A ship section transfer motion state intelligent scheduling method and system

Abstract

This invention discloses a dynamic intelligent scheduling method and system for ship segment transfer. The method involves grouping tasks and vehicles based on segment weight and vehicle tonnage; calculating vehicle preference for tasks and task preference for vehicles based on empty mileage, process number, and delivery time, achieving bidirectional dynamic matching; adaptively switching between production pull, low-energy consumption, and comprehensive scheduling modes based on yard occupancy; and dynamically adjusting weight parameters through deep learning and multi-objective optimization to form a closed-loop optimization cycle of "environmental change - parameter adjustment - scheduling - feedback." This invention achieves real-time scheduling in dynamic scenarios, improving efficiency by 42% compared to manual scheduling, reducing task delay rate to less than 2%, minimizing daily vehicle transport time variation to less than 0.5 hours, and significantly reducing empty mileage and energy consumption.

Description

Technical Field

[0001] This invention relates to the field of intelligent scheduling technology for the transfer of ship construction sections, and in particular to a dynamic intelligent scheduling method and system for the transfer of ship sections. Background Technology

[0002] In the shipbuilding process, sectional transfer is a crucial link connecting various production processes, and its scheduling efficiency directly affects the shipbuilding cycle, production costs, and resource utilization. With the increasing tonnage of shipbuilding, the refinement of production processes, and the parallel construction of multiple ships, sectional transfer scenarios are characterized by large workloads, high dynamism, and complex constraints, making it difficult for traditional scheduling methods to meet actual production needs.

[0003] Current ship segment transfer scheduling relies heavily on manual planning or algorithmic optimization, which presents several significant problems: First, when faced with dynamic emergencies such as task insertions, vehicle breakdowns, and road congestion, the decision-making response is delayed, easily leading to excessive deviations between planned and actual progress. Second, the lack of refined grouping and scheduling based on segment weight differences results in the mixed participation of flatbed trucks of different tonnages in scheduling. This not only presents the problem of smaller tonnage vehicles being unable to carry heavier segments but also leads to uneven vehicle loads, increased empty mileage, and serious energy waste.

[0004] Existing related patent CN120106497A uses a genetic algorithm combined with a generative adversarial network to optimize parameters. It incorporates indicators such as idle time and travel time through individual fitness calculations to meet priority scheduling requirements and solves the problem of algorithm parameter selection. However, this technology uses a centralized computing architecture and lacks a hierarchical collaborative system. All task sequence calculations rely on a single node, lacking real-time edge processing capabilities and unable to handle dynamic scenarios such as task insertion and road congestion. Furthermore, patent CN119539394A optimizes the segmented entry and exit order and stack location selection in the storage yard by integrating a genetic simulated annealing algorithm with heuristic rules, reducing reliance on dispatcher experience. However, this technology does not involve core aspects such as real-time vehicle path planning and dynamic task matching, lacking the ability to judge and dynamically adjust task balance within a group in real time. It cannot quickly recalculate paths and reallocate tasks in the event of emergencies such as vehicle malfunctions or road construction restrictions, resulting in insufficient scheduling efficiency. Summary of the Invention

[0005] Purpose of the invention: The present invention aims to provide a hierarchical, collaborative, dynamically adaptable, and energy-efficient intelligent scheduling method for ship segment transfer; another purpose of the present invention is to provide a dynamic intelligent scheduling system for ship segment transfer.

[0006] Technical solution: The ship section transfer dynamic intelligent scheduling method of the present invention includes the following steps:

[0007] (1) Determine the segmented transportation task groups and vehicle groups based on the production tasks and load tonnage;

[0008] (2) By the inter-group difference rate, determine whether the tasks between groups are balanced. If the tasks between groups are balanced, the vehicle sends a task request to its group; if the tasks between groups are not balanced, the vehicle sends a task request to other groups.

[0009] (3) Calculate the vehicle’s preference for candidate segmented tasks based on the empty mileage of the vehicle to the segmented storage location in the segmented transportation task, the current process of the segmented storage location in the segmented transportation task, and the difference between the estimated delivery time of the segmented transportation task and the current time.

[0010] (4) Based on the yard occupancy rate, the production pull mode, low energy consumption mode or comprehensive scheduling mode are adaptively switched. Based on the empty mileage of the vehicle to the segmented storage location in the segmented transportation task and the cumulative mileage of the vehicle, the vehicle preference of the candidate task in the current mode is calculated.

[0011] (5) Perform bidirectional dynamic matching based on the vehicle's preference for candidate segment tasks and the candidate tasks' preference for the vehicle under the current mode;

[0012] (6) The weights in the calculation of the vehicle's preference for candidate segment tasks and the candidate task's preference for the vehicle constitute a dynamic weight parameter set; based on the vehicle's total mileage and total transportation time, a deep learning model is established, and multi-objective optimization is performed with the vehicle's total mileage and total transportation time being minimized, and supervised learning is carried out.

[0013] (7) Compare the results before and after parameter tuning according to the window period, determine the distance improvement rate and time improvement rate, and construct the excitation function;

[0014] (8) The Actor-Critic or PPO framework, consisting of a policy network Actor and a value network Critic, is adopted. The input is a state vector and the output is the optimal dynamic weight parameter set. The state vector includes task characteristics, vehicle availability, tonnage grouping, yard occupancy rate, path congestion degree and historical execution feedback.

[0015] (9) Based on the optimal dynamic weight parameter set, recalculate the vehicle's preference for candidate segment tasks and the candidate task's preference for the vehicle under the current mode, and perform bidirectional dynamic matching again.

[0016] Furthermore, in step (2), the inter-group difference rate R is...

[0017]

[0018]

[0019] in, Let N be the task load coefficient of the vehicles in the i-th group. i Let n be the number of unfinished tasks in the i-th group. i The number of currently executable vehicles in group i;

[0020] The closer R is to 0, the more balanced the tasks are. If R > 0.5, the tasks are unbalanced between groups.

[0021] Furthermore, in step (3), the vehicle's preference for candidate segment tasks... for

[0022]

[0023] in, Let this be the empty mileage of the vehicle to the segmented storage location in the i-th task. Let be the current process number of the segment storage location in the i-th task. Let be the difference between the estimated delivery time of the i-th task and the current time. , and As weight.

[0024] Furthermore, in step (4), if the yard occupancy rate is ≥95%, the production pull mode is adopted, and subsequent processes are transported in segments with priority; if the yard occupancy rate is <60%, the low-energy consumption mode is adopted, following the principle of matching nearby; if the yard occupancy rate is greater than 60% and less than 95%, the comprehensive mode is adopted.

[0025] Furthermore, in step (4), the candidate task's preference for vehicles is...

[0026]

[0027] in, Let this be the empty mileage of the vehicle to the segmented storage location in the i-th task. This refers to the vehicle's cumulative mileage. and As weight.

[0028] Furthermore, in step (7), the excitation function for

[0029]

[0030]

[0031]

[0032] Among them, α, β, Dynamically configured in the cloud based on management preferences. For distance improvement rate, The total driving distance of the baseline scheme within a window period, To use the current weight Total driving distance during the same window period For time improvement rate, This represents the total completion time of the baseline plan within the window period. This represents the total completion time of the window period under the current weight.

[0033] The ship section transfer dynamic intelligent scheduling system of the present invention includes:

[0034] The cloud layer is used to handle task data aggregation, scheduling parameter generation and issuance, global visualization, and statistical analysis.

[0035] The edge layer is used to set up edge calculators according to the vehicle tonnage group. Each edge calculator receives the task grouping and matching parameters issued by the cloud server of the cloud layer, performs edge-side calculations in combination with the real-time status of the vehicles in its group, issues transportation instructions to the vehicles in its group, and reports the scheduling execution process data and statistical results to the cloud server.

[0036] The terminal layer includes several vehicles. The vehicle's onboard controller includes a positioning module, a sensing module, a communication module, and an execution control module. The positioning module provides vehicle position, attitude, and motion information; the sensing module acquires information about the vehicle's surrounding environment, obstacles, and operating status; the communication module enables bidirectional data communication between the vehicle and the edge layer; and the execution control module receives transportation instructions and drives the vehicle to complete driving, steering, braking, and hydraulic actuation actions.

[0037] The cloud server in the cloud layer issues task grouping and matching parameters to the edge calculator, the edge calculator issues transportation instructions to the vehicle, the vehicle reports status data to the edge calculator, and the edge calculator reports data to the cloud server, thus forming a closed-loop link of parameter issuance, instruction execution, status feedback and data reporting.

[0038] Furthermore, a task management service is set up in the cloud layer to create, publish, and maintain segmented transfer tasks and their status flow;

[0039] The cloud layer sets up a vehicle resource management service to maintain vehicle resource information, vehicle grouping information, and availability status;

[0040] The cloud layer sets up a map editing service to maintain the site road topology, work areas, key nodes, and no-entry or restricted-entry rules, providing a basic map for edge layer path planning;

[0041] The cloud layer sets up an AI parameter tuning model training service to train and adjust the matching parameters based on historical and real-time data, and then sends the trained matching parameters to the edge layer.

[0042] The cloud layer is configured with a report analysis service to perform statistical analysis on empty runs, energy consumption, mileage and overtime indicators and output reports.

[0043] The cloud layer is configured with a vehicle dynamic viewing service to display the real-time status of vehicles; and a segmented location viewing service is configured to display the segmented locations and their association information with the task.

[0044] The cloud layer is set up with a site status viewing service to display road traffic status, congestion status, or work area status;

[0045] The cloud layer is configured with a task progress viewing service to display the task execution progress and the status of key nodes;

[0046] A vehicle trajectory viewing service is set up in the cloud layer to display the vehicle's historical trajectory and operation playback;

[0047] A cloud-based system is configured to perform a balance assessment service, which is used to assess the balance of vehicle groups and provide a basis for resource optimization.

[0048] The cloud layer sets up an inter-group vehicle dispatch service to dispatch vehicles between different groups and synchronize the dispatch results to the edge layer for execution.

[0049] Furthermore, the edge layer sets up a vehicle and task matching service within the group, which is used to receive task grouping and matching parameters from the cloud, and dynamically match vehicles and tasks based on the real-time status of vehicles in the group and task requirements to form an executable allocation result.

[0050] The edge layer sets up a vehicle route planning service, which is used to calculate the driving route of the vehicle from the current location to the pickup point and delivery point based on the road topology formed by cloud map editing and the site and vehicle status information reported by the terminal, and to perform dynamic replanning when the site status changes.

[0051] The edge layer sets up a task balance judgment service between vehicles to judge the balance of task load, waiting time and execution time of vehicles in the group, and triggers task reallocation or path recalculation when there is an imbalance.

[0052] The edge layer is configured with a transportation mileage statistics service to track the mileage of vehicles during effective transportation processes.

[0053] An empty-run mileage statistics service is set up at the edge layer to count the mileage of vehicles without load;

[0054] An energy consumption statistics service is set up at the edge layer to collect energy consumption data during vehicle operation.

[0055] The edge layer converts the output of the above services into transportation instructions for individual vehicles and sends them to the terminal layer. It also summarizes the execution process data and statistical results and reports them to the cloud layer.

[0056] Furthermore, a vehicle positioning module is set up at the terminal layer, which outputs the vehicle's position and attitude information;

[0057] The terminal layer is equipped with a status perception module, which outputs environmental and operational status information from LiDAR and cameras;

[0058] The terminal layer is equipped with a vehicle navigation module, which performs navigation based on the path and instructions sent down from the edge layer;

[0059] The terminal layer is equipped with a vehicle control module, which uses the actuator and hydraulic controller of the execution control module to complete the vehicle movement and hydraulic execution.

[0060] The terminal layer is equipped with a task request module. When the vehicle is idle, completes a task, or encounters an abnormality, it sends a task request or status update to the edge layer through the communication module.

[0061] The terminal layer reports vehicle status data to the edge layer. Based on the status data, the edge layer performs real-time updates for intra-group matching, route planning, and task balancing, and then reissues the updated transportation instructions to achieve a dynamic scheduling closed loop.

[0062] Beneficial Effects: Compared with existing technologies, the significant advantages of this invention are: 1. The method of this invention groups tasks and vehicles according to segment weight and vehicle tonnage; calculates vehicle preference for tasks and task preference for vehicles based on empty mileage, process number, and delivery time, achieving bidirectional dynamic matching; adaptively switches between production pull, low-energy consumption, and comprehensive scheduling modes based on yard occupancy rate; and dynamically adjusts weight parameters through deep learning and multi-objective optimization, forming a closed-loop optimization of "environmental change - parameter adjustment - scheduling - feedback"; 2. The system of this invention constructs a dynamic scheduling system with cloud layer, edge layer, and terminal layer working collaboratively, realizing unified issuance of task grouping and matching parameters, real-time matching of vehicles and tasks within a group, vehicle route planning, task balance judgment among vehicles, rapid issuance of transportation instructions, vehicle status data feedback, mileage and energy consumption statistics, and report analysis, thereby improving the efficiency of ship segment transfer and reducing empty mileage and energy consumption, solving the problems of road congestion, task insertion, uncertain vehicle completion time, and discrepancies between plans and actual progress caused by segment movement within the yard during shipyard segment transfer. 3. The problem of large deviations from actual progress; 4. In this invention, vehicle route planning, task and vehicle allocation, and task sorting within the group are all processed in real time by the group's edge calculator, without needing to go through the cloud. Even if communication with the cloud is temporarily interrupted, the group's operations will still proceed normally; when an edge calculator in one group fails, it will not affect other groups; 5. This invention introduces a new tonnage class, requiring only the addition of one edge calculator node, making it easy to expand and enabling scheduling in dynamic environments caused by vehicle additions and subtractions, task insertions, and segmented relocations; 6. This invention does not require calculating the distance from the starting point to the destination of segmented transportation, but mainly analyzes the empty driving distance. For 100 tasks and 10 vehicle terminals concurrently requesting tasks, matching calculations can be completed within 100ms, improving transportation efficiency by 42% compared to manual scheduling, with a delay rate of less than 2%; 7. This invention considers both task balance and time balance, with the difference in daily transportation time for each vehicle being less than 0.5 hours. Dynamic scheduling is based on actual vehicle location, task execution status, and yard occupancy rate, reducing actual execution time by 20% compared to the global optimization scheduling scheme before transportation begins. Attached Figure Description

[0063] Figure 1 This is a hardware architecture diagram of the system of the present invention;

[0064] Figure 2 This is a diagram of the layered microservice architecture of the present invention;

[0065] Figure 3 This is a flowchart of the segmented transportation dynamic intelligent scheduling method of the present invention;

[0066] Figure 4 This is a flowchart illustrating the matching process between the task and the flatbed vehicle in this invention.

[0067] Figure 5 This is a flowchart of the dynamic intelligent parameter tuning process of the present invention. Detailed Implementation

[0068] This paper takes a shipyard section transfer scenario as an example. Within the same construction cycle, multiple ships are being built concurrently, with sections frequently transferred between processes such as fabrication, pre-outfitting, sandblasting, painting, and outfitting. During this transfer process, dynamic events such as task insertion, vehicle breakdowns, road congestion, and temporary road restrictions caused by section relocation in the storage yard occur simultaneously. To address the problems of delayed response, uneven vehicle load, and high empty-run mileage and energy consumption associated with traditional manual scheduling, this embodiment adopts a dynamic scheduling system that coordinates the cloud layer, edge layer, and terminal layer. The cloud layer uniformly issues task grouping rules and matching parameters; the edge layer performs real-time matching of vehicles and tasks within a group, path planning, and load balancing; and the terminal layer handles status awareness, command execution, and data feedback, forming a closed-loop dynamic scheduling system.

[0069] The dynamic intelligent scheduling method for ship segment transfer described in this invention comprises the following steps:

[0070] Step (1): Import the transfer task from the production management system and standardize the modeling of elements such as segment weight, start and end position, process number and deadline.

[0071] In this embodiment, a batch of transportation tasks are imported from the production management system within a transfer window on a specific workday. Each task includes at least a segment number, storage location coordinates, weight, destination coordinates, process number, and expected arrival time. To unify the task-driven logic, the process numbers for the five production steps—segment fabrication, pre-outfitting, sandblasting, painting, and outfitting—are defined as 1 to 5, respectively. When the production management system issues an urgent insert task, the task management service automatically adds the insert task to the task list and simultaneously pushes the task to the cloud and the corresponding edge group.

[0072] For ease of illustration, the imported tasks within this window period are assumed to consist of several segments. The example extracts a portion of the tasks to demonstrate the matching process. Task segment IDs include SEG245, SEG136, SEG938, SEG950, SEG214, SEG258, SEG331, SEG641, SEG666, and SEG817. Each segment stores coordinates, process numbers, and deadlines, as shown in the subsequent matching results. Segment weights are measured in tons, ranging from light to heavy loads, to demonstrate the necessity of grouping and allocation strategies.

[0073] Step (2): Initialize vehicle resources and availability status, and set the load capacity and number of flatbed trucks for each segment in the system.

[0074] Step (3) Group vehicles according to load tonnage and automatically assign tasks to the corresponding groups based on segment weight, while setting safety adaptation constraints and dynamic online mechanisms.

[0075] (31) Initialize flatbed truck resources in the vehicle resource management service, and enter the vehicle ID, load capacity class, daily availability status, cumulative mileage, and energy consumption model parameters for each flatbed truck. In the example, group A is the light-load vehicle group, including vehicles A1, A2, A3, and A4. Group B is the medium-load vehicle group, including vehicles B1, B2, and B3. Group C is the heavy-load vehicle group, including vehicles C1, C2, and C3. The real-time coordinates of each vehicle are reported by the terminal positioning module as the basic input for matching and path planning. Each vehicle is equipped with a "task request" trigger mechanism, which allows the vehicle to actively initiate a task request to the edge calculator of its group after it becomes idle or completes a previous task.

[0076] (32) In this embodiment, the segment weight is defined as D, and the vehicle load capacity thresholds are set as PA, PB, and PC, with PA < PB < PC. If PB < D < PC, the segment task is assigned to group C. If PA < D < PB, it is assigned to group B. If D < PA, it is assigned to group A. Vehicles are assigned to groups A, B, and C according to their rated load capacity. This grouping ensures that light-load vehicles do not undertake heavy-load tasks, avoiding operational risks caused by mismatch, and at the same time provides a basic constraint for the subsequent "heavy-load group can be allocated to light-load group".

[0077] (33) Once transportation begins, if a newly inserted task arrives at the cloud, the cloud will automatically assign it to the corresponding edge group according to the weight rules mentioned above and send it to the edge calculator of that group. If a new vehicle is added or a vehicle recovering from a fault comes online, the vehicle resource management service will update its availability status and assign it to a group according to its load weight, and synchronize it to the edge side to add it to the matching candidate set.

[0078] Step (4): Calculate the task load and inter-group difference rate for each group, and determine whether there is an imbalance based on the threshold.

[0079] To avoid a situation where a certain tonnage group experiences "task backlog and insufficient vehicles," leading to a prolonged overall completion time, this embodiment maintains inter-group load balancing judgment logic simultaneously on both the cloud and edge sides. The task load coefficient for each group's vehicles is set as the ratio of the number of currently incomplete tasks to the number of currently available vehicles. A further defined inter-group difference rate, R, is used to quantify the load difference between groups; the closer R is to 0, the more balanced the tasks are. When R > 0.5, inter-group task imbalance is considered, triggering an inter-group resource allocation strategy. Since small-load vehicles cannot handle large-tonnage segments, this embodiment sets a hard constraint prohibiting allocation from light-load groups to heavy-load groups, while allowing heavy-load groups to allocate to light-load or medium-load groups, and allowing medium-load groups to allocate to light-load groups. After allocation is triggered, the cloud generates allocation results and sends them to the edge for execution, allowing heavy-load group vehicles to enter light-load or medium-load groups to perform light-load or medium-load tasks after completing their heavy-load tasks, thereby improving overall resource utilization and reducing waiting time.

[0080] In this embodiment's window period, it is assumed that in the early stages, heavy-load tasks from Group C arrive in concentrated numbers, and road congestion leads to extended completion times. Group C's load factor is higher than other groups, and R exceeds 0.5. The cloud extends the subsequent executable task authorization for some Group C vehicles to Group B or Group A. After a Group C vehicle completes its current heavy-load task, it can initiate a task request to the edge calculator of its newly assigned group to execute light-load or medium-load tasks. This allocation process does not change the segmented weight grouping principle, but only changes the range of executable task sets for vehicles, thereby maintaining safety and adaptability.

[0081] Step (5) Calculate and sort the vehicle’s preference for candidate tasks, and select production pull, low energy consumption or comprehensive mode according to the yard occupancy rate, and dynamically adjust the weight of empty driving distance and process priority.

[0082] The vehicle-task matching system within the group employs a "two-way preference" and "mode-switchable" approach. First, the vehicle side calculates its preference for each candidate task for the remaining tasks, forming a preference list and prioritizing requests for tasks ranked higher. The preference design focuses on factors such as empty driving distance, process priority, and deadline urgency. Let d be the empty driving distance from the vehicle to the segmented storage location in the i-th task. i Let the current process number of this segment be p. i Let the difference between the estimated delivery time of the task and the current time be Δt. i If the system does not record the expected delivery time, then Δt will be used. i Assign a preset constant to avoid unreasonable task delays. Vehicle's task preference S task It can be constructed as a weighted combination of various indicators, S task A smaller value indicates that the task is more "suitable" for the vehicle, ranks higher in the priority list, and is sent to the vehicle with higher priority. task The value is calculated as follows:

[0083]

[0084] In the formula, Let this be the empty mileage of the vehicle to the segmented storage location in the i-th task. Let be the current process number of the segment storage location in the i-th task. Let be the difference (in seconds) between the estimated delivery time of the i-th task and the current time. If the system has not recorded the estimated delivery time, ;

[0085] To match different actual production rhythms, this invention supports three modes: "Production Pull Mode," "Low Energy Consumption Mode," and "Combined Mode." Schedulers can manually select a mode on the cloud interface; if not set, the system will automatically decide based on yard occupancy. Yard occupancy reflects the remaining stack space and the number of relocations. This embodiment sets the following decision logic: If yard occupancy is ≥95%, the Production Pull Mode is used, prioritizing the transport of subsequent process segments to free up stack space for previous processes, thereby alleviating yard congestion. If yard occupancy is <60%, the Low Energy Consumption Mode is used, primarily employing the principle of proximity matching to reduce empty running distance and energy consumption. If yard occupancy is between 60% and 95%, the Combined Mode is used, balancing production rhythm and energy consumption.

[0086] In this embodiment, the yard occupancy rate is set to 96% at a certain moment, and the system automatically selects the production pull mode. At this time, the weights of process priority and deadline in the vehicle's task preference are increased, while the weight of empty driving distance is relatively decreased, to ensure the production objective of "transferring the later-stage processes first to free up storage space, and then transferring the subsequent processes." To illustrate the differences brought about by mode switching, matching example results for the production pull mode, low-energy mode, and integrated mode are given below.

[0087] Step (6) calculates the task's preference for the requested vehicle and combines vehicle-side supplementary requests with edge-side iterative updates when multiple vehicles compete, to achieve bidirectional matching and conflict mitigation.

[0088] To avoid an imbalance between mileage and energy consumption caused by "the same vehicle always undertaking long-distance tasks," this embodiment further introduces a task preference degree S for the vehicle. vehicle This is used to measure the priority of a task among multiple requesting vehicles. A task's vehicle preference can be constructed by combining the distance of the vehicle to the task's starting point and the vehicle's cumulative mileage, prioritizing vehicles that are closer and have less cumulative mileage, thus achieving mileage and waiting balance. Let the cumulative mileage of the vehicle be L. j Let the distance from the vehicle to the starting point of the mission be d. j The task's preference for vehicles can vary with d. j With L j Increase and increase, S vehicle The smaller the value, the higher the priority.

[0089] The bidirectional matching process is executed as follows:

[0090] The edge maintains a list of remaining tasks and a set of available vehicles. When a task receives only one vehicle request, it accepts the request and completes the matching process. When a task receives multiple vehicle requests simultaneously, the task calculates the preference for each vehicle and selects the vehicle with the highest preference to complete the matching. Unselected vehicles are removed from their preference lists after being matched for a task, and then make requests to the remaining tasks. This process is repeated until the list of remaining tasks is empty. Matched vehicles do not make new requests until the task is completed, and matched tasks are removed from the task list to avoid duplicate assignment. This process can be processed concurrently on the edge, adapting to highly dynamic scenarios with multiple vehicles making requests simultaneously.

[0091] Step (7) Output example matching results under different scheduling modes, including the correspondence between vehicles and tasks, key preference degree and sorting basis.

[0092] To visually illustrate the matching differences of this invention under different modes, this embodiment selects vehicle coordinates and task elements at the same moment and performs matching operations in production pull mode, low-energy consumption mode, and comprehensive mode, respectively. Example vehicle and task coordinates, segment weights, process numbers, deadlines, and calculated S are provided. task With S vehicle The results are shown below. It should be understood that S in the table... task With S vehicle The example calculation results obtained according to the preference concept of this invention are used to reflect the sorting and selection logic.

[0093] (71) The results of the production pull mode are shown in Table 1. The minimum value of the task is 0.2489, and the maximum value is 0.8602. According to the rule of "the smaller the task, the higher the priority", the combination of tasks ranked higher reflects a strong preference for the task. The task of A3-SEG938 is 0.2489, which is the minimum value in this mode. This car has the highest priority for this task under the weight of the production pull strategy. The task of A4-SEG950 is 0.3873, and the task of A1-SEG245 is 0.3993, which are also in the lower range. Their request order will be significantly higher than the combination of tasks with larger tasks. Further combining the process number and the deadline, the production pull mode emphasizes "prioritizing subsequent processes and avoiding blockages caused by full yard load". In this table, the tasks with process number 5 include A2-SEG136, B1-SEG214, and C3-SEG817. Although A2-SEG136 has the highest process time and an earlier deadline, its Task=0.8602 is the highest value in the table, indicating that its empty driving distance during this window significantly reduces the vehicle side's willingness to request it. This demonstrates that the invention does not simply sort by "process / deadline time," but rather comprehensively weights empty driving distance, process time, and time difference. Looking at the Vehicle value on the task side, the highest Vehicle value in this mode is 0.8141 for A1-SEG245, followed by 0.7734 for C1-SEG641 and 0.7538 for C2-SEG666. This suggests that when multiple vehicles compete, the tasks corresponding to SEG245, SEG641, and SEG666 tend to choose these vehicles. This can be inferred from the vehicle selection criteria of "shorter distance or better cumulative mileage," thus achieving a secondary selection based on "efficiency and balance." It is worth noting that the Vehicle score for the A4-SEG950 is only 0.2784, which is a relatively low value in the table. This indicates that even if the vehicle side has a strong preference for the task, the task side may still reduce its vehicle selection priority due to factors such as the vehicle's cumulative mileage and location. This reflects the corrective effect of the two-way preference mechanism on local optima.

[0094] Table 1 S under Production Pull Mode task S vehicle Calculated values ​​and dynamic dispatch results

[0095]

[0096] (72) The results in the low-energy consumption mode are shown in Table 2. The minimum task value is 0.2723 and the maximum value is 0.7746. This mode emphasizes matching the nearest location to reduce empty runs and energy consumption. The combination with the smallest task value is A4-SEG950, which has the highest request priority; followed by C3-SEG817 (Stask=0.3691), and then A3-SEG938 (Stask=0.5032). Compared with the production pull mode, the task value of C3-SEG817 dropped significantly from 0.8302 to 0.3691, and its priority jumped from the lower to the higher position. This indicates that under the low-energy consumption weight, the empty run distance of this vehicle to this segment is more advantageous, thus overriding the difference caused by the process and the deadline. From the task perspective, the lowest energy consumption vehicle (Svehicle) in low-energy mode is 0.8432 for C1-SEG641, followed by 0.8006 for C2-SEG666 and 0.7740 for B1-SEG214. When vehicles generally initiate requests based on proximity, the task side can more easily select the vehicle with the best overall distance and accumulated mileage from multiple requests, thereby reducing the number of secondary rollbacks and reassignments and lowering the burden of edge recalculation.

[0097] Table 2 S under low energy consumption mode task With S vehicle Calculated value

[0098] (73) The results of the integrated mode are shown in Table 3. The minimum value of the task is 0.3611 and the maximum value is 0.7783; the minimum value of the vehicle is 0.2207 and the maximum value is 0.7497. The goal of the integrated mode is to balance production pull and low energy consumption, so its task distribution is between the two. From the minimum value of the task, the combination with the highest priority in the integrated mode is A4-SEG950 (Stask=0.3611), followed by A3-SEG938 (Stask=0.3845), and then B2-SEG258 (Stask=0.5163). Compared to the low-energy consumption mode, A4-SEG950 maintains a higher priority, but its task score increases from 0.2723 to 0.3611, reflecting that the comprehensive mode, while retaining the advantage of proximity, adds process and deadline factors, making the scoring more comprehensive rather than extremely biased towards distance. Compared to the production-pull mode, B2-SEG258's task score increases from 0.4312 to 0.5163, indicating that the comprehensive weight suppresses some cases of "excessive bias solely due to production pull". From the perspective of vehicles on the task side, A1-SEG245 has the highest vehicle score of 0.7497 in the comprehensive mode, indicating the strongest preference for this vehicle on the task side; B2-SEG258's vehicle score of 0.7082 is second; and C2-SEG666's vehicle score of 0.6445 is also relatively high. Compared to the production-driven mode, most vehicles showed a decrease in performance under the integrated mode. For example, B1-SEG214 decreased from 0.7102 to 0.5154, and C1-SEG641 decreased from 0.7734 to 0.6177. This indicates that the integrated mode emphasizes "balanced suppression" in vehicle selection on the task side, avoiding excessive favoritism of a few vehicles on the task side that leads to an imbalance in accumulated mileage and load. This aligns with the goal of "equal emphasis on task balance and time balance" in this invention.

[0099] Table 3 S under integrated mode task With S vehicle Calculated value

[0100]

[0101] Step (8) trains a parameter set based on historical and real-time data and designs reward and punishment rules so that the weights can be adaptively adjusted according to the working conditions to maintain the stability of the scheduling effect.

[0102] Because the tonnage ratio of segments, yard occupancy rate, and segment location distribution will drift during different construction cycles, keeping the weight coefficients in the preference calculation fixed for a long time may lead to a degradation in scheduling effectiveness. This embodiment introduces a dynamic intelligent parameter tuning mechanism, which constructs a parameter set {w} from the weight coefficients in the comprehensive mode. i}, for {w iA virtual scheduling simulation was performed to calculate the total travel distance DIS and the total transport time T. Multi-objective optimization was then used to simultaneously minimize both DIS and T to obtain the initial optimal parameter set {w}. i In actual scheduling and execution, the system collects historical transportation data per unit time, including task matching records, vehicle empty-run rate, vehicle load balancing, and task on-time completion rate. Excellent matching data with low empty-run rates and meeting on-time completion rates are selected for supervised learning, and a reward function is introduced to quantify the improvement effect before and after parameter tuning. The reward value adopts a benefit-minus-penalty form, with hard constraints imposed on on-time completion rate to ensure that mileage reduction is not sacrificed for delivery. Experience tuples are written into the training buffer, which uses a first-in-first-out strategy to retain the latest 5000 samples. When the system detects that the segmented tonnage ratio or yard occupancy rate distribution drift exceeds a threshold, the training frequency is increased and the window period is shortened to allow the model to adapt to environmental changes more quickly. Training uses an Actor-Critic or PPO framework consisting of a policy network (Actor) and a value network (Critic). The input state vector s includes at least task features, vehicle availability, tonnage grouping, yard occupancy rate, path congestion, and historical execution feedback. The output is a dynamic weight parameter set {w}. i After training, the Actor generates multiple candidate {w} sets in the current state. i The system performs Pareto front screening on candidate solutions based on their performance across multiple objective indicators, including transportation efficiency, energy consumption, space utilization, and on-time delivery rate, to obtain the optimal solution that satisfies the constraints. i} is sent to the edge calculator, so that {w i It applies real-time to the calculation of vehicle and task preferences, forming a closed-loop adaptive operation of "environmental change, parameter adjustment, scheduling, and feedback".

[0103] Step (9) handles dynamic events such as road closures, vehicle malfunctions, and task insertions, and quickly restores feasible scheduling through status reporting, route replanning, and task reassignment.

[0104] (91) If road construction at the storage yard results in temporary road closures or requires one-way traffic on a certain section, the dispatcher edits the road topology through the cloud map management module, updates the closure rules, and synchronizes them to the edge path planning service. After receiving the update, the edge side immediately triggers path replanning for vehicles that have not yet entered the closure section, recalculating feasible paths from their current location to the pickup and delivery points. For vehicles that have entered the vicinity of the closure section, the edge side combines the obstacle and congestion status reported by the terminal to select a detour or waiting strategy, and sends the updated instructions to the terminal for execution. This process does not require centralized global recalculation; the edge within the group can complete a rapid response, ensuring real-time performance in dynamic scenarios.

[0105] (92) If a vehicle malfunctions or is taken out of service for maintenance, the terminal immediately reports the abnormal status and stops issuing task requests. The edge side removes the vehicle from the executable set. If the vehicle has been matched but the task has not been completed, the edge side triggers task reallocation within the group, adds the task back to the remaining task list, and issues it to other vehicles. If the task has an urgent attribute or a close deadline, the edge side increases the weight of the deadline item when calculating the preference to speed up the matching. After the malfunctioning vehicle is restored, the dispatcher clicks the task request on the mobile terminal to reinstate the vehicle into the scheduling loop.

[0106] (93) After the insertion task enters the cloud, the cloud automatically groups it according to the weight rule and sends it to the corresponding edge. The edge adds the insertion task to the remaining task list and performs incremental matching based on the current vehicle request situation. If the insertion task causes a significant increase in the load within the group, the edge reports to the cloud, and the cloud determines whether to trigger inter-group vehicle dispatch based on the inter-group difference rate R. If the yard relocation makes the originally planned pickup point unreachable, the terminal perception module identifies the congestion or obstacles in the work area and reports it. The edge updates the site status and recalculates the path. If necessary, the task is temporarily suspended and other executable tasks are given priority. Execution is resumed after the relocation is completed, thereby reducing unnecessary waiting.

[0107] This invention provides a closed-loop process from task import, resource initialization, grouping, intra-group matching, path planning, balance judgment, dynamic parameter adjustment to dynamic event handling and statistical analysis. It enables real-time scheduling and adaptive optimization in complex dynamic environments such as road congestion, task insertion, uncertain vehicle completion time, and yard relocation. It can effectively improve the efficiency of ship segment transfer and reduce empty travel and energy consumption.

[0108] The ship section transfer dynamic intelligent scheduling system described in this invention can complete the collaborative deployment of the cloud layer, edge layer, and terminal layer, establish access channels for data such as task, vehicle, road, and yard status, and unify communication and permission configuration, as detailed below:

[0109] A dynamic scheduling system is constructed that coordinates the cloud layer, edge layer, and terminal layer to achieve unified issuance of task grouping and matching parameters, real-time matching of vehicles and tasks within a group, vehicle route planning, task balance judgment among vehicles, rapid issuance of transportation instructions, vehicle status data feedback, mileage and energy consumption statistics and report analysis, thereby improving the efficiency of ship segment transfer and reducing empty runs and energy consumption.

[0110] The system hardware architecture of this invention includes a cloud server at the cloud layer, multiple edge calculators at the edge layer, and a flatbed truck and vehicle controller at the terminal layer. The cloud server at the cloud layer is used to handle functions such as task data aggregation, scheduling parameter generation and issuance, global visualization, and statistical analysis. Edge calculators are set up at the edge layer for different tonnage groups. Each edge calculator receives task grouping and matching parameters from the cloud server, performs edge-side calculations based on the real-time status of the vehicles in its group, issues transportation instructions to the vehicles in its group, and simultaneously reports the scheduling execution process data and statistical results to the cloud server. The terminal layer comprises multiple flatbed trucks, grouped as follows: Group A: vehicles A1, A2...An; Group B: vehicles B1, B2...Bn; Group C: vehicles C1, C2...Cn. Each flatbed truck is equipped with an onboard controller, which includes at least a positioning module, a sensing module, a communication module, and an execution control module. The positioning module includes GNSS and IMU to provide vehicle position, attitude, and motion information; the sensing module includes LiDAR and cameras to acquire information about the vehicle's surrounding environment, obstacles, and operating status; the communication module includes 5G / WIFI to enable bidirectional data communication between the vehicle and the edge layer; and the execution control module includes a driver and a hydraulic controller to receive transportation instructions and drive the vehicle to perform actions such as driving, steering, braking, and hydraulic actuation. The cloud server issues task grouping and matching parameters to the edge calculator, the edge calculator issues transportation instructions to the flatbed trucks, the flatbed trucks report status data to the edge calculator, and the edge calculator reports data to the cloud server, thus forming a closed-loop link of parameter issuance, instruction execution, status feedback, and data reporting.

[0111] The software and services of this invention adopt a layered microservice architecture. The upper layer interfaces with a production management system, which provides production plans, task requirements, and resource constraints related to segmented transfer and inputs this information into the cloud layer. The cloud layer includes a task management service for creating, publishing, and maintaining segmented transfer tasks and their status transitions; a vehicle resource management service for maintaining vehicle resource information, vehicle grouping information, and availability; a map editing service for maintaining site road topology, work areas, key nodes, and traffic restriction rules, providing a basic map for edge layer path planning; an AI parameter tuning model training service for training and adjusting matching parameters based on historical and real-time data, and distributing the trained matching parameters to the edge layer; and a report analysis service for statistically analyzing indicators such as empty runs, energy consumption, mileage, and overtime and outputting reports. The system includes: a vehicle dynamics viewing service to display real-time vehicle status; a segmented location viewing service to display segmented locations and their association with tasks; a site status viewing service to display road traffic conditions, congestion status, or work area status; a task progress viewing service to display task execution progress and key node status; a vehicle trajectory viewing service to display historical vehicle trajectories and operation replays; a group balance judgment service to assess the balance of vehicle groups and provide a basis for resource optimization; and an inter-group vehicle allocation service to allocate vehicles between different groups and synchronize the allocation results to the edge layer for execution.

[0112] Furthermore, the edge layer sets up a vehicle-task matching service within a group. This service receives task grouping and matching parameters from the cloud and dynamically matches vehicles and tasks based on the real-time status of vehicles in the group and task requirements, resulting in an executable allocation. It also sets up a vehicle path planning service, which calculates the driving path from the current location to the pickup and delivery points based on the road topology generated by cloud map editing and the site and vehicle status information reported by the terminal, and dynamically replans the path when the site status changes. A vehicle task balancing judgment service is set up to assess the balance of task load, waiting time, and execution time among vehicles in the group, triggering task reassignment or path recalculation when imbalances occur. A transportation mileage statistics service is set up to count the mileage of vehicles performing effective transportation processes. An empty-run mileage statistics service is set up to count the mileage of vehicles traveling without load. An energy consumption statistics service is set up to count the energy consumption data during vehicle execution. The edge layer converts the output of these services into single-vehicle transportation instructions and sends them to the terminal layer, while simultaneously summarizing the execution process data and statistical results and reporting them to the cloud layer.

[0113] Furthermore, the terminal layer is equipped with vehicle positioning capabilities, with the positioning module outputting vehicle position and attitude information; state perception capabilities, with LiDAR and cameras outputting environmental and operational status information; vehicle navigation capabilities, executing navigation based on paths and instructions issued by the edge layer; vehicle control capabilities, with the actuators and hydraulic controllers of the execution control module completing vehicle movement and hydraulic execution; and task request capabilities, initiating task requests or status updates to the edge layer via the communication module when the vehicle is idle, has completed a task, or encounters an anomaly. The terminal layer continuously reports vehicle status data to the edge layer, and the edge layer performs real-time updates based on the status data for intra-group matching, path planning, and balance judgment, and then reissues the updated transportation instructions to achieve a dynamic scheduling closed loop.

Claims

1. A dynamic intelligent scheduling method for ship section transfer, characterized in that, Includes the following steps: (1) Determine the segmented transportation task groups and vehicle groups based on the production tasks and load tonnage; (2) By the inter-group difference rate, determine whether the tasks between groups are balanced. If the tasks between groups are balanced, the vehicle sends a task request to its group; if the tasks between groups are not balanced, the vehicle sends a task request to other groups. (3) Calculate the vehicle’s preference for candidate segmented tasks based on the empty mileage of the vehicle to the segmented storage location in the segmented transportation task, the current process of the segmented storage location in the segmented transportation task, and the difference between the estimated delivery time of the segmented transportation task and the current time. (4) Based on the yard occupancy rate, the production pull mode, low energy consumption mode or comprehensive scheduling mode are adaptively switched. Based on the empty mileage of the vehicle to the segmented storage location in the segmented transportation task and the cumulative mileage of the vehicle, the vehicle preference of the candidate task in the current mode is calculated. (5) Perform bidirectional dynamic matching based on the vehicle's preference for candidate segment tasks and the candidate tasks' preference for the vehicle under the current mode; (6) The weights in the calculation of the vehicle's preference for candidate segment tasks and the candidate task's preference for the vehicle constitute a dynamic weight parameter set; Based on the total mileage and total transportation time of the vehicle, a deep learning model is established to perform multi-objective optimization with the goal of minimizing the total mileage and total transportation time, and supervised learning is carried out. (7) Compare the results before and after parameter tuning according to the window period, determine the distance improvement rate and time improvement rate, and construct the excitation function; (8) The Actor-Critic or PPO framework, consisting of a policy network Actor and a value network Critic, is adopted. The input is a state vector and the output is the optimal dynamic weight parameter set. The state vector includes task characteristics, vehicle availability, tonnage grouping, yard occupancy rate, path congestion degree and historical execution feedback. (9) Based on the optimal dynamic weight parameter set, recalculate the vehicle's preference for candidate segment tasks and the candidate task's preference for the vehicle under the current mode, and perform bidirectional dynamic matching again.

2. The dynamic intelligent scheduling method for ship segment transfer according to claim 1, characterized in that, In step (2), the inter-group difference rate R is in, Let N be the task load coefficient of the vehicles in the i-th group. i Let n be the number of unfinished tasks in the i-th group. i The number of currently executable vehicles in group i; The closer R is to 0, the more balanced the tasks are. If R > 0.5, the tasks are unbalanced between groups.

3. The dynamic intelligent scheduling method for ship segment transfer according to claim 1, characterized in that, In step (3), the vehicle's preference for candidate segment tasks. for in, Let this be the empty mileage of the vehicle to the segmented storage location in the i-th task. Let be the current process number of the segment storage location in the i-th task. Let be the difference between the estimated delivery time of the i-th task and the current time. , and For weights.

4. The dynamic intelligent scheduling method for ship segment transfer according to claim 1, characterized in that, In step (4), if the yard occupancy rate is ≥95%, the production pull mode is adopted, and the subsequent processes are transported in segments with priority; if the yard occupancy rate is <60%, the low energy consumption mode is adopted, following the principle of matching nearby; if the yard occupancy rate is greater than 60% and less than 95%, the comprehensive mode is adopted.

5. The dynamic intelligent scheduling method for ship section transfer according to claim 1, characterized in that, In step (4), the candidate task's vehicle preference is: in, Let this be the empty mileage of the vehicle to the segmented storage location in the i-th task. This refers to the vehicle's cumulative mileage. and For weights.

6. The dynamic intelligent scheduling method for ship segment transfer according to claim 1, characterized in that, In step (7), the excitation function for Among them, α, β, Dynamically configured in the cloud based on management preferences. For distance improvement rate, The total driving distance of the baseline scheme within a window period, To use the current weight Total driving distance during the same window period For time improvement rate, This represents the total completion time of the baseline plan within the window period. This represents the total completion time of the window period under the current weight.

7. A dynamic intelligent scheduling system for ship section transfer, characterized in that, include: The cloud layer is used to handle task data aggregation, scheduling parameter generation and issuance, global visualization, and statistical analysis. The edge layer is used to set up edge calculators according to the vehicle tonnage group. Each edge calculator receives the task grouping and matching parameters issued by the cloud server of the cloud layer, performs edge-side calculations in combination with the real-time status of the vehicles in its group, issues transportation instructions to the vehicles in its group, and reports the scheduling execution process data and statistical results to the cloud server. The terminal layer includes several vehicles. The vehicle's onboard controller includes a positioning module, a sensing module, a communication module, and an execution control module. The positioning module provides vehicle position, attitude, and motion information; the sensing module acquires information about the vehicle's surrounding environment, obstacles, and operating status; the communication module enables bidirectional data communication between the vehicle and the edge layer; and the execution control module receives transportation instructions and drives the vehicle to complete driving, steering, braking, and hydraulic actuation actions. The cloud server in the cloud layer issues task grouping and matching parameters to the edge calculator, the edge calculator issues transportation instructions to the vehicle, the vehicle reports status data to the edge calculator, and the edge calculator reports data to the cloud server, thus forming a closed-loop link of parameter issuance, instruction execution, status feedback and data reporting.

8. The ship section transfer dynamic intelligent scheduling system according to claim 7, characterized in that, The cloud layer sets up a task management service for creating, publishing, and maintaining segmented transfer tasks and their status flow; The cloud layer sets up a vehicle resource management service to maintain vehicle resource information, vehicle grouping information, and availability status; The cloud layer sets up a map editing service to maintain the site road topology, work areas, key nodes, and no-entry or restricted-entry rules, providing a basic map for edge layer path planning; The cloud layer sets up an AI parameter tuning model training service to train and adjust the matching parameters based on historical and real-time data, and then sends the trained matching parameters to the edge layer. The cloud layer is configured with a report analysis service to perform statistical analysis on empty runs, energy consumption, mileage and overtime indicators and output reports. The cloud layer is configured with a vehicle dynamic viewing service to display the real-time status of vehicles; and a segmented location viewing service is configured to display the segmented locations and their association information with the task. The cloud layer is set up with a site status viewing service to display road traffic status, congestion status, or work area status; The cloud layer is configured with a task progress viewing service to display the task execution progress and the status of key nodes; A vehicle trajectory viewing service is set up in the cloud layer to display the vehicle's historical trajectory and operation playback; A cloud-based system is configured to perform a balance assessment service, which is used to assess the balance of vehicle groups and provide a basis for resource optimization. The cloud layer sets up an inter-group vehicle dispatch service to dispatch vehicles between different groups and synchronize the dispatch results to the edge layer for execution.

9. The ship section transfer dynamic intelligent scheduling system according to claim 7, characterized in that, The edge layer sets up a vehicle and task matching service within the group. This service receives task grouping and matching parameters from the cloud and dynamically matches vehicles and tasks based on the real-time status of vehicles in the group and task requirements, resulting in an executable allocation result. The edge layer sets up a vehicle route planning service, which is used to calculate the driving route of the vehicle from the current location to the pickup point and delivery point based on the road topology formed by cloud map editing and the site and vehicle status information reported by the terminal, and to perform dynamic replanning when the site status changes. The edge layer sets up a task balance judgment service between vehicles to judge the balance of task load, waiting time and execution time of vehicles in the group, and triggers task redistribution or path recalculation when there is an imbalance. The edge layer is configured with a transportation mileage statistics service to track the mileage of vehicles during effective transportation processes. An empty-run mileage statistics service is set up at the edge layer to count the mileage of vehicles without load; An energy consumption statistics service is set up at the edge layer to collect energy consumption data during vehicle operation. The edge layer converts the output of the above services into transportation instructions for individual vehicles and sends them to the terminal layer. It also summarizes the execution process data and statistical results and reports them to the cloud layer.

10. The ship section transfer dynamic intelligent scheduling system according to claim 7, characterized in that, The terminal layer is equipped with a vehicle positioning module, which outputs vehicle position and attitude information. The terminal layer is equipped with a status perception module, which outputs environmental and operational status information from LiDAR and cameras; The terminal layer is equipped with a vehicle navigation module, which performs navigation based on the path and instructions sent down from the edge layer; The terminal layer is equipped with a vehicle control module, which uses the actuator and hydraulic controller of the execution control module to complete the vehicle movement and hydraulic execution. The terminal layer is equipped with a task request module. When the vehicle is idle, completes a task, or encounters an abnormality, it sends a task request or status update to the edge layer through the communication module. The terminal layer reports vehicle status data to the edge layer. The edge layer performs real-time updates based on the status data for intra-group matching, route planning, and task balancing, and then reissues the updated transportation instructions to achieve a dynamic scheduling closed loop.

Images