Supply chain warehouse management system and method based on AI digital intelligence

Abstract

The invention discloses a supply chain warehouse management system and method based on AI digital intelligence, and relates to the technical field of intelligent warehousing, and the method comprises the steps that a data processing unit collects supply chain system data through a data processing module, and constructs an associated data network containing goods information, shipping space information and order information based on the supply chain system data; the goods allocation unit executes a goods allocation algorithm according to the associated data network, extracts an association score based on goods information, analyzes a collaborative ex-warehouse frequency, and calculates an affinity matrix; the storage scheme unit generates a storage position scheme based on the affinity matrix; the job scheduling unit constructs a scheduling model, performs task optimization and outputs a task allocation sequence; the order picking navigation unit is provided with an order picking path planning module, presents a path and guidance, collects execution data, and optimizes the system through analysis. The problems that in traditional warehouse management, the goods allocation efficiency is low, task scheduling is not optimized, and the goods picking path is unreasonable are solved.

Description

Technical Field

[0001] This application relates to the technical field of intelligent warehousing, and in particular to a supply chain warehousing management system and method based on AI digitalization. Background Art

[0002] In the prior art, traditional warehousing management systems mainly rely on manual experience for goods location allocation and picking planning, and there are many limitations. First, it is difficult for such systems to deeply explore the relevance between goods and unable to identify combinations of goods that are frequently shipped together based on historical order data. Second, the storage plans manually planned often cannot fully consider the optimal matching of warehouse location information and goods characteristics. Third, traditional systems lack intelligent optimization capabilities in operation scheduling and are difficult to efficiently allocate tasks while meeting various constraints. Finally, the picking path planning usually adopts a fixed mode, unable to dynamically adjust according to real-time situations and lacking a continuous optimization mechanism based on execution data. Summary of the Invention

[0003] In view of the above problems, this application is proposed.

[0004] Therefore, this application provides a supply chain warehousing management system and method based on AI digitalization, which can solve the problems mentioned in the background art.

[0005] To solve the above technical problems, this application provides the following technical solutions: In a first aspect, the present application provides an AI digitalized supply chain warehousing management system, including: a data processing unit for collecting supply chain system data through a data processing module and constructing an associated data network including goods information, storage location information, and order information based on the supply chain system data; a storage location allocation unit for executing a storage location allocation algorithm according to the associated data network, the storage location allocation algorithm extracting an association degree score of goods based on the goods information, analyzing the collaborative outbound frequency of historical orders according to the order information, and calculating an affinity matrix between goods; a storage plan unit for generating a storage location plan based on the affinity matrix; the process of generating the storage location plan includes constructing a mathematical model of the warehouse space; performing clustering analysis on the affinity matrix, dividing goods with affinity values exceeding a first preset threshold into the same goods group; mapping the goods group to the mathematical model of the warehouse space to generate a first storage location plan; setting a multi-objective optimization function, calculating the total picking path length and the storage location occupancy rate index of the first storage location plan; forming a second storage location plan by iteratively adjusting the goods position allocation in the first storage location plan to reduce the total picking path length and improve the storage location occupancy rate index; a job scheduling unit for constructing a job scheduling model according to the storage location plan, optimizing tasks in combination with the constraint conditions in the storage location plan, and outputting a task allocation sequence through the job scheduling model; a picking navigation unit for configuring a picking path planning module based on the task allocation sequence, presenting a picking path and operation instructions on a mobile terminal by using the picking path planning module, collecting execution data, and optimizing the supply chain warehousing management system by analyzing the execution data.

[0006] Preferably, the extracting the association degree score of goods based on the goods information includes: extracting a category identifier, a physical size, and storage requirements in the goods information to generate a goods feature vector, and performing a similarity calculation on the goods feature vector to obtain the association degree score.

[0007] Preferably, the analyzing the collaborative outbound frequency of historical orders according to the order information includes: counting the co-occurrence times of goods pairs in the historical orders within a preset time period, recording the outbound time interval of each group of goods pairs, and calculating the standardized collaborative outbound frequency; the calculating the affinity matrix between goods includes: setting an association degree weight parameter and a frequency weight parameter, and performing a weighted combination on the association degree score and the collaborative outbound frequency to form the affinity matrix.

[0008] Preferably, the mathematical model of the warehouse space abstractly represents the physical warehouse space through storage location coordinates, aisle layout, and area division.

[0009] Preferably, constructing the job scheduling model according to the storage location scheme includes: converting the storage location scheme into a topological graph structure; wherein, the topological graph structure represents storage locations with nodes and feasible walking paths with edges; establishing a resource constraint table; wherein, the resource constraint table includes the number of available pickers, the number of equipment, and time window parameters; calculating the shortest path distance matrix between nodes in the topological graph structure; constructing a task conflict detection matrix to record time overlaps and resource competitions between tasks; generating a mathematical expression of the job scheduling model according to the shortest path distance matrix and the task conflict detection matrix.

[0010] Preferably, optimizing tasks by combining the constraint conditions in the storage location scheme and outputting a task assignment sequence through the job scheduling model includes: receiving order data, and converting the order data into a set of picking tasks; calculating a task similarity matrix according to the goods composition and quantity of each task in the set of picking tasks; performing task merging based on the task similarity matrix, and combining tasks with similarity exceeding a merging threshold into batch tasks; using integer programming methods to perform time slot allocation and resource allocation for the batch tasks to generate an initial scheduling scheme; detecting resource conflict points in the initial scheduling scheme, and eliminating the resource conflict points by adjusting the task order; generating the task assignment sequence; wherein, the task assignment sequence includes the execution sequence number, start time, allocated resources, and picking path information of each task.

[0011] Preferably, configuring a picking path planning module based on the task assignment sequence includes: parsing the task assignment sequence, and extracting the set of picking points and order requirements of each task; constructing a path planning engine; wherein, the path planning engine includes a global path database and a local obstacle avoidance algorithm; querying the global path database according to the set of picking points to obtain a standard path; detecting whether there are temporary obstacles in the standard path; wherein, the temporary obstacles include channel congestion and area maintenance; if a temporary obstacle is detected, then starting the local obstacle avoidance algorithm to generate a detour path; if no temporary obstacle is detected, then keeping the standard path; calculating the final picking path and generating path visualization data for presentation on the mobile terminal.

[0012] Preferably, the collecting of the execution data and the optimization of the supply chain warehousing management system by analyzing the execution data include: deploying data collection points to record the actual execution trajectory, timestamp, and abnormal events of the picking operation; comparing the difference between the actual execution trajectory and the planned path to generate a deviation report; setting up a data analysis pipeline to statistically aggregate the deviation report and identify systematic deviation patterns; according to the systematic deviation patterns, correcting the weight parameters in the affinity matrix to obtain a correction result; updating the mapping relationship of the goods groups in the storage location plan and adjusting the time estimation parameters in the operation scheduling model to obtain an adjustment result; and transmitting back the correction result and the adjustment result to complete the closed-loop optimization of the supply chain warehousing management system.

[0013] In a second aspect, the present application also provides an AI digitalized supply chain warehousing management method, including: collecting supply chain system data through a data processing module and constructing an associated data network based on the supply chain system data; performing a storage location allocation algorithm according to the associated data network, where the storage location allocation algorithm extracts the association degree score of the goods based on the goods information, analyzes the collaborative outbound frequency of historical orders according to the order information, and calculates the affinity matrix between the goods; generating a storage location plan based on the affinity matrix; constructing an operation scheduling model according to the storage location plan, optimizing tasks in combination with the constraint conditions in the storage location plan, and outputting a task allocation sequence through the operation scheduling model; configuring a picking path planning module based on the task allocation sequence, presenting the picking path and operation guidelines on a mobile terminal by using the picking path planning module, collecting execution data, and optimizing the supply chain warehousing management system by analyzing the execution data.

[0014] Preferably, the associated data network includes goods information, warehouse location information, and order information.

[0015] Implementing the present application has the following beneficial effects: The present application provides an AI digitalized supply chain warehousing management system and method. By constructing an associated data network including goods information, warehouse location information, and order information through a data processing unit, it realizes the comprehensive integration and structured representation of supply chain system data; the storage location allocation unit calculates the affinity matrix based on the goods association degree score and the collaborative outbound frequency of historical orders, realizing the precise quantification of complex association relationships between goods; the storage plan unit generates a scientific storage location plan based on the affinity matrix, optimizing the utilization of warehousing space; the operation scheduling unit constructs an operation scheduling model and optimizes tasks in combination with constraint conditions, improving the resource allocation efficiency and the rationality of task execution order; the picking navigation unit provides intuitive operation guidelines on a mobile terminal and collects execution data for closed-loop optimization, realizing the continuous self-improvement of the system, and comprehensively improving the intelligent level and operation efficiency of warehousing management. Description of the Drawings

[0016] To more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following will briefly introduce the drawings required for the description of the embodiments of the present application or related technologies. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other related drawings can also be obtained based on these drawings.

[0017] Figure 1 It is a schematic diagram of the overall structure of a supply chain warehouse management system based on AI digitalization involved in the present application; Figure 2 It is an application environment diagram of a supply chain warehouse management system based on AI digitalization involved in the present application; Figure 3 It is a general flowchart of a supply chain warehouse management method based on AI digitalization involved in the present application. Detailed implementation manners

[0018] In order to make the objectives, technical solutions and advantages of the present application more clear, the following further details the present application in combination with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0019] Supply chain warehouse management based on AI digitalization is widely used in modern logistics systems. For example, it can be applied to various fields such as e-commerce distribution centers, auto parts warehouses, and retail supply chains. When an enterprise needs to optimize the warehouse layout or improve the order processing efficiency, the supply chain warehouse management system based on AI digitalization can provide the optimal warehouse management solution according to historical data.

[0020] In related technologies, traditional warehouse management systems mainly rely on manual experience for goods location allocation and picking planning, and cannot deeply explore the correlation between goods. However, enterprises do have the need to improve warehouse efficiency and reduce operating costs. How to better apply AI technology to supply chain warehouse management optimization is a relatively important issue.

[0021] Aiming at the problem in related technologies that intelligent goods location allocation and operation scheduling cannot be carried out based on the correlation of goods, a supply chain warehouse management system based on AI digitalization is proposed, as Figure 1As shown in the figure, the system includes: a data processing unit 200, configured to collect supply chain system data through a data processing module and construct an associated data network including goods information, warehouse location information, and order information based on the supply chain system data; a goods location allocation unit 300, configured to execute a goods location allocation algorithm according to the associated data network. The goods location allocation algorithm extracts the association degree score of goods based on the goods information, analyzes the collaborative outbound frequency of historical orders according to the order information, and calculates the affinity matrix between goods; a storage plan unit 400, configured to generate a storage location plan based on the affinity matrix; a job scheduling unit 500, configured to construct a job scheduling model according to the storage location plan, optimize tasks in combination with the constraint conditions in the storage location plan, and output a task allocation sequence through the job scheduling model; a picking navigation unit 600, configured to configure a picking path planning module based on the task allocation sequence, present the picking path and operation instructions on a mobile terminal by using the picking path planning module, collect execution data, and optimize the supply chain warehouse management system by analyzing the execution data.

[0022] The supply chain warehouse management system based on AI digitalization provided by the embodiments of this application can be applied to an application environment as Figure 2 shown in the figure. Among them, the management terminal 102 communicates with the server 104 through a network. The data storage system 106 can store the supply chain data that the server 104 needs to process. The data storage system can be integrated on the server 104 or placed on a cloud platform or a distributed storage system. The mobile terminal 108 is connected to the server 104 through a wireless network and is used to receive and display the picking path and operation instructions at the warehouse site. The warehouse site equipment 110 includes shelves, sorting equipment, and automation equipment, etc., and is connected to the server 104 through Internet of Things technology to upload the equipment status and execution data in real time.

[0023] In practical applications, the server 104 runs Figure 1Each functional unit of the system shown realizes the intelligent operation of supply chain warehousing management. Specifically, the data processing unit 200 collects the order data input from the management terminal 102 and the inventory data uploaded by the on-site warehouse equipment 110 through the interaction with the data storage system 106, and constructs a complete associated data network. The location allocation unit 300 analyzes the association between goods based on this network, calculates the affinity matrix, and provides a decision basis for the storage plan unit 400. The storage location plan generated by the storage plan unit 400 is displayed to the warehouse management personnel through the management terminal 102, and at the same time is transmitted to the operation scheduling unit 500 for task optimization. After the task allocation sequence output by the operation scheduling unit 500 is received by the picking navigation unit 600, it is converted into an intuitive operation guide and pushed to the mobile terminal 108 to guide the warehouse operators to perform the picking task. The actual execution data recorded by the mobile terminal 108 is transmitted back to the server 104 for continuous system optimization, forming a complete data closed-loop. In this way, the system realizes the digitalization and adaptive optimization of the entire warehousing management process.

[0024] Among them, the management terminal 102 can be, but is not limited to, various personal computers, laptop computers, smart phones, tablet computers, Internet of Things devices, and industrial control terminals. The industrial control terminal can be an intelligent warehousing management terminal, an intelligent scheduling panel, a warehouse management system (WMS) terminal, an enterprise resource planning (ERP) system terminal, etc. The mobile terminal 108 can be a handheld data terminal (PDT), a smart watch, a smart bracelet, a head-mounted device, etc. The head-mounted device can be a virtual reality (VR) device, an augmented reality (AR) device, a smart glasses, etc., and can be used for warehouse pickers to receive operation guides and navigation information. The server 104 can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services.

[0025] Furthermore, the data processing unit is used to collect supply chain system data through the data processing module and construct an associated data network including goods information, warehouse location information, and order information based on the supply chain system data.

[0026] Specifically, the data processing module can obtain data from various system sources in multiple ways such as API interface call, database connection, file import, or message queue.

[0027] In an alternative embodiment, the data processing module may process data using an ETL (Extract-Transform-Load) process. First, raw data is extracted from source systems such as enterprise resource planning (ERP) systems, warehouse management systems (WMS), transportation management systems (TMS), supplier management systems, and sales order systems through a data extractor; then, the extracted data is cleaned, format-converted, standardized, and deduplicated through a data converter; finally, the processed data is loaded into the data storage layer of the system through a data loader. The ETL process can be set to execute periodically, such as once per hour, day, or week, to ensure timely data updates.

[0028] In another alternative embodiment, the data processing module may adopt a real-time data collection method. By deploying data collection agents in source systems to monitor data change events, when data changes are detected, the changed data is immediately pushed into a message queue, and the data processing module consumes the data from the message queue to achieve near-real-time data collection. This method is applicable to scenarios with high requirements for data timeliness, such as warehouse management in the fast-moving consumer goods industry.

[0029] The data processing unit organizes the collected supply chain system data into a graph data structure, where nodes represent entities (such as goods, storage locations, orders, etc.), and edges represent the relationships between entities. The process of constructing the associated data network includes three steps: entity extraction, relationship identification, and network construction.

[0030] In an alternative embodiment, the goods information may include attributes such as goods ID, name, category, sub-category, brand, specification, physical dimensions (length, width, height), weight, volume, value, shelf life, storage condition requirements (temperature range, humidity range, light requirements, etc.), hazard level, supplier information, and goods status. Goods attributes can be customized according to industry characteristics and management needs. For example, in the pharmaceutical industry, specific attributes such as batch number, drug approval number, and expiration date may also be included.

[0031] In another alternative embodiment, the storage location information may include attributes such as storage location ID, warehouse area, shelf number, layer number, position number, coordinate information (X, Y, Z coordinates), type, capacity (maximum storage capacity, maximum load-bearing, etc.), occupancy status, applicable goods types, and special requirements. The storage location coordinates are represented by a three-dimensional coordinate system, where the X-axis and Y-axis represent horizontal positions, and the Z-axis represents vertical height, for accurately positioning each storage location in the warehouse.

[0032] In yet another alternative embodiment, the order information may include attributes such as order ID, customer information, creation time, expected delivery time, order status, priority, order details (including goods ID, quantity), payment information, and logistics information. The system can also record the historical status changes of orders for order tracking and analysis.

[0033] There are various types of relationships between entities in an associated data network, mainly including: 1. Goods - storage location relationship: indicating which storage location the goods are stored in, including attributes such as storage time and quantity; 2. Goods - order relationship: indicating which order the goods belong to, including attributes such as ordered quantity; 3. Goods - goods relationship: indicating the correlation between goods, such as goods that are often ordered together, components belonging to the same set, etc.; 4. Storage location - storage location relationship: indicating the physical adjacency relationship between storage locations, which is used for path planning and optimization.

[0034] In an alternative embodiment, the associated data network is stored using an attribute graph model and implemented using a graph database (such as Neo4j or JanusGraph) to achieve efficient relationship querying and traversal. Each node and edge can have multiple attributes, and the query languages of the graph database (such as Cypher or Gremlin) can conveniently perform complex relationship queries and path analysis. For example, through a query, all goods that are often shipped out together with a specific good can be found, or the shortest path from one storage location to another can be calculated.

[0035] In another alternative embodiment, the associated data network can be stored using a triple model (subject - predicate - object) and implemented using an RDF (Resource Description Framework) database. Triple - formatted data is more suitable for semantic querying and reasoning, and supports the use of the SPARQL query language for complex semantic queries. For example, queries such as "all dangerous goods stored in area A and weighing more than 50 kg" or "all goods that require refrigeration and are often ordered together with product X" can be made.

[0036] The data processing unit can also implement an incremental update mechanism to only process the changed data, reducing the system load and data processing time. The specific implementation method is to record the last update timestamp of the data in the source system, and only extract the records with a timestamp greater than the last processed time during each ETL process, thereby achieving incremental update.

[0037] Example 1: In an e - commerce warehousing scenario, extract the new order data in the last 24 hours from the order system, including information such as order number, customer information, order time, product list, and quantity; extract the current goods inventory information from the inventory management system, including goods ID, name, specification, inventory quantity, storage location, etc.; extract the storage location information from the warehouse management system, including storage location ID, area, shelf number, current occupancy status, etc. After the data processing module cleans and standardizes this data, an associated data network is constructed. In this network, information such as "all pending orders containing a specific product" or "all idle storage locations in a specific area" can be quickly queried.

[0038] Example 2: In the pharmaceutical warehousing scenario, the data processing unit needs to pay special attention to the batch and expiration date information of drugs. It extracts the basic drug information from the ERP system, the batch and expiration date information from the quality management system, and the storage location information from the warehouse management system. In the constructed associated data network, each drug node, in addition to its basic attributes, also includes specific attributes such as batch, production date, and expiration date. At the same time, the association relationships between drugs and storage locations, drugs and suppliers, and drugs and orders are established. In this way, drugs approaching expiration can be quickly identified, or the storage locations and inventory status of specific batches of drugs can be queried.

[0039] Example 3: In the automotive parts warehousing scenario, the data processing unit extracts part information from the parts management system, including part number, name, applicable vehicle models, specifications, etc.; extracts storage location information from the warehouse management system; and extracts production order information from the production planning system. In the constructed associated data network, "assembly relationships" are established between part nodes, indicating which parts need to be assembled together into a certain component. Through this relationship, the storage locations of parts can be optimized, and parts that are often used together can be placed in close proximity to improve the picking efficiency.

[0040] Through the above implementation methods, the data processing unit can effectively collect and integrate supply chain data from different source systems, construct a complete associated data network, and provide a data basis for subsequent location allocation, storage plan generation, operation scheduling, and picking navigation.

[0041] Furthermore, the location allocation unit is used to execute a location allocation algorithm based on the associated data network. The location allocation algorithm extracts the association degree score of goods based on the goods information, analyzes the collaborative outbound frequency of historical orders according to the order information, and calculates the affinity matrix between goods.

[0042] The location allocation algorithm first extracts the association degree score of goods based on the goods information. Specifically, extracting the association degree score of goods based on the goods information includes: extracting the category identifier, physical size, and storage requirements in the goods information, generating a goods feature vector, and performing a similarity calculation on the goods feature vector to obtain the association degree score.

[0043] In an alternative embodiment, the construction of the goods feature vector is to map the multi-dimensional attributes of goods into a numerical vector. The system extracts the category identifier, physical size, and storage requirement information of goods from the associated data network and constructs a feature vector. For each good , its feature vector can be expressed as: ; where represents the feature vector of the good ; Indicates the th category identification feature of the goods (such as can represent the main category, can represent the sub-category); Indicates the th physical size feature of the goods (such as can represent the length, can represent the width); Indicates the th storage requirement feature of the goods (such as can represent the temperature requirement, can represent the humidity requirement).

[0044] It should be noted that for discrete features such as category identification, one-hot encoding is used to convert them into numerical values; for continuous features such as physical size, standardization processing is used to map the numerical values to the interval [0, 1]. By constructing a feature vector in a unified format, standardized comparison of different goods can be achieved, laying a foundation for subsequent similarity calculation.

[0045] For the similarity calculation of feature vectors, in an alternative embodiment, cosine similarity is used to calculate the similarity between two goods feature vectors: ; wherein, represents the similarity between goods and goods ; and respectively represent the feature vectors of goods and goods ; represents the th component of vector represents the dimension of the feature vector; represents the dot product of vectors; represents the norm of the vector. The value range of cosine similarity is , and the closer the value is to 1, the more similar the two goods are.

[0046] It should be noted that by calculating the cosine value of the angle between feature vectors, the similarity degree of goods in the multi-dimensional feature space can be effectively measured, especially suitable for dealing with high-dimensional sparse features.

[0047] In another alternative embodiment, considering that different features have different influence degrees on similarity, the system uses weighted Euclidean distance to calculate the distance between feature vectors, and then converts the distance into similarity: ; ; Among them, represents the weighted Euclidean distance between the eigenvectors of goods and goods . represents the weight of the th feature, satisfying ; represents the converted similarity, and its value range is .

[0048] It should be noted that by considering the relative importance of features and setting different weights, the business requirements can be more accurately reflected. For example, for the storage of dangerous goods, the weight of the storage requirement features can be increased; for space optimization, the weight of the physical size features can be increased.

[0049] The calculated similarity value is directly used as the correlation score between goods , that is: That is: ; Among them, represents the correlation score between goods and goods , and the value range is

[0050] The larger the value, the higher the correlation between the two goods. The correlation score quantifies the similarity between different goods and provides the basic data for the subsequent calculation of the affinity matrix.

[0051] Next, the goods location allocation algorithm analyzes the collaborative outbound frequency of historical orders according to the order information. Specifically, analyzing the collaborative outbound frequency of historical orders according to the order information includes: counting the co-occurrence times of goods pairs in historical orders within a preset time period, recording the outbound time interval of each group of goods pairs, and calculating the standardized collaborative outbound frequency. In an optional embodiment, first set a preset time period such as the last 30 days, the last 90 days, or the last year), and extract all order data within this time period from the associated data network. For each pair of goods , the system counts the number of times they co-occur in the same order, denoted as ; at the same time, counts the number of times each appears in all orders, denoted as and ; Among them, represents goods and goods of the original co-warehouse-out frequency; indicates the goods and goods appear together in the same order; and respectively indicate the goods and goods appearances in all orders. It should be noted that the formula calculates a form of conditional probability, indicating the probability that another good is also warehoused out under the condition that the good with fewer appearances is warehoused out. This calculation method can handle the situation of unbalanced appearance frequencies of goods and avoid the co-occurrence relationship of low-frequency goods being masked by high-frequency goods.

[0052] In another alternative embodiment, not only the co-occurrence times are considered, but also the time interval between warehouse-outs is considered. For each pair of goods that appear in the same order , record their actual warehouse-out times and calculate the time interval between warehouse-outs , where represents the th co-occurrence. Then introduce a time decay factor and calculate the time-weighted co-warehouse-out frequency: ; where, represents the co-warehouse-out frequency considering the time interval; is the time decay coefficient, controlling the influence degree of the time interval on the co-occurrence; is the time decay factor, and the shorter the time interval, the closer the factor value is to 1. By introducing the time factor, this method can more accurately characterize the co-occurrence of goods at the operational level, which is beneficial to optimizing the picking path and reducing the operation time.

[0053] To make the co-warehouse-out frequency comparable, the original frequency is standardized: ; where, represents the standardized co-warehouse-out frequency; and are respectively the minimum and maximum original co-warehouse-out frequencies among all pairs of goods. The standardization process maps the co-warehouse-out frequency to the interval [0,1], which is convenient for unified calculation with the correlation score.

[0054] Finally, the location allocation algorithm calculates the affinity matrix between goods. Specifically, calculating the affinity matrix between goods includes: setting the correlation weight parameter and the frequency weight parameter, and performing a weighted combination of the correlation score and the co-warehouse-out frequency to form the affinity matrix.

[0055] In an alternative embodiment, the correlation weight parameter and the frequency weight parameter are set, and a linear weighted combination of the correlation score and the collaborative outbound frequency is performed to form an affinity matrix: ; wherein, represents the affinity value between goods and goods ; represents the correlation score between goods and goods ; represents the normalized collaborative outbound frequency between goods and goods ; and are the correlation weight parameter and the frequency weight parameter respectively, satisfying and . By adjusting the weight parameters, the system can balance the influence of goods characteristic similarity and collaborative outbound patterns according to business requirements, and implement a flexible goods location allocation strategy.

[0056] In another alternative embodiment, considering that there may be a non-linear relationship between the correlation and the collaborative frequency, the system calculates the affinity value by geometric weighting: ; wherein, the parameter meanings are the same as above. This geometric weighting method requires both indicators to be relatively high for the affinity value to be high, which emphasizes the comprehensiveness of the two indicators more, and avoids the situation where the affinity value is too high due to a single high indicator.

[0057] It should be noted that the affinity matrix is a symmetric matrix, and each element represents the affinity degree between goods and goods . The value range is [0, 1], and the larger the value, the higher the affinity degree between the two goods, and the more suitable they are to be placed in adjacent positions. The affinity matrix provides a mathematical basis for subsequent storage location optimization. By reasonably setting the relative positions between goods, the picking path length can be reduced and the picking efficiency can be improved.

[0058] Example 1: In the scenario of an e-commerce warehouse, the feature vectors of smartphones and mobile phone cases are extracted from the goods information. The feature vector of the smartphone is , indicating that it belongs to the electronic product category, has a medium volume and weight, and has no special storage requirements; the feature vector of the mobile phone case is , indicating that it also belongs to the electronic product category and has a slightly smaller volume and weight. The calculated correlation score between them 。

[0059] Analyze the order data for the past 60 days and find that smartphones appear in 2,000 orders, phone cases appear in 1,800 orders, and they both appear in 1,700 orders. Calculate the original co-warehousing frequency , the standardized co-warehousing frequency 。

[0060] Set the correlation weight parameter and the frequency weight parameter , and calculate the affinity value: ; This high affinity value indicates that smartphones and phone cases should be placed in close proximity for simultaneous picking, improving order processing efficiency.

[0061] Example 2: In the scenario of an auto parts warehouse, analyze the relationship between engine blocks and tires. Calculate their correlation score based on the feature vector (since they are quite different in terms of category, size, and storage requirements); analyze the standardized co-warehousing frequency from historical repair orders (they are rarely repaired together). Set and , and calculate the affinity value: ; This low affinity value indicates that engine blocks and tires can be placed in relatively distant locations to save warehouse space and meet their respective storage requirements.

[0062] Through the above-mentioned location allocation algorithm, the correlation and coordination between goods can be objectively quantified, a scientific affinity matrix can be constructed, providing data support for the optimization of subsequent storage location plans, thereby realizing intelligent warehousing space planning and improving warehouse operation efficiency.

[0063] Furthermore, the storage plan unit is used to generate a storage location plan based on the affinity matrix.

[0064] Specifically, the process of generating the storage location plan includes generating a first storage location plan and optimizing it according to the first storage location plan to obtain a second storage location plan.

[0065] In an alternative embodiment, first construct a mathematical model of the warehouse space. The mathematical model of the warehouse space abstractly represents the physical warehouse space through location coordinates, aisle layout, and area division. The specific implementation process is as follows: The mathematical model of the warehouse space uses a three-dimensional coordinate system to represent the location of each storage location. Each storage location is defined as a coordinate point (x, y, z) and also includes capacity parameters (such as maximum load-bearing capacity, available volume). The channel layout is represented by a set of edges connecting nodes. Each edge represents a passable path, recording its length and passing conditions. The area division divides the warehouse into multiple functional areas, and each area contains a set of storage locations with similar storage conditions.

[0066] It should be noted that in addition to having three-dimensional spatial position information, each storage location also has limit parameters describing its physical storage capacity, that is, capacity parameters. The specific meanings are as follows: Maximum load-bearing capacity: It refers to the maximum weight limit that the storage location can safely bear. For example, the maximum load-bearing capacity of a storage location on the upper layer of a certain shelf may be 50 kg, while the maximum load-bearing capacity of a ground storage location may be 500 kg. This parameter ensures that the total weight of the goods allocated to this storage location does not exceed the bearing capacity of the storage location, avoiding structural safety problems.

[0067] Available volume: It refers to the maximum space volume that the storage location can accommodate goods. For example, the available volume of a standard storage location may be 1.2 m × 0.8 m × 2.0 m = 1.92 cubic meters. This parameter ensures that the allocated goods can be completely placed in this storage location in terms of physical dimensions, avoiding problems of insufficient space.

[0068] The role of these capacity parameters is to serve as constraint conditions during storage location allocation to ensure that the system does not allocate overweight or oversized goods to storage locations with insufficient bearing capacity; helps calculate the actual utilization rate of storage locations and evaluate the space use efficiency; provides a calculation basis for the storage location occupancy rate index in the multi-objective optimization function.

[0069] For example, when the system needs to allocate storage locations for a batch of goods with a weight of 80 kg and a volume of 0.5 cubic meters, only storage locations with a maximum load-bearing capacity ≥ 80 kg and an available volume ≥ 0.5 cubic meters will be considered as candidate locations. This can avoid physically infeasible allocation schemes and ensure the actual feasibility of the storage scheme.

[0070] In another alternative embodiment, in addition to the basic spatial location information, the warehouse space mathematical model also quantifies dynamic operation parameters. Specifically, for a stereoscopic warehouse, the picking time differences of storage locations at different heights are recorded through a time coefficient matrix. For example, the time coefficient of the ground floor storage location is 1.0, while that of the top floor storage location may be 1.8, indicating that the picking time is 1.8 times that of the ground floor. For a partitioned warehouse, a regional transfer matrix is established to record the time or distance cost required to move from one area to another. For example, the transfer cost between adjacent areas may be 10 seconds, while that between long-distance areas may be 60 seconds. These precise dynamic parameters enable the warehouse space model to accurately reflect the actual operation conditions, providing precise constraint conditions and an evaluation basis for subsequent path planning and optimization.

[0071] Next, the storage scheme unit performs a clustering analysis on the affinity matrix and divides the goods with affinity values exceeding the first preset threshold into the same goods group.

[0072] It should be noted that the first preset threshold is a critical value for measuring the association strength between goods and is determined based on the numerical distribution characteristics of the goods affinity matrix and business requirements. In practical applications, this threshold is usually determined through the following steps: First, perform a statistical analysis on the affinity matrix to obtain the distribution of affinity values (such as mean, standard deviation, quantiles, etc.); then, based on the business characteristics, select an appropriate percentile as the initial threshold; finally, through iterative testing of the clustering results generated by different thresholds, select the value that can best balance the "association within the goods group" and the "quantity and size of the goods groups". For most warehousing scenarios, the first preset threshold is usually set between 0.6 and 0.8. When the first preset threshold is set to 0.6, the formed goods groups are fewer but larger in scale, and the association between goods within the group is relatively low; when the first preset threshold is set to 0.8, the number of formed goods groups is larger but smaller in scale, and the association between goods within the group is higher.

[0073] The storage scheme unit uses a hierarchical clustering algorithm to process the affinity matrix. In the specific implementation of the warehousing scenario: First, take individual goods as the initial groups; then, calculate the average affinity value between all pairs of groups; next, merge the two groups with the highest affinity value and exceeding the first preset threshold each time; finally, when there is no pair of groups with an affinity value higher than the first preset threshold, the clustering process ends. Through this method, the storage scheme unit does not need to preset the number of groups, but naturally forms groups according to the actual affinity relationship, avoiding the limitation of the need to specify the number of groups in traditional clustering methods.

[0074] Next, the storage scheme unit maps the goods groups to the warehouse space mathematical model to generate the first storage location scheme.

[0075] In an alternative embodiment, the storage planning unit calculates the comprehensive matching degree of each goods group with each warehouse area, while evaluating multiple key dimensions: storage condition matching degree (such as calculating the temperature requirement deviation value, humidity adaptability score, etc.), goods characteristic matching degree (such as the ratio of weight to shelf load-bearing capacity, the ratio of volume to available space, etc.), operational convenience (such as the inverse relationship between access frequency and distance to the entrance and exit), and safety requirements (such as the risk level and isolation requirement score). The storage planning unit combines these dimensional indicators into a single matching degree score through weight coefficients to find the most suitable area for each goods group.

[0076] In another alternative embodiment, the storage planning unit uses the Hungarian algorithm to solve the optimal allocation problem of goods groups to areas. This algorithm models the allocation of goods groups to areas as a maximum weight matching problem in a bipartite graph, and finds the allocation plan with the highest total matching degree through matrix operations. When constructing the matching matrix, the storage planning unit considers the area capacity constraint and the priority of goods groups, and through preprocessing, ensures that important goods groups are preferentially allocated the most suitable areas without violating the capacity limit.

[0077] Inside the area, the storage planning unit further allocates the goods groups to specific storage locations. In an alternative embodiment, the storage planning unit implements a "continuous space first" strategy: first, accurately calculate the space requirements of each goods group (considering the quantity, volume, and special requirements of the goods), and then apply a space search algorithm in the target area to find the largest continuous storage location block that meets the requirements. If a large enough continuous space is found, it is directly allocated; if the continuous space is insufficient, the storage planning unit will use a block allocation algorithm to find several larger storage location blocks while minimizing the distance between these blocks and reducing the degree of dispersion.

[0078] In another alternative embodiment, the storage planning unit analyzes the affinity relationship hierarchy within the goods group, and identifies the core goods subset and peripheral goods by constructing an affinity heat map. The storage planning unit allocates the core goods subset with the highest affinity to adjacent positions, and the goods with the second highest affinity to nearby positions, forming a concentric circle distribution pattern centered on high affinity. This distribution strategy realizes the optimal expression of affinity in a limited space and improves the picking efficiency.

[0079] Through the above steps of goods group mapping and storage location allocation, the storage planning unit successfully generates the first storage location plan, which specifies the clear storage location coordinates for each good.

[0080] After obtaining the first storage location plan, the storage planning unit sets a multi-objective optimization function and calculates the total picking path length and storage location occupancy rate indicators of the first storage location plan.

[0081] In an alternative embodiment, the storage scheme unit constructs a comprehensive multi-objective evaluation system. The total picking path length is calculated through order simulation: the storage scheme unit extracts representative order samples (such as the recent 1000 orders) from the historical order database, executes a path planning algorithm (such as the Dijkstra algorithm) for each order to calculate the shortest picking path, and then assigns weight coefficients according to the order frequency and priority to calculate the weighted total path length. The index of storage location occupancy rate is evaluated through multiple dimensions: on the one hand, calculate the ratio of the number of allocated storage locations to the total number of storage locations, and on the other hand, consider the volume utilization efficiency and calculate the ratio of the used space volume to the total available volume.

[0082] In another alternative embodiment, the storage scheme unit constructs an objective function using a weighted combination method: F = w1 × normalized total picking path length + w2 × (1 - storage location occupancy rate). Among them, the normalized total picking path length is a dimensionless index obtained by dividing the original path length by the theoretical maximum path length; w1 and w2 are adjustable weight parameters that satisfy w1 + w2 = 1 and the value range is between [0, 1]. This design of the multi-objective function enables warehouse managers to flexibly adjust the optimization objectives according to the actual business focus (efficiency priority or space utilization priority), making the storage scheme more adaptable to the actual needs of different types of warehouses.

[0083] Finally, the storage scheme unit forms the second storage location scheme by iteratively adjusting the goods location allocation in the first storage location scheme, reducing the total picking path length and improving the storage location occupancy rate index.

[0084] In an alternative embodiment, the storage scheme unit uses the simulated annealing algorithm for optimization. The initial temperature parameter T is set to a sufficiently high value (such as 100) to allow accepting worse solutions at the initial stage of optimization to jump out of the local optimum; the current solution S is initialized as the first storage location scheme. During the iteration process, the storage scheme unit generates a new solution S' within the neighborhood of the current solution. The neighborhood operations include: goods pair exchange (randomly select two goods to exchange positions), goods movement (move a single good to a new position), and goods group rotation (keep the relative positions within the group unchanged and move the entire goods group to a new position), etc. Calculate the change in the objective function ΔF. If the change is negative (the new solution is better), then accept the new solution; if the change is positive (the new solution is worse), then accept the new solution with a probability of exp(-ΔF / T). After each round of iteration, the temperature parameter is decreased by a coefficient α (such as 0.95): T = α × T, until the temperature drops to the preset lower limit or there is no improvement after consecutive multiple iterations. The simulated annealing algorithm controls the search strategy through the "temperature" parameter, which can effectively balance global exploration and local optimization and avoid falling into the local optimum solution.

[0085] In another optional embodiment, the storage scheme unit implements a genetic algorithm optimization framework. The framework first initializes the population, including the first storage location scheme and multiple random mutation schemes. After calculating the fitness (the inverse of the objective function) of each scheme, excellent individuals are selected as parents through tournament selection or roulette method. The storage scheme unit implements dedicated crossover operators, such as region-based crossover (the parent scheme exchanges cargo allocation between randomly selected regions) and cargo group-based crossover (the parent scheme exchanges the location allocation of a specific cargo group). Mutation operations include random cargo location changes, internal reorganization of cargo groups, etc. The new generation population consists of elite individuals (a small number of individuals with the highest fitness are retained) and offspring individuals, and iterative evolution is performed for multiple generations until the termination condition is met. Genetic algorithms can effectively handle large-scale complex cargo allocation problems by simulating natural selection and genetic processes.

[0086] In another optional embodiment, the storage solution unit constructs a "constraint-aware" operation framework: all neighborhood operations and genetic variations are subject to the physical constraints and business rules of the warehouse. In specific implementation, the storage solution unit maintains a constraint checking function library, including storage condition compatibility checks (such as temperature control requirements), physical restriction checks (such as weight and carrying capacity matching), safety specification checks (such as hazardous material isolation), etc. Before generating a new solution each time, the storage solution unit pre-verifies whether the operation violates the constraints; if so, the operation is automatically adjusted or abandoned to ensure that each candidate solution generated is feasible. This constraint-aware mechanism significantly improves the optimization efficiency and avoids the additional computational cost of processing infeasible solutions.

[0087] After the above optimization process, the storage plan unit finally generates a second storage location plan. Under the premise of meeting all constraints, this plan reduces the total length of the picking path and increases the cargo space occupancy rate, thereby achieving dual optimization of warehouse space and operational efficiency.

[0088] Example 1: In an e-commerce distribution center, the warehouse is divided into three main areas: normal temperature area, refrigerated area, and valuables area, with a total of 5,000 cargo locations. First, build an accurate three-dimensional warehouse model to record the location coordinates and area of each cargo location. For 1,000 SKUs of goods, the affinity matrix is calculated based on the aforementioned cargo location allocation algorithm. Set the first preset threshold to 0.75 (this value is determined by analyzing the value distribution of the affinity matrix and combining the operational characteristics of the e-commerce distribution center, which can ensure high correlation between goods in the group while avoiding the generation of too many small-scale goods groups), and divide these goods into 120 goods groups through a hierarchical clustering algorithm to ensure that the affinity value of the goods in each group exceeds the first preset threshold.

[0089] The storage plan unit precisely evaluates the multi-dimensional matching degree of each goods group with each area and uses the Hungarian algorithm for allocation. Specifically, the group of electronic products with high-frequency orders is allocated to the normal temperature area near the exit, the groups of food and drugs that need refrigeration are allocated to the refrigerated area, and the groups of precious jewels and limited-edition goods are allocated to the valuable goods area with a high security level. Inside the area, the "continuous space first" strategy is adopted to allocate the goods in the same group to adjacent storage locations, forming a clear first storage location plan.

[0090] The total picking path length of the first storage location plan is calculated to be 15,000 meters (simulated based on 1,000 recent historical orders), and the storage location occupancy rate is 75%. The weight parameters w1 = 0.6 and w2 = 0.4 are set (determined according to the business characteristics that this distribution center pays more attention to picking efficiency), and the multi-objective optimization function value is calculated to be 0.55.

[0091] The storage plan unit uses the simulated annealing algorithm to optimize the first storage location plan, with an initial temperature of 100 and a cooling coefficient of 0.95. Particular attention is paid to the commodity combinations that are often purchased together in high-frequency orders. By iteratively adjusting their relative positions, the picking path of common orders is significantly shortened. After 1,000 iterations, the second storage location plan is obtained, the total picking path length is reduced to 11,200 meters (a reduction of 25.3%), and the storage location occupancy rate is increased to 82%. The optimized function value drops to 0.408, which is 26.7% better than the initial plan.

[0092] Example 2: In an auto parts distribution center, the warehouse is divided into a heavy-duty area, a standard area, and a small parts area, with a total of 3,000 storage locations. The storage plan unit establishes an affinity matrix for 4,000 auto parts, and sets the first preset threshold to 0.8 (this value is determined to be a relatively high threshold after statistical analysis. The reason is that there is a strong technical correlation in the collaborative use of auto parts in vehicle maintenance, and it is necessary to ensure that the parts in the same group have very close functional connections). Through the spectral clustering algorithm, the auto parts are divided into 80 goods groups. Specifically, the parts that are often used together for engine maintenance (such as cylinder liners, pistons, connecting rods, etc.) are divided into the same group; the parts that are often used for chassis maintenance (such as shock absorbers, suspension springs, etc.) are divided into the same group.

[0093] According to the multi-dimensional matching degree evaluation, the storage plan unit allocates the engine parts group to the heavy-duty area equipped with heavy-duty handling equipment, the chassis parts group to the standard area, and the small parts such as dashboard components to the small parts area, forming a clear first storage location plan. Inside the area, the parts are arranged in the order of maintenance, so that the parts that need to be used first during maintenance are placed in more accessible positions.

[0094] The calculated average picking path length of the first storage location plan is 280 meters / order, and the occupancy rate of storage locations is 68%. The weight parameters are set as w1 = 0.5 and w2 = 0.5 (the distribution center attaches equal importance to efficiency and space utilization), and the initial value of the optimization function is 0.51. The storage plan unit is optimized using the genetic algorithm, starting from different initial solutions simultaneously, with a population size of 50, a crossover probability of 0.8, and a mutation probability of 0.1. The component layout for high-frequency maintenance orders is preferentially optimized, and the second storage location plan is obtained after 200 generations of evolution. In the optimized plan, different components of the same vehicle model are placed more concentratedly, the average picking path is shortened to 210 meters / order (a 25% reduction), and the occupancy rate of storage locations is increased to 76%. The value of the optimized function drops to 0.3825, which is a 25% improvement compared to the initial plan.

[0095] Through the above embodiments, the storage plan unit generates a scientific and reasonable storage location plan based on the affinity matrix, and continuously improves the quality of the plan through the multi-objective optimization algorithm, realizing the balanced optimization of the picking path length and space utilization rate, providing strong support for warehouse operation.

[0096] Furthermore, the job scheduling unit is used to construct a job scheduling model according to the storage location plan, optimize tasks in combination with the constraint conditions in the storage location plan, and output a task assignment sequence through the job scheduling model.

[0097] Specifically, constructing a job scheduling model according to the storage location plan includes: First, convert the storage location plan into a topological graph structure; among them, the topological graph structure represents storage locations with nodes and feasible walking paths with edges. The topological graph structure is a mathematical abstraction of the physical layout of the warehouse. Each node represents a storage location or intersection point, containing coordinate attributes and capacity information; each edge represents the channel connection between two nodes, containing attributes such as distance and width. This graph structure representation method can accurately describe the spatial relationship of the warehouse and provide a basis for subsequent path planning.

[0098] Then, establish a resource constraint table; among them, the resource constraint table includes the number of available pickers, the number of equipment, and time window parameters. The resource constraint table records all available resources and their limiting conditions for performing picking tasks. Human resources include the skill levels and work time arrangements of each picker; equipment resources include the types, quantities, and performance parameters of equipment such as forklifts, picking trucks, and AGVs; time window parameters define time constraints such as work shifts, rest times, and order delivery deadlines. These information together constitute the boundary conditions for task scheduling.

[0099] Next, calculate the shortest path distance matrix among the nodes in the topological graph structure. The distance matrix records the shortest travel distances between any two nodes in the warehouse and serves as the basis for subsequent path planning and operation time estimation. The calculation process usually uses the Dijkstra algorithm or the Floyd-Warshall algorithm. For large warehouses, a partition calculation strategy can be adopted to improve efficiency. The final result is a complete two-dimensional matrix, where each element represents the shortest path distance from one node to another node.

[0100] Then, construct a task conflict detection matrix to record the time overlap and resource competition among tasks. The conflict detection matrix is used to identify and quantify potential conflicts among tasks, mainly considering two dimensions: time overlap (whether the execution times of two tasks have an intersection) and resource competition (whether the same personnel or equipment are required). For each pair of tasks that may be executed in parallel, calculate the degree of conflict between them and record it in the matrix, providing a basis for subsequent conflict elimination.

[0101] Finally, based on the shortest path distance matrix and the task conflict detection matrix, generate the mathematical expression of the job scheduling model. The job scheduling model describes the constraint conditions and optimization objectives of task allocation and execution in a mathematical form, usually expressed as a mixed integer programming problem. The model includes decision variables (indicating that a task is assigned to a specific resource and starts at a specific time), an objective function (such as minimizing the total completion time), and a series of constraint conditions (such as resource capacity constraints, time window constraints, task sequence constraints, and conflict avoidance constraints).

[0102] Furthermore, combine the constraint conditions in the storage location plan for task optimization. The task assignment sequence output by the job scheduling model includes: First, receive order data and convert the order data into a set of picking tasks. Each order is parsed into one or more picking tasks, and each picking task contains information such as the list of goods to be picked, quantity, priority, and deadline. For large orders, the system may decompose them into multiple subtasks for parallel processing and resource allocation.

[0103] Then, according to the goods composition and quantity of each task in the set of picking tasks, calculate the task similarity matrix. The task similarity matrix measures the similarity degree of different tasks in terms of goods composition, usually calculated using the Jaccard similarity coefficient: the number of common goods types between two tasks divided by the number of goods type unions. Tasks with high similarity mean that they need to pick similar goods, and merging their execution can reduce duplicate paths and improve picking efficiency.

[0104] Next, task merging is performed based on the task similarity matrix, and tasks with similarity exceeding the merging threshold are combined into batch tasks. The merging threshold is the critical value for determining whether two tasks are similar enough to be merged. The merging threshold is usually determined according to the warehouse type and business requirements, and its value range is usually between 0.3 and 0.7. A too-high merging threshold will result in few tasks being merged, and the advantages of batch processing cannot be fully utilized; a too-low merging threshold may lead to too many unrelated tasks being merged, increasing the picking complexity and error rate. The task merging process usually adopts a greedy algorithm: starting from the pair of tasks with the highest similarity, merging is carried out until there are no eligible task pairs, and finally a set of batch tasks is formed.

[0105] Then, the integer programming method is used to perform time slot allocation and resource allocation for the batch tasks to generate an initial scheduling plan. The decision variables of the integer programming model represent that tasks are assigned to specific resources and start at specific times. The goal is to minimize the total completion time or maximize the resource utilization rate while satisfying various constraints. The model is solved by a solver (such as CPLEX or Gurobi) to obtain the specific start time, execution duration, and allocated resources for each batch task.

[0106] Next, the resource conflict points in the initial scheduling plan are detected, and the resource conflict points are eliminated by adjusting the task order. The conflict detection process scans the resource usage at each time point along the time axis, and marks the time period when the resource demand exceeds the available amount as a conflict point. For each conflict point, calculate the cost of adjusting the relevant tasks (such as the penalty for causing delays), and select the task with the minimum cost for time adjustment until all conflicts are resolved.

[0107] Finally, a task assignment sequence is generated; among them, the task assignment sequence contains the execution sequence number, start time, allocated resources, and picking path information of each task. The task assignment sequence is the final output of job scheduling, providing detailed execution guidance for warehouse operators and equipment. Each task entry in the sequence contains comprehensive information: task identification and execution sequence, planned start and end times, allocated personnel and equipment resources, a list of picking points arranged in sequence, and goods details. This detailed task assignment sequence enables the picking operation to be carried out efficiently and orderly.

[0108] Through the above steps, the job scheduling unit realizes the efficient conversion from the storage location plan to the specific execution plan, solves the multi-task scheduling problem in a complex warehousing environment, and provides scientific decision-making support for the daily operation of the warehouse.

[0109] Exemplarily, in an e-commerce distribution center, the operation scheduling unit processes 200 customer orders. First, the storage location plan is converted into a topological graph containing 8,000 nodes, and the complete shortest path distance matrix is calculated. The resource constraint table shows that there are 25 pickers and 18 electric picker trucks in the distribution center. The system calculates the similarity between all tasks, sets the merging threshold to 0.5 (this value is determined based on the analysis of the historical operation data of this warehouse and can achieve a balance between picking efficiency and accuracy), and merges 200 original orders into 120 batch tasks. After using the integer programming method for time slot and resource allocation, 15 resource conflicts are detected, mainly concentrated in the peak period from 10:00 am to 11:30 am. By adjusting the time of non-urgent tasks, all conflicts are successfully eliminated. The finally generated task assignment sequence precisely specifies the execution order, time, resources, and paths of each batch task, ensuring that all orders can be processed before the deadline.

[0110] Furthermore, the picking navigation unit is used to configure the picking path planning module based on the task assignment sequence, present the picking path and operation guidelines on the mobile terminal by using the picking path planning module, collect execution data, and optimize the supply chain warehouse management system by analyzing the execution data.

[0111] Specifically, configuring the picking path planning module based on the task assignment sequence includes: First, parse the task assignment sequence and extract the picking point set and order requirements for each task. The picking navigation unit receives the task assignment sequence output by the operation scheduling unit, parses each task, and extracts the picking point set (i.e., the list of storage location coordinates to be visited) and the picking order requirements (such as priority constraints or mandatory order constraints). These information are the basic input data for path planning.

[0112] Then, construct a path planning engine; among them, the path planning engine includes a global path database and a local obstacle avoidance algorithm. The global path database is a pre-calculated and stored set of common paths in the warehouse, containing the optimal paths from any starting point to the end point. This database is generated through offline calculation, considering static factors such as aisle width, shelf layout, and regional traffic, and can quickly respond to regular path queries. The local obstacle avoidance algorithm is a dynamic programming method that responds to environmental changes in real time and can quickly calculate a temporary detour path when an obstacle is detected. The combination of the two provides both efficient and flexible path planning capabilities.

[0113] Next, based on the picking point set, query the global path database to obtain the standard path. The standard path is the optimal path that connects all picking points under ideal conditions (without temporary obstacles), and is usually solved using the traveling salesman problem algorithm. The system first generates the access order of the picking points, then queries the global path database to obtain the optimal path segments between adjacent points, and finally connects these path segments to form a complete standard picking path.

[0114] Then, detect whether there are temporary obstacles in the standard path; among them, temporary obstacles include channel congestion and area maintenance. The system detects temporary obstacles in multiple ways: real-time sensor data (such as cameras, RFID readers), the location information of other operators, the temporary restriction information input by the administrator (such as a certain area is being cleaned or maintained), and the historical congestion probability model. This multi-source data fusion method can timely discover path obstacles that may affect the picking efficiency.

[0115] Subsequently, if a temporary obstacle is detected, start the local obstacle avoidance algorithm to generate a detour path; if no temporary obstacle is detected, keep the standard path. When the local obstacle avoidance algorithm detects an obstacle, it recalculates the alternative route of the affected path segment based on the current environmental state. This algorithm usually adopts the A* or RRT (rapidly-exploring random tree) method, which can find a feasible path to bypass the obstacle within a limited time, while minimizing the additional distance and time costs caused by the detour.

[0116] Finally, calculate the final picking path and generate path visualization data for presentation on the mobile terminal. The final picking path is the result of combining the standard path and the necessary detour paths. The system converts this path into visualization data, including 2D or 3D map display, turning prompts, distance indication, and estimated time, etc., and intuitively presents it to the picking operators through the mobile terminal (such as handheld devices, tablets or smart glasses), providing clear operation guidelines.

[0117] Furthermore, collecting execution data and optimizing the supply chain warehouse management system by analyzing the execution data includes: First, deploy data collection points to record the actual execution trajectory, timestamp, and abnormal events of the picking operation. The data collection points include the location tracking function of the mobile terminal, barcode scanning records, RFID detection points, and IoT sensors, etc. These collection points record the actual movement trajectory of the picking personnel, the timestamp of arriving at each picking point, and the abnormal events during the picking process (such as picking the wrong item, misplacing, path deviation, etc.). These detailed execution data provide the basis for subsequent analysis.

[0118] Then, compare the differences between the actual execution trajectory and the planned path to generate a deviation report. Compare the actual execution trajectory with the planned path, and calculate the path deviation (the spatial difference between the actual path and the planned path), time deviation (the difference between the actual time used and the planned time used), and operation deviation (such as the change in picking order). These deviation data are integrated into a structured deviation report, which contains information such as deviation type, occurrence location, duration, and severity.

[0119] Next, set up a data analysis pipeline to perform statistical aggregation on the deviation report and identify systematic deviation patterns. The data analysis pipeline uses machine learning and data mining methods to cluster and identify patterns in a large number of deviation reports, and discovers recurring deviation patterns. Systematic deviation patterns refer to deviations that occur repeatedly in multiple executions, such as a certain area always requiring extra time, a certain passageway often experiencing congestion, or the picking time of a certain type of goods being underestimated. These patterns reflect the structural differences between system parameters and the actual situation.

[0120] Subsequently, based on the systematic deviation patterns, correct the weight parameters in the affinity matrix to obtain the correction result. According to the deviation pattern analysis, dynamically adjust the association degree weight parameter and frequency weight parameter in the affinity matrix. For example, if it is found that although some goods have a high association degree score but often cause the path to be extended during actual picking, the system will correspondingly reduce the association degree weight; if the co-outbound frequency of some goods is underestimated, resulting in the goods that are often picked together being stored separately, the system will increase the frequency weight. This adaptive parameter adjustment makes the affinity matrix more accurately reflect the actual operation requirements.

[0121] Then, update the goods group mapping relationship in the storage location plan and adjust the time estimation parameters in the job scheduling model to obtain the adjustment result. Based on the corrected affinity matrix, the system correspondingly updates the goods group mapping relationship in the storage location plan to optimize the goods storage location. At the same time, the system adjusts the time estimation parameters in the job scheduling model according to the execution data, such as correcting the passage time in a specific area, updating the standard time for picking operations, and adjusting the resource capacity assessment. These adjustments ensure that future task planning is based on more accurate time and resource estimates.

[0122] Finally, transmit the correction result and the adjustment result back to complete the closed-loop optimization of the supply chain warehousing management system. The system transmits the corrected affinity matrix parameters, the updated storage location plan, and the job scheduling parameters back to the corresponding functional units to form a complete data closed-loop. The next round of planning will use these updated parameters, thereby achieving the continuous self-optimization of the system, making the warehousing management process more and more in line with the actual operating conditions, and continuously improving efficiency and accuracy.

[0123] Based on the same inventive concept, an embodiment of the present application further provides a supply chain warehousing management method based on AI digitalization. The implementation solutions provided by this system to solve problems are similar to the implementation solutions described in the above method. Therefore, the specific limitations in one or more embodiments of the supply chain warehousing management method based on AI digitalization provided below can refer to the limitations on a supply chain warehousing management system based on AI digitalization in the above text, and will not be elaborated here.

[0124] In an exemplary embodiment, as Figure 3 shown, a supply chain warehousing management method based on AI digitalization is provided, including the following steps 202 to step 210: Step 202: Collect supply chain system data through a data processing module, and construct an associated data network based on the supply chain system data; the associated data network includes goods information, warehouse location information, and order information; Step 204: Execute a goods location allocation algorithm according to the associated data network. The goods location allocation algorithm extracts the association degree score of the goods based on the goods information, analyzes the collaborative outbound frequency of historical orders according to the order information, and calculates the affinity matrix between the goods; Step 206: Generate a storage location plan based on the affinity matrix; Step 208: Construct a job scheduling model according to the storage location plan, optimize tasks by combining the constraint conditions in the storage location plan, and output a task allocation sequence through the job scheduling model; Step 210: Configure a picking path planning module based on the task allocation sequence, use the picking path planning module to present the picking path and operation guidelines on the mobile terminal, collect execution data, and optimize the supply chain warehousing management system by analyzing the execution data.

[0125] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as these combinations of technical features do not conflict, they should be considered as the scope recorded in the present application.

[0126] The above-described embodiments only represent several implementation manners of the present application. Their descriptions are relatively specific and detailed, but they should not be construed as limiting the patent scope of the present application. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present application, several modifications and improvements can still be made, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the appended claims.

Claims

1. A supply chain warehousing management system based on AI digitalization, characterized in that, Including: A data processing unit, configured to collect supply chain system data through a data processing module and construct an associated data network including goods information, warehouse location information, and order information based on the supply chain system data; A goods location allocation unit, configured to execute a goods location allocation algorithm according to the associated data network. The goods location allocation algorithm extracts an association degree score of goods based on the goods information, analyzes the collaborative outbound frequency of historical orders according to the order information, and calculates an affinity matrix between goods; A storage plan unit, configured to generate a storage location plan based on the affinity matrix; The process of generating the storage location plan includes constructing a mathematical model of the warehouse space; Performing clustering analysis on the affinity matrix, dividing goods with affinity values exceeding a first preset threshold into the same goods group; mapping the goods group to the mathematical model of the warehouse space to generate a first storage location plan; setting a multi-objective optimization function, and calculating the total picking path length and the goods location occupancy rate index of the first storage location plan; By iteratively adjusting the goods location allocation in the first storage location plan, reducing the total picking path length and increasing the goods location occupancy rate index to form a second storage location plan; A job scheduling unit, configured to construct a job scheduling model according to the storage location plan, perform task optimization in combination with the constraint conditions in the storage location plan, and output a task allocation sequence through the job scheduling model; A picking navigation unit, configured to configure a picking path planning module based on the task allocation sequence, present a picking path and operation instructions on a mobile terminal by using the picking path planning module, collect execution data, and optimize the supply chain warehouse management system by analyzing the execution data.

2. The supply chain warehouse management system based on AI digitalization according to claim 1, characterized in that: The extracting the association degree score of goods based on the goods information includes: extracting a category identifier, a physical size, and a storage requirement in the goods information to generate a goods feature vector, and performing a similarity calculation on the goods feature vector to obtain the association degree score.

3. The supply chain warehousing management system based on AI digitalization according to claim 2, characterized in that: The analyzing the collaborative outbound frequency of historical orders according to the order information includes: counting the co-occurrence times of goods pairs in the historical orders within a preset time period, recording the outbound time interval of each group of goods pairs, and calculating the standardized collaborative outbound frequency; the calculating the affinity matrix between goods includes: setting an association degree weight parameter and a frequency weight parameter, and performing a weighted combination on the association degree score and the collaborative outbound frequency to form the affinity matrix.

4. The supply chain warehousing management system based on AI digitalization according to claim 3, characterized in that: The mathematical model of the warehouse space abstractly represents the physical warehouse space through goods location coordinates, aisle layout, and area division; wherein, the goods location coordinates represent the position of each goods location through a three-dimensional coordinate system; the aisle layout is represented by a set of edges connecting nodes, each edge represents a passable path, and the length and passing conditions are recorded; the area division divides the warehouse into multiple functional areas, and each area contains a set of goods locations with the same storage conditions.

5. The supply chain warehousing management system based on AI digitalization according to claim 4, characterized in that: Constructing a job scheduling model according to the storage location scheme includes: converting the storage location scheme into a topological graph structure; wherein, the topological graph structure represents storage locations with nodes and feasible walking paths with edges; establishing a resource constraint table; wherein, the resource constraint table includes the number of available pickers, the number of devices, and time window parameters; calculating the shortest path distance matrix between nodes in the topological graph structure; constructing a task conflict detection matrix to record time overlaps and resource competitions between tasks; and generating a mathematical expression of the job scheduling model according to the shortest path distance matrix and the task conflict detection matrix.

6. The supply chain warehousing management system based on AI digitalization according to claim 5, wherein: Combining the constraint conditions in the storage location scheme for task optimization and outputting a task assignment sequence through the job scheduling model includes: Receiving order data and converting the order data into a set of picking tasks; Calculating a task similarity matrix according to the composition and quantity of goods in each task in the set of picking tasks; Performing task merging based on the task similarity matrix, and combining tasks with similarity exceeding a merging threshold into batch tasks; Using integer programming methods to perform time slot allocation and resource allocation for the batch tasks to generate an initial scheduling scheme; Detecting resource conflict points in the initial scheduling scheme and eliminating the resource conflict points by adjusting the task order; Generating the task assignment sequence; wherein, the task assignment sequence includes the execution sequence number, start time, allocated resources, and picking path information of each task.

7. The supply chain warehousing management system based on AI digitalization according to claim 6, wherein: Configuring a picking path planning module based on the task assignment sequence includes: Parsing the task assignment sequence and extracting the set of picking points and order requirements for each task; Constructing a path planning engine; wherein, the path planning engine includes a global path database and a local obstacle avoidance algorithm; Querying the global path database according to the set of picking points to obtain a standard path; Detecting whether there are temporary obstacles in the standard path; wherein, the temporary obstacles include passage congestion and area maintenance; If a temporary obstacle is detected, starting the local obstacle avoidance algorithm to generate a detour path; if no temporary obstacle is detected, keeping the standard path; Calculating the final picking path and generating path visualization data for presentation on the mobile terminal.

8. The supply chain warehousing management system based on AI digitalization according to claim 7, wherein: Collecting execution data and optimizing the supply chain warehouse management system by analyzing the execution data includes: Deploying data collection points to record the actual execution trajectory, time stamps, and abnormal events of the picking operation; Comparing the difference between the actual execution trajectory and the planned path to generate a deviation report; Setting up a data analysis pipeline to statistically aggregate the deviation report and identify systematic deviation patterns; According to the systematic deviation patterns, correcting the weight parameters in the affinity matrix to obtain a correction result; Updating the goods group mapping relationship in the storage location scheme and adjusting the time estimation parameters in the job scheduling model to obtain an adjustment result; Sending back the correction result and the adjustment result to complete the closed-loop optimization of the supply chain warehouse management system.

9. A supply chain warehousing management method based on AI digitalization, which adopts a supply chain warehousing management system based on AI digitalization as described in any one of claims 1 to 8, and is characterized in that, Including: Collecting supply chain system data through a data processing module and constructing an associated data network based on the supply chain system data; Execute the location allocation algorithm according to the associated data network. The location allocation algorithm extracts the association degree score of the goods based on the goods information, analyzes the collaborative outbound frequency of historical orders according to the order information, and calculates the affinity matrix between the goods; Generate a storage location plan based on the affinity matrix; Construct a job scheduling model according to the storage location plan, optimize the tasks in combination with the constraints in the storage location plan, and output a task allocation sequence through the job scheduling model; Configure a picking path planning module based on the task allocation sequence, use the picking path planning module to present the picking path and operation guidelines on the mobile terminal, collect execution data, and optimize the supply chain warehouse management system by analyzing the execution data.

10. The method for supply chain warehousing management based on AI digitalization according to claim 9, characterized in that: The associated data network includes goods information, warehouse location information and order information.

Images