Logistics storage intelligent scheduling method and system

Abstract

The invention relates to the technical field of logistics storage intelligent scheduling, and discloses a logistics storage intelligent scheduling method and system. The method comprises the following steps: constructing a virtual storage model and injecting a real-time operation data stream to form a storage dynamic twinborn body; implanting a self-adaptive scheduling logic core in the twin body, and performing multi-dimensional task analysis through the core to generate a task decomposition atlas; resource matching operation is carried out in the logic partition based on the atlas, and a resource matching scheme is output; generating a control instruction sequence containing time sequence constraint according to the scheme and issuing the control instruction sequence to a physical execution terminal; and collecting a feedback data stream to correct the twinborn state. Through real-time data flow driving, synchronous sensing and instantaneous response of the storage state are realized, and the scheduling deviation caused by data updating delay of a traditional system is solved. Through the task decomposition atlas, deep structured analysis and multi-resource collaborative matching of complex tasks are realized, and the overall utilization efficiency of resources and the task execution fluency are improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent scheduling technology for logistics and warehousing, specifically to an intelligent scheduling method and system for logistics and warehousing. Background Technology

[0002] Current logistics and warehousing systems largely rely on warehouse management or monitoring systems for scheduling and management. These systems typically make decisions based on periodically updated databases. Their operation involves taking static snapshots of inventory, equipment status, and other information at specific intervals, and scheduling algorithms are calculated based on these discontinuous data points. This discrete data supply method introduces an inherent delay between the warehouse status perceived by the scheduling system and the actual physical state. When dealing with high-frequency, fluctuating order flows and rapidly changing equipment status, such systems struggle to capture dynamic information in a timely manner, leading to decisions based on outdated information and insufficient accuracy and timeliness of scheduling instructions. The system's ability to respond to external changes is lagging, resulting in poor overall scheduling flexibility.

[0003] Existing technical solutions typically treat complex tasks such as receiving, picking, and packaging as a single unit at the task processing level, or perform only simple linear decomposition. Resource allocation logic largely relies on pre-defined, fixed rules. This approach ignores the multidimensional relationships and spatiotemporal dependencies between subtasks within complex tasks. Its resource matching process is relatively crude, lacking the ability to collaboratively optimize and coordinate multiple resources across different logical partitions. This easily leads to resource utilization conflicts, extended task waiting times, and low efficiency in cross-regional collaboration, limiting further improvements in overall warehouse throughput. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for intelligent scheduling of logistics and warehousing, so as to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides an intelligent scheduling method for logistics warehousing, the method comprising: Construct a virtual warehouse model corresponding to the physical warehouse space, wherein the virtual warehouse model contains multiple logical partitions; Real-time running data streams are injected into the virtual warehouse model to form a dynamic warehouse twin with dynamic response characteristics; an adaptive scheduling logic core is implanted into the dynamic warehouse twin. The adaptive scheduling logic core performs multi-dimensional task analysis on the warehouse dynamic twin to generate a task decomposition graph. Based on the task decomposition graph, resource matching calculations are performed within the logical partitions of the warehouse dynamic twin, and a resource matching scheme is output. According to the resource matching scheme, the strategy executor in the core of the adaptive scheduling logic is activated to generate a sequence of control instructions containing timing constraints. The control command sequence is sent to the physical execution terminal, and the feedback data stream from the physical execution terminal is collected; The state of the warehouse dynamic twin is corrected using the feedback data stream.

[0006] Preferably, the construction of the virtual storage model corresponding to the physical storage space includes: scanning the structural layout of the physical storage space, identifying the spatial coordinates and topological relationships of storage shelves, conveying paths, work stations and buffer areas, constructing logical partitions with the same topological structure in the virtual space based on the spatial coordinates and topological relationships, assigning a unique attribute identifier to each logical partition, and establishing a mapping relationship table between physical storage equipment and the logical partitions.

[0007] Preferably, injecting real-time operational data streams into the virtual warehouse model to form a dynamic warehouse twin with dynamic response characteristics includes: acquiring order data, inventory data, and equipment status data in real time from the warehouse management system to form a real-time operational data stream; synchronizing the real-time operational data streams to the corresponding logical partitions according to the mapping relationship table; setting up a data listener for each logical partition, and triggering a status refresh of the warehouse twin when the data in the logical partition is updated.

[0008] Preferably, the step of embedding the adaptive scheduling logic core into the warehouse dynamic twin includes: defining a set of scheduling rule bases, the scheduling rule bases including priority rules, path optimization rules and conflict avoidance rules; constructing a strategy selector, the strategy selector dynamically selecting applicable scheduling rules from the scheduling rule bases based on the current state of the warehouse dynamic twin; encapsulating the strategy selector and the scheduling rule bases into an adaptive scheduling logic core, and deploying it in the warehouse dynamic twin; The constructed strategy selector includes: establishing a state feature extraction module to collect in real time the order backlog, equipment operating load, and path passage efficiency indicators of each logical partition in the warehouse dynamic twin; designing a rule matching engine to calculate the similarity between the feature vector output by the state feature extraction module and the triggering condition of each rule in the scheduling rule base, sorting the rules according to the similarity calculation results, selecting the rule with the highest similarity as the applicable scheduling rule, and setting a dynamic rule switching mechanism so that when the state feature change of the warehouse dynamic twin exceeds the sensitivity threshold, the rule selection process is automatically re-executed.

[0009] Preferably, the step of performing multi-dimensional task analysis on the warehouse dynamic twin through the adaptive scheduling logic core to generate a task decomposition graph includes: receiving an externally input task request, the task request including material information, target location and time requirements; simulating the complete execution process of the task request in the warehouse dynamic twin; identifying the logical partitions and operation dependencies involved; decomposing the task request into multiple atomic operation steps; and generating a task decomposition graph with nodes and edges based on the operation dependencies.

[0010] Preferably, the step of performing resource matching calculations within the logical partitions of the warehouse dynamic twin based on the task decomposition graph and outputting a resource matching solution includes: traversing each atomic operation step in the task decomposition graph, querying the logical partitions and equipment resources required by it, checking the availability status of the logical partitions and equipment resources within the task time window, allocating available resource instances for each atomic operation step, and generating a resource matching scheme containing resource identifiers and execution time points.

[0011] Preferably, the step of activating the strategy executor in the adaptive scheduling logic core according to the resource matching scheme and generating a control instruction sequence containing timing constraints includes: the strategy executor determining the specific execution parameters of each atomic operation step according to the resource matching scheme and the scheduling rules in the adaptive scheduling logic core, arranging a strict execution timing sequence for all atomic operation steps to ensure that the time requirements of the task request are met, and compiling the atomic operation steps with execution parameters and timing sequence into a control instruction sequence that can be recognized by the physical execution terminal.

[0012] Preferably, the step of sending the control command sequence to the physical execution terminal and collecting the feedback data stream from the physical execution terminal includes: sending the control command sequence sequentially to the corresponding AGV, stacker crane or sorting device through a communication interface, and receiving the status code, location information and task completion progress returned by the physical execution terminal in real time to form a feedback data stream.

[0013] Preferably, the step of using the feedback data stream to correct the state of the warehouse dynamic twin includes: comparing the feedback data stream with the predicted state of the corresponding logical partition in the warehouse dynamic twin, calculating the state deviation value, and if the state deviation value exceeds a preset threshold, triggering the parameter calibration process of the virtual warehouse model to update the state attributes of the logical partition. The step of comparing the feedback data stream with the predicted state of the corresponding logical partition in the warehouse dynamic twin includes: performing time series alignment processing on the feedback data stream, extracting the actual state data of the physical execution terminal that is consistent with the timestamp of the predicted state; and selecting key state variables for each logical partition, including device location coordinates, task execution progress, and resource occupancy status. The calculation of the state deviation value includes: calculating the absolute error between the actual value and the predicted value for each key state variable, and using a weighted average algorithm to synthesize the absolute errors of multiple variables into a comprehensive state deviation value, wherein the weight coefficients are preset according to the degree of influence of the variables on the system stability.

[0014] Preferably, the present invention also includes a logistics and warehousing intelligent scheduling system, the system including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein when the processor executes the computer program, it implements the steps of the logistics and warehousing intelligent scheduling method described above.

[0015] Compared with the prior art, the beneficial effects of the present invention are: By injecting real-time operational data streams into a virtual warehouse model to form a dynamic twin, and embedding the core scheduling logic within it, the fusion of scheduling computation and state awareness is achieved. Any state change in the physical space is mapped in real time and drives the twin to evolve synchronously. The embedded scheduling logic can therefore directly perform calculations based on a continuously updated, high-fidelity virtual model, eliminating the latency introduced by data querying, transmission, and processing in traditional architectures. The scheduling system can respond instantly to environmental changes and generate control commands for dynamic scenarios, fundamentally improving the real-time performance and environmental adaptability of its decisions.

[0016] By performing multi-dimensional analysis of tasks and generating a structured task decomposition graph, complex macro-tasks are broken down into a network of sub-tasks with clear logic and spatiotemporal relationships. Resource matching operations, based on this graph, can perform refined identification and global optimization of the resource requirements of each sub-task within the logical partitions of the twin. This method comprehensively considers the cooperation relationships between different resource types and the constraints before and after task execution, thereby achieving efficient collaboration and seamless connection of cross-partition and multi-category resources when formulating resource allocation schemes. Overall resource utilization efficiency is optimized, and the smoothness and completion speed of complex task flows are improved. Continuous correction of the twin's state by feedback data streams further ensures that the task decomposition and resource matching schemes are always based on the real environment, continuously enhancing the reliability and executability of the output results. Attached Figure Description

[0017] Figure 1 This is a schematic diagram illustrating the working principle of the intelligent logistics and warehousing scheduling method described in this invention. Figure 2 A flowchart for constructing a virtual warehouse model; Figure 3 A flowchart for generating a task decomposition map. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Please see Figure 1 This invention provides an intelligent scheduling method for logistics warehousing. The method includes: constructing a virtual warehousing model corresponding to the physical warehousing space, the virtual warehousing model containing multiple logical partitions; injecting real-time running data streams into the virtual warehousing model to form a dynamic warehousing twin with dynamic response characteristics; embedding an adaptive scheduling logic core into the dynamic warehousing twin; performing multi-dimensional task analysis on the dynamic warehousing twin through the adaptive scheduling logic core to generate a task decomposition graph; performing resource matching calculations within the logical partitions of the dynamic warehousing twin based on the task decomposition graph to output a resource matching scheme; activating the strategy executor in the adaptive scheduling logic core according to the resource matching scheme to generate a control instruction sequence containing timing constraints; sending the control instruction sequence to the physical execution terminal and collecting feedback data streams from the physical execution terminal; and using the feedback data streams to correct the state of the dynamic warehousing twin.

[0020] Example 1: See Figure 2 In practice, a detailed scan of the physical storage space is performed using laser scanners or 3D imaging equipment to capture the spatial coordinates and topological relationships of storage shelves, conveyor paths, workstations, and buffer areas. These spatial coordinates and topological relationships include geometric positions, connectivity, and relative orientations. Based on the spatial coordinates and topological relationships, logical partitions with the same topological structure are constructed in the virtual space. These logical partitions are implemented using computer-aided design software, and each logical partition is assigned a unique attribute identifier using an alphanumeric encoding format. A mapping table between physical storage equipment and logical partitions is established and stored in a database table, containing equipment identifiers, logical partition identifiers, and coordinate mapping information. In practice, the scanning process involves multi-sensor fusion to ensure accurate extraction of spatial coordinates and topological relationships. The layout of logical partitions in the virtual space corresponds one-to-one with the physical storage space. Attribute identifiers are used to uniquely distinguish each logical partition, and the mapping table is periodically verified to maintain accuracy.

[0021] The topology analysis of logical partitions is implemented using an undirected graph adjacency list algorithm: Entities such as shelves, paths, and platforms in the physical warehouse are abstracted as nodes in a graph, and the connectivity between entities (e.g., the connection between shelves and conveyor paths, and the connection between paths and platforms) is abstracted as edges. The adjacency list data structure stores the relationships between nodes and edges. Specifically, each node is assigned a unique ID corresponding to the attribute identifier of the physical entity, and each entry in the adjacency list records the IDs of its directly connected nodes and the connection distance (in meters). This algorithm can quickly query the topological associations between any two entities by traversing the adjacency list, ensuring topological consistency between the virtual model and the physical space.

[0022] Real-time order data, inventory data, and equipment status data are acquired from the warehouse management system. Order data includes order number and bill of materials; inventory data includes inventory level and storage location; and equipment status data includes equipment operating status and fault information. This data is streamed via an application programming interface (API) to form a real-time operational data stream. This real-time operational data stream is synchronized to the corresponding logical partitions according to a mapping table. The synchronization process is based on identifier matching in the mapping table, and data updates trigger changes in the logical partition status. A data listener is set up for each logical partition. The data listener is implemented as a software module to continuously monitor the data inflow. The real-time operational data stream consists of three core data categories: Order data fields: order number, material code, required quantity, order creation time, and latest completion time; Inventory data fields: material code, storage logical partition ID, current inventory quantity, and inventory update timestamp; Equipment status data fields: equipment ID, equipment type, current operating status, coordinates of the logical partition, remaining battery power, and cumulative runtime. Data injection employs a second-level synchronization mechanism, establishing a long-lived connection with the warehouse management system via TCP / IP. It actively pulls the latest data every second, triggering a reconnection mechanism upon pull failure. Reconnection intervals are 1 second, 3 seconds, and 5 seconds respectively, generating an alarm log after 5 failed attempts. The data listener is triggered by field changes. Specifically, when the required quantity or latest completion time changes in the order data corresponding to a logical partition, the current inventory quantity changes in the inventory data, or the current operating status or the coordinates of the logical partition changes, the twin's status is automatically refreshed. The specific refresh process involves locking the logical partition's status data cache; replacing the old values ​​of the changed fields; recalculating the load factor of the logical partition (currently occupied resources divided by total resources); unlocking the cache and synchronizing it to the twin's global status database. When the data in a logical partition is updated, the warehouse dynamic twin's status is refreshed, involving recalculating the logical partition's dynamic attributes. Optionally, the data listener can adopt a publish-subscribe pattern for efficient data distribution.

[0023] In actual deployment, during the construction of logical partitions, graph theory algorithms are used to represent nodes and edges in the topology analysis. Spatial coordinates are defined using a Cartesian coordinate system, and attribute identifier codes are generated following international standard encoding rules. The establishment of the mapping table involves data cleaning and verification steps to prevent mapping errors. The specific implementation of the graph theory algorithm stores the associations between nodes and edges by constructing an adjacency matrix or adjacency list data structure. The row and column indices of the adjacency matrix correspond to node identifiers, and the matrix element values ​​represent the Boolean values ​​or weights of the edges. The adjacency list uses a linked list structure to record the set of adjacent nodes for each node, thus completely mapping the topology of the physical warehouse. Real-time data stream processing employs a streaming computing framework, such as Apache Kafka, to ensure low-latency data transmission. The implementation of the data listener includes event handling logic, automatically invoking the state refresh function when a data update event occurs. In some embodiments, the warehouse management system's data interface is configured as a RESTful API, supporting high-concurrency data access. Optionally, logical partitions can be further subdivided based on functional areas, such as by temperature zones or security levels.

[0024] Example 2: This example, based on the warehouse dynamic twin generated in Example 1, adds an adaptive scheduling logic core and dynamic rule matching function to achieve intelligent and flexible task parsing. Specifically, a scheduling rule base is defined, with the following specific rule definitions: Priority rules are sorted by weight of order urgency > material scarcity > task complexity. Order urgency is calculated as the latest completion time minus the current time divided by the estimated task execution time. Material scarcity is calculated as the historical 30-day average inventory minus the current inventory divided by the historical 30-day average inventory. Task complexity is determined by the total number of atomic operation steps in the task decomposition graph generated after multi-dimensional parsing of the task. Path optimization rules are based on Dijkstra's algorithm, using path length plus equipment congestion coefficient as the cost function. Equipment congestion coefficient is the number of currently occupied devices on the path divided by the maximum number of devices the path can accommodate. In conflict avoidance rules, when two atomic operation steps request the same resource, the principle of first-come, first-served applies. If the request time difference is less than 0.5 seconds, the task with the higher priority rule calculation result is given priority.

[0025] The scheduling rule base includes priority rules, path optimization rules, and conflict avoidance rules. Priority rules are used to determine the order in which tasks are executed, path optimization rules are used to calculate the optimal movement path, and conflict avoidance rules are used to resolve the problem of device resource contention.

[0026] A strategy selector is constructed. Based on the current state of the warehouse dynamic twin, the strategy selector dynamically selects applicable scheduling rules from the scheduling rule base. The strategy selector and the scheduling rule base are encapsulated into an adaptive scheduling logic core and deployed in the warehouse dynamic twin. The adaptive scheduling logic core is integrated into the operating framework of the warehouse dynamic twin as a software module.

[0027] A status feature extraction module is established. This module collects data on order backlog, equipment operating load, and path passage efficiency in each logical partition of the warehouse dynamic twin in real time through a data interface. The specific dimensions and sources of the data collection are as follows: the data on order backlog comes from the twin's order management module, and the collection frequency is consistent with the data listener (1 second / time); the data on equipment operating load comes from the twin's equipment management module, which is statistically analyzed by polling the operating status indicators (0=idle, 1=busy, 2=faulty) of each device; the data on path passage efficiency comes from the twin's path monitoring module, which counts the number of devices passing through the path every 10 seconds and converts it into the number of passes per unit time.

[0028] Order backlog is measured by the number of orders awaiting processing; equipment load is expressed as a percentage of equipment utilization; and path efficiency is calculated as the number of tasks passing through path nodes per unit time. A rule matching engine is designed. This engine calculates the similarity between the feature vectors output by the state feature extraction module and the trigger conditions of each rule in the scheduling rule base. The similarity calculation uses the cosine similarity method in the vector space model. Rules are ranked according to the similarity calculation results, and the rule with the highest similarity is selected as the applicable scheduling rule. A dynamic rule switching mechanism is set up to continuously monitor changes in the state features of the warehouse dynamic twin. When the changes in state features exceed a preset sensitivity threshold, the rule selection process is automatically re-executed. The specific calculation method of the state feature extraction module is as follows: order backlog is the number of orders awaiting processing in the current logical partition divided by the historical daily average number of orders processed in the partition; equipment load is the number of currently running devices in the logical partition divided by the total number of devices in the partition; and path efficiency is the number of times the current path passes through per unit time in the logical partition divided by the maximum number of times the path is designed to pass through. The rule matching process specifically matches the three-dimensional feature vector (order backlog, equipment operating load, and path efficiency) output by the state feature extraction module with the trigger condition vector of each rule. The trigger condition vector consists of manually preset threshold ranges. For example, the trigger condition vector for a certain priority rule might be: order urgency < 0.3, material scarcity > 0.5, and task complexity > 5. During matching, each dimension of the feature vector is checked to see if it falls within the threshold range of the trigger condition vector. If two or more dimensions meet the threshold, it is considered a preliminary match. Then, the optimal rule is selected by calculating the Euclidean distance. The trigger threshold determination criterion for the dynamic switching mechanism is: the change in state features equals the L1 norm of the absolute value of the current feature vector minus the feature vector from the previous match, i.e., the sum of the absolute values ​​of the changes in the three dimensions. The preset sensitivity threshold is 0.2. When the change in state features is ≥ 0.2, the trigger rule is reselected; when the change is < 0.2, the current rule remains unchanged.

[0029] The similarity calculation between the feature vector and the rule triggering condition uses the Manhattan distance matching algorithm: Let the feature vector output by the state feature extraction module be X=(x1,x2,x3), where x1 is the order backlog (quantified as the number of orders to be processed), x2 is the equipment operating load (quantified as the number of current tasks), and x3 is the path passage efficiency (quantified as the number of passages per unit time); the triggering condition vector of a certain rule in the scheduling rule base is Y=(y1,y2,y3). Then, the similarity calculation is achieved by calculating the Manhattan distance between the two vectors. The distance calculation formula is: Where L is the Manhattan distance, a smaller L value indicates a higher similarity. The rule matching engine sorts all rules by their L values ​​and selects the rule with the smallest L value as the applicable scheduling rule. The sensitivity threshold of the dynamic rule switching mechanism is set to a fixed value of 3 (i.e., switching is triggered when the sum of the changes in each dimension of the feature vector exceeds 3). For example, when x1 changes from 5 to 9, x2 changes from 3 to 4, and x3 changes from 8 to 7, the sum of the changes is 4+1+1=6>3, and the rule selection process is automatically re-executed.

[0030] The rules in the scheduling rule base are stored in the form of production rules. Each rule includes preconditions and a conclusion action. The strategy selector is implemented using object-oriented programming. The data acquisition frequency of the state feature extraction module is configurable to ensure real-time performance. The similarity calculation process of the rule matching engine can be expressed as the formula: in: Representing the eigenvector With the rule triggering condition vector Similarity value between them It is a multidimensional feature vector output by the state feature extraction module. It is the trigger condition vector of a rule in the scheduling rule base, symbol Represents the vector dot product. and Representing vectors respectively sum vector The Euclidean norm. This can be understood as the similarity value. The value ranges from 0 to 1, with a higher match rate indicating a closer match to 1. Optionally, the sensitivity threshold of the dynamic rule switching mechanism can be adjusted based on the volatility of warehouse operations, for example, by setting a lower threshold during peak periods to enhance response speed.

[0031] Specifically, the status feature extraction module collects data in the following ways: order backlog is calculated by querying the pending orders table of the logical partition to determine the total number of orders; equipment operating load is obtained by reading the current task queue length of the equipment controller; and path passage efficiency is calculated by counting the number of times equipment passes through the path in the past 5 minutes. Data collection employs a multi-threaded concurrent acquisition mechanism, with each logical partition allocated an independent acquisition thread. Mutual exclusion locks are used between threads to avoid data conflicts. The collected raw data is cleaned (outliers such as negative numbers or values ​​outside the reasonable range are removed), then encapsulated into feature vectors and output to the rule matching engine. The data acquisition cycle is consistent with the rule matching cycle, both being 1 second.

[0032] In actual deployment, the state feature extraction module periodically reads state snapshots from the database of the warehouse dynamic twin. The dimension of the feature vector V is consistent with the dimension of the trigger condition vector in the scheduling rule base. The rule matching engine traverses all rules in the scheduling rule base, calculates the similarity value S corresponding to each rule, and sorts the rule list in descending order. The strategy selector selects the rule at the top of the sorted list as the currently applicable scheduling rule and loads its conclusion action into the execution environment. The rule dynamic switching mechanism checks the change in Euclidean distance of the feature vector after each state feature update. If the change exceeds the sensitivity threshold, the rule reselection process is immediately triggered. In some embodiments, the maintenance of the scheduling rule base supports dynamic addition and deletion of rules to adapt to changes in business strategies. The normalization of feature vectors can improve the accuracy of similarity calculation. Optionally, the rule matching engine can use a caching mechanism to store recent matching results to improve computational efficiency.

[0033] Example 3: See Figure 3 This embodiment, based on the adaptive scheduling logic core implanted in Embodiment 2, adds multi-dimensional task analysis, task decomposition graph generation, and resource matching calculation functions to achieve structured processing and precise resource allocation for complex tasks. It receives externally input task requests, which include material information, target location, and time requirements. Material information is represented by material codes and quantities, target location is defined by logical partition coordinates, and time requirements are specified by the latest completion timestamp. In the warehouse dynamic twin, the complete execution flow of the task request is simulated. The simulation process is based on the current state data of the warehouse dynamic twin, identifying the logical partitions and operational dependencies involved in the task execution flow. Operational dependencies refer to the sequential constraints between atomic operation steps.

[0034] The task request is decomposed into multiple atomic operation steps, which are indivisible basic work units. A task decomposition graph with nodes and edges is generated based on operation dependencies. Nodes represent atomic operation steps, and edges represent operation dependencies. Each atomic operation step in the task decomposition graph is traversed to query the logical partitions and device resources it requires. Resource requirement information is predefined in the attributes of the atomic operation step. The availability status of logical partitions and device resources within the task time window is checked. The task time window is determined by the estimated execution time of the atomic operation step and its dependencies. Availability status is obtained by querying the resource reservation table of the dynamic twin of the warehouse. Available resource instances are allocated to each atomic operation step. Resource instances are specific physical devices or logical partition units, and a resource matching scheme containing resource identifiers and execution time points is generated. The resource matching scheme is output in the form of structured data records.

[0035] Task requests are received via a standard application programming interface (API), with data in JSON or XML format. The simulated execution flow is driven by a discrete event simulation engine. Operation dependencies are described using a directed acyclic graph (DAG) model, and the task decomposition graph is stored using a graph database. The resource requirements for atomic operation steps are defined, including resource type and duration, and a resource reservation table maintains the temporal status of all resources. The resource matching scheme generation process can be represented as a resource allocation function: ; Where: function This represents the resource allocation logic. This represents the result of assigning the i-th atomic operation step to the j-th resource instance. This represents the set of resource requirements for the i-th atomic operation step. This represents the available time slots for the j-th resource instance within the task time window. This represents the load weight of the k-th logical partition. It can be understood that the resource allocation function F needs to satisfy the constraints between the resource requirements of atomic operation steps and the availability of resource instances. Optionally, the resource matching operation can use a constraint programming algorithm to solve for the optimal allocation scheme.

[0036] In actual deployment, the task decomposition map generation process includes parsing the bill of materials for the task request, determining the required picking, handling, and handover operation sequence, and annotating each atomic operation step with its expected execution duration and resource type dependencies. The resource matching calculation module accesses the resource status snapshot of the warehouse dynamic twin and finds compatible resource instances that are not occupied within the task time window for each atomic operation step. The allocation process must ensure that the order of operation dependencies is satisfied. The output of the resource matching scheme includes a unique identifier for the resource instance allocated to each atomic operation step and the planned start and end times. In some embodiments, when multiple resource instances are available, the selection strategy can be based on the principle of shortest distance or lowest load. After the resource matching scheme is generated, the state of the relevant resources is temporarily locked to prevent conflicts.

[0037] Example 4: This example, based on the resource matching scheme output in Example 3, adds functions for generating and issuing control command sequences and collecting feedback data streams to achieve physical implementation of the scheduling strategy and feedback of execution status. Specifically, the strategy executor, according to the resource matching scheme (which includes resource identifiers and execution time points allocated to atomic operation steps), and combined with the scheduling rules in the core of the adaptive scheduling logic (including priority rules, path optimization rules, and conflict avoidance rules), determines the specific execution parameters for each atomic operation step. These parameters include movement speed, lifting height, and loading weight. A strict execution sequence is arranged for all atomic operation steps. This sequence is constructed based on the time points in the resource matching scheme and combined with operation dependencies to ensure that the time requirements of the task request are met. The atomic operation steps with execution parameters and sequence are compiled into a control command sequence recognizable by the physical execution terminal. The control command sequence adopts a protocol format. The control command sequence is sent sequentially to the corresponding AGV, stacker crane or sorting device through the communication interface, which includes Ethernet, wireless LAN or industrial fieldbus. The status code, location information and task completion progress returned by the physical execution terminal are received in real time to form a feedback data stream. The feedback data stream is persistently stored for subsequent analysis.

[0038] The strategy executor is a standalone software service that parses the resource allocation results in the resource matching scheme and calls rules from the scheduling rule base to calculate the optimal execution parameters. The orchestration of the execution timing needs to consider equipment dynamics constraints and path conflict avoidance. The compilation process of the control instruction sequence involves translating abstract operation instructions into equipment-specific low-level commands, such as translating the "pick up" instruction into joint control signals for the robotic arm. It can be understood that timing orchestration needs to satisfy strict sequence constraints, the logic of which can be expressed as a timing constraint function: Where: function Indicates the timing arrangement algorithm, This indicates the timing arrangement of the final generated control command sequence. This represents the set of execution parameters determined by the strategy executor. This indicates the deadline requirement obtained from the resource matching scheme. This represents the set of all preconditions between atomic operation steps.

[0039] In actual deployment, the policy executor generates specific control instructions for each atomic operation step, including target coordinates, action type, and parameter settings. These instructions are organized into an ordered list according to a pre-arranged sequence. The communication interface manager is responsible for encapsulating the instructions in the list into data packets according to the protocol and sending them to the correct physical execution terminal network address. After receiving and executing the instructions, the physical execution terminal returns an acknowledgment signal and real-time status data through the same communication link. Referring to Table 1, the feedback data stream contains key information about the instruction execution results.

[0040] Table 1: Feedback Data Stream Record Table It is understood that the status codes in the feedback data stream are used to immediately determine whether the instruction execution was successful, while the location information and progress percentage are used to track the overall task progress. In some embodiments, if the returned status code indicates an error or timeout, the policy executor will trigger an exception handling process. Optionally, the communication interface may employ a heartbeat mechanism to maintain the reliability of the connection with the physical execution terminal.

[0041] Example 5: This example utilizes the feedback data stream collected in Example 4 to add a state correction function to the warehouse dynamic twin, thereby achieving closed-loop optimization of the scheduling system. Specifically, the feedback data stream is compared with the predicted state of the corresponding logical partition in the warehouse dynamic twin. The feedback data stream includes the actual status code, location coordinates, and task progress percentage returned by the physical execution terminal. The predicted state is the expected state data obtained by the warehouse dynamic twin based on the simulation operation of the control command sequence. A state deviation value is calculated, which quantifies the degree of difference between the actual state and the predicted state. If the state deviation value exceeds a preset threshold, the parameter calibration process of the virtual warehouse model is triggered, updating the state attributes of the logical partition. The feedback data stream undergoes time series alignment processing. This time series alignment processing uses a timestamp matching algorithm to align the actual data points and predicted data points to a unified time axis, extracting the actual state data of the physical execution terminal that matches the predicted state timestamp. For each logical partition, key state variables are selected, including equipment location coordinates, task execution progress, and resource occupancy status. The selection of key state variables is based on their contribution to the representation of the system's operating state. For each key state variable, the normalized deviation between the actual value and the predicted value is calculated. The normalized deviation is obtained by dividing the absolute difference by the range of the variable. A weighted summation algorithm is used to synthesize the normalized deviations of multiple key state variables into a comprehensive state deviation value, where the weight coefficients are preset according to the degree of influence of the variables on the system stability.

[0042] The comparison operation between the feedback data stream and the predicted state is periodically executed in the state monitoring module. The time series alignment process uses linear interpolation to fill the small differences between the data collection time points and the predicted time points. The actual state data and corresponding timestamps of the physical execution terminals in the feedback data stream are extracted and compared with the predicted state timestamps of the digital twin. If the actual data timestamp is inconsistent with the predicted timestamp (the difference is within ±1 second), the actual state data corresponding to the predicted timestamp is calculated by linear interpolation. For example, if the predicted state corresponding to the predicted timestamp t0 is P0, the actual state corresponding to the timestamp t1 (t1 < t0) in the actual data is P1, and the actual state corresponding to the timestamp t2 (t2 > t0) is P2, then the actual state data P corresponding to t0 is P = P1 + (P2 - P1) * (t0 - t1) / (t2 - t1), ensuring the consistency of the comparison in the time dimension. The normalization process of the key state variables ensures that variables with different dimensions can participate in the comprehensive calculation. The normalization of the device position coordinates may be based on the overall size of the warehouse. The task execution progress is already a percentage and does not require additional normalization. The resource occupancy status can be converted into a utilization percentage. It can be understood that the calculation of the normalization deviation makes the deviation values of all key state variables fall within the same numerical range. The calculation process of the comprehensive state deviation value can be expressed by the formula: Where: represents the comprehensive state deviation value, represents the total number of selected key state variables, is the variable index, represents the preset weight coefficient of the nth key state variable, represents the actual value of the nth key state variable, represents the predicted value of the nth key state variable, represents the range of the nth key state variable, that is, the difference between the maximum and minimum values that the variable may take. Optionally, the weight coefficient can be assigned by analyzing the impact of the fluctuations of each variable on the overall system performance in historical operation data.

[0043] During actual deployment, the calculation frequency of the state deviation value is consistent with the update frequency of the feedback data stream. The preset threshold is set as a fixed value according to the system's requirement for state synchronization accuracy. When the comprehensive state deviation value exceeds the threshold, the parameter calibration process is automatically started. The process adjusts the parameters related to the dynamic characteristics of physical devices in the virtual warehouse model through an optimization algorithm, such as the maximum speed parameter of an AGV cart or the acceleration parameter of a stacker crane. The update of the logical partition state attributes directly acts on the core database of the warehouse digital twin, and the corrected state will be used for subsequent task parsing and resource matching operations. In some embodiments, the range of the key state variables This can be obtained from equipment specifications or historical operating data and set as a constant. Optionally, the parameter calibration process can be designed as a gradient descent iterative process to gradually reduce the state deviation value. It can be understood that through continuous comparison and correction based on normalized deviation, the warehouse dynamic twin can effectively track the real state of the physical system.

[0044] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0045] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for intelligent scheduling of logistics warehousing, characterized in that, The method includes: Construct a virtual warehouse model corresponding to the physical warehouse space, wherein the virtual warehouse model contains multiple logical partitions; Real-time running data streams are injected into the virtual warehouse model to form a dynamic warehouse twin with dynamic response characteristics; an adaptive scheduling logic core is implanted into the dynamic warehouse twin. The adaptive scheduling logic core performs multi-dimensional task analysis on the warehouse dynamic twin to generate a task decomposition graph. Based on the task decomposition graph, resource matching calculations are performed within the logical partitions of the warehouse dynamic twin, and a resource matching scheme is output. According to the resource matching scheme, the strategy executor in the core of the adaptive scheduling logic is activated to generate a sequence of control instructions containing timing constraints. The control command sequence is sent to the physical execution terminal, and the feedback data stream from the physical execution terminal is collected; The state of the warehouse dynamic twin is corrected using the feedback data stream.

2. The intelligent scheduling method for logistics warehousing as described in claim 1, characterized in that, The construction of a virtual storage model corresponding to the physical storage space includes: scanning the structural layout of the physical storage space, identifying the spatial coordinates and topological relationships of storage shelves, conveying paths, work stations, and buffer areas, constructing logical partitions with the same topological structure in the virtual space based on the spatial coordinates and topological relationships, assigning a unique attribute identifier to each logical partition, and establishing a mapping relationship table between physical storage equipment and the logical partitions.

3. The intelligent scheduling method for logistics warehousing as described in claim 2, characterized in that, The step of injecting real-time operational data streams into the virtual warehouse model to form a dynamic warehouse twin with dynamic response characteristics includes: acquiring order data, inventory data, and equipment status data in real time from the warehouse management system to form a real-time operational data stream; synchronizing the real-time operational data streams to the corresponding logical partitions according to the mapping relationship table; setting up a data listener for each logical partition, and triggering a status refresh of the warehouse twin when the data in the logical partition is updated.

4. The intelligent scheduling method for logistics warehousing as described in claim 1, characterized in that, The process of embedding an adaptive scheduling logic core into the warehouse dynamic twin includes: defining a set of scheduling rule bases, which include priority rules, path optimization rules, and conflict avoidance rules; constructing a strategy selector, which dynamically selects applicable scheduling rules from the scheduling rule bases based on the current state of the warehouse dynamic twin; encapsulating the strategy selector and the scheduling rule bases into an adaptive scheduling logic core and deploying it in the warehouse dynamic twin. The constructed strategy selector includes: establishing a state feature extraction module to collect in real time the order backlog, equipment operating load, and path passage efficiency indicators of each logical partition in the warehouse dynamic twin; designing a rule matching engine to calculate the similarity between the feature vector output by the state feature extraction module and the triggering condition of each rule in the scheduling rule base, sorting the rules according to the similarity calculation results, selecting the rule with the highest similarity as the applicable scheduling rule, and setting a dynamic rule switching mechanism so that when the state feature change of the warehouse dynamic twin exceeds the sensitivity threshold, the rule selection process is automatically re-executed.

5. The intelligent scheduling method for logistics warehousing as described in claim 4, characterized in that, The process of performing multi-dimensional task analysis on the warehouse dynamic twin through the adaptive scheduling logic core to generate a task decomposition graph includes: receiving externally input task requests, which include material information, target location, and time requirements; simulating the complete execution flow of the task request in the warehouse dynamic twin; identifying the logical partitions and operational dependencies involved; decomposing the task request into multiple atomic operation steps; and generating a task decomposition graph with nodes and edges based on the operational dependencies.

6. The intelligent scheduling method for logistics warehousing as described in claim 5, characterized in that, The step of performing resource matching operations within the logical partitions of the warehouse dynamic twin based on the task decomposition graph and outputting a resource matching solution includes: traversing each atomic operation step in the task decomposition graph, querying the logical partitions and equipment resources it requires, checking the availability status of the logical partitions and equipment resources within the task time window, allocating available resource instances for each atomic operation step, and generating a resource matching scheme containing resource identifiers and execution time points.

7. The intelligent scheduling method for logistics warehousing as described in claim 6, characterized in that, The step of activating the strategy executor in the adaptive scheduling logic core according to the resource matching scheme and generating a control instruction sequence containing timing constraints includes: the strategy executor determining the specific execution parameters of each atomic operation step according to the resource matching scheme and the scheduling rules in the adaptive scheduling logic core, arranging a strict execution timing sequence for all atomic operation steps to ensure that the time requirements of the task request are met, and compiling the atomic operation steps with execution parameters and timing sequence into a control instruction sequence that can be recognized by the physical execution terminal.

8. The intelligent scheduling method for logistics warehousing as described in claim 7, characterized in that, The step of sending the control command sequence to the physical execution terminal and collecting the feedback data stream from the physical execution terminal includes: sending the control command sequence sequentially to the corresponding AGV, stacker crane or sorting device through the communication interface, and receiving the status code, location information and task completion progress returned by the physical execution terminal in real time to form a feedback data stream.

9. The intelligent scheduling method for logistics warehousing as described in claim 8, characterized in that, The method of using the feedback data stream to correct the state of the warehouse dynamic twin includes: comparing the feedback data stream with the predicted state of the corresponding logical partition in the warehouse dynamic twin, calculating the state deviation value, and if the state deviation value exceeds a preset threshold, triggering the parameter calibration process of the virtual warehouse model and updating the state attributes of the logical partition. The step of comparing the feedback data stream with the predicted state of the corresponding logical partition in the warehouse dynamic twin includes: performing time series alignment processing on the feedback data stream, extracting the actual state data of the physical execution terminal that is consistent with the timestamp of the predicted state; and selecting key state variables for each logical partition, including device location coordinates, task execution progress, and resource occupancy status. The calculation of the state deviation value includes: calculating the absolute error between the actual value and the predicted value for each key state variable, and using a weighted average algorithm to synthesize the absolute errors of multiple variables into a comprehensive state deviation value, wherein the weight coefficients are preset according to the degree of influence of the variables on the system stability.

10. A logistics warehousing intelligent scheduling system, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the intelligent scheduling method for logistics warehousing as described in any one of claims 1 to 9.

Images