Automatic air rail scheduling method and system

Abstract

The invention discloses an automatic air rail scheduling method and system, and relates to air rail scheduling, and the method comprises the steps: obtaining air rail data; calculating the residual load of each vehicle according to the vehicle data set, and generating a to-be-distributed vehicle set V; performing priority ranking on the task data set to generate a task set U; generating a vehicle-task matching relationship according to the vehicle set V and the task set U; path planning is carried out on the vehicle-task matching relation through an improved Dijkstra algorithm, and an optimal scheduling scheme of vehicle scheduling is obtained; the improved shortest path algorithm Dijkstra is used for dynamically adjusting a path weight by introducing a track congestion degree coefficient; scheduling the sky rail vehicle according to the optimal scheduling scheme; aiming at the low path planning efficiency caused by the dynamic change of the track congestion of the sky track, the planning efficiency is improved by constructing a vehicle-task bipartite graph and combining a path planning strategy for dynamically adjusting the track congestion degree and the like.

Description

Technical Field

[0001] The present application relates to the field of skytrain scheduling, and in particular to an automated skytrain scheduling method and system. Background Art

[0002] As the core infrastructure of modern smart factories and intelligent logistics, automated skyrail systems are widely used in manufacturing production lines, large-scale storage centers, and port container transportation. Through a network of suspended rails, this system enables efficient transportation and scheduling of materials and products in three-dimensional space, effectively addressing bottlenecks in ground transportation systems, such as limited space and fixed routes.

[0003] The Skytrain rail network has limited capacity, and certain key sections (such as intersections and near loading and unloading points) are prone to forming bottlenecks. Traditional scheduling methods often use static shortest path planning strategies that fail to consider the dynamic congestion state of the track. This results in multiple vehicles being assigned to the same section simultaneously, causing route conflicts and system congestion, seriously impacting overall operational efficiency.

[0004] Furthermore, most existing technologies use greedy algorithms for vehicle-task matching, based solely on simple distance or time metrics. These algorithms fail to comprehensively consider the matching between vehicle load capacity and task material requirements, as well as the current state of system congestion, making it difficult to achieve optimal global resource allocation. Particularly during periods of high-demand workloads, this simplistic matching strategy leads to suboptimal scheduling, resulting in reduced system throughput and increased response times. Summary of the Invention

[0005] In view of the low efficiency of path planning caused by the dynamic changes of track congestion in the existing technology, this application provides an automated track scheduling method and system, which improves the planning efficiency by constructing a vehicle-task bipartite graph and combining a path planning strategy with dynamic adjustment of track congestion.

[0006] One aspect of the present application provides an automated empty rail scheduling method, comprising: S1, acquiring empty rail data, wherein the empty rail data includes a vehicle data set and a task data set; the vehicle data set includes the current coordinates and current load of each vehicle; the task data set includes the material requirements and task priority of each task; S2, calculating the remaining load of each vehicle based on the vehicle data set, and generating a set of vehicles to be assigned V; S3, prioritizing the task data set to generate a task set U; S4, generating a vehicle-task matching relationship based on the vehicle set V and the task set U; S5, performing path planning on the vehicle-task matching relationship through an improved Dijkstra algorithm to obtain an optimal scheduling scheme for vehicle scheduling; the improved shortest path algorithm Dijkstra dynamically adjusts the path weight by introducing a track congestion coefficient; S6, scheduling empty rail vehicles according to the optimal scheduling scheme.

[0007] Furthermore, S2 generates a set V of vehicles to be assigned, including: extracting the maximum load limit and current load value of each vehicle from the vehicle data set; calculating the remaining load by comparing the maximum load limit and the current load value; judging whether the remaining load is greater than the minimum load threshold required for the task; for vehicles that meet the load requirements, recording the identification number, current position coordinates and remaining load of the corresponding vehicle to generate the vehicle set V.

[0008] Furthermore, S4 generates a vehicle-task matching relationship, including: calculating the Manhattan distance between each available vehicle and each task to be executed based on the vehicle set V and the task set U; calculating the resource matching degree based on the vehicle's remaining load and material requirements; constructing a vehicle-task bipartite graph based on the Manhattan distance and resource matching degree; solving the vehicle-task bipartite graph to obtain the vehicle-task matching relationship.

[0009] Furthermore, based on the Manhattan distance and resource matching, a vehicle-task bipartite graph is constructed, including: taking the vehicle set V and the task set U as the two vertex sets of the bipartite graph respectively; calculating the vehicle v i With each task t j The weight coefficient w between ij , Calculated, where d ij is the Manhattan distance, r ij is the resource matching degree, α and β are the preset balance coefficients; when the weight coefficient w ij When the vehicle v is greater than the threshold θ, i With the task t j Establish a connecting edge e between ij , and w ij As the weight value of the corresponding edge; construct the adjacency matrix G as a bipartite graph, where when there is a connecting edge e ij When the matrix element g ij =w ij , when there is no connecting edge e ij When g ij =0.

[0010] Furthermore, the vehicle-task bipartite graph is solved to obtain the vehicle-task matching relationship, including: (1) constructing the time window matrix T according to the adjacency matrix G, and each element t ij Represents vehicle v i Can you do it in task t j Complete the corresponding task within the time window; (2) According to the time window feasibility matrix T, modify the adjacency matrix G to obtain G'; where, when t ij = 0, the corresponding g ij The value is set to the preset maximum value G max , when tij =1, maintain the original g ij The value remains unchanged; (3) Normalize the rows and columns of the matrix G' to obtain G"; (4) Mark the zero elements according to G", use the minimum number of horizontal and vertical lines to cover all zero elements, and calculate the total number of covering lines. If the total number of covering lines is greater than or equal to the matrix order n, the optimal matching solution is found; if the number of covering lines is less than n, execute step (5); (5) In the G" matrix, obtain the minimum value min among the elements not covered by the lines val , subtract min from all elements not covered by the line val Add min to all elements covered by both lines val , keep other element values unchanged, return to (4) and continue execution; (6) According to the final zero element position in the matrix G' obtained in step (5), select zero elements in different rows and columns to combine, and establish vehicle v according to the selected zero element position (i, j). i With the task t j The matching relationship between them is obtained to obtain the vehicle-task matching relationship.

[0011] Furthermore, the time window matrix T is constructed, including: obtaining task t from the task dataset j Time window parameters, including the earliest start time TE j and the latest end time TL j ; According to the Manhattan distance d ij , calculate vehicle v i Move from current position to task t j Time of location TA ij ;When TA ij ≤TL j When t ij =1 means vehicle v i Able to j Arrives within the time window of , otherwise t ij =0 means vehicle v i Unable to complete the task within the specified time j .

[0012] Furthermore, S5, the path planning of the vehicle-task matching relationship is performed by the improved Dijkstra algorithm to obtain the optimal scheduling scheme for vehicle scheduling, including: constructing an air-rail network topology graph G = (N, E) according to the vehicle-task matching relationship, wherein N represents a node set, the node set includes the vehicle current position node and the task position node; E represents an edge set, and an edge represents a passable track between each node; for each edge e∈E in the air-rail network, calculating the initial weight w e , w e =d e ×(1+γ×ce ), where d e is the actual track distance corresponding to edge e, c e is the track congestion coefficient, γ is the congestion impact factor; for each vehicle-task matching pair (v i ,t j ), with the vehicle's current position as the starting point and the task location as the end point, the shortest path P from the starting point to the end point is generated by the improved Dijkstra algorithm i And the corresponding total weight W i ; For all vehicle-task matching pairs (v i ,t j )'s shortest path set {P1,P2,.....,P n}, according to the total weight

[0013] {W1,W2,.....,W n} sorting; according to the weight sorting results, confirm each vehicle-task matching pair (v i ,t j ) path allocation, and obtain the optimal scheduling solution including path planning sequence and task execution order.

[0014] Track congestion coefficient c e , the calculation formula is: e =n e / u e , where n e is the number of vehicles currently assigned to edge e, and ue is the maximum capacity threshold of edge e.

[0015] Furthermore, the shortest path P from the starting point to the end point is generated by the improved Dijkstra algorithm. i And the corresponding total weight W i , including: initializing the distance array D, setting the starting point distance to 0 and the distances of other nodes to infinity; initializing the visited node set S to an empty set; initializing the predecessor node array P, which is used to record the predecessor node of each node in the shortest path; looping the following operations until the end point is added to the set S: i. Select the node u with the smallest distance value from the unvisited nodes and add it to the set S; ii. For each adjacent node v of node u, if D[u]+w uv <D[v], then update D[u]+w uv =D[v]; iii. Update the congestion coefficient c of the relevant edge according to the currently planned path e , iv. Dynamically adjust the weight of related edges, w e '=d e ×(1+γ×c e'); According to the predecessor node array P, trace back from the end point to the starting point to build the shortest path P i ; Total weight W i Equal to the value corresponding to the end point in the distance array D.

[0016] Another aspect of the present application further provides an automated sky rail scheduling system for executing an automated sky rail scheduling method of the present application.

[0017] Compared with the existing technology, the advantages of this application are:

[0018] Due to the intertwined constraints of limited track capacity and task time windows, route planning for empty rail transit systems struggles to adapt to dynamically changing system congestion, leading to inefficient vehicle scheduling and insufficient resource utilization. Existing technologies typically employ static route planning strategies or simple greedy algorithms, which are unable to effectively address dynamic changes in track congestion. They also often separate vehicle-task matching from route planning, resulting in poor overall system performance.

[0019] This application, on the one hand, constructs a vehicle-task bipartite graph and combines the weight coefficient By comprehensively considering distance and resource matching and applying the Hungarian algorithm, the solution ensures global optimality and avoids the local optimality trap of traditional greedy algorithms. Furthermore, the time window constraint is transformed into a feasibility matrix T. Using matrix transformation techniques (G→G'→G"), infeasible matches are naturally excluded from the optimal solution, ensuring the time feasibility of the scheduling solution and improving computational efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The present application will be further described in the form of exemplary embodiments, which will be described in detail with reference to the accompanying drawings. These embodiments are not limiting, and in these embodiments, the same numbers represent the same structures, wherein:

[0021] Figure 1 is an exemplary flow chart of an automated sky rail scheduling method according to some embodiments of the present application;

[0022] Figure 2 is a schematic diagram of an exemplary application scenario of an automated sky rail dispatching system according to some embodiments of the present application;

[0023] Figure 3 is an exemplary flow chart for calculating the remaining load of a vehicle according to some embodiments of the present application;

[0024] Figure 4 is an exemplary flowchart of vehicle-task matching according to some embodiments of the present application;

[0025] Figure 5This is a topological diagram of an aerial rail network according to some embodiments of the present application. DETAILED DESCRIPTION

[0026] The method and system provided in the embodiments of the present application are described in detail below with reference to the accompanying drawings.

[0027] like Figure 1 He Ru Figure 2 As shown, this embodiment obtains empty track data from the central control system, including the position coordinates and load conditions of 5 vehicles, as well as the material requirements and priorities of 3 task points. Calculate the remaining load of each vehicle, and select three available vehicles V1, V2, and V5 to form a set V. Sort tasks T1, T2, and T3 by priority to generate a task set U. According to the vehicle's remaining load and the task material requirements, a vehicle-task bipartite graph is established to obtain a matching relationship: V1→T1, V2→T3, V5→T2. Use the improved Dijkstra algorithm for path planning to avoid congested tracks and generate the optimal path plan (solid line part). The control system issues a scheduling instruction to control the three vehicles to travel to the target task point along the planned path.

[0028] For a detailed description of the three-dimensional effects of the scheduling system, see the following: Specifically, the vehicle set V information: V1: Empty railcar 1, remaining load: 200kg; V2: Empty railcar 2, remaining load: 280kg; V3: Empty railcar 3, status: In-flight; V4: Empty railcar 4, status: In-flight; V5: Empty railcar 5, remaining load: 120kg. The task set U information: T1: Finished product loading area, material requirement: 150kg, priority: High. T2: Finished product loading area, material requirement: 80kg, priority: Medium. T3: Assembly area, material requirement: 40kg, priority: Low.

[0029] S1. Obtaining ART data. In this embodiment, the ART dispatch system obtains real-time ART system status data from the central database of the ART control center via a unified data interface. This is achieved by establishing a secure connection channel with the central database and adopting a data subscription model. When the system status changes, the relevant data is automatically pushed to the dispatch system.

[0030] The system regularly collects status information for all vehicles in the sky-high rail network (for example, every 500 milliseconds). The vehicle data set is stored in a structured table format, with each record containing the following fields: vehicle identification number (unique identification code), current track unit number, three-dimensional coordinate values (X, Y, Z), current vehicle load value (unit: kg), vehicle status flag (idle / on mission / charging / fault), maximum load limit (factory setting), and other basic information. This data is obtained through collaborative detection between on-board sensors and track monitoring points and transmitted to a central database via a wireless communication network.

[0031] Task data is sent from the Manufacturing Execution System (MES) or Warehouse Management System (WMS) to the Skytrain scheduling system via a standard data interface. Each task record contains information such as the task identification number, material code, material quantity required (in kg), material pickup location coordinates, material delivery location coordinates, task creation timestamp, earliest start time, latest completion time, and task priority (e.g., 1-10, with higher values indicating higher priority). The system stores this task data in a task queue data structure for subsequent processing.

[0032] like Figure 3 As shown, in S2, the remaining load of each vehicle is calculated based on the vehicle dataset, generating a set V of vehicles to be assigned. In this embodiment, the empty rail scheduling system first needs to screen vehicles available for task assignment and evaluate their load capacity. After the system obtains the vehicle dataset in real time from the central control database, it executes the vehicle screening process through a dedicated data processing module. This process first filters out a subset of vehicles with an "idle" status (state_flag = 1) based on the vehicle status flag (state_flag), avoiding processing of vehicles that have already been assigned tasks or are faulty.

[0033] For each idle vehicle, the system reads its maximum load limit (max_load) and current load value (current_load). The maximum load limit is an inherent parameter of the vehicle, set according to the vehicle model and typically stored in the system's equipment parameter table. The current load value is obtained in real time by a high-precision weighing sensor installed on the vehicle, with an accuracy of ±0.5kg and updated every 5 seconds.

[0034] The system uses a floating-point calculation engine to calculate the remaining load: remaining_load = max_load - current_load. For example, if the maximum load of railcar No. 3 is 300kg and the current load is 75kg, then its remaining load is 225kg. The result is rounded to one decimal place and stored in a temporary data buffer.

[0035] Next, the system determines the load threshold. By analyzing the material requirements of all pending tasks in the current task queue, the system determines a minimum load threshold (min_threshold). This threshold also takes into account a safety margin (typically set to 1.1): min_threshold = min_material_demand × 1.1. The system only retains records for vehicles with a remaining_load greater than min_threshold.

[0036] For vehicles that meet the load requirements, the system creates a structured vehicle set V. This set is implemented using an optimized hash table data structure, with the vehicle identification number as the key value, and contains the following fields: vehicle_id: the vehicle's unique identification number (such as "AGV-2023-C056"); position_X, position_Y, position_Z: the vehicle's current three-dimensional coordinate values; remaining_load: the calculated remaining load value; battery_level: the percentage of remaining battery power (as auxiliary decision-making information); last_maintenance_time: the last maintenance timestamp; the set V is indexed and optimized to support fast multi-dimensional queries based on the remaining load value or position coordinates, providing efficient data access for subsequent matching algorithms. At the same time, the system sorts the vehicles in the set V in descending order according to the remaining load, so as to give priority to vehicles with strong load capacity in scenarios with higher load requirements. Specifically, in this embodiment, the original vehicle data is detailed in Table 1, and the idle vehicle screening results are detailed in Table 2.

[0037] Table 1 Original vehicle data

[0038]

[0039] Table 2 Idle vehicle screening results

[0040]

[0041]

[0042] As shown in Table 1 and Table 2, in this embodiment, the maximum load of the No. 3 empty rail vehicle is 300 kg, and the current load is 75 kg, so its remaining load is: REMAINING_LOAD = 300 kg - 75 kg = 225 kg; if the minimum material demand in the current task queue is 100 kg, then the minimum load threshold after considering the safety margin is: MIN_THRESHOLD = 100 kg × 1.1 = 110 kg.

[0043] The vehicle set V uses an optimized hash table data structure with vehicle ID as the key value:

[0044] AGV-2023-C147:{VEHICLE_ID:"AGV-2023-C147",POSITION_X:125.4,POSITION_Y:78.2,POSITION_Z: 0.0,REMAINING_LOAD:350.0,BATTERY_LEVEL:78,LAST_MAINTENANCE_TIME:"2025-03-1508:30:00"};

[0045] AGV-2023-C056:{VEHICLE_ID:"AGV-2023-C056",POSITION_X:210.6,POSITION_Y:156.3,POSITION_Z :0.0,REMAINING_LOAD:250.0,BATTERY_LEVEL:85,LAST_MAINTENANCE_TIME:"2025-04-0214:15:00"};

[0046] AGV-2023-C003:{VEHICLE_ID:"AGV-2023-C003",POSITION_X:56.8,POSITION_Y:98.4,POSITION_Z: 0.0,REMAINING_LOAD:225.0,BATTERY_LEVEL:90,LAST_MAINTENANCE_TIME:"2025-03-2510:45:00"};

[0047] AGV-2023-C021:{VEHICLE_ID:"AGV-2023-C021",POSITION_X:185.2,POSITION_Y:42.7,POSITION_Z: 0.0,REMAINING_LOAD:200.0,BATTERY_LEVEL:95,LAST_MAINTENANCE_TIME:"2025-04-0809:20:00"}.

[0048] In this embodiment, the system prioritizes tasks using a multi-factor comprehensive evaluation method to generate a task set U. The system first extracts all task records with a "pending" status from the task queue. Each task record contains a base priority value (base_priority), which is set by the upper-level production management system and ranges from 1 to 10 (10 being the highest priority).

[0049] The sorting process uses an improved priority calculation model to dynamically adjust the actual priority of the task by taking into account multiple factors: Time urgency assessment: The system calculates the deadline margin of the task, the formula is:

[0050] time_urgency = (deadline_time - current_time) / estimated_execution_time. A smaller time_urgency value indicates a more urgent task. When this value is less than 1.5, the system increases the base priority.

[0051] Production Line Relevance Assessment: The system queries the production line dependency table to determine whether the task is associated with a critical production line. If the task material is required by a critical production line, its priority is increased.

[0052] Task waiting time compensation: To prevent low-priority tasks from being executed for a long time, the system calculates the waiting time coefficient of the task: waiting_factor = (current_time - creation_time) / average_waiting_time; when the waiting_factor exceeds the set threshold (usually 2.0), the task priority is appropriately increased.

[0053] Taking all the above factors into consideration, the system calculates the overall priority value of each task:

[0054] final_priority = base_priority + urgency_bonus + production_line_bonus + waiting_bonus. The system uses a stable merge sort algorithm to sort tasks in descending order of final_priority. After sorting, the system stores the sorted results in a double-ended queue data structure, forming the task set U.

[0055] like Figure 4 As shown, in S4, a vehicle-task matching relationship is generated based on the vehicle set V and the task set U. In this embodiment, the sky rail scheduling system first constructs a distance matrix between vehicles and tasks. Specifically, the key parameter settings are detailed in Table 3.

[0056] Table 3 Key parameter settings

[0057] Parameter name Value illustrate α 0.4 Distance factor weight coefficient, which controls the influence of distance on matching β 0.6 Resource matching weight coefficient, which controls the impact of load matching on the results θ 0.3 Weight threshold, which determines whether a valid connection is established between the vehicle and the task v 2 m / s Standard vehicle speed of the SkyRail system Tdelay 10 seconds Scheduling delay compensation time to deal with factors such as system communication delay λ 0.8～1.5 Dynamic congestion coefficient, reflecting the impact of track congestion on travel time Gmax 999999 Alternative values for infeasible time matches to ensure they are not selected by the algorithm

[0058] The system uses the three-dimensional Manhattan distance calculation method, which is suitable for reflecting the actual movement characteristics of the sky rail system in three-dimensional space. For each vehicle v in the vehicle set V i And each task t in the task set U j , the system calculates its Manhattan distance d ij :d ij =|x i -x j |+|y i -y j |+|z i -z j |, where (x i ,y i ,z i ) is the vehicle v i The current space coordinates, (x j ,yj ,z j ) is task t j The spatial coordinates of the material collection point. The system uses an efficient vectorized calculation method to process the distance matrix and stores the calculation results in the distance matrix D. The calculation results of this embodiment are detailed in Table 4. This embodiment uses the three-dimensional Manhattan distance calculation method to reflect the actual movement characteristics of the sky rail system in three-dimensional space. When vehicle vi is exactly at task tj (dij = 0), the system sets 1 / dij to the preset maximum value of 10.0 to avoid numerical overflow.

[0059] Table 4 Manhattan distance matrix

[0060] Vehicle / Mission Task T1 Task T2 Task T3 Vehicle V1 120 85 210 Vehicle V2 175 150 65 Vehicle V3 95 180 145 Vehicle V4 160 190 110

[0061] At the same time, the system calculates the resource matching matrix R to evaluate the matching degree between each vehicle and each task in terms of payload: Among them, RL i For vehicle v i Residual load, MD j For task t j Material requirement, MD j,max is the maximum material requirement across all tasks. This formula ensures that the resource matching degree is within the interval [0, 1], with values closer to 1 indicating a higher matching degree. The matching degree is highest when the vehicle's remaining load just meets the task requirements (no load capacity is wasted). The resource matching results of this embodiment are detailed in Table 5. This application evaluates the matching degree by comparing the vehicle's remaining load with the task's material requirements. When the vehicle's remaining load just meets the task requirements, the matching degree is highest, avoiding wasted load capacity.

[0062] Table 5 Resource matching matrix

[0063] Vehicle / Mission Mission T1 (150kg) Mission T2 (80kg) Task T3 (40kg) V1(200kg) 1 0.83 0.75 V2(280kg) 1 0.54 0.95 V3 (225kg) 1 0.65 0.67 V4 (120kg) 0.8 0.67 0.5

[0064] In a preferred embodiment of the present invention, the Skytrain scheduling system uses a dedicated graph processing module to construct a vehicle-task bipartite graph. This module receives a distance matrix D and a resource matching matrix R as input and outputs an adjacency matrix G representing the relationships between vehicles and tasks.

[0065] In its implementation, the system first allocates an m×n two-dimensional floating-point array in memory as the data container for the adjacency matrix G, where m represents the number of vehicles in the vehicle set V and n represents the number of tasks in the task set U. The system uses contiguous storage to improve memory access efficiency and initializes all matrix elements to 0, indicating that there are no connections between any vehicle-task pairs.

[0066] For each vehicle-task pair (v i ,t j), the system performs a comprehensive weight calculation. During the weight calculation process, the system sets two key balance coefficients: α and β. In this embodiment, after experimental verification, the system achieves the best balance between distance optimization and resource matching when α = 0.4 and β = 0.6. The weight calculation formula is: When d ij = 0 (the vehicle is exactly at the task location), the system sets It is a predetermined maximum value (10.0 in this embodiment) rather than infinity to avoid numerical overflow problems. The weight coefficient matrix of this embodiment is detailed in Table 6. This embodiment combines the distance factor (the closer the better) and the resource matching degree (the more matching the better). The coefficients α = 0.4 and β = 0.6 have been verified by experiments to achieve the best balance between distance optimization and resource matching.

[0067] Table 6 Weight coefficient matrix

[0068] Vehicle / Mission Task T1 Task T2 Task T3 Vehicle V1 0.63 0.72 0.49 Vehicle V2 0.63 0.41 0.81 Vehicle V3 0.76 0.47 0.54 Vehicle V4 0.58 0.44 0.47

[0069] The system introduces a weight threshold θ for connection screening, which is determined to be 0.3 through offline simulation and actual operation data analysis. ij >θ, the system sets g in the adjacency matrix G ij =w ij , establish vehicle v i With the task t j The effective connection relationship between them; when w ij When ≤θ, keep g ij =0, indicating that the vehicle-task pair does not participate in the subsequent matching calculation, thereby reducing the amount of calculation and avoiding unreasonable matching.

[0070] After completing the construction of the basic bipartite graph, the system further considers the time window constraint, which is the key link to ensure that the task is completed on time. The system has designed a dedicated time constraint evaluation module, which is responsible for constructing and processing the time window feasibility matrix T. This module first extracts each task t from the task database j Time window parameters, including: earliest start time TE j (the earliest time the task can start execution); the latest completion time TL j (The deadline by which the task must be completed); the system uses a time prediction method based on a physical model to calculate the estimated time TA for the vehicle to reach the task location from its current location ij : Where: v is the standard vehicle speed of the sky rail system, which is set by the system parameter configuration file and is set to 2 m / s in this embodiment; T delayThis is the scheduling delay compensation time, used to account for uncertainties such as system communication delays and vehicle startup acceleration times. In this embodiment, it is set to 10 seconds. Traditional sky rail scheduling systems often handle time window constraints as a post-verification after path planning. This results in a large number of planned paths being discarded due to failure to meet the time window, resulting in wasted computing resources and inefficient scheduling. This application eliminates time-infeasible matches before solving the bipartite graph, fundamentally avoiding the waste of computing resources.

[0071] Specifically, the system further considers the impact of track congestion on time prediction and introduces a dynamic congestion coefficient λ (0.8≤λ≤1.5), which is automatically adjusted to the following value during peak hours: Time feasibility judgment uses an accurate comparison algorithm: when TA ij ≤TL j When the system determines that the vehicle v i Able to j Arrive before the specified latest completion time, set the time window feasibility matrix element t ij =1; otherwise, set t ij = 0, indicating that the match is not feasible in time. The time window feasibility matrix of this embodiment is detailed in Table 7. This embodiment eliminates temporally infeasible matches before solving the bipartite graph, fundamentally avoiding wasted computing resources. A dynamic congestion coefficient λ is also introduced during peak periods to enhance time prediction accuracy.

[0072] Table 7 Time window feasibility matrix

[0073]

[0074] The system modifies the adjacency matrix G based on the time window feasibility matrix T and generates the adjacency matrix G' considering the time constraint: when t ij = 0, the corresponding g ij Replaced with the preset maximum value G max (set to 9999 in this embodiment) to ensure that time-infeasible matches are automatically excluded in subsequent optimizations; when t ij =1, keep the original g ij The value remains unchanged, and temporally feasible matching candidates are retained. This process ensures that temporally infeasible matches are never selected during the execution of the Hungarian algorithm, while avoiding the complexity of modifying the optimization algorithm itself. In this embodiment, the modified adjacency matrix is shown in Table 8. For example, d31 = |56.8-110.5| + |98.4-75.2| + |0-0| = 53.7 + 23.2 + 0 = 76.9m; TA31 = 76.9 / 2 + 10 = 48.45s (calculated value) → 57.5s (considering congestion coefficient λ = 1.2); TL1 = 80s > TA31 = 57.5s, temporally feasible → t31 = 1.

[0075] Table 8 Corrected adjacency matrix

[0076]

[0077]

[0078] The system uses an improved Hungarian algorithm to solve the optimal matching problem of bipartite graphs. This algorithm module is designed to efficiently handle large-scale matching problems and is specifically optimized for the sky rail scheduling scenario. First, the system normalizes the adjacency matrix G' considering the time constraint: Row normalization: For each row of the matrix G', find the minimum non-infinite value min of the row row,i , and subtract min from each element of the row row,i , so that each row contains at least one zero element; column normalization: for the matrix after row normalization, perform the same operation on each column, find the column minimum and subtract it, ensuring that each column contains at least one zero element; after normalization, the system obtains the optimized matrix G", in which the zero elements represent the potential optimal matching positions.

[0079] The system then performs a covering line algorithm to cover all zero elements in the matrix with the minimum number of straight lines: Initial marking: The system first performs an initial marking of the zero elements in the matrix G', using an improved star marking algorithm (Star-Prime method) to ensure that there is at most one star-marked zero element in each row and column; Covering operation: The system covers all columns containing star-marked zero elements and performs an iterative covering process; Covering line counting: The system calculates the minimum number of horizontal and vertical lines required to cover all zero elements lines .

[0080] The system determines the relationship between the number of coverage lines and the matrix order n, where n = min (m, n) is the effective order of the matrix: when sum lines =n, it indicates that n independent zero elements have been found, that is, the optimal matching solution; when sum lines When <n, the matrix G needs to be further adjusted".

[0081] The matrix adjustment process uses an incremental method: the system scans the matrix G", and finds the minimum value min among all elements not covered by the covering line val ; Subtract min from all uncovered elements val , reduce the weight of uncovered areas; add min to all elements covered by both lines val , maintain numerical balance; the remaining elements (covered by a line) remain unchanged; after adjustment, the system returns to the covering line step and iterates until the optimal matching solution is found.

[0082] The final matching selection adopts the row-first scanning strategy: the system scans the final optimized matrix G", and for the asterisk zero element position (i, j) in each row, establishes the vehicle v i With the task t j The system records the matching results in a structured matching table, which contains the following fields: match id Matches the record's unique identifier; vehicle id Assigned vehicle identification; task id Corresponding task identifier; match weight Original matching weight; estimated arrival,time expected time of arrival complete,time Estimated completion time; matching timestamp The matching generates a timestamp; the system uses a transaction processing mechanism to ensure atomic updates to the matching table, avoiding data inconsistencies caused by concurrent operations. The matching results are also cached in a high-speed memory area for rapid access by the subsequent path planning module. The final vehicle-task matching results obtained in this embodiment are shown in Table 9.

[0083] Table 9 Vehicle-task matching results

[0084]

[0085] S5: Path planning is performed based on the vehicle-task matching relationships to obtain the optimal vehicle scheduling solution. In the implementation of the present invention, the system constructs a topological graph G(N, E) of the ASR network based on the vehicle-task matching relationships generated in step S4. Unlike traditional methods, this system does not construct a complete topological graph of the entire ASR network. Instead, it constructs a directional topological graph based on the determined matching relationships, improving computational efficiency.

[0086] In specific implementation, the system first obtains the vehicle-task matching relationship set from step S4, for example

[0087] {(v1,t3),(v2,t5),(v4,t1)}. For each matching pair (v i ,t j ), the system creates two special nodes in the topology graph: vehicle v i The current location node and task t j For example, if vehicle v2 is currently located at the factory coordinates (35, 42), and task t5 is located at the coordinates (78, 63), then these two locations are regarded as key nodes in the topology graph.

[0088] The system also connects these special nodes to the actual sky rail network. During implementation, the system obtains track connection information from the pre-stored sky rail system physical layout database. For each vehicle or task location, the system identifies the nearest track access point and establishes a temporary connection edge. For example, if the nearest track access point for location v2 is node N 14 (40,45), the system will create a line from v2 position to N 14 Similarly, if the nearest track access point at t5 is N 28 (75,60), the system will create 28 Virtual edge to t5.

[0089] Furthermore, the system allocates network resources based on the priority of matching relationships. Matches with higher priorities are given priority in the topology graph. This is achieved by incorporating a priority factor into edge weight calculations. For example, if (v1, t3) has a higher priority than (v4, t1), the edge weights on v1's possible paths will be reduced during calculation, increasing the likelihood of those edges being selected.

[0090] Based on the special nodes and virtual connection edges created in the above steps, the system further constructs a complete topology graph G. The final topology graph G contains the following components: Comprehensive node set N: contains three types of nodes; the current position nodes of all matching vehicles, such as v1 (15, 20), v2 (35, 42), v4 (50, 30); the position nodes of all matching tasks, such as t3 (60, 25), t5 (78, 63), t1 (45, 80). Track nodes related to the above positions in the actual sky rail network, including intersections, bifurcations and access points, such as N 14 (40,45),N 22 (55,45),N 23 (70,50),N 28 (75,60), etc. The topological structure diagram of the sky rail network in this embodiment is as follows Figure 5 shown.

[0091] Comprehensive edge set E: contains three types of edges; actual track edge: represents the actual track segment in the air-rail system, such as E 45 (N 14 →N 22 ), E 62 (N 22 →N 23 ); Vehicle access edge: a virtual edge from the vehicle’s current position to the nearest track access point, such as E v2 (v2→N 14 ); Task access edge: a virtual edge from the nearest track access point to the task location, such as E t5 (N 28 →t5).

[0092] The system uses an adjacency matrix or adjacency list to represent this topology. In actual implementation, considering that the sky rail network is usually sparse (i.e., there are few connections between nodes), the system prefers to use the adjacency list representation to save storage space. For example, for node N 22 , its adjacency table entries may be: N 22 →{(N 14 ,E 45 ,15m),(N 23 ,E 62 ,20m),(N 19 ,E 51 ,18m)}, indicating N 22 Connected to three nodes through edges E 45 、E 62 and E 51 , the corresponding track distances are 15 meters, 20 meters and 18 meters respectively.

[0093] To improve the efficiency of path planning, the system also builds a spatial index structure for the topological graph, such as a quadtree or R-tree. This enables the system to quickly locate adjacent nodes in space. For example, when it is necessary to find the nearest track access point for vehicle v2, the system can use the spatial index structure to find N nodes in O(logn) time complexity. 14 , without having to traverse all track nodes.

[0094] In actual operation, the topology graph G is updated in real time to reflect changes in the system state. For example, when a vehicle has moved from its initial position to a certain position on the track, its location node is updated in the topology graph; when a track is temporarily closed for maintenance, the corresponding edge is marked as unavailable or temporarily removed from the graph.

[0095] The final topological graph G contains not only spatial and connection information, but also weight information of each edge. The initial weight is calculated by formula w e =d e ×(1+γ×c e ) is calculated and adjusted based on the priority of the matching pair. For example, for the path edges related to the high-priority task t3, the system may apply a priority coefficient λ = 0.8, multiplying the original weight by λ, making these edges more likely to be selected during path planning.

[0096] In traditional rail scheduling systems, path planning usually only considers the physical distance factor, that is, a static weight calculation method is used to simply set the edge weight to the actual track distance d eThis approach is computationally simple and straightforward, but it has significant limitations. When multiple vehicles are assigned to the same track segment simultaneously, localized track congestion can easily occur because the weight calculation doesn't take real-time load into account. Each vehicle's path planning is independent of the others, making them unaware of the plans of other vehicles, leading to uneven allocation of system resources.

[0097] Therefore, for each edge e∈E in the air-rail network, the system adopts a weight calculation formula that comprehensively considers physical distance and congestion conditions: w e =d e ×(1+γ×c e ). In practical applications, the physical distance d of the edge e Usually obtained from the layout database of the sky rail system. For example, track segment N 22 to N 23 The actual length is 15 meters, then d e = 15. The congestion impact factor γ is set according to the overall operation strategy of the system. It may be set to a larger value such as 1.8 during high-traffic periods and reduced to 0.5 during low-peak periods to reflect the sensitivity to congestion in different situations. The calculation results of the track congestion coefficient in this embodiment are detailed in Table 10, and the calculation results of the edge weights are detailed in Table 11.

[0098] Table 10 Track congestion coefficient

[0099]

[0100] Table 11 Side weight calculation results

[0101]

[0102] Track congestion coefficient c e The calculation formula is During the implementation process, the system tracks the usage of each edge in real time. For example, the key track segment N of a factory 14 to N 22 The maximum capacity u e Set to 4 (indicates that the track section allows a maximum of 4 vehicles to pass safely at the same time). If there are currently 2 vehicles assigned to use this track section, then n e =2, congestion coefficient c e =2 / 4=0.5.

[0103] The system maintains a dynamic edge usage record table to record the current allocation of each edge, such as: edge E 45 (N 14 →N 22 ): Allocated vehicles v1, v3, current n e =2,u e =4,c e =0.5; side E62 (N 22 →N 23 ): Allocated vehicle v5, current n e =1,u e =3, c e =0.33; these congestion data are updated in real time as the vehicle scheduling plan is executed, ensuring that path planning is based on the latest network status.

[0104] For each vehicle-task matching pair (v i ,t j ), the system uses a modified Dijkstra algorithm to plan the shortest path from the vehicle's current location to the task location. This method innovatively improves on the traditional Dijkstra algorithm and is specifically targeted at the dynamic multi-vehicle scheduling scenarios unique to skytrain systems.

[0105] The traditional Dijkstra algorithm, a classic single-source shortest path algorithm, performs well in static road networks. Its key features are that edge weights remain constant throughout the algorithm's execution, and each vehicle's path planning is performed independently without affecting each other. In traditional implementations, the algorithm typically follows these steps: initialize a distance array; each time, select the unvisited node with the smallest distance; update the distances of that node's neighbors; and repeat until the target node is found.

[0106] However, this static path planning approach has significant shortcomings in highly dynamic, multi-vehicle environments like skytrain systems: Multiple vehicles independently selecting the "shortest path" inevitably leads to congestion on certain popular routes, while leaving other routes idle. This leads to uneven allocation of system resources and reduced overall efficiency. For example, in a factory material distribution scenario, if multiple vehicles need to travel from a warehouse to a production line simultaneously, traditional algorithms will cause all vehicles to converge on the same main road, creating a "virtual traffic jam" that actually prolongs overall delivery time.

[0107] Unlike traditional algorithms, this improved version implements two core innovations: dynamic weight adjustment within a single path plan, and global multi-path priority ranking. These two innovations enable the system to dynamically sense and respond to changing congestion conditions, achieving global resource optimization.

[0108] In this specific embodiment, the system first initializes three key arrays: the distance array D (recording the shortest distance from the starting point to each node), the visited node set S, and the predecessor node array P. For example, for the matching pair (v2, t5), the system sets the distance of the node at position v2 to 0 and the distances of all other nodes to infinity. The path sorting and optimal scheduling scheme in this embodiment are detailed in Table 12, and the final scheduling scheme is detailed in Table 13.

[0109] Table 12 Path sorting and optimal scheduling solution

[0110]

[0111] Table 13

[0112]

[0113] During the execution of the algorithm, the system selects the unvisited node u with the smallest distance value to join the set S in each iteration. For example, the first iteration must select the node at position v2 itself; later it may select the adjacent track access point N 14 , and so on. For each selected node, the system checks all its adjacent nodes v. If the total distance to v via node u is less than the currently known distance to v, the distance value and the predecessor node record are updated.

[0114] The first key innovation of this invention is the dynamic updating of congestion status. Unlike traditional algorithms that update the network status only after the path is determined, this algorithm updates the congestion status in real time during a single path calculation. Specifically, whenever it is determined that node v will be reached through node u (i.e., D[u]+w uv <D[v] holds), the system immediately updates the congestion coefficient of edge (u, v). For example, when determining edge E 45 (N 14 →N 22 ) will be used by vehicle v2, the system calculates the updated congestion: (the original 2 cars plus the new 1 car), and adjust the edge weights accordingly: w e '=15×(1+1.5×0.75)=31.875.

[0115] This real-time dynamic adjustment mechanism is technically significant: even when calculating the same path, if multiple possible paths share a common edge, once the algorithm determines that edge should be used to reach a node, the weight of that edge is immediately increased, potentially causing the algorithm to change its mind in subsequent iterations and choose an alternative path to reach the destination. This enables adaptive congestion avoidance within bicycle path planning, a fundamental difference from traditional algorithms.

[0116] The planning algorithm for a single path ends when the destination node (task location) is added to the set S. The system then backtracks through the predecessor node array P to construct the complete path. The system begins at the destination and gradually finds the previous node for each node using the predecessor node array until it reaches the starting point. The path is then reversed to obtain the ordered path sequence Pi from the starting point to the destination. The distance value D[destination] corresponding to the destination is the total weight Wi of the path.

[0117] The second key innovation of this invention is its global sorting and order confirmation mechanism. After completing path planning for all matching pairs, the system does not simply have all vehicles execute their respective paths simultaneously. Instead, it globally sorts the set of paths {P1, P2, ..., Pn} based on their total weight. Paths with lower weights represent lower travel costs and are prioritized. For example, if path P1 has a weight of 85.6, P2 has a weight of 104.2, and P3 has a weight of 92.8, the sorted execution order is P1 → P3 → P2.

[0118] This sorting mechanism naturally enables differentiated task prioritization: Vehicles matching urgent tasks are typically assigned lower priority, resulting in priority execution; while common tasks are executed sequentially as system resources permit. In actual application at an automobile manufacturer, this mechanism ensures that material supplies to key production lines are prioritized, reducing response time for critical material deliveries from an average of 12 minutes to 4 minutes.

[0119] Based on the sorting results, the system confirms the path assignments for each vehicle-task pair sequentially, rather than simultaneously. This sequential confirmation mechanism is another technical innovation of this invention. For each confirmed path, the system immediately updates the global congestion status table, allowing subsequent path planning to avoid areas that are about to become congested. For example, after P1 is confirmed, the ne value of all edges traversed by P1 is increased by 1, and the corresponding congestion coefficient ce and edge weight we are also updated accordingly. In this way, even if P3 does not foresee a potential conflict with P1 during planning, it can still avoid the congested section occupied by P1 during actual execution.

[0120] Ultimately, the system generates a complete scheduling plan, which includes the following key information: the precise path sequence for each vehicle, such as the path of vehicle v2: v2 position → N14 → N22 → N23 → N28 → t5 position; the estimated transit time of each path point, accurate to the second; the specific operations to be performed after arriving at the task location (such as loading or unloading specific materials); and the next instructions after the task is completed (such as returning to the standby area or executing the next task).

[0121] Through this path planning method based on the improved Dijkstra algorithm, the system can generate an efficient and robust vehicle scheduling plan while taking into account resource constraints and congestion avoidance, significantly improving the scheduling intelligence level and overall operating efficiency of the sky rail system.

[0122] S6: Dispatching the ATR vehicles according to the optimal scheduling plan. After obtaining the optimal scheduling plan, the system converts it into specific control instructions and issues them to each ATR vehicle. In practice, the system first converts each vehicle's path planning sequence into a series of path point coordinates and time points, forming the vehicle's movement trajectory. The system then transmits these instructions to the corresponding vehicle control unit via the communication network.

[0123] After receiving the dispatch instructions, the vehicle control unit controls the vehicle's starting, acceleration, deceleration, turning, and stopping according to the planned route and time schedule, achieving precise movement from its current location to the target location. At the same time, the system continuously monitors the vehicle's actual operating status. If it deviates from the planned route or an emergency occurs, an emergency response mechanism is triggered, and the route planning must be re-executed if necessary.

[0124] When the vehicle arrives at the task location, it performs the corresponding material loading and unloading operations, and returns to the standby state or performs the next assigned task after completing the task. In this way, the present invention realizes the fully automated scheduling control of the sky rail system and improves the system operation efficiency.

Claims

1. An automated empty rail dispatching method, characterized in that: include: S1, obtaining empty rail data, wherein the empty rail data includes a vehicle data set and a mission data set; The vehicle dataset includes the current coordinates and current load of each vehicle; The task data set includes the material requirements and task priority of each task; S2, based on the vehicle data set, calculate the remaining load of each vehicle and generate a set of vehicles to be assigned V; S3, prioritize the task data set and generate the task set U; S4, generating a vehicle-task matching relationship based on the vehicle set V and the task set U; S5, performing path planning on the vehicle-task matching relationship by using an improved Dijkstra algorithm to obtain an optimal scheduling solution for vehicle scheduling; the improved shortest path algorithm Dijkstra dynamically adjusts the path weight by introducing a track congestion coefficient; S6, dispatching the empty rail vehicles according to the optimal dispatching plan.

2. The automated empty rail dispatching method according to claim 1, characterized in that: S2, generate the set of vehicles to be assigned V, including: Extract the maximum load limit and current load value of each vehicle from the vehicle dataset; Calculate the remaining load by comparing the maximum load limit and the current load value; Determine whether the remaining load is greater than the minimum load threshold required for the task; For vehicles that meet the load requirements, the identification number, current position coordinates and remaining load of the corresponding vehicle are recorded to generate a vehicle set V.

3. The automated empty rail dispatching method according to claim 2, characterized in that: S4, generating a vehicle-task matching relationship, including: Based on the vehicle set V and the task set U, calculate the Manhattan distance between each available vehicle and each task to be executed; Calculate resource matching based on vehicle remaining load and material requirements; Construct a vehicle-task bipartite graph based on Manhattan distance and resource matching; Solve the vehicle-task bipartite graph and obtain the vehicle-task matching relationship.

4. The automated empty rail dispatching method according to claim 3, characterized in that: Construct a vehicle-task bipartite graph, including: The vehicle set V and the task set U are respectively regarded as two vertex sets of the bipartite graph; Calculate v for each vehicle i With each task t j The weight coefficient w between ij , Calculated, where d ij is the Manhattan distance, r ij is the resource matching degree, α and β are the preset balance coefficients; When the weight coefficient w ij When the vehicle v is greater than the threshold θ, i With the task t j Establish a connecting edge e between ij , and w ij As the weight value of the corresponding edge; Construct the adjacency matrix G as a bipartite graph where when there is a connecting edge e ij When the matrix element g ij =w ij , when there is no connecting edge e ij When g ij =0.

5. The automated empty rail dispatching method according to claim 4, characterized in that: Solve the vehicle-task bipartite graph to obtain the vehicle-task matching relationship, including: (1) According to the adjacency matrix G, construct the time window matrix T, each element t of the matrix T ij Represents vehicle v i Can you do it in task t j Complete the corresponding task within the time window; (2) According to the time window feasibility matrix T, the adjacency matrix G is modified to obtain G'; where, when t ij = 0, the corresponding g ij The value is set to the preset maximum value G max , when t ij =1, maintain the original g ij The value remains unchanged; (3) Normalize the rows and columns of the matrix G' to obtain G"; (4) Mark the zero elements according to G”, use the minimum number of horizontal and vertical lines to cover all zero elements, and calculate the total number of covered lines. If the total number of covered lines is greater than or equal to the matrix order n, the optimal matching solution is found; if the number of covered lines is less than n, execute step (5); (5) In the G' matrix, get the minimum value min among the elements not covered by the line val , subtract min from all elements not covered by the line val Add min to all elements covered by both lines val , keep other element values unchanged, return to (4) and continue execution; (6) According to the final zero element position in the matrix G' obtained in step (5), select zero elements in different rows and columns to combine, and establish the vehicle v according to the selected zero element position (i, j) i With the task t j The matching relationship between them is obtained to obtain the vehicle-task matching relationship.

6. The automated empty rail dispatching method according to claim 5, characterized in that: Construct the time window matrix T, including: Get task t from the task dataset j Time window parameters, including the earliest start time TE j and the latest end time TL j ; According to the Manhattan distance d ij , calculate vehicle v i Move from current position to task t j Time of location TA ij ; When TA ij ≤TL j When t ij =1 means vehicle v i Able to j Arrives within the time window of , otherwise t ij =0 means vehicle v i Unable to complete the task within the specified time j .

7. The automated empty rail dispatching method according to claim 5, characterized in that: S5, using the improved Dijkstra algorithm to perform path planning based on the vehicle-task matching relationship, including: Based on the vehicle-task matching relationship, a topological graph of the empty track network G = (N, E) is constructed, where N represents a node set, including the vehicle's current location node and the task location node; E represents an edge set, and an edge represents a passable track between nodes; For each edge e∈E in the air-rail network, calculate the initial weight w e , w e =d e ×(1+γ×c e ), where d e is the actual track distance corresponding to edge e, c e is the track congestion coefficient, γ is the congestion impact factor; For each vehicle-task matching pair (v i ,t j ), with the vehicle's current position as the starting point and the task location as the end point, the shortest path P from the starting point to the end point is generated by the improved Dijkstra algorithm i And the corresponding total weight W i ; For all vehicle-task matching pairs (v i ,t j )'s shortest path set {P1,P2,.....,P n }, according to the total weight {W1,W2,.....,W n } to sort; According to the weight sorting results, confirm each vehicle-task matching pair (v i ,t j ) path allocation, and obtain the optimal scheduling solution including path planning sequence and task execution order.

8. The automated empty rail dispatching method according to claim 7, characterized in that: Track congestion coefficient c e , the calculation formula is: e =n e / u e , where n e is the number of vehicles currently assigned to edge e, and ue is the maximum capacity threshold of edge e.

9. The automated empty rail dispatching method according to claim 7, characterized in that: Generate the shortest path P from the starting point to the end point through the improved Dijkstra algorithm i And the corresponding total weight W i ,include: Initialize the distance array D, set the starting point distance to 0, and the distances of other nodes to infinity; Initialize the visited node set S to an empty set; Initialize the predecessor node array P, which is used to record the predecessor node of each node in the shortest path; Loop through the following operations until the endpoint is added to the set S: i. Select the node u with the smallest distance value from the unvisited nodes and add it to the set S; ii. For each adjacent node v of node u, if D[u]+w uv <D[v], then update D[u]+w uv =D[v]; iii. Update the congestion coefficient c of the relevant edge based on the currently planned path e , iv. Dynamically adjust the weight of related edges, w e '=d e ×(1+γ×c e '); According to the predecessor node array P, trace back from the end point to the starting point to build the shortest path P i ; Total weight W i Equal to the value corresponding to the end point in the distance array D.

10. An automated sky rail dispatching system, characterized in that: include: At least one processing unit; used to execute instructions to implement the automated sky rail scheduling method according to any one of claims 1 to 9.

Images