Intelligent matching method and system based on purchase demand and execution, and storage medium

Abstract

The invention discloses an intelligent matching method and system based on purchase demands and execution and a storage medium, and relates to the technical field of information intelligent matching, and the method comprises the following steps: constructing a dynamic data association graph, and precisely synchronizing data in real time; constructing a global task scheduling model according to the synchronous data so as to dynamically allocate tasks based on loads; generating a demand list according to the synchronous data, and performing dynamic analysis and semantic completion on the demand list; continuously performing supplier dynamic matching and priority ranking according to a demand list dynamic analysis result; performing difference traceability and graded early warning aiming at the abnormity of the supplier demand matching process; dynamic indexes are monitored in real time, and automatic collaborative iteration is carried out. The technical problem that existing demand purchase matching is not high in intelligent degree is solved.

Description

Technical Field

[0001] This invention relates to the field of intelligent information matching technology, and more specifically, to an intelligent matching method, system, and storage medium based on procurement needs and execution. Background Technology

[0002] Currently, with the increasing complexity of enterprise procurement management needs, data-driven intelligent matching has become key to improving procurement efficiency. However, in procurement management scenarios, existing technologies have room for improvement at several levels. For example, in terms of data processing, demand orders, procurement records, and warehouse data are stored in different databases with low data synchronization frequency (e.g., hourly batch synchronization), which easily leads to information lag. On the other hand, repeated comparison of all data increases computing power consumption and makes it difficult to meet the requirements of real-time performance and efficiency.

[0003] Meanwhile, in the demand analysis stage, traditional vague expressions and implicit conditions place high demands on traditional rule engines, making it difficult to accurately extract key information and transform it into structured data. Furthermore, in the supplier matching process, the supplier evaluation dimensions are relatively singular, failing to fully consider dynamic factors such as real-time capacity and delivery flexibility, which affects the optimization effect of the matching results. In addition, in terms of anomaly handling, existing technologies can usually only indicate superficial differences such as quantity discrepancies, making it difficult to trace the root causes of differences such as demand changes, supplier omissions, or warehousing errors. This results in time-consuming and labor-intensive overall problem investigation, which is inefficient. Moreover, the early warning response mechanism is also lagging and fails to classify and handle anomalies of different severity levels, thus affecting the efficiency of problem resolution. Summary of the Invention

[0004] To address this, the present invention provides an intelligent matching method, system, and storage medium based on procurement needs and execution, thereby resolving the aforementioned technical problems in demand-based procurement matching in the prior art.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] A smart matching method based on procurement needs and execution includes the following steps:

[0007] Construct dynamic data association maps and synchronize data accurately in real time;

[0008] A global task scheduling model is built based on synchronized data to dynamically allocate tasks based on load;

[0009] A requirement order is generated based on the synchronized data, and the requirement order is dynamically parsed and semantically completed.

[0010] Based on the dynamic parsing results of the demand order, continue to perform dynamic matching and priority ranking of suppliers;

[0011] Implement differential source tracing and tiered early warning for anomalies in the supplier matching process;

[0012] It monitors dynamic indicators in real time and automatically iterates in a coordinated manner.

[0013] Based on the above technical solution, the present invention is further described as follows:

[0014] As a further aspect of the present invention

[0015] The construction of a dynamic data association map and the real-time accurate data synchronization specifically include:

[0016] A graph database is constructed, and the real-time relationships between "demand order - purchase order - supplier - warehouse - region" are stored in the graph database to form a dynamic data relationship graph. The demand order is used as the starting core node, and relationship edges are established with purchase orders, suppliers, warehouses and regions respectively. The edge connection represents the relationship between entities, such as the "creation relationship" between demand order and purchase order, the "supply relationship" between purchase order and supplier, etc. When a demand order involves a specific product category, the relevant supplier and its current availability information can be quickly located through the relationship edges based on the dynamic data relationship graph.

[0017] Configure a distributed data gateway, which is associated with and connected to different independent subsystems, including the demand management system, procurement system, warehousing system, and supplier system.

[0018] Collect multi-source heterogeneous data such as demand orders, purchase records, and warehouse data through a distributed data gateway;

[0019] Furthermore, the CDC (Change Data Capture) model is introduced, and the CDC model is used to capture changes in multi-source heterogeneous data in the databases of each independent subsystem in real time;

[0020] Specifically, the distributed data gateway monitors the binary log files of each independent subsystem database in real time based on the CDC model, and captures the data change events related to the log files, thereby triggering a real-time data synchronization mechanism to effectively reduce information lag and delay.

[0021] We continue to build incremental data identification rules based on real-time synchronized data information, so that each piece of data is given a unique identifier and version number. The CDC model only directly synchronizes data items whose version numbers have been upgraded. When the current data is first created, the version number is the initial value. When the data changes, the version number is automatically synchronized and upgraded. The CDC model only compares the current version number with the last synchronized version number. If the data version number has not changed, it is not transmitted synchronously. If the version number is upgraded, only the data items whose version number has been upgraded are transmitted synchronously.

[0022] As a further aspect of the present invention

[0023] The step of constructing a global task scheduling model based on synchronization data to dynamically allocate tasks based on load specifically includes:

[0024] A global task scheduling model is constructed, which loads the CPU and memory usage status information of cluster nodes (servers) in real time, and establishes a heartbeat monitoring mechanism with a frequency of once per second.

[0025] Nodes (servers) with a load rate of <30% are designated as idle nodes. The global task scheduling model automatically assigns a task package to an idle node based on the main task queue generated from the synchronized data. The mapping relationship between the task package and the corresponding node is recorded in real time in the "task-node" subgraph of the dynamic data association graph. When the node's parsing times out (>30 seconds) or fails, the subtask is automatically returned to the global task scheduler to be re-entered into the main task queue. At the same time, the "status" attribute of the corresponding task package node in the dynamic data association graph is updated. This ensures that the synchronized data can be reasonably distributed according to the node load, avoids data accumulation, reduces information lag, and guarantees the real-time efficiency of subsequent needs.

[0026] As a further aspect of the present invention

[0027] The process of generating a demand order based on synchronized data, and dynamically parsing and semantically completing the demand order, specifically includes:

[0028] A BERT pre-trained model is constructed, and a demand semantic understanding model is trained based on the BERT pre-trained model and a corpus of historical purchase orders. The demand semantic understanding model is input with demand order text generated by the CDC model synchronously, and the demand semantic understanding model parses and outputs explicit and implicit information in the demand order text. Explicit information includes clearly structured fields such as demand category, quantity, and region, while implicit information includes non-explicit intent fields. Combining current time series, historical collaboration, and special regulations data, an intent inference submodule is further constructed to complete the implicit information, and the overall parsing result after completion is verified for compliance. The expression for further verifying the compliance of the parsing is as follows:

[0029] S=α·P+β·H+γ·C

[0030] In the formula: S is the score for parsing the requirement form text; P is the score for the completeness of explicit information, which is based on the required and optional items in the standard requirement form and compared with the actual extracted explicit information. With 100 points as the benchmark, 30 points are deducted when a required item is missing and 10 points are deducted when an optional item is missing.

[0031] H represents the accuracy score for completing implicit information. This score is based on the current timeline, historical collaboration, and specially specified data, and is compared with the actual extracted implicit information. A manual verification process can be triggered afterward. The final score is based on 100 points. If there are missing or incorrect extracted items in the comparison results, 10 points will be deducted for each item.

[0032] C represents the supplier's real-time capacity adaptability score. Combining historical collaboration data with the supplier's real-time capacity data, the automatic splitting ratio of the demand quantity is optimized and calculated. That is, when the demand quantity exceeds the capacity of a single supplier, other suppliers are automatically introduced to split the demand quantity proportionally. The splitting is based on the current available capacity ratio of different suppliers.

[0033] The expression for calculating the supplier's real-time capacity adaptability score C is as follows:

[0034] Supplier real-time capacity adaptability score C = Total demand coverage score × 25% + Single supplier capacity adaptability score × 30% + Delivery time coordination score × 25% + Risk diversification score × 20%

[0035] Among them, the total demand coverage score means that the sum of the quantities undertaken by each supplier must equal the total demand, otherwise it is 0 points; the single supplier capacity matching score means that the quantity allocated to each supplier must be within its real-time adjustable capacity, otherwise it is 0 points; and the delivery time coordination score means that the delivery time of each supplier must fully meet the delivery time of the total demand, otherwise it is 0 points.

[0036] Risk diversification aims to avoid over-reliance on a single supplier. By diversifying the allocation, the overall risk caused by a sudden production failure of a particular supplier is reduced. The supplier with the largest allocation is used as the benchmark. If the largest supplier's allocation share is ≤60%, then 100 points are awarded; if the share is 60%-80%, then 50 points are awarded; if the share is >80%, then 0 points are awarded.

[0037] α, β, and γ are the weight coefficients of the corresponding terms, where α = 0.4, β = 0.3, and γ = 0.3.

[0038] When the text parsing score S of the demand order is greater than or equal to 80, it is considered a qualified demand parsing. Otherwise, the parsing will be readjusted for the explicit information completeness score P and / or implicit information completeness score H and / or supplier real-time capacity adaptability score C, which are not full marks.

[0039] As a further aspect of the present invention

[0040] The process of generating a demand order based on synchronized data, and dynamically parsing and semantically completing the demand order, specifically includes:

[0041] A dynamic graph embedding model is constructed to observe in real time whether there are areas with significant demand fluctuations, and these areas are mapped to the vertices of the graph structure of the dynamic graph embedding model. The static attribute of three-dimensional coordinate distance and the dynamic attribute of real-time capacity load rate are then embedded based on the dynamic graph embedding model. The quantification results of "static distance + dynamic load rate" are then fused with the parsing results of the demand semantic understanding model at a weight of 50% each, and the overall allocation ratio is then recalculated.

[0042] As a further aspect of the present invention

[0043] The step of dynamically matching and prioritizing suppliers based on the dynamic parsing results of the demand order specifically includes:

[0044] To avoid the drawbacks of traditional methods that only sort and match procurement needs and suppliers by time, a dynamic supplier capability matrix and a demand-supplier matching scoring model are constructed in a distributed manner. The dynamic supplier capability matrix updates the capabilities of matched suppliers in real time, and the demand-supplier matching scoring model sorts the matching degree. Finally, intelligent algorithms are used to dynamically optimize and ensure that high-value and urgent needs are prioritized.

[0045] Based on the supplier dynamic capability matrix, a dynamic profile package is constructed for each supplier. The contents of the dynamic profile package include:

[0046] Real-time supply S1: The quantity of goods that the supplier can immediately allocate;

[0047] Short-term supply S2: The quantity of goods that a supplier can quickly allocate within a specific time period;

[0048] Capacity elasticity coefficient E: reflects the supplier's ability to cope with demand fluctuations. E = (actual maximum capacity / conventional standard capacity) - 1. The larger the value, the better the capacity elasticity coefficient E.

[0049] Regional delivery cost D: The total cost for the supplier to deliver goods to the target area;

[0050] Cooperation fit F: Based on historical cooperation data, F = (Number of on-time deliveries / Total number of cooperations) × 0.6 + (Number of quality-compliant deliveries / Total number of cooperations) × 0.4;

[0051] Continuing with the demand-supplier matching scoring model, the following relationship is established:

[0052] M = M1 × 0.4 + M2 × 0.6

[0053] In the formula, M is the comprehensive matching score. The higher the comprehensive matching score, the higher the matching degree between the supplier and the demand.

[0054] Basic matching score M1: M1 = Category fit Q × 0.5 + Cooperation adaptability F × 0.5, where the category fit Q is 0 or 1 depending on the degree of matching between the supplier's category and the demand category;

[0055] Dynamic matching score M2: M2 = Supply capacity score A × 0.3 + Cost-effectiveness score B × 0.4 + Timeliness satisfaction score T × 0.2 + Capacity elasticity score E' × 0.1;

[0056] Supply capacity score A: A = (S1 / Demand quantity) × 50 + (S2 / Demand quantity) × 50. If the result exceeds 100, it is calculated as 100 points.

[0057] Cost-benefit analysis B: B = 100 - (Regional distribution cost D / Maximum acceptable cost) × 100;

[0058] Timeliness satisfaction score T: T = (actual delivery time / required delivery time) × 100. If the actual delivery time is longer than the required delivery time, then T = 0.

[0059] Capacity elasticity score E': E' = E × 100, converting the capacity elasticity coefficient into a score of 0-100, with any value exceeding 100 being counted as 100.

[0060] Therefore, through multi-dimensional calculation, a comprehensive matching score between suppliers and demands is obtained. The higher the comprehensive matching score, the higher the matching priority between suppliers and demands, and thus their priority is ranked.

[0061] To address real-time changes, a dynamic adjustment mechanism is further implemented as follows:

[0062] Supply adjustment coefficient K1: When the change rate of the supplier's real-time supply S1 exceeds 20%, K1 = adjusted S1 / original S1, and at this time the dynamic matching M2 is updated to M2 × K1;

[0063] Cost adjustment coefficient K2: When the change rate of regional delivery cost D exceeds 15%, K2 = D before adjustment / D after adjustment, and the dynamic matching M2 is updated to M2 × K2;

[0064] Emergency Request Adjustment: For emergency requests, the comprehensive matching score M needs to be multiplied by the urgency coefficient U. U ranges from 1.2 to 2.0 depending on the urgency of the request. The higher the urgency, the larger U is.

[0065] When a high-value order exceeds a preset threshold in the comprehensive matching score M calculation, the weight of the cooperation fit degree F is increased to 0.6, while the weight of the category fit degree Q is reduced to 0.4, and the capacity elasticity coefficient E is introduced for adjustment, i.e., M1 = (category fit degree Q × 0.4 + cooperation fit degree F × 0.6) × (1 + E × 0.2), in order to prioritize the stability of high-value orders and the ability to cope with fluctuations.

[0066] As a further aspect of the present invention

[0067] The aforementioned method for tracing discrepancies and providing tiered early warnings for anomalies in the supplier matching process specifically includes:

[0068] A difference feature library is constructed, and difference factor data for each key node in the entire procurement process is collected and quantified in real time. These key nodes include demand nodes, order nodes, production nodes, logistics nodes, and warehousing nodes. The calculation formulas and meanings of the difference factors corresponding to each key node are as follows:

[0069] ① Demand Node:

[0070] Requirement change frequency: Fv = number of changes / original planned cycle, used to measure the frequency of requirement changes;

[0071] Completeness of requirement description: Cd = number of complete items / total number of items × 100%, which reflects the level of detail in the requirement description. The difference value ΔCd is 100% - Cd.

[0072] ② Order Node:

[0073] Order response time: Tr = Supplier confirmation time - Order placement time, reflecting the supplier's response speed to the order;

[0074] Price deviation rate: Rp = (actual unit price - benchmark unit price) / benchmark unit price × 100%, which measures the degree of deviation between the actual purchase price and the benchmark price;

[0075] ③ Production nodes:

[0076] Production scheduling achievement rate: Pc = actual production scheduling output / planned production scheduling output × 100%, which is used to evaluate the completion of the production scheduling plan. The difference is 100% - Pc.

[0077] Material completeness rate: Cm = number of complete material items / total number of material items × 100%, which represents the completeness of the materials required for production. The difference value is 100% - Cm;

[0078] ④ Logistics nodes:

[0079] Transportation damage rate: Dt = number of damaged items / total transportation volume × 100%, which reflects the damage to goods during the logistics transportation process;

[0080] In-transit delay rate: Dd = (actual in-transit time - standard in-transit time) / standard in-transit time × 100%, which reflects the timeliness of logistics transportation;

[0081] ⑤ Inbound node:

[0082] Quantity matching rate: Qi = Actual quantity received / Order quantity × 100%, used to judge the consistency between the actual quantity received and the order quantity, and the difference value is 100% - Qi;

[0083] Quality pass rate: Qq = number of qualified products / number of samples × 100%, used to assess the quality compliance of goods entering the warehouse, and the difference is 100% - Qq;

[0084] Specification matching rate: Qs = number of matching specifications / total number of specifications × 100%. This checks the degree of matching between the specifications of the goods entering the warehouse and the required specifications. The difference value ΔQs is 100% - Qs.

[0085] Warehouse entry timeliness: Ti = actual entry time - planned entry time, which measures the timeliness of goods entering the warehouse;

[0086] Construct a propagation matrix and calculate the cumulative impact index;

[0087] Based on the difference factors extracted in the above steps, a propagation matrix Mij is further constructed. Mij is the influence coefficient of the i-th node on the j-th node. For example, M15 corresponds to the influence of ΔCd on ΔQs. Its initial value is determined by the correlation analysis of ΔCd and ΔQs, and Dd and Ti factors in historical collaborative data. For example, M15 = 0.4.

[0088] Obtain the node sensitivity Sj: It is obtained by training on the degree of influence of the historical data of the difference factors of other nodes on the difference factors of each node in the above steps. For example, the S5 of the inbound node is 0.9.

[0089] Calculate the propagation intensity Si at a single node:

[0090] Si = Σ(Difference value of node i × Mij × Sj)

[0091] For example, the propagation intensity from the demand node to the inbound node is S1 = ΔCd × M15 × S5;

[0092] Obtain the weight Wi of each node: Set the weight Wi according to the historical data influence range of the difference factor of each node in the above steps, and satisfy ΣWi=1, for example, W5=0.3 for the inbound node;

[0093] Further calculate the cumulative impact index Tc:

[0094] Tc=Σ(Si×Wi)

[0095] The cumulative impact index Tc is the core indicator for measuring the degree of impact of differences at each node on the entire procurement process. Its value reflects the severity of the impact of the difference, and it serves as a reference basis for tracing the source of the difference.

[0096] Generate real-time dynamic early warning thresholds, calculate early warning levels, and push them out, specifically as follows:

[0097] Directly associate with the historical data of the node difference factors extracted in the above steps, and take the 90th percentile value of the same type of differences in the past 12 months as the basic warning line T_base. The 90th percentile value means that in a set of data, 90% of the data is less than or equal to this value, and only 10% of the data is greater than this value;

[0098] Further determine the actual risk coefficient R of the current supply chain. The specific formula is as follows:

[0099] R = 1 + (0.4×Dd + 0.3×ΔCm + 0.3×Rp) / 100

[0100] The three node difference factors directly used in the formula include:

[0101] In-transit delay rate Dd: with a weight of 40%, used to reflect logistics risk;

[0102] Material completeness rate difference ΔCm: with a weight of 30%, used to reflect production risk;

[0103] Price deviation rate Rp: with a weight of 30%, used to reflect cost risk;

[0104] Then the calculation formula for the final warning threshold T_final is as follows:

[0105] T_final = T_base × R

[0106] When any of the difference values in the above steps ≥ T_final, or the cumulative impact index Tc ≥ T_final, then an alarm is immediately triggered, specifically including:

[0107] When any of the difference values in the above steps ≥ 1.5×T_final, and Tc ≥ 1.5×T_final, or any of the difference values in the above steps ≥ 2×T_final, or Tc ≥ 2×T_final, then a first-level alarm is triggered;

[0108] When T_final ≤ any of the difference values in the above steps < 1.5×T_final, and T_final ≤ Tc < 1.5×T_final, or any of the difference values in the above steps ≥ 1.5×T_final, but Tc < T_final, or Tc ≥ 1.5×T_final, but any of the difference values in the above steps < T_final, then a second-level alarm is triggered.

[0109] As a further solution of the present invention,

[0110] The real-time monitoring of dynamic indicators and automatic collaborative iteration specifically include:

[0111] Construct a three-dimensional dynamic index system of "efficiency-accuracy-timeliness";

[0112] The system automatically collects real-time data of the above indicators at regular intervals to form a dynamic trend chart, where the horizontal axis represents time and the vertical axis represents the indicator value. It also marks the baseline and actual data value lines and compares the average value of the indicators for the most recent three monitoring periods with the baseline value. By monitoring multiple data points, it avoids accidental fluctuations that could trigger erroneous operations.

[0113] The model warning process is triggered when the value falls below the threshold.

[0114] It is proposed that there is a normal fluctuation deviation. If the average value of the indicator in the window of the most recent 3 monitoring periods is within the normal fluctuation deviation compared with the benchmark value, then no warning will be triggered to avoid random fluctuations.

[0115] The above-mentioned trigger model early warning process automatically generates a fluctuation trajectory file and continuously records the average value of the next three windows. If it continues to remain within the normal fluctuation deviation range, it is marked as a stable fluctuation zone.

[0116] Based on the aforementioned trigger model warnings and stable fluctuation bands as optimization criteria, automatic collaborative iteration is performed to reduce the model warning trigger rate and update the normal fluctuation deviation values ​​according to changes in business scenarios.

[0117] An intelligent matching system based on the aforementioned intelligent matching method for procurement needs and execution, the intelligent matching system comprising:

[0118] The data synchronization module is used to build dynamic data association maps and synchronize data accurately in real time.

[0119] The task dynamic allocation module is used to build a global task scheduling model based on synchronization data, so as to dynamically allocate tasks based on load;

[0120] The dynamic demand parsing module is used to generate demand orders based on synchronized data, and to perform dynamic parsing and semantic completion on the demand orders.

[0121] The supplier dynamic matching module is used to continue dynamic matching and priority sorting of suppliers based on the dynamic parsing results of the demand order;

[0122] The discrepancy tracing and early warning module is used to trace discrepancies and issue tiered early warnings for anomalies in the supplier matching process.

[0123] The dynamic indicator monitoring and iteration module is used for real-time monitoring of dynamic indicators and automatic collaborative iteration.

[0124] A computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the method.

[0125] The present invention has the following beneficial effects:

[0126] This method and system can achieve real-time and accurate data synchronization by constructing a dynamic data association graph, collecting multi-source heterogeneous data through a distributed data gateway, capturing data changes in real time with the help of a CDC model, and reducing redundant transmission with incremental data identification rules, thereby improving data processing efficiency and high-concurrency support capabilities. Simultaneously, by constructing a global task scheduling model, it monitors node load in real time and dynamically allocates tasks, achieving efficient load-based task allocation and avoiding data accumulation and information lag. Furthermore, it constructs a demand semantic understanding model using a BERT pre-trained model, combines it with an intent reasoning submodule to complete implicit information and perform qualification verification, and integrates it with a dynamic graph embedding model. Real-time regional status effectively enables precise analysis and allocation ratio optimization of demand orders, making demand processing more aligned with actual scenarios. Furthermore, by constructing a dynamic supplier capability matrix and a demand-supplier matching scoring model, and employing a multi-dimensional scoring and dynamic adjustment mechanism, it achieves precise dynamic matching and reasonable priority ranking of suppliers, ensuring that high-value and urgent demands are met first, thus improving procurement efficiency. By constructing a difference feature library to quantify the difference factors at key nodes throughout the entire process, and combining this with a propagation matrix to calculate the cumulative impact index, dynamic early warning thresholds and tiered early warning mechanisms are set, enabling the traceability and tiered early warning of differences in the procurement process, thereby allowing for more timely risk response. Attached Figure Description

[0127] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. The structures, proportions, sizes, etc., drawn in this specification are only used to complement the content disclosed in the specification, so that those skilled in the art can understand and read them. Any modifications to the structure, changes in the proportional relationships, or adjustments to the size, without affecting the effects and objectives that the present invention can produce, should still fall within the scope of the technical content disclosed in the present invention.

[0128] Figure 1 This is a schematic diagram of the overall process of the intelligent matching method based on procurement needs and execution provided in an embodiment of the present invention.

[0129] Figure 2 This is a schematic diagram illustrating the architecture of an intelligent matching system based on procurement needs and execution, provided in an embodiment of the present invention.

[0130] Figure 3 This is a schematic diagram of the physical structure of an electronic device according to an embodiment of the present invention.

[0131] The attached diagram lists the components represented by each number as follows:

[0132] Data synchronization module 10; dynamic task allocation module 20; dynamic demand parsing module 30; dynamic supplier matching module 40; difference tracing and early warning module 50; dynamic indicator monitoring and iteration module 60;

[0133] Electronic device 70: processor 701, memory 702, internal bus 703. Detailed Implementation

[0134] The following specific embodiments illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0135] The terms "upper," "lower," "left," "right," and "middle" used in this specification are merely for clarity of description and are not intended to limit the scope of the invention. Any changes or adjustments to their relative relationships, without substantially altering the technical content, should also be considered within the scope of the invention.

[0136] like Figure 1 As shown, this embodiment of the invention provides a method for intelligent matching of procurement needs and execution. By constructing a dynamic data association graph, combining it with a distributed data gateway to collect multi-source heterogeneous data, and leveraging a CDC model to capture data changes in real time, coupled with incremental data identification rules to reduce redundant transmission, it achieves real-time and accurate data synchronization, improving data processing efficiency and high-concurrency support capabilities. Simultaneously, by constructing a global task scheduling model, it monitors node load in real time and dynamically allocates tasks, achieving efficient task allocation based on load, avoiding data accumulation and information lag. Furthermore, it constructs a demand semantic understanding model using a BERT pre-trained model, and combines this with an intent reasoning submodule to complete implicit information and perform qualification verification. By incorporating dynamic graphs into the model to integrate real-time regional status, the system effectively achieves precise analysis and allocation ratio optimization of demand orders, making demand processing more aligned with real-world scenarios. Furthermore, by constructing a dynamic supplier capability matrix and a demand-supplier matching scoring model, and employing multi-dimensional scoring and dynamic adjustment mechanisms, it achieves precise dynamic matching and reasonable priority ranking of suppliers, ensuring that high-value and urgent demands are met first, thus improving procurement efficiency. Moreover, by constructing a difference feature library to quantify difference factors at key nodes throughout the entire process, and combining this with a propagation matrix to calculate the cumulative impact index, dynamic early warning thresholds and tiered early warning mechanisms are set, enabling the tracing of differences and tiered early warnings in the procurement process, thereby allowing for more timely risk response. Specifically, the steps include the following:

[0137] S1: Construct a dynamic data association map and synchronize data accurately in real time;

[0138] The specific process is as follows:

[0139] A graph database is constructed, and the real-time relationships between "demand order - purchase order - supplier - warehouse - region" are stored in the graph database to form a dynamic data relationship graph. The demand order is used as the starting core node, and relationship edges are established with purchase orders, suppliers, warehouses and regions respectively. The edge connection represents the relationship between entities, such as the "creation relationship" between demand order and purchase order, the "supply relationship" between purchase order and supplier, etc. When a demand order involves a specific product category, the relevant supplier and its current availability information can be quickly located through the relationship edges based on the dynamic data relationship graph.

[0140] Configure a distributed data gateway, which is associated with and connected to different independent subsystems, including but not limited to the demand management system, procurement system, warehousing system, and supplier system.

[0141] The distributed data gateway collects heterogeneous data from multiple sources, including demand orders, purchase records, and warehouse data. Since the data involved in the procurement process is scattered across different databases (such as MySQL and Oracle) and business systems, the distributed data gateway effectively enables unified reception and forwarding of data requests from different independent subsystems. It can also convert different data formats, such as converting the JSON format of purchase orders to the XML format of the warehouse system. Furthermore, it can control data access permissions, such as restricting supplier systems to reading only purchase order data related to their own, thereby significantly improving reading efficiency, reducing redundant steps, and supporting high-concurrency scenarios through distributed multi-node deployment, avoiding information delays caused by single points of failure.

[0142] Furthermore, the CDC (Change Data Capture) model is introduced, and the CDC model is used to capture changes in multi-source heterogeneous data in the databases of each independent subsystem in real time;

[0143] Traditional data synchronization methods, such as full database scanning, consume a lot of computing power, and timed database scanning has data lag, making it difficult to obtain the latest data in a timely manner. Therefore, the CDC model can capture data change operations such as "addition" and "modification" in a more real-time and accurate manner, rather than directly connecting to the full data.

[0144] Specifically, the distributed data gateway monitors the binary log files (such as MySQL's binlog) of each independent subsystem database in real time based on the CDC model, and captures data change events related to the log files, thereby triggering a real-time data synchronization mechanism to effectively reduce information lag and delay.

[0145] For example, when the demand department creates a new purchase order, the CDC model captures the change of "new demand order" in real time and synchronizes it to the dynamic data association graph through the distributed data gateway, ensuring that the supplier matching module can immediately and accurately obtain the latest demand; when the warehousing system records "inbound quantity", the change data is synchronized to the difference feature library in real time through the distributed data gateway, thereby supporting subsequent shortage / surplus early warning.

[0146] As a preferred embodiment, in order to further reduce redundant data transmission, incremental data identification rules are built based on real-time synchronized data information to add a unique identifier (GUID) and version number to each piece of data. The CDC model only directly synchronizes data items whose version numbers have been upgraded or changed.

[0147] Specifically, when the current data is first created, the version number is an initial value. When the data changes, the version number is automatically updated and upgraded. The CDC model only compares the current version number with the version number of the last synchronization. If the data version number has not changed, it is not transmitted synchronously. If the version number is upgraded, only the data items with the upgraded version number are transmitted synchronously. Thus, based on the CDC model capturing data changes, the version number is further used to identify whether synchronization is needed, which significantly reduces the large amount of duplicate transmission caused by traditional full synchronization. This is especially useful in procurement scenarios involving a large amount of order, inventory, and supplier data. For example, repeatedly transmitting an unmodified purchase order would waste bandwidth and computing power, ultimately achieving "real-time, accurate, and low-redundancy" data synchronization.

[0148] S2: Construct a global task scheduling model based on synchronization data to dynamically allocate tasks based on load;

[0149] The specific process is as follows:

[0150] To improve data processing efficiency in a multi-node (server) cluster environment, a global task scheduling model is constructed. The CPU and memory usage status information of the cluster nodes (servers) are loaded in real time through the global task scheduling model, and a heartbeat monitoring mechanism with a frequency of once per second is established.

[0151] Nodes (servers) with a load rate of <30% are assumed to be idle nodes. The global task scheduling model automatically assigns a task package to an idle node based on the main task queue generated from the synchronized data. The mapping relationship between the task package and the corresponding node is recorded in real time in the "task-node" subgraph of the dynamic data association graph. When the node's parsing times out (>30 seconds) or fails, the subtask is automatically returned to the global task scheduler to be re-entered into the main task queue. At the same time, the "status" attribute of the corresponding task package node in the dynamic data association graph is updated. This ensures that the synchronized data can be reasonably distributed according to the node load, avoids data accumulation, reduces information lag, and ensures real-time and efficient operation for subsequent needs.

[0152] S3: Generate a requirement order based on the synchronized data, and perform dynamic parsing and semantic completion on the requirement order;

[0153] The specific process is as follows:

[0154] A BERT pre-trained model is constructed, and a demand semantic understanding model is trained based on the BERT pre-trained model and a corpus of historical purchase orders. The demand semantic understanding model is input with demand order text generated by the CDC model synchronous data, and the demand semantic understanding model parses and outputs explicit and implicit information in the demand order text. The explicit information includes explicit structured fields such as demand category, quantity, and region, while the implicit information includes non-explicit intent fields (such as the delivery time associated with "urgent" and the specific category completion of "seasonal fruits"). Combining the current time series, historical collaboration, and special regulations data, an intent reasoning submodule is further constructed to complete the implicit information, and the overall parsing result after completion is verified for compliance. For example, based on the current season and regional sales data, "urgent purchase" is parsed as "delivery time ≤ 72 hours", and "seasonal fruits" is completed as "watermelon, peach".

[0155] The expression for further verification of the analytical qualification is as follows:

[0156] S=α·P+β·H+γ·C

[0157] In the formula: S is the score for parsing the requirement form text; P is the score for the completeness of explicit information, which is based on the required and optional items in the standard requirement form and compared with the actual extracted explicit information. With 100 points as the benchmark, 30 points are deducted when a required item is missing and 10 points are deducted when an optional item is missing.

[0158] H represents the accuracy score for completing implicit information. This score is based on the current timeline, historical collaboration, and specially specified data, and is compared with the actual extracted implicit information. A manual verification process can be triggered afterward. The final score is based on 100 points. If there are missing or incorrect extracted items in the comparison results, 10 points will be deducted for each item.

[0159] C represents the supplier's real-time capacity adaptability score. Combining historical collaboration data with the supplier's real-time capacity data, the automatic splitting ratio of the demand quantity is optimized. That is, when the demand quantity exceeds the capacity of a single supplier, other suppliers are automatically introduced to split the demand quantity proportionally. The splitting is based on the current available capacity ratio of different suppliers. For example, when the delivery time is specified as 3 days and the delivery quantity is 100 units of a certain model of equipment, and the single supplier has an available capacity of 60 units in 3 days, and the second supplier has an available capacity of 30 units in 3 days, then a third supplier needs to be introduced. Let's assume that the third supplier has an available capacity of 30 units in 3 days. Then the automatic splitting ratio of the supplier is 2:1:1, that is, 50 units: 25 units: 25 units.

[0160] The expression for calculating the supplier's real-time capacity adaptability score C is as follows:

[0161] Supplier real-time capacity adaptability score C = Total demand coverage score × 25% + Single supplier capacity adaptability score × 30% + Delivery time coordination score × 25% + Risk diversification score × 20%

[0162] Among them, the total demand coverage score means that the sum of the quantities undertaken by each supplier must equal the total demand, otherwise it is 0 points; the single supplier capacity matching score means that the quantity allocated to each supplier must be within its real-time adjustable capacity, otherwise it is 0 points; and the delivery time coordination score means that the delivery time of each supplier must fully meet the delivery time of the total demand, otherwise it is 0 points.

[0163] Risk diversification aims to avoid over-reliance on a single supplier. By diversifying the allocation, the overall risk caused by a sudden production failure of a particular supplier is reduced. The supplier with the largest allocation is used as the benchmark. If the largest supplier's allocation share is ≤60%, then 100 points are awarded; if the share is 60%-80%, then 50 points are awarded; if the share is >80%, then 0 points are awarded.

[0164] α, β, and γ are the weight coefficients of the corresponding terms, where α = 0.4, β = 0.3, and γ = 0.3.

[0165] When the text parsing score S of the demand order is greater than or equal to 80 points, it is considered a qualified demand parsing. Otherwise, the parsing will be readjusted for the explicit information completeness score P and / or implicit information completeness score H and / or supplier real-time capacity adaptability score C, which are not full marks.

[0166] For example, in a real-world application scenario, a demand order requests the purchase of 100 units of a certain model of equipment. The demand semantic understanding model first extracts the equipment model and quantity from the explicit information. However, the demand order does not specify the delivery time. Therefore, the demand semantic understanding model automatically completes the delivery time to within 3 days based on historical collaboration data. It also optimizes the automatic splitting ratio of the demand quantity based on the current available capacity of different suppliers. For example, supplier A is allocated 70 units and supplier B is allocated 30 units. The above process ensures that the key information of the demand order can be accurately extracted and transformed into structured data. The original simple demand order with only "100 units X1 equipment" is transformed into clear structured data containing "model X1, quantity 100 units, delivery within 3 days, splitting A to supply 70 units / B to supply 30 units". This provides high-quality input for subsequent supplier matching and avoids problems such as stockouts and delays caused by incomplete information or unreasonable parsing.

[0167] As another preferred embodiment, a dynamic graph embedding model is constructed to observe in real time whether there are areas with significant demand fluctuations, and these areas are mapped to the graph structure vertices of the dynamic graph embedding model. The three-dimensional coordinate distance static attribute and the real-time capacity load rate dynamic attribute are then embedded based on the dynamic graph embedding model. The quantification results of "static distance + dynamic load rate" are then fused with the parsing results of the demand semantic understanding model according to a 50% weighting, and the overall allocation ratio is then recalculated.

[0168] For example, based on the analysis results of the demand semantic understanding model, supplier A allocates 80 units and supplier B allocates 20 units. At this time, the demand in a certain area suddenly increases by 30 units, and delivery is required within 3 days. According to the analysis results of the demand semantic understanding model, both suppliers A and B still have adjustable capacity. Therefore, the static attribute of the three-dimensional coordinate distance from the center of the demand area and the dynamic attribute of the real-time capacity load rate are further introduced. It is found that the static distance of supplier B is closer. Therefore, the new demand is preferentially allocated to supplier B, that is, supplier B is allocated 50 units. This is further combined with the analysis results of the demand semantic understanding model at a ratio of 8:2, each weighted at 50%, to obtain a new overall allocation ratio of 7:3. However, if the current load rate of suppliers A and B suddenly increases or their adjustable capacity cannot meet the new allocation demand, the allocation should be reduced or avoided. The quantitative results should continue to be calculated according to the current adjustable real-time capacity ratio of suppliers A and B. Alternatively, supplier C can be directly introduced, and supplier C allocates 30 units, resulting in a new overall allocation ratio of 8:2:3.

[0169] By using dynamic graph embedding models, the drawbacks of ignoring real-time regional conditions when parsing only the demand text using BERT are avoided. For example, sudden increases in supplier load or large distances affecting timeliness may occur. Dynamic graph embedding quantifies real-time status into data and further combines it with the results of demand text parsing, making the overall demand parsing more flexible and realistic, and reducing delays or resource waste caused by unexpected situations.

[0170] S4: Based on the dynamic parsing results of the demand order, continue to perform dynamic matching and priority ranking of suppliers;

[0171] The specific process is as follows:

[0172] To avoid the drawbacks of traditional methods that only sort and match procurement needs and suppliers by time, a dynamic supplier capability matrix and a demand-supplier matching scoring model are constructed in a distributed manner. The dynamic supplier capability matrix updates the capabilities of matched suppliers in real time, and the demand-supplier matching scoring model sorts the matching degree. Finally, intelligent algorithms are used to dynamically optimize and ensure that high-value and urgent needs are prioritized.

[0173] More specifically, a dynamic profile package is constructed for each supplier based on the supplier dynamic capability matrix. The contents of the dynamic profile package include:

[0174] Real-time supply S1: The quantity of goods that the supplier can immediately allocate;

[0175] Short-term supply S2: The quantity of goods that a supplier can quickly allocate within a specific time period;

[0176] Capacity elasticity coefficient E: reflects the supplier's ability to cope with demand fluctuations. E = (actual maximum capacity / conventional standard capacity) - 1. The larger the value, the better the capacity elasticity coefficient E.

[0177] Regional delivery cost D: The total cost for the supplier to deliver goods to the target area;

[0178] Cooperation fit F: Based on historical cooperation data, F = (Number of on-time deliveries / Total number of cooperations) × 0.6 + (Number of quality-compliant deliveries / Total number of cooperations) × 0.4;

[0179] Continuing with the demand-supplier matching scoring model, the following relationship is established:

[0180] M = M1 × 0.4 + M2 × 0.6

[0181] In the formula, M is the comprehensive matching score. The higher the comprehensive matching score, the higher the matching degree between the supplier and the demand.

[0182] Basic matching score M1: M1 = Category fit Q × 0.5 + Cooperation adaptability F × 0.5, where the category fit Q is 0 or 1 depending on the degree of matching between the supplier's category and the demand category;

[0183] Dynamic matching score M2: M2 = Supply capacity score A × 0.3 + Cost-effectiveness score B × 0.4 + Timeliness satisfaction score T × 0.2 + Capacity elasticity score E' × 0.1;

[0184] Supply capacity score A: A = (S1 / Demand quantity) × 50 + (S2 / Demand quantity) × 50. If the result exceeds 100, it is calculated as 100 points.

[0185] Cost-benefit analysis B: B = 100 - (Regional distribution cost D / Maximum acceptable cost) × 100;

[0186] Timeliness satisfaction score T: T = (actual delivery time / required delivery time) × 100. If the actual delivery time is longer than the required delivery time, then T = 0.

[0187] Capacity elasticity score E': E' = E × 100, converting the capacity elasticity coefficient into a score of 0-100, with any value exceeding 100 being counted as 100.

[0188] Therefore, through multi-dimensional calculation, a comprehensive matching score between suppliers and demands is obtained. The higher the comprehensive matching score, the higher the matching priority between suppliers and demands, and thus their priority is ranked.

[0189] To address real-time changes, a dynamic adjustment mechanism is further implemented as follows:

[0190] Supply adjustment coefficient K1: When the change rate of the supplier's real-time supply S1 exceeds 20%, K1 = adjusted S1 / original S1, and at this time the dynamic matching M2 is updated to M2 × K1;

[0191] Cost adjustment coefficient K2: When the change rate of regional delivery cost D exceeds 15%, K2 = D before adjustment / D after adjustment, and the dynamic matching M2 is updated to M2 × K2;

[0192] Emergency Request Adjustment: For emergency requests, the comprehensive matching score M needs to be multiplied by the urgency coefficient U. U ranges from 1.2 to 2.0 depending on the urgency of the request. The higher the urgency, the larger U is.

[0193] An optional implementation plan is as follows: when a high-value order exceeds a preset threshold in the comprehensive matching score M calculation, the weight of the cooperation fit degree F is increased to 0.6, while the weight of the category fit degree Q is reduced to 0.4, and the capacity elasticity coefficient E is introduced for adjustment, i.e., M1 = (category fit degree Q × 0.4 + cooperation fit degree F × 0.6) × (1 + E × 0.2), in order to prioritize the stability of high-value orders and the ability to cope with fluctuations;

[0194] The above solution introduces multiple quantitative formulas to make the supplier matching process more scientific and operable. The dynamic capability matrix can ensure the timeliness of data, and the dynamic adjustment mechanism can respond quickly to changes, effectively guaranteeing efficient matching in various scenarios and significantly improving procurement efficiency and satisfaction.

[0195] S5: Conduct differential source tracing and graded early warning for anomalies in the supplier matching process;

[0196] The specific process is as follows:

[0197] S501: Construct a difference feature library, and collect and quantify difference factor data for each key node in the entire procurement process in real time. The key nodes include demand nodes, order nodes, production nodes, logistics nodes, and warehousing nodes. The calculation formulas and meanings of the difference factors corresponding to each key node are as follows:

[0198] ① Demand Node:

[0199] Requirement change frequency: Fv = number of changes / original planned cycle, used to measure the frequency of requirement changes;

[0200] Completeness of requirement description: Cd = number of complete items / total number of items × 100%, which reflects the level of detail in the requirement description. The difference value ΔCd is 100% - Cd.

[0201] ② Order Node:

[0202] Order response time: Tr = Supplier confirmation time - Order placement time, reflecting the supplier's response speed to the order;

[0203] Price deviation rate: Rp = (actual unit price - benchmark unit price) / benchmark unit price × 100%, which measures the degree of deviation between the actual purchase price and the benchmark price;

[0204] ③ Production nodes:

[0205] Production scheduling achievement rate: Pc = actual production scheduling output / planned production scheduling output × 100%, which is used to evaluate the completion of the production scheduling plan. The difference is 100% - Pc.

[0206] Material completeness rate: Cm = number of complete material items / total number of material items × 100%, which represents the completeness of the materials required for production. The difference value is 100% - Cm;

[0207] ④ Logistics nodes:

[0208] Transportation damage rate: Dt = number of damaged items / total transportation volume × 100%, which reflects the damage to goods during the logistics transportation process;

[0209] In-transit delay rate: Dd = (actual in-transit time - standard in-transit time) / standard in-transit time × 100%, which reflects the timeliness of logistics transportation;

[0210] ⑤ Inbound node:

[0211] Quantity matching rate: Qi = Actual quantity received / Order quantity × 100%, used to judge the consistency between the actual quantity received and the order quantity, and the difference value is 100% - Qi;

[0212] Quality pass rate: Qq = number of qualified products / number of samples × 100%, used to assess the quality compliance of goods entering the warehouse, and the difference is 100% - Qq;

[0213] Specification matching rate: Qs = number of matching specifications / total number of specifications × 100%. This checks the degree of matching between the specifications of the goods entering the warehouse and the required specifications. The difference value ΔQs is 100% - Qs.

[0214] Warehouse entry timeliness: Ti = actual entry time - planned entry time, which measures the timeliness of goods entering the warehouse;

[0215] S502: Construct the propagation matrix and calculate the cumulative impact index;

[0216] Based on the difference factors extracted in step S501, a propagation matrix Mij is further constructed. Mij is the influence coefficient of the i-th node on the j-th node. For example, M15 corresponds to the influence of ΔCd on ΔQs. Its initial value is determined by the correlation analysis of ΔCd and ΔQs, and Dd and Ti factors in historical collaborative data. For example, M15 = 0.4.

[0217] Obtain the node sensitivity Sj: It is obtained by training on the degree of influence of the historical data of the difference factors of other nodes on the difference factors of each node in step S501. For example, the S5 of the inbound node is 0.9.

[0218] Calculate the propagation intensity Si at a single node:

[0219] Si = Σ(Difference value of node i × Mij × Sj)

[0220] For example, the propagation intensity from the demand node to the inbound node is S1 = ΔCd × M15 × S5;

[0221] Obtain the weight Wi of each node: Set the influence range of the historical data of the difference factor of each node in step S501, and satisfy ΣWi=1, for example, W5=0.3 for the inbound node;

[0222] Further calculate the cumulative impact index Tc:

[0223] Tc=Σ(Si×Wi)

[0224] The cumulative impact index Tc is the core indicator for measuring the degree of impact of differences at each node on the entire procurement process. Its value reflects the severity of the impact of the difference, and it serves as a reference basis for tracing the source of the difference.

[0225] S503: Generates real-time dynamic warning thresholds, calculates warning levels, and pushes them out. Specifically:

[0226] The historical data of the node difference factor extracted in step S501 are directly associated with the 90th percentile of the same type of difference in the past 12 months, which is used as the basic warning line T. The 90th percentile means that in a set of data, 90% of the data are less than or equal to the value, and only 10% of the data are greater than the value.

[0227] Taking "Quantity Difference Rate (ΔQi)" as an example: If the quantity difference rate data for the past 12 months are: 2, 3, 5, 5, 6, 7, 8, 9, 10, 12, 15, 20 (unit: %), after sorting these data from smallest to largest, the 90th percentile value is the 11th data (12 months × 90% = 10.8, rounded up to the 11th), which is 15%. Therefore, in the past 12 months, 90% of the quantity difference rates are ≤15%, and only 10% of the cases exceed 15%.

[0228] Further determine the actual risk coefficient R of the current supply chain, and the specific formula is as follows:

[0229] R = 1 + (0.4×Dd + 0.3×ΔCm + 0.3×Rp) / 100

[0230] The three node difference factors directly adopted in the formula include:

[0231] In-transit delay rate Dd: with a weight of 40%, used to reflect logistics risk;

[0232] Material completeness rate difference ΔCm: with a weight of 30%, used to reflect production risk;

[0233] Price deviation rate Rp: with a weight of 30%, used to reflect cost risk;

[0234] Then the calculation formula for the final warning threshold T_final is as follows:

[0235] T_final = T_base × R

[0236] When any difference value in step S501 ≥ T_final, or the cumulative impact index Tc ≥ T_final, an alarm will be triggered immediately, specifically including:

[0237] When any difference value in step S501 ≥ 1.5×T_final, and Tc ≥ 1.5×T_final, or any difference value in step S501 ≥ 2×T_final, or Tc ≥ 2×T_final, a first-level alarm will be triggered;

[0238] When T_final ≤ any difference value in step S501 < 1.5×T_final, and T_final ≤ Tc < 1.5×T_final, or any difference value in step S501 ≥ 1.5×T_final, but Tc < T_final, or Tc ≥ 1.5×T_final, but any difference value in step S501 < T_final, a second-level alarm will be triggered;

[0239] S6: Real-time monitor dynamic indicators and automatically cooperate and iterate;

[0240] The specific process is as follows: construct a three-dimensional dynamic index system of "efficiency - accuracy - timeliness";

[0241] Regularly and automatically collect the real-time data of the above indicators to form a dynamic trend chart, where the horizontal axis is time and the vertical axis is the index value. At the same time, mark the reference value line and the actual data value line, and compare the average value of the indicators in the recent 3 monitoring periods with the reference value to avoid false operations triggered by accidental fluctuations through multi-data point monitoring;

[0242] When the threshold is exceeded, trigger the model warning process. For example: the matching accuracy reference value is 95%, and the index values in the recent 3 monitoring periods are 94%, 93%, and 92% respectively, and the window average value 93% < 95% → trigger an alarm;

[0243] It is assumed that there is a normal fluctuation deviation of ±1%. If the index values ​​of the most recent 3 monitoring periods are 94%, 96%, and 93%, the window average of 94.3% ≈ 95% → will not trigger, in order to avoid random fluctuations.

[0244] The above-mentioned trigger model early warning process automatically generates a fluctuation trajectory file and continuously records the average value of the next three windows. If it is maintained in the 94%-95% range, it is marked as a stable fluctuation zone.

[0245] Based on the aforementioned trigger model warnings and stable fluctuation bands as optimization criteria, automatic collaborative iteration is performed to reduce the model warning trigger rate and update the normal fluctuation deviation values ​​according to changes in business scenarios.

[0246] Please refer to Figure 2 This invention also provides an intelligent matching system based on the aforementioned intelligent matching method for procurement needs and execution, the intelligent matching system comprising:

[0247] The data synchronization module 10 is used to construct a dynamic data association map and synchronize data accurately in real time.

[0248] The task dynamic allocation module 20 is used to build a global task scheduling model based on synchronization data, so as to dynamically allocate tasks based on load.

[0249] The dynamic demand parsing module 30 is used to generate demand orders based on synchronized data, and to perform dynamic parsing and semantic completion on the demand orders;

[0250] The supplier dynamic matching module 40 is used to continue dynamic matching and priority sorting of suppliers based on the dynamic parsing results of the demand order.

[0251] The discrepancy traceability and early warning module 50 is used to trace discrepancies and provide graded early warnings for abnormalities in the supplier matching process.

[0252] The dynamic indicator monitoring and iteration module 60 is used for real-time monitoring of dynamic indicators and automatic collaborative iteration.

[0253] Figure 3 This is a schematic diagram of the physical structure of an electronic device according to an embodiment of the present invention, such as... Figure 3 As shown, the electronic device 70 includes: a processor 701, a memory 702, and an internal bus 703; wherein, the processor 701 and the memory 702 communicate with each other through the internal bus 703;

[0254] The processor 701 is used to call program instructions in the memory 702 to execute the methods provided in the above-described method embodiments, such as: constructing a dynamic data association map and synchronizing data in real time with precision; constructing a global task scheduling model based on the synchronized data to dynamically allocate tasks based on load; generating a demand order based on the synchronized data and performing dynamic parsing and semantic completion on the demand order; continuing to perform dynamic matching and priority ranking of suppliers based on the dynamic parsing results of the demand order; performing source tracing and graded early warning for discrepancies in the supplier matching demand process; and monitoring dynamic indicators in real time and automatically coordinating and iterating.

[0255] This embodiment provides a non-transitory computer-readable storage medium that stores computer instructions. These instructions cause a computer to execute the methods provided in the above-described method embodiments, including, for example: constructing a dynamic data association graph and synchronizing data accurately in real time; constructing a global task scheduling model based on the synchronized data to dynamically allocate tasks based on load; generating a demand order based on the synchronized data and performing dynamic parsing and semantic completion on the demand order; continuing to dynamically match suppliers and prioritize them based on the dynamic parsing results of the demand order; tracing discrepancies and issuing tiered warnings for anomalies in the supplier matching process; and monitoring dynamic indicators in real time and automatically iterating collaboratively.

[0256] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps of the above method embodiments. The aforementioned storage medium includes various storage media capable of storing program code, such as ROM, RAM, magnetic disk, or optical disk.

[0257] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0258] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the parts that contribute to the prior art, can be embodied in the form of software products. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., including several instructions to cause a computer device (which may be a server or network device, etc.) to execute the methods of each embodiment or some parts of the embodiments.

[0259] Although the present invention has been described in detail above with general descriptions and specific embodiments, modifications or improvements can be made to it, which will be obvious to those skilled in the art. Therefore, all such modifications or improvements made without departing from the spirit of the present invention fall within the scope of protection claimed by the present invention.

Claims

1. An intelligent matching method based on procurement needs and execution, characterized in that, The steps include the following: Construct dynamic data association maps and synchronize data accurately in real time; A global task scheduling model is built based on synchronized data to dynamically allocate tasks based on load; A requirement order is generated based on the synchronized data, and the requirement order is dynamically parsed and semantically completed. Based on the dynamic parsing results of the demand order, continue to perform dynamic matching and priority ranking of suppliers; Implement differential source tracing and tiered early warning for anomalies in the supplier matching process; It monitors dynamic indicators in real time and automatically iterates in a coordinated manner.

2. The intelligent matching method based on procurement needs and execution as described in claim 1, characterized in that, The construction of a dynamic data association map and the real-time accurate data synchronization specifically include: A graph database is constructed, and the real-time relationships between "demand order - purchase order - supplier - warehouse - region" are stored in the graph database to form a dynamic data relationship graph. The demand order is used as the starting core node, and relationship edges are established with purchase orders, suppliers, warehouses and regions respectively. The relationship edges represent the relationship between entities, such as the "creation relationship" between demand orders and purchase orders, the "supply relationship" between purchase orders and suppliers, etc. When a demand order involves a specific product category, the relevant suppliers and their current availability information can be quickly located through the relationship edges based on the dynamic data relationship graph. Configure a distributed data gateway, which is associated with and connected to different independent subsystems, including the demand management system, procurement system, warehousing system, and supplier system. Collect multi-source heterogeneous data such as demand orders, purchase records, and warehouse data through a distributed data gateway; Furthermore, the CDC (Change Data Capture) model is introduced, and the CDC model is used to capture changes in multi-source heterogeneous data in the databases of each independent subsystem in real time; Specifically, the distributed data gateway monitors the binary log files of each independent subsystem database in real time based on the CDC model, and captures the data change events related to the log files, thereby triggering a real-time data synchronization mechanism to effectively reduce information lag and delay. We continue to build incremental data identification rules based on real-time synchronized data information, so that each piece of data is given a unique identifier and version number. The CDC model only directly synchronizes data items whose version numbers have been upgraded. When the current data is first created, the version number is the initial value. When the data changes, the version number is automatically synchronized and upgraded. The CDC model only compares the current version number with the last synchronized version number. If the data version number has not changed, it is not transmitted synchronously. If the version number is upgraded, only the data items whose version number has been upgraded are transmitted synchronously.

3. The intelligent matching method based on procurement needs and execution as described in claim 2, characterized in that, The step of constructing a global task scheduling model based on synchronization data to dynamically allocate tasks based on load specifically includes: A global task scheduling model is constructed, which loads the CPU and memory usage status information of cluster nodes (servers) in real time, and establishes a heartbeat monitoring mechanism with a frequency of once per second. Nodes (servers) with a load rate of <30% are designated as idle nodes. The global task scheduling model automatically assigns a task package to an idle node based on the main task queue generated from the synchronized data. The mapping relationship between the task package and the corresponding node is recorded in real time in the "task-node" subgraph of the dynamic data association graph. When the node's parsing times out (>30 seconds) or fails, the subtask is automatically returned to the global task scheduler to be re-entered into the main task queue. At the same time, the "status" attribute of the corresponding task package node in the dynamic data association graph is updated. This ensures that the synchronized data can be reasonably distributed according to the node load, avoids data accumulation, reduces information lag, and guarantees the real-time efficiency of subsequent needs.

4. The intelligent matching method based on procurement needs and execution as described in claim 3, characterized in that, The process of generating a demand order based on synchronized data, and dynamically parsing and semantically completing the demand order, specifically includes: A BERT pre-trained model is constructed, and a demand semantic understanding model is trained based on the BERT pre-trained model and a corpus of historical purchase orders. The demand semantic understanding model is input with demand order text generated by the CDC model synchronously, and the demand semantic understanding model parses and outputs explicit and implicit information in the demand order text. Explicit information includes clearly structured fields such as demand category, quantity, and region, while implicit information includes non-explicit intent fields. Combining current time series, historical collaboration, and special regulations data, an intent inference submodule is further constructed to complete the implicit information, and the overall parsing result after completion is verified for compliance. The expression for further verifying the compliance of the parsing is as follows: S=α·P+β·H+γ·C In the formula: S is the score for parsing the requirement form text; P is the score for the completeness of explicit information, which is based on the required and optional items in the standard requirement form and compared with the actual extracted explicit information. With 100 points as the benchmark, 30 points are deducted when a required item is missing and 10 points are deducted when an optional item is missing. H represents the accuracy score for completing implicit information. This score is based on the current timeline, historical collaboration, and specially specified data, and is compared with the actual extracted implicit information. A manual verification process can be triggered afterward. The final score is based on 100 points. If there are missing or incorrect extracted items in the comparison results, 10 points will be deducted for each item. C represents the supplier's real-time capacity adaptability score. Combining historical collaboration data with the supplier's real-time capacity data, the automatic splitting ratio of the demand quantity is optimized and calculated. That is, when the demand quantity exceeds the capacity of a single supplier, other suppliers are automatically introduced to split the demand quantity proportionally. The splitting is based on the current available capacity ratio of different suppliers. The expression for calculating the supplier's real-time capacity adaptability score C is as follows: Supplier real-time capacity adaptability score C = Total demand coverage score × 25% + Single supplier capacity adaptability score × 30% + Delivery time coordination score × 25% + Risk diversification score × 20% Among them, the total demand coverage score means that the sum of the quantities undertaken by each supplier must equal the total demand, otherwise it is 0 points; the single supplier capacity matching score means that the quantity allocated to each supplier must be within its real-time adjustable capacity, otherwise it is 0 points; and the delivery time coordination score means that the delivery time of each supplier must fully meet the delivery time of the total demand, otherwise it is 0 points. Risk diversification score: Avoid over-reliance on a single supplier. Reduce the overall risk caused by a sudden production failure of a single supplier by diversifying the allocation. The supplier with the largest allocation volume is used as the benchmark. If the largest supplier's allocation volume is ≤60% → 100 points, if the proportion is 60%-80% → 50 points, if the proportion is >80% → 0 points. α, β, and γ are the weight coefficients of the corresponding terms, where α = 0.4, β = 0.3, and γ = 0.

3. When the text parsing score S of the demand order is greater than or equal to 80, it is considered a qualified demand parsing. Otherwise, the parsing will be readjusted for the explicit information completeness score P and / or implicit information completeness score H and / or supplier real-time capacity adaptability score C, which are not full marks.

5. The intelligent matching method based on procurement needs and execution as described in claim 4, characterized in that, The process of generating a demand order based on synchronized data, and dynamically parsing and semantically completing the demand order, specifically includes: A dynamic graph embedding model is constructed to observe in real time whether there are areas with significant demand fluctuations, and these areas are mapped to the graph structure vertices of the dynamic graph embedding model. The static attribute of three-dimensional coordinate distance and the dynamic attribute of real-time capacity load rate are further embedded based on the dynamic graph embedding model. The quantification results of "static distance + dynamic load rate" are then fused with the parsing results of the demand semantic understanding model at a weight of 50% each, and the overall allocation ratio is recalculated.

6. The intelligent matching method based on procurement needs and execution as described in claim 5, characterized in that, The step of dynamically matching and prioritizing suppliers based on the dynamic parsing results of the demand order specifically includes: To avoid the drawbacks of traditional methods that only sort and match procurement needs and suppliers by time, a dynamic supplier capability matrix and a demand-supplier matching scoring model are constructed in a distributed manner. The dynamic supplier capability matrix updates the capabilities of matched suppliers in real time, and the demand-supplier matching scoring model sorts the matching degree. Finally, intelligent algorithms are used to dynamically optimize and ensure that high-value and urgent needs are prioritized. Based on the supplier dynamic capability matrix, a dynamic profile package is constructed for each supplier. The contents of the dynamic profile package include: Real-time supply S1: The quantity of goods that the supplier can immediately allocate; Short-term supply S2: The quantity of goods that a supplier can quickly allocate within a specific time period; Capacity elasticity coefficient E: reflects the supplier's ability to cope with demand fluctuations. E = (actual maximum capacity / conventional standard capacity) - 1. The larger the value, the better the capacity elasticity coefficient E. Regional delivery cost D: The total cost for the supplier to deliver goods to the target area; Cooperation fit F: Based on historical cooperation data, F = (Number of on-time deliveries / Total number of cooperations) × 0.6 + (Number of quality-compliant deliveries / Total number of cooperations) × 0.4; Continuing with the demand-supplier matching scoring model, the following relationship is established: M = M1 × 0.4 + M2 × 0.6 In the formula, M is the comprehensive matching score. The higher the comprehensive matching score, the higher the matching degree between the supplier and the demand. Basic matching score M1: M1 = Category fit Q × 0.5 + Cooperation adaptability F × 0.5, where the category fit Q is 0 or 1 depending on the degree of matching between the supplier's category and the demand category; Dynamic matching score M2: M2 = Supply capacity score A × 0.3 + Cost-effectiveness score B × 0.4 + Timeliness satisfaction score T × 0.2 + Capacity elasticity score E' × 0.1; Supply capacity score A: A = (S1 / Demand quantity) × 50 + (S2 / Demand quantity) × 50. If the result exceeds 100, it is calculated as 100 points. Cost-benefit analysis B: B = 100 - (Regional distribution cost D / Maximum acceptable cost) × 100; Timeliness satisfaction score T: T = (actual delivery time / required delivery time) × 100. If the actual delivery time is longer than the required delivery time, then T = 0. Capacity elasticity score E': E' = E × 100, converting the capacity elasticity coefficient into a score of 0-100, with any value exceeding 100 being counted as 100. Therefore, through multi-dimensional calculation, a comprehensive matching score between suppliers and demands is obtained. The higher the comprehensive matching score, the higher the matching priority between suppliers and demands, and thus their priority is ranked. To address real-time changes, a dynamic adjustment mechanism is further implemented as follows: Supply adjustment coefficient K1: When the change rate of the supplier's real-time supply S1 exceeds 20%, K1 = adjusted S1 / original S1, and at this time the dynamic matching M2 is updated to M2 × K1; Cost adjustment coefficient K2: When the change rate of regional delivery cost D exceeds 15%, K2 = D before adjustment / D after adjustment, and the dynamic matching M2 is updated to M2 × K2; Emergency Request Adjustment: For emergency requests, the comprehensive matching score M needs to be multiplied by the urgency coefficient U. U ranges from 1.2 to 2.0 depending on the urgency of the request. The higher the urgency, the larger U is. When a high-value order exceeds a preset threshold in the comprehensive matching score M calculation, the weight of the cooperation fit degree F is increased to 0.6, while the weight of the category fit degree Q is reduced to 0.4, and the capacity elasticity coefficient E is introduced for adjustment, i.e., M1 = (category fit degree Q × 0.4 + cooperation fit degree F × 0.6) × (1 + E × 0.2), in order to prioritize the stability of high-value orders and the ability to cope with fluctuations.

7. The intelligent matching method based on procurement needs and execution as described in claim 6, characterized in that, The aforementioned method for tracing discrepancies and providing tiered early warnings for anomalies in the supplier matching process specifically includes: A difference feature library is constructed, and difference factor data for each key node in the entire procurement process is collected and quantified in real time. These key nodes include demand nodes, order nodes, production nodes, logistics nodes, and warehousing nodes. The calculation formulas and meanings of the difference factors corresponding to each key node are as follows: ① Demand Node: Requirement change frequency: Fv = number of changes / original planned cycle, used to measure the frequency of requirement changes; Completeness of requirement description: Cd = number of complete items / total number of items × 100%, which reflects the level of detail in the requirement description. The difference value ΔCd is 100% - Cd. ② Order Node: Order response time: Tr = Supplier confirmation time - Order placement time, reflecting the supplier's response speed to the order; Price deviation rate: Rp = (actual unit price - benchmark unit price) / benchmark unit price × 100%, which measures the degree of deviation between the actual purchase price and the benchmark price; ③ Production nodes: Production scheduling achievement rate: Pc = actual production scheduling output / planned production scheduling output × 100%, which is used to evaluate the completion of the production scheduling plan. The difference is 100% - Pc. Material completeness rate: Cm = number of complete material items / total number of material items × 100%, which represents the completeness of the materials required for production. The difference value is 100% - Cm; ④ Logistics nodes: Transportation damage rate: Dt = number of damaged items / total transportation volume × 100%, which reflects the damage to goods during the logistics transportation process; In-transit delay rate: Dd = (actual in-transit time - standard in-transit time) / standard in-transit time × 100%, which reflects the timeliness of logistics transportation; ⑤ Inbound node: Quantity matching rate: Qi = Actual quantity received / Order quantity × 100%, used to judge the consistency between the actual quantity received and the order quantity, and the difference value is 100% - Qi; Quality pass rate: Qq = number of qualified products / number of samples × 100%, used to assess the quality compliance of goods entering the warehouse, and the difference is 100% - Qq; Specification matching rate: Qs = number of matching specifications / total number of specifications × 100%. This checks the degree of matching between the specifications of the goods entering the warehouse and the required specifications. The difference value ΔQs is 100% - Qs. Warehouse entry timeliness: Ti = actual entry time - planned entry time, which measures the timeliness of goods entering the warehouse; Construct a propagation matrix and calculate the cumulative impact index; Based on the difference factors extracted in the above steps, a propagation matrix Mij is further constructed. Mij is the influence coefficient of the i-th node on the j-th node. For example, M15 corresponds to the influence of ΔCd on ΔQs. Its initial value is determined by the correlation analysis of ΔCd and ΔQs, and Dd and Ti factors in historical collaborative data. For example, M15 = 0.

4. Obtain the node sensitivity Sj: It is obtained by training on the degree of influence of the historical data of the difference factors of other nodes on the difference factors of each node in the above steps. For example, the S5 of the inbound node is 0.

9. Calculate the propagation intensity Si at a single node: Si = Σ(Difference value of node i × Mij × Sj) For example, the propagation intensity from the demand node to the inbound node is S1 = ΔCd × M15 × S5; Obtain the weight Wi of each node: Set the weight Wi according to the historical data influence range of the difference factor of each node in the above steps, and satisfy ΣWi=1, for example, W5=0.3 for the inbound node; Further calculate the cumulative impact index Tc: Tc=Σ(Si×Wi) The cumulative impact index Tc is the core indicator for measuring the degree of impact of differences at each node on the entire procurement process. Its value reflects the severity of the impact of the difference, and it serves as a reference basis for tracing the source of the difference. Generate real-time dynamic warning thresholds, calculate warning levels and push them, specifically as follows: Directly associate with the historical data of the node difference factors extracted in the above steps, and take the 90th percentile value of the same type of differences in the past 12 months as the basic warning line T_base. The 90th percentile value means that in a set of data, 90% of the data is less than or equal to this value, and only 10% of the data is greater than this value; Further determine the actual risk coefficient R of the current supply chain. The specific formula is as follows: R = 1+(0.4×Dd + 0.3×ΔCm + 0.3×Rp) / 100 The three node difference factors directly used in the formula in step v above include: In-transit delay rate Dd: with a weight of 40%, used to reflect logistics risks; Material kit completeness rate difference ΔCm: with a weight of 30%, used to reflect production risks; Price deviation rate Rp: with a weight of 30%, used to reflect cost risks; Then the calculation formula for the final warning threshold T_final is as follows: T_final = T_base × R When any of the difference values in the above steps ≥ T_final, or the cumulative impact index Tc ≥ T_final, an alarm is immediately triggered, specifically including: When any of the difference values in the above steps ≥ 1.5×T_final, and Tc ≥ 1.5×T_final, or any of the difference values in the above steps ≥ 2×T_final, or Tc ≥ 2×T_final, a first-level alarm is triggered; When T_final ≤ any of the difference values in the above steps < 1.5×T_final, and T_final ≤ Tc < 1.5×T_final, or any of the difference values in the above steps ≥ 1.5×T_final, but Tc < T_final, or Tc ≥ 1.5×T_final, but any of the difference values in the above steps < T_final, a second-level alarm is triggered.

8. The intelligent matching method based on procurement requirements and execution according to claim 7, wherein The real-time monitoring of dynamic indicators and automatic collaborative iteration specifically includes: Construct a three-dimensional dynamic indicator system of "efficiency - accuracy - timeliness"; [[ID=*16]]Regularly and automatically collect real-time data of the above indicators to form a dynamic trend chart, where the horizontal axis is time and the vertical axis is the indicator value. At the same time, mark the reference value line and the actual data value line, and compare the average value of the indicators in the recent 3 monitoring cycles with the reference value. Through multi-data point monitoring, avoid accidental fluctuations from triggering misoperations; Trigger the model warning process when it is lower than the threshold; Suppose there is a normal fluctuation deviation. If the comparison between the window average value formed by the indicator values in the recent 3 monitoring cycles and the reference value is within the normal fluctuation deviation, no warning is triggered to avoid accidental fluctuations; Automatically generate a fluctuation trajectory file through the above-triggered model warning process, and continuously record the average values of the subsequent 3 windows. If it continues to maintain within the normal fluctuation deviation range, it is marked as a stable fluctuation band; Based on the above-triggered model warning and stable fluctuation band as the optimization basis for automatic collaborative iteration, reduce the model warning trigger rate, and update the normal fluctuation deviation value according to the changes in the business scenario.

9. An intelligent matching system based on the intelligent matching method for procurement needs and execution as described in claim 8, characterized in that, The intelligent matching system includes: A data synchronization module, used to construct a dynamic data association map and synchronize data in real time and accurately; A task dynamic allocation module, used to construct a global task scheduling model based on the synchronized data to dynamically allocate tasks based on the load; The dynamic demand parsing module is used to generate demand orders based on synchronized data, and to perform dynamic parsing and semantic completion on the demand orders. The supplier dynamic matching module is used to continue dynamic matching and priority sorting of suppliers based on the dynamic parsing results of the demand order. The discrepancy tracing and early warning module is used to trace discrepancies and issue tiered early warnings for anomalies in the supplier matching process. The dynamic indicator monitoring and iteration module is used for real-time monitoring of dynamic indicators and automatic collaborative iteration.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1-8.

Images