A method and system for collaborative optimization of dock resources based on constraint programming and adaptive large neighborhood search

Abstract

This invention discloses a method and system for collaborative optimization of dock resources based on constraint programming and adaptive large neighborhood search. The method first constructs a first-stage constraint programming model based on the loading network, gantry crane resources, and site data. This model solves the model to determine the time schedule for loading operations and to define "time windows" for entry and exit of segments. The method then constructs a second-stage adaptive large neighborhood search model, using the "time windows" as constraints, iteratively searching for the physical layout positions of segments in a continuous two-dimensional space. During the iteration process, if a conflict-free position cannot be found within the time window, the time is allowed to be postponed and penalized, achieving spatiotemporal linkage. This invention achieves collaborative optimization of dock time scheduling and spatial layout through a decoupling approach of using CP-SAT to handle rigid time constraints and ALNS to handle continuous spatial constraints. This ensures the physical feasibility of the solution and improves the response speed of dynamic replanning.

Description

Technical Field

[0001] This invention relates to the field of intelligent ship manufacturing and production management technology. Specifically, it relates to a method and system for integrated collaborative optimization of the loading plan (time dimension), site layout (spatial dimension), and gantry crane / high crane resources (equipment dimension) in the digital dock production process using a hybrid intelligent optimization algorithm. Background Technology

[0002] Ship assembly and loading are among the most complex stages in shipbuilding, and their production efficiency heavily relies on the scientific scheduling of core resources such as dock space, gantry cranes, and overhead cranes. In traditional dock production management, the preparation of loading plans, the layout planning of assembly areas, and the scheduling of crane operations are often separate and sequential processes. Schedulers and managers rely heavily on their personal work experience for scheduling. For example, the planning department first formulates the loading plan (time sequence), then the site department finds available storage areas based on the plan (spatial layout), and finally the lifting team arranges crane operations based on the results of the first two steps.

[0003] This experience-dependent and fragmented scheduling model has the following significant drawbacks when dealing with the mass construction of modern ships (such as the parallel production of multiple ships, the coordinated hoisting of ultra-large sections by two machines, and the strict constraints of batch launching schedules): 1) Unable to achieve collaborative optimization: Due to the fragmentation of time planning, spatial layout and resource scheduling, each solution is only locally optimal and cannot achieve the global optimization goal of "shortest loading cycle, highest site utilization, and least crane waiting time".

[0004] 2) Poor optimization effect and frequent resource conflicts: It is difficult to exhaust all feasible solutions under complex constraints based on experience alone, resulting in uneven resource allocation, which can easily lead to site conflicts, logistics channel blockages, or long periods of empty running and waiting for gantry cranes, resulting in poor optimization effect.

[0005] 3) High difficulty in replanning and poor dynamic adjustment capability: The dock site is full of uncertainties. When encountering delays in section entry, weather changes (such as excessive wind speed causing crane shutdown), or sudden equipment failures, existing technical solutions cannot respond quickly. Manually readjusting the plan is time-consuming and labor-intensive, and it is difficult to assess the impact of the adjustment on the overall situation, resulting in a serious disconnect between the plan and actual execution, i.e., high difficulty in replanning.

[0006] Therefore, there is an urgent need for a collaborative optimization method that can simultaneously take into account the constraints of the loading process, site space, and crane resources, in order to solve the technical pain points of existing technologies that cannot be optimized simultaneously, rely on experience, and are difficult to dynamically adjust. Summary of the Invention

[0007] To address the shortcomings of existing dock scheduling schemes, which rely on manual experience and are fragmented in terms of time planning, spatial layout, and resource scheduling, the present invention aims to provide a dock resource collaborative optimization method. This method can achieve integrated and automated solution of loading plans (time), site layout (space), and crane resources (equipment), and has the ability to quickly perform dynamic replanning when disturbances occur on the production site (such as segment delays or equipment failures).

[0008] To achieve the above objectives, the present invention employs the following technical solution: The first aspect of this invention is to provide a method for collaborative optimization of dock resources based on constraint programming and adaptive large neighborhood search, the method comprising the following steps: 1) Time and resource scheduling: Based on the dock's loading network data, gantry crane resource data, and site resource data, a first-stage constraint programming model is constructed and solved; to generate a time schedule containing various loading operations, and to determine the entry time window and exit time window for each segment that needs to be stored in the yard; 2) Spatial layout optimization: Using the entry time window and exit time window as constraints, a second-stage adaptive large neighborhood search model is constructed; 3) Iterative solution and linkage: Through iterative search of the adaptive large neighborhood search model algorithm, the physical layout position of each segment is found in the continuous two-dimensional space of the yard; during the iteration process, if a segment cannot find a conflict-free physical layout position within the entry time window, the entry time of the segment is allowed to be postponed, and the postponement amount is included as a penalty in the cost function of the adaptive large neighborhood search model. 4) Scheme output: When the adaptive large neighborhood search model algorithm converges or reaches the termination condition, the output is a yard layout scheme containing the precise physical coordinates of each segment, and combined with the time schedule of the loading operation, a final time-space collaborative optimization scheme for dock resources is formed.

[0009] Furthermore, the first-stage constrained programming model in step 1) includes at least one or more of the following constraints: 2-1) Ensure that the loading operation meets the constraints of the pre-process relationships; 2-2) Ensure that the occupancy of gantry crane resources at any given time does not exceed the cumulative constraint of its available quantity limit; 2-3) Ensure that the total area of ​​the yard occupied by the segments does not exceed the site capacity limit, and exclude the yard accumulation constraints of specific unavailable areas.

[0010] Further, in step 1), the first-stage constrained programming model is optimized using a bi-objective function, which includes: 3-1) Minimize the overall ship assembly completion cycle; 3-2) Minimize the time difference between the completion milestones of multiple half-ships.

[0011] Further, in step 3), the iterative search of the adaptive large neighborhood search model algorithm is performed by adaptively selecting destruction and repair operators; the destruction operator is selected from at least one of conflict focusing operator, time bucket operator and region operator; the repair operator is selected from at least one of space-priority repair operator, spatiotemporal collaborative repair operator and region exchange operator.

[0012] Furthermore, the execution logic of the conflict focusing operator includes: 5-1) Evaluate the number of geometric overlap conflicts in each segment of the current layout; 5-2) Prioritize the segment with the highest number of overlapping conflicts as the target for destruction; 5-3) When there are few conflicts in the layout, it degenerates into prioritizing the selection of objects to be destroyed based on area.

[0013] Furthermore, the cost function of the adaptive large neighborhood search model described in step 3) includes one or more of the following penalty terms: 6-1) Overlap penalty term for penalizing the geometric overlap area between segments; 6-2) Penalty items for exceeding the boundary distance of segments; 6-3) Delay penalty items that penalize the amount of time delay in segmented entry; 6-4) Aggregation penalty term that penalizes the spatial dispersion of segments with the same ship number.

[0014] Furthermore, the method also includes a parallel evaluation acceleration step: during a single iteration of step 3) of claim 1, multiple candidate layout solutions are generated in parallel, and the cost function of each candidate solution is calculated using a process pool, from which the candidate solution with the optimal cost is selected for subsequent iterations.

[0015] Furthermore, the method also includes a dynamic replanning step, specifically: in response to a disturbance event occurring at the production site, using the current actual production progress as a new initial condition, automatically triggering and re-executing the steps of claim 1 to generate an updated dock resource collaborative optimization scheme.

[0016] A second aspect of the present invention is to provide a dock resource collaborative optimization system based on constraint programming and adaptive large neighborhood search, the system comprising one or more modules configured to perform the following operations: 1) Time and resource scheduling module: configured to execute step 1) of claim 1, based on the network data, gantry crane resource data and site resource data, to construct and solve the first-stage constraint programming model to generate time scheduling and segmented entry and exit time windows; 2) Spatial layout optimization module: configured to perform steps 2) and 3) of claim 1, receive a time window, construct and iteratively solve the second-stage adaptive large neighborhood search model, and process the delay penalty during the iteration process to find the physical layout location; 3) Solution output module: configured to execute step 4 of claim 1, and at the end of the iteration, combine the time schedule and physical layout position to output the final time-space co-optimization solution.

[0017] A third aspect of the present invention is to provide a computer device including a processor and a memory, the memory storing a computer program that, when executed by the processor, implements the method as described in any one of claims 1 to 8.

[0018] A fourth aspect of the present invention is to provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method as described in any one of claims 1 to 8.

[0019] Compared with the prior art, the present invention has the following beneficial effects: 1) Achieved coordinated optimization of time and space: By decoupling the complex dock scheduling problem into two stages, leveraging the advantages of CP-SAT in handling logical constraints and ALNS in handling continuous spatial optimization, and through the "time window" and "delay penalty" mechanism to achieve flexible linkage, the fundamental problem of global optimization that cannot be achieved in the background technology is solved.

[0020] 2) Ensured the physical feasibility of the solution: Through collision detection and overlap penalty based on vectorized matrix operations in the ALNS stage, and strict constraints on resources such as the gantry crane dual-machine collaboration in the CP-SAT stage, it was ensured that the final output loading plan and yard layout scheme were physically executable and conflict-free.

[0021] 3) Improved response speed of dynamic replanning: When unexpected situations such as segmented delays or equipment failures occur on site, this method does not need to rely on manual rescheduling as in the background technology. Instead, it can automatically trigger the rescheduling process and quickly recalculate a new solution with the least global disturbance within minutes, which significantly improves the rigid synchronization capability of the plan.

[0022] 4) Improved efficiency in solving complex problems: By introducing a parallel evaluation mechanism, multiple candidate solutions can be evaluated simultaneously using multi-core CPU resources, which significantly shortens the single-round time and overall convergence time of the ALNS algorithm, meeting the performance requirements of production scheduling. Attached Figure Description

[0023] Figure 1 This is an overall flowchart of the dock resource collaborative optimization method provided in the embodiments of the present invention. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings.

[0025] Example 1

[0026] See Figure 1 This embodiment provides a dock resource collaborative optimization method based on constrained programming (CP-SAT) and adaptive large neighborhood search (ALNS). The method runs on one or more computer servers, each containing a processor and memory, with the memory storing instructions for executing the method.

[0027] S1, Data Loading and First-Stage Model Building: The core of this step is to map and load the business data stored in the database into a standardized solver input data structure required for optimizing the solver.

[0028] S1-1) Data Source: In this embodiment, the data source is the relevant data tables in an enterprise database (such as MySQL). These data (such as segments, total segments, and process networks) can be idempotently imported into the database from source data files such as Excel using a data import script.

[0029] S1-2) Data Mapping: Data loading is performed through the data loading function. This process converts the fields in the data table into canonical objects that the solver can recognize. a) Segmentation specification object: Define the geometric dimensions (length, width) and weight of the segment; b) Process specification objects: Define the duration of the process, its prerequisites, and key resource requirements. For example, based on the process type or configuration, mark the number of gantry crane resources required (e.g., 1 or 2 units). c) Yard specification objects: Define the geometric boundaries of the yard area (such as the northeast yard, southwest yard); d) Crane specifications: Define the quantity and capacity of gantry cranes / high cranes.

[0030] S1-3) Model Construction: Instantiate the loaded and mapped data into a first-stage constrained programming model instance.

[0031] S2, Perform the first stage of constraint programming solution (time window determination): This step uses the CP-SAT solver to handle the time-bound, strongly constrained, and resource scheduling problems.

[0032] S2-1) Add constraints: In the first-stage model, construct the following core constraints: a) Process constraints: Strictly adhere to the prerequisite relationships defined in the process specification object; b) Gantry crane resource constraints: Use cumulative constraint functions to ensure that operations marked as requiring dual-machine collaboration must occupy the resources of two cranes simultaneously, and the total resource usage does not exceed the upper limit of available cranes; c) Yard Occupancy Constraint: Also using a cumulative constraint function, the occupancy of each segment (based on its area) is treated as a resource that accumulates over time. This constraint is only used to ensure that the total occupied area does not exceed the site capacity limit and excludes certain unusable areas (such as the dock area).

[0033] S2-2) Setting a dual objective function: The objective function in this embodiment aims to optimize the batch production rhythm.

[0034] a) Primary objective: Minimize the overall ship completion cycle; b) Secondary objective: Minimize the time difference between the completion milestones of multiple half-ships to ensure the synchronization of batch launches.

[0035] S2-3) Solution and Output: The CP-SAT solver searches for the optimal solution within the set time limit. After the solution is completed, two key outputs are generated: a) Loading schedule: A schedule file containing the start and end times of all loading operations; b) Yard Occupancy Plan: The core output of this phase. This plan is a data structure that does not contain (X,Y) coordinates. Instead, it maps and generates a discrete entry and exit time window, as well as a list of candidate partitions that are allowed to be stacked, for each segment that needs to be stacked.

[0036] S3. Perform the second stage of spatial layout optimization (coordinate determination): This step uses the ALNS algorithm to solve the physical layout and collision detection problems of segments in the "continuous two-dimensional space" under the "time window" constraint determined in S2.

[0037] S3-1) Initial Solution Generation (Optional): An initial solution can be generated before the iteration begins. For example, the segments are sorted according to the entry time window generated in S2, and the space-first repair operator is called to place them one by one to obtain a feasible initial layout.

[0038] S3-2) ALNS Iteration: The algorithm runs within a simulated annealing framework. In each iteration, a combination of "destroy" and "repair" operators is selected through an adaptive weighting mechanism.

[0039] S3-3) Destroy operator: Used to remove a set of segments from the current layout solution and return them to the placement pool. This embodiment may include: a) Conflict-focused destruction: Assess the number of geometric overlaps of each segment, sort them in descending order of conflict count, and prioritize removing the segments with the most severe overlaps; if conflicts are few, degenerate to area-based priority. b) Time Bucket Destruction: Randomly select a time window and remove all segments that remain on the field during that time period; c) Area Destruction: Calculate the "pressure" of each yard partition based on the number of conflicts and area aggregation, randomly select high-pressure partitions and remove some segments from them.

[0040] S3-4) Repair operator: used to re-insert segments from the placement pool into the layout. This embodiment may include: a) Space-priority repair: Insert in descending order of segment area, traverse candidate partitions using a Skyline-like scanning method to find feasible placement locations, and allow slight delays if necessary; b) Spatiotemporal collaborative repair: Prioritize placement within the original time window given by S2; if this fails, then within the allowed delay range, locally rearrange the segment and its temporally overlapping neighboring segments. c) Area swap repair: When a segment cannot be placed into the original candidate area, attempt to initiate an area swap in the high-load partition, and rearrange the area allocation of the two segments at the same time.

[0041] S3-5) Cost Function and Spatiotemporal Linkage: ALNS evaluates the cost function of new solutions. This function mainly consists of a penalty term: a) Overlap penalty: Penalize the geometric overlap area between any two segments in the storage yard; b) Boundary violation penalty: A penalty is imposed for the distance a segment exceeds the boundary of its location; c) Delay Penalty: This is crucial for achieving spatiotemporal linkage. When a repair operator (such as spatial priority repair) cannot find a non-overlapping location within the "time window" given by S2, it is allowed to postpone the entry time of that segment for a set maximum delay period (e.g., 24 hours). This delay in hours will be included as a penalty in the cost function. In a preferred embodiment, this cost function... F It can be defined as the weighted sum of all penalties, for example: F = W1*P overlap + W2 * P out of bounds + W3 * Pshift +W4 *P cluster , where P overlap For the overlap penalty, P out of bounds As a punishment for crossing the boundary, P shift To postpone punishment, P cluster To gather punishment, and W 1， W 2， W3 and W4 are their respective weighting coefficients; d) Clustering Penalty (Optional): To encourage close arrangement of sections within the same vessel, an additional "clustering within the same vessel" penalty can be applied. The penalty value is accumulated based on the distance of the center point of sections with the same vessel number from the average position of the group.

[0042] S3-6) Spatial search optimization: To improve the efficiency of layout and collision detection, this embodiment adopts a regional snapshot caching mechanism to cache the coordinates of active segments within a specific time window, and combines vectorized matrix operations to batch determine whether candidate positions overlap with already placed segments, thus avoiding inefficient loop judgments.

[0043] S4. Output the collaborative optimization solution: When the ALNS iteration reaches the termination condition (such as the set number of iterations or time limit), stop the iteration and output the optimal solution with the lowest current cost.

[0044] S4-1) Job scheduling: The output is a job scheduling file, the content of which comes from the CP-SAT solution results of S2.

[0045] S4-2) Yard Layout Scheme: The output is a yard layout file. This file contains the precise physical coordinates of each segment in the yard, whether it has been rotated, and the final determined entry / exit time (including the delay amount).

[0046] Example 2

[0047] In S3 of Example 1, to further shorten the solution time of the ALNS phase, parallel evaluation acceleration is introduced. Specifically, after the "destroy-repair" step in each iteration of ALNS, instead of generating a single candidate solution, multiple (e.g., 4) "destroy-repair" operations are performed in parallel to generate multiple candidate solutions. Subsequently, a process pool is used to distribute the cost evaluation tasks (mainly collision detection) of these candidate solutions to multiple CPU cores for parallel computation. After the main process collects all evaluation results, it selects the candidate solution with the lowest cost and submits it to the acceptance criteria of simulated annealing for judgment. This method can significantly reduce the time consumed in a single iteration. To ensure compatibility on operating systems such as Windows, the entry script that calls this parallel mechanism should be placed within a protected block of the main program.

[0048] Example 3

[0049] Based on Example 1 or 2, a dynamic replanning process is added. When disturbances occur on-site (such as segmented delays or equipment failures), the scheduler can trigger rescheduling through the system. The system will use the current actual production progress (e.g., completed processes, occupied space) as new initial conditions and re-execute the complete process from S1 to S4. Due to the high solution speed of the hybrid algorithm of CP-SAT and ALNS, the system can quickly generate a new, globally optimal collaborative solution that takes into account on-site disturbances, enabling flexible responses to unexpected situations.

[0050] Example 4

[0051] This embodiment provides a dock resource collaborative optimization based on constraint programming and adaptive large neighborhood search to implement the above method. The system includes: 1) Time and resource scheduling module: It is configured to execute S1, which, based on the network data, gantry crane resource data and site resource data, constructs and solves the first-stage constraint programming model to generate time scheduling and segmented entry and exit time windows; 2) Spatial layout optimization module: It is configured to execute S2 and S3, receive time windows, build and iteratively solve the second-stage adaptive large neighborhood search model, and handle the delay penalty during the iteration process in order to find the physical layout location; 3) Solution output module: Configured to execute S4, which combines time scheduling and physical layout location to output the final time-space co-optimization solution when the iteration ends.

Claims

1. A method for collaborative optimization of dock resources based on constraint programming and adaptive large neighborhood search, characterized in that, The method includes the following steps: 1) Time and resource scheduling: Based on the dock's loading network data, gantry crane resource data, and site resource data, a first-stage constraint programming model is constructed and solved; to generate a time schedule containing various loading operations, and to determine the entry time window and exit time window for each segment that needs to be stored in the yard; 2) Spatial layout optimization: Using the entry time window and exit time window as constraints, a second-stage adaptive large neighborhood search model is constructed; 3) Iterative solution and linkage: Through the iterative search of the adaptive large neighborhood search model algorithm, the physical layout location of each segment is found in the continuous two-dimensional space of the yard; during the iteration process, if a segment cannot find a conflict-free physical layout location within the entry time window, the entry time of the segment is allowed to be postponed, and the number of hours of postponement is included as a penalty in the cost function of the adaptive large neighborhood search model. 4) Scheme output: When the adaptive large neighborhood search model algorithm converges or reaches the termination condition, the output is a yard layout scheme containing the precise physical coordinates of each segment, and combined with the time schedule of the loading operation, a final time-space collaborative optimization scheme for dock resources is formed.

2. The dock resource collaborative optimization method based on constraint programming and adaptive large neighborhood search as described in claim 1, characterized in that, Step 1) The first-stage constrained programming model includes at least one or more of the following constraints: 2-1) Ensure that the loading operation meets the constraints of the pre-process relationships; 2-2) Ensure that the occupancy of gantry crane resources at any given time does not exceed the cumulative constraint of its available quantity limit; 2-3) Ensure that the total area of ​​the yard occupied by the segments does not exceed the site capacity limit, and exclude the yard accumulation constraints of specific unavailable areas.

3. A dock resource collaborative optimization method based on constraint programming and adaptive large neighborhood search as described in claim 1 or 2, characterized in that, In step 1), the first-stage constrained programming model is optimized using a bi-objective function, which includes: 3-1) Minimize the overall ship assembly completion cycle; 3-2) Minimize the time difference between the completion milestones of multiple half-ships.

4. The dock resource collaborative optimization method based on constraint programming and adaptive large neighborhood search as described in claim 1, characterized in that, In step 3), the iterative search of the adaptive large neighborhood search model algorithm is performed by adaptively selecting destruction and repair operators; the destruction operator is selected from at least one of conflict focusing operator, time bucket operator and region operator; the repair operator is selected from at least one of space-priority repair operator, spatiotemporal collaborative repair operator and region exchange operator.

5. The dock resource collaborative optimization method based on constraint programming and adaptive large neighborhood search as described in claim 4, characterized in that, The execution logic of the conflict focusing operator includes: 5-1) Evaluate the number of geometric overlap conflicts in each segment of the current layout; 5-2) Prioritize the segment with the highest number of overlapping conflicts as the target for destruction; 5-3) When there are few conflicts in the layout, it degenerates into selecting objects to destroy based on area priority.

6. The dock resource collaborative optimization method based on constraint programming and adaptive large neighborhood search as described in claim 1, characterized in that, The cost function of the adaptive large neighborhood search model described in step 3) includes one or more of the following penalty terms: 6-1) Overlap penalty term for penalizing the geometric overlap area between segments; 6-2) Penalty items for exceeding the boundary distance of segments; 6-3) Delay penalty items that penalize the amount of time delay in segmented entry; 6-4) Aggregation penalty term that penalizes the spatial dispersion of segments with the same ship number.

7. The dock resource collaborative optimization method based on constraint programming and adaptive large neighborhood search as described in claim 1, characterized in that, The method further includes a parallel evaluation acceleration step: during a single iteration of step 3) of claim 1, multiple candidate layout solutions are generated in parallel, and the cost function of each candidate solution is calculated using a process pool, from which the candidate solution with the best cost is selected for subsequent iterations.

8. The dock resource collaborative optimization method based on constraint programming and adaptive large neighborhood search as described in claim 1, characterized in that, The method also includes a dynamic replanning step, specifically: in response to a disturbance event occurring at the production site, using the current actual production progress as a new initial condition, automatically triggering and re-executing the steps of claim 1 to generate an updated dock resource collaborative optimization scheme.

9. A dock resource collaborative optimization system based on constraint programming and adaptive large neighborhood search, characterized in that, The system includes one or more modules configured to perform the following operations: 1) Time and resource scheduling module: configured to execute step 1) as described in claim 1, based on the network data, gantry crane resource data and site resource data, to construct and solve the first-stage constraint programming model to generate time scheduling and segmented entry and exit time windows; 2) Spatial layout optimization module: configured to perform steps 2) and 3) of claim 1, receive a time window, construct and iteratively solve the second-stage adaptive large neighborhood search model, and process the delay penalty during the iteration process to find the physical layout location; 3) Solution output module: configured to execute step 4 of claim 1, and at the end of the iteration, combine the time schedule and physical layout position to output the final time-space co-optimization solution.

10. A computer device comprising a processor and a memory, characterized in that, The memory stores a computer program that, when executed by a processor, implements the method as described in any one of claims 1 to 8.

11. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1 to 8.

Images