Artificial intelligence-based track trolley intelligent control and scheduling method

Abstract

The application relates to an artificial intelligence-based track trolley intelligent control and scheduling method, and relates to the field of science popularization exhibition halls.The method comprises collecting track occupation state data and position data of each trolley, performing noise reduction and calibration on the collected data to obtain standardized data, dynamically allocating running routes for multiple trolleys and avoiding collisions according to pre-stored track layout data and the standardized data to generate a scheduling signal, controlling the start-stop, speed and steering of each trolley according to the scheduling signal, monitoring the running state in real time and outputting a warning signal to trigger shutdown in response to an exception, and synchronously driving the exhibition hall multimedia equipment to perform linkage display according to the running state of each trolley.The application enables multiple trolleys to dynamically avoid obstacles and run in a complex track network, and linkage of the multimedia equipment provides an immersive interactive experience.

Description

Technical Field

[0001] This application relates to the field of science popularization exhibition halls, and in particular to an intelligent control and scheduling method for track vehicles based on artificial intelligence. Background Technology

[0002] In multimedia interactive scenarios such as science exhibition halls and cultural tourism exhibition halls, track-mounted vehicles are widely used to carry exhibits or guide visitors along pre-set routes, conveying information to visitors in conjunction with displays and audio equipment within the exhibition hall. As the scale of exhibition halls expands and the demand for interaction increases, the single-vehicle, single-line display mode can no longer meet the needs of multiple visitors experiencing the exhibits simultaneously. The parallel operation of multiple vehicles on the same track network has become a common deployment method.

[0003] Existing exhibition hall track-based trolley systems mostly employ a fixed-program control scheduling method. This means that during deployment, each trolley is pre-programmed with a fixed route and timetable, and during operation, each trolley starts and stops strictly according to the pre-programmed sequence. This fixed scheduling method can meet basic operational needs on simple tracks with a single or small number of vehicles. However, when the track network has branching, crossing, or loop structures, and multiple trolleys are running in parallel, conflicts between fixed routes cannot be detected and resolved in real time. Because the fixed program lacks the ability to perceive the real-time occupancy status of the tracks, when a section of track is temporarily occupied due to the preceding trolley slowing down or stopping, subsequent trolleys still enter along their original routes, creating a collision risk. Maintenance personnel are forced to avoid conflicts by increasing the departure interval between vehicles or limiting the number of vehicles running simultaneously, which directly reduces the utilization rate of the track network and the exhibition hall's throughput.

[0004] Another limitation of fixed scheduling is the lack of autonomous identification and response capabilities for abnormal states. When obstacles appear on the track or a vehicle experiences a mechanical failure, the fixed program cannot autonomously identify the anomaly and take measures such as stopping the machine or reassigning the route. It can only rely on manual intervention by on-site personnel, resulting in a delayed response and potential safety hazards. Due to the lack of data linkage between the scheduling program and anomaly monitoring, even if the system is equipped with sensors, the data collected by the sensors is only used for post-event recording and cannot be fed back to the scheduling decision-making process in real time during operation.

[0005] The aforementioned shortcomings in scheduling capabilities also affect the multimedia interaction effects of the exhibition hall. Under fixed scheduling mode, the vehicle's operating status information cannot be transmitted to the exhibition hall's display and audio equipment in real time. The playback of multimedia content can only run independently according to a fixed schedule, and the display content cannot be dynamically switched according to the actual position of the vehicle. This results in the multimedia content seen by visitors being disconnected from the actual movement trajectory of the vehicle, leading to a stiff interactive experience. Summary of the Invention

[0006] In order to dynamically allocate operating routes for multiple vehicles based on the real-time track status, autonomously avoid collisions, monitor anomalies in real time, and link with multimedia equipment in the exhibition hall, this application provides an artificial intelligence-based intelligent control and scheduling method for track vehicles.

[0007] Firstly, this application provides an intelligent control and scheduling method for a railcar based on artificial intelligence, which adopts the following technical solution: An intelligent control and scheduling method for railcars based on artificial intelligence includes the following steps: S1. Collect track occupancy status data and position data of each trolley; S2. The occupancy status data and the position data are denoised and calibrated to remove invalid data and obtain standardized track status data and standardized trolley position data. S3. Based on the pre-stored track layout data, the standardized track status data, and the standardized vehicle position data, dynamically allocate running routes for multiple vehicles and avoid collisions, and generate scheduling signals; S4. Control the start, stop, speed and steering of each trolley according to the scheduling signal; S5. Monitor the running status and track status of each trolley in real time, and in response to the detection of an anomaly, output an early warning signal and trigger the corresponding trolley to stop; S6. Synchronously drive the multimedia equipment in the exhibition hall to display information based on the operating status of each vehicle.

[0008] By employing the aforementioned technical solution, noise reduction, calibration, and invalid data removal are performed on the track occupancy status data and the position data of each vehicle. The standardized data is then input into the AI ​​intelligent scheduling algorithm, transforming the input source for scheduling decisions from raw sensor signals into standardized data that has undergone quality control. Based on this, the scheduling algorithm dynamically allocates routes to multiple vehicles and avoids collisions according to the track layout data. The generated scheduling signals simultaneously drive both vehicle control and anomaly warning processes. The anomaly warning results are then fed back to the vehicle control layer to trigger a shutdown, forming a closed loop from data acquisition to scheduling decision-making to execution control to anomaly response. The multimedia interactive display is synchronously driven according to the operating status of each vehicle, establishing a real-time data channel between the audiovisual content of the exhibition hall and the physical movement of the vehicles, eliminating the problem of disconnect between multimedia content and vehicle positions in a fixed-time playback mode.

[0009] Optionally, in S3, each track segment in the track layout data is maintained in three states: idle, reserved, and occupied. When assigning the running route to the vehicle, the status of each track segment traversed by the running route is switched from the idle state to the reserved state; When the trolley enters the corresponding track segment, the state of the corresponding track segment is switched from the reserved state to the occupied state; When the trolley leaves the corresponding track segment, the state of the corresponding track segment is switched from the occupied state to the idle state; The track segment in the reserved state and the corresponding reserved time window information serve as inputs for allocating the running route and avoiding collisions.

[0010] By adopting the above technical solution, the status of each track segment is expanded from two types (idle and occupied) to three types: idle, reserved, and occupied. When assigning routes to the vehicles, the status of the target track segment is switched to reserved, allowing other vehicles to be aware of track segments that are about to be occupied during their path planning. Compared to path planning based solely on the current snapshot of occupancy, the three-state model incorporates the temporal dimension of track occupancy information into the planning stage, reducing route conflicts caused by tracks being idle during planning but occupied during execution. The reservation time window information also serves as input for route allocation and collision avoidance, allowing the downstream processes of cost calculation and collision detection to share the same set of reservation information, avoiding data inconsistencies caused by the two processes maintaining occupancy status independently.

[0011] Optionally, in S3, a passage cost is calculated for each track segment in the track layout data. The passage cost is a weighted composite of a track segment length factor, a current state factor, a historical congestion factor, and a fairness compensation factor. The current state factor of the track segment in the reserved state is higher than the current state factor of the track segment in the idle state and lower than the current state factor of the track segment in the occupied state; The historical congestion factor is determined based on the statistical distribution of the state transition timestamps of each track segment; When assigning the running route to the vehicle, the sum of the travel costs of each track segment is used as the objective function.

[0012] Optionally, in S3, for each car, based on the car's current position, speed, acceleration, and the speed limit constraints of each track segment ahead, the earliest arrival time and latest arrival time of the car to each track segment ahead are calculated to obtain the motion envelope. The motion envelope of each vehicle is mapped to a spatiotemporal conflict diagram with track segment number as the horizontal axis and time as the vertical axis. The spatiotemporal conflict diagram uses the track segment in the reservation state and the corresponding reservation time window as the initial occupied area. In response to the existence of an overlapping region between the motion envelopes of the two vehicles on the spatiotemporal conflict map, a collision risk is determined between the two vehicles.

[0013] By adopting the above technical solution, the motion envelope is calculated based on the current position, speed, acceleration, and speed limit constraints of each vehicle on the track ahead. This transforms collision detection from a point estimation method that compares the distance between two vehicles frame by frame to a region estimation method based on time intervals. The motion envelope defines the occupied time interval of each track segment using the earliest and latest arrival times, naturally accommodating the positional uncertainty caused by speed fluctuations and acceleration / deceleration, and avoiding missed detections caused by speed changes in the point estimation method. After mapping the motion envelope to a spatiotemporal conflict map with the track segment number as the horizontal axis and time as the vertical axis, collision detection is transformed into two-dimensional region overlap determination. The reserved track segments and reserved time windows are pre-filled into the conflict map as initial occupied areas, allowing the new vehicle's envelope projection to directly intersect with the pre-filled area for determination, eliminating the need for vehicle-by-vehicle and segment-by-segment comparisons.

[0014] Optionally, in S3, the running route of each car is divided into a locked segment and a candidate segment. The locked segment is a continuous sequence of track segments in front of the current position of the car and does not change during operation. The candidate segment is the remaining route segment after the locked segment. The number of track segments included in the locking segment is determined according to the current speed of the trolley; the higher the current speed, the more track segments the locking segment includes. In response to the collision risk occurring within the candidate segment, incremental replanning is performed on the candidate segment to generate a corrected path.

[0015] By adopting the above technical solution, the running route of each vehicle is divided into a locked segment and a candidate segment. The locked segment remains unchanged during operation, while the candidate segment allows for incremental replanning based on collision detection results. The length of the locked segment is dynamically adjusted according to the vehicle's current speed; the higher the speed, the longer the locked segment, ensuring that the distance covered by the locked segment is not less than the braking safety distance at the current speed. This division restricts the scope of path correction to the candidate segment, and the vehicle's trajectory within the locked segment is not affected by replanning, avoiding frequent changes in the vehicle's trajectory during the journey caused by full-path replanning. When a collision risk occurs within the candidate segment, incremental replanning is performed only on the candidate segment, reducing the search space from the full path to the track subgraph corresponding to the candidate segment, thus reducing the computational load of path search.

[0016] Optionally, in response to the collision risk occurring within the locking segment of the two vehicles, a yielding cost is calculated for each of the two vehicles, the yielding cost including braking distance cost, path extension cost, and cascading impact cost; The cascading cost is the additional overlap between the new path after yielding and other vehicles on the spatiotemporal conflict map. The party with the lowest yielding cost is selected as the yielding party. The locked segment of the yielding party is shortened to the current track segment of the yielding party, and the remaining part of the original locked segment is downgraded to the candidate segment and the incremental replanning is triggered.

[0017] By adopting the above technical solution, when a collision risk occurs within the locked segment of two vehicles, a negotiated yielding agreement is initiated because the locked segment cannot be avoided through candidate segment replanning. The yielding cost is a weighted composite of three components: braking distance cost, path extension cost, and cascading impact cost. The cascading impact cost incorporates the additional overlap between the new path after yielding and other vehicles on the spatiotemporal conflict map, ensuring that the yielding decision considers not only the direct costs of the conflicting parties but also the cascading impact on third-party vehicles. After the party with the lowest yielding cost yields, the yielding party's locked segment is shortened to its current track segment, and the remaining portion of the original locked segment is downgraded to a candidate segment and triggers incremental replanning. This allows the yielding party to begin calculating the new path during deceleration, eliminating the window period of passively waiting for the conflict segment to be released after yielding.

[0018] Optionally, before performing the incremental replanning or the yielding, the corrected path is projected onto the spatiotemporal conflict map to detect new overlapping areas with other vehicles. The number of vehicles involved in the new overlap is the affected range value. If the affected range value exceeds a preset affected threshold, the corresponding path correction scheme is abandoned. In response to the fact that the spillover range values ​​of all candidate path correction schemes exceed the preset spillover threshold, each vehicle involved in the collision risk reduces to a preset safe speed and maintains its current path.

[0019] By adopting the above technical solution, before performing incremental replanning or yielding, the corrected path is projected onto the spatiotemporal conflict map to pre-detect new overlaps with other vehicles, using the number of vehicles involved in the new overlap as the ripple range value. When the ripple range value exceeds a preset ripple threshold, the corresponding correction scheme is abandoned, avoiding the diffusion effect of a single path correction triggering a chain reaction of replanning for multiple vehicles. When the ripple range values ​​of all candidate correction schemes exceed the threshold, each vehicle involved in the collision risk reduces to a preset safe speed and maintains its current path, choosing not to modify any path as a fallback strategy. This feedforward pre-assessment shifts the detection of chain conflicts from post-correction remediation to pre-correction prevention, fundamentally avoiding the oscillating cycle of repeated modifications and the generation of new conflicts in the subsequent iterative resolution.

[0020] Optionally, when allocating the running route to the vehicle, a preset number of alternative routes ranked by passage cost value are cached at the same time to form a alternative route pool for the vehicle. When performing the incremental replanning, alternative paths that are compatible with the current locked segment are retrieved from the alternative path pool first. The compatibility condition is that the locked segment portion of the alternative path is consistent with the current locked segment. The alternative path pool is incrementally updated each time the trolley passes through a track segment, eliminating alternative paths that are incompatible with the updated locked segment and adding new alternative paths.

[0021] By adopting the above technical solution, when allocating routes for the vehicle, a preset number of alternative paths ranked by travel cost are cached simultaneously, forming an alternative path pool. During incremental replanning, alternative paths compatible with the currently locked segment are retrieved from the alternative path pool first. Once a match is found, the path is switched directly, reducing the response time for path correction from the path search level to the table lookup level. The alternative path pool is incrementally updated every time the vehicle passes through a track segment, eliminating alternative paths incompatible with the updated locked segment and adding new alternative paths. This ensures that the alternative path pool is always synchronized with the vehicle's current position and the locked segment status, avoiding the invalidation of alternative paths in the static cache due to changes in track status.

[0022] Optionally, in step S5, the warning threshold is dynamically adjusted according to the running density parameter, where the running density parameter is the ratio of the number of trolleys currently running on the track to the total number of track segments. In step S1, the acquisition frequency is adjusted in conjunction with the operating density parameter and the track area type where each vehicle is located. The adjustment of the warning threshold and the adjustment of the acquisition frequency are performed based on the same operating density parameter.

[0023] By adopting the above technical solution, the warning threshold and acquisition frequency are dynamically adjusted based on the operation density parameter, which is the ratio of the number of trolleys currently running on the track to the total number of track segments. The warning threshold and acquisition frequency are executed in conjunction with the same operation density parameter. When the operation density increases, the warning threshold tightens while the acquisition frequency increases simultaneously, ensuring that the temporal resolution of the location data relied upon for warning judgment matches the severity of the warning. When the operation density decreases, the warning threshold loosens while the acquisition frequency decreases simultaneously, reducing the data transmission and preprocessing load in low-density scenarios. This linkage mechanism avoids the mismatch that may occur when the warning threshold and acquisition frequency are adjusted independently. For example, when the warning threshold is tightened but the acquisition frequency remains low, the warning judgment relies on coarse location data, which can easily lead to misjudgments.

[0024] Optionally, in S3, before each car departs, a global running route is pre-allocated based on the track layout data and the current occupancy status data. During the operation of each car, the running route is locally corrected based on the real-time standardized track status data and standardized car position data. The running route generated in the pre-allocation stage is stored as a safe route after verification. In response to the scheduling response time exceeding the preset timeout threshold, the mode is switched to degraded operation mode, and each car runs at low speed in one direction along the safe route. The warning threshold is tightened to the most stringent level, and the multimedia linkage display is switched to the general loop playback mode.

[0025] By adopting the above technical solution, global operating routes are pre-allocated based on track layout data and current occupancy status data before each vehicle departs. During operation, the operating routes are locally corrected based on real-time standardized data. The scheduling calculation is divided into two stages: offline pre-allocation and online correction. The operating routes generated in the pre-allocation stage are stored as safe routes after verification. When the scheduling response time exceeds a preset timeout threshold, the system switches to a degraded operation mode, and each vehicle runs at low speed in one direction along the safe route. The pre-allocated routes are reused as degraded safe routes, eliminating the need to maintain a separate safe route dataset. Moreover, the degraded routes are calculated based on the most recent track status and remain synchronized with the current track layout. In the degraded operation mode, the warning threshold is tightened to the most stringent level, and the multimedia linkage display switches to a general loop playback mode, enabling the system to maintain basic safety monitoring and exhibition hall operation while degrading scheduling capabilities.

[0026] Optionally, in S6, an event-driven synchronization mechanism is adopted: multiple trigger point positions are preset in the track layout data, and a position event is generated in response to each car arriving at the trigger point position. The position event is also used for scheduling decisions and switching of multimedia linkage display content.

[0027] By adopting the above technical solution, the multimedia interactive display uses an event-driven synchronization mechanism. Multiple trigger point positions are preset in the track layout data. When each vehicle reaches a trigger point, a position event is generated. These position events are used simultaneously for scheduling decisions and switching of multimedia interactive display content. Compared to polling vehicle positions at a fixed frequency and continuously pushing synchronization commands to multimedia devices, the event-driven mechanism only generates communication when a vehicle reaches a preset trigger point. The operation between two trigger points does not incur synchronization communication overhead, reducing the communication volume of multimedia interaction from being proportional to the polling frequency to being proportional to the number of trigger points.

[0028] Optionally, S3 further includes: performing statistical analysis on the accumulated historical running trajectory during non-running periods, identifying track segments with high occupancy frequency and collision warning trigger rate, and updating the analysis results as the historical congestion factor in the passage cost calculation.

[0029] By employing the above technical solution, statistical analysis is performed on accumulated historical running trajectories during off-peak hours to identify track segments with high occupancy frequency and collision warning trigger rates. The analysis results are then updated into historical congestion factors in the passage cost calculation. In subsequent scheduling cycles, the cost function for path planning will automatically bypass track segments with a high concentration of historical collision warnings, upgrading path allocation from reactive decision-making based solely on the current state to predictive decision-making incorporating historical statistics. The update of historical congestion factors is performed offline during off-peak hours, without consuming real-time computing resources during runtime.

[0030] Optionally, the system also includes a visitor interaction terminal, which is configured to receive the exhibition theme selected by the visitor and transmit the selection signal corresponding to the exhibition theme to the AI ​​intelligent scheduling module; the AI ​​intelligent scheduling module adjusts the running route and stopping order of each vehicle according to the selection signal, and the multimedia linkage module synchronously switches the content of the corresponding exhibition theme.

[0031] By adopting the above technical solution, the system includes a visitor interaction terminal. After receiving the exhibition theme selected by the visitor, the terminal transmits the selection signal to the AI ​​intelligent scheduling module. The scheduling module adjusts the running routes and stopping order of each vehicle according to the selection signal, and the multimedia linkage module synchronously switches the content of the corresponding exhibition theme. This interactive closed loop expands the information flow of the exhibition hall from a one-way output from the system to the visitor to a two-way loop from visitor intention to scheduling decision to vehicle movement to multimedia content. This allows the running routes of the tracked vehicles to be dynamically adjusted in response to the visitor's real-time selection, rather than being solely preset by the system.

[0032] Optionally, the difference between the latest arrival time and the earliest arrival time of each track segment is determined by connecting the speed limit constraints of the corresponding track segment segment by segment, and the difference increases with the difference in the number of segments between the track segment and the track segment where each car is currently located. The look-ahead time range of the spatiotemporal conflict map is divided into a near-end window and a far-end window, and the length of the near-end window is equal to the passage time corresponding to the locked segment. In response to the overlapping region being located within the near-end window, the incremental replanning is performed on the candidate segment; in response to the overlapping region being located within the far-end window, a collision warning is recorded but the incremental replanning is not performed.

[0033] By adopting the above technical solution, the width of the motion envelope of each track segment, i.e., the difference between the latest arrival time and the earliest arrival time, is determined segment by segment based on the speed limit constraints of the corresponding track segment. The difference increases with the difference in the number of segments between the track segment and the current track segment of the vehicle. The envelope of the near-end track segment is narrow, the data is fresh and has low uncertainty, resulting in high accuracy and low false alarm rate in collision detection; the envelope of the far-end track segment is wide, accommodating the cumulative uncertainty of speed fluctuations and data delays, and preserving a safety margin. The look-ahead time range of the spatiotemporal conflict map is divided into a near-end window and a far-end window. The length of the near-end window is equal to the travel time corresponding to the locked segment, ensuring that the coverage of the near-end window is strictly aligned with the locked segment. Overlaps detected within the near-end window directly trigger incremental replanning, and the collision risk within the corresponding locked segment must be responded to immediately; overlaps detected within the far-end window only record a collision warning but do not trigger replanning, and the collision risk within the corresponding candidate segment is left to be reassessed after the vehicle advances to that segment and enters the near-end window. This alignment avoids overreacting to false overlaps at the far end, while maintaining responsiveness to real collision risks at the near end.

[0034] Optionally, an overlap area value is calculated for the overlapping region located within the near-end window, and the collision risk is divided into three levels based on the overlap area value: If the overlap area value is less than the first area threshold, it is marked as a slight overlap and no response is executed; In response to an overlap area value between a first area threshold and a second area threshold, it is marked as medium overlap. The speed of the vehicle involved in the collision risk is reduced to narrow the corresponding motion envelope. In the next detection cycle, the overlap area value of the narrowed motion envelope is recalculated and the level is updated. In response to an overlap area value greater than a second area threshold, the candidate segment is marked as strongly overlapping, and the incremental replanning is performed on the candidate segment.

[0035] By employing the above technical solution, the overlapping area value of overlapping regions within the near-end window is calculated and categorized into three levels: micro-overlap, medium-overlap, and strong-overlap. Micro-overlap does not trigger a response, filtering out false collision alarms caused by the uncertainty expansion of the envelope edge. Medium-overlap reduces collision risk by adjusting the vehicle's speed to narrow the corresponding motion envelope. In the next detection cycle, the overlapping area value of the narrowed envelope is recalculated and the level is updated, forming a feedback loop of deceleration, envelope narrowing, and reduced overlapping area. This allows some medium-risk collisions to be resolved automatically through speed fine-tuning without triggering path replanning. Strong-overlap directly triggers incremental replanning, corresponding to collision risks that must be resolved through path correction. This three-level classification matches the cost of collision response with the severity of collision risk, avoiding the waste of computational resources caused by triggering replanning for all overlaps.

[0036] Optionally, before performing the detection of the overlapping region, a coarse screening is performed on the motion envelope of each car in the track segment dimension to identify track segments through which the motion envelopes of two or more cars pass. Only the cars that pass through the same track segment are subjected to overlap detection in the time dimension. The common orbital segments are marked as hot segments, and the remaining orbital segments are marked as cold segments. The time axis of the hot segments in the spatiotemporal conflict diagram adopts a first time resolution, and the time axis of the cold segments in the spatiotemporal conflict diagram adopts a second time resolution. The first time resolution is higher than the second time resolution. The spatiotemporal conflict graph maintains a rolling window along the time axis. Incremental updates are performed on hot segments in each scheduling cycle, while the incremental update frequency of cold segments is lower than that of hot segments. A full reconstruction of the spatiotemporal conflict graph is performed every preset number of cycles.

[0037] By adopting the above technical solution, before performing overlapping region detection, a coarse screening is first performed on the motion envelopes of each vehicle at the track segment dimension to identify track segments where the motion envelopes of two or more vehicles share the same path. Only vehicle pairs sharing the same track segment are subjected to time-dimensional overlap detection. Since in typical exhibition hall scenarios, most track segments are traversed by only a single vehicle's envelope at any given time, the coarse screening can filter out most non-collision vehicle pairs, reducing the computational load of fine detection from being proportional to the number of vehicle pairs to being proportional only to the number of vehicle pairs sharing a common segment. The coarse screening results are also used to label hot and cold segments. Hot segments are displayed on the time axis of the spatiotemporal conflict graph with a fine-grained resolution to capture fine temporal overlaps, while cold segments are displayed with a coarse-grained resolution to reduce storage and computational overhead. The spatiotemporal conflict graph maintains a rolling window along the time axis. Incremental updates are performed on hot segments in each scheduling cycle, while the incremental update frequency for cold segments is lower than that for hot segments, concentrating the computational load of incremental updates on track segments with concentrated collision risks. Every preset number of cycles, a full reconstruction of the spatiotemporal conflict map is performed to correct incremental cumulative errors, so that the conflict map remains consistent with the actual orbital state after a long period of operation.

[0038] In summary, this application includes at least one of the following beneficial technical effects: 1. By performing noise reduction, calibration, and invalid data removal on track occupancy status data and vehicle position data, the data is input into the AI ​​scheduling algorithm to dynamically allocate running routes for multiple vehicles and avoid collisions. At the same time, it links anomaly warnings and multimedia displays, forming a closed loop from data acquisition to scheduling decision-making to execution control to anomaly response to multimedia linkage. This enables multiple vehicles to be dynamically scheduled and run in complex track networks according to real-time status, eliminating the problem that route conflicts cannot be detected and resolved in real time under fixed program control.

[0039] 2. By expanding the track segment status from two types (idle and occupied) to three types (idle, reserved, and occupied), the target track segment is marked as reserved during the route allocation stage. The reservation time window information is used as input for both passage cost calculation and collision detection. This allows the route planning to be aware of track segments that are about to be occupied, reducing route conflicts caused by track segments being idle during planning but occupied during execution.

[0040] 3. By calculating the motion envelope based on the car's current position, speed, acceleration, and speed limit constraints of each track segment, and mapping the motion envelope to a spatiotemporal conflict graph with track segment number as the horizontal axis and time as the vertical axis, two-dimensional overlap detection is performed. This transforms collision detection from frame-by-frame point estimation to time interval estimation that accommodates speed uncertainty. Combined with the segmentation of path-locked segments and candidate segments, the negotiated yielding protocol, and the feedforward pre-evaluation of the conflict impact range, the path correction maintains the stability of the car's motion while avoiding the spread of cascading replanning. Attached Figure Description

[0041] Figure 1 This is a flowchart of an intelligent control and scheduling method for a track vehicle based on artificial intelligence, according to one embodiment of the present invention.

[0042] Figure 2 This is a block diagram of an artificial intelligence-based intelligent control and scheduling system for a track vehicle according to one embodiment of the present invention.

[0043] Figure 3 This is a schematic diagram of a spatiotemporal conflict diagram and a motion envelope in one embodiment of the present invention.

[0044] Figure 4 This is a schematic diagram of the segmentation of the locked segment and the candidate segment in one embodiment of the present invention.

[0045] Figure 5 This is a flowchart for assessing the scope of impact in one embodiment of the present invention.

[0046] Explanation of reference numerals in the attached figures: 110. Track status acquisition module; 120. Data preprocessing module; 130. AI intelligent scheduling module; 140. Car control module; 150. Anomaly warning module; 160. Multimedia linkage module; 210. Motion envelope; 220. Overlapping area; 230. Initial occupied area; 310. Locked segment; 320. Candidate segment; 330. Collision risk location. Detailed Implementation

[0047] The present application will be further described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are merely illustrative of the application and are not intended to limit the scope of the application.

[0048] This application discloses an intelligent control and scheduling method for tracked vehicles based on artificial intelligence, applied to a multi-vehicle parallel scheduling scenario in a multimedia interactive exhibition hall. The tracked vehicle is a small, electrically driven vehicle that runs along a fixed track, with dimensions ranging from 0.3m×0.2m×0.15m to 0.5m×0.3m×0.2m. Driven by an onboard motor, its operating speed ranges from 0.5m / s to 2m / s. It is used to carry exhibit models or guide visitors along a preset route. The track layout data is organized in the form of a directed graph, where nodes correspond to the endpoints and bifurcation points of track segments, and directed edges correspond to track segments. Each edge carries a segment number, physical length, speed limit, and one-way / two-way attributes. The exhibition hall's multimedia equipment includes 3D displays and voice broadcasting devices installed in each exhibition area.

[0049] The following example uses a science exhibition hall with 20 track segments and 2 branching points as a scenario, with 3 cars operating simultaneously, numbered car A, car B, and car C. The track segments are numbered T1 to T20, with T7 and T14 being branching points. The speed limit for each segment is set between 0.8 m / s and 2 m / s based on the curve radius and the parking requirements of the exhibition booths.

[0050] like Figure 1 As shown, the method includes the following steps S1 to S6. S1 involves collecting track status data and trolley position data. The output of S1 is transmitted to S2 for noise reduction and calibration. The output of S2 is standardized track status data and standardized trolley position data, which are transmitted to S3. S3 dynamically allocates running routes for multiple trolleys and avoids collisions based on pre-stored track layout data and the output of S2, generating a scheduling signal. The scheduling signal is transmitted to S4 to control the start, stop, speed, and steering of each trolley. S5 monitors the running status of each trolley and the track status in real time. In response to the detection of an anomaly, a warning signal is output and the corresponding trolley is stopped. S6 synchronously drives the multimedia equipment in the exhibition hall for a linked display based on the running status of each trolley.

[0051] S1. Collect track occupancy status data and the position data of each trolley. Track occupancy status data is collected by laser sensors installed at the entrance and exit of each track segment. The laser sensors detect the presence of obstructions on the track segment by emitting and receiving laser beams. An obstruction corresponds to the track segment being occupied, and no obstruction corresponds to an empty segment. Trolley position data is collected by position encoders installed on the drive wheel axle of each trolley. The position encoders convert the number of rotations of the drive wheel into the absolute position coordinates of the trolley on the track. The basic value of the acquisition frequency is 10 frames / second, that is, all data is read from the sensors and encoders once every 0.1 seconds. The collected raw data is transmitted to the data preprocessing stage in real time.

[0052] In other embodiments, track occupancy data can also be collected by infrared sensors. Infrared sensors identify occupancy status by detecting changes in infrared reflection signals on the track segment, making them suitable for dusty exhibition hall environments, and this is also a method for collecting track occupancy data.

[0053] S2. Noise reduction and calibration are performed on the occupancy status data and position data to remove invalid data, resulting in standardized track status data and standardized trolley position data. Noise reduction uses a sliding window mean filter with a window length of 5 frames. The arithmetic mean of the sampled values ​​from 5 consecutive frames is taken as the filtered output of the current frame, eliminating noise caused by instantaneous sensor jitter. Calibration is performed on the zero-point offset of the position encoder. The system records the zero-point readings of each encoder before opening each day. During operation, the encoder output value is subtracted from the zero-point reading to obtain the calibrated position value. The invalid data removal condition is as follows: sampling points with a single-frame position deviation exceeding a preset jump threshold are considered invalid and discarded. The jump threshold is calculated based on the trolley's maximum speed and acquisition cycle. For example, a maximum speed of 2 m / s and an acquisition cycle of 0.1 seconds correspond to a maximum single-frame displacement of 0.2 m. The jump threshold is set to 1.5 times 0.2 m, i.e., 0.3 m. Single-frame displacements exceeding 0.3 m are considered abnormal encoder jumps. After noise reduction, calibration, and elimination, the data becomes standardized track status data and standardized vehicle position data, which are transmitted to the AI ​​intelligent scheduling stage via TCP / IP protocol.

[0054] S3. Based on pre-stored track layout data, standardized track state data, and standardized trolley position data, dynamically allocates operating routes for multiple trolleys and avoids collisions, generating scheduling signals. The track layout data is pre-stored in the system memory, including segment numbers T1 to T20 for 20 track segments, directed connections between segments, branch direction definitions for two bifurcation points T7 and T14, speed limits for each segment, and unidirectional / bidirectional attributes for each segment. The scheduling signals contain the segment sequence of the operating route assigned to each trolley, the target speed value for each segment, and start / stop instructions. The internal logic of S3 involves multiple collaborative mechanisms, including track segment state management, passage cost calculation, collision detection, and path correction, which will be elaborated below.

[0055] Optionally, in S3, each track segment in the track layout data has three states: idle, reserved, and occupied. In the traditional two-state model, path planning can only make decisions based on the current occupancy snapshot and cannot detect track segments that are about to be occupied. For example, if car A has just been assigned a route from T5 to T8 but has not yet entered T5, T5 to T8 is still shown as idle in the two-state model. Car B's path planning may include T5 to T8 in the optional route, leading to a conflict between the two cars during subsequent execution. The three-state model eliminates this blind spot by introducing the reserved state.

[0056] When assigning routes to the vehicles, the status of each track segment along the route is switched from idle to reserved. Taking vehicle A as an example, when the AI ​​intelligent scheduling assigns route T3-T5-T7-T10 to vehicle A, the status of T5, T7, and T10 changes from idle to reserved at the time of assignment completion. When vehicle A enters T5, the status of T5 changes from reserved to occupied. When vehicle A leaves T5 and enters T7, the status of T5 changes from occupied to idle, and the status of T7 changes from reserved to occupied. Each track segment performs these three-state transitions sequentially according to this rule.

[0057] After car A's route allocation is completed, during the time period before car A actually enters T5, car B's path planning reads that T5 to T10 are in a reserved state. These track segments are considered resources that will soon become unavailable, and the path planning automatically detours to an alternative route that does not include T5 to T10. The reserved track segments and their corresponding reservation time windows serve as input for both subsequent passage cost calculation and collision detection. The reservation time window information records the start time and expected release time of each track segment's reservation. The cost calculation stage adjusts the passage cost for that segment accordingly, and the collision detection stage pre-fills the initial occupied area 230 in the spatiotemporal conflict map based on this information.

[0058] Furthermore, in S3, a passage cost is calculated for each track segment in the track layout data. The passage cost is a weighted composite of the track segment length factor, the current state factor, the historical congestion factor, and the fairness compensation factor. The track segment length factor is the normalized value of the physical length of the track segment; the longer the physical length, the larger the track segment length factor.

[0059] The current state factor is assigned a value based on the three states of the orbital segment: The current state factor corresponding to the idle state is based on a value, such as 0.1; The current state factor corresponding to the reserved state is higher than that of the idle state, for example, 0.5; The current state factor corresponding to the occupied state is the maximum value, for example, 1.0.

[0060] The cost of a reserved status lies between that of an idle and an occupied status, constituting a soft penalty mechanism. Route planning avoids hard switching between reserved and unreserved segments, instead smoothly guiding the route to the lower-cost unreserved segment. The historical congestion factor is determined based on the statistical distribution of state transition timestamps for each track segment. The system records the timestamps for each track segment's transitions from idle to reserved, from reserved to occupied, and from occupied to idle. The number of state transitions per unit time is counted as the historical occupancy frequency for that segment; a higher occupancy frequency results in a higher historical congestion factor. The fairness compensation factor is determined based on the cumulative waiting time of each vehicle. When a vehicle's cumulative waiting time deviates from the global average waiting time due to multiple yields or route replanning, the cost of the requested track segment is reduced through the fairness compensation factor, allowing the vehicle to receive more priority in subsequent route allocations.

[0061] When assigning routes to the car, the objective function is the sum of the travel costs of each track segment. The goal is to search the directed graph for the path with the minimum cost from the starting point to the ending point. For example, if car B starts from T1 and travels to T15, there are two possible routes: Route A passes through T1-T3-T7-T10-T15, with a total travel cost of 0.2+0.15+0.5+0.3+0.18=1.33; Route B passes through T1-T4-T9-T12-T15, with a total travel cost of 0.2+0.12+0.18+0.22+0.18=0.9. Route B has a lower total cost than Route A, so Route B is assigned to car B. The travel cost of T7 in Route A is 0.5, which is higher than the cost of the corresponding segment in Route B. This is because T7 is currently in a reserved state, and the current state factor increases the travel cost of T7.

[0062] Furthermore, in S3, for each vehicle, based on its current position, speed, acceleration, and the speed limits of the preceding track segments, the earliest and latest arrival times for each track segment are calculated, resulting in the motion envelope 210. The earliest arrival time corresponds to the scenario where the vehicle continues to travel at the maximum allowed speed for each segment, while the latest arrival time corresponds to the scenario where the vehicle travels at the minimum reasonable speed. These two time values ​​define the time interval during which the vehicle occupies that track segment, naturally accommodating the positional uncertainty caused by speed fluctuations and acceleration / deceleration.

[0063] In point estimation collision detection, which compares the distance between two vehicles frame by frame, the two vehicles may meet the safety distance requirement at a certain frame, but due to differences in acceleration and deceleration, they may have crossed the same track segment between two frames, and point estimation cannot capture inter-frame collisions. The motion envelope 210 describes the occupancy of each vehicle on each track segment using time intervals rather than time points, transforming collision detection from point estimation to interval estimation.

[0064] Taking car A as an example, car A is currently located at T5 with a speed of 1.5 m / s and an acceleration of 0.2 m / s². 2 The physical length of T6 is 3m, and the speed limit is 2m / s. The earliest arrival time of car A at the entrance of T6 is calculated based on its current speed and acceleration. It accelerates to the speed limit of T6 at the maximum permissible acceleration and then passes through at a constant speed. The latest arrival time of car A at the entrance of T6 is calculated based on the minimum reasonable speed. It gradually decelerates from its current speed to the minimum reasonable speed and then passes through at a constant speed. The difference between the two time values ​​is the envelope width of T6. The same calculation is performed sequentially for subsequent track segments such as T7 and T8. The envelope width of each segment varies due to different speed limit constraints.

[0065] Map the motion envelopes 210 of each vehicle onto a spatiotemporal conflict diagram with track segment number as the horizontal axis and time as the vertical axis. For example... Figure 3 As shown, the horizontal axis of the spatiotemporal conflict diagram represents track segment numbers T5 to T10, and the vertical axis represents time, in seconds. Figure 3 In the diagram, solid rectangles represent the motion envelope 210 of car A, and dashed rectangles represent the motion envelope 210 of car B. The horizontal width of each rectangle corresponds to a track segment, and the vertical height corresponds to the difference between the latest and earliest arrival times of that segment, i.e., the width of the envelope of that segment. The envelope of car A extends rightward from T5 to T9, and the envelope of car B extends rightward from T7 to T10. The envelopes of the two cars overlap vertically at T7. Figure 3 The area filled with intersecting lines is the overlapping area 220. The area filled with diagonal lines above T8 is the initial occupied area 230, corresponding to the track segment T8 in the reservation state and the corresponding reservation time window. This area has been pre-filled in the spatiotemporal conflict map before the envelope projection. After the new car's envelope is projected onto the conflict map, it directly intersects with the pre-filled initial occupied area 230, without needing to compare it with each reserved car individually.

[0066] In response to the overlapping region 220 of the motion envelopes 210 of the two vehicles on the spatiotemporal conflict map, a collision risk is determined between the two vehicles. Figure 3 As shown, the time interval occupied by the envelope of car A on T7 is from 2.0 seconds to 3.5 seconds, and the time interval occupied by the envelope of car B on T7 is from 2.8 seconds to 4.2 seconds. The two intervals overlap between 2.8 seconds and 3.5 seconds. Figure 3 The area filled by the intersecting lines at T7 corresponds to the overlapping area 220, which indicates that there is a collision risk between car A and car B at T7.

[0067] In other embodiments, collision detection can also employ a frame-by-frame distance comparison method. This involves calculating the track distance between any two vehicles in each captured frame, and determining a collision risk when the distance is less than a preset safety distance. While the frame-by-frame distance comparison method has lower computational complexity, it cannot cover inter-frame collisions and is also a collision avoidance implementation method.

[0068] Furthermore, in S3, the running route of each vehicle is divided into a locked segment 310 and a candidate segment 320. For example... Figure 4 As shown, taking the route of car A as an example, Figure 4 The horizontal direction is arranged from left to right, with the current track segment T5 of the car A and the subsequent track segments T6 to T15 on the running route. The track segments are connected by arrows to indicate the direction of travel. Figure 4 In the diagram, T5 is filled with gray to indicate the track segment where car A is currently located; T6, T7, and T8 are marked with solid lines and thick borders to indicate locked segment 310; and T9, T10, and T15 are marked with dashed lines to indicate candidate segment 320. Figure 4 The collision risk location 330 is marked with a cross above T9. Figure 4 The brackets below indicate the range of locked segment 310 and candidate segment 320: locked segment 310 covers 3 segments from T6 to T8 and is marked as not changing during operation; candidate segment 320 covers T9 to T15 and is marked as allowing incremental replanning.

[0069] Locked segment 310 is a sequence of continuous track segments following the trolley's current track segment along its running route, and it remains unchanged during operation. Candidate segment 320 is the remaining route segment after locked segment 310, allowing for corrections based on collision detection results. The number of track segments included in locked segment 310 is determined by the trolley's current speed; the higher the current speed, the more track segments locked segment 310 includes. Higher speeds result in longer braking distances, and locked segment 310 must cover the safe braking distance at the current speed, ensuring that the trolley within locked segment 310 has sufficient distance to brake even if a change command is received. Figure 4 As shown, the current speed of car A is 1.5 m / s. Locked segment 310 consists of the three track segments T6-T7-T8 ahead, and candidate segment 320 consists of T9-T10-T15. When car A accelerates to 2 m / s, locked segment 310 extends to four track segments ahead.

[0070] In response to a collision risk occurring within candidate segment 320, incremental replanning is performed on candidate segment 320. The search scope of the incremental replanning is limited to the orbital subgraph corresponding to candidate segment 320 and does not involve any changes to locked segment 310. Figure 4As shown, the collision risk location 330 is located at T9, and T9 is within the candidate segment 320 of car A. Incremental replanning searches for alternative paths within the range of T9 to T15, replacing T9 with T11 for detour. The trajectory of the locked segment 310T6-T7-T8 is not affected.

[0071] In other embodiments, the relationship between the length of the locking segment 310 and the speed can also be achieved by a nonlinear mapping based on the braking distance formula. That is, the braking distance is calculated based on the braking deceleration and the current speed and then converted into the number of track segments. This is also an implementation method that determines the length of the locking segment 310 based on the current speed.

[0072] Furthermore, in response to the collision risk occurring within the locked segment 310 of the two vehicles, since the locked segment 310 cannot be avoided through incremental replanning of the candidate segment 320, a negotiated yielding agreement is initiated. The yielding cost is calculated separately for each vehicle, and is a weighted composite of braking distance cost, path extension cost, and cascading cost. The braking distance cost is the number of track segments required for the yielding vehicle to decelerate from its current speed to its yielding speed; the more segments required for deceleration, the greater the braking distance cost. The path extension cost is the sum of the increased travel costs of the new route after yielding compared to the original route; the greater the detour distance, the greater the path extension cost. The cascading cost is the number of new overlaps between the new route after yielding and other vehicles on the spatiotemporal conflict map; the more vehicles involved in the new overlap, the greater the cascading cost.

[0073] In the following yielding scenario, car B's route is T2-T4-T5-T7-T11-T15, which differs from the route in the cost calculation example above. Both car A and car B's locking segment 310 includes T7. Car A is currently at T5, with locking segment 310 being T6-T7-T8, while car B is currently at T4, with locking segment 310 being T5-T7-T11. When car A yields, the braking distance cost is 2 segments, the path extension cost is the increased travel cost of 0.35 after the route detours through T6-T9, and the cascading cost is the new overlap with car C at T9. These three costs are multiplied by their respective weights to synthesize the total yielding cost. When car B yields, the braking distance cost is 1 segment, the path extension cost is the increased travel cost of 0.28 after the route detours through T8-T12, and the cascading cost is 0, meaning there is no new overlap with other cars. After comparison, the total cost of yielding to car B is less than that of car A, so car B is chosen as the yielding party.

[0074] After car B is identified as the yielding vehicle, its locked segment 310 is shortened to its current track segment T4. Segments T5, T7, and T11 of the original locked segment 310 are downgraded to candidate segments 320, triggering incremental replanning for these downgraded candidate segments. The new path for car B is calculated during deceleration, and it resumes operation immediately following the new path after deceleration ends.

[0075] Furthermore, before performing incremental replanning or yielding, the corrected path is projected onto the spatiotemporal conflict map to detect newly added overlapping areas 220 with other vehicles. The number of vehicles involved in the newly added overlap is the affected range value. Figure 5 As shown, the impact assessment is completed before the actual implementation of the path correction, which is a feedforward pre-assessment. It shifts the detection of chain conflicts from post-implementation remediation after the implementation of the path correction to pre-implementation prevention before implementation.

[0076] Following the scenario after car B yields, car B's new path is T4-T8-T12-T15. Projecting this path onto the spatiotemporal conflict graph, a new overlap is detected between car B's new envelope and car C's at T12, with a affected range value of 1. Assuming a preset affected range threshold of 2, and since the affected range value of 1 does not exceed the threshold, this correction scheme is executed.

[0077] If another alternative path for car B is T4-T9-T13-T15, after being projected onto the spatiotemporal conflict map, it has a new overlap with car A on T9 and a new overlap with car C on T13. The affected range value is 2, which is equal to the preset affected range threshold. This corrective solution is abandoned, and the system takes the next candidate solution from the alternative path pool and re-executes the affected range evaluation.

[0078] In response to all candidate path correction schemes exceeding the preset impact threshold, all vehicles involved in the collision risk reduce their speed to a preset safe speed and maintain their current path. In high-density operation scenarios where the number of vehicles increases to 5, for example, adding vehicles D and E to the baseline scenario of 3 vehicles, all alternative correction schemes for vehicle B, after projection, overlap with more than 2 other vehicles. The system does not perform any path correction; vehicles A and B simultaneously reduce their speed to the preset safe speed of 0.5 m / s, maintaining their current path until the collision-risk track segment is naturally cleared by the preceding vehicle, after which normal scheduling resumes. The reason for choosing not to modify the path as a fallback strategy is that when the global vehicle density is high, any path change may trigger a chain reaction of replanning for other vehicles; not changing the path is the option with the lowest global cost.

[0079] Optionally, when assigning routes to the vehicles, a preset number of alternative paths with the highest toll costs are cached simultaneously, forming a pool of alternative paths for each vehicle. For example, the preset number is 3, meaning that the top 3 cost-based routes are stored for each vehicle. Each path in the alternative path pool stores the complete segment sequence and the total toll cost.

[0080] When performing incremental replanning, the alternative path compatible with the currently locked segment 310 is retrieved first from the candidate path pool. The compatibility condition is: a continuous sequence of segments of the same length as the currently locked segment is extracted from the candidate path after the current track segment of the vehicle. This sequence is completely identical to the segment sequence of the currently locked segment, that is, only the candidate segment 320 after the locked segment is different. Taking vehicle A at time T4 as an example, the current speed is 1.5 m / s, the current running route is T1-T4-T5-T7-T10-T15, and the locked segment is T5-T7-T10. The candidate path pool stores three paths: The first alternative path is T1-T4-T5-T7-T10-T12-T15. The segment sequence within the coverage area of ​​locked segment 310 is T5-T7-T10, which is consistent with the current locked segment T5-T7-T10 and is compatible. The second alternative path is T1-T4-T5-T8-T11-T15. The segment sequence within the coverage area of ​​the locked segment is T5-T8-T11. The second segment is T8, while the second segment of the currently locked segment is T7, which is inconsistent and incompatible. The third alternative path is T1-T4-T5-T7-T10-T14-T15, and the segment sequence within the locked segment coverage is T5-T7-T10, which is consistent with the currently locked segment T5-T7-T10 and is compatible.

[0081] The alternative path pool is incrementally updated each time the vehicle passes through a track segment. After vehicle A passes through T5, the locked segment progresses from T5-T7-T10 to T7-T10-T15. Paths in the alternative path pool that do not include T7-T10 in the locked segment 310 are eliminated. The system adds new alternative paths to 3 within the track subgraph corresponding to candidate segment 320 after locked segment 310 (T7-T10), and recalculates and sorts the cost of each alternative path based on the latest weight value of the current multi-factor dynamic passage cost.

[0082] Optionally, the width of the motion envelope 210 for each track segment, i.e., the difference between the latest and earliest arrival times, is determined segment by segment based on the speed limit constraints of the corresponding track segment. The difference increases with the difference in the number of segments between the track segment and the current track segment of each vehicle. Near-end track segments are closer to the current position of the vehicle, resulting in fresher data and lower cumulative uncertainty in speed changes, leading to a narrower envelope width and higher collision detection accuracy. Far-end track segments are farther from the current position, where the uncertainty of speed fluctuations and data delays accumulates segment by segment with the difference in the number of segments, resulting in a wider envelope width and preserving a safety margin.

[0083] Taking car A currently at position T5 with a speed of 1.5 m / s as an example. The segment difference between T6 and T5 is 1, the speed limit of T6 is 2 m / s, and the envelope width of T6 is calculated based on the time difference between the maximum and minimum speeds under the speed limit constraint of T6, which is 0.3 seconds. The segment difference between T8 and T5 is 3, and the envelope width, accumulated over the speed ranges under the respective speed limits of T6 and T7, is 1.2 seconds. The segment difference between T10 and T5 is 5, and the envelope width accumulates to 2.5 seconds. The expansion of the far-end envelope width causes almost all far-end overlap detections to report positive results. If all far-end overlaps trigger replanning, the system will be overloaded.

[0084] The look-ahead time range of the spatiotemporal conflict map is divided into a near-end window and a far-end window. The length of the near-end window is equal to the travel time corresponding to the locked segment 310. Taking the locked segment 310T6-T7-T8 of car A as an example, the travel time of the locked segment 310 is calculated based on the physical length of the three track segments and the speed limit value, for example, 4.5 seconds. The near-end window is the time range 4.5 seconds forward from the current moment. The time range covered by the near-end window is strictly aligned with the track range covered by the locked segment 310, so that collision detection within the near-end window is only for the unchangeable path segments.

[0085] If the overlapping region 220 is within the near-end window and the collision risk occurs within candidate segment 320, incremental replanning is performed on candidate segment 320. If the overlapping region 220 is within the near-end window and the collision risk occurs within locked segment 310, a negotiated yielding protocol is initiated. If the overlapping region 220 is within the far-end window, a collision warning is recorded but incremental replanning is not performed. The system is reassessed after the vehicle continues to move forward and the far-end segment enters the near-end window. For example, if the envelopes of vehicle A and vehicle C overlap on T10, and T10 is within the far-end window of vehicle A, the system records the warning but does not trigger path correction. After vehicle A passes T7, locked segment 310 advances to T8-T9-T10. T10 enters the near-end window, and the system recalculates the envelope width and determines whether the overlap still exists. If it still exists, incremental replanning or a yielding protocol is triggered.

[0086] Furthermore, the overlap area value is calculated for the overlapping region 220 located within the near-end window, and the collision risk is divided into three levels based on the overlap area value. The overlap area value is the two-dimensional area of ​​the overlapping region 220 of the two vehicle envelopes on the track segment axis and time axis on the spatiotemporal conflict map, simultaneously encoding the duration of overlap in the time dimension and the number of overlapping track segments in the spatial dimension. Even within the near-end window, the edge of the envelope may still produce a small false overlap due to uncertainty expansion, that is, the two vehicle envelopes only brush past each other at the edge, and the actual collision probability is very low. If incremental replanning is triggered for all near-end overlaps, it will consume unnecessary computational resources. The three-level classification matches the cost of collision response with the severity of collision risk.

[0087] If the overlap area is less than the first area threshold, it is marked as a minor overlap, and no response is executed. The first area threshold corresponds to the maximum tolerable area rubbed by the edge of the envelope. For example, if the overlap area between car A and car B on T7 is 0.05 segments per second, and the first area threshold is 0.1 segments per second, 0.05 is less than 0.1, so it is marked as a minor overlap, and the system does not execute any response. This overlap may disappear on its own in subsequent detection cycles due to the natural changes in the speed of the two cars.

[0088] In response to an overlap area value falling between a first area threshold and a second area threshold, it is marked as medium overlap. The speed of the vehicles involved in reducing the collision risk is adjusted to narrow the corresponding motion envelope 210. In the next detection cycle, the overlap area value of the narrowed motion envelope 210 is recalculated, and the level is updated. The second area threshold corresponds to the minimum hazardous area that must be mitigated through path correction. For example, if the overlap area value between vehicles A and B on T7 is 0.3 segments per second, falling between the first area threshold of 0.1 and the second area threshold of 0.5, it is marked as medium overlap. The system sends a speed reduction command to vehicle A, reducing its speed from 1.5 m / s to 1.0 m / s. This speed reduction decreases the speed uncertainty of vehicle A, narrowing the motion envelope 210. In the next detection cycle, the narrowed envelope of vehicle A is recalculated, and the overlap area value decreases from 0.3 to 0.04 segments per second, below the first area threshold of 0.1. The level is updated from medium overlap to slight overlap, and the collision risk is automatically mitigated through speed fine-tuning without triggering path replanning.

[0089] In response to an overlap area value greater than the second area threshold, a strong overlap is identified, and incremental replanning is performed on candidate segment 320. For example, if the overlap area value between car A and car B on T7 is 0.8 segments per second, which is greater than the second area threshold of 0.5, it is identified as a strong overlap. The envelopes of the two cars deeply overlap on T7, and speed adjustment alone cannot resolve this. This triggers incremental replanning of candidate segment 320 to search for an alternative path for car A or car B to bypass T7.

[0090] Optionally, before performing the detection of overlapping regions 220, a coarse screening is first performed on the motion envelopes 210 of each vehicle at the track segment dimension to identify track segments traversed by two or more vehicles whose motion envelopes 210 share the same track segment. Only vehicle pairs that share the same track segment are subjected to time-dimension overlap detection. The coarse screening operation involves traversing the set of track segments traversed by the envelope of each vehicle, finding the intersection of the segment sets of any two vehicles, and if the intersection is not empty, the vehicle pair proceeds to fine screening; otherwise, it is skipped.

[0091] Following the baseline scenario, the envelope segment sets of the three cars are as follows: Car A passes through T5 to T10 (6 segments), Car B passes through T7 to T13 (7 segments), and Car C passes through T14 to T20 (7 segments). Three pairs are generated: the intersection of the segment sets of Car A and Car B is T7 to T10 (4 segments), which is not empty, so it proceeds to detailed inspection; the intersection of the segment sets of Car B and Car C is empty, so it is skipped; the intersection of the segment sets of Car A and Car C is also empty, so it is skipped. In summary, only one of the three pairs proceeds to detailed inspection.

[0092] Track segments that share a common path are marked as "hot segments," and the remaining segments are marked as "cold segments." In the example above, T7 to T10 are marked as hot segments, and T1 to T6 and T11 to T20 are marked as cold segments. The time axis of hot segments in the spatiotemporal conflict map uses a first time resolution, while the time axis of cold segments uses a second time resolution, with the first time resolution being higher than the second. For example, the time axis resolution for hot segments T7 to T10 is 0.1 seconds, while the time axis resolution for cold segments is 1.0 second. Using fine-grained resolution for hot segments captures detailed temporal overlaps, while using coarse-grained resolution for cold segments reduces storage and computational overhead. In an exhibition hall containing 20 track segments, hot segments typically account for only 20% to 30% of the total segments, and the storage requirements for coarse-grained cold segments are reduced by approximately an order of magnitude compared to a uniform fine-grained resolution across the entire map.

[0093] The spatiotemporal conflict graph maintains a rolling window along the time axis, and incremental updates are performed on hot segments every scheduling cycle. Incremental updates include three operations: sliding the window start point forward one time step and deleting expired occupancy entries; adding an envelope prediction for one time step at the end of the window and writing it into the corresponding hot segment's occupancy area; and performing overlap detection only on the hot segments involved in the update. The incremental update frequency for cold segments is lower than that for hot segments; for example, hot segments are updated once per scheduling cycle, while cold segments are updated once every 5 scheduling cycles. A full reconstruction of the spatiotemporal conflict graph is performed every preset number of cycles, for example, once every 50 scheduling cycles. This recalculates the envelopes of all 20 orbital segments and projects them onto the conflict graph, correcting accumulated state deviations and time axis alignment errors during incremental updates. When an orbital segment changes from a hot segment to a cold segment, its time axis is merged from fine-grained to coarse-grained; when it changes from a cold segment to a hot segment, reverse refinement is performed.

[0094] S4. Control the start, stop, speed, and direction of each trolley according to the dispatch signals. The dispatch signals contain the target speed value, start / stop instructions, and direction instructions for each trolley.

[0095] After receiving the dispatch signal, the onboard motor drive controller of the trolley converts the target speed value into a motor speed control command, driving the trolley along the running route. On straight sections, the trolley travels at a constant target speed. Upon reaching a fork in the road, a steering command specifies which branch the trolley should enter, and the motor drive controller controls the switch mechanism to change tracks. When entering a speed-limited section, the target speed value decreases to the speed limit for that section, and the motor decelerates to the corresponding speed.

[0096] Taking car A as an example, car A travels at a target speed of 1.5 m / s on straight section T5. Upon reaching the fork point T7, it enters the left branch line T10 according to the steering command. When entering the speed-limited section T10, the target speed is reduced to the speed limit value of 1.0 m / s for that section. The car control module 140 synchronously feeds back the current speed, position, and operating status data to the abnormal warning link for real-time monitoring by S5.

[0097] S5. Monitor the running status and track status of each trolley in real time, and in response to the detection of an anomaly, output an early warning signal and trigger the corresponding trolley to stop.

[0098] Anomalies are categorized into track obstacles, vehicle mechanical malfunctions, and abnormal movements not caused by scheduling commands. Track obstacles are identified by unexpected occlusion signals detected by laser sensors. For example, if a section of track is not occupied by any vehicle, and the laser sensor continuously reports occlusion, it is identified as a track obstacle. Vehicle mechanical malfunctions are identified by anomalies in position encoder data. For example, if the encoder data of vehicle A remains static for several consecutive frames, but the target speed in the scheduling signal is not zero, it is identified as a drive mechanism malfunction. Abnormal movements not caused by scheduling commands are identified by the deviation between the current speed and the target speed in the scheduling signal. For example, if the actual speed of vehicle B continuously deviates from the target speed and the deviation exceeds a preset deviation threshold, it is identified as abnormal movement.

[0099] In response to the detection of any of the above-mentioned anomalies, the system outputs a warning signal to the management personnel terminal and simultaneously triggers the corresponding trolley to stop. The warning signal includes the anomaly type, the track segment number of the anomaly location, and the time of occurrence. The stop command is sent to the motor drive controller of the corresponding trolley through the scheduling signal channel. The controller sets the target speed to zero and initiates braking.

[0100] Optionally, the warning threshold can be dynamically adjusted based on the operating density parameter, which is the ratio of the number of vehicles currently running on the track to the total number of track segments. Continuing with the baseline scenario, three vehicles operate simultaneously, with a total of 20 track segments, resulting in an operating density parameter of 3 / 20 = 0.15. The speed deviation threshold for abnormal movement is set to 0.3 m / s when the operating density parameter is 0.15; that is, a difference between the actual speed and the target speed exceeding 0.3 m / s is considered abnormal. When the fourth vehicle joins the operation, the operating density parameter increases to 4 / 20 = 0.2, the distance between vehicles decreases, and the consequences of a collision become more severe. The speed deviation threshold is then tightened to 0.2 m / s, making anomaly detection more sensitive.

[0101] The adjustment of the warning threshold and the acquisition frequency are performed based on the same operating density parameter. When the operating density parameter is 0.15, the acquisition frequency remains at the base value of 10 frames / second; when the operating density parameter is increased to 0.2, the acquisition frequency increases to 15 frames / second. Tightening the warning threshold while simultaneously increasing the acquisition frequency ensures that the temporal resolution of the location data upon which the warning decision depends matches the severity of the warning. Lowering the operating density parameter relaxes the warning threshold while simultaneously reducing the acquisition frequency, decreasing data transmission and preprocessing load in low-density scenarios.

[0102] S6. Synchronously drive the multimedia equipment in the exhibition hall to display information based on the operating status of each vehicle.

[0103] The exhibition hall's multimedia equipment includes 3D displays and voice broadcasting devices installed in each booth area. The system pushes the real-time location data of each vehicle to the control interface of the multimedia equipment via the TCP / IP protocol. The 3D displays switch the corresponding display screen according to the booth area where the vehicle is currently located, and the voice broadcasting devices simultaneously play the corresponding popular science explanations. Taking vehicle A arriving at booth T10 as an example, the 3D display screen switches to the astronomy-themed display screen corresponding to T10, and the voice device begins broadcasting astronomical explanations. After vehicle A leaves T10, the 3D display screen stops displaying the current image and waits for the next vehicle to arrive.

[0104] Optionally, in S3, global routes are pre-allocated based on track layout data and current occupancy status data before each vehicle departs. During the operation of each vehicle, local corrections are made to the routes based on real-time standardized track status data and standardized vehicle position data. Pre-allocation is performed before the exhibition hall opens. The system calculates the minimum-cost path from each of the three vehicles to their respective starting points and ending points based on the current status of each track segment, generating three global routes. The calculations during the pre-allocation phase are not subject to real-time constraints, allowing for a longer search time to obtain better path solutions.

[0105] The routes generated during the pre-allocation phase are stored as safe routes after verification. Verification includes checking for spatiotemporal conflicts between the three routes and ensuring each route meets the speed limits and unidirectional / bidirectional attribute constraints of the track segment. Once verified, the routes are stored in the system memory. If the scheduling response time exceeds a preset timeout threshold during operation (e.g., if the timeout threshold is 0.5 seconds and a scheduling calculation by the AI ​​intelligent scheduling takes 0.7 seconds, exceeding the threshold), the system switches to a degraded operation mode. In degraded operation mode, the three vehicles each run at low speed in one direction along their stored safe routes, reducing the speed to 0.5 m / s. The warning threshold is tightened to the most stringent level, and the multimedia interactive display switches to a general loop playback mode, where the 3D display loops general promotional content for the exhibition hall, and the voice device loops a general welcome message. Degraded operation mode continues until the AI ​​intelligent scheduling resumes normal response and automatically exits.

[0106] Optionally, S6 employs an event-driven synchronization mechanism. Multiple trigger point locations are preset in the track layout data, set at booth area entrances and key bifurcation points. Following the baseline scenario, five trigger points are set at the T3 entrance, T7 bifurcation point, T10 booth area entrance, T14 booth area entrance, and T18 exit area entrance. A location event is generated in response to each vehicle reaching a trigger point. The location event includes the vehicle number, trigger point number, and arrival time, and is used for scheduling decisions and switching multimedia interactive display content. Taking vehicle A reaching trigger point T10 as an example, the location event notifies the AI ​​intelligent scheduling module 130 to update vehicle A's position baseline for subsequent route correction, and simultaneously notifies the multimedia equipment to switch to the astronomical theme display content corresponding to T10. No synchronous communication occurs between two trigger points; the communication volume decreases from being proportional to the polling frequency to being proportional to the number of trigger points.

[0107] Optionally, S3 also includes: performing statistical analysis on the accumulated historical running trajectories during non-running periods, identifying track segments with high occupancy frequency and collision warning trigger rate, and updating the analysis results as historical congestion factors in the passage cost calculation.

[0108] Following the baseline scenario, after one week of exhibition hall operation, the system statistically analyzes all daily operational trajectories after closing time. The statistical method involves iterating through the three-state transition timestamp records for each track segment on that day, calculating the number of times the track transitions from idle to reserved within a unit of time as the occupancy frequency, and calculating the number of times a collision warning signal is triggered in that segment as the collision warning trigger rate. The statistical results show that the occupancy frequency of the T7 bifurcation point ranks first among the 20 track segments, while its collision warning trigger rate ranks second. The collision warning trigger rate of the T14 bifurcation point ranks first. The system adjusts the historical congestion factors for T7 and T14 from their initial values, increasing the passage cost value of T7 and T14 in the cost function of path planning in subsequent scheduling cycles, automatically favoring route allocation to avoid T7 and T14. The update of the historical congestion factors is performed offline during closing time, without consuming real-time computing resources during runtime. As exhibition hall operational data continues to accumulate, the historical congestion factors are updated and iterated daily, and path allocation gradually upgrades from reactive decision-making based solely on the current state to predictive decision-making combining historical statistics.

[0109] Optionally, the system also includes a visitor interaction terminal. The visitor interaction terminal is a touchscreen installed at the entrance of the exhibition hall, displaying a list of selectable exhibition themes, such as astronomy, geography, and biology. The visitor interaction terminal is configured to receive the exhibition theme selected by the visitor and transmit the selection signal corresponding to the theme to the AI ​​intelligent scheduling module 130. The selection signal includes a theme number; for example, astronomy corresponds to theme number 1, geography to theme number 2, and biology to theme number 3. The AI ​​intelligent scheduling module 130 adjusts the running routes and stopping order of each vehicle according to the selection signal. Taking the visitor selecting the astronomy theme as an example, the theme number in the selection signal is 1, which is transmitted to the AI ​​intelligent scheduling module 130. The scheduling module queries the list of booths corresponding to theme number 1, which is T10 and T14. It adjusts the running route of vehicle A to prioritize passing through T10 and T14, and changes the stopping order from the default T10→T14→T18 to T10→T14, skipping T18, which is unrelated to the astronomy theme. The multimedia linkage module 160 synchronously switches the content corresponding to the displayed theme. The 3D display screen and voice device of T10 load the display images and narration of the astronomy theme, and T14 synchronously loads the second set of display content of the astronomy theme. The audience's choice of the exhibition hall entrance, the adjustment of the car route, and the switching of multimedia content form a closed loop from the audience's intention to the scheduling decision, the movement of the car, and the multimedia display.

[0110] like Figure 2 As shown, this application also discloses an intelligent control and scheduling system for a track vehicle based on artificial intelligence, including a track status acquisition module 110, a data preprocessing module 120, an AI intelligent scheduling module 130, a vehicle control module 140, an anomaly warning module 150, and a multimedia linkage module 160.

[0111] The track status acquisition module 110 is configured to acquire track occupancy status data and position data of each trolley, corresponding to step S1 in the method embodiment. The track status acquisition module 110 consists of laser sensors installed at the entrance and exit of each track segment and position encoders installed on the drive wheel axles of each trolley. The laser sensors acquire occupancy status data of each track segment, and the position encoders acquire position data of each trolley. The output of the track status acquisition module 110 is transmitted to the data preprocessing module 120 via the TCP / IP protocol.

[0112] The data preprocessing module 120 is configured to denoise and calibrate the occupancy status data and position data, removing invalid data to obtain standardized track status data and standardized vehicle position data, corresponding to step S2 in the method embodiment. The data preprocessing module 120 is deployed on an embedded processor and performs sliding window mean filtering, zero-point offset calibration, and jump threshold removal. The output of the data preprocessing module 120 is transmitted to the AI ​​intelligent scheduling module 130 via the TCP / IP protocol.

[0113] The AI ​​intelligent scheduling module 130 is configured to dynamically allocate routes and avoid collisions for multiple vehicles based on pre-stored track layout data, standardized track state data, and standardized vehicle position data, generating scheduling signals, corresponding to step S3 in the method embodiment. The AI ​​intelligent scheduling module 130 is deployed on an embedded processor and integrates three-state management logic, multi-factor passage cost calculation logic, motion envelope 210 calculation and spatiotemporal conflict map collision detection logic, locked segment 310 and candidate segment 320 segmentation and incremental replanning logic, negotiated yielding protocol logic, and impact range assessment logic. The output of the AI ​​intelligent scheduling module 130 is the scheduling signal, which is simultaneously transmitted to the vehicle control module 140 and the multimedia linkage module 160.

[0114] The trolley control module 140 is configured to control the start, stop, speed, and direction of each trolley according to the scheduling signal, corresponding to step S4 in the method embodiment. The trolley control module 140 is an on-board motor drive controller installed on each trolley. After receiving the scheduling signal, it converts the target speed value into a motor speed control command, controlling the switch mechanism to switch the track direction at the bifurcation point. The trolley control module 140 feeds back the current speed, position, and running status data of each trolley to the anomaly warning module 150.

[0115] The anomaly warning module 150 is configured to monitor the real-time operating status and track status of each vehicle. In response to an anomaly detection, it outputs a warning signal and triggers the corresponding vehicle to stop, corresponding to step S5 in the method embodiment. The anomaly warning module 150 is a monitoring program deployed on the system server. It receives sensor data from the track status acquisition module 110 and operating status data fed back by the vehicle control module 140, and performs track obstacle detection, mechanical fault detection, and abnormal motion detection. The warning signal is pushed to the management terminal, and the stop command is sent to the motor drive controller of the corresponding vehicle through the scheduling signal channel.

[0116] The multimedia linkage module 160 is configured to synchronously drive the multimedia equipment in the exhibition hall to perform a linked display based on the running status of each vehicle, corresponding to step S6 in the method embodiment. The multimedia linkage module 160 includes 3D displays and voice broadcasting devices for each booth area in the exhibition hall, as well as control interfaces for these devices. The control interface receives real-time position data of each vehicle output by the AI ​​intelligent scheduling module 130 via the TCP / IP protocol, and switches the corresponding display screen and narration according to the current booth area of ​​the vehicle.

[0117] The signal transmission relationships between the modules are as follows: the track status acquisition module 110 transmits raw data to the data preprocessing module 120; the data preprocessing module 120 transmits standardized data to the AI ​​intelligent scheduling module 130; the AI ​​intelligent scheduling module 130 transmits scheduling signals to the trolley control module 140 and the multimedia linkage module 160; the trolley control module 140 feeds back operating status data to the anomaly warning module 150; and the anomaly warning module 150 transmits a stop command to the trolley control module 140 through the scheduling signal channel. All modules communicate via the TCP / IP protocol.

[0118] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0119] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0120] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method for intelligent control and scheduling of railcars based on artificial intelligence, characterized in that, Includes the following steps: S1. Collect track occupancy status data and position data of each trolley; S2. The occupancy status data and the position data are denoised and calibrated to remove invalid data and obtain standardized track status data and standardized trolley position data. S3. Based on the pre-stored track layout data, the standardized track status data, and the standardized vehicle position data, dynamically allocate running routes for multiple vehicles and avoid collisions, and generate scheduling signals; S4. Control the start, stop, speed and steering of each trolley according to the scheduling signal; S5. Monitor the running status and track status of each trolley in real time, and in response to the detection of an anomaly, output an early warning signal and trigger the corresponding trolley to stop; S6. Synchronously drive the multimedia equipment in the exhibition hall to display information based on the operating status of each vehicle.

2. The intelligent control and scheduling method for track vehicles based on artificial intelligence according to claim 1, characterized in that, In S3, each track segment in the track layout data is maintained in three states: idle, reserved, and occupied. When assigning the running route to the vehicle, the status of each track segment traversed by the running route is switched from the idle state to the reserved state; When the trolley enters the corresponding track segment, the state of the corresponding track segment is switched from the reserved state to the occupied state; When the trolley leaves the corresponding track segment, the state of the corresponding track segment is switched from the occupied state to the idle state; The track segment in the reserved state and the corresponding reserved time window information serve as inputs for allocating the running route and avoiding collisions.

3. The intelligent control and scheduling method for track vehicles based on artificial intelligence according to claim 2, characterized in that, In step S3, a passage cost is calculated for each track segment in the track layout data. The passage cost is a weighted composite of a track segment length factor, a current state factor, a historical congestion factor, and a fairness compensation factor. The current state factor of the track segment in the reserved state is higher than the current state factor of the track segment in the idle state and lower than the current state factor of the track segment in the occupied state; The historical congestion factor is determined based on the statistical distribution of the state transition timestamps of each track segment; When assigning the running route to the vehicle, the sum of the travel costs of each track segment is used as the objective function.

4. The intelligent control and scheduling method for track vehicles based on artificial intelligence according to claim 3, characterized in that, In S3, for each car, the earliest arrival time and the latest arrival time of the car to the track segment ahead are calculated based on the car's current position, speed, acceleration and the speed limit constraints of each track segment ahead, and the motion envelope (210) is obtained. The motion envelope (210) of each car is mapped to a spatiotemporal conflict diagram with track segment number as the horizontal axis and time as the vertical axis. The spatiotemporal conflict diagram uses the track segment in the reservation state and the corresponding reservation time window as the initial occupied area (230). In response to the existence of an overlapping region (220) between the motion envelopes (210) of the two vehicles on the spatiotemporal conflict map, a collision risk is determined between the two vehicles.

5. The intelligent control and scheduling method for track vehicles based on artificial intelligence according to claim 4, characterized in that, In S3, the running route of each car is divided into a locked segment (310) and a candidate segment (320). The locked segment (310) is a continuous sequence of track segments in front of the current position of the car and does not change during operation. The candidate segment (320) is the remaining route segment after the locked segment (310). The number of track segments included in the locking segment (310) is determined according to the current speed of the trolley; the higher the current speed, the more track segments the locking segment (310) includes. In response to the collision risk occurring within the candidate segment (320), incremental replanning is performed on the candidate segment (320) to generate a corrected path.

6. The intelligent control and scheduling method for track vehicles based on artificial intelligence according to claim 5, characterized in that, In S3, in response to the collision risk occurring within the locking segment (310) of the two vehicles, a yielding cost is calculated for each of the two vehicles, the yielding cost including braking distance cost, path extension cost and cascading cost; The cascading cost is the additional overlap between the new path after yielding and other vehicles on the spatiotemporal conflict map. The party with the lowest yielding cost is selected as the yielding party. The locked segment (310) of the yielding party is shortened to the track segment where the yielding party is currently located. The remaining part of the original locked segment (310) is downgraded to the candidate segment (320) and the incremental replanning is triggered.

7. The intelligent control and scheduling method for track vehicles based on artificial intelligence according to claim 6, characterized in that, Before performing the incremental replanning or the yielding, the corrected path is projected onto the spatiotemporal conflict map to detect new overlapping areas with other vehicles (220). The number of vehicles involved in the new overlap is the affected range value. If the affected range value exceeds a preset affected threshold, the corresponding path correction scheme is abandoned. In response to the fact that the spillover range values ​​of all candidate path correction schemes exceed the preset spillover threshold, each vehicle involved in the collision risk reduces to a preset safe speed and maintains its current path.

8. The intelligent control and scheduling method for track vehicles based on artificial intelligence according to claim 5, characterized in that, When assigning the running route to the vehicle, a preset number of alternative routes with the highest passage cost are cached at the same time to form a alternative route pool for the vehicle. When performing the incremental replanning, alternative paths that are compatible with the current locked segment (310) are retrieved from the alternative path pool first. The compatibility condition is that the locked segment (310) of the alternative path is consistent with the current locked segment (310). The alternative path pool is incrementally updated each time the trolley passes through a track segment, eliminating alternative paths that are incompatible with the updated locked segment (310) and supplementing new alternative paths.

9. The intelligent control and scheduling method for track vehicles based on artificial intelligence according to claim 5, characterized in that, The difference between the latest arrival time and the earliest arrival time of each track segment is determined by connecting the speed limit constraints of the corresponding track segment segment by segment, and the difference increases with the difference in the number of segments between the track segment and the track segment where each car is currently located. The look-ahead time range of the spatiotemporal conflict map is divided into a near-end window and a far-end window, and the length of the near-end window is equal to the passage time corresponding to the locked segment (310). In response to the overlapping region (220) being located within the near-end window, the incremental replanning is performed on the candidate segment (320); in response to the overlapping region (220) being located within the far-end window, a collision warning is recorded but the incremental replanning is not performed.

10. The intelligent control and scheduling method for track vehicles based on artificial intelligence according to claim 4, characterized in that, Before performing the detection of the overlapping region (220), a coarse screening is performed on the motion envelope (210) of each car in the track segment dimension to identify the track segments through which the motion envelope (210) of two or more cars pass. Only the cars that pass through the same track segment are subjected to time dimension overlap detection. The common orbital segments are marked as hot segments, and the remaining orbital segments are marked as cold segments. The time axis of the hot segments in the spatiotemporal conflict diagram adopts a first time resolution, and the time axis of the cold segments in the spatiotemporal conflict diagram adopts a second time resolution. The first time resolution is higher than the second time resolution. The spatiotemporal conflict graph maintains a rolling window along the time axis. Incremental updates are performed on hot segments in each scheduling cycle, while the incremental update frequency of cold segments is lower than that of hot segments. A full reconstruction of the spatiotemporal conflict graph is performed every preset number of cycles.

Images