Multi-type order insertion conflict resolution method and device based on large model, and medium

Abstract

The embodiment of the invention discloses a multi-type order insertion conflict resolution method and device based on a large model and a medium, and relates to the technical field of large models, and the method comprises the steps: obtaining a multi-source order insertion request, carrying out the semantic analysis of each order insertion request through a pre-trained clothing industry large model, and obtaining a multi-source order insertion request; determining a current order insertion type corresponding to each order insertion request and a corresponding conflict matrix, wherein the order insertion type comprises an explosive order returning type, an inventory shortage order insertion type, a quality inspection remedy order insertion type, a VIP order adding type and a fast reverse order tracking type; according to the conflict matrix, determining an order insertion conflict pair with a resource conflict, and obtaining a real-time resource index of the order insertion conflict pair in a real-time service system based on the order insertion conflict pair; based on the real-time resource index data, according to a preset triggering condition of a basic action element, a conflict resolution strategy of the order insertion conflict pair is matched, and the basic action element comprises a preemption action element, an extension action element and a segmentation action element. False conflict misjudgment and hidden conflict missed judgment are eliminated, and the defect of strategy and service disjunction is avoided.

Description

A method, device, and medium for resolving multiple types of single-insertion conflicts based on a large model. Technical Field

[0001] This specification relates to the field of large model technology, and in particular to a method, device and medium for resolving multiple types of insertion conflicts based on large models. Background Technology

[0002] In the field of intelligent manufacturing in the apparel industry, there is an inherent contradiction between the stability and flexibility of production plans. The volatile nature of market demand, especially sudden order requests from sales, supply chains, and clients, often drastically impacts established production plans, leading to resource conflicts on production lines. These orders are not homogeneous but rather represent different business scenarios with varying priorities, constraints, and business objectives. Typical order types include, but are not limited to: repeat orders for best-selling items due to market popularity (with extremely high delivery time requirements); VIP orders arising from temporary demands from VIP clients (emphasizing customer experience and priority guarantees); inventory shortage orders forced by shortages of specific fabrics and accessories (requiring rapid adjustments to processes and material substitutions); quality control remedial orders that must be addressed immediately after batch issues are discovered during quality inspection (concerning quality closure and brand reputation); and rapid response orders dynamically initiated based on pre-sale data (pursuing ultimate market responsiveness). These various types of orders often occur concurrently during certain events, competing for the same production capacity, equipment, or material resources, resulting in complex resource conflicts.

[0003] Currently, the apparel industry generally relies on planning and scheduling personnel for manual conflict resolution. Schedulers need to rely on years of experience to simultaneously grasp massive amounts of heterogeneous information, such as the process characteristics of the entire product line, the real-time load status of each production line, the dynamics of fabric and accessory inventory, equipment compatibility, and customer priorities. They also need to rely on subjective experience to judge conflicts and formulate strategies. However, relying on manual conflict resolution requires personnel with extremely high experience, who need to be deeply familiar with the process logic and production line operation status of all product lines. On the other hand, when there are large numbers of order requests and multiple types of high-frequency concurrency, manual processing faces a serious information overload problem. Schedulers cannot fully analyze the implicit resource dependencies between orders within a limited time, which can easily lead to missed or misjudged conflicts, resulting in problems such as delayed scheduling decisions, frequent production line changes, material mismatch, and delivery delays.

[0004] Therefore, under the existing order insertion conflict handling model that is dominated by human experience, when multiple types of orders are inserted concurrently and in large quantities, there are problems such as high threshold for human operation experience and information overload leading to inaccurate conflict identification. As a result, dispatchers can only rely on human experience, lack decision reference information, and are unable to make efficient conflict resolution decisions. Summary of the Invention

[0005] This specification provides one or more embodiments of a method, device, and medium for resolving multi-type order insertion conflicts based on a large model, which is used to solve the following technical problem: In the existing order insertion conflict handling mode dominated by human experience, when multiple types of orders are concurrent and the number is large, there is a problem that the human operation experience threshold is high and information overload leads to inaccurate conflict identification. As a result, the dispatcher can only rely on human experience, lacks decision reference information, and cannot make efficient conflict resolution decisions.

[0006] One or more embodiments of this specification adopt the following technical solution: One or more embodiments of this specification provide a method for resolving multi-type order insertion conflicts based on a large model. The method includes: acquiring multi-source order insertion requests; performing semantic parsing on each order insertion request using a pre-trained large model of the apparel industry to determine the current order insertion type corresponding to each order insertion request; performing coupled analysis of business entity sharing and real-time business indicators based on multiple current order insertion types to determine the corresponding conflict matrix, wherein the order insertion types include popular item return orders, inventory shortage order insertions, quality inspection remedial order insertions, VIP additional orders, and quick response follow-up orders; determining order insertion conflict pairs with resource conflicts according to the conflict matrix; acquiring real-time resource indicator data of the order insertion conflict pairs in the real-time business system based on the order insertion conflict pairs; and matching the conflict resolution strategy of the order insertion conflict pairs according to the triggering conditions of preset basic action elements based on the real-time resource indicator data, wherein the basic action elements include preemption action elements, extension action elements, and segmentation action elements.

[0007] This specification provides one or more embodiments of a multi-type single-interruption conflict resolution device based on a large model, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the above-described method.

[0008] This specification provides one or more embodiments of a non-volatile computer storage medium storing computer-executable instructions configured to perform the above-described method.

[0009] The at least one technical solution adopted in the embodiments of this specification can achieve the following beneficial effects: By deeply analyzing text semantics through a large model, extracting business primary keys and associating them with real-time business indicators, conflicts are identified based on business logic rather than surface features, ensuring that conflict pair screening is based on real resource competition relationships; a quantitative conflict matrix is ​​constructed, which is dynamically calculated based on shared business entities and multi-dimensional dependencies, thereby accurately identifying order insertion conflict pairs that truly have resource competition relationships and effectively filtering out pseudo-conflicts that only have superficial associations such as related customers, thus achieving accurate conflict identification; at the same time, the order insertion type is embedded as a core parameter in the construction of the conflict matrix and the triggering conditions of action elements, so that the conflict resolution strategy is deeply coupled with business objectives. In this system, for example, reorders of popular items automatically trigger a preemptive strategy due to high timeliness requirements, while orders due to inventory shortages are prioritized and extended due to material constraints, thus avoiding the defect of strategy being disconnected from business operations. Based on real-time resource indicators, basic action elements are triggered and matched. The generation of conflict resolution strategies does not apply fixed rules, but rather dynamically combines the most suitable optimization solutions based on the specific economic value of both parties in the conflict, the current degree of resource scarcity, and the real-time status of production capacity pressure. This elevates conflict handling from passive response to proactive optimization, eliminating false conflict misjudgments and hidden conflict omissions, significantly improving the scientific nature of production line resource scheduling and the stability of production plans, realizing the transformation from surface-characteristic conflicts to business logic conflicts, and providing technical support for flexible production in intelligent garment manufacturing. Attached Figure Description

[0010] To more clearly illustrate the technical solutions in the embodiments or prior art of this specification, the drawings used in the description of the embodiments or prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this specification. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In the drawings: Figure 1 is a schematic flowchart of a multi-type insertion conflict resolution method based on a large model provided in an embodiment of this specification; Figure 2 is an example diagram of process parameters corresponding to a product type provided in an embodiment of the specification; Figure 3 is a schematic structural diagram of a multi-type insertion conflict resolution device based on a large model provided in an embodiment of this specification. Detailed Implementation

[0011] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this specification, and not all embodiments. Based on the embodiments of this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this specification.

[0012] Currently, the industry treats order insertions as undifferentiated, independent events, using a first-come, first-served approach or prioritizing orders based on a single dimension (such as order amount or delivery date). When multiple order insertion types occur concurrently, it's impossible to accurately identify which orders truly have resource conflicts. For example, simply having the same customer ID might incorrectly classify two orders sharing the same VIP customer as highly conflicting, ignoring the fact that they produce different products, use different materials, and actually have no resource competition. Conversely, it might also miss a strong conflict between two orders belonging to a best-selling repeat order and a fast-response follow-up order, respectively. Despite different customers, both are highly dependent on the same scarce fabric and core production line; without coordinated scheduling, this could lead to severe bottlenecks. Due to a lack of understanding of the semantics of order insertion types, existing methods cannot distinguish between pseudo-conflicts (such as overlapping customer IDs but no resource overlap) or capture implicit conflicts (such as the combined pressure of different order insertion types on the same bottleneck process), resulting in inaccurate conflict identification. This inaccuracy directly leads to a disconnect between scheduling decisions and actual business operations, causing avoidable conflicts to spread passively across production lines, warehouses, and the supply chain. Ultimately, this manifests as frequent line changes, material mismatches, delivery delays, and decreased capacity utilization, severely restricting the flexible response capabilities and overall production efficiency of garment manufacturing enterprises.

[0013] This specification provides a method for resolving multi-type order insertion conflicts based on a large model. It should be noted that the execution entity in this specification can be a server or any device with data processing capabilities. Figure 1 is a flowchart illustrating a multi-type order insertion conflict resolution method based on a large model provided in this specification. As shown in Figure 1, the method mainly includes the following steps: Step S101: Obtain multi-source order insertion requests; perform semantic parsing on each order insertion request using a pre-trained large model for the apparel industry to determine the current order insertion type corresponding to each request; and perform coupled analysis of business entity sharing and real-time business indicators based on multiple current order insertion types to determine the corresponding conflict matrix.

[0014] The types of order insertions include repeat orders for best-selling items, order insertions for inventory shortages, order insertions for quality inspection remediation, VIP order additions, and quick-response order follow-ups. In one embodiment of this specification, multi-source order insertion requests are first obtained. "Multi-source" refers to different business systems from which the requests originate, including at least the Supply Chain Management System (SCM), Customer Relationship Management System (CRM), the early warning module of the Manufacturing Execution System (MES), and the Transportation Management System (TMS). Multi-source order insertion requests refer to real-time order insertion data from different business systems within the enterprise, including repeat orders for best-selling items triggered by market demand in the Sales Management System (e.g., an emergency order of 100 DRESS-001 dresses for the 618 promotion, requiring delivery within 48 hours), order addition requests submitted by VIP customers in the Customer Relationship Management System (e.g., a temporary order of 50 haute couture dresses by VIP customer Zhang San recorded in the CRM system, requiring priority processing), and order insertion requests for inventory shortages triggered by material shortages in the Supply Chain Management System (e.g., a temporary order of 50 haute couture dresses by VIP customer Zhang San recorded in the CRM system, requiring priority processing), and order insertion requests for inventory shortages triggered by material shortages in the Supply Chain Management System (e.g., a temporary order of 50 haute couture dresses by VIP customer Zhang San recorded in the Supply Chain Management System, requiring priority processing). The alert includes FAB-001 (fabric inventory below the safety threshold, requiring urgent production line adjustments), quality inspection remedial order requests generated after batch issues are discovered in the quality inspection system (e.g., batch B-20250601 recorded in the quality inspection system has defects, requiring 100 pieces to be produced within 24 hours), and rapid response order requests dynamically initiated based on pre-sale data (e.g., a rapid response activity pushed by the pre-sale platform to add 200 lightweight jackets for the Double 11 pre-sale, requiring delivery within 72 hours). All order requests are transmitted in real-time via an enterprise-grade API gateway in structured text format (such as JSON). The acquisition method can be set to automatically poll the API interfaces of sales, CRM, supply chain, quality inspection, and pre-sale platforms every 5 minutes, receiving new order data through a message queue (such as Kafka) to ensure data timeliness and storing it in a temporary data buffer. It should be noted that multi-source order insertion requests include not only real-time order insertion requests, but also unprocessed historical order insertion requests that have not yet been put into production. Furthermore, in this embodiment of the specification, a multi-source order insertion request can be two popular product reorder insertion requests and one inventory shortage insertion request. In other words, there may be two order insertion requests of the same type among the multi-source order insertion requests, or there may be multiple order insertion requests with different order insertion types. This embodiment of the specification uses the case of multi-source order insertion requests corresponding to different order insertion types as an example for illustration.

[0015] The pre-trained large-scale model for the apparel industry is used to perform semantic parsing on each order insertion request to determine the current order insertion type and corresponding conflict matrix for each request. This is achieved through the following steps: First, the large-scale model for the apparel industry is pre-trained. The large-scale model for the apparel industry in this embodiment is based on the open-source general-purpose model BERT-base and is finely tuned using 100,000 apparel industry order corpora (including five types of order insertion history records: popular item reorders, inventory shortage orders, quality inspection remedial orders, VIP additional orders, and quick-response follow-up orders). This model has embedded the apparel industry's unique business knowledge system to ensure its deep coupling with the apparel production scenario. First, the architecture of the large-scale model for the apparel industry is based on the Transformer encoder and has been specifically modified for the characteristics of the apparel industry. Its necessary modules and layers include: an embedding layer that integrates a dictionary of apparel industry professional terms (such as "hot-selling items", "silk", and "overlock") and a dedicated word segmenter; a named entity recognition layer based on BIO-annotated sequences, used to identify business primary keys in text; a data interface layer that interacts in real time with external business databases (such as ERP, PLM, and CRM systems); a cross-modal fusion layer, used to concatenate and interact with the text encoding vector and the structured business indicator vector obtained from the query; and a top multi-classification layer that outputs the probability distribution corresponding to five types of order insertion, such as "hot-selling item reorder".

[0016] The training of the large-scale apparel industry model involves two stages: The first stage is pre-training, which uses a large-scale unlabeled text corpus collected from apparel industry technical documents, historical order contracts, production logs, and supply chain reports. Self-supervised tasks such as masked language models are employed to train the model, enabling it to master general semantic and grammatical knowledge in the apparel field. The second stage is supervised fine-tuning, which uses a manually labeled set of historical order insertion requests. Each sample includes the original text description, the labeled business primary key entity and its type, and the business type label ultimately determined by experts. Key training parameters used in this stage include: a step-decreasing learning rate with an initial value of a specific value; a batch size set according to GPU memory; a cross-entropy loss function; and the AdamW optimizer with a weight decay coefficient to prevent overfitting. The second step involves semantic parsing of each order insertion request using the pre-trained large-scale apparel industry model to determine the current order insertion type and corresponding conflict matrix for each request.

[0017] First, the text description of the order insertion request is input into a pre-trained large-scale model for the apparel industry. The text description is segmented and entity-labeled to determine multiple business primary keys. These business primary keys are codes that uniquely identify business entities in the system, including apparel product primary keys, fabric and accessory primary keys, customer primary keys, production line primary keys, and pre-sale activity primary keys. The received text description of the order insertion request is then input into the pre-trained large-scale model for the apparel industry. This pre-trained model has added customized modules for the apparel manufacturing field to its basic architecture. First, the segmentation and entity labeling of the text description are performed by the model's embedded preprocessing and understanding module. Segmentation divides natural language text into the smallest semantic units according to semantic boundaries. The segmentation process uses a dedicated segmenter that integrates a professional terminology database for the apparel industry, dividing the input continuous text sequence into lexical units (Tokens) with independent semantic meaning. For example, "618 promotion urgently adding 100 DRESS-001 dresses" is segmented into "618 promotion," "urgent," "additional," "100 pieces," "DRESS-001," and "dress." Next, entity labeling identifies key business entities and assigns them unique identifiers to the system. The entity labeling function is implemented by the named entity recognition layer inside the model, which has been fine-tuned by the sequence labeling task. This layer can identify and label key business entities that are explicitly mentioned in the text and belong to predefined categories. The directly output entity set constitutes the first business primary key set.

[0018] For example, in the text "618 promotion urgently adds 100 DRESS-001 dresses," the model will identify and label DRESS-001 as the primary key for the clothing product and "618 promotion" as the primary key for the pre-sale event. However, the complete business profile of an order insertion request often relies on more entities that are not explicitly stated in the text but are closely related through business logic. Therefore, the primary key parsing of related business operations is performed. Using the first set of primary keys identified in the previous step as an index, an online query is initiated to the enterprise's real-time business system through the data interface layer of the large model integration. Based on the type of the identified primary key, the query is routed to the corresponding business database. For example, using the apparel product primary key (DRESS-001), the Product Lifecycle Management (PLM) or Bill of Materials (BOM) system is queried to obtain the list of primary keys for the necessary fabrics and accessories for producing this style (e.g., fabric F-2024-SILK, accessories Z-2024-BUTTON). The Customer Relationship Management (CRM) or Marketing system is queried to parse out the core customer primary key that initiated the order for this product (e.g., e-commerce platform ID or key distributor code). The real-time timestamp corresponding to the order insertion request and the existing pre-sale activity list are then queried to obtain the corresponding pre-sale activity primary key. Generally, the pre-sale activity primary key is obtained from the text description of the order insertion request. If no pre-sale activity primary key is found after identifying it through the text description, it is determined that the order insertion request does not contain a pre-sale activity primary key. Simultaneously, for this specific order insertion requirement, the production line primary key of the product is obtained from the Manufacturing Execution System (MES). Through real-time relational queries and dynamic generation, a set of related business primary keys is obtained. The first set of business primary keys and the set of related business primary keys are merged and deduplicated to determine multiple business primary keys. For example, DRESS-001 is marked as the primary key for clothing products, referring to the system code that uniquely identifies clothing products and is used to associate product attributes; 618 activity is marked as the primary key for pre-sale activities, referring to the system code that uniquely identifies promotional activities and is used to associate activity attributes; VIP customer Zhang San is marked as the primary key for customers, referring to the system code that uniquely identifies customers and is used to associate customer level; SLK-888 silk gauze is marked as the primary key for fabrics and accessories, referring to the specific fabric or accessory code; and the production line primary key refers to the production line identifier bound to a specific production task.

[0019] Through a large-scale model of the apparel industry, the business database is queried in real time. The model calls its integrated data interface adapters, which are pre-configured with connection parameters. These adapters, based on the specific business primary key value identified in the previous step, initiate low-latency real-time query requests to the corresponding enterprise business systems (such as Product Lifecycle Management (PLM), Enterprise Resource Planning (ERP), Customer Relationship Management (CRM), and Warehouse Management System (WMS)) to obtain the business metrics corresponding to the business primary key. Business metrics refer to real-time business data associated with the business primary key. For example, based on the apparel product primary key, the system can query the sales growth rate and repurchase rate of that style within a preset time period; based on the fabric and accessories primary key, it can query the real-time inventory level and safety stock threshold of that material; based on the customer primary key, it can query the customer's credit rating and cooperation importance score; for the production line primary key, it can query the quality inspection report system to obtain the batch defect rate and rework count; and for the pre-sale activity primary key, it can query the marketing system to obtain the activity conversion rate and regional best-selling index. The heterogeneous, dynamically changing numerical data returned by the query is normalized and then ordered into a fixed-dimensional business indicator vector. This business indicator vector refers to a fixed-dimensional vector formed after normalizing a set of numerical values ​​corresponding to multiple business indicators obtained from a pre-trained large-scale apparel industry model's real-time query of the business database, and then converting them into a single vector. The business indicator vector serves as the core data bridge connecting the large-scale model and the business system, enabling the model to perform accurate semantic parsing based on real-time business status. For example, given inventory ratio ∈ [0,1] and customer level ∈ [0,100], an indicator vector is generated, such as the product primary key query result [sales growth rate, repurchase rate]. Simultaneously, the model's text encoder (such as a Transformer Encoder) independently performs deep semantic analysis on the original text description, generating a high-dimensional original semantic text encoding vector. This vector contains the contextual semantics and potential intent of the order description.

[0020] The original semantic text encoding corresponding to the text description and the business indicator vector are concatenated through a cross-modal fusion layer. A fully connected layer then performs feature interaction and dimensionality reduction to determine the joint vector corresponding to the order insertion request. This joint vector integrates unstructured semantic information and structured real-time business status information. Finally, using a large-scale apparel industry model and the joint vector, the probability distribution of the order insertion type corresponding to the request is output to determine the current order insertion type. The joint vector is then input into a multi-classification layer at the top of the model, activated by a softmax function. This classification layer, trained on a large amount of labeled data (text description and corresponding order insertion type labels), calculates the probability that the joint vector belongs to one of five preset categories: best-selling reorder, inventory shortage order insertion, quality inspection remedial order insertion, VIP additional order, and quick-response follow-up order. A five-dimensional probability distribution vector is output. Determining the current order insertion type involves selecting the category corresponding to the maximum value in this probability distribution as the final judgment of the business nature of the order insertion request.

[0021] Compared with conventional methods, this technical solution achieves semantic-level parsing of order insertion types through a pre-trained large model of the apparel industry, fundamentally solving the core problem of inaccurate conflict identification caused by ignoring the differences in order insertion types in the background technology; by deeply parsing the semantics of the text through the large model, the business primary key is extracted and associated with real-time business indicators, and conflicts are identified based on business logic rather than surface features, ensuring that the screening of conflict pairs is based on the real resource competition relationship.

[0022] Based on the business primary key and business indicator vector corresponding to each order insertion request, a coupling analysis of business entity sharing and real-time business indicators is performed to determine the conflict matrix. The specific implementation method is as follows: First, based on the number of current order insertion types, an initial matrix of order corresponding to the number is constructed, where the rows and columns of the initial matrix correspond to a pair of order insertion types; if there are N order insertion requests to be processed in the current system, an N-row N-column square matrix is ​​initialized in memory as the initial matrix, where the row index i and column index j of the matrix correspond to the i-th and j-th specific order insertion requests, respectively. Since this embodiment takes multiple order insertion requests of different types as an example, it can also be understood that the row index i and column index j correspond to the i-th and j-th specific order insertion types, respectively. The initial value of each element M[i][j] in the matrix is ​​set to zero, indicating that no conflict relationship has been analyzed yet. This square matrix is ​​the initial matrix.

[0023] Obtain the business primary key corresponding to each order type in each order type pair. Take the intersection of the business primary keys of the order type pairs to determine the shared business primary key for each order type pair. Here, the business primary key refers to the complete set of business primary keys for each order request obtained through the aforementioned large model parsing and real-time association query. For each different pair of order requests in the matrix (i.e., each i,j pair, i≠j), extract their respective business primary key sets and calculate their intersection. Shared entity = business primary key of order i ∩ business primary key of order j (only retaining the same primary key type, such as product ID ∩ product ID). This intersection result is the shared business primary key. For example, sharing the primary key of raw materials and accessories means competing for the same batch of materials. For example, if the primary key for order A is [Product ID=DRESS-001, Material ID=FAB-001, Customer ID=CUST-777], and the primary key for order B is [Product ID=DRESS-001, Material ID=FAB-002, Customer ID=CUST-888], then the shared primary key for both order A and order B is the product primary key. In practical applications, if the shared primary key contains a product primary key, it typically also includes a material primary key and / or a production line primary key.

[0024] When the shared business primary key is not null, it indicates that the two insertion types in the insertion pair share a shared business primary key, excluding the case where i=j. Dependency strength is calculated only when the shared entity is not null. Based on the shared business primary key, corresponding dependency dimensions are matched, including resource dependency, time dependency, and process dependency. This can be achieved through a pre-defined mapping rule library, associating different types of shared business primary keys with three predefined dependency dimensions: resource dependency, time dependency, and process dependency. The degree of conflict between multiple insertion types is quantified using these three dimensions. For example, resource dependency strength calculation is triggered only when a shared entity contains both a clothing product primary key and a fabric / accessory primary key, or contains a fabric / accessory primary key. When a shared entity contains a clothing product primary key, it indicates that the two orders belong to the same product. In this case, the fabrics and accessories used by the same product are generally the same. Therefore, a shared entity usually contains both a clothing product primary key and a fabric / accessory primary key, resulting in a resource conflict related to fabrics and accessories. Another scenario is that the shared entity contains a fabric / accessory primary key, meaning that two different clothing products use the same fabrics and accessories, which also results in a resource conflict related to fabrics and accessories. Time dependency strength calculation is based on delivery time conflicts. When a shared entity contains a pre-sale activity primary key, it means that the two orders belong to the same pre-sale activity, and their corresponding delivery times are usually the same. When a shared entity contains a customer primary key, it implies that the two orders belong to the same customer order, but there may be differences in delivery times. Therefore, time dependency strength calculation is triggered only when the shared entity does not contain a pre-sale activity primary key or contains a customer primary key. Based on process compatibility at the process dimension, when a shared entity contains a garment product primary key or a production line primary key, it indicates that two insert orders are producing the same garment product or using the same production line, and their process parameters are completely identical. Therefore, process similarity calculation is only triggered when the shared entity does not contain a garment product primary key or a production line primary key. Following the above rules, at least one required dependency dimension is matched based on the shared business primary key of each insert order type pair.

[0025] Based on the business metric vector, the dependency strength component corresponding to the dependency dimension is calculated. Resource dependency strength quantifies the intensity of competition between two order requests due to shared key resources (materials and accessories). When multiple orders have high demand for the same limited resource, and the available quantity of that resource is insufficient, the resource dependency strength approaches 1, indicating severe conflict; conversely, if the resource is sufficient or the demand is low, the dependency strength approaches 0, indicating no significant conflict. The calculation of resource dependency strength accurately quantifies the intensity of competition between two order requests due to shared key resources (materials and accessories primary key). Resource dependency strength = max{0, 1 - (available inventory / total demand)}. When the shared entity contains a material and accessories primary key, it indicates that the two orders share the same material and accessories. In this case, the total demand needs to be calculated based on physical consumption, specifically obtained by summing the material and accessories demand of order i and order j. Available inventory is the real-time inventory obtained based on the associated material and accessories ID. The demand for raw materials and accessories = the quantity of products required corresponding to the order request × the consumption of raw materials and accessories per unit. This consumption is accurately obtained from the BOM (Bill of Materials). When the available resources are insufficient, i.e., the available inventory is less than the total demand, the resource dependency strength value is between (0,1). The smaller the available inventory, the closer the dependency strength is to 1, indicating a more severe resource conflict. For example, if the inventory is 20 and the total demand is 50, the dependency strength is 0.6. When the available resources are sufficient, i.e., the available inventory is not less than the total demand, the resource dependency strength is 0, indicating that the resources are sufficient to meet the demand and there is no significant conflict. When the resources are completely unavailable, i.e., the available inventory is 0, the resource dependency strength is 1, indicating a severe conflict.

[0026] Time dependency strength = 1 - (time overlap ratio). Time overlap ratio = max(0, min(delivery time_i, delivery time_j) - max(current time, delivery start time_i, delivery start time_j)) / (max(delivery time_i, delivery time_j) - min(delivery time_i, delivery time_j)), calculated only when delivery times overlap; otherwise, time dependency strength = 0. In the technical solution of this application, the strength component corresponding to the process dimension refers to process similarity, that is, quantifying the degree of conflict between two order insertion requests due to the special characteristics of the production process. The core idea is that the more similar the production processes required by two order insertion requests, the higher their dependence on the same set of process equipment, special tooling fixtures, or skilled workers with specific skills, and therefore the greater the possibility of conflict in the process dimension. Therefore, the process similarity is the process similarity in the production process corresponding to the two insert orders, that is, the process similarity between the process parameters of insert order i and the process parameters of insert order j. Figure 2 is an example diagram of process parameters corresponding to a product type provided in the embodiment of the specification. As shown in Figure 2, for example, order USC602030174 includes at least a variety of process parameters such as flat lapel, half-canvas, and lapel width of 8.3cm. In this example system, each process parameter is displayed through the corresponding process code.

[0027] As shown in Figure 2, each order insertion request (corresponding to a specific order) is associated with a complete product process configuration table in the MES system, which is a list consisting of dozens or even hundreds of process codes and parameters, as shown in the attached figure. These process codes (such as 0001 representing "flat lapel", 000B representing "half-canvas lining", 0012 representing "single-row two-button closure", etc.) constitute the complete set of process requirements for the order insertion request. When two different order insertion requests need to occupy the same special process resources, for example, both need to execute the "0001 flat lapel" process, which can only be completed on specific dedicated equipment such as a dedicated lapel shaping machine, then these two order insertion requests have a dependency relationship in the process dimension.

[0028] The following section, referring to Figure 2, details the calculation process of process similarity using the process parameters shown in Figure 2 as an example: First, extract the process parameter vector. For each order insertion request i, using its garment product primary key as an index, query the product process configuration table in the MES system to extract all process codes associated with the order insertion request, forming a multi-dimensional process parameter vector Ai. For example, the set of process codes for order i can be represented as: Si = {0001, 000B, 0012, 00B4,00F7, 00G2, 00N2, 00V4, 00VF, 0101, 0201, ……} (a total of 268 codes, as shown in Figure 2). Next, process similarity is calculated using the Jaccard similarity coefficient algorithm based on set overlap. The matching degree between two process code sets is calculated as follows: Process similarity S_process = |Si ∩ Sj| / |Si ∪ Sj|, where |Si ∩ Sj| represents the number of process codes shared by the two order requests, and |Si ∪ Sj| represents the total number of unique process codes in the two order requests. The physical meaning of this formula is the proportion of processes shared by the two order requests to their total number of process types; a higher proportion indicates a higher degree of overlap in their process-level requirements.

[0029] For example, suppose order i has 268 processes (as shown in Figure 2) and order j has 245 processes, and they share 220 process codes. Then, S_process = 220 / (268 + 245 - 220) = 220 / 293 ≈ 0.75. The dependency strength component in the process dimension is directly set as the calculated process similarity S_process. The business logic is that the higher the similarity of the process code sets of the two order requests, the higher the overlap in their needs for the same set of process equipment, tooling fixtures, or specialized skilled workers, and therefore the greater the possibility of resource conflicts in the process dimension. For example, if S_process is calculated to be 0.75, then the component corresponding to the process dimension is 0.75.

[0030] This application directly sets the process similarity component to the calculated process similarity S_process. The business logic is that the higher the process similarity between two order insertion requests, the greater the overlap in their needs for the same set of process equipment, tooling fixtures, or specialized skilled workers, thus increasing the likelihood of resource conflicts at the process level. For example, if S_process is calculated to be 0.85, then the process similarity component is also 0.85.

[0031] The comprehensive dependency strength of each insert type pair is determined by using dependency strength components. This can be achieved by learning weights from a historical conflict case database using a weighted fusion method, obtaining resource dependency weights, time dependency weights, and process dimension weights, or by setting weights based on expert experience, such as setting the same weights for all three dimensions (e.g., resource dependency weight, time dependency weight, and process dimension weight) to 1 / 3. Based on the shared business primary key of each insert type pair, the corresponding dependency dimensions are matched, and the corresponding dimension weights and dependency strength components are determined. The comprehensive dependency strength is then calculated by weighted summation. The comprehensive dependency strength is used to determine the element values ​​of the insert type pair in the initial matrix, constructing a conflict matrix. An N×N matrix is ​​initialized (N = number of inserts). For each insert pair (i, j), if i = j, then M[i][j] = 0, indicating self-dependency is ignored. If the shared entity is an empty set, M[i][j] = 0, indicating no shared entity and no conflict. Otherwise, M[i][j] = comprehensive dependency strength, with values ​​∈ [0, 1]. Output the conflict matrix M, where M[i][j] represents the dependency strength between insertion i and insertion j.

[0032] Compared to conventional methods, this technical solution constructs a conflict matrix through coupled analysis of business entity sharing and real-time business metrics, fundamentally solving the core problem of inaccurately identifying conflicting order insertions. It identifies genuine resource competition relationships based on the intersection of business primary keys and captures the essence of conflicts through quantitative analysis of dependent dimensions (resources, time, and processes), ensuring the construction of the conflict matrix is ​​grounded in the objective fact of shared business entities. Simultaneously, the order insertion type serves as the semantic basis for the matrix's rows and columns, ensuring deep coupling between conflict intensity calculation and business scenarios. This avoids the vicious cycle of false conflict misjudgment and missed implicit conflict detection in existing technologies, providing accurate input for subsequent conflict resolution strategies and improving the scientific nature and production efficiency of production line resource scheduling.

[0033] Step S102: Determine the order insertion conflict pairs with resource conflicts based on the conflict matrix, and obtain the real-time resource indicator data of the order insertion conflict pairs in the real-time business system based on the order insertion conflict pairs.

[0034] Based on the conflict matrix, conflicting order pairs with resource conflicts are identified, specifically through the following steps: First, the preset matrix elements of the conflict matrix are traversed, and order pairs with element values ​​greater than zero are filtered to construct a candidate conflict pair set, and the shared business primary key of each candidate conflict pair is determined. The preset matrix elements typically refer to all elements in the upper or lower triangular region of the conflict matrix. For each accessed element M[i][j], its element value is checked to see if it is greater than zero. If it is, it indicates that there is a comprehensive dependency strength between order requests i and j, i.e., a possibility of conflict exists. The candidate conflict pair set is determined for all order request pairs (i, j) corresponding to elements that meet this condition, excluding order pairs with no dependency relationship M[i][j]=0, thus narrowing the scope of subsequent calculations. Simultaneously, during the construction of this set, the shared business primary key of each candidate conflict pair is extracted or associated based on the index. For each candidate order pair (i, j), a set intersection operation is performed, and the shared entity set S is obtained. ij =Primary key list(i)∩Primary key list(j), the intersection is only performed within the same primary key type, such as comparing product ID with product ID, and not comparing across types.

[0035] Then, based on the shared business primary key of each candidate conflict pair, a conflict correction index and a scenario sensitivity index are determined for each candidate conflict pair. The conflict correction index is used to mark whether the conflict pair needs correction, and the scenario sensitivity index reflects the sensitivity of the conflict scenario. Specifically, when the shared business primary key is only the customer primary key and / or the pre-sale activity primary key, the conflict correction index of the candidate conflict pair is set to 0; otherwise, it is set to 1. If the shared business primary key is only the customer primary key and / or the pre-sale activity primary key, and does not contain any production resource-related primary keys (such as raw materials, processes associated with specific production line equipment, etc.), it is determined that the conflict mainly stems from business-level associations rather than direct competition for physical capacity or materials. Therefore, its conflict correction index is set to 0, indicating that the conflict may be the lowest under the current production resource scheduling scenario, and this conflict intensity needs to be corrected to no substantial resource conflict. If the shared business primary key contains any primary key directly related to production resources (such as raw material primary keys, process primary keys associated with specific bottleneck equipment), the conflict correction index is set to a baseline value greater than zero, for example, set to 1.

[0036] In one embodiment, the threshold used to filter conflict pairs is a dynamic threshold, dynamically calculated based on the business context of the current conflict scenario, ensuring that more stringent conflict judgment criteria are applied to highly sensitive scenarios (such as VIP customers and best-selling products). First, the scenario sensitivity index reflects the sensitivity of the conflict scenario, based on the shared entity set S. ij If the customer ID is in the database, then query the CRM system to obtain the customer level L_c corresponding to that customer ID (with a value of 0–100). Let the customer sensitivity factor = L_c / 100; based on the shared entity set S... ijIf the customer ID and product ID are in the data, then query the sales system to obtain the sales growth rate G_s ((this week's sales - last week's sales) / last week's sales) for the product ID in the past 7 days. Let the product sensitivity factor = min(G_s, 1.0), and take the maximum value between the customer sensitivity factor and the product sensitivity factor to determine the scenario sensitivity index of the candidate conflict pair. It should be noted that the scenario sensitivity index belongs to [0,1]. The larger the value, the more sensitive the scenario is, and the lower the dependency strength is required to trigger the conflict judgment.

[0037] In the historical conflict case database, historical conflict events with the same primary key as the shared business are filtered out to calculate the mean dependency strength μ of the historical conflict event. Using the scenario sensitivity index and the mean dependency strength, the dynamic threshold corresponding to each candidate conflict pair is calculated. Specifically, the dynamic threshold formula is T=μ×(1-scenario sensitivity index). It should be noted that T should be constrained to [0.3, 0.8] to prevent misjudgment due to extreme values. The hard boundary is set by the company's operational experience, but the boundary value itself does not participate in the threshold calculation logic.

[0038] Finally, the modified dependency strength of each candidate conflict pair is determined by combining the conflict correction index with the comprehensive dependency strength of each candidate conflict pair in the conflict matrix. In other words, the modified dependency strength of the candidate conflict pair is determined by multiplying the conflict correction index with the element value of the candidate conflict pair in the conflict matrix. Based on the modified dependency strength of each candidate conflict pair and the corresponding dynamic threshold, the single-entry conflict pairs with resource conflicts are identified. That is, the candidate conflict pairs with modified dependency strength greater than the corresponding dynamic threshold are selected as single-entry conflict pairs with resource conflicts.

[0039] By coordinating conflict correction metrics, scenario sensitivity metrics, and dynamic thresholds, the problem of inaccurately identifying conflicting order insertions is fundamentally solved. Conflict correction and scenario sensitivity metrics are dynamically set by sharing business primary key types, triggering correction only when product / material / batch primary keys are shared, and setting sensitivity to 0 when only customer / activity primary keys are shared, automatically filtering out pseudo-conflicts. Simultaneously, the calculation of dynamic thresholds ensures conflict judgments are adapted to business scenarios, avoiding strategy bias caused by fixed thresholds. Correcting dependency strength accurately quantifies the true conflict intensity, transforming conflict identification from the surface characteristics of matrix element values ​​to the business logic of resource competition inherent in shared business primary keys, effectively eliminating false conflict misjudgments and missed implicit conflicts, providing accurate input for subsequent conflict resolution strategies, and ensuring that production line resource scheduling is based on real business competition relationships.

[0040] Based on the order insertion conflict pair, real-time resource indicator data for the order insertion conflict pair is obtained in the real-time business system. Specifically, this includes: obtaining the real-time business primary key parameters for each conflicting order in the real-time business system based on the business primary key of each conflicting order in the order insertion conflict pair; the business primary key refers to the code that uniquely identifies a business entity in the system, including the apparel product primary key, such as DRESS-001, used to associate product attributes; the fabric and accessories primary key, such as FAB-001, used to associate material attributes; the customer primary key, such as VIP-001, used to associate customer level; the production batch primary key, such as B-20250601, used to associate production batch information; and the pre-sale activity primary key, such as 618, used to associate promotional activities. The real-time business primary key parameters for each conflicting order are obtained in the real-time business system, which refers to various professional databases that are updated synchronously with the enterprise's operational status, such as ERP systems, supply chain management systems, CRM systems, and production management systems. The data is obtained in real time by querying the databases of various business systems (such as the product master data table of the ERP system, the raw material inventory table of the supply chain system, the customer level table of the CRM system, and the batch information table of the production system) through the API interface. The parameters include real-time business data such as product unit price, real-time inventory of raw materials, customer level, production line load rate, and maximum production capacity of the production line, to ensure data timeliness.

[0041] By using the real-time business primary key parameters of each conflicting order, the real-time resource indicator data for that conflicting order pair is determined. This real-time resource indicator data includes one or more of the following: product business value indicator, production line load rate indicator, unit capacity demand indicator, and inventory ratio indicator. The product business value indicator reflects the quantitative value of the conflicting order; the production line load rate indicator reflects the quantitative status of the current production line load; the unit capacity demand indicator reflects the quantitative intensity of the capacity demand from the conflicting order; and the inventory ratio indicator reflects the quantitative degree of material shortage.

[0042] Specifically, by using the primary key of the clothing product for each conflicting order, the corresponding product unit price is obtained. The product business value index is determined by comparing the product unit price with the pre-obtained industry average price. For example, by querying the product master data table through the ERP system, the current unit price of DRESS-001 is obtained as 1200 yuan. The industry average price is a benchmark value calculated by the system based on historical data, such as 800 yuan. The product business value index is calculated as the ratio of the product unit price to the industry average price, reflecting the relative level of the commercial value of the order, such as 1200 / 800=1.5. Using the primary key of the fabric and accessories, obtain the corresponding real-time inventory level. Based on this real-time inventory level and a pre-defined safety stock threshold, determine the inventory ratio indicator. For example, by querying the fabric and accessories inventory table through the supply chain system, obtain the current inventory level of FAB-001 as 500 units. The safety stock threshold is a baseline value set by the system based on material characteristics and historical consumption, such as 300 units. The inventory ratio indicator is calculated as the ratio of the real-time inventory level to the safety stock threshold, reflecting the degree of material shortage, such as 500 / 300≈1.67. Using the primary key of the production line, obtain the corresponding real-time production line load rate and maximum production line capacity. Based on the ratio of the order demand for each conflicting order to the maximum production line capacity, determine the unit capacity demand indicator. Use the real-time production line load rate to determine the production line load rate indicator. For example, by querying the batch information table through the production system, we can obtain information such as the current production line load rate being 95%, the maximum production line capacity being 1000 units / day, and an order A requiring 100 units. The unit capacity demand index is calculated as the ratio of the order demand to the maximum production line capacity, reflecting the intensity of the order's occupancy on the production line. For example, 100 / 1000 = 0.1. The production line load rate index directly uses the 95% obtained from the query as the index value. This process ensures that each real-time resource index is calculated based on real-time data from the business system, and that the index definition is deeply coupled with the business scenario. The product business value index reflects commercial value; high-value orders, such as repeat orders of best-selling items, have high indexes. The inventory ratio index reflects the degree of material scarcity; a low inventory ratio indicates a shortage. The unit capacity demand index reflects the intensity of capacity occupancy; a high demand ratio indicates high occupancy. The production line load rate index reflects the current production line load; a high load rate indicates tight capacity. Following the above method, we obtain real-time resource index data for the order conflict pair.

[0043] Compared to conventional methods, this technical solution obtains resource metrics by associating them with the real-time business system through business primary keys, fundamentally solving the problem of inaccurate resource metrics caused by differences in order types. Conventional methods rely solely on order amount or customer level for metric calculation, failing to distinguish the semantics of business scenarios. For example, they confuse the high-value metrics of VIP customer orders with the high-time-sensitivity value of repeat orders for popular items, leading to distorted product business value metrics. Alternatively, they use only static inventory thresholds to judge material shortages, ignoring real-time dynamic changes in inventory, making inventory ratio metrics unable to reflect the true degree of shortage. In contrast, this solution automatically associates business metrics with the business system through business primary keys, ensuring that each metric is calculated based on real-time business data, thus accurately reflecting the product business value. The system uses a dynamic ratio of product unit price to industry average price to reflect commercial value, avoiding deviations caused by fixed amounts. The inventory ratio indicator accurately captures material shortage status based on the real-time ratio of real-time inventory to safety stock threshold. The unit capacity demand indicator objectively quantifies capacity occupancy intensity by using the real-time ratio of order insertion demand to the production line's maximum capacity. The production line load rate indicator directly uses real-time system query values ​​to reflect the actual load on the production line. This real-time data acquisition mechanism based on business primary keys ensures that resource indicator calculations are based on the objective facts of business entity relationships, rather than subjective settings, ensuring deep coupling between indicators and business scenarios, and significantly improving the scientific nature of production line resource scheduling and the stability of production plans.

[0044] Step S103: Based on real-time resource indicator data, match the conflict resolution strategy for the insertion conflict pair according to the preset triggering conditions of the basic action element.

[0045] The basic action element includes preemption action element, extension action element, and segmentation action element.

[0046] In one embodiment of this specification, firstly, atomic operation base action elements for combining into a resolution strategy are defined, and trigger conditions deeply bound to business logic are set for three types of core action elements: preemption action elements, extension action elements, and splitting action elements. Firstly, the preemption action element is defined as a scheduling operation that allows an order insertion request to preferentially occupy a specific resource time window or material quota originally allocated to another request; the extension action element is defined as delaying the entire production plan of the current order insertion request to avoid the current resource bottleneck period; the splitting action element is defined as decomposing a complete order insertion request's production task into multiple sub-batches, which are then arranged in parallel or interleaved to different production resources or time windows.

[0047] The triggering conditions for the preemptive action element are related to the current order type, product business value index, and production line load rate index. The preemptive action element must meet the following conditions: the product business value index is higher than a preset business value threshold, and the production line load rate is lower than a preset load rate threshold. It should be noted that when the current order type is a best-selling reorder or VIP add-on order, the business value threshold corresponding to the preemptive action element is lower than the business value threshold corresponding to other order types. In other words, a lower business value threshold is specifically set for best-selling reorders or VIP add-on orders. For best-selling reorders or VIP add-on orders, even if their product business value index is not very high, as long as this lower exclusive threshold is reached, and the target production line's production line load rate index is lower than the preset load rate threshold, the preemptive action element is triggered, reflecting a high degree of protection for the timeliness and customer relationship of such orders. In one example, the business value threshold corresponding to other order types can be set to 1.3, the business value threshold corresponding to best-selling reorders or VIP add-on orders can be set to 1.15, and the preset load rate threshold can be set to 90%.

[0048] The triggering conditions for the extended action element are related to the current order type and the inventory ratio indicator. The extended action element must meet the condition that the inventory ratio indicator is lower than a preset inventory ratio threshold. When the current order type is an inventory shortage order, the inventory ratio threshold for the extended action element is higher than the inventory ratio threshold corresponding to other order types. For orders triggered by material shortages, a more cautious strategy is adopted. Only when the material's inventory ratio indicator is higher than a higher threshold (i.e., inventory is relatively more abundant) is the relatively mild strategy of extension considered; conversely, if the inventory ratio is very low, extension may be ineffective, and other strategies need to be considered, reflecting the special handling logic for orders with shortage risk. In one example, the inventory ratio threshold corresponding to other order types can be set to 0.5, and the inventory ratio threshold corresponding to an inventory shortage order can be set to 0.65.

[0049] The triggering condition for this splitting action element is related to the current order type and the unit capacity demand indicator. The splitting action element must meet the requirement that the unit capacity demand indicator is higher than the preset unit capacity demand threshold. When the current order type is a quick-response follow-up order, the unit capacity demand threshold of the splitting action element is lower than the unit capacity demand threshold corresponding to other order types. For quick-response follow-up orders that pursue extreme response speed, the threshold for triggering the splitting strategy is lowered. Even if the order's theoretical occupancy ratio on a single production line is not high (below the lower dedicated threshold), splitting may still be triggered. The purpose is to make the most of the scattered and fragmented capacity gaps to start and complete part of the production as quickly as possible, thereby achieving rapid market response. In one example, the unit capacity demand threshold corresponding to other order types can be set to 0.5, and the unit capacity demand threshold corresponding to quick-response follow-up orders can be set to 0.35.

[0050] Then, based on the real-time resource indicator data, and according to the preset triggering conditions of the basic action elements, the conflict resolution strategy for the order insertion conflict pair is matched. Specifically, this is achieved as follows: Based on the real-time resource indicator data and the triggering conditions of the basic action elements, the initial action element set corresponding to each conflicting order in the order insertion conflict pair is determined. The initial action element set is the set of all basic action elements that simultaneously satisfy their respective triggering conditions. For example, if a conflicting order has a high product business value indicator and a low production line load rate, it includes a preemption action element; if the inventory ratio indicator is low, it includes an extension action element; if the unit capacity demand indicator is high, it includes a splitting action element. In one example, each conflicting order may correspond to one or more initial action elements. If each conflicting order corresponds to one initial action element, then this initial action element is used as the conflict resolution action for this order.

[0051] If multiple initial action elements exist for each conflicting order, the action elements are filtered as follows: Obtain the customer primary key parameter and pre-sale activity primary key parameter for each conflicting order. Using the customer primary key parameter, determine the customer level of the conflicting order. The customer level can be a level coefficient or a grade. Using the pre-sale activity primary key parameter and the current timestamp, determine the time urgency coefficient of the conflicting order. Obtain the pre-sale committed delivery time and pre-sale start time corresponding to the pre-sale activity primary key parameter. Calculate the delivery duration period using the relative difference between the pre-sale committed delivery time and the current timestamp. Determine the pre-sale period using the pre-sale committed delivery time and the pre-sale start time. Determine the time urgency coefficient using the ratio of the delivery duration period to the pre-sale period. It should be noted that, in the absence of a pre-sale activity primary key parameter, the time urgency coefficient is calculated based on the required delivery time and the current time. For example, the delivery duration period is calculated using the relative difference between the required delivery time and the current timestamp. The reserved delivery period is determined using the required delivery time and the order placement time. The time urgency coefficient is determined using the ratio of the delivery duration period to the reserved delivery period.

[0052] Using customer level and time urgency coefficient as constraints, the initial action element set is dynamically filtered to determine the action element set for each conflicting order. Customer level is the constraint for preemptive action elements, and time urgency coefficient is the constraint for extended action elements. Here, the preemptive action element for two conflicting orders can be selected by comparing their customer levels, and the extended action element for two conflicting orders can be selected by comparing their time urgency coefficients. For example, in the initial action element set, order A = {preemption, segmentation}, and order B = {preemption, extension}. If the customer level of order A is lower than that of order B, the preemption action element for order A is filtered out; that is, the segmentation action element for order A is selected. Then, the time urgency coefficients of order A and order B are compared. If the time urgency coefficient of order B is greater than that of order A, the extended action element for order B is filtered out. Thus, the action element set for order A = {segmentation} and order B = {preemption} is obtained.

[0053] In addition to the methods described above, customer-level thresholds and time urgency coefficient thresholds can also be set. For example, in the above example, assuming the time urgency coefficient of order B is less than that of order A, the extended action elements of order A should be filtered. However, since order A has no extended action elements, a threshold can be used to filter the extended action elements of orders with a time urgency coefficient not less than the preset urgency coefficient threshold. If, after filtering, an order still has multiple action elements, each action element is treated as a solution. The time, cost, and other changes after resolving each solution are simulated and quantified to select the optimal solution. Based on the action element set of each conflicting order, at least one conflict resolution strategy for the conflict pair is determined, where each conflict resolution strategy includes at least one basic action element.

[0054] In one embodiment of this specification, after the conflict resolution strategy matching is completed, the complete decision reference chain is presented to the scheduler in a structured and visual form through the scheduling decision interface.

[0055] Taking the conflict between a clothing company's handling of a popular repeat order A (adding 5,000 cloud skirts) and a quick-response repeat order B (adding 3,000 warm sweatshirts) during the Double 11 promotion as an example, the conflict was identified as resource competition between the two due to their shared core production line LINE-03 and scarce fabric FAB-001. Based on real-time resource indicators, an initial strategy was matched, triggering a preemptive action element for the popular repeat order A (prioritizing production line occupancy) and triggering a splitting action element for the quick-response repeat order B (splitting it into multiple batches for flexible production).

[0056] The interface synchronously displays four-dimensional reference information, including a conflict pair details area, a strategy logic explanation area, an impact simulation and prediction area, and an interactive adjustment area. The conflict pair details area clearly indicates the order type, shared business primary key, and real-time indicator trends. The strategy logic explanation area uses natural language to explain the decision-making basis; for example, repeat orders for popular items are time-sensitive businesses, and the system triggers preemption based on their high business value indicators; rapid response orders require quick action but have high capacity load, and splitting can reduce the impact of a single batch. The impact simulation and prediction area dynamically presents multi-dimensional estimated effects after strategy execution, such as changes in delivery time, production line changeover frequency, and material consumption fluctuations. The interactive adjustment area provides sliders to adjust the number of split batches, drop-down menus to switch action element combinations, and checkboxes to add on-site constraints, such as production line LINE-03 requiring maintenance after 14:00 today.

[0057] The dispatcher first refers to the strategy logic description area, combined with the company's business priorities for the day, such as confirming the market window priority for repeat orders of best-selling items, to verify the rationality of the system's judgment; then considers the impact prediction, assessing the potential impact of the strategy on the overall production rhythm, such as determining whether the delay is within the customer's tolerance threshold; when it is learned that the quick-response follow-up order B is related to strategic customers and the fabric inventory has been temporarily replenished, the dispatcher selects to enable the extended action element in the interaction area and fine-tunes the batch cutting, updating the simulation effect in real time. After the extended action element is activated, the material supply pressure is relieved, the batch cutting is optimized to 4 batches, and the delivery delay is significantly shortened, helping them to weigh the pros and cons.

[0058] Finally, the scheduler clicks "confirm execution," transforming the adjusted strategy into standardized scheduling instructions. This is automatically synchronized to the MES to update production line scheduling, pushed to the WMS to adjust material distribution, and generates an operation log containing adjustment details and decision-making basis for full traceability. By using structured reference content, the decision-making dilemma of relying on human experience in traditional models is resolved, while retaining the final discretion of humans in complex business scenarios. This allows conflict resolution to combine the accuracy of algorithms with the flexibility of humans, significantly improving the scientific nature, efficiency, and interpretability of scheduling decisions.

[0059] The at least one technical solution adopted in the embodiments of this specification can achieve the following beneficial effects: By deeply analyzing text semantics through a large model, extracting business primary keys and associating them with real-time business indicators, conflicts are identified based on business logic rather than surface features, ensuring that conflict pair screening is based on real resource competition relationships; a quantitative conflict matrix is ​​constructed, which is dynamically calculated based on shared business entities and multi-dimensional dependencies, thereby accurately identifying order insertion conflict pairs that truly have resource competition relationships and effectively filtering out pseudo-conflicts that only have superficial associations such as related customers, thus achieving accurate conflict identification; at the same time, the order insertion type is embedded as a core parameter in the construction of the conflict matrix and the triggering conditions of action elements, so that the conflict resolution strategy is deeply coupled with business objectives. In this system, for example, reorders of popular items automatically trigger a preemptive strategy due to high timeliness requirements, while orders due to inventory shortages are prioritized and extended due to material constraints, thus avoiding the defect of strategy being disconnected from business operations. Based on real-time resource indicators, basic action elements are triggered and matched. The generation of conflict resolution strategies does not apply fixed rules, but rather dynamically combines the most suitable optimization solutions based on the specific economic value of both parties in the conflict, the current degree of resource scarcity, and the real-time status of production capacity pressure. This elevates conflict handling from passive response to proactive optimization, eliminating false conflict misjudgments and hidden conflict omissions, significantly improving the scientific nature of production line resource scheduling and the stability of production plans, realizing the transformation from surface-characteristic conflicts to business logic conflicts, and providing technical support for flexible production in intelligent garment manufacturing.

[0060] This specification also provides a multi-type single-interpolation conflict resolution device based on a large model, as shown in FIG3. The device includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the above-described method.

[0061] This specification also provides a non-volatile computer storage medium storing computer-executable instructions configured to perform the above-described method.

[0062] The above are merely one or more embodiments of this specification and are not intended to limit this specification. Various modifications and variations can be made to the one or more embodiments of this specification by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of one or more embodiments of this specification should be included within the scope of the claims of this specification.

Claims

1. A method for resolving multiple types of single-entry conflicts based on a large model, characterized in that, The method includes: acquiring multi-source order insertion requests; performing semantic parsing on each order insertion request using a pre-trained large-scale apparel industry model to determine the current order insertion type corresponding to each request; performing coupled analysis of business entity sharing and real-time business indicators based on multiple current order insertion types to determine the corresponding conflict matrix, wherein the order insertion types include popular item reorders, inventory shortage orders, quality inspection remedial orders, VIP order additions, and quick-response follow-up orders; identifying order insertion conflict pairs with resource conflicts based on the conflict matrix; acquiring real-time resource indicator data of the order insertion conflict pairs in the real-time business system based on the order insertion conflict pairs; and matching the conflict resolution strategy of the order insertion conflict pairs according to the triggering conditions of preset basic action elements based on the real-time resource indicator data, wherein the basic action elements include preemption action elements, extension action elements, and segmentation action elements.

2. The multi-type insertion conflict resolution method based on a large model according to claim 1, characterized in that, The system uses a pre-trained large-scale model of the apparel industry to perform semantic parsing on each order insertion request, determining the current order insertion type and corresponding conflict matrix for each request. Specifically, this includes: inputting the text description of the order insertion request into the pre-trained large-scale model of the apparel industry; performing word segmentation and entity annotation on the text description to determine multiple business primary keys, including apparel product primary keys, fabric and accessories primary keys, customer primary keys, production line primary keys, and pre-sale activity primary keys; querying the business database in real time using the large-scale model of the apparel industry to obtain the business indicators corresponding to the business primary keys, and determining the business indicator vector; concatenating the original semantic text encoding corresponding to the text description and the business indicator vector through a cross-modal fusion layer to determine the joint vector corresponding to the order insertion request; outputting the order insertion type probability distribution corresponding to the order insertion request using the large-scale model of the apparel industry and the joint vector; selecting the category corresponding to the maximum value in the order insertion type probability distribution to determine the current order insertion type corresponding to the order insertion request; and performing coupling analysis of business entity sharing and real-time business indicators based on the business primary key and business indicator vector corresponding to each order insertion request to determine the conflict matrix.

3. The multi-type insertion conflict resolution method based on a large model according to claim 2, characterized in that, Based on the business primary key and business indicator vector corresponding to each order insertion request, a coupling analysis of business entity sharing and real-time business indicators is performed to determine a conflict matrix. Specifically, this includes: constructing an initial matrix of order corresponding to the number of current order insertion types, where each row and column of the initial matrix corresponds to an order insertion type pair; obtaining the business primary key corresponding to each order insertion type in each order insertion type pair, taking the intersection of the business primary keys of the order insertion type pairs to determine the shared business primary key of each order insertion type pair; when the shared business primary key is not null, matching the corresponding dependency dimension based on the shared business primary key, and calculating the dependency strength component corresponding to the dependency dimension based on the business indicator vector, where the dependency dimension includes resource dependency, time dependency, and process dimension; determining the comprehensive dependency strength of each order insertion type pair through the dependency strength component, and determining the element values ​​of the order insertion type pair in the initial matrix based on the comprehensive dependency strength to construct the conflict matrix.

4. The multi-type insertion conflict resolution method based on a large model according to claim 1, characterized in that, The determination of order insertion conflict pairs with resource conflicts based on the conflict matrix specifically includes: traversing the preset matrix elements of the conflict matrix, filtering order insertion pairs with element values ​​greater than zero to construct a candidate conflict pair set, and determining the shared business primary key of each candidate conflict pair; determining the conflict correction index and scenario sensitivity index of each candidate conflict pair based on the shared business primary key, wherein when the shared business primary key is only the customer primary key and / or the pre-sale activity primary key, the conflict correction index of the candidate conflict pair is set to 0, otherwise it is set to 1; filtering historical conflict events with the same shared business primary key in the historical conflict case database to calculate the average dependency strength of the historical conflict events, and calculating the dynamic threshold corresponding to each candidate conflict pair using the scenario sensitivity index and the average dependency strength; determining the corrected dependency strength of the candidate conflict pair using the conflict correction index and the comprehensive dependency strength of each candidate conflict pair, so as to determine the order insertion conflict pairs with resource conflicts in the candidate conflict pair set based on the corrected dependency strength and the corresponding dynamic threshold of each candidate conflict pair.

5. The multi-type insertion conflict resolution method based on a large model according to claim 1, characterized in that, Based on the order insertion conflict pair, real-time resource indicator data of the order insertion conflict pair is obtained in the real-time business system. Specifically, this includes: obtaining the real-time business primary key parameter of each conflicting order in the real-time business system based on the business primary key of each conflicting order; determining the real-time resource indicator data of the order insertion conflict pair through the real-time business primary key parameter of each conflicting order, wherein the real-time resource indicator data includes any one or more of the following: product business value indicator, production line load rate indicator, unit capacity demand indicator, and inventory ratio indicator.

6. The multi-type insertion conflict resolution method based on a large model according to claim 5, characterized in that, By using the real-time business primary key parameters of each conflicting order, the real-time resource indicator data of the conflicting order pair is determined. Specifically, this includes: obtaining the corresponding product unit price using the apparel product primary key of each conflicting order; determining the product business value indicator using the product unit price and the pre-obtained industry average product price; obtaining the corresponding real-time inventory quantity using the fabric and accessories primary key; determining the inventory ratio indicator using the real-time inventory quantity and the pre-obtained safety stock threshold; obtaining the corresponding real-time production line load rate and maximum production line capacity using the production line primary key; determining the unit capacity demand indicator using the ratio of the order demand quantity of each conflicting order to the maximum production line capacity; and determining the production line load rate indicator using the real-time production line load rate.

7. The multi-type insertion conflict resolution method based on a large model according to claim 5, characterized in that, Based on the real-time resource indicator data, and according to the preset triggering conditions of basic action elements, a conflict resolution strategy for the order insertion conflict pair is matched. Specifically, this includes: determining the initial action element set corresponding to each conflicting order in the order insertion conflict pair according to the real-time resource indicator data and the triggering conditions of the basic action elements; obtaining the customer primary key parameter and pre-sale activity primary key parameter of each conflicting order; determining the customer level of the conflicting order using the customer primary key parameter; determining the time urgency coefficient of the conflicting order using the pre-sale activity primary key parameter and the current timestamp; dynamically filtering the initial action element set using the customer level and the time urgency coefficient as constraints to determine the action element set for each conflicting order; and determining at least one conflict resolution strategy for the order insertion conflict pair through the action element set of each conflicting order, wherein the conflict resolution strategy includes at least one basic action element.

8. The multi-type insertion conflict resolution method based on a large model according to claim 1, characterized in that, The triggering conditions for the preemptive action element are related to the current order type, product business value indicators, and production line load rate indicators. Specifically, when the current order type is a best-selling repeat order or a VIP add-on order, the business value threshold corresponding to the preemptive action element is lower than the business value threshold corresponding to other order types. The triggering conditions for the extended action element are related to the current order type and inventory ratio indicators. Specifically, when the current order type is an inventory shortage order, the inventory ratio threshold of the extended action element is higher than the inventory ratio threshold corresponding to other order types. The triggering conditions for the segmentation action element are related to the current order type and unit capacity demand indicators. Specifically, when the current order type is a quick-response follow-up order, the unit capacity demand threshold of the segmentation action element is lower than the unit capacity demand threshold corresponding to other order types.

9. A multi-type insertion conflict resolution device based on a large model, characterized in that, The device includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method as described in any one of claims 1-8.

10. A non-volatile computer storage medium storing computer-executable instructions, characterized in that, The computer-executable instructions are configured to perform the method as described in any one of claims 1-8.

Images