Intelligent warehouse digital twin dynamic modeling method

Abstract

This invention relates to the field of computer-aided modeling and physical system simulation technology, and discloses a dynamic modeling method for intelligent warehousing and distribution digital twins. The method involves acquiring the geometric parameters, spatial coordinates, motion constraint parameters, interface states, and state sampling data of the warehousing structure entity, the load-bearing entity, and the operating equipment entity; constructing a geometric model, a motion constraint model, and an event interaction model to generate a digital twin simulation model; converting the state sampling data into model events and constructing a directed event dependency graph; determining the local coupling domain based on the model residuals; forming a baseline simulation branch and a correction simulation branch; and verifying the model events, geometric boundaries, motion continuity, space occupancy mutual exclusion, and load transfer conservation. When load transfer conflicts occur, local state rollback, model event rearrangement, and playback are performed. The correction simulation branch with reduced model residuals and passing verification is selected and written into the digital twin simulation model to generate a model version.

Description

Technical Field

[0001] This invention relates to the field of computer-aided modeling and physical system simulation technology, and in particular to a dynamic modeling method for intelligent warehousing and distribution digital twins. Background Technology

[0002] With the application of computer simulation technology in the operation analysis of warehousing equipment, computer simulation models can be established based on the geometric parameters, spatial coordinates, motion constraint parameters, and interface states of the warehousing structure, supporting entities, and operating equipment entities. These models can then simulate the state changes of conveying equipment, stacking equipment, handling equipment, and supporting entities within the physical warehousing environment according to preset equipment operation logic. Compared to offline simulations that primarily rely on preset parameters, digital twin technology can also receive state sampling data from the smart warehousing and distribution physical environment, mapping the operating states of physical entities to corresponding model objects and continuously updating the digital twin simulation model.

[0003] For example, Chinese patent application CN115659693A discloses a three-dimensional warehouse digital twin system and method. The system acquires warehouse-related data through a data acquisition module, performs three-dimensional simulation based on warehouse physical parameters, equipment operation logic, and related data through a system simulation module, and presents the operating status of warehouse equipment and supporting entities through a visualization module to support the simulation, monitoring, and backtracking of the warehouse operation process.

[0004] In the physical scenario of smart warehousing and distribution, the physical equipment is typically in continuous motion, and load transfer occurs between interconnected pieces of equipment. Because different state acquisition interfaces may have different data sampling periods, transmission delays, and data arrival orders, the state sampling data received by the digital twin simulation model may not match the actual order of changes in the corresponding physical states. When state sampling data is written to a model object that has evolved to a subsequent state, model residuals may arise between the predicted state of the model object and the corresponding state sampling data, further leading to inconsistencies between the execution order of model events and the physical operation process.

[0005] Current digital twin simulation methods typically correct for model state deviations by adjusting relevant model parameters, restoring historical model states, or re-executing a simulation process. However, in the physical scenario of smart warehousing and distribution, there are simultaneous coupling relationships among the storage structure entities, load-bearing entities, and operating equipment entities, including interface contacts, space occupancy, motion constraints, and load transfers. A state change in one model object may propagate to multiple related model objects. If only the model objects that generate model residuals are corrected, the boundary states or space occupancy states between adjacent model objects may become inconsistent. If the entire digital twin simulation model is restored and re-simulated, model objects unaffected by model deviations will repeatedly participate in state restoration, model event sequencing, and simulation calculations.

[0006] Furthermore, during the transfer of a load entity from one model object to another, there is a correspondence between the load exit state and the load entry state. When the execution order of model events changes or the load transfer state is not fully recorded, situations may arise in the digital twin simulation model such as mismatched exit and entry events, or the same load corresponding to multiple occupancy states simultaneously. If model parameters are adjusted solely based on reducing model residuals without simultaneously verifying the timing of model events, spatial occupancy relationships, and load transfer relationships, the corrected model state may still be inconsistent with the actual operating state in the smart warehousing and distribution physical scenario.

[0007] Therefore, the main technical problem that the existing technology needs to solve is: when there is a deviation between the state sampling data of the smart warehousing and distribution physical scenario and the state of the digital twin simulation, or when there is a conflict between the load transfer state and the state of the digital twin, how to determine the range of the associated model affected by the deviation or conflict, and how to perform local dynamic correction of the digital twin simulation model under the condition of restoring and verifying the consistency of the model event timing, spatial occupancy relationship and load transfer relationship. Summary of the Invention

[0008] The purpose of this invention is to provide a dynamic modeling method for digital twins in smart warehousing and distribution, in order to solve the problems that when there is a deviation between the state sampling data and the state of the digital twin simulation in the physical scenario of smart warehousing and distribution, or when there is a conflict between the load transfer state and the state of the digital twin simulation, the correction range of the digital twin simulation model is difficult to determine, the execution order of model events is inconsistent, and the local correction results may not meet the physical constraints.

[0009] To achieve the aforementioned objectives, this invention provides a dynamic modeling method for intelligent warehousing and distribution digital twins, comprising: The system acquires entity identifiers, geometric parameters, spatial coordinates, motion constraint parameters, interface status, and status sampling data of the storage structure entities, load-bearing entities, and operating equipment entities in the smart warehousing and distribution physical scenario. It then constructs geometric models, motion constraint models, and event interaction models, and generates digital twin simulation models based on entity identifiers. The model instances corresponding to each entity are then identified as model objects.

[0010] State sampling data is converted into model events that act on the corresponding model objects. A directed event dependency graph is constructed based on the state transition relationships of the model objects and the interface contact relationships, space occupation relationships and load transfer relationships between the model objects. The model events are executed in the topological order of the directed event dependency graph. A state snapshot is recorded before the model events are executed, and the model residual between the predicted state and the state sampling data is determined.

[0011] The event replay interval is determined based on the model events that continuously generate model residuals. The model object corresponding to the model residual is determined as the starting model object. The model object associated with the starting model object is selected along the directed event dependency graph to form the initial local coupling domain.

[0012] Candidate model parameters are determined from the geometric parameters, motion constraint parameters, and event interaction parameters of the model objects within the initial local coupling domain, forming a baseline simulation branch and a correction simulation branch. In the baseline simulation branch and the correction simulation branch, model events within the event playback interval are replayed, and geometric boundary verification, motion continuity verification, space occupancy mutual exclusion verification, and load transfer conservation verification are performed.

[0013] The boundary state sequences of the boundary model objects in the benchmark simulation branch and the correction simulation branch are compared. When the boundary state sequences are inconsistent, the initial local coupling domain is expanded until the boundary state sequences are consistent, thus forming the local coupling domain and simulation boundary conditions.

[0014] The exit and entry events of the same load are combined into a load transfer event. When the exit and entry events are not paired or the same load corresponds to multiple occupied states, the event conflict interval is determined, the model objects in the local coupling domain are restored to the state snapshot before the event conflict interval, and the model events in the event conflict interval are reordered and replayed.

[0015] Select a corrected simulation branch that passes geometric boundary verification, motion continuity verification, space occupancy mutual exclusion verification, and load transfer conservation verification, and whose model residual is less than that of the baseline simulation branch. Generate local simulation state patches based on the corrected simulation branch, and then generate a new model version based on the parent model version and the local simulation state patches. If no corrected simulation branch meets the selection criteria, retain the current model version.

[0016] Compared with existing technologies, the above technical solution has the following advantages: 1. This invention constructs the storage structure entities, load-bearing entities, and operating equipment entities in the smart warehousing and distribution physical scenario into geometric models, motion constraint models, and event interaction models, respectively, and forms a digital twin simulation model according to the entity identifier association. This enables the model objects to simultaneously represent the spatial position, motion constraints, interface status, space occupancy status, and load transfer status of the entities, providing a unified model foundation for the computer simulation, state evolution, and dynamic correction of the smart warehousing and distribution physical system.

[0017] 2. This invention constructs a directed event dependency graph based on state transition relationships, interface contact relationships, space occupancy relationships, and load transfer relationships. It then determines the local coupling domain along the directed event dependency graph based on the initial model object corresponding to the model residual. This allows the model correction range to be limited to the associated model objects affected by the model state deviation, avoiding repeated state recovery and model event replay for unaffected model objects. This reduces the computational load and model event processing load during the dynamic correction process of the digital twin simulation model.

[0018] 3. This invention compares the boundary state sequences of boundary model objects in the baseline simulation branch and the correction simulation branch, and expands the current coupling domain when the boundary state sequences are inconsistent, and forms frozen boundary model objects and simulation boundary conditions when the boundary state sequences are consistent. This allows the local model correction results to maintain connection with the model state outside the local coupling domain at the boundary of the local coupling domain, reducing the possibility that local parameter adjustments will continue to propagate to irrelevant model objects and cause new simulation state deviations.

[0019] 4. This invention associates exit and entry events of the same load with load transfer tokens. When it detects that exit and entry events are not matched, the same load corresponds to multiple occupancy states, or the load transfer token has not been consumed, it determines the event conflict interval, performs state rollback, reorders and replays the model events on the local model object, so that the load transfer process in the digital twin simulation model meets the requirements of event timing, space occupancy and unique load occupancy, and reduces duplicate occupancy and discontinuous load states caused by asynchronous state sampling data.

[0020] 5. This invention forms a baseline simulation branch, a first disturbance simulation branch, a second disturbance simulation branch, and a candidate correction simulation branch. Combined with model residual comparison, geometric boundary verification, motion continuity verification, space occupancy mutual exclusion verification, and load transfer conservation verification, the correction simulation branch that satisfies physical constraints and reduces model residuals is selected and written into the digital twin simulation model. This enables model parameter correction not only to reduce the deviation between the predicted state and the state sampling data, but also to maintain the consistency between the model event execution process and the geometric conditions, motion conditions, and load transfer conditions of the smart warehousing and distribution physical scene, thereby improving the reliability of the dynamic modeling results of the digital twin simulation model. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of the overall process of the intelligent warehousing and distribution digital twin dynamic modeling method of the present invention; Figure 2 This is a schematic diagram of the process for local coupling domain expansion and boundary state sequence verification of the present invention; Figure 3 This is a schematic diagram of the process for load transfer conflict identification, model state rollback, and model event rearrangement and replay of the present invention. Figure 4 This is a schematic diagram of the process for correcting simulation branch generation, model residual comparison, and multidimensional physical constraint verification in this invention. Figure 5 This is a schematic diagram illustrating the mapping between the physical timeline of the state sampling data and the simulation timeline of the model events in this invention. Figure 6 This is a schematic diagram of the topological evolution of the directed event dependency graph constructed by the present invention based on state transition relationships, interface contact relationships, space occupancy relationships and load transfer relationships; Figure 7 This is a schematic diagram illustrating the state changes of the load transfer token when the carrier entity of the present invention performs load transfer between the source model object and the target model object. Figure 8 This is a schematic diagram of the data mapping structure between the intelligent warehousing and distribution physical scenario and the digital twin simulation model of the present invention; Figure 9 This is a comparison curve showing the change of the position model residuals of the baseline simulation branch and the correction simulation branch of this invention over simulation time. Detailed Implementation

[0022] The following is combined with Figures 1 to 9This paper provides a detailed explanation of a dynamic modeling method for digital twins in smart warehousing and distribution. Taking the pallet transfer process in a smart warehousing and distribution physical scenario as an example, this embodiment illustrates the process after state sampling data enters the digital twin simulation model, including model object generation, model event construction, directed event dependency graph formation, local coupling domain determination, load transfer conflict identification, local model state rollback, candidate model parameter correction, and model version generation.

[0023] In this embodiment, the smart warehousing and distribution physical scenario refers to a physical system composed of warehousing structure entities, load-bearing entities, operating equipment entities, and the physical connection relationships, space occupancy relationships, and load transfer relationships between these entities. The digital twin simulation model refers to a computer simulation model established in a computer according to entity identifiers, capable of representing the geometric state, motion state, interface state, and load state of each entity in the smart warehousing and distribution physical scenario, and capable of executing model state evolution based on state sampling data.

[0024] The number of entities, spatial dimensions, sampling period, and motion constraint parameters listed in this embodiment are used to illustrate the specific implementation process. While maintaining the logic of entity identification association, model event generation, directed event dependency graph construction, local coupling domain expansion, model event playback, and model version update, the corresponding parameters can be configured according to the actual dimensions of the smart warehousing and distribution physical scene, the type of operating equipment, and the status acquisition conditions. In this embodiment, the spatial occupancy grid size, quantization accuracy, residual event cluster merging conditions, model residual weights, normalization benchmark, and verification threshold are configured according to the actual dimensions of the smart warehousing and distribution physical scene, the type of operating equipment, and the status acquisition conditions.

[0025] like Figure 8 As shown, the smart warehousing and distribution physical scenario includes the storage structure entity, the load-bearing entity, and the operating equipment entity. The storage structure entity includes the shelf Shelf_A and the conveyor equipment Conveyor_C02. The second layer of the shelf Shelf_A is set as the storage location Shelf_A_Layer2, and the storage location Shelf_A_Layer2 is set as the connection interface IFA-02. The load-bearing entity is the pallet Pallet_889. The operating equipment entity includes the automated guided vehicle AGV-05 and the robotic arm AGV_Arm_02 installed on the automated guided vehicle AGV-05.

[0026] The length of the smart warehousing and distribution physical scenario is Width is Height is Using the ground point in the southwest corner of the smart warehousing and distribution physical scenario as the origin of a unified spatial coordinate system, and taking the direction from west to east as... The positive direction of the axis is taken as the direction from south to north. The positive direction of the axis is taken as the vertically upward direction. Positive direction of the axis.

[0027] The connection center coordinates of the storage location Shelf_A_Layer2 are: The IFA-02 connection interface is located at... The allowable deviation of the connection in the axial direction is ,exist The allowable deviation of the connection in the axial direction is ,exist The allowable deviation of the connection in the axial direction is The allowable deviation for docking is used to limit the maximum coordinate difference between the docking centers of the robotic arm AGV_Arm_02 and the storage location Shelf_A_Layer2 when they form an effective interface contact.

[0028] The external dimensions of the Automated Guided Vehicle (AGV-05) are as follows: The initial center coordinates are Rated operating speed is Maximum operating speed is The current braking acceleration is The initial coordinates of the end effector of the robotic arm AGV_Arm_02 are: End-effector along The maximum telescopic distance in the axial direction is ,along The maximum lifting distance in the axial direction is The dimensions of Pallet_889 are as follows: The load mass is .

[0029] Configure entity identifiers for each entity. The entity identifier for storage location Shelf_A_Layer2 is OBJ-RACK-A-L2, the entity identifier for automated guided vehicle AGV-05 is OBJ-AGV-05, the entity identifier for robotic arm AGV_Arm_02 is OBJ-ARM-02, the entity identifier for conveyor equipment Conveyor_C02 is OBJ-CONV-C02, and the entity identifier for pallet_889 is OBJ-PALLET-889.

[0030] The entity identifier refers to a data index used to uniquely identify an entity between the smart warehousing and distribution physical scenario and the digital twin simulation model. Based on the entity identifier, the digital twin simulation server assigns geometric parameters, spatial coordinates, motion constraint parameters, interface states, and state sampling data from different state acquisition interfaces to the corresponding entities, preventing the mixing of data from different entities.

[0031] The status sampling data is generated by the status acquisition interface in the smart warehousing and distribution physical scenario. Specifically, the physical location is collected by an ultra-wideband positioning device, with a sampling frequency of [missing information]. The operating speed of the AGV-05 automated guided vehicle is collected by a wheel speed encoder, with a sampling frequency of [missing information]. The position of the end effector of the robotic arm AGV_Arm_02 is acquired by a joint encoder, with a sampling frequency of [missing information]. The interface status of the IFA-02 connector and the load presence status of the Pallet_889 are collected by a proximity switch and a load detection device, respectively, with a time resolution of [missing information]. .

[0032] Each state sampling data point includes at least the acquisition sequence number, physical timestamp, reception timestamp, entity identifier, state type, and state value. For location state sampling data, the state value includes... Axis coordinates axis coordinates and Axis coordinates; for motion state sampling data, state values ​​include running speed, motion direction, and braking status; for interface state sampling data, state values ​​include locked status, released status, and interface contact status; for load state sampling data, state values ​​include unoccupied status, occupied status, and transfer status.

[0033] The digital twin simulation server constructs geometric models, motion constraint models, and event interaction models based on entity identifiers.

[0034] The geometric model refers to a set of model data used to represent the external outline, spatial coordinates, geometric motion boundaries, and docking interface positions of the corresponding entity. For example, the geometric model of the automated guided vehicle (AGV-05) includes length... ,width ,high The geometric bounding volume and the spatial coordinates of the center of the geometric bounding volume; the geometric model of the storage location Shelf_A_Layer2 includes the geometric bounding volume of the docking area, the coordinates of the docking center, and the allowable docking deviation.

[0035] The motion constraint model refers to a set of model data used to represent the allowed motion states and conditions for changes in motion states of the corresponding entity. The motion constraint model includes the range of running speed, acceleration range, braking acceleration, allowed motion direction, running path, joint motion range, and the allowed range of state changes between adjacent simulation time steps.

[0036] The event interaction model refers to a set of model data used to represent the triggering conditions, interface conditions, and state transition rules when physical interactions occur between different model objects. The event interaction model includes at least interface contact relationships, space occupancy relationships, and load transfer relationships. Interface contact relationships represent the interface identifier, location conditions, and state conditions when two model objects form a valid connection; space occupancy relationships represent the process of a model object occupying, releasing, and requesting space occupancy grids; and load transfer relationships represent the state transition process of the same load exiting one model object and entering another.

[0037] Associate the geometric model, motion constraint model, and event interaction model corresponding to the same entity identifier to form a model instance of the corresponding entity, and determine the model instance as a model object.

[0038] In this embodiment, a model object refers to a simulation instance in a digital twin simulation model that corresponds to a physical entity and can independently record its state and execute model events. Each model object includes at least a geometric state record, a motion state record, an event interface record, and a state version record. The geometric state record is used to store spatial coordinates and geometric boundaries; the motion state record is used to store running speed, acceleration, and braking states; the event interface record is used to store interface states, space occupancy states, and load states; and the state version record is used to store the state of the model object under different simulation timestamps and different model versions.

[0039] The various model objects are combined according to the spatial and interface connection relationships in the physical scenario of smart warehousing and distribution to generate a digital twin simulation model. The digital twin simulation model is not just a three-dimensional model for displaying the shape of the entity, but a computer simulation model of the physical system that can receive state sampling data, generate model events, execute model state evolution, output predicted states, and perform state replay.

[0040] After a model event is executed, the initial state of the corresponding model object is taken from its state before the event. The motion constraint model updates the spatial coordinates, velocity, and motion state of the model object based on the motion state changes corresponding to the model event, the current running speed, acceleration, braking acceleration, running path, and simulation time step. The geometric model updates the position of the geometric bounding volume and the space occupied by the model object based on the updated spatial coordinates. The event interaction model updates the interface state, space occupancy state, and load state of the model object based on the model event type, interface contact relationship, space occupancy relationship, and load transfer relationship. The updated spatial coordinates, motion state, interface state, space occupancy state, and load state together constitute the predicted state of the model object.

[0041] like Figure 1As shown, after the automated guided vehicle (AGV)-05 moves to the storage location Shelf_A_Layer2, the robotic arm AGV_Arm_02 makes contact with the docking interface IFA-02 and transfers the pallet Pallet_889 from the shelf model object to the robotic arm model object. Some of the state sampling data generated during this process are shown in Table 1.

[0042] Table 1. State Sampling Timing Data Table

[0043] The digital twin simulation server reads the entity identifier and state type from the state sampling data, and compares the current state value in the state sampling data with the previous state value of the corresponding model object. When the current state value is different from the previous state value, a model event is generated that acts on the corresponding model object.

[0044] The model event refers to a data record used to transform one or more model objects from a pre-execution state to a post-execution state. Each model event includes at least a model event identifier, event type, entity identifier of the affected model object, physical timestamp, simulation timestamp, pre-execution state, and post-execution state.

[0045] The digital twin simulation server determines the actual occurrence time of state sampling data based on physical timestamps. When multiple state sampling data have the same physical timestamp, the order of the corresponding state sampling data is determined according to the acquisition sequence number. When the entity identifier, physical timestamp, state type, and state value are all the same and the acquisition sequence number is repeated, the corresponding state sampling data is identified as duplicate state sampling data, and model events are not generated repeatedly.

[0046] The event sequence Seq_1024 corresponds to the transport equipment entering the deceleration model event, the event sequence Seq_1025 corresponds to the shelf interface releasing the model event, the event sequence Seq_1026 corresponds to the pallet exiting the model event, the event sequence Seq_1027 corresponds to the robotic arm interface contacting the model event, the event sequence Seq_1028 corresponds to the pallet entering the model event, and the event sequence Seq_1029 corresponds to the transport equipment completing the braking model event.

[0047] The exit and entry events of the same carrier entity between two interconnected model objects are combined into a cross-interface event pair. In this embodiment, the pallet exit event of pallet_889 exiting the shelf model object and the pallet entry event of pallet_889 entering the robotic arm model object are combined into a cross-interface event pair.

[0048] The cross-interface event pairs represent two corresponding model events used to indicate the transfer of the same load across a physical interface. Specifically, the exit model event records the state change of the load leaving the source model object, and the enter model event records the state change of the load entering the target model object.

[0049] like Figure 5 As shown, a time anchor sequence between the physical time axis and the simulation time axis is established based on the physical timestamps of multiple cross-interface event pairs.

[0050] The physical timeline represents the actual sequence of occurrence of state sampling data in the smart warehousing and distribution physical scenario, while the simulation timeline represents the execution sequence of model events in the digital twin simulation model. Time anchors are used to establish a correspondence between a physical timestamp and a simulation timestamp.

[0051] Each physical timestamp is converted to a second value relative to the same time base. In this embodiment, the physical timestamp 2026-04-12 10:00:05.000 corresponds to the simulation timestamp. The physical timestamp 2026-04-12 10:00:06.000 is mapped to the simulation timestamp. State sampling data located between two adjacent time anchor points are mapped to the simulation time axis according to the following formula:

[0052] In the formula, This represents the simulation timestamp corresponding to the state sampling data, in seconds. The value in seconds is the result of converting the physical timestamp of the state sampling data. This represents the simulation timestamp corresponding to the previous time anchor point, in seconds. This represents the simulation timestamp corresponding to the next time anchor point, in seconds. This represents the second value obtained by converting the physical timestamp corresponding to the previous time anchor point, in seconds. This represents the second value obtained by converting the physical timestamp corresponding to the next time anchor point, in seconds.

[0053] According to the mapping relationship, the simulation timestamp corresponding to the tray exit model event is: When the digital twin simulation server receives a tray exit model event, the digital twin simulation model has already executed up to the simulation timestamp. Since the simulation timestamp corresponding to the tray exit model event is earlier than the maximum simulation timestamp among the currently executed model events, the tray exit model event is determined to be a delayed model event.

[0054] The delayed model event refers to a model event that is received by the digital twin simulation server after the digital twin simulation model has been executed to the next simulation timestamp, and whose corresponding simulation timestamp is located before the current simulation timestamp.

[0055] Subsequent model events reachable from delayed model events along the directed event dependency graph are identified as model events to be replayed. Delayed model events and model events to be replayed constitute the candidate event set for incremental topological sorting and subsequent model event replay.

[0056] like Figure 6 As shown, each model event is set as an event node in a directed event dependency graph. The directed event dependency graph is a data structure composed of event nodes and directed edges. Event nodes are used to represent model events, and directed edges are used to represent the sequential execution relationship that must be satisfied between two model events.

[0057] In this embodiment, the event of the transportation equipment entering the deceleration model is set as an event node. Set the event of the transportation equipment completing braking model as an event node. Set the shelf interface release model event as an event node. Set the robotic arm interface contact model event as an event node. Set the tray exit model event as an event node. Set the tray entry event as an event node. .

[0058] The directed edges in a directed event dependency graph include state transition edges, interface contact edges, space occupancy edges, and load transfer edges.

[0059] State transition edges are used to represent the sequential relationship between two consecutive state changes of the same model object. In this embodiment, the automated guided vehicle AGV-05 should first enter the deceleration state before completing braking. Therefore, the deceleration model event occurs when the vehicle enters the deceleration state. And the transportation equipment completes the braking model event Establish state transition edges between them.

[0060] The interface contact edge is used to represent the prerequisite events required to form an interface contact state. In this embodiment, the docking interface IFA-02 should first change from a locked state to a released state before the robotic arm model object can form interface contact with the docking interface IFA-02. Therefore, the release model event at the shelf interface is used. Contact Model Events with Robotic Arm Interface An interface contact edge is established between them; only after the robotic arm model object forms an interface contact can the pallet_889 exit the shelf model object. Therefore, in the robotic arm interface contact model event... Tray exit model event Establish an interface contact edge between them.

[0061] Space occupancy edges represent the sequential relationship where one model event releases space before another model event can request to occupy that space. The digital twin simulation server divides the smart warehousing and distribution physical scenario into space occupancy grids, with each grid having a size of [size missing]. When a model event releases the corresponding space-occupying grid and another model event requests to occupy the same space-occupying grid, a space-occupying edge is established between the model event that released the space-occupying grid and the model event that requested to occupy the space-occupying grid.

[0062] In this embodiment, the robotic arm AGV_Arm_02 can only enter the docking space after the automated guided transport device AGV-05 completes braking. Therefore, the braking model event of the transport device is completed... Contact Model Events with Robotic Arm Interface Establish space-occupying edges between them.

[0063] Load transfer edges are used to represent the sequential relationship between exiting and entering the model for the same load. The model object corresponding to the load output end is designated as the source model object, and the model object corresponding to the load receiving end is designated as the target model object. In this embodiment, the shelf model object is the source model object, and the robotic arm model object is the target model object. The pallet exits the model event... and tray entry model event Establish load transfer edges between them.

[0064] After the delayed model events are inserted into the directed event dependency graph, incremental topological sorting is performed on the delayed model events and the model events to be replayed.

[0065] During incremental topology sorting, the in-degree of each visited event node is recorded. The in-degree of an event node refers to the number of directed edges pointing to that event node. Event nodes with an in-degree of zero are added to the unsorted queue. Event nodes are output sequentially from the unsorted queue, and directed edges pointing from output event nodes to their successors are deleted. When the in-degree of a successor event node becomes zero, the corresponding successor event node is added to the unsorted queue, thus forming the updated topology order.

[0066] When a delayed model event is inserted into a directed event dependency graph, a directed cycle is determined to be formed in the directed event dependency graph when the newly established state transition edge, interface contact edge, space occupancy edge, or load transfer edge forms a closed path with the original directed path in the directed event dependency graph.

[0067] When a directed event dependency graph forms a directed cycle, a strongly connected subgraph search is performed. A strongly connected subgraph is a set of event nodes in which any two event nodes can be reached from each other via a directed path.

[0068] For each event node, the strongly connected subgraph search records its first visit number and earliest backtracking number. The first visit number indicates the order in which the event node was first searched, and the earliest backtracking number indicates the visit number of the earliest visited event node that can be reached from that event node along the directed edges. Visited event nodes that have not yet been searched are pushed onto the traversal stack.

[0069] When the first access sequence number of an event node is equal to the earliest backtracking sequence number, event nodes are popped sequentially from the top of the traversal stack until the event node is popped, forming a strongly connected subgraph. For multiple strongly connected subgraphs containing the newly formed directed edges, the strongly connected subgraph with the fewest event nodes is selected as the minimum strongly connected subgraph.

[0070] In a set of delayed model event data in this embodiment, event nodes were detected. Event Nodes and event nodes The closed directed path, therefore the determined minimum strongly connected subgraph includes event nodes. Event Nodes and event nodes Read the simulation timestamps of each model event within the smallest strongly connected subgraph, and determine the interval between the earliest and latest simulation timestamps as the candidate event conflict interval.

[0071] The digital twin simulation server executes model events in the topological order of the directed event dependency graph, and records a snapshot of the state of the corresponding model object before executing each model event.

[0072] A state snapshot refers to a data record that saves the current state of a corresponding model object before the execution of a model event. A state snapshot includes a state snapshot identifier, model version identifier, model object entity identifier, spatial coordinates, running speed, motion state, interface state, space occupancy state, load occupancy state, load transfer token state, and simulation timestamp. State snapshots are used to restore the model object to its state before the execution of the model event if the execution order of the model event changes or if the calibration simulation branch fails verification.

[0073] After the model event is executed, the digital twin simulation model obtains the predicted state of the model object according to the aforementioned model state evolution process. The predicted state is compared with the state sampling data with the corresponding physical timestamp to determine the location model residual, state transition time model residual, and space occupancy model residual, respectively.

[0074] The location model residuals are determined according to the following formula:

[0075] In the formula, This represents the location model residuals, in meters. , and Representing the predicted states respectively Axis coordinates axis coordinates and Axis coordinates, in meters; , and These represent the state sampling data respectively. Axis coordinates axis coordinates and Axis coordinates, in meters.

[0076] Simulated timestamps The predicted spatial coordinates of the transportation equipment model object are The spatial coordinates in the state sampling data are Therefore, we can conclude that:

[0077] The residuals of the state transition time model are determined according to the following formula:

[0078] In the formula, This represents the residual of the state transition time model, in seconds. The timestamp, in seconds, indicates the time when the corresponding state in the digital twin simulation model changes. This represents the simulation timestamp of state changes after the state sampling data is mapped to the simulation time axis, in seconds.

[0079] The residuals of the space occupancy model are determined according to the following formula:

[0080] In the formula, This represents the residual of the space occupancy model; This indicates the number of space occupancy grids whose predicted space occupancy state does not match the corresponding space occupancy state in the state sampling data; This indicates the total number of grid cells used in the comparison.

[0081] To compare the direction of model residual changes under different parameter conditions across multiple perturbation simulation branches, and further determine the composite model residuals:

[0082] In the formula, Represents the residuals of the integrated model; Indicates the residual weights of the location model; Represents the residual weights of the state transition time model; This represents the residual weights of the space occupancy model; The reference value for the normalized residuals of the location model is expressed in meters. This represents the normalized baseline for the residuals of the state transition time model, in seconds.

[0083] In this embodiment, the location model residual weights The residual weight of the state transition time model is 0.5. The residual weight of the space occupancy model is 0.3. The baseline for the normalized residuals of the location model is 0.2. for State transition time model residual normalization benchmark for The sum of the three weights is .

[0084] Based on the reachability relationships in the directed event dependency graph and the spatial adjacency relationships between model objects, the continuously generated model residuals and their corresponding model events are grouped into residual event clusters.

[0085] The residual event cluster refers to a group of model residuals and model events that are temporally continuous, mutually reachable in the directed event dependency graph, and whose corresponding model objects are physically adjacent. In this embodiment, when there are no more than three reachable paths between two adjacent model events, and the simulation time interval between two model events is no greater than [missing information], [the cluster is defined as follows]. Furthermore, the minimum distance between the shared space occupied by the corresponding model objects and the mesh or geometric bounding volume is not greater than [missing value]. When the time comes, the corresponding model residuals will be merged into the same residual event cluster.

[0086] Model residuals are continuously generated between the event when the transportation equipment enters the deceleration model and the event when the transportation equipment completes the braking model. The earliest event when the transportation equipment enters the deceleration model and the latest event when the transportation equipment completes the braking model are defined as the event replay interval, and the transportation equipment model object is defined as the starting model object.

[0087] The event replay interval refers to the simulation time range determined for re-executing a set of interrelated model events. The start point of the event replay interval is the model event with the earliest simulation timestamp in the residual event cluster, and the end point is the model event with the latest simulation timestamp in the residual event cluster.

[0088] like Figure 2As shown, the process extends forward along the outgoing edges of the directed event dependency graph from the initial model object, selecting robotic arm model objects, pallet model objects, and shelf model objects whose states are affected by the model events corresponding to the initial model object; and extends backward along the incoming edges of the directed event dependency graph from the selected successor model objects, supplementing the preceding shelf interface model objects and path occupancy model objects.

[0089] Forward expansion along outgoing edges is used to determine which model objects the state change of the initial model object may propagate to; backward expansion along incoming edges is used to supplement the preceding model objects on which the state change of the subsequent model object depends.

[0090] The initial model object, subsequent model objects, and preceding model objects are defined as the current coupling domain. The current coupling domain is a temporary set of model objects formed during the expansion of the local coupling domain, and the edge of the current coupling domain may still contain model objects affected by changes in the candidate model parameters.

[0091] Within the current coupling domain, a baseline simulation branch that retains the current model parameters and a correction simulation branch that changes the candidate model parameters are formed.

[0092] The baseline simulation branch refers to a simulation branch that keeps the model parameters in the current digital twin simulation model unchanged and executes the playback according to the redefined model event sequence. The calibration simulation branch refers to a simulation branch that changes only at least one candidate model parameter while keeping other model parameters and simulation boundary conditions unchanged.

[0093] The two simulation branches replay the model events within the same event replay interval. After the replay is complete, the boundary state sequence of each boundary model object at the edge of the current coupled domain is obtained in the two simulation branches.

[0094] The boundary model object refers to a model object located at the edge of the current coupling domain and connected to a model object outside the current coupling domain by a directed edge. The boundary state sequence refers to multiple state records of the boundary model object arranged according to simulation timestamps within the event playback interval.

[0095] The boundary state sequence includes quantized spatial coordinates, quantized running speed, quantized simulation timestamp, interface status word, space occupancy status word, and load status word. Before comparing the boundary state sequences, the spatial coordinates are... Quantify the running speed according to Quantization is performed, and the simulation timestamps are categorized according to... Quantify it.

[0096] If the quantized spatial coordinates, quantized running speed, quantized simulation timestamp, interface status word, space occupancy status word, and load status word of the same boundary model object in two simulation branches are identical, then the corresponding two sets of boundary state sequences are determined to be consistent.

[0097] When two sets of boundary state sequences of the same boundary model object are consistent, it indicates that the changes in the candidate model parameters have not propagated outside the boundary model object, and the boundary model object is identified as a frozen boundary model object.

[0098] The frozen boundary model object refers to a model object whose boundary state sequence is consistent in the baseline simulation branch and the correction simulation branch, and which can serve as a local simulation boundary.

[0099] When the two sets of boundary state sequences are inconsistent, it indicates that the changes in the candidate model parameters have propagated to the boundary model object. The model object that is directly connected to the boundary model object through the directed event dependency graph will be incorporated into the current coupling domain, and the model event replay and boundary state sequence comparison will be re-executed.

[0100] The digital twin simulation model comprises 146 model objects. After two expansions, all boundary model objects at the edge of the current coupling domain are identified as frozen boundary model objects, and the current coupling domain consisting of 12 model objects is identified as the local coupling domain.

[0101] The local coupling domain refers to a set of model objects that includes the initial model object and associated model objects affected by changes in the parameters of the corresponding model residuals or candidate models, with the frozen boundary model object as the boundary.

[0102] The boundary state sequence of each frozen boundary model object is set as the simulation boundary condition. The simulation boundary condition is used to limit the state connection relationship between the local coupled domain and the model objects outside the local coupled domain when performing model event playback within the local coupled domain.

[0103] When pallet_889 is transferred from the shelf model object to the robotic arm model object, the event interaction model generates a load transfer token Token_Pallet_889.

[0104] The load transfer token is used to uniquely associate the exit and entry events of the same load with the model event, and records the data of the load transfer status between the source model object, the interface transfer area, and the target model object.

[0105] The load transfer token includes the load identifier Pallet_889, the source model object entity identifier OBJ-RACK-A-L2, the target model object entity identifier OBJ-ARM-02, the interface identifier IFA-02, and the transfer sequence number Transfer_00036. The load identifier and the transfer sequence number together constitute a unique index for the load transfer token.

[0106] like Figure 7As shown, when Pallet_889 exits the shelf model object, an exit model event associated with the load transfer token is generated, and the load occupancy status in the source model object is updated from occupied to in progress.

[0107] When Pallet_889 enters the robotic arm model object, the load transfer token is read according to the combined unique index, and the load transfer token is changed from the transit state to the consumption state in an atomic state transition manner. At the same time, an entry model event is generated, and the load occupancy state in the target model object is updated to the occupied state.

[0108] The atomic state transition refers to reading the load transfer token and converting the load transfer token to a consumable state as an indivisible data operation. No other entering model event is allowed to read the same load transfer token before this data operation is completed.

[0109] After an inbound model event completes, the load transfer token is converted to a consumed state. Load transfer tokens that have been converted to a pending or consumed state cannot be read again by other inbound model events.

[0110] Table 2 Load Transfer Token Status Change Table

[0111] The digital twin simulation server uses the candidate event conflict interval as the initial retrieval range for abnormal load transfer events, and detects entry events that are not associated with load transfer tokens, entry events that are repeatedly associated with the same load transfer token, load transfer tokens that have not been converted to a consumed state, source model objects and target model objects that are simultaneously in an occupied state, and interface identifiers that are inconsistent between exit events and entry events.

[0112] When any of the aforementioned situations is detected, the corresponding exit model event and enter model event will be combined into an abnormal load transfer event.

[0113] Because the pallet exit model event arrives late, the pallet entry model event has already caused the robotic arm model object to enter the occupied state before the pallet exit model event is inserted into the directed event dependency graph, while the shelf model object remains in the occupied state. Therefore, the same pallet_889 corresponds to two occupied states at the same time, and the corresponding load transfer event is identified as an abnormal load transfer event.

[0114] Tracing back from the abnormal load transfer event along the directed event dependency graph, model events that have passed the load transfer conservation check are selected as candidate starting events. When multiple candidate starting events exist, the model event with the latest simulation timestamp and closest to the abnormal load transfer event in the topological order is selected as the starting event of the event conflict interval.

[0115] Tracing forward from the abnormal load transfer event along the directed event dependency graph, the model event that corresponds to the load forming a unique occupancy state is selected as the candidate termination event. When multiple candidate termination events exist, the model event that appears first in the topological order is selected as the termination event of the event conflict interval.

[0116] The event conflict interval refers to the simulation time range from the last model event in which the load state still satisfies the load transfer conservation condition to the first model event in which the load returns to the unique occupancy state.

[0117] like Figure 3 As shown, the model objects within the local coupling domain are restored to a snapshot of their state before the starting event, while the state of the model objects outside the local coupling domain remains unchanged.

[0118] During the recovery process, model objects are recovered in reverse order of the directed event dependency graph topology. First, the states formed by subsequent model events are undone, then the states formed by preceding model events are undone. The recovered data includes spatial coordinates, running speed, motion state, interface state, space occupancy state, load occupancy state, and load transfer token state.

[0119] The incremental topology sorting is re-executed for model events within the event conflict interval, and the model events of shelf interface release, robotic arm interface contact, pallet exit, pallet entry, and transportation equipment braking are replayed in sequence.

[0120] After playback, the Pallet_889 tray corresponds to only one occupied state or one transitioning state under any simulation timestamp.

[0121] Compare the replayed state of the boundary model object with the state of the boundary model object at the corresponding moment before rollback. If they match, complete the replay of the event conflict interval; if they do not match, incorporate the model object directly connected to the corresponding boundary model object through the directed event dependency graph into the local coupling domain, and re-execute state restoration, model event sorting, and model event replay.

[0122] When determining candidate model parameters, geometric parameters, motion constraint parameters, and event interaction parameters are distinguished based on the change pattern of model residuals in the residual event cluster.

[0123] When the model residuals before and after a model object enters a motion state maintain a positional deviation in the same direction, the corresponding geometric parameters are determined as candidate model parameters. This same-direction positional deviation means that the positive and negative directions of the position model residuals on the same coordinate axis remain unchanged during multiple state sampling processes.

[0124] For example, the storage location Shelf_A_Layer2 follows the same path during multiple pallet connection processes. Generate in the positive direction of the axis to The fixed positional deviation will affect the IFA-02 connector. The axis coordinates are determined as candidate model parameters.

[0125] When the model residuals accumulated during the motion of the model object accumulate during the motion process and remain after the model object exits the motion state, the corresponding motion constraint parameters are determined as candidate model parameters.

[0126] In this embodiment, the position model residual of the transportation equipment model object before entering the deceleration state is: During deceleration, the position model residual is changed from Increase to After braking is completed, the position model residual remains at [value]. to Therefore, the braking acceleration of the automated guided vehicle AGV-05 was determined as a candidate model parameter.

[0127] When a state transition time model residual is generated between the exit model event and the enter model event in a cross-interface event pair, and the state transition time model residual no longer accumulates after the enter model event, the corresponding event interaction parameter is determined as a candidate model parameter.

[0128] For example, in another set of connection data, tray exit model events and tray entry model events occur consecutively. The fixed time deviation, and the model event time after the tray enters the model event does not continue to shift, so the interface response delay of the connection interface IFA-02 is determined as a candidate model parameter.

[0129] like Figure 4 As shown, taking the braking acceleration of the automated guided vehicle (AGV-05) as an example, the current braking acceleration is... The initial disturbance step size is determined according to 5% of the allowable adjustment range of braking acceleration. This embodiment adopts... .

[0130] respectively and The first perturbation simulation branch is formed with parameter values ​​increasing sequentially, so as to... and This forms a second perturbation simulation branch with successively decreasing parameter values.

[0131] Each disturbance simulation branch keeps other model parameters and simulation boundary conditions unchanged, and replays the model events contained in the same residual event cluster. The position model residual, state transition time model residual, space occupancy model residual, and combined model residual are calculated for each disturbance simulation branch.

[0132] When the difference between the residuals of the integrated model of two adjacent perturbation simulation branches changes from negative to positive, or from positive to negative, it is determined that the direction of change of the integrated model residuals has reversed, and the parameter interval between the corresponding two candidate model parameters is determined.

[0133] The verification data for each simulation branch are shown in Table 3.

[0134] Table 3 Simulation Branch Validation and Model Residual Table

[0135] Braking acceleration is Adjust to At that time, the residuals of the integrated model changed from decreasing to increasing, therefore... to The parameter range is defined. Within the parameter range, using... , and Each of these forms a simulation branch to be selected for correction.

[0136] Geometric boundary verification is used to detect whether the geometric bounding volume of model objects exceeds the allowable movement boundary, and whether geometric penetration occurs between model objects that are not allowed to contact each other. The minimum distance between the robotic arm model object and the shelf model object in the non-connection area is not less than [a certain value]. The geometric boundary check passes when none of the model objects exceed the geometric motion boundary.

[0137] Motion continuity verification uses The simulation time step is used to calculate the velocity change between two adjacent simulation time steps. The velocity change is considered complete when the difference between the velocity change and the theoretical velocity change resulting from the corresponding braking acceleration within the simulation time step is no greater than [value missing]. The motion continuity check passes when the running speed does not exceed the maximum running speed.

[0138] Space occupancy mutual exclusion check is used to detect whether the same space occupancy grid is occupied by two model objects that are not allowed to coexist at the same simulation timestamp. When the interface is active, the source model object and the target model object are allowed to share the preset interface space occupancy grid; other space occupancy grids are not allowed to be occupied by two model objects simultaneously.

[0139] The load transfer conservation check is used to detect whether each load transfer token is associated with only one exit model event and one enter model event, and to detect whether the same load has only one occupied state or one transfer state at any simulation timestamp.

[0140] The earliest model event in the residual event cluster is determined as the first start event, and the latest model event is determined as the first end event. The first event replay interval is formed from the first start event to the first end event. In this embodiment, the first event replay interval is... to .

[0141] The first event replay interval is used to verify the direct impact of candidate model parameters on the current residual event cluster.

[0142] The first termination event is designated as the second starting event. Starting from the first termination event, a forward search is performed along the directed event dependency graph to read the subsequent model events corresponding to each frozen boundary model object. The boundary state change model event with the earliest simulation timestamp is then designated as the second termination event. In this embodiment, the second termination event occurs at... ,Depend on to This forms the second event replay interval.

[0143] The second event replay interval is used to verify whether the candidate model parameters continue to affect the frozen boundary model object after the residual event cluster ends, thereby avoiding the reduction of model residuals only in the current residual interval and the generation of new boundary state deviations in subsequent model events.

[0144] The candidate correction simulation branch first performs geometric boundary verification, motion continuity verification, space occupancy mutual exclusion verification, and load transfer conservation verification within the first event playback interval. The candidate correction simulation branch that passes all verifications continues to play back the second event playback interval and performs the aforementioned verifications again.

[0145] Branch_Candidate_02 passed all checks in both the first and second event replay intervals. The residuals of the position model, state transition time model, and space occupancy model of the candidate correction simulation branch were all less than the residuals of the model corresponding to the baseline simulation branch. Furthermore, the state of the frozen boundary model object at the time corresponding to the second termination event was consistent with the corresponding state in the baseline simulation branch. Therefore, Branch_Candidate_02 was determined as the correction simulation branch.

[0146] The braking acceleration in the calibration simulation branch is The current model parameters are Therefore, the parameter increment of the candidate model is .

[0147] For the first in the local coupling domain The model object in the first The continuous state at each sampling time point, the local model object state increment is determined according to the following formula:

[0148] In the formula, Indicates the first The model object in the first The continuous state increment of the local model object at each sampling time; This represents the continuous state values ​​in the correction simulation branch; This represents the continuous state values ​​in the baseline simulation branch.

[0149] The continuous state includes spatial coordinates, running speed, and model event execution time. For continuous states containing multiple components, the local model object state increment is calculated for each state component. For discrete states such as interface states, space occupancy states, and load occupancy states, instead of subtracting values, state replacement pairs are formed by combining the original state value and the replacement state value. For example, changing the space occupancy state from "occupied" to "released" is recorded as a discrete state replacement pair.

[0150] The current model version before writing the candidate model parameter increment and the local model object state increment is determined as the parent model version, and the parent model version is identified as MV_20260412_100005_03.

[0151] The parent model version refers to the digital twin simulation model version before the local model correction results are written. Subsequent model versions are based on the parent model version and have local simulation state patches written in.

[0152] The incremental parameters of candidate models, the continuous state increments of local model objects, discrete state replacement pairs, the set of model events in the event playback interval, and the corresponding topological order of model events are encapsulated into a local simulation state patch SP_PALLET889_0007.

[0153] The local simulation state patch refers to a data set that records only the model parameters and model object states that have changed relative to the parent model version within a local coupled domain. The local simulation state patch is not a copy of the complete digital twin simulation model.

[0154] The local simulation state patch SP_PALLET889_0007 also records the parent model version identifier, the set of entity identifiers corresponding to the local coupling domain, the event playback interval, the first event playback interval, the second event playback interval, and the boundary state sequence of the frozen boundary model object. The parent model version and the local simulation state patch are combined to form the model version MV_20260412_100005_04.

[0155] When undoing a model version, model objects within the local coupling domain are restored in the reverse order of the model events' topological sequence. First, the discrete state replacement pairs corresponding to subsequent model events are undoed, then the corresponding continuous state increments are subtracted. Subsequently, the state increments corresponding to preceding model events are undoed, and model restoration is completed based on the boundary state sequence of the parent model version and the frozen boundary model objects.

[0156] When multiple local simulation state patches exist, the local coupling domain, event playback interval, and model event set corresponding to each local simulation state patch are compared. The three local simulation state patches formed in this embodiment are shown in Table 4.

[0157] Table 4. Write Relationship Table for Multiple Local Simulation State Patches

[0158] The local coupling domains corresponding to local simulation state patch SP_CONV_C02_0011 and local simulation state patch SP_PALLET889_0007 do not share a common model object, and there is no directed reachable path between the two model event sets. Therefore, the two local simulation state patches can be written to the same parent model version in parallel.

[0159] The absence of a directed reachable path means that starting from any model event corresponding to a local simulation state patch, one cannot reach any model event corresponding to another local simulation state patch by following the directed edges in the directed event dependency graph.

[0160] The local simulation state patch SP_IFA02_0013 contains a connection interface model object, which is also located within the local coupling domain corresponding to the local simulation state patch SP_PALLET889_0007. Furthermore, the robotic arm interface contact model event corresponding to SP_PALLET889_0007 can reach the interface state correction model event corresponding to SP_IFA02_0013 along the directed event dependency graph. Therefore, SP_PALLET889_0007 is written first, and then the model event corresponding to SP_IFA02_0013 is replayed in the written digital twin simulation model.

[0161] Model objects belonging to two local coupling domains and connected to both local coupling domains are defined as common boundary model objects. When the spatial coordinates, running speed, interface status word, space occupancy status word, and load status word of the common boundary model object are consistent with the corresponding status after the previous local simulation status patch is written, the next local simulation status patch is written.

[0162] Before setting the model version to the current model version, convert the state changes of the equipment entities to be executed into future model events.

[0163] The state changes of the equipment entity to be executed are formed by the state change data in the smart warehousing and distribution physical scenario that is ready to be sent to the equipment entity. The future model event refers to the model event that is generated and inserted into the digital twin simulation model based on the state changes of the equipment entity to be executed before the physical equipment actually executes the operation.

[0164] In this embodiment, future model events include the AGV-05 start-up model event, path occupancy model event, conveyor interface contact model event, pallet exit from robotic arm model event, pallet entering conveyor model event, and conveyor departure model event.

[0165] Future model events are inserted into a directed event dependency graph, and forward simulation is executed starting from the current simulation time corresponding to the model version, until all load transfer tokens corresponding to each future load transfer event have completed exiting and entering model events and have been converted to the consumed state. The forward simulation interval is formed from the current simulation time corresponding to the model version to the time when the last load transfer token is converted to the consumed state.

[0166] Perform space occupancy mutual exclusion checks, load transfer conservation checks, and frozen boundary model object state checks on the forward simulation interval. After passing all checks, generate predicted event signatures according to the topological order of model events.

[0167] The predicted event signature is a state sequence composed of multiple ordered event records obtained from forward simulation. Each ordered event record includes, in sequence, a topology number, an event type, an entity identifier, an interface identifier, and a load identifier; if an interface identifier or a load identifier does not exist, a null value is written in the corresponding field.

[0168] The partial predicted event signatures generated in this embodiment are shown in Table 5.

[0169] Table 5 Predicted Event Signature Table

[0170] The model version that has passed the forward simulation verification is set as the current model version. After the working equipment entity generates new state sampling data, the new state sampling data is converted into feedback model events, and actual event signatures are formed according to the actual occurrence order of the feedback model events.

[0171] The feedback model event refers to the model event converted from the state sampling data generated after the actual execution of the work equipment entity. The actual event signature adopts the same field order as the predicted event signature.

[0172] Starting with the ordered event record with the smallest topology sequence number, the actual event signature and the predicted event signature are compared item by item. When the event type, entity identifier, interface identifier, and payload identifier under the corresponding topology sequence number are all the same, the comparison continues to the next ordered event record.

[0173] When any field is different or the order of events is different for the first time, the feedback model event corresponding to the ordered event record is identified as the first inconsistent feedback model event.

[0174] When the actual event signature and the predicted event signature are all consistent, the current model version is marked as the confirmed model version, and the feedback model event is written into the digital twin simulation model.

[0175] When the actual event signature is inconsistent with the predicted event signature, the local simulation state patch corresponding to the current model version is revoked, the parent model version is restored, and the model object corresponding to the first inconsistent feedback model event is set as the new starting model object. The process of determining the local coupling domain, determining the candidate model parameters, selecting the corrected simulation branch, and generating the model version is then re-executed.

[0176] Comparative Implementation Method 1 To verify the aforementioned implementation process, a comparative implementation method 1 was set up in the same digital twin simulation platform. Comparative implementation method 1 adopts the basic approach of warehouse digital twin modeling, simulation time management, and historical state playback. After detecting model residuals, it restores all model objects from the most recent global state snapshot and re-executes subsequent model events. It does not construct local coupling domains, set load transfer tokens, or use dual-event playback intervals and local simulation state patches.

[0177] Both this embodiment and Comparative Embodiment 1 use 146 model objects, perform 100 pallet transfer simulations respectively, and set up [something] in 50 of the simulations. to The state sampling data reception delay was recorded for each simulation. The data included the number of model objects participating in the re-simulation, the number of model events re-executed, the model correction calculation time, the corrected position model residual, the number of repeated space occupancy times, the number of load state discontinuities, and the consistency rate between actual event signatures and predicted event signatures. The obtained data is shown in Table 6.

[0178] Table 6 Simulation data table of this embodiment and comparative embodiment 1

[0179] The data in Table 6 are simulation data obtained under the same simulation hardware, the same model object, the same state sampling data, and the same delayed model events. Table 6 is used to compare the overall effect of the combined processing and global recovery and global re-simulation methods in this embodiment, and is not used to determine the independent contribution of each individual processing step.

[0180] By determining the state transition relationships, interface contact relationships, space occupancy relationships, and load transfer relationships between model events through a directed event dependency graph, and forming a local coupling domain based on the boundary state sequences of the baseline simulation branch and the correction simulation branch, model objects unaffected by model residuals can be excluded from state recovery and model event playback.

[0181] By using load transfer tokens, space occupancy mutual exclusion checks, and load transfer conservation checks, correction simulation branches with repetitive space occupancy states or discontinuous load states can be excluded before model parameters are written.

[0182] like Figure 9 As shown, the position model residuals of the baseline simulation branch gradually increase with simulation time, while the position model residuals of the calibration simulation branch gradually decrease after model event replay and stabilize at a certain value. nearby.

[0183] It should be noted that the embodiments of the present invention have better implementability and are not intended to limit the present invention in any way. Any person skilled in the art may use the above-disclosed technical content to change or modify it into equivalent effective embodiments. However, any modifications or equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention shall still fall within the scope of the technical solution of the present invention.

Claims

1. A method for dynamic modeling of intelligent warehousing and distribution digital twins, characterized in that, include: Obtain entity identifiers, geometric parameters, spatial coordinates, motion constraint parameters, interface status, and status sampling data of the storage structure entities, load-bearing entities, and operating equipment entities in the smart warehousing and distribution physical scenario; construct geometric models, motion constraint models, and event interaction models; generate digital twin simulation models by association; and determine model instances as model objects. State sampling data is converted into model events. A directed event dependency graph is constructed based on state transition, interface contact, space occupancy and load transfer relationships. Model events are executed in topological order and state snapshots are recorded to obtain the predicted state and model residuals. The event replay interval and the starting model object are determined by the model events corresponding to the model residuals, and the associated model objects are selected along the directed event dependency graph to form the initial local coupling domain; Candidate model parameters are determined from the geometric parameters, motion constraint parameters, and event interaction parameters of the initial local coupled domain. A baseline simulation branch and a correction simulation branch are formed. Model events are replayed and geometric boundary, motion continuity, space occupancy mutual exclusion, and load transfer conservation checks are performed. Compare the boundary state sequences of the two simulation branches, and expand the initial local coupling domain when they are inconsistent until they are consistent, thus forming the local coupling domain and simulation boundary conditions; Combine the load exit and enter model events into a load transfer event. When the two model events are not paired or the same load corresponds to multiple occupancy states, determine the event conflict interval, roll back the model objects in the local coupling domain, rearrange and replay the model events. Select the calibration simulation branch that passes the verification and whose model residual is less than the baseline simulation branch, write it into the digital twin simulation model and generate a model version. If no calibration simulation branch meets the conditions, retain the current model version.

2. The intelligent warehousing and distribution digital twin dynamic modeling method according to claim 1, wherein the state sampling data is converted into model events acting on the corresponding model object, including: Configure a physical timestamp and acquisition sequence number for the state sampling data, and determine the occurrence order of the state sampling data according to the physical timestamp and acquisition sequence number; Identify exit and entry state changes of the same carrier entity between two interconnected model objects from the state sampling data, and form cross-interface event pairs by combining the corresponding exit model events and entry model events. Based on the physical timestamps of multiple cross-interface event pairs, a time anchor sequence between the physical time axis and the simulation time axis is established, and each model event is mapped to the simulation time axis; Model events whose simulation timestamps are earlier than the simulation timestamps corresponding to the current model state are identified as delayed model events. The successor model events of the delayed model events are identified as model events to be replayed along the directed event dependency graph, serving as a candidate event set for incremental topology sorting and model event replay.

3. The intelligent warehousing and distribution digital twin dynamic modeling method according to claim 2, constructing the directed event dependency graph includes: Establish state transition edges between model events corresponding to continuous state changes of the same model object; Establish an interface contact edge between the pre-interface contact model event that makes two model objects make interface contact; When multiple model objects compete for the same space, a space occupancy edge is established between the model event that releases the space and the model event that requests the space; The model object corresponding to the load output end is determined as the source model object, the model object corresponding to the load receiving end is determined as the target model object, and a load transfer edge is established between the model event of the load exiting the source model object and the model event of the load entering the target model object. Insert the delayed model event into the directed event dependency graph, and perform incremental topological sorting on the delayed model event and its successor model event; When the established state transition edge, interface contact edge, space occupancy edge, or load transfer edge forms a directed loop, determine the minimum strongly connected subgraph containing the directed edge that caused the formation of the directed loop, and determine the time interval corresponding to the model event contained in the minimum strongly connected subgraph as the candidate event conflict interval.

4. The intelligent warehousing and distribution digital twin dynamic modeling method according to claim 3, forming the local coupling domain includes: From the starting model object, extend forward along the outgoing edges of the directed event dependency graph to select the successor model object whose state is affected by the model event corresponding to the starting model object; Extend backward from the successor model object along the incoming edge of the directed event dependency graph, and select the predecessor model object on which the successor model object depends for a state change; The starting model object, the successor model object, and the preceding model object are determined as the current coupling domain. The model events within the event playback interval are replayed in the benchmark simulation branch and the correction simulation branch, respectively, to obtain the boundary state sequence of each boundary model object at the edge of the current coupling domain. When the two boundary state sequences corresponding to the same boundary model object are consistent, the boundary model object is determined as a frozen boundary model object; when the two boundary state sequences corresponding to the same boundary model object are inconsistent, the model object directly connected to the boundary model object through the directed event dependency graph is incorporated into the current coupling domain, and the model event replay and boundary state sequence comparison are re-executed. The current coupling domain of all edge model objects is defined as the local coupling domain, and the boundary state sequence of each frozen boundary model object is set as the simulation boundary condition.

5. The intelligent warehousing and distribution digital twin dynamic modeling method according to claim 3, wherein the event interaction model generates a load transfer token when the load transfer event occurs, and the load transfer token includes a load identifier, an entity identifier of the source model object, an entity identifier of the target model object, an interface identifier, and a transfer sequence number; When the load exits the source model object, an exit model event associated with the load transfer token is generated, and the load occupancy status in the source model object is updated to the transfer status. When a load enters the target model object, the load transfer token is read and an entry event associated with the load transfer token is generated, and the load occupancy status in the target model object is updated to the occupied status. Each load transfer token is associated with only one exit model event and one enter model event, and is converted to a consumed state after the enter model event is executed.

6. The intelligent warehousing and distribution digital twin dynamic modeling method according to claim 5, determining the event conflict interval includes: The candidate event conflict interval is used as the initial search range for abnormal load transfer events. The system detects entry events that are not associated with load transfer tokens, entry events that are repeatedly associated with the same load transfer token, load transfer tokens that have not been converted to a consumed state, the source model object and the target model object that are both in an occupied state, and the interface identifiers of the exit event and the entry event that are inconsistent. The load transfer event corresponding to any detected anomaly is identified as an abnormal load transfer event; Tracing back from the abnormal load transfer event along the directed event dependency graph, the model event with the latest simulation timestamp and closest to the abnormal load transfer event in the topological order, which passes the load transfer conservation check, is determined as the starting event of the event conflict interval; Tracing forward from the abnormal load transfer event along the directed event dependency graph, the model event that corresponds to the load re-establishing a unique occupancy state and appears first in the topological order is determined as the termination event of the event conflict interval.

7. The intelligent warehousing and distribution digital twin dynamic modeling method according to claim 6, comprising rolling back the model objects within the local coupling domain and replaying the model events within the event conflict interval, including: Restore each model object within the local coupling domain to a snapshot of its state before the starting event, while keeping the state of each model object outside the local coupling domain unchanged; The model events within the event conflict interval are reordered according to the directed event dependency graph and then replayed sequentially. The replay of the event conflict interval is completed when the state of the boundary model object after replay is consistent with the state of the boundary model object at the corresponding moment before rollback. When the state of the boundary model object after replay is inconsistent with the state of the boundary model object at the corresponding moment before rollback, the model objects directly connected to the boundary model object through the directed event dependency graph are incorporated into the local coupling domain, and state recovery, model event sorting, and model event replay are re-executed.

8. The intelligent warehousing and distribution digital twin dynamic modeling method according to claim 4, determining the candidate model parameters, includes: Based on the reachability relationships in the directed event dependency graph and the spatial adjacency relationships between the model objects, the continuously generated model residuals and the model events that generate the model residuals are grouped into residual event clusters. When the model residuals of the model object before entering the motion state and after exiting the motion state maintain the same positional deviation in the same direction, the geometric parameters corresponding to the model object are determined as candidate model parameters. The model residuals accumulated during the motion of the model object and retained after the model object exits the motion state are determined as candidate model parameters. When a state transition time difference occurs between the exit model event and the enter model event in the cross-interface event pair, and the state transition time difference does not extend beyond the enter model event, the event interaction parameters of the corresponding interface are determined as candidate model parameters.

9. The intelligent warehousing and distribution digital twin dynamic modeling method according to claim 8, forming a candidate correction simulation branch, including: For each candidate model parameter, at least two first perturbation simulation branches with successively increasing parameter values ​​and at least two second perturbation simulation branches with successively decreasing parameter values ​​are formed; While keeping the model parameters other than the candidate model parameters and the simulation boundary conditions unchanged, the model events contained in the residual event cluster are replayed in the first perturbation simulation branch and the second perturbation simulation branch, respectively. The composite model residual is determined based on the location model residual, the state transition time model residual, and the space occupancy model residual, and the parameter range in which the direction of change of the composite model residual reverses between adjacent disturbance simulation branches is determined. Multiple candidate correction simulation branches are formed within the parameter range.

10. The intelligent warehousing and distribution digital twin dynamic modeling method according to claim 9, performing verification on the candidate correction simulation branch, including: The earliest model event in the residual event cluster is determined as the first start event, the latest model event in the residual event cluster is determined as the first termination event, and the interval between the first start event and the first termination event is determined as the first event replay interval. The first termination event is determined as the second start event, the model event in which the frozen boundary model object undergoes its first state change after the first termination event is determined as the second termination event, and the period between the second start event and the second termination event is determined as the second event replay interval. Within the first event playback interval, the geometric boundary verification, motion continuity verification, space occupancy mutual exclusion verification, and load transfer conservation verification are performed sequentially on each candidate correction simulation branch, and candidate correction simulation branches that fail any of the verifications are deleted. For the retained candidate correction simulation branches, continue to replay the model events within the second event replay interval, and perform the geometric boundary verification, motion continuity verification, space occupancy mutual exclusion verification, and load transfer conservation verification again.

Images