Rail transit rapid arrangement method and system

Abstract

The invention relates to the technical field of rail transit arrangement, and discloses a rail transit rapid arrangement method and system, and the method comprises the steps: collecting multi-dimensional data through real-time interaction between a vehicle-mounted sensor and a ground dispatching center, and building a passenger flow change model to predict passenger flow density data; on the basis of the multi-dimensional data, a rail transit marshalling adjustment instruction is generated through an arrangement unit, the front vehicle power and the front road condition are called through a traffic database, and a correction instruction of the rail transit marshalling adjustment instruction is output through a distribution model; the correction instruction is input into the mixed verification unit for edge correction, the correction instruction is optimized to optimize the rail transit marshalling adjustment instruction, rail transit is achieved, and the accuracy of arranging the rail transit result is greatly improved through double-layer correction.

Description

Technical Field

[0001] The present invention relates to the technical field of rail transit scheduling, and discloses a rail transit rapid scheduling method and system. Background Art

[0002] Current rail transit marshaling and scheduling mainly rely on physical coupling and fixed marshaling modes, which have significant defects. Traditional marshaling requires decision-making and manual operation through ground interlocking equipment, which takes several minutes and cannot meet the surge in capacity demand during peak hours. The existing methods are based on fixed timetables and static priority strategies, and lack the ability to dynamically adapt to sudden changes in passenger flow and equipment health status, resulting in a high empty load rate during off-peak hours. Physical marshaling relies on fixed traction and braking parameters, which makes it difficult to balance safety intervals and energy consumption limits in emergency scenarios. Summary of the Invention

[0003] The purpose of this section is to summarize some aspects of embodiments of the present invention and briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section and the abstract and title of this application to avoid obscuring the purpose of this section, the abstract and the title of the invention, and such simplifications or omissions should not be used to limit the scope of the present invention.

[0004] In order to solve the above technical problems, the main purpose of the present invention is to provide a rail transit rapid scheduling method, comprising: Through real-time interaction between on-board sensors and the ground dispatch center to collect multi-dimensional data, a passenger flow change model is established to predict passenger flow density data; Based on multi-dimensional data, the marshaling unit generates rail transit formation adjustment instructions. The traffic database retrieves the preceding vehicle power and road conditions ahead, and outputs correction instructions for the rail transit formation adjustment instructions through the distribution model. The correction instructions are input into the hybrid verification unit for edge correction, and the correction instructions are optimized to optimize the rail transit formation adjustment instructions.

[0005] As a preferred solution of the rail transit rapid scheduling method of the present invention, wherein: The multi-dimensional data includes passenger flow density, rail transit energy consumption, and equipment health status; The rail transit energy consumption and the equipment health status are used to set a priority strategy, which prioritizes power demand and equipment health; The priority strategy includes peak hours, off-peak hours and off-peak hours.

[0006] As a preferred solution of the rail transit rapid scheduling method of the present invention, wherein: The passenger flow change model includes a data processing layer, an input layer, a feature extraction layer, a fusion prediction layer and a mixed loss layer; The feature extraction layer includes spatiotemporal feature fusion, obtains passenger flow and passenger flow transfer features between stations through graph convolutional networks, and enhances the feature identification of key time periods; The feature extraction layer also includes multimodal feature fusion, which includes a multi-channel input structure, encodes passenger flow and time features, and sorts the encoding results; The fusion prediction layer sets up a bottom layer to capture short-term passenger flow fluctuation characteristics and a high-level learning model of long-term passenger flow trends and cycles. The fusion prediction layer also includes an enhanced attention mechanism to focus on the impact of historical time periods on the next moment and enhance the contribution of feature dimensions; The passenger flow change prediction model is provided with a three-terminal parallel output layer, and the three-terminal parallel output layer includes a peak prediction terminal, a flat peak prediction terminal and a valley peak prediction terminal.

[0007] As a preferred solution of the rail transit rapid scheduling method of the present invention, wherein: The orchestration unit includes multi-source data interface, dynamic decision making and instruction verification; The multi-source data interface receives the peak, flat and valley peak prediction values ​​and real-time carriage passenger capacity data from the three-terminal parallel output layer of the passenger flow change model, as well as the preceding vehicle power parameters and road condition information retrieved from the traffic database; Dynamic decision-making builds a flexible allocation rule base based on multi-dimensional data. The rule base includes priority strategies and matching verification between train length and platform capacity. The instruction verification simulates the formation to adjust the instruction execution effect, verifies the feasibility of the instruction, verifies whether the traction system load exceeds the limit after coupling, and generates a visual alarm.

[0008] As a preferred solution of the rail transit rapid scheduling method of the present invention, wherein: The dynamic decision-making implementation method includes: Dynamically adjust input feature weights based on time period. If a formation adjustment instruction conflicts with the road conditions ahead, an online game model is triggered to generate a compromise speed curve based on the power performance of the leading vehicle and the braking ability of the following vehicle. If the station capacity is insufficient, priority will be given to splitting the train groups and dynamically allocating the stop location time; The rail transit formation adjustment instruction generation rules include generating mandatory association instructions during peak hours, ignoring the rail transit maintenance status, and calling standby formations; and generating degraded operation instructions during off-peak hours.

[0009] As a preferred solution of the rail transit rapid scheduling method of the present invention, wherein: The method for outputting correction instructions from the distribution model includes: Based on the marshaling status data collected in real time by the on-board edge computing node, the marshaling status instructions are dynamically adjusted through correction factors; Through distributed communication protocols, each rail transit system adjusts operating parameters based on local decisions; The distribution model also includes instruction isolation, which includes physical isolation, logical isolation and timing isolation; The physical isolation allocates independent communication channels for instructions of different priorities; The logical isolation sets up a sandbox environment in the vehicle controller, restricting hardware access rights to non-safety-related instructions and allowing only safe instructions to directly control the traction brake system; The timing isolation adopts a time-triggered architecture and polls the instruction queue according to a fixed cycle to ensure that high-priority instructions are executed first and only instructions in the same control domain are processed in a single cycle.

[0010] As a preferred solution of the rail transit rapid scheduling method of the present invention, wherein: The hybrid verification unit includes an input layer, a verification and optimization layer, an edge optimization layer and an output layer; The input layer receives correction instructions from the distribution model, real-time rail transit marshaling status data and safety threshold library; The verification and optimization layer simulates the running status of the train after the execution of the correction instructions to verify whether it meets the power matching, braking safety and energy consumption constraints; The edge optimization layer uses a constrained genetic algorithm to minimize the risk score and energy consumption increment as the objective function to generate an optimized correction instruction set, which includes the marshaling length adjustment priority and dynamic speed curve; The output layer sends the optimized correction instructions to the vehicle controller and simultaneously updates the instruction execution record of the traffic database.

[0011] As a preferred solution of the rail transit rapid scheduling method of the present invention, wherein: The edge optimization layer also collects historical optimization instruction execution result data to build a feedback training set and dynamically adjusts the risk weight and energy consumption weight; If safety requirements conflict with energy-saving requirements, a hierarchical optimization strategy is adopted, giving priority to forced deceleration when the braking distance is insufficient, regardless of energy consumption limits; If there is a conflict in the coordination of multiple groups, resources are allocated evenly; The edge optimization layer compares the indicators before and after the execution of the optimization instruction in real time and generates a verification report. If the indicator deviation exceeds the tolerance threshold, the instruction rollback mechanism is triggered, and the group configuration is automatically restored to the previous stable state. A fault code is sent to the cloud dispatch center to start the manual intervention process.

[0012] As a preferred solution of the rail transit rapid scheduling system of the present invention, wherein: The data acquisition module integrates the vehicle-mounted sensors and the communication interface of the ground dispatch center to collect multi-dimensional data in real time; The passenger flow prediction module outputs the predicted probabilities of peak, off-peak and valley periods and passenger flow fluctuation characteristics; The orchestration unit includes a multi-source data interface, dynamic decision-making, and command verification. The multi-source data interface receives passenger flow prediction results, power parameters of the preceding vehicle, and road conditions ahead. The dynamic decision-making generates formation adjustment instructions and executes conflict resolution logic. The command verification simulates the execution effect of the instruction and generates a visual alarm. The distribution model obtains the correction factor and corrects the instruction through communication; The hybrid verification unit verifies the power matching, braking safety and energy consumption constraints of the formation; The edge optimization submodule uses a constrained algorithm to generate an optimization instruction set and dynamically adjusts the optimization weights through an adaptive learning module; The feedback closed unit compares the indicator deviation before and after the instruction execution, triggering the rollback mechanism or manual intervention process.

[0013] As a preferred solution of the rail transit rapid scheduling system of the present invention, wherein: The edge optimization submodule integrates the learning model and dynamically updates the fitness function weights based on historical instruction execution data; Conflict resolution: when multiple trains compete for the same coupling section, priority is sorted by the remaining battery capacity. When safety requirements conflict with energy-saving requirements, the vehicle is forced to reduce speed to meet the braking distance constraint, regardless of energy consumption limits; Communication isolation allocates a high-priority channel of the sliced ​​network for emergency braking commands, sets up a sandbox environment in the on-board controller, restricts the access rights of non-safety commands to the traction braking system, and only allows safety commands to directly control the hardware.

[0014] Beneficial effects of the present invention: A compromise speed curve is generated through a distributed game model to resolve the conflict between the power of the leading vehicle and the braking of the trailing vehicle. Flexible priority strategies are combined, with forced coupling during peak hours and downgraded operation during valley hours to reduce the no-load rate. The hybrid verification unit constrains energy consumption and risks to achieve automatic speed reduction when the braking safety distance is insufficient.

[0015] In the event of sudden large passenger flow, the marshaling unit generates rail transit marshaling adjustment instructions to give priority to allocating the right of way to large trains, shorten the response time of standby marshaling coupling, and realize dynamic matching of the leading vehicle's power performance and the trailing vehicle's braking capability based on the distribution model.

[0016] A hybrid verification unit is set up to perform edge correction on the correction instructions, and the accuracy of the rail transit arrangement results is greatly improved through double-layer correction. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. Those skilled in the art can also derive other drawings based on these drawings without inventive effort. Among them: Figure 1 This is a flow chart of a rail transit rapid scheduling method according to the present invention; Figure 2 A diagram illustrating an application method of a passenger flow change model for a rapid rail transit scheduling method according to the present invention; Figure 3 A diagram showing a working method of a game model for a rail transit rapid scheduling method according to the present invention; Figure 4 A monitoring diagram of a system for arranging a rail transit rapid arranging method according to the present invention; Figure 5 This is a curve simulation diagram of real-time monitoring and prediction of passenger flow at Lujiazui Station among multiple stations in a rail transit rapid scheduling method of the present invention; Figure 6 This is a curve simulation diagram of real-time monitoring and prediction of passenger flow at the multi-station Medieval Avenue Station using a rail transit rapid scheduling method according to the present invention. DETAILED DESCRIPTION

[0018] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0019] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0020] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.

[0021] Example 1 like Figure 1 As shown, a rail transit rapid scheduling method includes: Through real-time interaction between on-board sensors and the ground dispatch center, multi-dimensional data is collected and a passenger flow change model is established to predict passenger flow density data.

[0022] Furthermore, the multidimensional data includes passenger flow density, rail transit energy consumption, and equipment health status; Passenger flow density is obtained through platform cameras and on-board infrared counters; rail transit energy consumption is obtained through traction system current monitoring; and equipment health is obtained through bearing temperature sensors and vibration monitoring.

[0023] The platform camera uses the target detection algorithm to count the number of people entering and leaving the station in real time, the on-board infrared counter monitors the distribution of passengers in the carriage, and combines the gate card swiping data to cross-verify the passenger flow density.

[0024] Eliminate false detections caused by lighting changes or occlusions, and dynamically calibrate the deviation between infrared counts and video statistics.

[0025] The current monitoring module of the rail transit traction system collects the current values ​​of the train during acceleration, constant speed and braking in real time, and calculates the instantaneous energy consumption based on the voltage data. When the high current lasts longer than the threshold, it is marked as abnormal energy consumption, triggering a device health check.

[0026] The bearing temperature sensor and vibration monitoring module collect the status of mechanical components in real time, and generate early warning signals when the temperature is abnormal or the vibration frequency changes suddenly.

[0027] The multi-dimensional data includes passenger flow density, rail transit energy consumption, and equipment health status; The rail transit energy consumption and the equipment health status are used to set a priority strategy, which prioritizes power demand and equipment health; The priority strategy includes peak hours, off-peak hours and off-peak hours.

[0028] The passenger flow change model includes a data processing layer, an input layer, a feature extraction layer, a fusion prediction layer and a mixed loss layer; The data processing layer is used to fill missing data with the mean of adjacent time periods, align passenger flow, energy consumption, and equipment data by timestamp, and construct a unified spatiotemporal data matrix.

[0029] The feature extraction layer includes spatiotemporal feature fusion, obtains passenger flow and passenger flow transfer features between stations through graph convolutional networks, and enhances the feature identification of key time periods; Furthermore, the spatiotemporal feature fusion constructs a site topology relationship diagram through a graph convolutional network, with sites as nodes and edge weights as historical passenger flow correlations, to extract cross-site passenger flow transfer characteristics. For example, during the morning rush hour, node A represents concentrated travel from a residential area to the commercial area of ​​node B, which is the passenger flow transfer feature from node A to node B.

[0030] like Figure 2 As shown in FIG, the application method of the passenger flow prediction model includes: The site association weights during holidays and concert periods are dynamically increased to strengthen the impact of special events.

[0031] The feature extraction layer also includes multimodal feature fusion, which includes a multi-channel input structure, encodes passenger flow and time features, and sorts the encoding results; The fusion prediction layer sets up a bottom layer to capture short-term passenger flow fluctuation characteristics and a high-level learning model of long-term passenger flow trends and cycles. The fusion prediction layer also includes an enhanced attention mechanism to focus on the impact of historical time periods on the next moment and enhance the contribution of feature dimensions; The passenger flow change prediction model is provided with a three-terminal parallel output layer, and the three-terminal parallel output layer includes a peak prediction terminal, a flat peak prediction terminal and a valley peak prediction terminal.

[0032] Furthermore, a preferred technical solution includes the passenger flow change model further comprising: The fusion prediction layer includes underlying short-term fluctuation capture, long-term trend learning and three-end parallel output.

[0033] Furthermore, the short-term fluctuations are captured by the underlying network to analyze real-time passenger flow changes, including sudden large passenger flows on special holidays and small passenger flows on weekdays, etc., to identify transient fluctuation patterns, such as a sudden drop in the number of passengers entering the station due to temporary flow restrictions at rail transit entrances, and a sudden increase in passenger flow at tourist attractions during statutory holidays.

[0034] Long-term trend learning uses high-level network modeling to model periodic patterns, and combines attention mechanisms to focus on similar historical periods to enhance prediction robustness. Periodic patterns include the postponement of the peak hour every Friday evening and relatively even distribution of passenger flow at stations from Tuesday to Thursday evenings.

[0035] The historical similar period includes the passenger flow distribution of the same type of stations in one month, excluding holidays, concerts, etc., for one week.

[0036] The three terminals output in parallel: the peak prediction terminal outputs the probability of future passenger flow overload; the flat peak prediction terminal predicts the fluctuation range of regular passenger flow; and the valley peak prediction terminal identifies low passenger flow periods and triggers energy-saving operation.

[0037] Furthermore, the hybrid loss layer sets a loss function, and by iterating the mean square error between the passenger flow predictions and actual values ​​for multiple time periods at each station on the three ends every day of a week, it outputs the loss value of the predicted value, and corrects and compensates for short-term fluctuations and long-term trend learning. For example, if long-term trend learning predicts that the passenger flow at the y station z time period in the second week of x month increases sharply, the passenger flow at the y station z time period in the third week of x month is less than the passenger flow at the y station z time period in the second week of x month, and the passenger flow at the y station z time period in the fourth week of x month is less than the passenger flow at the y station z time period in the second week of x month, then the loss of the predicted value at the y station z time period in the second week of x month is used to compensate for the emergency situation at the y station z time period in the second week of x month.

[0038] If the rail transit scheduling system is affected by a special event, such as a concert at station y during time z in the second week of month x, the weight of stations associated with station y will be dynamically increased. The attention weight of station y and its associated stations will be increased through the passenger flow change model, thereby increasing the threshold of the passenger flow prediction range.

[0039] The orchestration unit includes multi-source data interface, dynamic decision making and instruction verification; The multi-source data interface receives the peak, flat and valley peak prediction values ​​and real-time carriage passenger capacity data from the three-terminal parallel output layer of the passenger flow change model, as well as the preceding vehicle power parameters and road condition information retrieved from the traffic database; The multi-source data interface module builds a standardized channel for heterogeneous data and uses multi-source information fusion technology to solve the problem of data format differences; The data collection layer deploys a lightweight data agent program to receive the peak, off-peak and valley period prediction values ​​output by the passenger flow prediction model and the real-time passenger volume data reported by the on-board sensors in real time, and converts heterogeneous data such as bus signals and consultation messages into the same format through the protocol conversion engine.

[0040] A two-layer structure of real-time data cache and historical data warehouse is established. The power parameters of the preceding vehicle are stored in a time series database, and the road condition information is stored in a spatial topology database.

[0041] Dynamic decision-making builds a flexible allocation rule base based on multi-dimensional data. The rule base includes priority strategies and matching verification between train length and platform capacity. The instruction verification simulates the formation to adjust the instruction execution effect, verifies the feasibility of the instruction, verifies whether the traction system load exceeds the limit after coupling, and generates a visual alarm.

[0042] The dynamic decision-making implementation method includes: Dynamically adjust input feature weights based on time period. If a formation adjustment instruction conflicts with the road conditions ahead, an online game model is triggered to generate a compromise speed curve based on the power performance of the leading vehicle and the braking ability of the following vehicle. If the station capacity is insufficient, priority will be given to splitting the train groups and dynamically allocating the stop location time; The rail transit formation adjustment instruction generation rules include generating mandatory association instructions during peak hours, ignoring the rail transit maintenance status, and calling standby formations; and generating degraded operation instructions during off-peak hours.

[0043] A specific implementation method of a dynamic strategy includes: The multi-source data interface receives the peak, off-peak and off-peak forecast values ​​and real-time carriage passenger capacity data output by the three-terminal parallel output layer. During peak hours and stations, the goal is to maximize capacity, and the spare train formations of rail transit are forcibly connected, the departure interval is shortened to the limit, and some equipment maintenance alarms are ignored. The aforementioned equipment maintenance alarms include air conditioning filter cleaning and passenger window cleaning. During off-peak hours and stations, energy conservation is the core, and the train formations are split to reduce the empty load rate, the departure interval is extended, and non-essential on-board equipment is turned off. During off-peak hours: some trains are suspended for maintenance, the backbone trains are retained, and only basic services are maintained.

[0044] like Figure 5 Shown are the actual passenger flow detection curve and predicted passenger flow curve at 16 points at Lujiazui Station; like Figure 6 Shown are the actual passenger flow detection curve and predicted passenger flow curve at 16 points on Century Avenue; Figure 5 and Figure 6 The passenger flow monitoring and prediction chart for different stations at the same time shows that the actual passenger flow in Lujiazui is 1,000 people, and the predicted passenger flow is 900, with an error of less than 10%. The actual passenger flow in Century Avenue is 900 people, and the predicted passenger flow is 880, with an error of less than 10%.

[0045] Furthermore, dynamic decision-making is based on a hybrid decision-making mechanism of rule reasoning and case reasoning to achieve multi-objective optimization.

[0046] Through timestamp analysis, the peak, flat and valley time period rules are automatically matched, a three-level weight system is set up, with safety weight greater than efficiency weight greater than energy consumption weight, a three-dimensional platform model is established to calculate the effective parking time, and the train length and platform stay time are dynamically matched.

[0047] Input features are dynamically weighted. The weight of passenger flow features is increased at major stations during peak hours. The weight of features at each station is balanced during off-peak hours. During off-peak hours, rail transit scheduling focuses on energy consumption features.

[0048] Furthermore, the dynamic decision-making implementation also includes a conflict detection mechanism. The conflict monitoring uses a topology analysis algorithm to detect logical conflicts between the formation adjustment path and the status of the signal machine ahead and the track occupancy.

[0049] A specific implementation method of topology analysis algorithm detection: The dynamic path topology map is constructed based on the electronic map of the rail transit network. The nodes represent the endpoints of the track sections, and the edge weights include attributes such as section length, slope, speed limit, etc. It accesses the road condition information ahead in real time and stores and dynamically updates the node status through the graph database.

[0050] Based on the traction power and acceleration of the leading vehicle and the braking distance and deceleration rate of the trailing vehicle, the time the trailing vehicle will occupy the preceding section at its current speed is obtained. If the predicted occupancy time window of the trailing vehicle overlaps with the red light period of the preceding signal or the track occupancy period, a conflict marker is triggered. The conflict marker classifies conflicts into hard and soft conflicts. Hard conflicts require immediate replanning, while soft conflicts can be adjusted later. Conflict resolution and replanning call for an online game model, generating a compromise speed curve based on the power curve of the leading vehicle and the braking curve of the trailing vehicle, ensuring that the trailing vehicle decelerates to a safe interval before entering the conflict section. Through on-board edge nodes, the system communicates with adjacent trains to negotiate the release of switch control or adjust the parking position, giving priority to high-priority trains.

[0051] When insufficient power is detected in the vehicle ahead, a braking and traction power balance equation is established, and a compromise speed curve is obtained through iterative calculation. A specific implementation method of the braking and traction power balance equation is as follows: The braking parameters of the front and rear vehicles in rail transit are collected. The braking parameters of the front vehicle collect the current, voltage, and speed of the traction motor, and the real-time traction power P1=I×V×η is calculated, where η is the motor efficiency, I is the current, and v is the voltage.

[0052] The rear vehicle braking parameters collect regenerative braking current, braking resistor temperature, and wheel speed, and calculate the real-time braking power P2=k×I×ω, where k is the energy conversion coefficient and ω is the braking efficiency.

[0053] The power weight is adjusted according to the time period type, with traction power being prioritized during peak hours and regenerative braking energy recovery being emphasized during off-peak hours.

[0054] The Newton-Raphson method is used to iteratively calculate the maximum allowable speed that meets power balance, and further generate a compromise speed curve to ensure that the traction power of the leading vehicle and the braking power of the trailing vehicle are dynamically matched. If the current speed is greater than the maximum allowable speed, the regenerative braking priority is increased, and more braking power is allocated to the trailing vehicle; if the current speed is less than the maximum allowable speed, the leading vehicle is allowed to moderately increase traction.

[0055] Specifically, the Newton-Raphson method is a numerical method for solving the roots of equations. The principle is to use the tangent of the function to approximate the roots of the equation, select an initial point, calculate the intersection of the tangent through the derivative of the function, and gradually approximate the roots. It is used to solve single roots, repeated roots and complex roots. In this application, the speed curve is compromised.

[0056] The method for outputting correction instructions from the distribution model includes: Marshalling status data collected in real time based on onboard edge computing nodes; Through distributed communication protocols, each rail transit system adjusts operating parameters based on local decisions; The distribution model also includes instruction isolation, which includes physical isolation, logical isolation and timing isolation; The physical isolation allocates independent communication channels for instructions of different priorities; The logical isolation sets up a sandbox environment in the vehicle controller, restricting hardware access rights to non-safety-related instructions and allowing only safe instructions to directly control the traction brake system; The timing isolation adopts a time-triggered architecture and polls the instruction queue according to a fixed cycle to ensure that high-priority instructions are executed first and only instructions in the same control domain are processed in a single cycle.

[0057] The correction factor calculation is based on the edge computing framework to achieve localized real-time decision-making, and the adaptive adjustment of rail transit scheduling instructions is achieved through real-time rail transit scheduling dynamic allocation.

[0058] Lightweight data collection agents are deployed on vehicle-mounted edge nodes to obtain rail transit train power output, braking pressure, and passenger distribution in real time.

[0059] Data noise is eliminated through sensor fusion technology, and a sliding time window is used to count the extreme values ​​and changing trends of the marshaling power output, braking pressure, and passenger distribution.

[0060] Compare the current status with the safe operation traction power threshold and braking response time standard, identify the deviation parameters, and automatically adjust the correction weight of each parameter according to the fault level and operating field: The weights of power output, braking pressure, and passenger distribution parameters are dynamically adjusted, and a correction instruction set is generated based on the adjustment results, including traction compensation coefficient, braking timing offset, and passenger flow, resulting in an increase in the number of rail transit formation groups.

[0061] A preferred solution is that rail transit rapid scheduling can also set up a distributed communication protocol architecture, build a layered communication system, achieve multi-node collaborative decision-making and instruction consistency assurance, allocate low-latency channels through high-priority instructions, monopolize bandwidth, and use multicast communication channels for regular instructions. When a high-priority instruction is detected, the low-priority channel cache data is automatically cleared.

[0062] Furthermore, the traction and braking control instructions of the physical isolation layer are transmitted through an independent bus, which is physically isolated from the Ethernet channel of the vehicle information system. The safety-critical controller adopts dual independent power supply and automatically switches to the supercapacitor backup power supply when the main power supply fails.

[0063] Furthermore, in the sandbox environment in the logical isolation layer, non-secure instructions run in a virtualized container and can only access mapped IO ports; secure instructions directly access control registers through hardware-level memory protection; the on-board controller loads a whitelist policy when it starts, and only authorizes instruction sets that have passed security certification to access the underlying hardware; during operation, the behavioral analysis engine monitors abnormal access attempts and triggers instruction execution interruption.

[0064] Furthermore, the timing isolation layer divides the control cycle into fixed time slots, assigns high-priority instructions to the first time slot, and prohibits concurrent execution of cross-control domain instructions within the same cycle.

[0065] The instruction queue adopts a double-buffered queue structure. The foreground queue executes the current cycle instruction, and the background queue preloads the next cycle instruction. When an instruction execution timeout is detected, the watchdog reset is immediately triggered and the system returns to a safe state.

[0066] The hybrid verification unit includes an input layer, a verification and optimization layer, an edge optimization layer and an output layer; The input layer receives correction instructions from the distribution model, real-time rail transit marshaling status data and safety threshold library; The verification and optimization layer simulates the running status of the train after the execution of the correction instructions to verify whether it meets the power matching, braking safety and energy consumption constraints; In this application, a preferred implementation method includes: Among them, in the power matching, the total traction force of the marshaling ≥ the traction force required by the current slope × safety factor; In braking safety, the emergency braking distance is ≤ the distance to the obstacle ahead - the redundancy distance; In the energy consumption constraint, the energy consumption increment of a single adjustment instruction is ≤ the remaining capacity of the current energy storage module.

[0067] The edge optimization layer uses a constrained genetic algorithm to minimize the risk score and energy consumption increment as the objective function, generating an optimized correction instruction set, including the group length adjustment priority and dynamic speed curve; The output layer sends the optimized correction instructions to the vehicle controller and simultaneously updates the instruction execution records of the traffic database.

[0068] In this application, a specific implementation method of a hybrid verification unit includes: The input layer builds a multi-source data fusion channel to achieve precise alignment of instruction stream and state stream.

[0069] Data access and cleaning convert the correction instructions output by the distribution model, real-time data from rail transit on-board sensors, and rail transit equipment safety threshold library into standardized data frames.

[0070] Hardware timestamps are used to mark each data source, and a sliding window compensation mechanism is used to eliminate timing misalignment caused by network transmission delays.

[0071] A three-level threshold tree is established according to the rail transit power system, braking system, and energy consumption system. When equipment aging or equipment maintenance is detected, the emergency threshold branch is automatically loaded.

[0072] A further verification and optimization layer implements instruction pre-verification based on the virtual execution environment of the digital twin.

[0073] Power matching verification: input the traction force parameters in the correction command, calculate the theoretical acceleration curve, compare the rail transit power system parameters, and output power matching verification.

[0074] The load matching test compares the dynamic relationship between theoretical acceleration and actual train set weight and track gradient, and triggers an alarm code if any limit is exceeded.

[0075] Braking safety verification calculates the complete stopping distance based on the current speed, brake pressure correction value, and track adhesion coefficient, superimposes the position information of the obstacle in front, and generates a braking compensation instruction when the braking end point is predicted to invade the danger zone.

[0076] Energy consumption constraint analysis simulates the energy flow of the traction braking system based on the speed curve in the correction instruction, and predicts the energy consumption increment per unit mileage. If the predicted energy consumption exceeds the threshold and there is no safety risk, the instruction is marked as "optimizable" and handed over to the edge optimization layer.

[0077] Furthermore, the edge optimization layer integrates the evolutionary algorithm of multi-objective optimization to achieve instruction set re-optimization.

[0078] The modified instruction set is converted into a gene sequence, where the group length = gene segment 1, the speed curve = gene segment 2, and the passenger flow prediction = gene segment 3. A death penalty strategy is adopted to directly eliminate gene segments that violate safety thresholds and have high risk scores, and replace the eliminated instructions with other instructions.

[0079] The evolutionary optimization process uses the original modified instructions as seed individuals and generates the initial population through random perturbations.

[0080] The group length adjustment uses single-point crossover to verify each gene segment individually to ensure that the initial gene sequence is correct and retain legal grouping combinations. Gaussian noise is introduced to repeatedly verify the gene sequence converted from the modified instruction set. Ultimately, the individuals with the highest fitness in each generation are retained and directly enter the next generation.

[0081] Establish a two-dimensional coordinate system of risk and energy consumption, eliminate dominated solutions, and select the optimal solution based on the current rail transit operation mode: If the passenger flow forecast is high, the safety priority mode is triggered and the solution with the lowest risk score is selected; If the passenger flow forecast is low, the economic mode is triggered and the solution with the smallest energy consumption increment is selected.

[0082] Furthermore, the output layer instructions build an instruction lifecycle management system, digitally sign the optimized instruction set and append a modified instruction set ID, establish an instruction traceability chain, and synchronously issue key control instructions through the redundant CAN bus. Non-real-time instructions such as the group length adjustment plan are transmitted in batches through the network.

[0083] Furthermore, instruction corrections are stored in the instruction database, and a timing database is used to record instruction issuance time, execution results, and device response curves. The actual execution data is compared with the prediction model to automatically calibrate the simulation parameters of the verification layer.

[0084] In the present application, a preferred implementation method of the edge optimization layer is as follows: For example, the edge optimization layer may further include a learning unit, a conflict optimization unit, and a feedback unit; The learning unit collects the execution result data of historical optimization instructions, builds a feedback training set, and dynamically adjusts the risk weight and energy consumption weight; The learning unit builds a layered online incremental learning framework, establishes a real-time data access interface at the bottom layer, supports two-way communication with trackside sensors, on-board controllers, and cloud databases, builds an incremental learning engine in the middle layer, extracts features from newly input execution result data, automatically identifies data pattern changes, designs a strategy output interface at the top layer, and establishes a two-way interactive channel with the conflict optimization unit and feedback unit to collect and store the execution data of historical optimization instructions in real time.

[0085] The established historical data storage module uses a time series database to store the optimization instruction set and execution results.

[0086] Through the weight distribution statistical database, the value distribution of historical risk weights and energy consumption weights is analyzed according to fixed cycles. When continuous safety alarms are detected, the weight adjustment process is automatically started: the current risk weight is gradually increased, and the downward adjustment of the energy consumption weight is frozen. The normal adjustment mechanism is restored after no new safety alarms are detected for N consecutive cycles.

[0087] If safety requirements conflict with energy-saving requirements, a hierarchical optimization strategy is adopted through the conflict optimization unit, giving priority to forced deceleration when the braking distance is insufficient, ignoring energy consumption restrictions; If there is a conflict in the coordination of multiple groups, resources are allocated evenly; The feedback unit is used to compare the indicators before and after the execution of the optimization instruction in real time and generate a verification report. If the indicator deviation exceeds the tolerance threshold, the instruction rollback mechanism is triggered, and the group configuration is automatically restored to the previous stable state. A fault code is sent to the cloud dispatch center to start the manual intervention process.

[0088] A specific implementation of the edge optimization layer: The learning unit builds an online incremental learning framework to achieve dynamic evolution of optimization strategies and balance the weight distribution between safety and energy efficiency.

[0089] Stores optimization instruction sets and execution results, including site passenger flow forecasts, actual risk values, energy consumption values, equipment response delays, etc.

[0090] The instruction parameters are discretized and encoded, and the adjustment range of rail transit formation length is mapped to three levels of small, medium and large classifications, and passenger flow density prediction is performed in conjunction with it.

[0091] Filter successful instructions whose actual risk scores are lower than the predicted values ​​and whose energy consumption meets the standards to strengthen the current strategy. If a failed instruction that causes equipment overload or emergency braking is captured, it will be marked as a high-risk scenario feature.

[0092] The historical distribution of risk weights and energy consumption weights is calculated. When continuous safety alarms are detected, the risk value is automatically increased, the current operating mode is identified, and the learning unit is prohibited from adjusting the weights in an emergency.

[0093] The conflict optimization unit uses a hierarchical decision tree to resolve multi-objective conflicts. The conflict optimization unit also includes multi-group resource conflict resolution.

[0094] The hierarchical decision tree calculates the difference ΔS between the theoretical braking distance and the actual available braking range in real time: when ΔS < 0, a forced speed reduction command is immediately generated, overwriting the original speed curve; the speed reduction amplitude is dynamically calculated based on the absolute value of ΔS, and is reduced to the safe speed threshold at the lowest. It also triggers the energy consumption limit release protocol, allowing the traction system to operate at over-rated power.

[0095] To resolve conflicts among multiple marshaling resources, a marshaling resource competition matrix is ​​established to quantify the intensity of each marshaling's demand for track, passenger flow density, and platform resources. The maximum-minimum fair allocation principle is adopted: the minimum resource guarantee for all marshalings is determined; the remaining resources are dynamically allocated according to demand priority, the marshaling authority is doubled during peak hours, and the priority of marshalings that are stranded for more than 2 minutes is automatically increased, and they are forcibly inserted into the dispatch queue.

[0096] Furthermore, the feedback unit builds a two-way verification channel for the instruction execution effect, achieving rapid isolation and recovery of abnormal states.

[0097] The indicator deviation calculation collects the post-execution data in real time, compares it item by item with the predicted value of the optimization instruction, and uses the sliding average algorithm to eliminate the instantaneous indicator deviation.

[0098] Furthermore, a three-level alarm system is established to output the three-level deviation comparison results to determine whether to trigger the instruction rollback.

[0099] Level 1 deviations are logged without intervention; level 2 deviations trigger local parameter fine-tuning; level 3 deviations immediately freeze the control output.

[0100] The command rollback includes calling the historical configuration library to load the previous stable state parameters; issuing recovery commands through the security channel to synchronously reset the on-board controller; locking the automatic optimization function and switching to the preset safe operation mode.

[0101] Fault handling automatically generates a fault code package containing timestamp, deviation data, and environmental parameters; it is uploaded to the dispatch center through a dedicated channel, triggering the manual console wake-up protocol.

[0102] If an on-site emergency occurs, the on-board system will initiate a degradation control strategy, shut down non-core loads, issue a delay notice through the passenger information system, and activate emergency evacuation instructions.

[0103] By exchanging real-time data through a shared memory pool and adopting an event-driven architecture to ensure millisecond-level response, the rail transit marshaling optimization system can ultimately achieve autonomous evolution and safe operation in complex environments.

[0104] like Figure 3 As shown, further, the specific implementation method of the game model includes: The input of the game model includes the leading car marshaling information and the waiting car marshaling information. Among them, the leading car marshaling information takes maintaining traction efficiency as the core goal and needs to avoid speed attenuation caused by insufficient power; the rear car marshaling information prioritizes ensuring braking safety and needs to reserve sufficient braking distance to prevent the risk of rear-end collision; the dispatching center: is responsible for the global priority strategy and dynamically allocates the scheduling of passenger flow resources to adapt to the station.

[0105] Game rules include conflict detection, dynamic game and distributed communication.

[0106] Conflict monitoring includes both hard and soft conflicts; Hard conflicts include overlapping rail transit formations and complete conflict of time windows, which require immediate re-planning.

[0107] Soft contention includes partial resource contention, insufficient station capacity, and allows for delayed adjustment or dynamic priority assignment.

[0108] Dynamic games include speed curve negotiation and resource competition arbitration.

[0109] Speed ​​curve negotiation is a process in which the leading and following vehicles generate a compromise speed curve through iterative calculation based on real-time power-brake parameters.

[0110] The compromise speed curve includes speed broken lines for peak period, off-peak period and off-peak period.

[0111] During peak hours, priority is given to ensuring the traction power of the leading vehicle, allowing the following vehicle to increase braking to shorten the safety distance.

[0112] During off-peak and off-peak periods, focus on recovering energy from regenerative braking of the following vehicle to reduce overall energy consumption.

[0113] Furthermore, resource competition arbitration splits the train formations and dynamically allocates dwell time and departure intervals to give priority to meeting the needs of stations with large passenger flows.

[0114] Furthermore, distributed communication includes physical channel isolation and logical sandbox isolation, and polls the instruction queue at a fixed period to ensure that high-priority instructions are executed first.

[0115] Physical channel isolation includes emergency braking instructions, which are transmitted through independent high-priority slice networks and have exclusive bandwidth.

[0116] Logical sandbox isolation includes non-safe instructions running in virtualized containers, and only safe instructions can directly control the traction and braking system.

[0117] like Figure 4 As shown, The energy consumption monitoring line chart on the left panel shows two energy consumption indicators, with the horizontal axis representing time and the vertical axis representing energy consumption value.

[0118] The left panel displays the health status of the equipment: traction system: normal status is marked in green, braking system: pending inspection is marked in yellow, indicating that maintenance is required; door system: normal status is marked in green.

[0119] The middle area displays a map showing the train operation area, marking key locations as virtual simulation locations, and showing the track lines and station distribution. All stations are marked with serial numbers 1 to 15, including A1 corresponding to 1 station, A2 corresponding to 7 stations, and A3 corresponding to 4 stations.

[0120] The right panel includes a list of train dispatch information, including train 1 (T2024011801), adjustment type: marshaling split; execution time, location, and status.

[0121] Train 2 (T2024011802) adjustment type: dynamic marshaling completed; execution time; location: A2 station; Train 3 (T2024011803) adjustment type: emergency marshaling canceled; execution time; location; status.

[0122] The status bar at the bottom shows the system status: normal operation; normal data communication; 2 pending alarms.

[0123] Performance indicators show: CPU utilization: 45%; memory utilization: 60%; network latency: 25ms; data processing rate: 2.5MBs.

[0124] Energy consumption is related to equipment health: Energy consumption monitoring data is used for dynamic marshaling decisions, and marshaling is divided into groups to reduce energy consumption; The "pending inspection" status of the brake system affects the safety of emergency marshaling operations and the alarm must be handled with priority.

[0125] Scheduling operation logic: The train split T2024011801 is to be executed at station A1, which may be an energy-saving strategy during off-peak hours; Dynamic marshaling T2024011802 has been completed, adapting the marshaling length to the passenger flow; Emergency formation T2024011803 has been canceled and the operation has been terminated due to conflict or equipment status.

[0126] Train locations A1 and A2 stations correspond to the map-marked areas Chapel Acres and South County; The distance between locations is represented by the scale values ​​500, 200, and 900, which are used to calculate the departure interval.

[0127] System Status and Alarms: CPU and memory usage are moderate, and network latency is as low as 25ms, indicating that the system load is normal and data processing efficiency meets the standards. Pending alarms are related to brake system inspections, train formation conflicts, or other abnormal events and require further investigation.

[0128] Example 2 The data acquisition module integrates the vehicle-mounted sensors and the communication interface of the ground dispatch center to collect multi-dimensional data in real time; The passenger flow prediction module outputs the predicted probabilities of peak, off-peak and valley periods and passenger flow fluctuation characteristics; The orchestration unit includes a multi-source data interface, dynamic decision-making, and command verification. The multi-source data interface receives passenger flow prediction results, power parameters of the preceding vehicle, and road conditions ahead. The dynamic decision-making generates formation adjustment instructions and executes conflict resolution logic. The command verification simulates the execution effect of the instruction and generates a visual alarm. The distribution model obtains the correction factor and corrects the instruction through communication; The hybrid verification unit verifies the power matching, braking safety and energy consumption constraints of the formation; The edge optimization submodule uses a constrained algorithm to generate an optimization instruction set and dynamically adjusts the optimization weights through an adaptive learning module; The edge optimization submodule integrates the learning model and dynamically updates the fitness function weights based on historical instruction execution data; Conflict resolution rules include prioritizing train battery levels when multiple trains compete for the same coupling section. When safety requirements conflict with energy-saving requirements, the vehicle is forced to reduce speed to meet the braking distance constraint, regardless of energy consumption limits; Communication isolation allocates a high-priority channel of the sliced ​​network for emergency braking commands. A sandbox environment is set up in the on-board controller to restrict the access rights of non-safety commands to the traction braking system, allowing only safe commands to directly control the hardware. The feedback closed unit compares the indicator deviation before and after the instruction execution, triggering the rollback mechanism or manual intervention process.

[0129] It is important to note that the configuration and arrangement of the present application, as illustrated in various exemplary embodiments, are illustrative only. Although only two embodiments are described in detail in this disclosure, those reading this disclosure should readily understand that the dimensions, scales, structures, shapes, and proportions of various components, as well as parameter values ​​(e.g., temperature, pressure, etc.), mounting arrangements, use of materials, colors, and orientations, can be varied without materially departing from the subject matter described herein. For example, components shown as integrally formed may be comprised of multiple parts or components, the positions of components may be inverted or otherwise altered, and the nature, number, or position of discrete components may be modified or changed. Therefore, all such modifications are intended to be encompassed within the scope of this invention. The order or sequence of any process or method steps may be altered or reordered according to alternative embodiments. Any "means-plus-function" clause is intended to cover structures that perform the functions described herein, and not only structural equivalence but also structural equivalents. Other substitutions, modifications, changes, and omissions may be made in the design, operating conditions, and arrangement of the exemplary embodiments without departing from the scope of this invention. Therefore, the invention is not limited to the specific embodiments, but extends to various modifications that still fall within the scope of the appended claims.

[0130] Additionally, in order to provide a concise description of exemplary embodiments, all features of an actual embodiment may not be described (ie, those features that are not relevant to the best mode presently contemplated for carrying out the invention or those that are not relevant to implementing the invention).

[0131] It will be appreciated that in the development of any actual implementation, as in any engineering or design project, numerous implementation-specific decisions may be made. Such a development effort is complex and time-consuming, but for those of ordinary skill having the benefit of this disclosure, the development effort will be a routine task of design, fabrication, and production without undue experimentation.

[0132] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A rail transit rapid scheduling method, characterized in that: include: Through real-time interaction between on-board sensors and the ground dispatch center to collect multi-dimensional data, a passenger flow change model is established to predict passenger flow density data; Based on multi-dimensional data, the marshaling unit generates rail transit formation adjustment instructions. The traffic database retrieves the preceding vehicle power and road conditions ahead, and outputs correction instructions for the rail transit formation adjustment instructions through the distribution model. The correction instructions are input into the hybrid verification unit for edge correction, and the correction instructions are optimized to optimize the rail transit formation adjustment instructions.

2. A rail transit rapid scheduling method according to claim 1, characterized in that: The multi-dimensional data includes passenger flow density, rail transit energy consumption, and equipment health status; The rail transit energy consumption and the equipment health status are used to set a priority strategy, which prioritizes power demand and equipment health; The priority strategy includes peak hours, off-peak hours and off-peak hours.

3. A rail transit rapid scheduling method according to claim 1, characterized in that: The passenger flow change model includes a data processing layer, an input layer, a feature extraction layer, a fusion prediction layer and a mixed loss layer; The feature extraction layer includes spatiotemporal feature fusion, obtains passenger flow and passenger flow transfer features between stations through graph convolutional networks, and enhances the feature identification of key time periods; The feature extraction layer also includes multimodal feature fusion, which includes a multi-channel input structure, encodes passenger flow and time features, and sorts the encoding results; The fusion prediction layer sets up a bottom layer to capture short-term passenger flow fluctuation characteristics and a high-level learning model of long-term passenger flow trends and cycles. The fusion prediction layer also includes an enhanced attention mechanism to focus on the impact of historical time periods on the next moment and enhance the contribution of feature dimensions; The passenger flow change prediction model is provided with a three-terminal parallel output layer, and the three-terminal parallel output layer includes a peak prediction terminal, a flat peak prediction terminal and a valley peak prediction terminal.

4. A rail transit rapid scheduling method according to claim 1, characterized in that: The orchestration unit includes multi-source data interface, dynamic decision making and instruction verification; The multi-source data interface receives the peak, flat and valley peak prediction values ​​and real-time carriage passenger capacity data from the three-terminal parallel output layer of the passenger flow change model, as well as the preceding vehicle power parameters and road condition information retrieved from the traffic database; Dynamic decision-making builds a flexible allocation rule base based on multi-dimensional data. The rule base includes priority strategies and matching verification between train length and platform capacity. The instruction verification simulates the formation to adjust the instruction execution effect, verifies the feasibility of the instruction, verifies whether the traction system load exceeds the limit after coupling, and generates a visual alarm.

5. A rail transit rapid scheduling method according to claim 4, characterized in that: The dynamic decision-making implementation method includes: Dynamically adjust input feature weights based on time period. If a formation adjustment instruction conflicts with the road conditions ahead, an online game model is triggered to generate a compromise speed curve based on the power performance of the leading vehicle and the braking ability of the following vehicle. If the station capacity is insufficient, priority will be given to splitting the train groups and dynamically allocating the stop location time; The rail transit formation adjustment instruction generation rules include generating mandatory association instructions during peak hours, ignoring the rail transit maintenance status, and calling standby formations; and generating degraded operation instructions during off-peak hours.

6. A rail transit rapid scheduling method according to claim 1, characterized in that: The method for outputting correction instructions from the distribution model includes: Based on the marshaling status data collected in real time by the on-board edge computing node, the marshaling status instructions are dynamically adjusted through correction factors; Through distributed communication protocols, each rail transit system adjusts operating parameters based on local decisions; The distribution model also includes instruction isolation, which includes physical isolation, logical isolation and timing isolation; The physical isolation allocates independent communication channels for instructions of different priorities; The logical isolation sets up a sandbox environment in the vehicle controller, restricting hardware access rights to non-safety-related instructions and allowing only safe instructions to directly control the traction brake system; The timing isolation adopts a time-triggered architecture and polls the instruction queue according to a fixed cycle to ensure that high-priority instructions are executed first and only instructions in the same control domain are processed in a single cycle.

7. The rail transit rapid scheduling method according to claim 1, characterized in that: The hybrid verification unit includes an input layer, a verification and optimization layer, an edge optimization layer and an output layer; The input layer receives correction instructions from the distribution model, real-time rail transit marshaling status data and safety threshold library; The verification and optimization layer simulates the running status of the train after the execution of the correction instructions to verify whether it meets the power matching, braking safety and energy consumption constraints; The edge optimization layer uses a constrained genetic algorithm to minimize the risk score and energy consumption increment as the objective function to generate an optimized correction instruction set, which includes the marshaling length adjustment priority and dynamic speed curve; The output layer sends the optimized correction instructions to the vehicle controller and simultaneously updates the instruction execution records of the traffic database.

8. A rail transit rapid scheduling method according to claim 7, characterized in that: The edge optimization layer also collects historical optimization instruction execution result data to build a feedback training set and dynamically adjusts the risk weight and energy consumption weight; If safety requirements conflict with energy-saving requirements, a hierarchical optimization strategy is adopted, giving priority to forced deceleration when the braking distance is insufficient, regardless of energy consumption limits; If there is a conflict in the coordination of multiple groups, resources are allocated evenly; The edge optimization layer compares the indicators before and after the execution of the optimization instruction in real time and generates a verification report. If the indicator deviation exceeds the tolerance threshold, the instruction rollback mechanism is triggered, and the group configuration is automatically restored to the previous stable state. A fault code is sent to the cloud dispatch center to start the manual intervention process.

9. A rail transit rapid scheduling system, which is implemented based on a rail transit rapid scheduling method according to any one of claims 1 to 8, characterized in that: The method comprises the following specific steps: The data acquisition module integrates the vehicle-mounted sensors and the communication interface of the ground dispatch center to collect multi-dimensional data in real time; The passenger flow prediction module outputs the predicted probabilities of peak, off-peak and valley periods and passenger flow fluctuation characteristics; The orchestration unit includes a multi-source data interface, dynamic decision-making, and command verification. The multi-source data interface receives passenger flow prediction results, power parameters of the preceding vehicle, and road conditions ahead. The dynamic decision-making generates formation adjustment instructions and executes conflict resolution logic. The command verification simulates the execution effect of the instruction and generates a visual alarm. The distribution model obtains the correction factor and corrects the instruction through communication; The hybrid verification unit verifies the power matching, braking safety and energy consumption constraints of the formation; The edge optimization submodule uses a constrained algorithm to generate an optimization instruction set and dynamically adjusts the optimization weights through an adaptive learning module; The feedback closed unit compares the indicator deviation before and after the instruction execution, triggering the rollback mechanism or manual intervention process.

10. A rail transit rapid scheduling system according to claim 9, characterized in that: The edge optimization submodule integrates the learning model and dynamically updates the fitness function weights based on historical instruction execution data; Conflict resolution rules include prioritizing train battery levels when multiple trains compete for the same coupling section. When safety requirements conflict with energy-saving requirements, the vehicle is forced to reduce speed to meet the braking distance constraint, regardless of energy consumption limits; Communication isolation allocates a high-priority channel of the sliced ​​network for emergency braking commands, sets up a sandbox environment in the on-board controller, restricts the access rights of non-safety commands to the traction braking system, and only allows safety commands to directly control the hardware.

Images