A four-way vehicle warehouse storage location-task-path dynamic joint optimization method

Abstract

The application discloses a four-way vehicle warehouse storage location-task-path dynamic joint optimization method, and relates to the technical field of vehicle scheduling.The method comprises the following steps: obtaining a task set based on a rolling window, constructing a conflict propagation graph to identify affected tasks, adaptively initializing a population by warm-start, iteratively optimizing by multi-objective differential evolution, simulating execution based on a time-space reservation table, calculating the sum of completion time and waiting time delay, device load variance and library capacity hotspot entropy, outputting a Pareto frontier and issuing frozen tasks.The application can eliminate collision and waiting deviation between planning and execution, support flow order increment hot start, and realize multi-objective balance of efficiency, load and hotspot.

Description

Technical Field

[0001] This invention relates to the field of vehicle scheduling technology, and in particular to a dynamic joint optimization method for four-way vehicle storage location-task-path. Background Technology

[0002] In a four-way vehicle-intensive storage system, the completion of inbound and outbound tasks involves three coupled decision-making processes: location allocation, equipment allocation, and path planning. Location selection affects path length and elevator load; equipment allocation determines the cost of changing floors and waiting delay; and path planning is constrained by the stack depth of the locations and the spatiotemporal conflicts between multiple vehicles on the same floor. Existing research mainly follows three routes: first, the integrated optimization route, which models the three types of decisions in a unified manner and uses metaheuristic methods such as Variable Neighborhood Search Genetic Algorithm (VNSGA) to solve batch tasks in one go; second, the serial process route, which makes decisions step by step in the order of "task sorting → location selection → path planning"; and third, the optimization of specific task allocation and sorting for dual-elevator scenarios.

[0003] However, the above methods have significant drawbacks in actual industrial deployments. First, existing integrated optimization models explicitly assume "no collision conflicts among multiple vehicles on the same level," resulting in an idealized lower limit for completion time. In actual execution, collision avoidance waiting causes a 15% to 30% deviation, leading to a disconnect between "optimal planning" and "optimal execution." Second, existing methods are geared towards static batch tasks and cannot handle scenarios with dynamically arriving streaming orders. Resolving from scratch when a new order is inserted is not only computationally expensive (on the order of several minutes) but also disrupts the already issued execution plan, resulting in rollback costs. Third, existing models mostly focus on the shortest completion time as a single objective. Under long-term operation, tasks are concentrated on a few near-lifter locations and vehicles, causing uneven equipment load and concentrated warehouse hotspots, neglecting the trade-offs between completion time, load balancing, and hotspot dispersion. Fourth, the serial process architecture lacks feedback loops. Deadlocks or congestion discovered by downstream path planning cannot drive upstream location and equipment reselection, resulting in a disconnect between theory and engineering implementation.

[0004] In summary, existing technologies lack a dynamic joint optimization method that embeds spatiotemporal conflict constraints into fitness evaluation, supports incremental hot start of streaming orders, takes into account multiple objective trade-offs, and has a feedback loop. Summary of the Invention

[0005] The technical problem to be solved by this invention is to address the shortcomings of existing technologies, specifically by providing a four-way vehicle-warehouse storage location-task-path dynamic joint optimization method, as detailed below: 1) In a first aspect, the present invention provides a dynamic joint optimization method for four-way vehicle storage location-task-path based on rolling window and warm-start incremental solution, the specific technical solution of which is as follows: S1. Obtain the set of four-way vehicle storage path scheduling tasks within the current scrolling planning window; wherein, the set of four-way vehicle storage path scheduling tasks includes: unfinished old tasks and newly arrived outbound or inbound tasks. S2. Construct a conflict propagation graph, perform a breadth-first search in the conflict propagation graph to identify the complete set of affected tasks, and determine the tasks in the set of affected tasks as tasks to participate in re-optimization. S3. Determine the set of frozen tasks, perform adaptive population initialization, and generate the initial population; S4. The initial population is iteratively optimized using a multi-objective differential evolution algorithm to generate offspring individuals; S5. Call the encoding repair operator to perform legality repair on the offspring individuals, evaluate the fitness of the repaired offspring individuals, and calculate the three objective function values ​​for each offspring individual; S6. Based on the three objective function values, perform non-dominated sorting and crowding distance calculation on the union of the parent and offspring populations, and select individuals for the next generation population. S7. Repeat steps S4 to S6 until the termination condition is met, and output the Pareto front solution set of the current rolling planning window. S8. Select a solution from the Pareto front solution set as the final solution, send the tasks belonging to the frozen task set in the final solution to the warehouse control system for execution, and update the space-time reservation table.

[0006] Based on the above solution, the present invention can be further improved as follows.

[0007] Furthermore, the process of generating the initial population specifically includes: The rate of change is calculated based on the ratio of the number of tasks in the affected task set to the total number of tasks in the current rolling planning window; the seed inheritance ratio is dynamically calculated based on the rate of change; the historical Pareto front solution set is obtained; seed individuals are inherited from the historical Pareto front solution set according to the seed inheritance ratio, otherwise all initial population individuals are randomly generated; the gene segments belonging to the affected task set in the inherited seed individuals are perturbed and resampled, and the gene segments of the frozen task set are directly retained to generate some initial population individuals; the remaining initial population individuals are randomly generated to make the initial population size reach a preset value.

[0008] Furthermore, the process of generating offspring individuals specifically includes: In each generation of evolution, differential evolution mutation and structure-preserving crossover operations are performed on each individual in the population to generate offspring individuals; wherein, the structure-preserving crossover operation is performed on a single task basis, exchanging the location decision variables, equipment decision variables, and path decision variables as a whole group.

[0009] Furthermore, it also includes: after generating offspring individuals, performing a variable neighborhood local search on the offspring individuals with a preset probability. The variable neighborhood local search includes: swapping the location decision variables of two affected tasks in the location exchange neighborhood, redistributing the equipment decision variables of a single task in the equipment load weighted average neighborhood, and replacing the path decision variables of a single task with the path in its path pool that minimizes the waiting delay of the spatiotemporal reservation table in the path replacement neighborhood.

[0010] Furthermore, the process of calculating the three objective function values ​​for each offspring individual is as follows: The path for each task in the offspring individual is simulated based on the spatiotemporal reservation table; when the time windows of multiple four-way vehicles overlap at the same track node, conflict resolution is performed according to task priority and the waiting delay is accumulated; the three objective functions include: minimizing the sum of completion time and accumulated waiting delay, minimizing equipment load variance, and maximizing the entropy of warehouse access hotspots.

[0011] Furthermore, the conflict propagation graph uses unfrozen tasks in the task set as nodes, and the paths between tasks with spatiotemporal overlap, shared four-way vehicle edges, and shared elevator time windows as edges; the newly arrived task is used as a seed node.

[0012] Furthermore, the legality repair includes, in sequence: location conflict repair, depth stack last-in-first-out order repair, double hoist layer switching constraint repair, and path pool mismatch repair.

[0013] 2) Secondly, the present invention also provides a four-way vehicle storage location-task-path dynamic joint optimization system based on rolling window and warm-start incremental solution, the specific technical solution of which is as follows: The acquisition module is used to: acquire the set of four-way vehicle storage route scheduling tasks within the current scrolling planning window; wherein, the set of four-way vehicle storage route scheduling tasks includes: incomplete old tasks and newly arrived outbound or inbound tasks; The construction module is used to: construct a conflict propagation graph, perform a breadth-first search in the conflict propagation graph, identify the complete set of affected tasks, and determine the tasks in the set of affected tasks as tasks to participate in re-optimization; The determination module is used to: determine the set of frozen tasks, perform adaptive population initialization, and generate the initial population; The generation module is used to iteratively optimize the initial population using a multi-objective differential evolution algorithm to generate offspring individuals; The repair module is used to: call the encoding repair operator to perform legality repair on the offspring individuals, evaluate the fitness of the repaired offspring individuals, and calculate the three objective function values ​​for each offspring individual; The selection module is used to: perform non-dominated sorting and crowding distance calculation on the union of the parent and offspring populations based on the three objective function values, and select individuals for the next generation population; The repeat module is used to: repeat the repair module and the selection module until the termination condition is met, and output the Pareto front solution set of the current rolling planning window; The update module is used to: select a solution from the Pareto front solution set as the landing solution, send the tasks belonging to the frozen task set in the landing solution to the warehouse control system for execution, and update the spatiotemporal reservation table.

[0014] Based on the above solution, the present invention can be further improved as follows.

[0015] Furthermore, the process of generating the initial population specifically includes: The rate of change is calculated based on the ratio of the number of tasks in the affected task set to the total number of tasks in the current rolling planning window; the seed inheritance ratio is dynamically calculated based on the rate of change; the historical Pareto front solution set is obtained; seed individuals are inherited from the historical Pareto front solution set according to the seed inheritance ratio, otherwise all initial population individuals are randomly generated; the gene segments belonging to the affected task set in the inherited seed individuals are perturbed and resampled, and the gene segments of the frozen task set are directly retained to generate some initial population individuals; the remaining initial population individuals are randomly generated to make the initial population size reach a preset value.

[0016] Furthermore, the process of generating offspring individuals specifically includes: In each generation of evolution, differential evolution mutation and structure-preserving crossover operations are performed on each individual in the population to generate offspring individuals; wherein, the structure-preserving crossover operation is performed on a single task basis, exchanging the location decision variables, equipment decision variables, and path decision variables as a whole group.

[0017] Furthermore, it also includes: a search module for: after generating offspring individuals, performing a variable neighborhood local search on the offspring individuals with a preset probability, the variable neighborhood local search including: a cargo location exchange neighborhood for swapping the cargo location decision variables of two affected tasks, a device reallocation neighborhood for redistributing the device decision variables of a single task according to the inverse weight of the device load, and a path replacement neighborhood for replacing the path decision variables of a single task with the path in its path pool that minimizes the waiting delay of the spatiotemporal reservation table.

[0018] Furthermore, the process of calculating the three objective function values ​​for each offspring individual is as follows: The path for each task in the offspring individual is simulated based on the spatiotemporal reservation table; when the time windows of multiple four-way vehicles overlap at the same track node, conflict resolution is performed according to task priority and the waiting delay is accumulated; the three objective functions include: minimizing the sum of completion time and accumulated waiting delay, minimizing equipment load variance, and maximizing the entropy of warehouse access hotspots.

[0019] Furthermore, the conflict propagation graph uses unfrozen tasks in the task set as nodes, and the paths between tasks with spatiotemporal overlap, shared four-way vehicle edges, and shared elevator time windows as edges; the newly arrived task is used as a seed node.

[0020] Furthermore, the legality repair includes, in sequence: location conflict repair, depth stack last-in-first-out order repair, double hoist layer switching constraint repair, and path pool mismatch repair.

[0021] 3) In a third aspect, the present invention also provides an electronic device, the electronic device including a processor coupled to a memory, the memory storing at least one computer program, the at least one computer program being loaded and executed by the processor to enable the electronic device to perform any of the above methods.

[0022] 4) In a fourth aspect, the present invention also provides a computer-readable storage medium storing at least one computer program, which is loaded and executed by a processor to enable a computer to implement any of the above methods.

[0023] The beneficial effects of the four-way vehicle storage location-task-path dynamic joint optimization method based on rolling window and warm-start incremental solution provided by this invention are as follows: This invention processes the dynamic arrival of streaming orders through a rolling window framework, utilizes a breadth-first search of a conflict propagation graph to identify the complete set of affected tasks, and performs incremental re-optimization only on affected tasks while preserving already issued tasks. This solves the problems of existing technologies being unable to handle online order flows with one-time batch solutions and requiring re-solution from zero for new order arrivals. By embedding a spatiotemporal reservation table into fitness evaluation and directly incorporating the waiting delay caused by conflict resolution into the completion time objective, it addresses the problem of planning and execution deviations caused by ignoring spatiotemporal conflicts of four-way vehicles at the same level in joint optimization modeling. Through a multi-objective differential evolution algorithm, it simultaneously optimizes the three objectives of completion time, equipment load variance, and warehouse capacity hotspot entropy and outputs a Pareto front solution set, solving the problems of uneven equipment load and concentrated warehouse capacity hotspots caused by single-objective optimization. By using a three-stage unified gene encoding and structure-preserving crossover operator, it allows path decisions to participate in evolution as explicit variables, solving the problems of lack of feedback loops in serial process architectures and the inability of downstream constraints to influence upstream decisions. This achieves improvements in planning and execution consistency, real-time response capability of order flows, and multi-objective trade-off capabilities.

[0024] It should be noted that the beneficial effects of the technical solutions of the second to fourth aspects of the present invention and their corresponding possible implementations can be found in the above description of the technical effects of the first aspect and its corresponding possible implementations, and will not be repeated here. Attached Figure Description

[0025] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings: Figure 1 This is a flowchart illustrating a four-way vehicle storage location-task-path dynamic joint optimization method according to an embodiment of the present invention. Figure 2 This is a schematic diagram of the event-driven solution process closed loop of a four-way vehicle storage location-task-path dynamic joint optimization method according to an embodiment of the present invention. Figure 3 This is a schematic diagram of a four-step joint method for repairing a dynamic joint optimization method for four-way vehicle storage location-task-path according to an embodiment of the present invention; Figure 4 This is a schematic diagram of the three-segment chromosome structure and decoding of a four-way vehicle storage location-task-path dynamic joint optimization method according to an embodiment of the present invention. Figure 5 This is a schematic diagram of the full fitness evaluation process of SRT embedding in a four-way vehicle storage location-task-path dynamic joint optimization method according to an embodiment of the present invention. Figure 6 This is a schematic diagram of the incremental SRT evaluation process of a four-way vehicle storage location-task-path dynamic joint optimization method according to an embodiment of the present invention. Figure 7 This is a schematic diagram of the conflict propagation graph construction and BFS propagation process of a four-way vehicle storage location-task-path dynamic joint optimization method according to an embodiment of the present invention. Figure 8 This is a schematic diagram of the adaptive warm-start population initialization process of a four-way vehicle storage location-task-path dynamic joint optimization method according to an embodiment of the present invention; Figure 9 This is a schematic diagram of the SPX structure maintaining the cross-three-branch logic of a four-way vehicle storage location-task-path dynamic joint optimization method according to an embodiment of the present invention; Figure 10 This is a schematic diagram of the MODE main loop process of a four-way vehicle storage location-task-path dynamic joint optimization method according to an embodiment of the present invention; Figure 11 This is a schematic diagram of the architecture of an electronic device according to the present invention. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.

[0027] like Figures 1 to 10 As shown in the figure, an embodiment of the present invention provides a dynamic joint optimization method for four-way vehicle storage location-task-path based on rolling window and warm-start incremental solution, which includes the following steps: S1. Obtain the set of four-way vehicle storage path scheduling tasks within the current scrolling planning window; wherein, the set of four-way vehicle storage path scheduling tasks includes: unfinished old tasks and newly arrived outbound or inbound tasks. S2. Construct a conflict propagation graph, perform a breadth-first search in the conflict propagation graph to identify the complete set of affected tasks, and determine the tasks in the set of affected tasks as tasks to participate in re-optimization. S3. Determine the set of frozen tasks, perform adaptive population initialization, and generate the initial population; S4. The initial population is iteratively optimized using a multi-objective differential evolution algorithm to generate offspring individuals; S5. Call the encoding repair operator to perform legality repair on the offspring individuals, evaluate the fitness of the repaired offspring individuals, and calculate the three objective function values ​​for each offspring individual; S6. Based on the three objective function values, perform non-dominated sorting and crowding distance calculation on the union of the parent and offspring populations, and select individuals for the next generation population. S7. Repeat steps S4 to S6 until the termination condition is met, and output the Pareto front solution set of the current rolling planning window. S8. Select a solution from the Pareto front solution set as the final solution, send the tasks belonging to the frozen task set in the final solution to the warehouse control system for execution, and update the space-time reservation table.

[0028] Taking a certain intelligent warehousing system as an example, the system uses a four-way shuttle system as its physical foundation. A four-way shuttle system refers to an automated three-dimensional warehouse system in which four-way shuttles can travel in four directions (front, back, left, and right) on the rack tracks.

[0029] For ease of understanding, the technical terms used below will be explained in a consistent manner: Four-way vehicle warehousing route scheduling task: In a four-way vehicle intensive warehousing system, this refers to a complete inbound / outbound operation unit that needs to be executed by one or more four-way vehicles. An inbound task specifies the information of the goods to be inbound and the goal of storing the goods in an available storage location, while an outbound task specifies the information of the goods to be outbound and the goal of retrieving the goods from a designated storage location.

[0030] Rolling planning window: refers to the time range considered in each optimization during dynamic scheduling, and only tasks that arrive or expire within that time range are uniformly planned and solved.

[0031] Freeze threshold: This refers to a time-length parameter used to determine which tasks have been issued for execution and cannot be changed. Tasks whose time elapsed since the current moment is less than or equal to the freeze threshold are considered to have been issued and frozen, and their decision variables remain unchanged in subsequent optimizations.

[0032] Conflict propagation graph: A graph structure used to describe the conflict propagation relationships between tasks. Each node in the graph represents an unfrozen task, and each edge represents a conflict propagation path between two tasks.

[0033] Breadth-first search: refers to a graph traversal algorithm that starts from the seed node and visits neighboring nodes layer by layer until the stopping condition is met.

[0034] The set of tasks affected: refers to the set of tasks that need to be re-optimized due to the addition of new tasks, including tasks that directly conflict with the new tasks and tasks that indirectly conflict through the time window of the shared four-way vehicle or the shared elevator.

[0035] Unfrozen tasks: These are tasks that are more than the freezing threshold away from the current time. These tasks have not yet been issued for execution, and their decision variables can be adjusted during the optimization process.

[0036] Pareto front solution set: In a multi-objective optimization problem, this refers to the set of all non-dominated solutions. A solution dominates another solution if the former is not inferior to the latter on all objectives and is superior to the latter on at least one objective.

[0037] Spatiotemporal reservation table: This refers to a data structure used to record the occupancy status of each track node or elevator connection point within a warehousing system during different time intervals. Each occupancy record includes a node identifier, occupancy start time, occupancy end time, occupant vehicle identifier, associated task identifier, freeze flag, and task conflict priority score.

[0038] Hot start: refers to using the solution of a previous similar problem as an initial solution or seed when solving an optimization problem, in order to accelerate the convergence process.

[0039] Multi-objective differential evolution algorithm: refers to a class of evolutionary algorithms that combine differential evolution algorithm with multi-objective optimization strategy. Typical strategies include using non-dominated sorting and crowding distance for individual selection.

[0040] Structure-preserving cross-operation: refers to a cross-operation based on a single task as the basic unit, ensuring that the location decision variables, equipment decision variables, and path decision variables of a task are inherited as a whole from a parent individual.

[0041] Variable neighborhood local search: refers to a local search method that searches alternately in multiple different types of neighborhood structures in order to escape local optima.

[0042] Encoding repair operators: refer to a set of operation sequences used to repair illegal chromosome individuals into legal individuals, including location conflict repair, depth stack last-in-first-out order repair, double elevator layer-swapping constraint repair, and path pool mismatch repair.

[0043] Last-In-First-Out (LIFO) order in depth stacks: This refers to the constraint rule that, within the same depth stack, goods should be removed from the deeper storage location (located at the bottom of the stack) before goods from the shallower storage location (located at the top of the stack).

[0044] Equipment load variance: refers to the variance of the cumulative occupancy time of all equipment (including four-way vehicles, freight hoists, and truck hoists), used to measure the degree of load balance among equipment.

[0045] Warehouse capacity access hotspot entropy: refers to the information entropy value calculated based on the access frequency of each storage location or storage area, used to measure the degree of hotspot balance in the distribution of warehouse capacity access.

[0046] Non-dominated ranking: refers to the process of ranking individuals in a population according to Pareto dominance, with individuals having smaller non-dominated stratum numbers being better.

[0047] Crowding distance: In multi-objective optimization, it is a metric used to measure the density of a solution on the Pareto front, and is used to maintain the diversity of the solution set.

[0048] Before providing a specific explanation, it is important to clarify that the specific values ​​involved in all the following examples are merely illustrative and not specific limitations. They can be adjusted as needed in actual applications.

[0049] The specific implementation process of S1 includes: Real-time monitoring of task arrival events. When a new outbound or inbound task is received, the current time is obtained, and a 15-minute time interval starting from the current time is determined as the current rolling planning window. All incomplete old tasks and newly arrived outbound or inbound tasks are retrieved from the task pool. Tasks that arrive or expire within the aforementioned 15-minute time interval are included in the four-way vehicle warehousing path scheduling task set within the current rolling planning window.

[0050] By using a rolling planning window mechanism, the dynamic scheduling problem in the infinite time domain is divided into a static optimization problem in the finite time domain, which can handle orders arriving in a streaming manner while controlling the scale of each optimization.

[0051] The specific implementation process of S2 includes: Construct a conflict propagation graph. The node set of the conflict propagation graph consists of all unfrozen tasks within the current rolling planning window that are more than 2 minutes away from the current time (i.e., the freeze threshold). Using newly arrived outbound or inbound tasks as seed nodes, perform a breadth-first search on this graph. During the search, sequentially visit the neighbor nodes of the seed node, and then the neighbor nodes of those neighbor nodes, until the search depth reaches the upper limit of 3 or the ratio of the number of identified affected tasks to the total number of unfrozen tasks exceeds 0.6, at which point propagation stops. All visited nodes constitute the set of affected tasks, which are then identified as tasks participating in the re-optimization.

[0052] By using a conflict propagation graph and breadth-first search, not only can tasks that directly conflict with new tasks be captured, but also tasks that are indirectly affected can be captured along the shared four-way vehicle link and the shared elevator time window link, thus avoiding suboptimal scheduling caused by missing indirectly affected tasks.

[0053] The specific implementation process of S3 includes: Determine the frozen task set. The frozen task set includes all tasks that are less than or equal to 2 minutes away from the current time (i.e., the freeze threshold). These tasks have been issued to the warehouse control system for execution, and their decision variables cannot be changed during the optimization process. Perform adaptive population initialization to generate the initial population.

[0054] By freezing thresholds, tasks that have already been issued are decoupled from tasks that need to be optimized, ensuring that optimization calculations and physical execution are carried out in parallel, and that tasks that have already started will not be rolled back due to the arrival of new orders.

[0055] The specific implementation process of S4 includes: A multi-objective differential evolution algorithm is used to iteratively optimize the initial population. The population size is fixed at 80. Differential evolution mutation and structure-preserving crossover operations are performed sequentially on each individual in the population to generate offspring. The formula for the differential evolution mutation operation is: .in For the generated mutant individuals, , , These are three distinct parent individuals randomly selected from the current population. This is the scaling factor, with a value of 0.5.

[0056] This invention employs a three-segment unified gene coding approach to represent the three types of decision variables for each task in an integrated manner. For a given set of variables... A scrolling window for each task, where the chromosome individual is defined as... ,in: For the sequence of decision variables for cargo location, The location identifier assigned to task i; For the sequence of equipment decision variables, each It includes three sub-variables: four-way vehicle number, cargo hoist number, and truck hoist number; For the path index decision variable sequence, The value is an integer between 1 and K, pointing to the path pool pre-generated for this task. The first in There are 10 candidate paths. By using the path index as an explicit decision variable to participate in the mutation and crossover operations of differential evolution, path selection, cargo location allocation, and equipment allocation are optimized synchronously in the same search space.

[0057] For each task, the system pre-calculates its path pool. The path pool is generated by running a Top-K A algorithm on the residual graph after removing the space-time occupancy of already issued tasks, for each task's (origin, location) pair. The algorithm returns K candidate paths with increasing time cost. K is set to 6. When the location decision variables of a task are changed during the evolution process, the original path pool for that task becomes invalid and needs to be regenerated during the validity repair phase.

[0058] Employing a multi-objective differential evolution algorithm, it can search in multiple objective directions simultaneously and output a Pareto front solution set, allowing scheduling decision-makers to flexibly select the solution to implement based on current business preferences.

[0059] The specific implementation process of S5 includes: The code repair operator is invoked to perform legality repair on the offspring individuals. Legality repair includes, in sequence, location conflict repair, depth stack last-in-first-out order repair, double-elevator layer-switching constraint repair, and path pool mismatch repair. After repair, the fitness of the repaired offspring individuals is evaluated, and three objective function values ​​are calculated for each offspring individual.

[0060] As an alternative to full evaluation, this method also provides an incremental SRT evaluation mechanism to accelerate fitness calculation during the evolutionary process. The specific steps are: comparing parent chromosomes... With offspring chromosomes Extracting the set of tasks that have changed genes If |C| / A value >0.4 indicates excessive variation, in which case a full evaluation should be rolled back. Otherwise, remove all task occupancy records from the current SRT in C, and revert the corresponding cumulative equipment hours and warehouse access frequency values. Then, according to the task arrival time order, rewrite the new paths of the tasks in C (taken from...). Reinsert the SRT, perform conflict resolution and wait delay accumulation. Finally, update the three objective function values ​​only for the task increment involved in C. The computational complexity of this incremental evaluation is O(n log n). In typical scenarios, it can reduce the time consumption by 60% to 80% compared to full evaluation.

[0061] The coding repair operator ensures the validity of solutions generated by crossover and mutation operations under constraints such as cargo location conflict, depth stack order, hoisting machine, and path consistency, thus avoiding invalid solutions during evolution. Fitness evaluation directly incorporates spatiotemporal conflicts into the objective function calculation, eliminating the deviation between optimal planning and optimal execution.

[0062] The specific implementation process of S6 includes: Based on the three objective function values, a non-dominated sort and crowding distance calculation are performed on the union of the parent and offspring populations. The non-dominated sort stratifies the population, with individuals numbered 1 in the non-dominated layer forming the optimal Pareto front for the current population. For individuals within the same non-dominated layer, the crowding distance is calculated. A larger crowding distance indicates a lower density and better diversity within the Pareto front. Individuals are selected sequentially according to their non-dominated layer numbers, from smallest to largest. When not all individuals within a particular non-dominated layer can be selected for the next generation, the individual with the largest crowding distance from that layer is prioritized for retention, until the next generation population reaches a size of 80.

[0063] The combined use of non-dominated sorting and crowding distance maintains the diversity of the Pareto front while preserving convergence, thus preventing the population from falling into local optima too early.

[0064] The specific implementation process of S7 includes: Repeat steps S4 through S6 until the maximum number of generations (40) is reached. Upon reaching the termination condition, output the Pareto front solution set for the current rolling planning window. This solution set contains multiple non-dominated solutions, each representing a combination of location allocation schemes, equipment allocation schemes, and path allocation schemes.

[0065] By setting the maximum number of generations as the termination condition, the computational time complexity of the algorithm is kept under control, thus meeting the real-time scheduling requirements of the warehousing system.

[0066] The specific implementation process of S8 includes: A solution is selected from the Pareto front solution set as the final solution. The selection method is as follows: range normalization is performed on each objective function value in the Pareto front solution set. ,in, , The value is a very small positive number to prevent division by zero errors. A weighted score is calculated based on business preference weights: , For business strategy parameters (note that day / night shifts and promotional / off-seasons differ), a weighted score is calculated for each solution based on day shift business preference weights. The solution with the highest score is selected as the implementation solution. Tasks belonging to the frozen task set from the implementation solutions are sent to the warehouse control system for execution, and the freeze flag of the corresponding record in the space-time reservation table is set to true. The system also receives the actual task execution time, actual collision waiting time, and actual hoist occupancy information reported by the warehouse control system, and updates the space-time reservation table accordingly for use in the next round of rolling window optimization.

[0067] By adopting a solution selection strategy that combines range normalization with business preference weighting, the optimization focus can be flexibly adjusted according to different business periods (day shift, night shift, promotion season, off-season). At the same time, the execution feedback loop ensures that the spatiotemporal reservation table always remains consistent with the physical system state.

[0068] Furthermore, the process of generating the initial population specifically includes: The rate of change is calculated based on the ratio of the number of tasks in the affected task set to the total number of tasks in the current rolling planning window; the seed inheritance ratio is dynamically calculated based on the rate of change; the historical Pareto front solution set is obtained; seed individuals are inherited from the historical Pareto front solution set according to the seed inheritance ratio, otherwise all initial population individuals are randomly generated; the gene segments belonging to the affected task set in the inherited seed individuals are perturbed and resampled, and the gene segments of the frozen task set are directly retained to generate some initial population individuals; the remaining initial population individuals are randomly generated to make the initial population size reach a preset value.

[0069] In another embodiment of this scheme, the rate of change is calculated. . .

[0070] The seed inheritance ratio is dynamically calculated based on the rate of change. The formula for calculating the seed inheritance ratio is: .in, To achieve the maximum inheritance ratio, This represents the minimum inheritance ratio.

[0071] Obtain the historical Pareto front solution set. The historical Pareto front solution set refers to the Pareto front solution set output and saved at the end of the previous rolling window optimization. If there is no historical Pareto front solution set for the first run, all initial population individuals are directly generated randomly.

[0072] When a historical Pareto front solution set exists, seed individuals are inherited from it according to the seed inheritance ratio. Specifically, if the historical Pareto front solution set contains m solutions, then the number of offspring derived from each non-dominated solution is... , where NP is the population size.

[0073] For the inherited seed individuals, only the gene segments belonging to the affected task set are perturbed and resampled. The perturbed resampling method is as follows: the cargo location decision variable is uniformly and randomly sampled from the feasible cargo location set of the task; the four-way vehicle decision variable is sampled with a weighted average of the inverse of the current equipment load, where equipment load refers to the total occupied time of the four-way vehicle in the spatiotemporal reservation table; the cargo hoist decision variable is sampled with a weighted average of the inverse of the occupied time; the truck hoist decision variable defaults to the same value as the cargo hoist decision variable; the path index decision variable is uniformly and randomly sampled between 1 and the path pool size 6. After perturbation, the encoding repair operator is called to perform legality repair. For gene segments in the frozen task set, the original values ​​are directly retained without any modification. After the above processing is completed, a portion of the initial population individuals are generated.

[0074] The remaining individuals in the initial population are randomly generated to bring the initial population size to a preset value of 80. The random generation method involves uniformly sampling all decision variables for each task within the feasible region and then calling the encoding repair operator to perform legality repair.

[0075] By adaptively adjusting the seed inheritance ratio based on the rate of change, it can fully utilize historical high-quality solutions to accelerate convergence under small perturbations and ensure random diversity to prevent getting trapped in local optima under large changes, thus achieving efficient hot start in the rolling window scenario.

[0076] Furthermore, the process of generating offspring individuals specifically includes: In each generation of evolution, differential evolution mutation and structure-preserving crossover operations are performed on each individual in the population to generate offspring individuals; wherein, the structure-preserving crossover operation is performed on a single task basis, exchanging the location decision variables, equipment decision variables, and path decision variables as a whole group.

[0077] In another embodiment of this scheme, in each generation of evolution, a mutation vector is generated for each individual in the population. The mutation vector is generated using the standard mode of the multi-objective differential evolution algorithm, that is, three distinct individuals are randomly selected from the current population, and the mutated individual is obtained by adding a scaling factor of 0.5 to the first individual and multiplying the difference between the second and third individuals.

[0078] The structure-preserving crossover operation generates offspring individuals. This operation operates on a single task unit. For each task within the current rolling planning window, the complete set of decision variables (including location, equipment, and routing variables) for that task is obtained from the mutated individual with a probability of 0.8, and from the parent individual with a probability of 0.2. This operation ensures that the three decision variables for a task are always inherited as a whole from a single parent, preventing cross-task splicing where location decision variables are taken from the mutated individual while equipment decision variables are taken from the parent individual.

[0079] The structure maintains the cross operation to avoid the traditional uniform cross operation from destroying the semantic association between the location, equipment and path within the task, ensuring higher legality of the solution after cross and reducing the workload of subsequent coding repair operators.

[0080] Furthermore, it also includes: after generating offspring individuals, performing a variable neighborhood local search on the offspring individuals with a preset probability. The variable neighborhood local search includes: swapping the location decision variables of two affected tasks in the location exchange neighborhood, redistributing the equipment decision variables of a single task in the equipment load weighted average neighborhood, and replacing the path decision variables of a single task with the path in its path pool that minimizes the waiting delay of the spatiotemporal reservation table in the path replacement neighborhood.

[0081] In another embodiment of this scheme, after generating offspring individuals through the structure-preserving crossover operation, a variable neighborhood local search is performed on the offspring individuals with a preset probability. The preset probability is set according to the neighborhood type: the trigger probability for the location exchange neighborhood is 0.3, the trigger probability for the equipment reallocation neighborhood is 0.3, and the trigger probability for the path replacement neighborhood is 0.4.

[0082] The operation of the location swap neighborhood is as follows: randomly select two affected tasks and swap their location decision variables. This operation aims to explore the impact of different location allocation combinations on completion time.

[0083] The operation of equipment reallocation neighborhood is as follows: randomly select an affected task, and reallocate the task's equipment decision variables (including four-way vehicle decision variables, freight hoist decision variables, and truck hoist decision variables) according to the inverse weight of the current equipment load. Equipment load refers to the total occupied time of the equipment in the spatiotemporal reservation table. This operation aims to explore the impact of different equipment allocation combinations on equipment load balancing.

[0084] The path replacement neighborhood operation works as follows: randomly select an affected task, obtain the waiting latency generated by each candidate path in the path pool of that task during simulated execution in the spatiotemporal reservation table, and select the path index corresponding to the candidate path with the minimum waiting latency as the new path decision variable for that task. This operation aims to directly optimize path selection using the congestion information fed back from the spatiotemporal reservation table.

[0085] The acceptance criterion for variable neighborhood local search is as follows: if the new solution after neighborhood operation is superior to the original solution in terms of Pareto dominance, then the new solution is accepted; otherwise, the new solution is accepted based on the simulated annealing probability, which is calculated as P. accept = .

[0086] Where Δ is the difference between the new solution and the original solution in the first objective function (the sum of completion time and waiting time) (if a multi-objective approach is adopted, the increment of the aggregated index is used). To simulate the annealing temperature. The annealing temperature decreases linearly with the number of generations g: = in, Given the initial temperature, G max This represents the maximum number of iterations.

[0087] The variable neighborhood local search adds local refinement capabilities on the basis of global evolutionary search. In particular, the path replacement neighborhood directly uses the congestion waiting information in the spatiotemporal reservation table to guide path selection, which can effectively reduce the collision waiting latency in actual execution.

[0088] Furthermore, the process of calculating the three objective function values ​​for each offspring individual is as follows: The path for each task in the offspring individual is simulated based on the spatiotemporal reservation table; when the time windows of multiple four-way vehicles overlap at the same track node, conflict resolution is performed according to task priority and the waiting delay is accumulated; the three objective functions include: minimizing the sum of completion time and accumulated waiting delay, minimizing equipment load variance, and maximizing the entropy of warehouse access hotspots.

[0089] In another embodiment of this scheme, the path of each task in the child individual is simulated based on the spatiotemporal reservation table. The simulation is performed in the following order: all occupancy records with the freeze flag set to true are copied from the base spatiotemporal reservation table to the temporary spatiotemporal reservation table for evaluation. The equipment time accumulation array and the warehouse access frequency array are initialized to zero. All tasks in the current rolling planning window are sorted in ascending order of arrival time to obtain the evaluation sequence.

[0090] For each task in the evaluation sequence, the location decision variables, equipment decision variables, and path index decision variables for that task are obtained from the offspring individuals. The corresponding node sequence is retrieved from the pre-computed path pool for that task based on the path index. The release time for that task is calculated, and the release time is the larger of the task's arrival time and the end time of the previous task on the four-way vehicle assigned to that task.

[0091] Simulated driving is performed along each track node of the task path. For each node, a time window is formed based on the four-way vehicle's entry and exit times at that node. The set of occupancy records in the temporary spatiotemporal reservation table used for evaluation, which have the same node identifier and overlap with the time window, is queried. If the query result is not empty, the conflict priority score between the current task and each conflicting task is calculated. Conflict Priority Score , Let be the deadline for task i; This is the current simulation time. Let be the number of remaining path nodes for task i; The indicator function takes a value of 1 when the four-way vehicle is carrying cargo, and a value of 0 otherwise; α, β, and γ are weighting coefficients, with values ​​of 0.4, 0.4, and 0.2 respectively. The cargo status indicator is set to 1 when the four-way vehicle is loading cargo, and 0 otherwise. If the priority score of the current task is lower than that of a conflicting task, the current task needs to wait for a duration equal to the end time of the conflicting task's occupation minus the planned entry time of the current task. The time windows of all subsequent path nodes of the current task are shifted backward by this waiting duration, accumulating the waiting delay of the task.

[0092] Insert the path occupancy record (including the offset time window) of the current task after conflict resolution into the temporary spatiotemporal reservation table for evaluation. Accumulate the equipment man-hours of the four-way vehicle, the cargo hoist, and the truck hoist assigned to this task. Accumulate the access frequency of the warehouse area where the assigned storage location of this task is located.

[0093] After all tasks are completed, calculate the three objective function values. The first objective function value, f1, is the sum of the completion time and the accumulated waiting time. ,in The largest task completion time among all tasks. Let be the cumulative waiting delay for task i. The second objective function value is the variance of the cumulative equipment hours for all equipment (including the four-way vehicle, the freight hoist, and the truck hoist). , Where J is the set of all equipment (including four-way vehicles, freight hoists, and truck hoists). Let j be the cumulative occupancy time of device j. This represents the average cumulative usage time of all devices. The third objective function is the negative database access hotspot entropy, calculated as f3 = −H({freq(l)}). , , where freq(l) is the frequency of access to storage location l or storage area l.

[0094] By embedding the spatiotemporal reservation table into the fitness evaluation process, the waiting delay caused by conflict resolution is directly included in the completion time target, eliminating the deviation between optimal planning and optimal execution. Simultaneously, the three objective functions correspond to completion efficiency, equipment balancing, and hotspot dispersion, respectively, achieving multi-objective trade-off optimization.

[0095] Furthermore, the conflict propagation graph uses unfrozen tasks in the task set as nodes, and the paths between tasks with spatiotemporal overlap, shared four-way vehicle edges, and shared elevator time windows as edges; the newly arrived task is used as a seed node.

[0096] In another embodiment of this scheme, the conflict propagation graph uses the unfrozen tasks within the current rolling planning window as the set of nodes. Unfrozen tasks are those that are more than 2 minutes away from the current time (i.e., the freezing threshold).

[0097] The edges of a conflict propagation graph are divided into three categories: The first type is the spatiotemporal overlap edge. For any two unfrozen tasks, if a path in the candidate path set of these two tasks overlaps within the estimated time window interval of a certain track node, then a spatiotemporal overlap edge is established between the two task nodes. Here, the estimated time window adopts a fast estimation method: the ideal travel time without considering collision waiting plus a preset elastic buffer time.

[0098] The second type is the shared four-way vehicle edge. For any two unfrozen tasks, if the feasible four-way vehicle sets of these two tasks intersect, then a shared four-way vehicle edge is established between the two task nodes. The feasible four-way vehicle set refers to the set of four-way vehicles that can execute the task, pre-selected based on the task's start and end point locations, storage depth, and current equipment load.

[0099] The third type is the shared elevator time window edge. For any two unfrozen tasks, if the estimated time windows of these two tasks on the same cargo elevator or the same vehicle elevator overlap, then a shared elevator time window edge is established between these two task nodes.

[0100] Using newly arrived outbound or inbound tasks as seed nodes, perform a breadth-first search in the aforementioned conflict propagation graph to identify the set of affected tasks.

[0101] The definition of the three types of edges comprehensively covers the main paths of conflict propagation in the four-way vehicle storage system. In particular, the shared four-way vehicle edge can capture the indirect impact chain caused by vehicle resource competition, ensuring the completeness of the identification of the set of affected tasks.

[0102] Furthermore, the legality repair includes, in sequence: location conflict repair, depth stack last-in-first-out order repair, double hoist layer switching constraint repair, and path pool mismatch repair.

[0103] In another embodiment of this scheme, the legality repair is performed in four steps in sequence, with each step repairing a typical type of illegal situation.

[0104] The first step is to resolve storage location conflicts. A reverse mapping is established from storage location identifiers to the task list occupying those locations. If a storage location is occupied by multiple tasks simultaneously, the task with the earliest arrival time is retained in ascending order. For the remaining conflicting tasks, the occupied storage locations are removed from the feasible storage location set of each conflicting task, and the nearest available storage location in the Manhattan distance is selected for reallocation. The Manhattan distance is the sum of the absolute differences between the row coordinates and column coordinates of two storage locations in the rack coordinate system.

[0105] The second step is to correct the Last-In-First-Out (LIFO) order of the depth stack. Group tasks by depth, where a single depth stack refers to a group of storage locations sharing the same inbound / outbound passage with increasing depth. Check the outbound order of tasks at different depths within the same depth stack. If a task at a greater depth (closer to the bottom of the stack) arrives later than a task at a lesser depth (closer to the top of the stack), this violates the LIFO constraint of the depth stack (i.e., goods at the bottom of the stack must be out before goods at the top). This constraint is corrected by swapping the storage location assignments of these two tasks, ensuring that the task corresponding to the deeper storage location has an earlier arrival time.

[0106] The third step is to repair the dual-lift constraint for floor switching. For tasks that require transfers between different floors (i.e., tasks that require floor switching), check whether the decision variables for the freight lift and the truck lift are the same. If they are the same, the dual-lift constraint is violated (the same lift cannot be used as both a freight lift and a truck lift simultaneously). From the remaining available lifts, sort them in ascending order of their current occupancy time, and select the lift with the shortest occupancy time to replace the truck lift's decision variable.

[0107] The fourth step is path pool mismatch repair. The location conflict repair in the first three steps may have altered the location decision variables for some tasks. When a task's location decision variables change, the original pre-computed path pool (generated based on old locations) for that task will no longer be valid. For these tasks whose locations have been changed, a new path pool is generated again using the Top-K A-Star algorithm based on the residual graph. The residual graph refers to the graph obtained after removing spatiotemporal nodes occupied by frozen tasks from the warehouse track network graph. The Top-K A-Star algorithm returns K candidate paths with increasing costs, where K is 6. The system sets the path index decision variable to 1 by default, pointing to the candidate path with the lowest cost.

[0108] The four-step repair operator processes constraints sequentially from basic to advanced based on constraint priority. Each repair step provides a valid input state for subsequent steps, ensuring that the final output chromosome individual satisfies all warehouse operation constraints. The path pool mismatch repair, as the final step, ensures that path information is updated promptly after a storage location change, avoiding inconsistencies between paths and storage locations.

[0109] This invention also provides a dynamic joint optimization system for four-way vehicle storage location-task-path based on rolling window and warm-start incremental solution, the specific technical solution of which is as follows: The acquisition module is used to: acquire the set of four-way vehicle storage route scheduling tasks within the current scrolling planning window; wherein, the set of four-way vehicle storage route scheduling tasks includes: incomplete old tasks and newly arrived outbound or inbound tasks; The construction module is used to: construct a conflict propagation graph, perform a breadth-first search in the conflict propagation graph, identify the complete set of affected tasks, and determine the tasks in the set of affected tasks as tasks to participate in re-optimization; The determination module is used to: determine the set of frozen tasks, perform adaptive population initialization, and generate the initial population; The generation module is used to iteratively optimize the initial population using a multi-objective differential evolution algorithm to generate offspring individuals; The repair module is used to: call the encoding repair operator to perform legality repair on the offspring individuals, evaluate the fitness of the repaired offspring individuals, and calculate the three objective function values ​​for each offspring individual; The selection module is used to: perform non-dominated sorting and crowding distance calculation on the union of the parent and offspring populations based on the three objective function values, and select individuals for the next generation population; The repeat module is used to: repeat the repair module and the selection module until the termination condition is met, and output the Pareto front solution set of the current rolling planning window; The update module is used to: select a solution from the Pareto front solution set as the landing solution, send the tasks belonging to the frozen task set in the landing solution to the warehouse control system for execution, and update the spatiotemporal reservation table.

[0110] It should be noted that the beneficial effects of the four-way vehicle storage location-task-path dynamic joint optimization system based on rolling window and warm-start incremental solution provided in the above embodiments are the same as those of the four-way vehicle storage location-task-path dynamic joint optimization method based on rolling window and warm-start incremental solution, and will not be repeated here. Furthermore, the system provided in the above embodiments is only illustrated by the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the system can be divided into different functional modules according to the actual situation to complete all or part of the functions described above. In addition, the system and method embodiments provided in the above embodiments belong to the same concept, and their specific implementation process is detailed in the method embodiments, and will not be repeated here.

[0111] like Figure 11 As shown, an electronic device 300 according to an embodiment of the present invention includes a processor 320 coupled to a memory 310. The memory 310 stores at least one computer program 330, which is loaded and executed by the processor 320 to enable the electronic device 300 to implement any of the above-mentioned methods. Specifically: The electronic device 300 can vary considerably due to differences in configuration or performance. It may include one or more processors 320 (Central Processing Units, CPUs) and one or more memories 310. The memories 310 store at least one computer program 330, which is loaded and executed by the processors 320 to enable the electronic device 300 to implement the four-way vehicle storage location-task-path dynamic joint optimization method based on scrolling window and warm-start incremental solution provided in the above embodiment. Of course, the electronic device 300 may also have wired or wireless network interfaces, a keyboard, and input / output interfaces for input / output. It may also include other components for implementing device functions, which will not be elaborated upon here.

[0112] An embodiment of the present invention provides a computer-readable storage medium storing at least one computer program, which is loaded and executed by a processor to enable a computer to implement any of the above-described methods.

[0113] Alternatively, the computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a compact disc read-only memory (CD-ROM), magnetic tape, a floppy disk, and an optical data storage device, etc.

[0114] In an exemplary embodiment, a computer program product or computer program is also provided, which includes computer instructions stored in a computer-readable storage medium. A processor of an electronic device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the electronic device to perform any of the methods described above.

[0115] It should be noted that the terms "first" and "second" in the specification and claims of this application are used to distinguish similar objects and represent a limitation on a specific order or sequence. Where appropriate, the order of use for similar objects can be interchanged so that the embodiments of this application described herein can be implemented in an order other than that shown or described.

[0116] Those skilled in the art will recognize that this invention can be implemented as a system, method, or computer program product. Therefore, this disclosure can be specifically implemented in the following forms: it can be entirely hardware, entirely software (including firmware, resident software, microcode, etc.), or a combination of hardware and software, generally referred to herein as a "circuit," "module," or "system." Furthermore, in some embodiments, the invention can also be implemented as a computer program product contained in one or more computer-readable media, which includes computer-readable program code.

[0117] Any combination of one or more computer-readable media may be used. A computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. A computer-readable storage medium can be, but is not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this document, a computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in connection with an instruction execution system, apparatus, or device.

[0118] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A dynamic joint optimization method for four-way vehicle storage location-task-path based on rolling window and warm-start incremental solution, characterized in that, include: S1. Obtain the set of four-way vehicle storage path scheduling tasks within the current scrolling planning window; wherein, the set of four-way vehicle storage path scheduling tasks includes: unfinished old tasks and newly arrived outbound or inbound tasks. S2. Construct a conflict propagation graph, perform a breadth-first search in the conflict propagation graph to identify the complete set of affected tasks, and determine the tasks in the set of affected tasks as tasks to participate in re-optimization. S3. Determine the set of frozen tasks, perform adaptive population initialization, and generate the initial population; S4. The initial population is iteratively optimized using a multi-objective differential evolution algorithm to generate offspring individuals; S5. Call the encoding repair operator to perform legality repair on the offspring individuals, evaluate the fitness of the repaired offspring individuals, and calculate the three objective function values ​​for each offspring individual; S6. Based on the three objective function values, perform non-dominated sorting and crowding distance calculation on the union of the parent and offspring populations, and select individuals for the next generation population. S7. Repeat steps S4 to S6 until the termination condition is met, and output the Pareto front solution set of the current rolling planning window. S8. Select a solution from the Pareto front solution set as the final solution, send the tasks belonging to the frozen task set in the final solution to the warehouse control system for execution, and update the space-time reservation table.

2. The four-way vehicle storage location-task-path dynamic joint optimization method based on rolling window and warm-start incremental solution as described in claim 1, characterized in that, The process of generating the initial population specifically includes: The rate of change is calculated based on the ratio of the number of tasks in the affected task set to the total number of tasks in the current rolling planning window; the seed inheritance ratio is dynamically calculated based on the rate of change; the historical Pareto front solution set is obtained; seed individuals are inherited from the historical Pareto front solution set according to the seed inheritance ratio, otherwise all initial population individuals are randomly generated; the gene segments belonging to the affected task set in the inherited seed individuals are perturbed and resampled, and the gene segments of the frozen task set are directly retained to generate some initial population individuals; the remaining initial population individuals are randomly generated to make the initial population size reach a preset value.

3. The four-way vehicle storage location-task-path dynamic joint optimization method based on rolling window and warm-start incremental solution as described in claim 1, characterized in that, The process of generating offspring individuals specifically includes: In each generation of evolution, differential evolution mutation and structure-preserving crossover operations are performed on each individual in the population to generate offspring individuals; wherein, the structure-preserving crossover operation is performed on a single task basis, exchanging the location decision variables, equipment decision variables, and path decision variables as a whole group.

4. The four-way vehicle storage location-task-path dynamic joint optimization method based on rolling window and warm-start incremental solution as described in claim 3, characterized in that, Also includes: After generating offspring individuals, a variable neighborhood local search is performed on the offspring individuals with a preset probability. The variable neighborhood local search includes: a location exchange neighborhood for swapping the location decision variables of two affected tasks, a device reallocation neighborhood for redistributing the device decision variables of a single task according to the inverse weight of the device load, and a path replacement neighborhood for replacing the path decision variables of a single task with the path in its path pool that minimizes the waiting delay of the spatiotemporal reservation table.

5. The four-way vehicle storage location-task-path dynamic joint optimization method based on rolling window and warm-start incremental solution as described in claim 1, characterized in that, The specific process for calculating the three objective function values ​​for each offspring individual is as follows: The path for each task in the offspring individual is simulated based on the spatiotemporal reservation table; when the time windows of multiple four-way vehicles overlap at the same track node, conflict resolution is performed according to task priority and waiting delay is accumulated. The three objective functions include: minimizing the sum of completion time and cumulative waiting delay, minimizing device load variance, and maximizing the entropy of warehouse access hotspots.

6. The four-way vehicle storage location-task-path dynamic joint optimization method based on rolling window and warm-start incremental solution as described in claim 1, characterized in that, The conflict propagation graph uses unfrozen tasks in the task set as nodes, and paths between tasks with spatiotemporal overlap, shared four-way vehicle edges, and shared elevator time windows as edges; the newly arrived task is used as a seed node, wherein the unfrozen task is a task whose time to freeze exceeds the current time threshold.

7. The four-way vehicle storage location-task-path dynamic joint optimization method based on rolling window and warm-start incremental solution as described in claim 1, characterized in that, The legality repair includes, in sequence: location conflict repair, depth stack last-in-first-out order repair, double hoist layer switching constraint repair, and path pool mismatch repair.

8. A four-way vehicle storage location-task-path dynamic joint optimization system based on rolling window and warm-start incremental solution, characterized in that, include: The acquisition module is used to: acquire the set of four-way vehicle storage route scheduling tasks within the current scrolling planning window; wherein, the set of four-way vehicle storage route scheduling tasks includes: incomplete old tasks and newly arrived outbound or inbound tasks; The construction module is used to: construct a conflict propagation graph, perform a breadth-first search in the conflict propagation graph, identify the complete set of affected tasks, and determine the tasks in the set of affected tasks as tasks to participate in re-optimization; The determination module is used to: determine the set of frozen tasks, perform adaptive population initialization, and generate the initial population; The generation module is used to iteratively optimize the initial population using a multi-objective differential evolution algorithm to generate offspring individuals; The repair module is used to: call the encoding repair operator to perform legality repair on the offspring individuals, evaluate the fitness of the repaired offspring individuals, and calculate the three objective function values ​​for each offspring individual; The selection module is used to: perform non-dominated sorting and crowding distance calculation on the union of the parent and offspring populations based on the three objective function values, and select individuals for the next generation population; The repeat module is used to: repeat the repair module and the selection module until the termination condition is met, and output the Pareto front solution set of the current rolling planning window; The update module is used to: select a solution from the Pareto front solution set as the landing solution, send the tasks belonging to the frozen task set in the landing solution to the warehouse control system for execution, and update the spatiotemporal reservation table.

9. An electronic device, characterized in that, The electronic device includes a processor coupled to a memory storing at least one computer program, which is loaded and executed by the processor to enable the electronic device to perform the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores at least one computer program, which is loaded and executed by a processor to enable the computer to perform the method as described in any one of claims 1 to 7.

Images