Intelligent logistics multi-candidate wave sorting system and method

Abstract

The application relates to the field of automatic sorting technology and discloses an intelligent logistics multi-candidate wave sorting system and method, which is provided with a business management layer, a scheduling and monitoring layer, an electric control control layer and a hardware execution layer. The business management layer performs order compliance verification and labeling, generates a candidate wave task in combination with a target function generated by time efficiency, space and equipment load calculation, the scheduling and monitoring layer decomposes the task into an instruction sequence and performs adjustment and routing reconstruction, and the hardware execution layer completes material extraction, transportation, pretreatment and sorting. The system compares the buffer zone inventory and the threshold to adjust the operation rhythm; the spare part link performs comparison and speed synchronous transfer based on a feature vector; when a fault occurs, a node topology matrix is reconstructed to perform path planning; when the load is full, an address overwrite is performed by using a backup mapping table; the system calculates the partial derivative of the loss function based on the execution data to realize self-learning update of the weight parameter. The application realizes full-link precise cooperation and dynamic fault tolerance, and improves operation efficiency and stability.

Description

Technical Field

[0001] This invention relates to the field of automated sorting technology, and in particular to an intelligent logistics multi-candidate wave picking system and method. Background Technology

[0002] With the continuous expansion of e-commerce, the warehousing and logistics sector faces the demand for high-frequency, small-batch, and time-sensitive order processing during peak sales periods. The core technical bottleneck of existing logistics picking systems in dealing with such highly volatile business scenarios lies in their overly simplistic candidate wave planning logic. Existing systems typically only perform simple candidate wave merging based on fixed time windows or fixed order sizes, failing to perform multi-dimensional correlation calculations involving order fulfillment deadlines, the physical distribution of inventory units, the dynamic load of handling equipment, and the real-time remaining capacity of sorting bins. This planning approach leads to data silos between the warehousing picking and last-mile sorting stages, making it difficult to accurately match the operational rhythms between different processes. This easily induces material backlogs and equipment idleness at local operation nodes, limiting the overall throughput efficiency of the system under high-concurrency scenarios.

[0003] Furthermore, existing systems lack sufficient adaptability to anomalies and architectural flexibility. In actual operation, when encountering situations such as temporary order changes, abnormal inventory data, sorting hardware failures, or full physical load of sorting bins, the system often lacks automated logical refactoring and anomaly routing capabilities, still requiring significant manual intervention for task redistribution, thus reducing the system's operational continuity. Simultaneously, the traditional system architecture has a high degree of hardware-software coupling, making it difficult to accommodate the mixed picking needs of multiple physical entities such as bins, pallets, and single packages, and also unable to meet the differentiated configurations of multi-owner operating rules. This lack of robustness and scalability results in poor system stability in complex engineering environments, making it difficult to support the high availability and multi-scenario adaptability requirements of modern logistics centers. Summary of the Invention

[0004] The purpose of this invention is to provide an intelligent logistics multi-candidate wave picking system and method, which solves the problems of rigid candidate wave planning, mismatch of the whole-chain operation rhythm, and weak fault tolerance and adaptability under abnormal conditions in traditional logistics systems when dealing with highly volatile e-commerce orders.

[0005] To achieve the above objectives, the present invention provides the following technical solution: The first aspect of this invention provides an intelligent logistics multi-candidate wave picking system. This system establishes a four-layer communication architecture consisting of a business management layer, a scheduling and monitoring layer, an electronic control layer, and a hardware execution layer, achieving closed-loop control across the entire chain from order logic processing to physical hardware execution. The business management layer uses a dynamic candidate wave planning module to tagged and transformed order execution, combining multi-dimensional constraints such as timeliness, space, and equipment load to generate candidate wave task documents. The scheduling and monitoring layer decomposes these documents into a set of instruction sequences containing physical coordinate information and receives real-time status data from the hardware execution layer, enabling dynamic adjustment and route reconstruction in abnormal situations. The electronic control layer drives the hardware execution layer through an industrial bus to complete material extraction, conveying, preprocessing, and sorting actions.

[0006] The system's innovation lies in resolving the conflict between timeliness and hardware load balancing in large-scale order processing through a multi-level feedback mechanism. The dynamic candidate wave planning module utilizes scalar scores in the timeliness dimension to ensure priority processing of urgent orders, reduces the round-trip physical distance of stacker cranes in the aisle dimension through scalar scores in the spatial dimension, and combines scalar scores in the equipment dimension to prevent material backlog at specific conveyor nodes. Under the constraint of meeting the upper limit of the physical capacity of the sorting slots, the system outputs the optimal candidate wave partitioning scheme, thereby improving the overall throughput efficiency of the warehousing unit.

[0007] The system possesses end-to-end adaptive traffic adjustment capabilities during execution. The scheduling and monitoring layer collects the physical inventory of goods within the conveying buffer unit through a real-time monitoring module, comparing the inventory value with the upper congestion threshold and the lower hunger threshold. When material congestion risk occurs at the sorting front end, the system issues instructions to reduce the operating speed of the upstream warehousing and picking unit and activates a circular buffer track for physical storage; when the supply of items at the sorting front end is insufficient, the system increases the retrieval frequency and releases the stored materials. This closed-loop adjustment mechanism based on inventory awareness can eliminate physical blockages caused by mismatched hardware actions.

[0008] The system also boasts high operational reliability and robustness. In equipment failure scenarios, the system reconstructs the physical node topology graph matrix, modifies the weights of connected edges corresponding to offline physical devices to infinity, and performs shortest path numerical calculations using the remaining connected graph, thereby physically avoiding failure points. In scenarios with full grid capacity, the system automatically overwrites the destination address pointer through a backup mapping database table and calculates the estimated physical time difference based on the running speed of the carrying cart assembly, generating precise action trigger timestamps to ensure the continuity of sorting operations and avoid system-wide downtime due to localized hardware saturation.

[0009] A second aspect of this invention provides an intelligent logistics multi-candidate wave picking method, which is applied to the aforementioned intelligent logistics multi-candidate wave picking system and includes the following core processing flow: First, the system receives outbound order data packets and performs compliance checks. Through multiple verifications of ownership, inventory, destination, and integrity, orders that meet the criteria are converted into numerical or enumerated tag variables and stored in the pending order pool. This step standardizes heterogeneous order data, providing a unified data foundation for subsequent algorithmic processing.

[0010] Secondly, the system reads the order pool to be processed and performs multi-dimensional combination optimization calculations. By calculating scalar scores in terms of timeliness, space, and equipment, it iterative evaluation is performed on candidate wave combinations. Under the premise of ensuring that the physical equipment load and grid capacity do not exceed the limits, candidate wave task document files guiding physical operations are generated and sent to the scheduling and monitoring layer for task decomposition.

[0011] Next, the hardware execution layer performs material extraction and conveying according to the instruction sequence. In the conveying buffer stage, the system senses the physical quantity of goods in the buffer area in real time and dynamically adjusts the upstream extraction speed. In the pre-processing stage, the system collects the actual weight, identification string, and dimensions of the goods to form a multi-dimensional physical feature vector, which is then compared against the expected parameter matrix. Only goods that pass the verification enter the synchronous feeding process. The inlet platform operates according to a preset acceleration curve, ensuring that the belt speed and the sorting host speed are synchronized within a preset range, achieving smooth transfer even at high speeds.

[0012] Finally, the system performs full-process data statistics and strategy self-learning optimization. The system collects the actual total execution time, number of overdue orders, and equipment idle rate for each candidate wave, generating execution feature vectors for each candidate wave. When the accumulated data reaches a preset threshold representing the minimum sample size required to trigger parameter updates, the system initiates a parameter iteration program. By constructing a true execution loss function that includes overdue penalties and idle losses, gradient descent logic is used to calculate the partial derivative of the loss function with respect to the weight parameter set, thereby outputting the updated normalized weight values.

[0013] The innovation of this method lies in its implementation of a parameter feedback closed loop based on actual execution deviations. The system can automatically fine-tune the influence weights of each dimension in the dynamic candidate wavelet planning based on historical execution efficiency, gradually bringing the candidate wavelet partitioning strategy closer to the optimal state under the current physical environment. This self-learning mechanism enhances the system's adaptability to business fluctuations and complex engineering environments.

[0014] In summary, the present invention has at least one of the following beneficial technical effects: 1. This invention utilizes a dynamic candidate wave planning module in the business management layer. By combining order fulfillment deadline timestamps, physical location coordinate distance costs, and theoretical equipment load estimates, it performs combined optimization calculations across timeliness, space, and equipment dimensions. Under the premise of meeting the upper limit constraints of equipment load and sorting grid capacity, it avoids downstream congestion and process disconnect caused by traditional fixed candidate wave division, and achieves precise matching of picking and sorting rhythms.

[0015] 2. This invention establishes a four-layer communication architecture, and the scheduling and monitoring layer collects the physical inventory of goods inside the delivery buffer unit. By comparing the inventory value with the upper limit congestion threshold and the lower limit hunger threshold, the system dynamically adjusts the extraction frequency of the upstream warehouse picking unit and coordinates with the ring buffer track for physical dwelling. This closed-loop adjustment mechanism based on inventory perception effectively smooths out business traffic fluctuations and eliminates stacking blockages caused by speed differences between hardware units.

[0016] 3. This invention constructs a reconstruction mechanism covering scenarios such as order changes, equipment failures, and full grid capacity. When equipment is offline, the system automatically plans the optimal backup path by modifying the weights of connected edges in the topology graph matrix to infinity and utilizing the remaining connected graph. When grid capacity is full, the system automatically overwrites pointers and calculates timestamps through the backup mapping database table to perform loading, effectively preventing system-wide shutdowns caused by local hardware saturation or failures, and ensuring stable and uninterrupted logistics operations. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the system architecture of the present invention; Figure 2 This is a schematic diagram of the method flow of the present invention; Figure 3 This is a schematic diagram comparing system throughput within a 24-hour work cycle in a specific application embodiment of the present invention; Figure 4 This is a schematic diagram of the parameter self-learning iteration and overdue order rate convergence curve in a specific application embodiment of the present invention. Detailed Implementation

[0018] See attached document Figure 1 This invention provides an intelligent logistics multi-candidate wave picking system, which may include: a business management layer, a scheduling and monitoring layer, an electronic control layer, and a hardware execution layer. These layers are interconnected via a communication network for sequential data interaction, instruction issuance, and status feedback.

[0019] The business management layer is deployed within the warehouse management system to receive external outbound order data and generate candidate wave task orders. The business management layer includes an order management module, a dynamic candidate wave planning module, an inventory management module, a picking list management module, a shipper rule configuration module, and a data statistics module.

[0020] The order management module receives outbound orders through a standardized data interface and extracts and stores various fields from these orders. The dynamic candidate wave planning module communicates with the order management module and, based on set timeliness, equipment, space, and grid constraints, merges and splits outbound orders, outputting candidate wave task orders. The inventory management module records and updates the location coordinates and quantity information of target SKUs within the warehouse. The picking list management module receives candidate wave task orders and converts them into corresponding picking list format data. The owner rule configuration module receives and stores operational parameters for different owners, which are then used by the dynamic candidate wave planning module. The data statistics module collects and stores the execution status records of orders and candidate wave tasks.

[0021] The scheduling and monitoring layer is deployed within the equipment monitoring and scheduling system, communicating with the business management layer. It receives candidate wave task orders, provides progress feedback to the business management layer, and issues equipment scheduling instructions to the electrical control layer. The scheduling and monitoring layer includes an equipment collaborative scheduling module, a path planning module, a real-time monitoring module, a fault diagnosis module, an anomaly handling module, and a data interaction module.

[0022] The equipment collaborative scheduling module breaks down candidate wave task orders into picking sub-tasks, conveying sub-tasks, supplying sub-tasks, and sorting sub-tasks, and generates action instruction sequences corresponding to specific physical equipment. The path planning module calculates the running route parameters of the hardware execution layer equipment between each physical node. The real-time monitoring module receives and displays the operating status data and location coordinates of each device in the hardware execution layer. The fault diagnosis module compares the operating status data with preset working thresholds and generates corresponding fault codes when values ​​exceed limits. The exception handling module sends a request to the equipment collaborative scheduling module to update the scheduling task upon receiving a fault code or a specific exception trigger signal. The data interaction module performs communication protocol parsing and data formatting between the scheduling monitoring layer, the business management layer, and the electrical control layer.

[0023] The electronic control layer communicates with the scheduling and monitoring layer, and also establishes electrical and communication connections with the various physical devices in the hardware execution layer. The electronic control layer receives action command sequences, drives the hardware execution layer to perform specific mechanical actions, and collects sensor signals from the hardware execution layer.

[0024] The electrical control layer includes a main controller, which is a Siemens S7-1500 series programmable logic controller. The main controller is connected to the control nodes of the hardware execution layer via a PROFINET industrial Ethernet architecture. The electrical control layer is configured with three operating modes: manual control mode, local automatic control mode, and online automatic control mode, supporting mode switching via a switch or host computer commands. A safety interlock logic loop is configured within the electrical control layer to cut off the drive signal of the corresponding device when an interlock condition is detected.

[0025] The hardware execution layer, controlled by the electrical control layer, is a collection of equipment that completes the physical transfer of materials. The hardware execution layer includes a warehousing and picking unit, a conveying and buffering unit, a pre-processing unit for parts, a sorting execution unit, a sorting compartment processing unit, and a safety protection unit.

[0026] The warehouse picking unit includes aisle stacker cranes, multi-level shuttles, and elevators. The aisle stacker cranes and multi-level shuttles are located within the automated warehouse aisles and are used to retrieve the corresponding boxes or pallets based on the location coordinates of the target SKU. The elevators are located at the ends of the aisles and are used to receive the boxes or pallets delivered by the aisle stacker cranes or multi-level shuttles and vertically transport them to the picking station at a designated elevation.

[0027] The conveyor buffer unit connects the warehouse picking unit and the supply pre-processing unit. The conveyor buffer unit includes roller conveyors, belt conveyors, lifting transfer machines, and circular shuttles. Roller conveyors and belt conveyors are used to carry and move goods horizontally. Lifting transfer machines are used to change the direction of goods at the branching and intersection points of the conveyor lines. Circular shuttles are used to transfer goods between multiple conveyor nodes. The control nodes of the conveyor buffer unit are controlled by an electrical control layer and are used to perform the conveying, merging, and splitting of goods, as well as the buffering operations within the area.

[0028] The pre-processing unit is located at the output end of the conveyor buffer unit. The pre-processing unit includes a dynamic weighing device, a barcode reader, a shape detection device, and an infeed table. The dynamic weighing device acquires the mass parameters of the passing goods. The barcode reader scans and parses the barcode or QR code data on the surface of the goods. The shape detection device measures the three-dimensional dimensions of the goods. The infeed table transfers the goods to the carrying surface of the sorting execution unit according to a preset acceleration curve.

[0029] The sorting execution unit is located downstream of the pre-processing unit. The sorting execution unit includes a sorting host, a rack track system, a drive system, a carrier trolley assembly, and an unloading mechanism. The sorting host is either a swing wheel sorter or a cross-belt sorting host. The rack track system is a closed-loop or linear physical track. The drive system provides traction, driving the carrier trolley assembly to move continuously on the rack track system. The unloading mechanism is installed at a node of the carrier trolley assembly or the sorting host, controlled by trigger signals from the electronic control layer, and generates a lateral force at the target physical location to discharge the goods to the sorting unit.

[0030] The sorting unit is located on the output side of the sorting execution unit. The sorting unit includes physical sorting slots, a full-slot detection device, an audible and visual alarm device, and receiving slots. The number of sorting slots is set to no less than 200. The full-slot detection device is installed in each sorting slot and is used to acquire data on the stacking status of goods within the slot using photoelectric or gravity sensors. The audible and visual alarm device is communicatively connected to the full-slot detection device and is triggered when the full-slot detection device outputs a high-level signal. The receiving slot is used to receive goods for which barcode information parsing failed or whose destination address is abnormal. The sorting unit has a locked and unlocked state, and its state switching is determined by control commands from the scheduling and monitoring layer.

[0031] Safety protection units are distributed within the physical areas of the aforementioned units. Each safety protection unit includes an emergency stop device, an audible and visual alarm device, a safety door interlock device, an obstacle detection device, and an overspeed detection device. The emergency stop device and the safety door interlock device are connected to the hard-wired safety circuit of the electronic control layer. The obstacle detection device and the overspeed detection device are used to send real-time monitoring signals to the electronic control layer. When any sensor in the safety protection unit is triggered, the system interrupts operation in the corresponding area.

[0032] See attached document Figure 2 The present invention provides a smart logistics multi-candidate wave picking method, which may include order receiving and standardized preprocessing steps executed within the business management layer.

[0033] The business management layer receives outbound order data packets from upstream enterprise resource planning systems or e-commerce platforms via a standardized application programming interface (API). The data packets are formatted using JSON or XML. The order management module within the business management layer extracts fields from the outbound order data packets to generate an initial order set. For any order in the initial order set The business management team parses it into a feature vector containing multiple dimensions: ; In this feature vector, Indicates the owner's identification code; This represents a matrix containing the types of the minimum required inventory units and their corresponding quantities. Indicates the physical fulfillment deadline timestamp for the order; Indicates the destination physical compartment code or delivery area code; This represents a collection of basic order information fields, including the recipient's address and contact information. This indicates transpose.

[0034] Business management for the initial order set Each order The compliance verification process is executed. This process comprises four independent subroutines: title verification, total inventory verification, destination compliance verification, and information integrity verification. The system defines a comprehensive verification function. The calculation logic is as follows: ; In this formula, For the ownership verification function, when the identifier code If the cargo owner relationship exists in the cargo owner relationship database stored in the business management layer, the function outputs a value of 1; otherwise, it outputs a value of 0.

[0035] This is a function to verify the total inventory. A matrix is ​​defined. The included first The minimum required quantity of inventory unit is The system database records the current total physical available inventory of this smallest inventory unit as follows: Total available inventory The calculation method is to subtract the locked inventory already allocated to other candidate waves from the total physical inventory. When for All of them All meet hour, The output value is 1; if any Make , The output value is 0.

[0036] For destination compliance verification functions, when encoding If the function exists in the system's physical grid mapping routing table, it outputs a value of 1; otherwise, it outputs a value of 0.

[0037] For information integrity verification functions, when the set If the length of the required field is greater than 0 and the data type conforms to the database column attribute definition, the function will output a value of 1; otherwise, it will output a value of 0.

[0038] When the comprehensive verification function At that time, business management determined the order Verification passed, and the order was processed. Write to the pending order pool set In the middle. When When the business management determines that the order verification has failed, it will transfer the order to the abnormal order database table and send a response message containing the verification failure field code to the upstream system through the application programming interface.

[0039] For writing to the pending order pool set For each order that passes validation, the business management layer performs a tagging process. The system extracts the order's field characteristics and converts them into numeric or enumerated tag variables.

[0040] The system performs a performance timeliness tag conversion operation: obtains the standard timestamp of the current server system. Calculate the remaining fulfillment time difference for the order. The system internally stores preset time intervals, when... When the time is less than or equal to 120 minutes, the system maps it to a time-sensitive tag value of 1; when When the time is greater than 120 minutes and less than or equal to 240 minutes, the time-sensitive label value is 2; the corresponding discrete integer label values ​​are generated sequentially according to this logic.

[0041] System Extraction Convert to owner category index value, extract Convert to destination physical region index value, read matrix The quantity field determines the total number of items in the order. If the total number of items is greater than 1, a boolean label for "multiple items" is added to the order; if the total number of items is equal to 1, a boolean label for "single item" is added. After the above labels are generated, the business management layer writes the generated label vector into the pending order pool. The corresponding order data row in the table is used as the input feature parameters read by the subsequent dynamic candidate wavelet planning module.

[0042] ; The business management layer includes a dynamic candidate wave planning module, which reads order data and tag features from the pending order pool. This module divides the pending order pool into multiple disjoint candidate wave sets by executing combinatorial optimization calculation instructions. The generated candidate wave sets are defined as being applicable to any given candidate wave. The dynamic candidate wavelet planning module calculates its comprehensive evaluation objective function. The objective function is used to calculate the weighted sum of the numerical values ​​of each order feature within a candidate wavelet across multiple preset dimensions. ; In this formula, The score is a scalar score for the timeliness dimension; Scoring is given as a scalar value for the spatial dimension; Balance scalar scores for the device dimension; These are the three-dimensional weight coefficient values ​​stored in the cargo owner rule configuration module, which are read by the business management layer and set internally by the system. The business management layer uses iterative adjustments to the partitioning scheme by changing the included order combinations to find ways to optimize the partitioning. Candidate wavelet data combinations that reach the preset convergence threshold.

[0043] Scalar score for execution timeliness by business management The calculation. The system extracts candidate waves. Fulfillment deadline stamps for all orders within the system and compared with the system's current timestamp Perform subtraction.

[0044] ; In this formula, Used to filter out the order with the smallest remaining time difference from the current time in the candidate wave. This is a time decay normal number stored in the system. This is an exponentially increasing term with a decaying effect.

[0045] Business management execution space dimension scalar score The calculation. The system database stores the physical three-dimensional coordinate data of each storage location in the automated warehouse. For candidate waves... Any two target minimum inventory units contained within and The system extracts the corresponding cargo location coordinates. and And calculate the physical acquisition distance cost between the two. .

[0046] ; In this formula, This represents the sum of physical distance costs between all allocated minimum inventory units within a candidate wave. This sum decreases as the difference in physical coordinates of the target cargo distribution within the same candidate wave decreases. The calculated output value increases.

[0047] During the generation of candidate batches, the business management layer performs equipment-level constraint comparisons. The system reads the preset rated processing limit parameters of the warehouse picking unit, conveyor buffer unit, and sorting execution unit, and marks them accordingly. The system determines the candidate wavelet based on the wavelet count. The total quantity of goods contained herein is used to calculate the theoretical load estimate for this candidate wave allocated to the above three equipment units, and these estimates are respectively labeled as follows: The system substitutes the above parameters into the equipment load constraint inequality: ; In this formula, stage The load, To configure the security load factor in the database, the value range is limited to (0,1]. The set traversal operator means that this constraint must be true simultaneously in three specific phases: Item acquisition phase During the circulation of items within the system, The system categorizes data based on destination. When the estimated load for any dimension of a candidate wave exceeds the corresponding inequality upper limit, the business management layer halts the generation of that candidate wave's data combination and returns the excess order data to the pending order pool. Under the condition that the above inequality constraints are satisfied, the system calculates... The sample variance of each ratio to its respective upper limit is used to calculate the reciprocal of the sample variance, and the reciprocal is assigned to the device dimension balanced scalar score. .

[0048] The business management layer performs capacity constraint comparisons at the grid level. The system reads the physical destination mapping routing table and accumulates and calculates candidate wavelets. Assign to the target sorting grid The total quantity of physical goods, marked as The system reads the physical capacity limit parameter of the sorting compartment through the communication interface. And the current number of items occupied in that sorting slot. .

[0049] Business management determines the following inequality conditions: ; in, Record candidate waves The set of all target grids pointed to. When the inequality determination result of any target grid is false, the system restricts the orders assigned to that conflicting target grid to be merged into the current candidate wave set.

[0050] Once a candidate wave passes all the aforementioned constraint inequalities and the objective function is calculated, the business management layer assigns a unique candidate wave identifier. The business management layer generates a candidate wave task document file, whose data format includes the candidate wave identifier, timeliness tag, order detail data rows, minimum inventory unit extraction list, and target sorting grid code. The business management layer sends an inventory lock command to the database to deduct the corresponding available inventory and sends a device status heartbeat query request to the scheduling and monitoring layer. When a normal status response message is received, the business management layer distributes the candidate wave task document file to the scheduling and monitoring layer through the data interaction module.

[0051] This invention provides a smart logistics multi-candidate wave picking method, which may include candidate wave picking and equipment action control steps executed in the scheduling monitoring layer and the electronic control layer.

[0052] The scheduling and monitoring layer receives candidate wave task documents from the business management layer via the data interaction module. The equipment collaborative scheduling module of the scheduling and monitoring layer extracts the minimum inventory unit extraction list and the target sorting grid code field from the candidate wave task document. Based on the physical node topology map stored internally by the system, the equipment collaborative scheduling module decomposes the candidate wave tasks into picking sub-task sets, conveying sub-task sets, supply sub-task sets, and sorting sub-task sets. The equipment collaborative scheduling module maps these sub-task sets to the corresponding hardware execution layer physical device identifier codes, generating a set of instruction sequences containing physical device identifier codes, action instruction codes, and target physical coordinates.

[0053] The scheduling and monitoring layer sends a sequence of instructions corresponding to the picking sub-task set to the electrical control layer. The electrical control layer parses the instruction sequence and outputs electrical signals to drive the stacker crane and multi-level shuttle within the warehouse picking unit. The stacker crane and multi-level shuttle move to the three-dimensional storage location coordinates specified in the instruction sequence and physically retrieve the corresponding bins or pallets. Subsequently, the electrical control layer drives the elevator to transport the bins or pallets along the vertical guide rails to the designated physical picking station.

[0054] At the picking station, a handheld RFID terminal scans the barcode on the surface of the retrieved goods, generating a scan feature string which is then uploaded to the scheduling and monitoring layer. The scheduling and monitoring layer performs a string equality comparison between the scan feature string and the expected data in the candidate wave task document file. When the comparison result is true, the scheduling and monitoring layer sends a release level signal to the control node of the picking station, and the goods are physically transferred to the roller conveyor of the conveying buffer unit. When the comparison result is false, the scheduling and monitoring layer sends an interception alarm signal.

[0055] Upon entering the transport buffer unit, the path planning module of the scheduling and monitoring layer reads the time-sensitivity tag value of the candidate wave to which the goods belong. For goods with a time-sensitivity tag value of 1, the path planning module assigns a direct physical route with the fewest physical nodes. For goods with a time-sensitivity tag value greater than 1, the path planning module assigns a regular transport route. Based on the assigned route, the electrical control layer sequentially triggers the motor drive circuits of the roller conveyor, belt conveyor, and lifting transfer machine in chronological order to control the physical transfer and merging operations of the goods between the transport nodes.

[0056] The real-time monitoring module of the scheduling and monitoring layer executes a dynamic adjustment program for the entire operation rhythm. The system sets up a buffer zone on the physical conveyor line at the front end of the sorting machine. The real-time monitoring module calculates the current physical inventory of goods in the buffer zone by collecting the number of times the photoelectric sensors in the buffer zone are blocked, and records this value. The system database has a pre-written lower limit hunger threshold. With upper limit congestion threshold .

[0057] The scheduling and monitoring layer continuously compares data at set time intervals. The numerical relationship between the comparison results and the two thresholds mentioned above. When the scheduling and monitoring layer determines that the sorting front-end processing capacity is saturated, it sends a parameter update instruction to the electronic control layer to reduce the operating speed of the warehouse picking unit. At the same time, it triggers the circular shuttle in the conveying buffer unit to transfer the subsequently arriving goods to the circular buffer track for physical storage.

[0058] When the comparison results meet At that time, the scheduling and monitoring layer determines that the supply of parts at the sorting front end is insufficient, and issues a parameter update instruction to the electronic control layer to increase the retrieval frequency of the warehouse picking unit, and instructs the goods remaining on the circular buffer track to rejoin the main conveyor line. When the comparison result meets the requirements... Meanwhile, the scheduling and monitoring layer maintains the baseline operating speed parameters of each unit. Through the above-mentioned conditional branch judgment and control command issuance, the system controls the cargo displacement rate of each node in the entire physical link to be within the set constraint range.

[0059] This invention provides an intelligent logistics multi-candidate wave picking method, which may include physical feature acquisition and sorting action control steps performed in the supply preprocessing unit and the sorting execution unit.

[0060] Goods enter the supply pre-processing unit from the physical end of the conveying buffer unit. Within the supply pre-processing unit, a dynamic weighing device collects the actual mass value of the goods, a barcode reader captures the optical features of the goods' surface and decodes them into an actual identification string, and a shape detection device obtains the three-dimensional physical boundary coordinates of the goods and calculates their actual dimensions using an infrared or laser sensor array. These sensor nodes package the actual mass value, the actual identification string, and the actual dimensions into a multi-dimensional physical feature vector and send it to the scheduling and monitoring layer via industrial Ethernet.

[0061] The scheduling and monitoring layer receives the aforementioned multi-dimensional physical feature vectors and reads the expected parameter matrix from the corresponding candidate wave task document file stored by the business management layer. The scheduling and monitoring layer extracts the actual identity string from the multi-dimensional physical feature vectors and performs a full-character matching search in the expected parameter matrix. When a corresponding expected record is found, the scheduling and monitoring layer further calculates the absolute value of the difference between the actual quality value and the expected quality value. The system has a tolerance threshold parameter. When the absolute value of the difference is less than or equal to the tolerance threshold and the actual identity string matches successfully, the scheduling and monitoring layer determines that the cargo feature verification has passed and outputs a release status code with a Boolean value of "true". If the actual identity string is not found, or the absolute value of the difference is greater than the tolerance threshold, the scheduling and monitoring layer outputs an abnormal status code with a Boolean value of "false".

[0062] When the release status code is false, the electronic control layer sends an interception command to the pre-processing unit, controlling the physical steering mechanism to transfer the corresponding goods to the physical anomaly buffer line. When the release status code is true, the electronic control layer sends a synchronous supply command to the inbound table. The control node of the inbound table reads the current physical operating speed parameters of the sorting host, generates the corresponding acceleration control curve, and drives the variable frequency motor at the bottom of the inbound table. The variable frequency motor drives the belt to accelerate until the difference between the linear velocity of the belt surface and the physical linear velocity of the carrier assembly of the sorting host falls within the preset allowable range, and then the goods are horizontally transferred onto the physical bearing surface of the carrier assembly.

[0063] After the goods enter the sorting host, the path planning module of the scheduling and monitoring layer extracts the target sorting slot code bound to the actual identity string. Based on the physical layout parameters of the rack track system, the path planning module calculates the displacement distance and estimated arrival time of the carrier trolley assembly from its current physical coordinates to the physical coordinates of the target sorting slot. The scheduling and monitoring layer converts this estimated arrival time into an action trigger timestamp for the electronic control layer and sends it to the electronic control layer. When the system clock reaches the action trigger timestamp, the electronic control layer outputs a high-level trigger signal to the drive solenoid valve of the unloading mechanism. The physical components of the unloading mechanism perform lateral displacement or pulley rotation, applying a lateral physical force to the side of the goods, causing the goods to detach from the carrier trolley assembly and fall into the target sorting slot.

[0064] During the sorting process, the full-grid detection device of the grid processing unit periodically collects the physical stacking height parameter within the target sorting grid. The full-grid detection device sends this physical stacking height parameter to the scheduling and monitoring layer. The scheduling and monitoring layer compares this parameter with the maximum physical capacity of the grid recorded in the database. When the physical stacking height parameter reaches a preset percentage of the maximum physical capacity, the scheduling and monitoring layer updates the status attribute field of the target sorting grid to a locked state and sends an activation level signal to the audible and visual alarm device. For goods in subsequent candidate wave tasks that point to this locked grid, the scheduling and monitoring layer reads the physical sorting grid backup mapping table, updates the target sorting grid code to the backup code, and recalculates the action trigger timestamp. If the backup mapping table is empty, the scheduling and monitoring layer routes the corresponding goods to the abnormal return end of the sorting host.

[0065] The scheduling and monitoring layer is equipped with a candidate wave completion counter. Each time the unloading mechanism successfully executes a physical loading action and places goods into the sorting slot, the counter value increments by one. The scheduling and monitoring layer compares the counter value with the total number of orders recorded in the candidate wave task document file at set clock cycles. When the counter value equals the total number of orders, the scheduling and monitoring layer generates a candidate wave physical completion signal and sends it back to the business management layer via the data interaction module. Upon receiving the candidate wave physical completion signal, the business management layer updates the execution status field of the candidate wave in the database to a closed-loop status and triggers the inventory management module to perform corresponding deduction and offset calculations on the total physical inventory value.

[0066] This invention provides an intelligent logistics multi-candidate wave picking method, which may include abnormal data monitoring, differential calculation and physical route reset steps executed in the business management layer and the scheduling monitoring layer.

[0067] The business management layer is configured with a background monitoring service program that continuously receives data update messages from external systems via a communication interface, and simultaneously receives physical status data collected by the hardware execution layer through the scheduling and monitoring layer. The business management layer uses this received data to trigger order change processing or inventory anomaly handling processes.

[0068] In the order change processing flow, when the data update message contains an order identifier code from a candidate wave task document file that is currently in physical execution, the exception handling module of the business management layer reads the message structure and extracts the updated minimum inventory unit matrix. The business management layer retrieves the original minimum inventory unit matrix bound to the order identifier code from the database and performs a difference calculation by subtracting the values ​​of the same type of goods in the two matrices, outputting a difference matrix containing the difference results. The business management layer iterates through the numerical elements in the difference matrix, extracting the numerical elements greater than zero and constructing a new demand submatrix, and extracting the numerical elements less than zero and constructing a canceled demand submatrix.

[0069] In the inventory discrepancy handling process, when the physical quantity of goods collected by the RFID handheld terminal or sensors of the warehouse picking unit is less than the expected quantity issued in the candidate wave task document, the electronic control layer generates an inventory discrepancy parameter and uploads it to the business management layer through the scheduling and monitoring layer. Based on this inventory discrepancy parameter, the business management layer deducts the total physical available inventory of the corresponding product item at that location coordinate in the database by an equal amount. The business management layer compares the deducted total physical available inventory with the total demand for the same type of goods that have not yet been picked in the current candidate wave. When the deducted total physical available inventory is less than the total demand, the business management layer calculates the quantity difference between the two and incorporates this difference into the aforementioned canceled demand sub-matrix.

[0070] For the non-empty data characteristics within the newly added demand sub-matrix, the business management layer generates a compensation task order. The data fields of the compensation task order include the newly added minimum inventory unit code, the physical increment value, and the target sorting grid code assigned in the original candidate wave task order file. The business management layer sends this compensation task order to the task receiving port of the scheduling and monitoring layer. The equipment coordination scheduling module of the scheduling and monitoring layer parses the compensation task order, generates an additional picking instruction sequence containing the three-dimensional physical coordinates of the target location, and sends it to the electrical control layer. The electrical control layer outputs pulse signals to drive the stacker crane and hoist in the aisle to the three-dimensional physical coordinates of the target location to retrieve the physical goods and release them to the physical merging node of the conveyor buffer unit to catch up with the physical execution progress of the original candidate wave.

[0071] For non-empty data features within the cancelled demand submatrix, the business management layer generates a status freeze command and synchronizes it to the scheduling and monitoring layer. The scheduling and monitoring layer extracts the actual identity string or barcode code of the cancelled goods and calls the real-time monitoring module to query the current physical node coordinate parameters corresponding to that string.

[0072] When the physical node coordinate parameters returned by the query show that the goods are in the warehouse picking unit and have not completed the physical separation action, the scheduling and monitoring layer sends an emergency stop command to the electrical control layer to stop the operation sequence of the goods, and sends a reverse return command sequence to drive the execution equipment to send the corresponding bin or pallet back to the storage location of the automated warehouse according to the original physical path.

[0073] When the physical node coordinates returned by the query indicate that the goods are already on the closed-loop physical route of the conveying buffer unit or sorting host, the path planning module of the scheduling and monitoring layer triggers an update operation of the routing address mapping table. The system locks control over the original target sorting slot code and modifies the destination address pointer of the goods to the physical code of the return buffer area pre-set on the hardware execution layer side. When the goods physically move with the conveying equipment to the diversion node or unloading node corresponding to the return buffer area, the electrical control layer outputs an action trigger level signal to the corresponding lifting transfer motor or unloading mechanism solenoid valve, applying physical thrust or traction force to cause the cancelled goods to leave the main material flow and enter the return buffer area. Subsequently, the scheduling and monitoring layer marks the actual identity string record of the goods as intercepted and uploads it to the business management layer to update the system database record.

[0074] This invention provides a smart logistics multi-candidate wave picking method, which may include equipment status monitoring, task freezing, topology reconstruction and instruction sequence reallocation steps executed in the scheduling monitoring layer and the electronic control layer.

[0075] The electrical control layer continuously reads the sensor level signals and driver status words of each physical device in the hardware execution layer via industrial Ethernet at a set sampling period. When the electrical control layer detects that the driver status word of a physical device matches the system's preset fault mask parameters, or that the communication heartbeat message response time of the physical device exceeds a set value in milliseconds, the electrical control layer generates a hardware interrupt message containing the physical device's identification code and fault code, and uploads the message to the communication receiving port of the scheduling and monitoring layer.

[0076] The fault diagnosis module of the scheduling and monitoring layer parses the received hardware interrupt message and changes the running status field bound to the physical device identifier in the database from online to offline. The path planning module of the scheduling and monitoring layer stores a physical node topology graph matrix representing the connection relationships of the hardware execution layer in system memory. When an offline status update command is received, the path planning module modifies the vertex coordinates corresponding to the offline physical device and the weight values ​​of all connected edges in the physical node topology graph matrix to infinity, generating an updated remaining connected graph matrix.

[0077] The device collaborative scheduling module in the scheduling and monitoring layer triggers the task recycling logic, retrieving a set of instruction sequences from the system database that were assigned to the offline physical device and whose execution status field is either "not started" or "issued but not physically acted upon." The device collaborative scheduling module freezes this set of instruction sequences, suspending the transmission of subsequent action trigger level signals to the electrical control layer.

[0078] For the aforementioned frozen instruction sequence set, the device collaborative scheduling module retrieves a set of available physical device identifiers from the device attribute table in the database that have the same type of fields and the same rated processing limit parameters as the offline physical device. The module then selects the physical device identifier with the smallest number of currently backlogged instruction sequences from the available physical device identifier set, overwrites and modifies the fields of the original physical device identifier in the frozen instruction sequence set, and generates a new mapping relationship. The module then unfreezes the sequence and sends the modified instruction sequence set to the electrical control layer to drive the alternative physical device to perform physical retrieval or handling actions.

[0079] For a cargo entity whose current physical coordinates are located upstream of an offline physical device and whose physical path is via that offline physical device, the path planning module of the scheduling and monitoring layer extracts the current physical coordinate node identifier and the target sorting slot physical node identifier of the cargo. Using these two node identifiers as the starting and ending points, the path planning module performs shortest path numerical calculations on the generated remaining connected graph matrix, outputting a sequence of backup physical nodes that does not contain edges with infinite weights. The scheduling and monitoring layer converts this backup physical node sequence into an action trigger timestamp sequence and sends it to the electrical control layer to drive the lifting and transferring machine in the conveying buffer unit, changing the physical conveying direction of the cargo entity.

[0080] When the uploaded offline physical device identifier corresponds to a sorting host in the sorting execution unit, the scheduling and monitoring layer sends a candidate wave suspension request data packet containing the current candidate wave identifier to the business management layer. After receiving the data packet, the business management layer modifies the status field of the corresponding candidate wave task document file to the paused execution status and stops issuing new candidate wave tasks to the scheduling and monitoring layer.

[0081] During this period, the equipment coordination scheduling module of the scheduling and monitoring layer sends a system-level buffer diversion command to the electrical control layer. Based on this command, the electrical control layer continuously outputs a high-level signal, driving the circular shuttle and related transfer and sorting mechanisms within the conveying buffer unit. The electrical control layer controls these mechanisms to sequentially transfer all physical goods currently on the main conveyor line but not yet entering the supply pre-processing unit to the preset physical backup buffer track area for physical storage, avoiding backlog caused by physical blockage of the main conveyor line. When the scheduling and monitoring layer receives a message indicating that the sorting host's status word has returned to a fault-free code, it restores the execution status of the candidate wave task document and issues a release command, driving the output motor of the backup buffer track area to start, and sequentially re-entering the stored physical goods into the main conveyor line.

[0082] This invention provides an intelligent logistics multi-candidate wave picking method, which may include physical status monitoring, grid attribute updating and sorting destination address redirection steps performed in the grid processing unit and scheduling monitoring layer.

[0083] The full-load detection device within the sorting unit is installed on the inner wall or bottom of each physical sorting slot. This device collects the stacking height or total mass parameters of the goods within the sorting slot and converts these parameters into continuous analog voltage signals or digital data. The full-load detection device transmits these signals or data to the input module of the electronic control layer via hardwired connection or industrial bus. The electronic control layer then encapsulates these signals into status messages and continuously uploads them to the scheduling and monitoring layer. The scheduling and monitoring layer has a capacity threshold field configured in its internal database for each physical sorting slot. The scheduling and monitoring layer extracts the values ​​from the status messages at set clock cycles and compares them with the corresponding capacity threshold field. When the received parameter value is greater than or equal to the value set in the capacity threshold field, the scheduling and monitoring layer determines that the physical sorting slot has reached full load.

[0084] After determining the physical sorting slot is at full capacity, the scheduling and monitoring layer modifies the corresponding status attribute field in the system database from available to locked. Simultaneously, the scheduling and monitoring layer sends an alarm trigger command to the electrical control layer. After parsing the command, the electrical control layer outputs a drive voltage to the control relay of the audible and visual alarm device installed outside the physical sorting slot, activating the operating circuit and causing the audible and visual alarm device to emit a fixed-frequency sound wave and a flashing LED signal.

[0085] For subsequent goods awaiting sorting in the candidate wave task document file whose target address points to a physical sorting grid that is already locked, the scheduling and monitoring layer triggers the backup grid mapping logic. The scheduling and monitoring layer has a backup mapping database table pre-loaded in its memory. This database table records the mapping pointer relationship between the primary sorting grid code and at least one backup sorting grid code, and synchronously updates and records the current status attribute fields of each backup sorting grid.

[0086] The path planning module of the scheduling and monitoring layer extracts the original primary sorting grid code bound to the goods to be sorted and performs pointer retrieval in the backup mapping database table. The path planning module reads the backup sorting grid codes associated with the original primary sorting grid codes sequentially according to the priority sequence set in the table, and reads their corresponding status attribute fields. When the retrieval program matches a backup sorting grid record with an available status attribute field, the scheduling and monitoring layer stops the retrieval and overwrites the target sorting grid code field in the data row of the goods to be sorted, updating its value to the available backup sorting grid code.

[0087] After the overwrite operation is completed, the path planning module reads the physical node coordinate parameters of the rack track system. The path planning module uses the current physical coordinates of the carrying trolley assembly carrying the goods as the starting point and the updated physical interface coordinates of the backup sorting slot as the ending point, calculating the arc distance between them along the physical running trajectory. The scheduling and monitoring layer extracts the current linear velocity parameter of the carrying trolley assembly, performs a division operation of distance and velocity, and obtains the estimated physical time difference for the goods to arrive at the backup sorting slot. The scheduling and monitoring layer adds the current system time to this estimated physical time difference to generate an update trigger timestamp, and sends it to the electrical control layer. When the system hardware clock reaches this update trigger timestamp, the electrical control layer outputs a drive level to the electromagnetic coil of the unloading mechanism at the corresponding physical location of the backup sorting slot, triggering a physical unloading action and dropping the goods into the backup sorting slot.

[0088] When the retrieval program traverses the backup mapping database table and finds that the status attribute fields of all associated backup sorting slots are locked, or that the original primary sorting slot has no associated backup mapping record in this table, the scheduling and monitoring layer executes an abnormal backflow instruction sequence. The scheduling and monitoring layer modifies the target sorting slot code field of the goods to be sorted to the receiving slot code of the physical system's end branch. The scheduling and monitoring layer issues a hold instruction to not execute the original unloading action, and the electronic control layer maintains the carrier trolley assembly's continued movement along the main closed-loop physical track, causing the goods to physically detour until they enter the receiving slot area. The scheduling and monitoring layer records the goods' actual identity string and marks its execution status as suspended and pending in the database until the original primary sorting slot or backup sorting slot is physically cleared and its status attribute fields are reset to available by the system.

[0089] This invention provides an intelligent logistics multi-candidate wave picking method, which may include execution data collection, loss function calculation and weight parameter numerical iteration steps performed within the business management layer.

[0090] The data statistics module within the business management layer triggers a data archiving procedure for the corresponding candidate wave identifier code upon receiving a signal indicating the completion of physical execution for any candidate wave. The module extracts the actual start and end timestamps of the candidate wave from the database's operation log table and performs a subtraction operation to calculate the total actual physical execution time. Simultaneously, the module extracts the fulfillment deadline timestamps of all orders included in the candidate wave and compares them with the actual end timestamps. When the actual end timestamp is greater than the fulfillment deadline, the system marks the order as overdue and counts the total number of overdue orders within the candidate wave. The module also extracts the driver operating status timeline data of each physical device uploaded from the electronic control layer, calculates the ratio of the time each physical device is in standby mode to the total actual physical execution time, and generates a device idle rate parameter. Finally, the module packages the total actual physical execution time, the total number of overdue orders, and the device idle rate parameter into a candidate wave execution feature vector and associates it with the candidate wave identifier code, writing it to the feature database table on the system hard drive.

[0091] The business management layer has a pre-set parameter iteration trigger counter in memory. When the number of newly added candidate wave execution feature vectors in the feature database table reaches a preset numerical threshold, the business management layer initiates the weight parameter iteration calculation program. The business management layer extracts the set of newly added candidate wave execution feature vectors and calculates the actual execution loss function for these candidate waves. The actual execution loss function is a linearly weighted sum of the sample mean of the total number of overdue orders and the sample mean of the equipment idle rate parameter. Its output value characterizes the degree of physical execution deviation of the system under the current set of weight parameters. In this loss function, the independent variable weight parameter set vector contains three dimensions, namely... . These correspond to the multiplier weights of the scalar scores for timeliness, space, and equipment in the dynamic candidate wave planning module.

[0092] The business management layer invokes the floating-point unit within the microprocessor to execute a sequence of weight update instructions based on gradient descent logic. The calculation formula is as follows: ; In this formula, Indicates the first The current weight vector value stored in the database within each calculation cycle; This represents the updated weight vector value after calculation and output; This represents the normal step size configured in the system parameter table, and its value range is set to (0, 0.1]. This represents the vector of partial derivatives of the actual execution loss function with respect to the weight parameter set vector. The business management layer modifies the values ​​of the weight vector in the opposite direction of the gradient, causing the calculated output value of the loss function to decrease in the next cycle.

[0093] Floating-point arithmetic unit output Subsequently, the business management layer performs a physical constraint boundary truncation and projection normalization process on the updated weight vector values. The system sets the lower limit threshold for each weight component to zero, i.e., for... Each component in Make a judgment. When At that time, the system performs an assignment operation. Subsequently, the system performs vector component summation calculation. The business management team divides each component by the summation result. Output the final normalized weight values.

[0094] The business management layer overwrites the final normalized weight values ​​into the corresponding parameter memory address in the cargo owner rule configuration module. Before the next calculation cycle arrives, the dynamic candidate wave planning module reads the overwritten weight values ​​as multiplier parameters for calculating the candidate wave comprehensive evaluation objective function, controlling the generation process of candidate wave task combinations. After the calculation is complete, the business management layer resets the parameter iteration trigger counter to zero.

[0095] Specific application examples: During a major promotional event at a large e-commerce logistics center, there is a massive, high-frequency outbound demand with stringent time requirements. The intelligent logistics multi-candidate wave picking system of this invention executes its operations according to the following process: The business management layer receives an outbound order data package containing 100,000 orders, extracts fields, performs compliance checks, and generates a pool of orders to be processed. The dynamic candidate wave planning module extracts the order fulfillment deadline timestamp, target physical location coordinates, and current equipment load, and calculates scalar scores for timeliness, space, and equipment balance, respectively. The system weights and sums these three scalar scores to derive a comprehensive evaluation objective function, generating the first batch of candidate wave task documents while satisfying the constraints of equipment load limits and sorting slot capacity limits.

[0096] The scheduling and monitoring layer receives the task document for the candidate wave and decomposes it into a set of instruction sequences for distribution. At this time, the electrical control layer drives the stacker crane and multi-level shuttle car of the hardware execution layer to the target cargo location at its three-dimensional physical coordinates to retrieve the material box.

[0097] During the conveying and buffering process, the real-time monitoring module collects the number of times the photoelectric sensors in the buffer area of ​​the conveying and buffering unit are blocked, and calculates that the current physical inventory of goods has reached 85% (the preset upper limit congestion threshold is set at 80%). The scheduling and monitoring layer immediately issues a parameter update instruction to the electronic control layer to reduce the operating speed of the warehouse picking unit, and triggers the circular shuttle to transfer some goods to the circular buffer track for physical storage, thus avoiding physical blockage of the downstream sorting execution unit.

[0098] In the sorting process, when the full-grid detection device of a sorting grid detects that the physical goods stacking height has reached the capacity threshold field, the scheduling and monitoring layer modifies the grid status attribute field to the locked state, retrieves the available backup sorting grid code in the backup mapping database table, overwrites the target sorting grid code field in the data row of the goods to be sorted, calculates the estimated physical time difference based on the current line velocity, generates a new action trigger timestamp to execute physical loading, and realizes automatic fault tolerance without stopping the machine.

[0099] As the operation progresses, the data statistics module extracts the total actual physical execution time, the total number of overdue orders, and the equipment idle rate parameters from the first 50 candidate waves to generate candidate wave execution feature vectors. Since the preset numerical threshold has been reached, the business management layer calculates the actual execution loss function, iteratively updates the partial derivative execution weight parameters, and automatically adjusts the weight ratio of timeliness and equipment load in the planning of subsequent candidate waves.

[0100] Experimental verification and effect comparison: To verify the effectiveness of the present invention, a comparative experiment was conducted between the system of the present invention and the traditional fixed-time-window candidate wave picking system under the same hardware equipment scale and historical promotional order dataset.

[0101] As attached Figure 3 The diagram illustrates a comparison of the hourly throughput of the two systems over a 24-hour work cycle. During peak business periods (hours 8-12 and 18-22), the traditional system, employing fixed candidate wavelets, experiences a sudden surge in front-end picking volume exceeding the back-end picking capacity, leading to physical congestion and a significant drop in throughput. In contrast, the system of this invention achieves precise matching between picking and sorting rhythms through closed-loop dynamic adjustment of the inventory value in the transport buffer stage and multi-dimensional dynamic candidate wavelet planning, maintaining a high and stable throughput throughout the day. Its peak processing capacity is approximately 35% higher than that of the traditional system.

[0102] As attached Figure 4As shown, this paper illustrates the convergence trend of the overdue order rate as the system continuously executes candidate wave tasks. In the initial stage of system startup, a small number of overdue orders exist because the initial weight parameters do not yet fully match the current physical distribution. Once the accumulated samples reach the numerical threshold triggering parameter updates, the system executes a weight update instruction sequence based on gradient descent logic. As the number of iterative candidate waves increases, the actual execution loss function gradually decreases, and the overdue order rate exhibits a significant step-like decrease, eventually stabilizing at a low level below 0.2%, fully verifying the authenticity and effectiveness of the self-learning update mechanism for weight parameters in this invention.

Claims

1. An intelligent logistics multi-candidate wave picking system, characterized in that, include: The business management layer is used to receive outbound order data packets, extract fields from the outbound order data packets to generate an initial order set, perform compliance verification and tagging processing on the initial order set to generate a pool of orders to be processed, divide the pool of orders to be processed based on the set constraints, and generate and issue candidate wave task document files. The scheduling and monitoring layer is used to receive the candidate wave task document file, decompose the candidate wave task into a set of instruction sequences containing physical device identification codes and target physical coordinates, and send the set of instruction sequences to the electronic control layer. At the same time, it receives the running status data of the hardware execution layer to perform dynamic adjustment and abnormal route reconstruction. The electronic control layer is used to receive the instruction sequence set, output electrical signals to drive the hardware execution layer to perform mechanical actions, and collect sensor signals from the hardware execution layer and upload them to the scheduling and monitoring layer. The hardware execution layer is controlled by the electronic control layer and is used to perform physical extraction, transfer, pretreatment and sorting of materials.

2. The intelligent logistics multi-candidate wave picking system according to claim 1, characterized in that, The business management layer includes a dynamic candidate wave planning module, which is used to calculate the comprehensive evaluation objective function of the candidate waves; The calculation logic of the comprehensive evaluation objective function includes: Calculate the timeliness dimension scalar score of the difference between the fulfillment deadline timestamp and the current timestamp for all orders within the candidate wave; calculate the spatial dimension scalar score of the distance cost of the physical location coordinates of the minimum target inventory unit within the candidate wave; calculate the equipment dimension balance scalar score of the inverse of the variance of the theoretical load estimate allocated to the equipment in the candidate wave. The dynamic candidate wave planning module performs a weighted summation of the three scalar scores and generates the candidate wave task document file when it meets the preset convergence threshold used to characterize the numerical stability of the objective function, as well as the upper limit constraints of the equipment load capacity and the upper limit constraints of the physical capacity of the sorting slot.

3. The intelligent logistics multi-candidate wave picking system according to claim 1, characterized in that, The hardware execution layer includes a warehouse picking unit, a conveying buffer unit, a component preprocessing unit, a sorting execution unit, and a sorting slot processing unit; The warehouse picking unit includes an aisle stacker crane, a multi-level shuttle car and a hoist, used to retrieve bins or pallets to the picking station according to the coordinates of the target location. The conveying and buffering unit is connected between the warehouse picking unit and the supply preprocessing unit, and is used to perform the conveying and buffering operations of goods; The pre-processing unit for the goods is located at the output end of the conveying buffer unit and is used to collect physical characteristic parameters of the goods and transfer the goods to the sorting execution unit. The sorting execution unit includes a sorting host, a frame track system, and a carrier trolley assembly, which is used to drive the goods to move continuously on the physical track; The sorting compartment processing unit is located on the output side of the sorting execution unit and includes physical sorting compartments and a full-compartment detection device, used to receive goods discharged from the sorting host.

4. The intelligent logistics multi-candidate wave picking system according to claim 3, characterized in that, The scheduling and monitoring layer includes a real-time monitoring module, which is used to collect the number of times the photoelectric sensor in the buffer area of ​​the conveying buffer unit is blocked, and to calculate the physical cargo inventory value in the buffer area at the current time. When the physical inventory of goods exceeds the preset upper limit congestion threshold used to characterize the saturation of the sorting front-end processing capacity, the scheduling and monitoring layer sends a parameter update instruction to the electronic control layer to reduce the operating speed of the warehouse picking unit, and triggers the circular shuttle to transfer the goods to the circular buffer track for physical storage. When the physical inventory of goods is less than the preset lower limit hunger threshold used to characterize insufficient supply at the sorting front end, the scheduling and monitoring layer issues a parameter update instruction to increase the extraction frequency of the warehouse picking unit.

5. The intelligent logistics multi-candidate wave picking system according to claim 3, characterized in that, The pre-processing unit for the supplies includes a dynamic weighing device, a barcode reader, and a shape detection device, used to obtain the actual weight value, actual identity string, and actual shape dimension value of the goods, and package them into a multi-dimensional physical feature vector and upload them to the scheduling and monitoring layer. The scheduling and monitoring layer is used to compare the multidimensional physical feature vector with the expected parameter matrix. When the absolute value of the difference between the actual quality value and the expected quality value is less than or equal to the tolerance threshold used to characterize the allowable range of quality deviation and the actual identity string is successfully matched, a synchronous supply instruction is generated and sent to the import station. The infeed platform is used to run according to a preset acceleration curve until the difference between the linear velocity of the belt surface and the physical linear velocity of the carrying trolley assembly falls within a preset range to perform horizontal transfer.

6. The intelligent logistics multi-candidate wave picking system according to claim 3, characterized in that, The business management layer is used to perform differential calculations based on data update messages to generate a difference matrix containing a new demand submatrix and a canceled demand submatrix. For the newly added demand submatrix, the business management layer generates a compensation task order and sends it to the scheduling and monitoring layer, driving the hardware execution layer to go to the three-dimensional physical coordinates of the target cargo location to extract the physical goods to the physical merging node; For the cancelled demand submatrix, the scheduling and monitoring layer queries the current physical node coordinate parameters of the goods. When the goods are in the warehouse picking unit, a reverse return instruction sequence is issued. When the goods are in the closed-loop physical route of the transport, the destination address pointer is modified to the physical code of the return buffer area to execute the interception.

7. The intelligent logistics multi-candidate wave picking system according to claim 1, characterized in that, The scheduling and monitoring layer includes a path planning module and a device collaborative scheduling module; When the scheduling and monitoring layer receives a hardware interruption message indicating that a physical device has failed and is offline, it modifies the weight value of the connected edge of the corresponding vertex in the physical node topology graph matrix to an infinite value and generates an updated remaining connected graph matrix. The device collaborative scheduling module is used to freeze the set of instruction sequences allocated to offline physical devices, retrieve the set of available physical device identifiers in the physical device attribute table, and then overwrite and modify the original physical device identifier field in the frozen instruction sequence set before issuing it. The path planning module is used to perform shortest path numerical calculations on the remaining connected graph matrix and output a sequence of backup physical nodes to control the physical transport direction of goods.

8. The intelligent logistics multi-candidate wave picking system according to claim 3, characterized in that, The full-grid detection device of the sorting grid processing unit is used to collect parameters of the physical goods in the sorting grid and upload them to the scheduling and monitoring layer. When the parameter is greater than or equal to the capacity threshold field used to characterize the physical goods stacking height or total mass of the sorting compartment reaching the loading limit, the scheduling and monitoring layer will modify the status attribute field of the corresponding physical sorting compartment to the locked state. The scheduling and monitoring layer is used to retrieve the backup sorting grid code associated with the original main sorting grid code and whose status attribute field is available in the backup mapping database table, overwrite the target sorting grid code field in the data row of goods to be sorted, and calculate the expected physical time difference in combination with the current physical coordinates and linear velocity parameters, generate an update trigger timestamp and send it to the electronic control layer to execute the physical loading action.

9. The intelligent logistics multi-candidate wave picking system according to claim 2, characterized in that, The business management layer includes a data statistics module; the data statistics module is used to extract parameters such as the total actual physical execution time, the total number of overdue orders, and the equipment idle rate, and package them into candidate wave execution feature vectors. When the number of newly added candidate wave execution feature vectors reaches a preset numerical threshold that represents the minimum sample size threshold for triggering parameter updates, the business management layer calculates the true execution loss function obtained by weighted summation of the sample mean of the total number of overdue orders and the sample mean of the equipment idle rate parameter, executes a weight update instruction sequence based on gradient descent logic, calculates the partial derivative of the true execution loss function with respect to the weight parameter set vector, outputs the updated weight vector values, performs truncation and projection normalization processing, and overwrites them to the parameter memory address for the dynamic candidate wave planning module to call.

10. A smart logistics multi-candidate wave picking method, characterized in that, The intelligent logistics multi-candidate wave picking system according to any one of claims 1-9 includes the following steps: Receive outbound order data packets, extract fields to perform compliance checks including ownership, total inventory, destination, and information integrity, and convert them into numeric or enumerated tag variables to generate a pool of orders to be processed; Read the order pool to be processed, perform combined optimization calculations based on timeliness, space and equipment dimensions to obtain a comprehensive evaluation objective function, and generate candidate wave task document files after satisfying the upper limit constraints of equipment load capacity and sorting slot physical capacity. The candidate wave task document file is decomposed into a set of instruction sequences and sent to the electronic control layer to drive the hardware execution layer to extract the material box or pallet at the corresponding physical location coordinates. The physical inventory of goods in the buffer zone is collected during the transport and buffering process, and the operating speed is dynamically adjusted by comparing it with the preset upper limit congestion threshold or the preset lower limit hunger threshold, which characterizes the processing capacity of the sorting front end. In the pre-processing stage of the supply, multi-dimensional physical feature vectors are collected for numerical and string comparison. After the comparison is successful, the import table is controlled to perform synchronous transfer at a certain speed. At the set action trigger time stamp, the sorting host unloading mechanism is controlled to discharge the goods into the physical sorting slot. The system collects data from the entire execution process to generate a real execution loss function, calculates partial derivatives, and iteratively updates the execution weight parameters.

Images