoiding delay fluctuations caused by chaotic forwarding sequences of multiple nodes (traditional Ethernet often experiences delay fluctuations of 20-50 milliseconds in similar scenarios); for example, if a metallic foreign object rolls towards the track at a curve due to centrifugal force, its intrusion data can be transmitted from the detection node to the decision node within 0.1 seconds through the TSN channel, gaining a critical window for the dispatcher to issue avoidance instructions; while if traditional transmission methods are used, delay fluctuations may cause instructions to lag by 1-2 seconds, significantly increasing safety risks; 5G-A Private Network: 5G-A base stations are deployed in marshalling yards and dedicated line hub areas to achieve wide-area coverage along the track. At the same time, 5G-A has strong anti-electromagnetic interference capabilities and can be adapted to strong electromagnetic environments such as railway traction power supply systems, ensuring the stability of data transmission in scenarios outside tunnels and at the edge of blind spots. The collaborative working process between 5G-A private network and Time-Sensitive Network (TSN) is as follows: Within the tunnel, time-sensitive network (TSN) is prioritized to ensure low-latency transmission, while 5G-A private network is used to supplement coverage outside the blind zone. This forms a collaborative transmission mode of deterministic low latency within the tunnel and wide-area stable coverage outside the blind zone, ensuring uninterrupted and low-latency transmission of data intruded by foreign objects throughout the entire process. In S20, the specific process of prioritizing foreign object intrusion data and non-foreign object intrusion data, and uploading the highest priority foreign object intrusion data first, is as follows: All data collected and transmitted by small-scale track inspection equipment (millimeter-wave radar, high-definition camera, vibration sensor, etc.) is prioritized, with foreign object intrusion data marked as the highest priority (priority level 1) and assigned a dedicated transmission channel: Configure data transmission scheduling rules at communication nodes (such as relay stations and edge gateways) along the track: When foreign object intrusion data and non-foreign object intrusion data (such as daily monitoring data of equipment status) coexist, the foreign object intrusion data is marked as the highest priority (priority level 1), and the non-foreign object intrusion data is marked as the lowest priority (priority level 0). The transmission equipment will prioritize scheduling the foreign object intrusion data with priority 1, so that it directly occupies the communication bandwidth and bypasses the transmission queue of non-urgent data. For example, when a metal foreign object is detected encroaching on a section of dedicated track, its data can be transmitted in queue, reaching the decision node 3-5 times faster than ordinary data, avoiding additional delays caused by network congestion. In S20, the specific process of deploying edge cache nodes at key nodes of the small-scale operation track and establishing a collaborative mechanism for local storage and cloud upload of foreign object intrusion data is as follows: At key nodes of the short-distance track (including but not limited to: dedicated line entrances and exits, marshalling yard throat area, inside of curves, etc.), deploy edge cache nodes (using high-performance industrial servers with local storage and data forwarding functions) to establish a collaborative mechanism for local storage and cloud upload of foreign object intrusion data; When foreign object intrusion data is transmitted to this node, it will be temporarily stored locally first, and then pushed to the cloud dispatch center simultaneously through a high-speed link. If a decision-making command needs to be issued quickly (such as train braking or locomotive avoidance), the foreign object intrusion data can be retrieved directly from the edge cache node and the command can be generated to ensure the speed of decision issuance. For long-term data backup or global analysis, the cloud can also synchronously obtain complete foreign object intrusion data, achieving the dual effect of rapid local response and global cloud backup; Foreign object data is temporarily stored locally and simultaneously pushed to the cloud, ensuring both the speed of decision-making and the preservation of data backups; S30: By using a historical foreign object movement database, a lightweight LSTM model is constructed to predict foreign object intrusion parameters. Combined with the foreign object intrusion data identified by the detection equipment, a dual coordinate system is constructed. Based on the dual coordinate system, the lightweight LSTM model is optimized to ensure that the deviation between the predicted trajectory and the actual trajectory is within a preset range. In S30, the specific process of constructing a lightweight LSTM model to predict foreign object intrusion limits using a historical foreign object movement database and combining this with the foreign object intrusion limit data identified by the detection equipment to construct a dual coordinate system is as follows: A lightweight LSTM model is constructed, consisting of two LSTM layers (each containing 64 neurons), one fully connected layer (containing 32 neurons), and three output neurons, corresponding to the longitudinal and lateral displacements of the foreign object within a preset time (e.g., 3 seconds). Model pruning techniques are employed to remove connection weights that contribute less than 5% to the prediction results, compressing the model size to a preset range (e.g., within 5MB) to meet the requirements of edge computing. Based on known and non-foreign object intrusion data within the historical cycle of railway short-haul scheduling, a historical foreign object movement database was constructed. This database was divided into a training set, a validation set, and a test set in an 8:1:1 ratio. The training set was used for model parameter learning, and the validation set was used for hyperparameter tuning. The training parameters were set as follows: the optimizer was Adam, the initial learning rate was 0.001 (decreasing to 0.5 times the original rate if the validation set loss did not decrease for 5 consecutive rounds), the batch size was set to 32, and the number of iterations was preset to 200 rounds (training stopped when the validation set accuracy was ≥90% and the mean squared error (MSE) <0.01). Model training was then completed. The trained lightweight LSTM model is deployed on the trackside edge computing node (integrated with detection equipment and edge cache nodes) to provide computing power support for real-time prediction of foreign object trajectories; After the detection device uploads the foreign object intrusion limit data, the edge computing node calls the lightweight LSTM model to perform real-time inference and outputs the foreign object intrusion limit prediction parameters, namely the longitudinal and lateral displacement of the foreign object within a preset time (such as 3 seconds). Extract the real-time specific coordinates of the foreign object (e.g., K1+200m, +0.3m) from the foreign object intrusion data identified by the detection equipment. Combine this with the longitudinal and lateral displacements of the foreign object within a preset future time to calculate the predicted coordinates of the foreign object at the preset future time (e.g., K1+215m, +0.5m). Finally, construct a dual coordinate system (i.e., the dual coordinate system includes: the predicted coordinates at the preset future time and the real-time specific coordinates of the foreign object). In S30, the lightweight LSTM model is optimized based on dual coordinates to ensure that the deviation between the predicted trajectory and the actual trajectory is within a preset range. The specific process is as follows: In the input feature vector set of the lightweight LSTM model, two new core feature vectors are added: the radius of curvature of the curve and the angle between the initial velocity direction of the foreign object and the tangent of the curve. Differentiated weight coefficients are assigned for curve scenarios. When a foreign object enters a curved area with a curvature radius ≤300m, the centrifugal force influence factor built into the lightweight LSTM model dynamically increases from the baseline value of 0.2 to 0.4, accurately depicting the influence of the curve on the lateral movement trend of the foreign object. The weights of the curve-related feature vectors in the input gate and forget gate of the LSTM unit are strengthened (weight coefficients are increased by 20% to 30%) to enhance the lightweight LSTM model's ability to learn the lateral displacement patterns of foreign objects in curve scenarios and ensure the accuracy of trajectory prediction at curves. The prediction time window of the fixed lightweight LSTM model is 3 seconds (matching the critical time period of emergency response for short-haul trains). The time step is optimized from the traditional 5 steps to 3 steps, focusing on the instantaneous trajectory changes of foreign objects in a short period of time (such as sudden changes in rolling direction and speed fluctuations), thereby improving the model's real-time fitting accuracy for dynamic scenes. A new feature vector representing the remaining track length (i.e., the straight-line distance from the current position of the foreign object to the end of the track) is added and embedded into the decision logic of the model output layer: when the remaining track distance is less than 50 meters, the lightweight LSTM model automatically triggers a longitudinal displacement prediction amplitude attenuation mechanism, dynamically reducing the longitudinal displacement prediction value according to the proportion of the remaining distance (e.g., attenuating by 30% when there is 30 meters remaining), avoiding prediction deviations caused by physical obstructions at the end of the track, and ensuring that the deviation between the predicted trajectory and the actual trajectory is within a preset range (e.g., the difference between the predicted value and the actual value is no more than 10 centimeters). It should be noted that when building the lightweight LSTM model, a dynamic adjustment mechanism for the centrifugal force influence factor is built in: the centrifugal force influence factor is used to quantify the influence of the centrifugal force on the lateral movement trend of the foreign object in the curve, and the baseline value is set to 0.2 (corresponding to the straight track scenario). When a foreign object is detected entering a curve with a radius of curvature ≤300m, the lightweight LSTM model automatically and dynamically increases the centrifugal force influence factor from a baseline value of 0.2 to 0.4 to accurately characterize the effect of the curve on the lateral movement of the foreign object. At the same time, in the input gate and forget gate of the LSTM unit, the weights of the feature vectors related to the curve (including features associated with the centrifugal force influence factor) are strengthened (the weight coefficient is increased by 20% to 30%), thereby enhancing the model's ability to learn the lateral displacement pattern of foreign objects in curve scenarios.

[0026] Through the above multi-dimensional optimization, the lightweight LSTM model can stably control the prediction deviation of foreign object trajectory within a preset range of ≤10 cm in complex scenarios with small operating tracks (especially curves and track end areas). S40: Based on the dual coordinates and the arrival time of the shuttle train, analyze whether there is a conflict between the shuttle train and the foreign object intrusion, and adjust the shuttle train according to the analysis results. In S40, the specific process for analyzing whether there is a conflict between the short-haul train and the foreign object encroachment is as follows: The object movement duration refers to the duration during which an object moves from its real-time specific coordinates to its predicted coordinates at a future preset time. The object movement duration is consistent with the preset prediction time of the lightweight LSTM model (such as the preset 3 seconds mentioned above, i.e., the object movement duration is fixed at 3 seconds). It should be noted that the foreign object movement duration is essentially the time window in which the foreign object is in the intrusion state corresponding to the predicted coordinates—that is, from the current moment until a preset time in the future (such as 3 seconds), the foreign object will continue to be in the track intrusion area where the predicted coordinates are located (if the foreign object is dynamically rolling, the predicted coordinates will be dynamically updated with time, but its movement duration is always equal to the model's preset prediction time). Arrival time refers to the total time required for the minibus to travel from its current location to completely pass through the track section corresponding to the predicted coordinates of the foreign object at the future preset time. The calculation requires combining the real-time speed of the train and the length of the track section (track section length = train length + track safety redundancy length, the safety redundancy length is preset to 2 meters to ensure that the train body completely passes through the encroachment area). The specific calculation logic is as follows: Arrival time = (mileage of the starting point of the track section corresponding to the predicted coordinates of the foreign object at the future preset time - mileage of the train's current location) ÷ real-time speed of the train; For example, if the current location of the train is K1+150m, the starting point of the track section corresponding to the predicted coordinates of the foreign object at the preset time is K1+200m, and the real-time speed of the train is 10km / h (≈2.78m / s), then the arrival time = (200-150)÷2.78≈18 seconds.

[0027] The determination of whether there is a conflict is based on the overlap between the time window of foreign object intrusion (i.e., the time of foreign object movement) and the time window of train arrival and passage through the intrusion area (i.e., the arrival time). If the train arrival time is greater than the foreign object movement time (e.g., arrival time 18 seconds > foreign object movement time 3 seconds), it means that when the train has completely passed through the track section corresponding to the predicted coordinates of the foreign object, the foreign object has exceeded the intrusion time window preset by the lightweight LSTM model (or has been cleared). There is no spatial or temporal overlap between the two, and it is determined that the small-scale train and the foreign object intrusion do not have a conflict relationship. If the train arrival time is less than or equal to the foreign object movement time (e.g., 2 seconds of arrival time is less than or equal to 3 seconds of foreign object movement time), it means that the train has entered and started to pass through the intrusion area within the time window when the foreign object is still in the intrusion state corresponding to the predicted coordinates. There is a spatial and temporal overlap between the two, and it is determined that the small-scale train and the foreign object intrusion have a conflict relationship. It should be noted that the calculation of arrival time must focus on the core requirement of the train passing through completely. The track section on which the calculation is based not only includes the location of the predicted coordinates of the foreign object, but also covers the track range corresponding to the length of the train body (e.g., if the train length is 15 meters, then the track section length = 15 meters + 2 meters safety redundancy = 17 meters). This avoids ignoring the collision risk between the train body and the foreign object by only calculating the arrival point of the train head. At the same time, the foreign object movement time strictly matches the preset prediction time of the lightweight LSTM model (e.g., 3 seconds) to ensure consistency with the technical parameters in the trajectory prediction section above, and to ensure the continuity and accuracy of the analysis process. In S40, the specific process of adjusting the short-distance trains based on the analysis results is as follows: If the analysis results indicate that the short-distance train and the foreign object intrusion into the limit are in conflict, then the short-distance train will be braked urgently through the train control system. Example 2 like Figure 2 As shown in Embodiment 1, the present invention provides an intelligent dispatching and operation management system based on railway short-haul scheduling, comprising: Data acquisition module: Detection equipment is deployed at key nodes of the small-scale operation track to identify and upload data on foreign object intrusion. The detection equipment is linked with the track circuit signals for adjustment, and the sampling frequency of the detection equipment is dynamically controlled. Delayed transmission module: Based on 5G-A private network and time-sensitive network (TSN), a fast channel for uploading foreign object intrusion data is constructed, and a priority preemption mechanism for foreign object intrusion data is established. Foreign object intrusion data and non-foreign object intrusion data are prioritized and uploaded first. Edge cache nodes are deployed at key nodes of the small operation track to establish a collaborative mechanism for local storage and cloud upload of foreign object intrusion data. Dual coordinate generation module: By using a historical foreign object movement database, a lightweight LSTM model is constructed to predict foreign object intrusion parameters and combined with the foreign object intrusion data identified by the detection device to construct dual coordinates. Based on the dual coordinates, the lightweight LSTM model is optimized to ensure that the deviation between the predicted trajectory and the actual trajectory is within a preset range. Mini-train control module: Based on dual coordinates and the arrival time of the mini-train, analyze whether there is a conflict between the mini-train and the foreign object intrusion limit, and control the mini-train according to the analysis results.

[0028] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for intelligent scheduling, operation and management based on railway short-haul shift scheduling, characterized by: include: S10: Deploy detection equipment at key nodes of the short-running track to identify and upload foreign object intrusion data, link the detection equipment with the track circuit signals, and dynamically control the sampling frequency of the detection equipment. S20: Based on the 5G-A private network and the Time Sensitive Network (TSN), a fast channel for uploading foreign object intrusion data is constructed, and a priority preemption mechanism for foreign object intrusion data is established. Foreign object intrusion data and non-foreign object intrusion data are prioritized and the highest priority foreign object intrusion data is uploaded first. At key nodes of the small operation track, edge cache nodes are deployed to establish a collaborative mechanism for local storage and cloud upload of foreign object intrusion data. S30: By using a historical foreign object movement database, a lightweight LSTM model is constructed to predict foreign object intrusion parameters. Combined with the foreign object intrusion data identified by the detection equipment, a dual coordinate system is constructed. Based on the dual coordinate system, the lightweight LSTM model is optimized to ensure that the deviation between the predicted trajectory and the actual trajectory is within a preset range. S40: Based on the dual coordinates and the arrival time of the shuttle train, analyze whether there is a conflict between the shuttle train and the foreign object intrusion, and adjust the shuttle train according to the analysis results.

2. The intelligent scheduling and operation management method based on railway short-haul scheduling according to claim 1, characterized in that: The specific process for deploying detection equipment at key nodes of the small-scale operational track is as follows: Millimeter-wave radar, high-definition cameras, and vibration sensors are deployed at key nodes of the small-scale operation track to collect and detect data on foreign object intrusion. All detection equipment includes: millimeter-wave radar, high-definition camera, and vibration sensor directly connected to the track edge sensing node. The millimeter-wave radar collects the reflective cross-sectional area (RCS) and three-dimensional position coordinates of the foreign object; the high-definition camera collects the color, texture, and shape visual features of the foreign object; the vibration sensor collects the vibration waveform energy and main frequency distribution dynamic characteristics of the rail. The foreign object intrusion data includes: the real-time specific coordinates of the foreign object.

3. The intelligent scheduling and operation management method based on railway short-haul shift scheduling according to claim 1, characterized in that: The specific process of dynamically adjusting the sampling frequency of the detection equipment is as follows: The detection equipment is connected to the track circuit signal system at the hardware level to obtain the track circuit occupancy status and real-time train speed data in real time. When the track circuit signal shows that it is occupied and the speed of the small train is <15km / h, it is determined to be a high-frequency detection demand field; when the track circuit signal shows that it is not occupied and the speed of the small train is greater than 15km / h, it is determined to be a low-frequency detection demand field. In high-frequency scenarios: The frame rate of the millimeter-wave radar has been dynamically increased from the conventional 30 frames per second to 100 frames per second, and the angular resolution has been simultaneously optimized to 0.5°. The frame rate of the high-definition camera has been increased from the usual 25 frames per second to 100 frames per second, avoiding image ghosting caused by camera adjustment movements; In normal scenarios, the frame rate drops back to 25 frames per second. The sampling rate of the vibration sensor has been increased from the conventional 200Hz to 500Hz, and the low-frequency filtering channel has been closed to fully capture the subtle vibration waveforms generated by the rolling of foreign objects. In normal scenarios, the frequency drops back to 200Hz.

4. The intelligent scheduling and operation management method based on railway short-haul shift scheduling according to claim 1, characterized in that: The specific process of constructing a fast data upload channel for foreign object intrusion based on 5G-A private network and Time Sensitive Network (TSN) is as follows: A fast data upload channel for foreign object intrusion is constructed through a 5G-A private network and a Time-Sensitive Network (TSN), wherein: Time-Sensitive Networking (TSN): Adopting an industrial-grade Ethernet architecture, TSN switches are deployed in communication cabinets along the track. Through a preset time-triggered mechanism, the deterministic latency of data transmission is guaranteed to be ≤5 milliseconds.

5. The intelligent scheduling and operation management method based on railway short-haul shift scheduling according to claim 1, characterized in that: The specific process of prioritizing foreign object intrusion data and non-foreign object intrusion data, and uploading the highest priority foreign object intrusion data first, is as follows: All data collected and transmitted by the small-scale track inspection equipment is prioritized, with foreign object intrusion data marked as the highest priority and assigned a dedicated transmission channel: Set data transmission scheduling rules at communication nodes along the track: When foreign object intrusion data and non-foreign object intrusion data exist simultaneously, the foreign object intrusion data is marked as the highest priority and the non-foreign object intrusion data is marked as the lowest priority. The transmission equipment will prioritize scheduling the foreign object intrusion data, directly occupy the communication bandwidth, and bypass the transmission queue of non-urgent data.

6. The intelligent scheduling and operation management method based on railway short-haul shift scheduling according to claim 1, characterized in that: The specific process for constructing the dual coordinates is as follows: A lightweight LSTM model is constructed, with 2 LSTM layers, 1 fully connected layer, and 3 neurons in the output layer, corresponding to the longitudinal and lateral displacements of the foreign object within a preset time period, thereby compressing the model volume to a preset range. The historical foreign object movement database was divided into a training set, a validation set, and a test set in an 8:1:1 ratio. The training set was used for model parameter learning, and the validation set was used for hyperparameter tuning. The training parameters were set as follows: the optimizer was Adam, the initial learning rate was 0.001, the batch size was set to 32, and the number of iterations was preset to 200 rounds to complete the model training. After the detection device uploads the foreign object intrusion data, the edge computing node calls the lightweight LSTM model to output the longitudinal and lateral displacements of the foreign object within a preset time period in the future. Based on the real-time specific coordinates of the foreign object, combined with the longitudinal and lateral displacements of the foreign object within a preset future time, the predicted coordinates of the foreign object at the preset future time are calculated. Finally, a dual coordinate system is constructed, which includes: the predicted coordinates at the preset future time and the real-time specific coordinates of the foreign object.

7. The intelligent scheduling and operation management method based on railway short-haul shift scheduling according to claim 1, characterized in that: The specific process for ensuring that the deviation between the predicted trajectory and the actual trajectory is within a preset range is as follows: In the input feature vector set of the lightweight LSTM model, new feature vectors are added for the curvature radius of the curve and the angle between the initial velocity direction of the foreign object and the tangent of the curve. When the foreign object enters the curve region with a curvature radius ≤ 300m, the centrifugal force influence factor built into the lightweight LSTM model is increased from the baseline value of 0.2 to 0.

4. The weights of the curve-related feature vectors in the input gate and forget gate of the LSTM unit are strengthened. The prediction time window of the lightweight LSTM model is fixed at 3 seconds, and the time step is optimized to 3 steps. When the remaining track distance is less than 50 meters, the lightweight LSTM model automatically triggers the longitudinal displacement prediction amplitude attenuation mechanism, dynamically reducing the longitudinal displacement prediction value according to the proportion of the remaining distance, ensuring that the deviation between the predicted trajectory and the actual trajectory is within the preset range.

8. The intelligent scheduling and operation management method based on railway short-haul shift scheduling according to claim 1, characterized in that: The specific process for analyzing whether there is a conflict between the small-scale train and the foreign object intrusion is as follows: Foreign object movement time refers to the duration during which a foreign object moves from its real-time specific coordinates to a predicted coordinate at a future preset time. Arrival time refers to the total time required for the miniature train to travel from its current location to the track section corresponding to the predicted future time coordinates of the object; If the train's arrival time is greater than the foreign object's movement time, it is determined that the short-distance train and the foreign object's intrusion into the limit do not have a conflict relationship. If the train's arrival time is less than or equal to the time the foreign object moves, it is determined that there is a conflict between the short-distance train and the foreign object's intrusion.

9. The intelligent scheduling and operation management method based on railway short-haul shift scheduling according to claim 1, characterized in that: The specific process of adjusting the short-distance trains based on the analysis results is as follows: If the analysis results indicate a conflict between the short-distance train and the foreign object encroaching on the limit, then the train control system will apply emergency braking to the short-distance train.

10. An intelligent dispatching and operation management system based on railway short-distance shift scheduling, characterized in that: include: Data acquisition module: Detection equipment is deployed at key nodes of the small-scale operation track to identify and upload data on foreign object intrusion. The detection equipment is linked with the track circuit signals for adjustment, and the sampling frequency of the detection equipment is dynamically controlled. Delayed transmission module: Based on 5G-A private network and time-sensitive network (TSN), a fast channel for uploading foreign object intrusion data is constructed, and a priority preemption mechanism for foreign object intrusion data is established. Foreign object intrusion data and non-foreign object intrusion data are prioritized and uploaded first. Edge cache nodes are deployed at key nodes of the small operation track to establish a collaborative mechanism for local storage and cloud upload of foreign object intrusion data. Dual coordinate generation module: By using a historical foreign object movement database, a lightweight LSTM model is constructed to predict foreign object intrusion parameters and combined with the foreign object intrusion data identified by the detection device to construct dual coordinates. Based on the dual coordinates, the lightweight LSTM model is optimized to ensure that the deviation between the predicted trajectory and the actual trajectory is within a preset range. Mini-train control module: Based on dual coordinates and the arrival time of the mini-train, analyze whether there is a conflict between the mini-train and the foreign object intrusion limit, and control the mini-train according to the analysis results.

Images