Intelligent scheduling method and system for trade warehousing

Abstract

The application discloses a kind of intelligent scheduling method and system of trade warehousing, it is related to warehousing logistics technical field, including the following module: multi-source data aggregation module, for collecting customs supervision state data, shipping schedule data and cargo physical attribute data, and carry out data cleaning and space-time alignment processing, generate multi-dimensional heterogeneous trade data sequence.In the application, through multi-source data aggregation module, break information island, carry out depth fusion with external trade data and internal warehousing data, trade potential energy evaluation module utilizes discrete time Markov chain and time-varying potential field model, can prospectively predict the customs clearance state and the urgency of goods, realizes the intelligent, dynamic evaluation of goods delivery priority, significantly improve the accuracy and timeliness of scheduling decision, virtual packing deduction module ensures the reliability of packing scheme by embedded physical stability check, generates loading topology dependent graph containing delivery sequence constraint.

Description

Technical Field

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

[0002] With the increasing frequency of global trade and the increasing complexity of supply chains, trade warehousing, as a key link connecting production and consumption, has a decisive impact on overall logistics costs and customer satisfaction in terms of efficiency and intelligence. Traditional trade warehousing scheduling and management usually rely on human experience or automated systems based on fixed rules. This management model exposes many technical defects and limitations when facing a complex and ever-changing trade environment.

[0003] Existing warehousing systems suffer from information silos in data acquisition. External trade data, such as customs supervision status and shipping schedules, often exist independently from internal warehousing management data (such as cargo physical attributes and inventory locations), lacking effective aggregation and integration mechanisms. Traditional outbound priority is usually based on simple first-in-first-out (FIFO) or order urgency, failing to fully utilize historical data and real-time status to predict future customs clearance risks and time pressures. In outbound operations, the efficiency and quality of packing operations directly affect subsequent transportation costs and cargo safety. Traditional warehousing scheduling systems lack responsiveness to emergencies or abnormal situations. Existing trade warehousing scheduling systems have significant technical problems in data integration, urgency assessment, packing and relocation coordination, anomaly correction, and equipment collaborative control, urgently requiring a more intelligent, efficient, and robust scheduling system to address these challenges. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies by proposing an intelligent scheduling method and system for trade warehousing.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: an intelligent scheduling system for trade warehousing, comprising the following modules: a multi-source data aggregation module, used to collect customs supervision status data, shipping schedule data, and cargo physical attribute data, and perform data cleaning and spatiotemporal alignment processing to generate a multidimensional heterogeneous trade data sequence; a trade potential energy assessment module, used to receive the multidimensional heterogeneous trade data sequence, predict the customs clearance state transition probability through a discrete-time Markov chain, and calculate the urgency of cargo release by combining a time-varying potential energy field model, and filter and output a set of high-potential-energy goods; and a virtual packing simulation module, used to receive the set of high-potential-energy goods, and perform virtual packing simulation based on an improved genetic strategy and DBL heuristic rules in a virtual three-dimensional space. The system performs iterative packing simulations to generate a loading topology dependency graph containing constraints on the order of goods leaving the warehouse. A reverse timing scheduling module receives the loading topology dependency graph, calculates the spatial entropy between the current physical warehouse location and the ideal outbound timing based on reverse-order logic, and generates a pre-shifting task queue for high-potential-energy goods. A collaborative path control module receives the pre-shifting task queue, assigns tasks using the KM bipartite graph matching algorithm, plans the handling path based on the A* search algorithm with a time window, and generates equipment collaborative control instructions. A dynamic anomaly correction module monitors real-time state changes in multi-dimensional heterogeneous trade data sequences, dynamically updates the high-potential-energy goods set when a circuit breaker threshold is triggered, and synchronously corrects the equipment collaborative control instructions.

[0006] As a further description of the above technical solution: The generation of multidimensional heterogeneous trade data sequences includes: accessing the customs data platform, port EDI system, and warehouse management system through preset heterogeneous communication interfaces to capture customs release status messages, estimated port cut-off time messages, and physical attribute messages of inventory goods in real time; based on the preset correspondence between cargo batch numbers and bill of lading numbers, associating and matching customs release status messages, estimated port cut-off time messages, and physical attribute messages to construct multi-source raw data frames; performing data cleaning and spatiotemporal alignment processing on the multi-source raw data frames, using interpolation algorithms to fill in missing timestamp data, and uniformly mapping all time data to the system's standard time axis to calculate the relative remaining port cut-off time; performing feature encoding and standardization processing on the aligned data, converting the text-type customs release status into discrete numerical features, mapping the relative remaining port cut-off time and physical attribute data to preset numerical intervals to generate normalized feature vectors; and serializing and encapsulating the normalized feature vectors according to the time dimension to generate a multidimensional heterogeneous trade data sequence for input to the trade potential assessment module.

[0007] As a further description of the above technical solution: The generation of high-potential-energy goods sets includes: extracting commodity coding features from multidimensional heterogeneous trade data sequences, and using the K-Means clustering algorithm to divide historical trade data into... Each commodity category subspace; for the current commodity category subspace, the customs supervision status within the historical time window is statistically analyzed from status... transition to state Based on the frequency of occurrence, construct the initial state transition probability matrix for this category. Real-time calculation of the duration of the current cargo's status under the current regulatory conditions. The state inertia coefficient is calculated by calling the preset Logistic decay function. Using the state inertia coefficients to determine the initial state transition probability matrix Nonlinear correction is performed: the diagonal elements are strengthened to maintain the probability of the current state, while the off-diagonal elements are proportionally weakened to reduce the probability of state transition. Simultaneously, normalization is performed to generate a dynamically resident weighted matrix describing the nonhomogeneous process. Construct the current state distribution vector of the goods. The current state is 1, and the rest are 0. Execution is performed using a dynamically resident weighted matrix. Step-by-step discrete-time Markov chain iterative operation Extract the component value corresponding to the release status from the result vector, and output the predicted release probability value of the goods within a preset time window in the future. Establish a gravitational field containing the target. Repulsive field of time The time-varying potential energy field model; the target gravitational field is constructed based on the predicted release probability value, and the formula is: This indicates the passability of goods to be allowed to leave the warehouse. The target gravitational field strength coefficient is typically set to a range of values. The time repulsion field is based on the remaining time window from the current moment to the cutoff time. Constructed using an improved Coulomb repulsive potential field function. This indicates the exponential increase in the urgency of shipping as the shipping date approaches. The time repulsion field strength coefficient is typically set to a range of values. The comprehensive trade potential energy value of the goods is calculated by weighted superposition of the target gravitational field and the time repulsion field. ; Traverse all goods in inventory awaiting shipment and calculate their... The algorithm then uses a max-heap sorting algorithm to select the top-K items with the highest potential energy values, or those with potential energy values ​​exceeding a preset activation threshold. The goods generate a high-potential-energy cargo set.

[0008] As a further description of the above technical solution: The generation of the loading topology dependency graph includes: acquiring the physical attribute data of each cargo in the high-potential-energy cargo set, including length, width, height, and load-bearing capacity; constructing a gene sequence using an integer-based encoding method to generate an initial loading population; each gene bit represents a unique identifier ID for a cargo, and the order of the gene sequence represents the attempted placement order in the virtual container; performing a decoding operation on each individual in the initial loading population, calling the DBL heuristic rule, extracting cargo one by one according to the gene sequence order, and executing the DBL coordinate calculation process to calculate the candidate placement coordinates of each cargo in the three-dimensional coordinate system of the virtual container; initializing the candidate coordinate set, the initial state only contains the origin coordinates of the virtual container. Based on the sorting rule of Z-axis depth first > Y-axis height second > X-axis width last, the points in the candidate coordinate set are sorted. The sorted candidate coordinate points are then traversed sequentially and used as anchor points for the current cargo. Collision detection is performed: it is determined whether the space volume occupied by the current cargo at the anchor point overlaps with the already occupied space set, and whether it exceeds the physical boundary of the virtual container. If the collision detection passes, an embedded physical stability check is further performed: the effective contact area ratio between the bottom surface of the current cargo and the top surface of the cargo already placed below is calculated. If the effective contact area ratio is higher than a preset physical stability threshold, the anchor point is confirmed as the final placement coordinate of the cargo. The volume of the cargo is added to the occupied space set, and new candidate points are generated based on the vertex coordinates of the cargo's upper right, upper left, and lower front corners, updating the candidate coordinate set; otherwise, a penalty function mechanism is triggered, applying a non-linear penalty weight to the fitness value of the individual, eliminating physically infeasible solutions; a dual-objective fitness evaluation system is constructed, including maximizing space utilization and minimizing the combined centroid; the fitness variance of all individuals in the current population is calculated to characterize the population diversity index; the crossover probability of the genetic algorithm is dynamically adjusted based on this index. ) and the probability of variation ( When the population diversity index falls below a preset warning threshold, the mutation probability is automatically increased to escape local optima. After a preset number of iterations, the globally optimal loading scheme on the Pareto optimal front is output. This scheme includes the final three-dimensional coordinates of each item in the high-potential-energy cargo set within the virtual container. and rotational attitude parameters; based on the globally optimal loading scheme, a virtual ray source is set along the negative Z-axis direction of the container outbound operation to perform geometric projection detection on each piece of cargo; for any two pieces of cargo and Determine whether the physical obstruction condition is met: that is, the goods On the projection plane of the outbound direction, with the goods There are overlapping regions ( ), and goods The depth coordinates are located on the cargo The outbound route ahead ( If the conditions are met, then the goods are deemed acceptable. For goods Physical occlusion of the foreground, establishing a system based on point to One-way dependency edges; using goods in the high-potential-energy goods set as nodes and one-way dependency edges as connections, construct an adjacency matrix describing the loading dependency logic. , of which elements Indicates goods Blocking goods Traverse the adjacency matrix Each non-zero element is mapped to an edge in the directed graph to construct the initial directed graph structure. A depth-first search is performed on the initial directed graph structure to detect loops. If a closed loop is detected, the current loading scheme is determined to have a logical deadlock, and a globally optimal loading scheme is regenerated. If there is no closed loop, the directed graph structure is instantiated into a loading topology dependency graph, and the in-degree of each node is calculated, where nodes with an in-degree of 0 represent goods that can be physically removed at the current moment.

[0009] As a further description of the above technical solution: The generation of the pre-shift task queue includes: receiving the loading topology dependency graph, performing a topology sorting algorithm on it, and transforming the graph structure into a linear ideal outbound sequence. ;in, The goods with an in-degree of 0 and the highest priority in the sequence represent the goods that should be loaded first. If there are parallel branches in the graph, a secondary sort is performed based on the comprehensive trade potential energy value of the goods in the high-potential-energy goods set, with those having higher potential energy values ​​ranked at the top of the sequence. The current physical storage coordinates of each goods in the ideal outbound sequence are obtained. And calculate the Manhattan distance from that coordinate to the warehouse departure platform. Define spatial entropy. This is used to characterize the degree of disorder between the current physical storage state and the ideal outbound timing; the calculation formula is: ,in The weighting coefficient is related to the order position of the goods in the ideal outbound sequence. There is a negative correlation (i.e., the higher the order of a cargo, the greater its distance weight; if it is farther from the platform, the entropy value increases significantly); the calculated spatial entropy value... Compare with the preset recombination activation threshold; if If the threshold is exceeded, the subsequent shift task generation process is triggered; a dynamic reorganization area is defined in the temporary storage area for shipments in the warehousing system, which contains several preset buffer grids; mapping is performed based on reverse time sequence logic: the ideal outbound sequence is... Mapped to the dynamic reorganization area, the mapping rule is: the first item in the sequence. Mapped to the buffer grid closest to the outbound platform, the last item in the sequence is sorted. Mapping to the buffer grid furthest from the outbound platform determines the target buffer coordinates for each item within the reassembly zone. For each item in the ideal outbound sequence, construct a shift task from the physical storage coordinates to the target buffer coordinates, and generate a task tuple. Arrange all task tuples in reverse order of the ideal outbound sequence (i.e., process goods at the end of the sequence first to avoid blocking the front path) or based on the unlocking order of dependency relationships, and encapsulate them to generate a pre-shifted task queue, which serves as the input basis for the collaborative path control module.

[0010] As a further description of the above technical solution: The generation of equipment collaborative control instructions includes: parsing the pre-shifted task queue, extracting task tuples, and constructing a task set. ; Obtain the coordinates, power levels, and real-time load status of all automated material handling equipment within the warehouse, and construct a set of available equipment. Construct a weighted matching model for a bipartite graph, where the edge weights are... Defined as the device-task response cost; introduces a comprehensive trade potential energy value as an adjustment factor to nonlinearly reduce the response cost of high-potential-energy tasks, making them easier for nearby devices to capture; uses the KM algorithm to solve for the optimal match and outputs the assignment result; initializes the process including spatial dimensions. With the time dimension The system uses a global spatiotemporal grid map and introduces the concept of a dynamic congestion potential energy field. For known high-frequency operation areas or intersections with planned routes, congestion potential energy values ​​are superimposed on the corresponding spatiotemporal nodes. This value decays over time and is used to characterize the traffic density of the area at a specific future time. For each pair of matches, a path is planned using the improved time window A* algorithm based on potential energy gradient: During the heuristic search process, the evaluation function is reconstructed as follows: ; For the actual cost of movement, Congestion potential energy was introduced for Manhattan distance estimation. , To mitigate the impact of the collision coefficient, a strategy of trading space for time is employed to adjust the collision coefficient. , The value of this value is negatively correlated with the overall trade potential energy of the goods currently being transported; the higher the potential energy of the goods, the better. The smaller the potential energy, the less sensitive the algorithm is to congestion, and the more inclined it is to plan shortcuts; the lower the potential energy of the goods, the better. The larger the priority level, the more likely it is to plan detour routes to reduce overall system congestion, allowing low-priority vehicles to proactively take detours and giving fast lanes to vehicles with greater potential energy; during the search process, if the current node is in the next time step... If the path falls within the occupancy window of other high-priority devices, an in-place waiting strategy or partial replanning will be implemented. After successful path planning, the generated path node sequence will be spatiotemporally locked in the global spatiotemporal grid map, and the congestion potential value of the area covered by the path will be updated synchronously. This influences the path planning of subsequent low-priority devices; and generates a collision-free spatiotemporal path sequence. ; It represents the horizontal and vertical physical coordinates in the warehouse grid map, which precisely correspond to the physical location of the shelves or aisles; The system time slice represents a discrete time slice, each time slice corresponding to the standard duration required for the device to perform a single atomic action (such as moving one unit, lifting, or rotating); the spatiotemporal path sequence is converted into machine code containing underlying driving parameters, including motor speed, steering angle, and lifting height, and encapsulated into device collaborative control instructions and issued.

[0011] As a further description of the above technical solution: The dynamic anomaly correction module's operating mechanism includes: allocating a fixed-length circular buffer in system memory, and constructing a sliding time window listener that follows a first-in, first-out (FIFO) principle, with its window length set to... Each time step; the multidimensional heterogeneous trade data sequence generated by the preceding module is mapped into the listener in real time to form the feature matrix of the current time window. ,in This is the normalized feature vector at the current moment; whenever a new data frame is received... The listener performs a sliding operation: removing the oldest feature vector. and merged into The time span of the window remains constant; inside the sliding time window listener, the vectors of two adjacent frames in the feature matrix of the current time window are analyzed. and Perform dual-channel difference analysis: Discrete state channel (for customs supervision status): Extract the corresponding discrete numerical features of customs status from the feature vector and calculate the Hamming distance. ;like This indicates a discrete transition in the state; further query the preset risk state mapping table to determine whether the new state belongs to the "inspection", "detention", or "refund" set; if it does, a hard blocking event is determined to have occurred; continuous time-sensitive channel (for ship schedule cut-off time): extract the corresponding relative remaining cut-off time value from the feature vector and calculate the first-order difference value. Define the time jitter threshold. ,like and If the remaining time is significantly shortened beyond the normal elapsed rate, a soft time-sensitive event is determined to have occurred. When any of the above events is detected, the system determines that a circuit breaker threshold has been triggered, generating an abnormal interruption signal containing the unique identifier ID of the affected goods and a mutation type tag. Based on the unique identifier ID in the abnormal interruption signal, the target goods are located in the current high-potential-energy goods set. According to the mutation type tag, the potential energy reconstruction logic is executed: if it is a hard blocking event, the predicted release probability value of the goods in the discrete-time Markov chain is forcibly set to zero, causing its comprehensive trade potential value to drop to negative infinity, thereby removing the goods from the high-potential-energy goods set and creating a vacant computing power slot; if it is a soft time-sensitive event, the post-mutation... The remaining port cutoff time is used to recalculate the time repulsion field strength in the time-varying potential energy field model and update its comprehensive trade potential energy value. For the generated empty computing power slots, the remaining goods waiting to be shipped in the inventory are traversed, and the maximum heap sort algorithm is used to select the candidate goods with the highest current potential energy value that are not in the set, and they are added to the high potential energy goods set to complete the dynamic update of the set. The system's task status register is queried to check whether the affected goods have generated tasks in the pre-shift task queue. If they have been generated, it is further determined whether the task has been assigned to a specific device through the KM bipartite graph matching algorithm. If the task is in the "assigned and executing" state, the target device ID for executing the task is locked, and the current spatiotemporal path sequence of the device is obtained. and the current spatiotemporal node The system sends a task interruption signal to the target device, causing it to perform a soft braking operation, allowing it to stop at the nearest dockable node and marking that node as the new temporary starting point. It then invokes the virtual packing simulation module and the reverse timing scheduling module to generate a new relocation task for the newly added candidate cargo. Starting from the temporary starting point and ending at the target buffer coordinates of the new task, it calls the improved time window A* algorithm based on potential energy gradients for path replanning, generating a new spatiotemporal path sequence. Use it to cover the original In the remaining segment, the corrected device collaborative control instructions are generated and issued, while the global spatiotemporal grid resources occupied by the original task are released and the congestion potential energy value is updated.

[0012] As a further description of the above technical solution: It also includes a virtual-real mapping feedback calibration module, whose operating mechanism includes: after the automated handling equipment completes the execution of a collaborative control command, it triggers the visual sensing equipment deployed on the outbound platform to collect the current depth point cloud data inside the container; it performs background filtering on the depth point cloud data and extracts the physical point cloud cluster of the latest put-in goods. ,in For the number of point clouds, For the first The spatial coordinates of each point are calculated; the actual landing posture of the cargo in physical space is calculated using Principal Component Analysis (PCA); the geometric centroid of the physical point cloud cluster is calculated. : Construct the covariance matrix : For the covariance matrix Perform eigenvalue decomposition to obtain the eigenvector corresponding to the largest eigenvalue. The direction of this feature vector is identified as the principal axis direction of the cargo, combined with the geometric centroid. Generate the actual landing posture including center coordinates and rotation angle. ; Invoke the globally optimal loading scheme generated in the virtual packing simulation module to extract the theoretical placement posture of the cargo. ; Calculate the difference between the actual landing posture and the theoretical landing posture, and construct the loading deviation vector. : Determine the loading deviation vector If the module length exceeds the preset tolerance threshold, then a bounding box expansion operation is performed on the virtual model representing the current goods in the occupied space set in the virtual 3D space to obtain the original dimensions from the goods' physical attribute data. The corrected physical dimensions of the expansion are calculated using the following formula, which includes the length, width, and height. : , , , This is the error sensitivity coefficient, used to amplify the weight of the impact of deviation on space occupancy; its value range is typically set to [value range missing]. ; The safety buffer gap, measured in millimeters, is used to allow for minor extra space in case of instability during robotic arm grasping or stacking; it is based on the center coordinates of the actual placement posture. With expansion physical dimensions The system regenerates the physical occupancy voxel of the cargo and updates the occupied space set accordingly. It extracts all unexecuted successor cargo nodes from the loading topology dependency graph. Based on the updated occupied space set, it recalculates the available corner point set within the virtual container. It performs secondary collision detection on the theoretical placement posture of the successor cargo: it checks whether the successor cargo overlaps with the updated physical occupancy voxel in volume at its theoretical position. If spatial interference conflict is detected, it determines that the original globally optimal loading scheme is invalid. It keeps the currently loaded cargo stationary and only re-triggers the DBL decoding and genetic optimization process in the virtual container loading deduction module for the remaining unloaded high-potential-energy cargo set to generate a new remaining loading scheme and corresponding equipment collaborative control instructions to eliminate the physical collision risk caused by accumulated errors.

[0013] As a further description of the above technical solution: A smart scheduling method for trade warehousing, applied to any one of the systems described in claims 1-8, includes the following steps: Step 1: Construction of multidimensional heterogeneous data sequence: Collect customs release status messages, estimated cut-off time messages, and physical attribute messages of inventory goods through heterogeneous communication interfaces to construct multi-source raw data frames; perform spatiotemporal alignment and interpolation filling on the multi-source raw data frames, and perform feature encoding and standardization processing to generate multidimensional heterogeneous trade data sequence. Step 2: Trade Potential Assessment and Goods Screening: Construct an initial state transition probability matrix based on historical trade data and calculate the state dwell time of goods under the current regulatory state; call the dwell decay function to generate a dynamic dwell weighting matrix, and use this matrix to perform multi-step discrete-time Markov chain iterative operation to obtain the predicted release probability value; establish a time-varying potential energy field model containing the target gravitational field and the time repulsive field, and calculate the comprehensive trade potential energy value by weighting and superimposing the two, and screen and output a set of high potential energy goods accordingly; Step 3: Virtual Packing and Topology Generation: An initial loading population is constructed from the high-potential-energy cargo set using an integer permutation-based encoding method; DBL decoding with embedded physical verification is performed: while calculating the candidate placement coordinates according to the DBL (Deep-Bottom-Left) heuristic, the effective contact area ratio is calculated to perform embedded physical stability verification, and a penalty function mechanism is used to eliminate physically infeasible solutions; Adaptive multi-objective co-evolution is performed: the crossover and mutation probabilities are dynamically adjusted based on the population diversity index to output the globally optimal loading scheme; geometric projection detection is performed based on the globally optimal loading scheme to identify physical occlusion predecessor relationships, and a loading topology dependency graph is constructed and output; Step 4: Reverse Scheduling and Task Generation: Execute a topology sorting algorithm on the loading topology dependency graph to generate an ideal outbound sequence; calculate the physical storage coordinates and spatial entropy value of the goods in the sequence. When the entropy value exceeds the threshold, map the goods to the target buffer coordinates of the dynamic reorganization area based on the reverse timing logic to generate a pre-shift task queue. Step 5: Collaborative Path Planning and Command Issuance: Construct a bipartite graph weighted matching model, introduce a comprehensive trade potential value to adjust the response cost, and use the KM algorithm to complete task assignment; overlay a dynamic congestion potential energy field on the global spatiotemporal grid map, and use an improved time window A* algorithm based on potential energy gradient to plan a collision-free spatiotemporal path sequence; among which, the avoidance coefficient is used. The algorithm's sensitivity to congestion is adjusted, and this coefficient is negatively correlated with the overall trade potential of goods; the path sequence is encapsulated into equipment collaborative control commands and issued. Step 6: Virtual-Real Feedback Calibration and Closed-Loop Control: After the equipment executes the command, depth point cloud data is collected and physical point cloud clusters are extracted; principal component analysis is used to calculate the actual landing posture of the cargo and compare it with the theoretical landing posture to generate a loading deviation vector; when the deviation exceeds the threshold, bounding box expansion operation is performed to update the occupied space set, and secondary collision detection is performed on subsequent cargo; if spatial interference conflict is detected, local replanning is triggered; at the same time, a sliding time window listener is used to perform dual-channel differential analysis on the multidimensional heterogeneous trade data sequence; when a hard blocking event or a soft time-sensitive event is identified and triggers the circuit breaker threshold, the high-potential-energy cargo set is dynamically updated and the corrective equipment cooperative control command is preempted.

[0014] The present invention has the following beneficial effects: 1. In this invention, firstly, a multi-source data aggregation module breaks down information silos, deeply integrating external trade data with internal warehousing data. The trade potential assessment module, utilizing discrete-time Markov chains and time-varying potential field models, can proactively predict the customs clearance status and urgency of goods leaving the warehouse, achieving intelligent and dynamic evaluation of goods' outbound priority. This significantly improves the accuracy and timeliness of scheduling decisions. The virtual packing simulation module not only considers space utilization but also ensures the reliability of the packing scheme through embedded physical stability verification, generating a loading topology dependency graph containing outbound order constraints. The reverse time-series scheduling module, based on this topology graph, through... Spatial entropy evaluation and reverse logic generation of pre-shift tasks effectively resolve the contradiction between packing schemes and storage space constraints, reduce unnecessary shifts, and improve overall operational efficiency. The dynamic anomaly correction module can monitor the state mutations of multidimensional heterogeneous trade data sequences in real time, and dynamically update the high-potential cargo set according to the mutation type (hard block or soft timeliness), and promptly correct equipment collaborative control commands. This circuit breaker mechanism and rapid response capability enable the system to quickly adjust scheduling strategies when facing emergencies such as changes in customs policies and adjustments to shipping schedules, avoiding the problems of passive response and low efficiency in traditional systems, and significantly improving the robustness and adaptability of the system.

[0015] 2. In this invention, the collaborative path control module uses the KM bipartite graph matching algorithm for task assignment and plans the transport path based on the A* search algorithm with a time window. It also introduces a dynamic congestion potential energy field and an avoidance coefficient λ, enabling conflict-free and highly efficient collaborative operation of multiple devices in complex warehousing environments. In particular, the negative correlation between the avoidance coefficient λ and the comprehensive trade potential value of the goods ensures that high-priority goods can pass first, further optimizing the overall scheduling efficiency. The virtual-real mapping feedback calibration module collects actual loading data through visual sensing devices and compares it with the theoretical loading plan to calculate the loading deviation. When the deviation exceeds a threshold, the system can automatically perform bounding box expansion and local replanning, effectively correcting the difference between the virtual model and physical reality, avoiding the risk of physical collisions caused by error accumulation, and ensuring the executability and accuracy of the scheduling plan. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the system architecture of the present invention. Detailed Implementation

[0017] 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.

[0018] Reference Figure 1This invention provides an embodiment of an intelligent scheduling system for trade warehousing, comprising the following modules: a multi-source data aggregation module, used to collect customs supervision status data, shipping schedule data, and cargo physical attribute data, and perform data cleaning and spatiotemporal alignment processing to generate a multidimensional heterogeneous trade data sequence; a trade potential energy assessment module, used to receive the multidimensional heterogeneous trade data sequence, predict the customs clearance state transition probability through a discrete-time Markov chain, and calculate the urgency of cargo release by combining a time-varying potential energy field model, and filter and output a set of high-potential-energy cargo; and a virtual packing simulation module, used to receive the set of high-potential-energy cargo, and perform iterative operations in a virtual three-dimensional space based on an improved genetic strategy and DBL heuristic rules. The system includes a container simulation module to generate a loading topology dependency graph containing constraints on the order of goods leaving the warehouse; a reverse timing scheduling module to receive the loading topology dependency graph, calculate the spatial entropy value between the current physical warehouse location and the ideal outbound timing based on reverse order logic, and generate a pre-shifting task queue for high-potential-energy goods; a collaborative path control module to receive the pre-shifting task queue, assign tasks using the KM bipartite graph matching algorithm, plan the handling path based on the A* search algorithm with a time window, and generate equipment collaborative control instructions; and a dynamic anomaly correction module to monitor the state changes of multidimensional heterogeneous trade data sequences in real time, dynamically update the set of high-potential-energy goods when the circuit breaker threshold is triggered, and synchronously correct the equipment collaborative control instructions.

[0019] Example 1: The system establishes connections with external data sources through pre-defined heterogeneous communication interfaces. Through the customs data platform interface, it captures customs release status messages in real time, including cargo declaration numbers and regulatory status codes (such as "released," "under inspection," and "pending declaration"). Through the port EDI (Electronic Data Interchange) system interface, using the bill of lading number as an index, it captures the estimated arrival time (ETA), estimated departure time (ETD), and cut-off time for the corresponding vessel in real time, generating an estimated cut-off time message. Through the internal warehouse management system (WMS) interface, it obtains the physical attributes (length, width, height, weight, pressure rating, and packaging type) of the inventory goods, generating a physical attribute message for the inventory goods.

[0020] Since the aforementioned messages originate from different systems and lack a unified association, data association is performed based on a pre-defined unique correspondence between the cargo batch number (BatchID) and the bill of lading number (B / LNo.).

[0021] The system uses the cargo batch number as the primary key to perform row-level concatenation of customs release status messages, estimated cut-off time messages, and physical attribute messages for the same batch of goods, constructing a cleaned, multi-source raw data frame. .

[0022] against To address potential issues such as missing or inconsistent timestamps, the system performs the following steps: Timeline mapping: Establishing a unified system standard timeline. The update times of customs data and port data are uniformly mapped onto this time axis. Interpolation fill: For a certain moment... Missing timestamp data is filled in using interpolation algorithms (such as linear interpolation). Assume... and If the data for a given time is known, then the missing time... ( ) data values The calculation is as follows: Calculation of relative remaining cut-off time: Based on the system's standard timeline, calculate the current time. Estimated cut-off time for the ship The difference, i.e., the relative remaining cut-off time. : This indicator is used to quantify the time-repulsive field strength in subsequent tests.

[0023] To eliminate the dimensional differences between data of different dimensions, a normalized feature vector is generated. Discrete numerical feature transformation: For customs clearance status in text format, label encoding is used to convert it into discrete numerical features. For example: Set the mapping rule as {"Released":1, "Pending Declaration":2, "Under Inspection":3, "Detained":4, "Rejected":5}. Numerical mapping and normalization: for the relative remaining cut-off time. and physical property data (set as) , including long ,Width ,high ,Heavy ), mapping it to a preset numerical range (e.g. Normalized values ​​are generated using the Min-Max normalization formula. : in, For raw data (such as or ), and These are the preset minimum and maximum bounds for this feature dimension.

[0024] After the above processing, for each time step The system generates a normalized feature vector containing all key information. : ,in, This refers to the normalized relative remaining cut-off time for Hong Kong. The normalized physical properties.

[0025] The system normalizes the generated feature vectors according to the time dimension. Perform serialization and encapsulation. Set a sliding window length or history length. , will continue The vector combination at each time step ultimately generates a multidimensional heterogeneous trade data sequence for input to the trade potential assessment module. : The sequence It includes the evolution trend and physical characteristics of goods over time, providing a standardized data foundation for subsequent prediction of customs clearance probability and calculation of the urgency of shipment.

[0026] Example 2: The system extracts commodity code (HSCode) features from multidimensional heterogeneous trade data sequences. Considering the significant differences in customs clearance characteristics among different types of goods (such as fresh produce and electronic products), the system first uses the K-Means clustering algorithm to divide the historical trade data into... Each product category subspace.

[0027] For the commodity category subspace to which the goods to be evaluated belong, statistically analyze the transitions of customs supervision status within that category over a historical time window. Let the customs supervision status space be denoted as . (For example: =Submitted, =Under inspection =Already released, etc.).

[0028] Construct the initial state transition probability matrix for this category. , of which Line number Column elements Indicates from state Leap to The probability of: At this point, we get It is a static matrix based on historical average levels. To reflect the impact of "how long the current cargo has been in a certain state" on future states, the system calculates the state dwell time of the current cargo in the current regulatory state in real time. The state inertia coefficient is calculated by calling the preset Logistic decay function. This function characterizes how the probability of a state changing decreases as the dwell time increases (i.e., the inertia of maintaining the current state increases): ,in, These are preset empirical constants used to fit the shape of the Logistic curve. The state inertia coefficient is utilized. For the initial state transition probability matrix Perform nonlinear corrections to generate a dynamically resident weighted matrix. .

[0029] The revised rules are as follows: Enhance the probability of diagonal elements maintaining their current state: Decrease off-diagonal elements (probability of state transition): ,in This is the normalization adjustment factor. Row normalization: ensures that the sum of the probabilities of each row of the matrix is ​​1. At this time, it is generated It is a dynamic matrix describing a nonhomogeneous process, which changes with... The state changes in real time with the changes in the cargo. Construct the current state distribution vector of the cargo. If the current cargo is in a certain state... Then the first vector One component is 1, and the rest are 0. This is achieved using a dynamic resident weighting matrix. implement Step-by-step discrete-time Markov chain iterative operation to calculate the future 1st... State distribution at each time step : From the result vector Extract the corresponding "released" status (set to...). The component value of ) is the predicted release probability value of the goods within a future preset time window. A state dwell time was introduced. If a batch of goods has been stuck in the "inspection" state for 3 days without any action, the probability of it being released in the next hour is usually lower than that of goods that have just entered the inspection state for 1 hour (or may be higher depending on the data distribution). By using a dynamic dwell time weighting matrix, the algorithm can perceive the impact of time passage on the probability, which solves the lag problem of the static model.

[0030] The system establishes a target gravitational field ( ) and time repulsion field ( The time-varying potential energy field model quantifies the outflow dynamics. Target gravitational field ( This indicates the degree to which goods are released by customs and are feasible for shipment. It is based on the predicted release probability value. Build: ,in, The target gravitational field strength coefficient is set to a value range of [value range missing]. This logarithmic function form ensures that the gravitational gain resulting from the increase in probability exhibits diminishing marginal returns. (Time repulsion field) This indicates the urgency of cargo leaving the site as the cut-off time approaches. Based on the relative remaining cut-off time calculated in Example 1 (denoted as...). Constructed using an improved Coulomb repulsive potential field function: ,in, The time repulsion field intensity coefficient is set to a value range of [value range missing]. .when At this time, the repulsive field strength increases exponentially, forcibly pushing the goods out of the warehouse. By linearly weighting and superimposing the two component potential fields, the comprehensive trade potential energy value of the goods at the current moment can be calculated. : ,in, and These are the weighting coefficients, and The system iterates through all goods awaiting shipment in the inventory and calculates the cost of each item. .

[0031] The potential energy values ​​of all goods are maintained using a max-heap sorting algorithm. Method one is to directly select the top-K goods at the top of the heap. Method two is to select all... Goods with a preset activation threshold. The final output list of goods is the set of high-potential-energy goods, which will serve as the input object for the virtual packing simulation module.

[0032] Example 3: The system acquires the physical attribute data (length) of each item in the high-potential-energy cargo set. ,Width ,high and load-bearing capacity rating).

[0033] The encoding method uses integer permutations. Assume the set contains... Items, the set of item IDs is .

[0034] A chromosome is represented as The sequence index represents the order in which virtual bins are attempted. The system generates this randomly. These chromosomes constitute the initial loaded population.

[0035] For each individual in the population, perform the following decoding process: Initialization: Create virtual containers, initialize the occupied space set to empty, and initialize the candidate coordinate set to... (i.e., origin). DBL coordinate calculation: Extract goods according to gene sequence order. For each goods, traverse all points in the candidate coordinate set. Based on " Axis (depth) priority > Axis (height) is the next most important factor. Candidate points are sorted according to the rule of "width last". Collision detection: The first candidate point after sorting is used as the anchor point. It is then determined whether the geometry generated by the cargo at that anchor point exceeds the boundary of the virtual container. Or, it may overlap in volume with existing geometry in an already occupied space set. Embedded physical stability check: If collision detection passes, calculate the effective contact area ratio. : .in, This represents the total area of ​​the bottom surface of the current cargo. This refers to the contact area between the bottom surface of the current cargo and the top surface of the cargo already placed below (or the bottom surface of the container). With the preset physical stability threshold (For example, 0.75) Compare: If If the verification passes, the anchor point is confirmed as the final placement coordinate. Add the cargo to the already occupied space set. Generate new candidate points based on the cargo's new vertices (top right, top left, bottom front) and update the candidate coordinate set. If... If this occurs, the penalty function mechanism is triggered. The system constructs a bi-objective fitness function: maximizing space utilization. Minimize the center of gravity of the combination: (Assuming the Z-axis represents the depth / vertical direction, the goal is to lower or move the center of gravity further inward.)

[0036] Calculate the variance of the fitness values ​​of all individuals in the current population to obtain the population diversity index. Genetic parameters, including crossover probability, are dynamically adjusted based on this indicator. and mutation probability The adjustment formula is as follows: in A preset warning threshold is set. When diversity is low, the mutation probability is increased to escape local optima. After iterative optimization, the globally optimal loading scheme located on the Pareto front is output, including the final coordinates of each item. and rotational attitude. Based on the globally optimal loading scheme, the system sets a virtual ray source along the outbound operation direction (negative Z-axis direction). For any two items... and The system performs geometric projection detection to determine if physical occlusion exists. The decision logic includes two sub-conditions: Projection overlap: Calculate the cargo... and goods The projected rectangle on the XY plane (the section perpendicular to the outbound direction). If the overlapping area... The first condition is met. Depth order: Compare the depth coordinates of the two. If the goods... depth coordinates Smaller than goods depth coordinates (Right now , indicating goods Located in the cargo (The goods must be located ahead of the outbound route) and meet the second condition. If both conditions are met simultaneously, then the goods are determined to be... For goods The physical occlusion of the forerunner. At this point, the system establishes a path consisting of... point to One-way dependent edges ( ), indicating "must be removed first" Only then can it be removed The system uses goods in the high-potential-energy goods set as nodes and unidirectional dependency edges as edges to construct an initial directed graph structure. This structure corresponds to an adjacency matrix. ,in: The system performs a depth-first search (DFS) on the graph structure to detect loops. If a closed loop is found (e.g., A blocks B, B blocks A), it indicates a deadlock in the loading scheme, and a new scheme needs to be generated. If no closed loop is found, the graph is instantiated as a loading topology dependency graph. The system traverses the graph and calculates the in-degree of each node. Nodes with an in-degree of 0 (i.e., no cargo blocking them) are marked as objects that can be physically manipulated at the current moment, providing basic data for the subsequent reverse timing scheduling module.

[0037] Example 4: The system receives a loading topology dependency graph containing nodes and unidirectional dependent edges. To transform the complex network of dependencies into a linear job order, the system executes a topological sorting algorithm.

[0038] The algorithm flow is as follows: Maintain an in-degree table to record the current in-degree of each node in the graph. Initialize a "zero-in-degree queue" and add all nodes with an in-degree of 0 to this queue. The secondary sorting strategy for parallel branches: When there are multiple nodes in the zero-in-degree queue (i.e., multiple goods are not physically blocked at the current moment and can be shipped out in parallel), the system calls the high-potential-energy goods set data generated in Example 2 and reads the comprehensive trade potential value of these goods. Pop up first The highest-priority node. When physical conditions allow, goods with high trade urgency are prioritized at the top of the sequence. Each time a node is popped, it is added to the ideal outbound sequence. Then, "delete" the node and its emanating edges, and update the in-degree of the remaining nodes. Repeat this process until the graph is empty. This ultimately generates a linear, ideal outbound sequence. .in, The representative should be the first cargo to be loaded. This represents the last cargo to be loaded. The system retrieves the sequence. Each item Current physical storage coordinates in the 3D library Set the center coordinates of the departure platform as follows: Calculate the Manhattan distance from each shipment to the outbound platform. : Define the spatial entropy value. This metric quantifies the degree of disorder between the "current inventory distribution" and the "ideal outbound order" (i.e., the order of goods / services urgently needed to be shipped). If it's buried at the deepest part of the warehouse, the entropy value will be extremely high. The calculation formula is as follows: ,in, The weighting coefficient is related to the goods in the ideal outbound sequence. Sort position in ( express The correlation is negative. This embodiment uses an inverse proportional weighting function: ,in This is a normalization constant. This means that... The distance to the platform has the largest weight; if it is far from the platform, it will greatly increase the overall entropy value. ;and The distance weight is the smallest, and its position has little influence.

[0039] The calculated spatial entropy value With the preset recombination activation threshold Compare: If If the current inventory distribution is determined to be extremely disordered, potentially leading to outbound congestion and triggering subsequent relocation processes, then... The current distribution is deemed acceptable, and no pre-shift task is generated. A dynamic reorganization area is designated within the temporary shipping area of ​​the warehousing system. This area includes... A preset buffer grid These grids are arranged in order of distance from the departure platform, i.e. .

[0040] The system performs mapping operations based on reverse timing logic to generate the ideal outbound sequence. Goods in the sequence are assigned to the buffer grid: the first item in the sequence. (Most urgent) Map to the buffer grid closest to the departure platform. Sort the sequence by... Goods Mapped to the Nearby buffer grid The last item in the sequence. Mapped to the farthest buffer grid This determines the target buffer coordinates for each item within the reorganization zone. .

[0041] For sequences Each item in Construct the shift task tuple: ,in, Inherited from the comprehensive trade potential value of the goods. To prevent path deadlock during transport (i.e., first...) It was moved to the entrance and blocked the way. (The path out), the system arranges and encapsulates tasks in reverse order of the ideal outbound sequence.

[0042] The generated pre-shift task queue for: This queue logic ensures that the equipment first moves the goods that are "delivered last and placed deep in the reorganization area" and then moves the goods that are "delivered first and placed at the door" last, thus achieving orderly stacking from the inside out. This queue will serve as the input basis for the collaborative path control module.

[0043] Example 5: The system parses the pre-shift task queue and extracts the previous task. A tuple of tasks to be executed is used to construct a task set. Simultaneously, by using the warehouse control system (WCS) interface, all automated handling equipment in an "idle" or "about to be idle" state is acquired to construct a set of available equipment. .

[0044] Construct a weighted matching model for a bipartite graph. The task in the middle is the left node. The device in the diagram is the right-hand node.

[0045] Calculate each edge weight This weight is defined as the "device-task response cost".

[0046] First, calculate the basic physical cost. (Based on Manhattan distance and battery level): Where (x_{start,i},y_{start,i})$ are the starting coordinates of the task. The current coordinates of the device. This represents the remaining battery percentage of the device. This is the power weighting coefficient.

[0047] Introducing the comprehensive trade potential value calculated in Example 2 As a regulating factor, the base cost is non-linearly reduced. (The corrected weights are then used.) The calculation formula is: ,in, Potential energy sensitivity coefficient ( If the task of Very high (high potential energy goods), then Approaching 0, making The weights decrease significantly. Under the minimum weight matching logic, smaller weights represent lower costs, making high-potential tasks more easily "captured" by nearby devices. The KM (Kuhn-Munkres) algorithm is used to find the minimum weight perfect matching (or the negative form of the maximum weight matching) of this bipartite graph, outputting the optimal assignment result. .

[0048] System initialization global spatiotemporal grid diagram . These are physical raster coordinates. This is a discretized system time slice. The standard duration for a single atomic action (moving one unit) is set to... Then the first Each time slice corresponds to physical time A dynamic congestion potential energy field is introduced. For areas covered by existing planned routes, the congestion potential energy value is calculated. Let a certain node At any moment If the node is occupied by another device, then the node will be in the nearest time. The congestion potential energy is: .in This represents the baseline congestion intensity. This formula characterizes how congestion potential energy decreases as the time difference increases. For each pair of matches... Plan the path from the current location of the equipment to the mission objective.

[0049] In the heuristic search process, nodes are defined. Evaluation function : The definitions are as follows: : The actual cost (number of steps) to move from the starting point to the current node. : The Manhattan distance estimate from the current node to the destination. : The congestion potential energy value of the current spatiotemporal node. Avoidance coefficient. Adjustment. The value of this makes it related to the overall trade potential value of goods. Negative correlation: .in To achieve the maximum avoidance coefficient, To adjust the gain. High potential energy cargo ( big): The algorithm ignores congestion potential energy. , Mainly composed of The decision leans towards planning straight, intersecting shortcuts (seizing the fast lane). Low-potential goods ( Small): If a certain node High, then A significant increase will cause the algorithm to favor searching. Low-lying nodes allow for the planning of alternative routes (active avoidance).

[0050] When searching for expanded nodes, check the next time step. Does the target mesh fall within the occupied time window of other high-priority devices (i.e., (If the "impassable" threshold is reached). If a conflict occurs, a wait-in-place strategy is implemented: insert a new path into the path. This means the coordinates remain the same, but the time increases. If there are no conflicts, continue the search.

[0051] After successful path planning, a node sequence is generated. The system immediately performs spatiotemporal locking on the path in the global spatiotemporal grid map, and assigns the corresponding nodes... Mark as occupied and trigger an update of the dynamic congestion potential field, increasing the area's capacity. This value can affect the planning of subsequent low-priority devices.

[0052] The final generated spatiotemporal path sequence The system analyzes the changes between adjacent spatiotemporal nodes and converts them into machine code: If This generates the instruction MOVE_X_POS. If... and The command WAIT is generated. Upon reaching the destination, the command LIFT_UP (lift) is generated. This code is then encapsulated into a device coordination control command and sent to the corresponding automated handling equipment via a wireless network for execution.

[0053] Example 6 The system allocates a fixed-length memory space. (For example A circular buffer is constructed. A sliding time window listener following the first-in-first-out (FIFO) principle is built. The multidimensional heterogeneous trade data sequence generated in Example 1 is used. Mapped into the listener in real time. At any given moment. The listener maintains the feature matrix of the current time window. : .in This is the normalized feature vector at the current moment. Each time a new data frame is generated... Arrive, Remove and merged into .

[0054] Inside the listener, extract the vectors of two adjacent frames. and Perform dual-channel difference analysis: Discrete state channel (for customs supervision status): Extract discrete numerical features from the vector. and Calculate the Hamming distance. : ,like Query the preset risk status mapping table. For example, if If the corresponding status is "under inspection" or "detained", then a hard blocking event is determined to be triggered.

[0055] Continuous time-sensitive channel (for ship cut-off times): Extracts the relative remaining cut-off time from the vector. and Calculate the first-order difference value. : Define the time jitter threshold (For example Hours, and the normal passage of time must be taken into account. ).like If the remaining time suddenly decreases significantly, and the shipping schedule is brought forward, then a soft time-sensitive event is determined to be triggered. When any of the above events is triggered, the system generates an abnormal interruption signal, which includes the unique identifier ID of the affected cargo and a mutation type tag.

[0056] The system locates the target cargo in the high-potential-energy cargo set generated in Example 2 based on a unique identifier (ID).

[0057] Execute potential energy reconstruction logic based on mutation type label: Case A: Hard blocking event, forcibly modify the predicted release probability value in Example 2. According to the formula in Example 2 ,at this time Typically, penalties are set to reduce the overall trade potential value. Remove the item from the collection to create an empty computing power slot.

[0058] Scenario B: Soft time-sensitive event, obtain the remaining cut-off time after the mutation. Recalculate the time repulsion field strength in Example 2: Update the total trade potential value, E_{total}'=w_1\cdotU_{att}+w_2\cdotU_{rep}'$. If The number of items increases dramatically, so their order in the set is adjusted. For vacant computing power slots, the system iterates through the remaining goods awaiting shipment in the inventory and uses a max-heap sort algorithm to pop the current... The highest-ranking candidate goods are added to the collection.

[0059] The system queries the task status register. If the affected goods are already in the pre-shift task queue generated in Example 4 and have been assigned to the equipment in Example 5... The system locks the device and obtains its current spatiotemporal path sequence. and the current moment The spacetime node where it is located .

[0060] Soft braking and temporary dwell: Issue a task interruption signal. The equipment is in... Search for the nearest safe docking node (non-intersection, non-main road) and mark it as the temporary starting point. .

[0061] New task generation: For newly added candidate goods, invoke the logic of Examples 3 and 4 to quickly calculate their target buffer coordinates in the dynamic reorganization area. .

[0062] Path replanning: Starting from, To complete the process, the improved time window A algorithm based on potential energy gradient from Example 5 is invoked. Key parameter adjustments: Since candidate goods typically possess extremely high comprehensive trade potential energy values, according to the formula in Example 5: At this time, the avoidance coefficient If the value is extremely small, the algorithm will ignore the congestion potential energy value. They devised a shortcut to bypass the obstacles.

[0063] Instruction overwrite: Generate a new spatiotemporal path sequence .use Overwrite the original path The remaining segments that have not yet been executed. Simultaneously, resources occupied by the original task in the global spatiotemporal grid are released, and the congestion potential energy values ​​of the relevant areas are updated. The revised instructions are encapsulated into new device coordination control instructions and then issued.

[0064] Example 7: Once the automated handling equipment completes a command execution (i.e., the goods are placed inside the container), the system initiates a calibration process. This involves calling upon visual sensing devices (such as TOF cameras or structured light cameras) deployed at the outbound platform to collect depth point cloud data of the current container's interior.

[0065] The system preprocesses the data (denoising and background removal) and extracts the physical point cloud clusters representing the most recently placed goods. .in For the number of point clouds, For the first The spatial coordinates of the points.

[0066] To obtain the precise attitude of the cargo in physical space, the system uses Principal Component Analysis (PCA) to calculate the geometric centroid of the physical point cloud cluster. : Construct the covariance matrix : .right Perform eigenvalue decomposition to obtain eigenvalues. and its corresponding unit eigenvector The eigenvector corresponding to the largest eigenvalue. This refers to the principal axis direction of the cargo (usually corresponding to the length direction of the cargo). Combined with the geometric centroid... The system generates the actual placement posture of the cargo in physical space: .in Using the centroid coordinates, For based on Calculated rotation angle.

[0067] The system backtracks the virtual packing simulation module (Example 3) to extract the theoretical placement of the cargo from the globally optimal loading scheme: Calculate the difference between the two and construct the loading deviation vector. : The specific component calculations are as follows: , , , .

[0068] The system calculates the loading deviation vector. The modulus length (or weighted modulus length) is determined and compared with the preset tolerance threshold.

[0069] If the deviation exceeds the threshold, it indicates a significant error in the physical loading (such as cargo slippage, expansion, or robotic arm positioning error), and the virtual environment must be updated to prevent subsequent collisions.

[0070] The system performs bounding box inflation on the virtual model representing the current cargo within the occupied space set. It reads the physical attribute message generated in Example 1 to obtain the original dimensions of the cargo. The corrected physical dimensions of the expansion are calculated using the following formula. : , , .in: This is the error sensitivity coefficient (e.g., 1.2), used to moderately amplify the impact of deviations and ensure safety redundancy. A safety buffer gap (e.g., 5mm) is provided to allow for minor space in case of instability during robotic arm grasping or stacking. The system utilizes the center coordinates of the actual landing posture. and expansion physical dimensions Reconstruct the physical occupancy voxel of the cargo and use it to update the set of occupied space in the virtual environment.

[0071] Based on the updated set of occupied space, the system performs a chain check: recalculates the set of available corner points within the virtual container (removing corner points occupied by the expanded volume). It extracts all unloaded successor cargo nodes from the loading topology dependency graph generated in Example 3. For each successor cargo, based on its theoretical placement posture in the globally optimal loading scheme, a secondary collision detection is performed: determining whether its theoretical volume overlaps with the updated physical occupied voxel.

[0072] Branch decision: If no conflict: continue with the original plan. If a spatial interference conflict is detected: determine that the original plan is invalid. Keep the currently loaded cargo (and its expanded volume) stationary. Roll back the unloaded cargo to the high-potential-energy cargo set. Trigger a local replanning instruction: re-invoke the DBL decoding and adaptive multi-objective co-evolution process in the virtual packing simulation module (Example 3). Generate a new remaining loading plan and corresponding equipment cooperative control instructions, thereby eliminating the risk before a physical collision occurs.

[0073] Example 8: Step S1: Construction of multidimensional heterogeneous data sequences The system initiates a data acquisition process, connecting in parallel to the customs data platform, port EDI system, and warehouse management system via heterogeneous communication interfaces. Based on the association between cargo batch numbers and bills of lading numbers, it captures customs release status messages (including discrete status codes), estimated cut-off time messages (including timestamps), and physical attribute messages of inventory goods in real time, constructing a multi-source raw data frame. This data frame undergoes spatiotemporal alignment processing: all time data is mapped to the system's standard timeline, missing data is filled using interpolation algorithms, and the relative remaining cut-off time is calculated. Feature encoding is performed on the aligned data: the text-based customs clearance status is converted into discrete numerical features. ,Will The data is then normalized using Min-Max methods based on physical properties. This ultimately generates a multidimensional heterogeneous trade data sequence encapsulated along the time dimension. .

[0074] Step S2: Trade potential assessment and goods screening The system invokes the trade potential assessment module. First, based on the commodity category subspace obtained through K-Means clustering, it obtains the initial state transition probability matrix. Second, it calculates the state dwell time of goods under the current regulatory state. The state inertia coefficient is calculated by calling the Logistic decay function. Generate a dynamic resident weighted matrix describing a nonhomogeneous process. Again, utilizing Perform multi-step discrete-time Markov chain iterative computation and output the predicted release probability value. Next, a time-varying potential energy field model is constructed to calculate the target gravitational field. Repulsive field of time Finally, the weighted average trade potential value is calculated. And high-potential-energy cargo sets are selected by max-heap sort.

[0075] Step S3: Virtual bin packing and topology generation The system invokes the virtual bin packing simulation module. An initial loading population is constructed using integer permutation encoding. During decoding, candidate placement coordinates are calculated based on the DBL heuristic rule, and an embedded physical stability check is performed simultaneously: the effective contact area ratio is calculated. ,like If the load falls below the physical stability threshold, a penalty function mechanism is used to eliminate the solution. Adaptive multi-objective co-evolution is performed, dynamically adjusting crossover and mutation probabilities based on population diversity indicators to output the globally optimal loading scheme (including 3D coordinates and orientation). Based on this scheme, geometric projection detection is performed along the outbound direction. If two goods satisfy projection overlap and depth order, a unidirectional dependency edge is established, ultimately constructing a loading topology dependency graph representing the outbound order constraints.

[0076] Step S4: Reverse Scheduling and Task Generation The system calls the reverse timing scheduling module. A topology sorting algorithm is executed on the loading topology dependency graph to generate an ideal outbound sequence. .

[0077] Calculate the Manhattan distance from the physical storage coordinates of the goods in the sequence to the outbound platform, and then calculate the spatial entropy value based on the weights. .

[0078] like If the threshold is exceeded, the goods are mapped to the target buffer coordinates of the dynamic reorganization area based on the reverse timing logic, and a pre-shift task queue is generated.

[0079] Step S5: Collaborative Path Planning and Command Issuance The system invokes the collaborative path control module. A bipartite graph weighted matching model is constructed, incorporating a comprehensive trade potential energy value to adjust the response cost of the equipment-task relationship. The KM algorithm is used to complete task assignment. A global spatiotemporal grid map is initialized and a dynamic congestion potential energy field is overlaid. For each matching pair, an improved time window A algorithm based on potential energy gradients is used to plan the path.

[0080] In this process, the avoidance coefficient is used Adjusting the congestion potential value Sensitivity: This causes high-potential-energy goods to tend to plan shortcuts, while low-potential-energy goods tend to take detours. The generated collision-free spatiotemporal path sequence is encapsulated into equipment collaborative control commands for issuance.

[0081] Step S6: Virtual and Real Feedback Calibration and Closed-Loop Control It contains two parallel closed-loop sub-processes: Physical Deviation Calibration: After the equipment executes the command, it acquires depth point cloud data and extracts physical point cloud clusters. Principal Component Analysis (PCA) is used to calculate the actual landing posture, and this is compared with the theoretical landing posture to generate a loading deviation vector. If the deviation exceeds the limit, a bounding box expansion operation is performed to update the occupied space set in the virtual environment, and secondary collision detection is performed on subsequent cargo. If spatial interference conflicts are detected, local replanning is triggered.

[0082] Trade data anomaly correction: A sliding time window listener is used to perform dual-channel differential analysis on multidimensional heterogeneous trade data sequences. When a hard blocking event (abrupt change in customs status) or a soft timeliness event (significantly earlier shipping schedule) triggers the circuit breaker threshold, the high-potential cargo set is dynamically updated (e.g., detaining cargo is removed and reserve cargo is added). At the same time, for affected in-transit tasks, a task interruption signal is sent, and the preemptive correction equipment collaborative control command is preempted based on the new potential distribution.

[0083] Example 9: Application of Intelligent Scheduling System in Trade Warehousing Data Acquisition and Preprocessing: The system establishes connections with external data sources through pre-defined heterogeneous communication interfaces, capturing multi-source data in real time: Customs data: Obtains cargo declaration numbers and regulatory status codes (such as "released" and "under inspection") through the customs data platform interface. Shipping schedule data: Obtains the estimated arrival time (ETA), estimated departure time (ETD), and cut-off time of vessels through the port EDI system interface. Warehouse data: Obtains the physical attributes of inventory goods (such as length, width, height, weight, and load-bearing capacity) through the warehouse management system (WMS) interface.

[0084] The system associates data from different sources based on the unique relationship between cargo batch number and bill of lading number, and merges them into multi-source raw data frames. To ensure data consistency, data cleaning and spatiotemporal alignment are performed, and interpolation algorithms are used to fill in missing timestamp data.

[0085] Trade potential assessment and goods screening: Multidimensional Heterogeneous Data Generation: Through feature encoding and standardization, standardized feature vectors are generated from different types of data (such as customs status, shipping schedules, and cargo physical attributes), and then serialized according to the time dimension. Potential Energy Calculation: Based on the predicted release probability and remaining port cut-off time, the system constructs a time-varying potential energy field model. By integrating the target gravitational field and the time repulsive field, the comprehensive trade potential energy of the cargo is calculated. High-Potential-Energy Cargo Selection: Using a max-heap sort algorithm, cargoes with the highest potential energy values ​​are selected to form a high-potential-energy cargo set for subsequent steps.

[0086] Virtual binning and topology generation: Loading scheme optimization: Based on the physical properties of high-potential-energy goods, the system constructs an initial loading population and performs DBL (Deep-Bottom-Left) decoding to simulate cargo loading. Collision detection and physical verification: Based on the candidate placement coordinates of each cargo, collision detection and embedded physical stability verification are performed to ensure the feasibility of the loading scheme. Generation of loading topology dependency graph: Based on the virtual packing scheme, the dependencies between cargoes are calculated, and a loading topology dependency graph is generated to ensure the logical consistency of the outbound order.

[0087] Reverse scheduling and task generation: Topology sorting: The loading topology dependency graph is sorted to generate an ideal outbound order. Task queue generation: Based on the storage location and outbound order of each item, the spatial entropy value is calculated and mapped to the dynamic reorganization area to generate a pre-shift task queue, ensuring that the handling tasks are carried out in an orderly manner.

[0088] Collaborative path planning and command issuance: Task and Equipment Matching: By constructing a bipartite graph weighted matching model, the system assigns tasks based on the response cost of the equipment (such as Manhattan distance, power consumption, etc.) and the comprehensive potential energy value of the task, using the KM algorithm. Path Planning: Path planning is performed based on an improved A* algorithm, considering the comprehensive trade potential energy value of the equipment during the planning process, avoiding congested areas, and ensuring efficient and collision-free path planning. Command Issuance: The planned path is converted into equipment collaborative control commands and issued to the automated equipment for execution via a wireless network.

[0089] Virtual and real feedback calibration and closed-loop control: Deep point cloud data acquisition: After the equipment performs its task, the deployed visual sensing devices acquire deep point cloud data inside the container, providing real-time feedback on the actual placement posture of the cargo. Loading deviation calibration: The actual placement posture of the cargo is calculated using Principal Component Analysis (PCA) and compared with the theoretical placement posture. If the deviation exceeds a set threshold, a bounding box expansion operation is performed to update the occupied space set, preventing subsequent collisions. Anomaly correction: Multidimensional heterogeneous trade data is monitored through a sliding time window listener. When a sudden change in state or a soft time-sensitive event is detected, the system dynamically adjusts the high-potential-energy cargo set and simultaneously corrects the equipment's collaborative control commands.

[0090] This embodiment demonstrates the application of an intelligent scheduling system in trade warehousing. Through steps such as multi-source data collection, trade potential assessment, virtual packing, reverse scheduling, route planning, and virtual-real feedback calibration, the system achieves efficient cargo scheduling and route planning, ensuring efficient and accurate operation during the warehousing process.

[0091] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An intelligent scheduling system for trade warehousing, characterized in that: Includes the following modules: The multi-source data aggregation module collects customs supervision status data, shipping schedule data, and cargo physical attribute data, and performs data cleaning and spatiotemporal alignment to generate a multidimensional heterogeneous trade data sequence. The trade potential energy assessment module receives the multidimensional heterogeneous trade data sequence, predicts the customs clearance state transition probability using a discrete-time Markov chain, calculates the urgency of cargo release using a time-varying potential energy field model, and filters and outputs a set of high-potential-energy cargo. The virtual packing simulation module receives the high-potential-energy cargo set, performs iterative packing simulations in a virtual three-dimensional space based on an improved genetic strategy and DBL heuristic rules, and generates a sequence containing approximately [data missing - likely related to cargo release order]. The system consists of a loading topology dependency graph; a reverse timing scheduling module, which receives the loading topology dependency graph, calculates the spatial entropy value between the current physical storage location and the ideal outbound timing based on reverse order logic, and generates a pre-shifting task queue for high-potential-energy goods; a collaborative path control module, which receives the pre-shifting task queue, assigns tasks using the KM bipartite graph matching algorithm, plans the handling path based on the A* search algorithm with a time window, and generates equipment collaborative control instructions; and a dynamic anomaly correction module, which monitors the state mutation of the multidimensional heterogeneous trade data sequence in real time, dynamically updates the high-potential-energy goods set when the circuit breaker threshold is triggered, and synchronously corrects the equipment collaborative control instructions.

2. The intelligent scheduling system for trade warehousing according to claim 1, characterized in that: The generation of multidimensional heterogeneous trade data sequences includes: The system connects to the customs data platform, port EDI system and warehouse management system through preset heterogeneous communication interfaces to capture customs release status messages, estimated cut-off time messages for ships and physical attribute messages of inventory goods in real time. Based on the pre-defined correspondence between cargo batch number and bill of lading number, customs release status messages, estimated cut-off time messages, and physical attribute messages are associated and matched to construct multi-source raw data frames. Data cleaning and spatiotemporal alignment are performed on the original data frames from multiple sources. Interpolation algorithms are used to fill in the missing timestamp data, and all time data are uniformly mapped to the system's standard time axis to calculate the relative remaining cut-off time. The aligned data is subjected to feature encoding and standardization. The customs clearance status of text type is converted into discrete numerical features. The relative remaining port interception time and physical attribute data are mapped to a preset numerical range to generate a normalized feature vector. The normalized feature vectors are serialized and encapsulated according to the time dimension to generate a multidimensional heterogeneous trade data sequence for input to the trade potential assessment module.

3. The intelligent scheduling system for trade warehousing according to claim 1, characterized in that: The generation of high-potential-energy cargo sets includes: Extract commodity coding features from multidimensional heterogeneous trade data sequences, and use the K-Means clustering algorithm to divide historical trade data into... Each commodity category subspace; for the current commodity category subspace, the customs supervision status within the historical time window is statistically analyzed from status... transition to state Based on the frequency of occurrence, construct the initial state transition probability matrix for this category. ; Real-time calculation of the current status dwell time of goods under the current regulatory status. The state inertia coefficient is calculated by calling the preset Logistic decay function. Using the state inertia coefficients to determine the initial state transition probability matrix Nonlinear correction is performed: the diagonal elements are strengthened to maintain the probability of the current state, while the off-diagonal elements are proportionally weakened to reduce the probability of state transition. Simultaneously, normalization is performed to generate a dynamically resident weighted matrix describing the nonhomogeneous process. ; Construct the current state distribution vector of the goods The current state is 1, and the rest are 0. Execution is performed using a dynamically resident weighted matrix. Step-by-step discrete-time Markov chain iterative operation Extract the component value corresponding to the release status from the result vector, and output the predicted release probability value of the goods within a preset time window in the future. ; Establish a gravitational field containing the target Repulsive field of time The time-varying potential energy field model; the target gravitational field is constructed based on the predicted release probability value, and the formula is: This indicates the passability of goods to be allowed to leave the warehouse. The target gravitational field strength coefficient is typically set to a range of values. The time repulsion field is based on the remaining time window from the current moment to the cutoff time. Constructed using an improved Coulomb repulsive potential field function. This indicates the exponential increase in the urgency of shipping as the shipping date approaches. The time repulsion field strength coefficient is typically set to a range of values. ; The comprehensive trade potential energy value of the goods is calculated by weighting and superimposing the target gravitational field and the time repulsion field. ; Traverse all goods in inventory awaiting shipment and calculate their... The algorithm then uses a max-heap sorting algorithm to select the top-K items with the highest potential energy values, or those with potential energy values ​​exceeding a preset activation threshold. The goods generate a high-potential-energy cargo set.

4. The intelligent scheduling system for trade warehousing according to claim 1, characterized in that: The generation of the loading topology dependency graph includes: Obtain the physical attribute data of each cargo in the high-potential-energy cargo set, including length, width, height and load-bearing level, and construct a gene sequence using an integer-based encoding method to generate an initial loading population; each gene bit represents a unique identifier ID of a cargo, and the order of the gene sequence represents the order in which virtual packing is attempted to be placed. For each individual in the initial loading population, a decoding operation is performed, invoking the DBL heuristic rule to extract cargo one by one according to the gene sequence order. The DBL coordinate calculation process is then executed to calculate the candidate placement coordinates for each cargo in the three-dimensional coordinate system of the virtual container. The candidate coordinate set is initialized, initially containing only the origin coordinates of the virtual container. Based on the sorting rule of Z-axis depth first > Y-axis height second > X-axis width last, the points in the candidate coordinate set are sorted. The sorted candidate coordinate points are then traversed sequentially and used as anchor points for the current cargo. Collision detection is performed: it is determined whether the space volume occupied by the current cargo at the anchor point overlaps with the already occupied space set, and whether it exceeds the physical boundary of the virtual container. If the collision detection passes, an embedded physical stability check is further performed: the effective contact area ratio between the bottom surface of the current cargo and the top surface of the cargo already placed below is calculated. If the effective contact area ratio is higher than a preset physical stability threshold, the anchor point is confirmed as the final placement coordinate of the cargo. The volume of the cargo is added to the occupied space set, and new candidate points are generated based on the vertex coordinates of the cargo's upper right, upper left, and lower front corners, updating the candidate coordinate set; otherwise, the penalty function mechanism is triggered, applying a non-linear penalty weight to the fitness value of the individual and eliminating physically infeasible solutions. A dual-objective fitness evaluation system is constructed, encompassing both maximizing space utilization and minimizing combinatorial centroid; the fitness variance of all individuals in the current population is calculated to characterize population diversity; and the crossover probability of the genetic algorithm is dynamically adjusted based on this index. ) and the probability of variation ( When the population diversity index falls below a preset warning threshold, the mutation probability is automatically increased to escape local optima. After a preset number of iterations, the globally optimal loading scheme on the Pareto optimal front is output. This scheme includes the final three-dimensional coordinates of each item in the high-potential-energy cargo set within the virtual container. and rotational attitude parameters; Based on the globally optimal loading scheme, a virtual ray source is set along the negative Z-axis direction of the container outbound operation to perform geometric projection detection on each piece of cargo; for any two pieces of cargo... and Determine whether the physical obstruction condition is met: that is, the goods On the projection plane of the outbound direction, with the goods There are overlapping regions ( ), and goods The depth coordinates are located on the cargo The outbound route ahead ( If the conditions are met, then the goods are deemed acceptable. For goods Physical occlusion of the foreground, establishing a system based on point to One-way dependent edges; Using goods in the high-potential-energy goods set as nodes and unidirectional dependency edges as connections, construct an adjacency matrix describing the loading dependency logic. , of which elements Indicates goods Blocking goods Traverse the adjacency matrix Each non-zero element is mapped to an edge in the directed graph to construct the initial directed graph structure. A depth-first search is performed on the initial directed graph structure to detect loops. If a closed loop is detected, the current loading scheme is determined to have a logical deadlock, and a globally optimal loading scheme is regenerated. If there is no closed loop, the directed graph structure is instantiated into a loading topology dependency graph, and the in-degree of each node is calculated, where nodes with an in-degree of 0 represent goods that can be physically removed at the current moment.

5. The intelligent scheduling system for trade warehousing according to claim 1, characterized in that: The generation of the pre-shift task queue includes: Receive the loaded topological dependency graph, perform a topological sorting algorithm on it, and transform the graph structure into a linear ideal outbound sequence. ;in, The goods with an in-degree of 0 and the highest priority in the sequence represent the goods that should be loaded first. If there are parallel branches in the graph, the goods are sorted in a secondary order based on their comprehensive trade potential value in the high-potential-energy goods set, with those having higher potential values ​​ranked at the top of the sequence. Obtain the current physical storage coordinates of each item in the ideal outbound sequence. And calculate the Manhattan distance from that coordinate to the warehouse departure platform. Define spatial entropy. This is used to characterize the degree of disorder between the current physical storage state and the ideal outbound timing; the calculation formula is: ,in The weighting coefficient is related to the order position of the goods in the ideal outbound sequence. There is a negative correlation (i.e., the higher the order of a cargo, the greater its distance weight; if it is farther from the platform, the entropy value increases significantly); the calculated spatial entropy value... Compare with the preset recombination activation threshold; if If the threshold is exceeded, the subsequent shift task generation process will be triggered; A dynamic reorganization area is designated within the temporary storage area for outbound shipments in the warehousing system. This area contains several pre-set buffer grids. Mapping is performed based on reverse time-series logic to convert the ideal outbound sequence... Mapped to the dynamic reorganization area, the mapping rule is: the first item in the sequence. Mapped to the buffer grid closest to the outbound platform, the last item in the sequence is sorted. Mapping to the buffer grid furthest from the outbound platform determines the target buffer coordinates for each item within the reassembly zone. ; For each item in the ideal outbound sequence, construct a shift task from the physical storage coordinates to the target buffer coordinates, and generate a task tuple. Arrange all task tuples in reverse order of the ideal outbound sequence (i.e., process goods at the end of the sequence first to avoid blocking the front path) or based on the unlocking order of dependency relationships, and encapsulate them to generate a pre-shifted task queue, which serves as the input basis for the collaborative path control module.

6. The intelligent scheduling system for trade warehousing according to claim 1, characterized in that: The generation of equipment collaborative control commands includes: Parse the pre-shifted task queue, extract task tuples, and construct a task set. ; Obtain the coordinates, power levels, and real-time load status of all automated material handling equipment within the warehouse, and construct a set of available equipment. Construct a weighted matching model for a bipartite graph, where the edge weights are... Defined as the response cost of the device-task; introduce the comprehensive trade potential value as an adjustment factor to nonlinearly reduce the response cost of high potential energy tasks, making them easier for nearby devices to capture; use the KM algorithm to solve for the best match and output the assignment result; Initialization includes spatial dimensions With the time dimension The system uses a global spatiotemporal grid map and introduces the concept of a dynamic congestion potential energy field. For known high-frequency operation areas or intersections with planned routes, congestion potential energy values ​​are superimposed on the corresponding spatiotemporal nodes. This value decays over time and is used to characterize the traffic density of the area at a specific future time. For each pair of matches, a path is planned using the improved time window A* algorithm based on potential energy gradient: During the heuristic search process, the evaluation function is reconstructed as follows: ; For the actual cost of movement, Congestion potential energy was introduced for Manhattan distance estimation. , To mitigate the impact of the collision coefficient, a strategy of trading space for time is employed to adjust the collision coefficient. , The value of this value is negatively correlated with the overall trade potential energy of the goods currently being transported; the higher the potential energy of the goods, the better. The smaller the potential energy, the less sensitive the algorithm is to congestion, and the more inclined it is to plan shortcuts; the lower the potential energy of the goods, the better. The larger the priority, the more likely it is to plan detour routes to reduce overall system congestion, allowing low-priority vehicles to take detours and giving the fast lane to vehicles with greater potential energy. During the search process, if the current node is in the next time step... If the path falls within the occupancy window of other high-priority devices, an in-place waiting strategy or partial replanning will be implemented. After successful path planning, the generated path node sequence will be spatiotemporally locked in the global spatiotemporal grid map, and the congestion potential value of the area covered by the path will be updated synchronously. This can affect the path planning of subsequent low-priority devices. Generate collision-free spatiotemporal path sequences ; It represents the horizontal and vertical physical coordinates in the warehouse grid map, which precisely correspond to the physical location of the shelves or aisles; The system time slice represents a discrete time slice, each time slice corresponding to the standard duration required for the device to perform a single atomic action (such as moving one unit, lifting, or rotating); the spatiotemporal path sequence is converted into machine code containing underlying driving parameters, including motor speed, steering angle, and lifting height, and encapsulated into device collaborative control instructions and issued.

7. The intelligent scheduling system for trade warehousing according to claim 1, characterized in that: The operating mechanism of the dynamic anomaly correction module includes: Allocate a fixed-length circular buffer in system memory, and construct a sliding time window listener that follows the first-in, first-out (FIFO) principle, with its window length set to [value missing]. Each time step; the multidimensional heterogeneous trade data sequence generated by the preceding module is mapped into the listener in real time to form the feature matrix of the current time window. ,in This is the normalized feature vector at the current moment; whenever a new data frame is received... The listener performs a sliding operation: removing the oldest feature vector. and merged into To keep the time span of the window constant; Inside the sliding time window listener, the vectors of two adjacent frames in the feature matrix of the current time window are... and Perform dual-channel difference analysis: Discrete state channel (for customs supervision status): Extract the corresponding discrete numerical features of customs status from the feature vector and calculate the Hamming distance. ;like This indicates a discrete transition in the state; further query the preset risk state mapping table to determine whether the new state belongs to the "inspection", "detention", or "rejection" set; if it does, a hard blocking event is determined to have occurred; continuous time-sensitive channel (for the ship's cut-off time): extract the corresponding relative remaining cut-off time value from the feature vector and calculate the first-order difference value. Define the time jitter threshold. ,like and If the remaining time is significantly shortened beyond the normal elapsed rate, a soft time-sensitive event is determined to have occurred. When any of the above events is detected, the system determines that the circuit breaker threshold is triggered and generates an abnormal interruption signal containing the unique identifier ID of the affected goods and the mutation type label. Based on the unique identifier ID in the abnormal interruption signal, locate the target cargo in the current high potential energy cargo set; execute the potential energy reconstruction logic according to the mutation type label: if it is a hard blocking event, force the predicted release probability value of the cargo in the discrete-time Markov chain to zero, causing its comprehensive trade potential value to drop to negative infinity, thereby removing the cargo from the high potential energy cargo set and forming an empty computing power slot. If it is a soft time-sensitive event, the intensity of the time repulsion field in the time-varying potential energy field model is recalculated using the remaining cut-off time after the mutation, and its comprehensive trade potential energy value is updated; for the generated empty computing power slots, the remaining goods waiting to be shipped in the inventory are traversed, and the maximum heap sorting algorithm is used to select the candidate goods with the highest current potential energy value that are not in the set, and they are filled into the high potential energy goods set to complete the dynamic update of the set. The system's task status register is queried to check if a task has been generated in the pre-shift task queue for the affected goods. If so, it is further determined whether the task has been assigned to a specific device using the KM bipartite graph matching algorithm. If the task is in the "assigned and executing" state, the target device ID executing the task is locked, and the current spatiotemporal path sequence of the device is obtained. and the current spatiotemporal node ; A task interruption signal is sent to the target device, driving it to perform a soft braking operation, causing it to stop at the nearest dockable node and marking that node as the new temporary starting point; the virtual packing simulation module and the reverse timing scheduling module are invoked to generate a new relocation task for the newly added candidate cargo; starting from the temporary starting point and ending at the target buffer coordinates of the new task, the improved time window A* algorithm based on potential energy gradient is invoked for path replanning; a new spatiotemporal path sequence is generated. Use it to cover the original In the remaining segment, the corrected device collaborative control instructions are generated and issued, while the global spatiotemporal grid resources occupied by the original task are released and the congestion potential energy value is updated.

8. The intelligent scheduling system for trade warehousing according to claim 1, characterized in that: It also includes a virtual-real mapping feedback calibration module, the operation mechanism of which includes: After the automated handling equipment completes the execution of a collaborative control command, it triggers the visual sensing devices deployed on the outbound platform to collect depth point cloud data of the current container; background filtering is performed on the depth point cloud data to extract the physical point cloud cluster of the most recently placed goods. ,in For the number of point clouds, For the first The spatial coordinates of each point are calculated; the actual landing posture of the cargo in physical space is calculated using Principal Component Analysis (PCA); the geometric centroid of the physical point cloud cluster is calculated. : Construct the covariance matrix : For the covariance matrix Perform eigenvalue decomposition to obtain the eigenvector corresponding to the largest eigenvalue. The direction of this feature vector is identified as the principal axis direction of the cargo, combined with the geometric centroid. Generate the actual landing posture including center coordinates and rotation angle. ; The globally optimal loading scheme generated in the virtual packing simulation module is invoked to extract the theoretical placement posture of the cargo. ; Calculate the difference between the actual landing posture and the theoretical landing posture, and construct the loading deviation vector. : ; Determine the loading deviation vector If the module length exceeds the preset tolerance threshold, then a bounding box expansion operation is performed on the virtual model representing the current goods in the occupied space set in the virtual 3D space to obtain the original dimensions from the goods' physical attribute data. The corrected physical dimensions of the expansion are calculated using the following formula, which includes the length, width, and height. : , , , This is the error sensitivity coefficient, used to amplify the weight of the impact of deviation on space occupancy; its value range is typically set to [value range missing]. ; The safety buffer gap, measured in millimeters, is used to allow for minor extra space in case of instability during robotic arm grasping or stacking; it is based on the center coordinates of the actual placement posture. With expansion physical dimensions Regenerate the physical occupancy voxel of the cargo and update the occupied space set with it; Extract all unexecuted successor cargo nodes from the loading topology dependency graph; recalculate the set of available corner points within the virtual container based on the updated set of occupied space; perform secondary collision detection on the theoretical placement posture of the successor cargo: detect whether the successor cargo overlaps in volume with the updated physical occupied voxel at its theoretical position; if spatial interference conflict is detected, determine that the original globally optimal loading scheme is invalid; keep the currently loaded cargo stationary, and only for the remaining unloaded high-potential-energy cargo set, re-trigger the DBL decoding and genetic optimization process in the virtual container loading deduction module to generate new remaining loading schemes and corresponding equipment collaborative control instructions to eliminate the risk of physical collisions caused by accumulated errors.

9. An intelligent scheduling method for trade warehousing, characterized in that: This method is applied to any one of the systems described in claims 1-8, and includes the following steps: Step 1: Construction of multidimensional heterogeneous data sequences By collecting customs release status messages, estimated cut-off time messages, and physical attribute messages of inventory goods through heterogeneous communication interfaces, a multi-source raw data frame is constructed. Spatiotemporal alignment and interpolation filling are performed on the multi-source raw data frame, and feature encoding and standardization are carried out to generate a multi-dimensional heterogeneous trade data sequence. Step 2: Trade Potential Assessment and Goods Screening An initial state transition probability matrix is ​​constructed based on historical trade data, and the state dwell time of goods under the current regulatory state is calculated. The dwell decay function is called to generate a dynamic dwell weighting matrix. This matrix is ​​used to perform multi-step discrete-time Markov chain iterative calculations to obtain the predicted release probability value. A time-varying potential energy field model containing the target gravitational field and the time repulsive field is established. The two are weighted and superimposed to calculate the comprehensive trade potential energy value. Based on this, a set of high potential energy goods is selected and output. Step 3: Virtual Packing and Topology Generation An initial loading population is constructed from a high-potential-energy cargo set using an integer permutation-based encoding method. DBL decoding with embedded physical verification is performed: while calculating candidate landing coordinates based on the DBL (Deep-Bottom-Left) heuristic, the effective contact area ratio is calculated to perform embedded physical stability verification, and a penalty function mechanism is used to eliminate physically infeasible solutions. Adaptive multi-objective co-evolution is performed: crossover and mutation probabilities are dynamically adjusted based on population diversity indicators to output the globally optimal loading scheme. Geometric projection detection is performed based on the globally optimal loading scheme to identify physical occlusion precursor relationships, constructing and outputting a loading topology dependency graph. Step 4: Reverse Scheduling and Task Generation A topology sorting algorithm is executed on the loading topology dependency graph to generate an ideal outbound sequence; the physical storage coordinates and spatial entropy value of the goods in the sequence are calculated. When the entropy value exceeds the threshold, the goods are mapped to the target buffer coordinates of the dynamic reorganization area based on the reverse timing logic to generate a pre-shift task queue. Step 5: Collaborative Path Planning and Command Issuance A bipartite graph weighted matching model is constructed, and a comprehensive trade potential energy value is introduced to adjust the response cost. The KM algorithm is used to complete task assignment. A dynamic congestion potential energy field is superimposed on the global spatiotemporal grid map, and a collision-free spatiotemporal path sequence is planned using an improved time window A* algorithm based on potential energy gradient. Among these, the avoidance coefficient is used. The algorithm's sensitivity to congestion is adjusted, and this coefficient is negatively correlated with the overall trade potential of goods; the path sequence is encapsulated into equipment collaborative control commands and issued. Step 6: Virtual and Real Feedback Calibration and Closed-Loop Control After the equipment executes the command, it collects depth point cloud data and extracts physical point cloud clusters; it uses principal component analysis to calculate the actual landing posture of the cargo and compares it with the theoretical landing posture to generate a loading deviation vector; when the deviation exceeds the threshold, it performs bounding box expansion operation to update the occupied space set and performs secondary collision detection on subsequent cargo; if spatial interference conflict is detected, it triggers local replanning; at the same time, it uses a sliding time window listener to perform dual-channel differential analysis on the multidimensional heterogeneous trade data sequence; when a hard blocking event or a soft time-sensitive event is identified that triggers the circuit breaker threshold, it dynamically updates the high-potential-energy cargo set and preempts the corrective equipment cooperative control command.

Images