Intelligent decision support method based on dynamic weight distribution

By applying a dynamic weight allocation method based on attention mechanism in the supply chain, the problem that traditional methods are difficult to capture the dynamic relationship between data is solved, and more efficient resource allocation and supply chain optimization are achieved.

CN120218497APending Publication Date: 2025-06-27NAT SECURITY COLLEGE NAT DEFENSE UNIV OF THE CHINESE PEOPLES LIBERATION ARMY
View PDF 0 Cites 2 Cited by

Patent Information

Application Number
CN202510277548.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-03-10
Publication Date
2025-06-27

AI Technical Summary

Technical Problem

Traditional supply chain multi-dimensional indicator evaluation methods are difficult to capture the dynamic relationship between data, resulting in inefficient information utilization, affecting the accuracy of decision-making and the optimization of resource allocation.

Method used

The dynamic weight allocation method based on attention mechanism is adopted, and the importance of each decision-making object on different indicators is calculated by constructing a query matrix and a key matrix, and the dynamic weight allocation of multidimensional data of the supply chain is realized.

Benefits of technology

It improves the accuracy of resource allocation and the intelligence level of supply chain optimization, can adapt to the weight, adapt to complex supply chain environments, and improve the accuracy of decision-making.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN120218497A_ABST
    Figure CN120218497A_ABST
Patent Text Reader

Abstract

The invention discloses an intelligent decision support method based on dynamic weight distribution, and the method comprises the steps: constructing a supply chain data source integration module, and collecting and integrating supply chain data from different sources; a multi-source data preprocessing module is constructed, an information hierarchical structure system is constructed from five dimensions of delivery cycle, supply stability, transportation cost, quality qualification rate and supply risk, standardization processing is performed on supply chain data, and it is ensured that all indexes can be compared on the same scale; constructing a dynamic weight distribution module, calculating the importance of each decision object on different indexes by constructing a query matrix and a key matrix based on an attention mechanism, and performing dynamic weight distribution on multi-dimensional data in the supply chain; an intelligent decision analysis and monitoring module is constructed, the most critical indexes are identified, and managers are helped to make intelligent decisions in the aspects of supplier selection, storage and scheduling in combination with dynamic weight calculation results. According to the invention, efficient and accurate decision support is provided for supply chain resource optimization and management.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of multi-dimensional index evaluation of the supply chain, and specifically relates to an intelligent decision-making support method based on dynamic weight allocation. Background Art

[0002] With the rapid development of deep learning technology, the attention mechanism has made remarkable progress in the fields of natural language processing, computer vision, etc. By calculating the similarity between the query and the key, the attention mechanism can dynamically adjust the weight allocation of the values, thereby effectively enhancing the model's attention to key information and improving the accuracy of feature extraction and the overall performance. In supply chain management, especially in the field of multi-dimensional index evaluation and resource allocation, in the face of the challenges of multi-source heterogeneous data, traditional weight allocation methods often struggle to capture the dynamic relationships between data, resulting in low information utilization efficiency, which in turn affects the accuracy of decision-making and the optimization of resource allocation. Existing multi-source data processing methods, such as weight setting based on static rules, weighted methods based on statistical analysis, and feature extraction based on a single deep learning model, although can process data to a certain extent, still have limitations when dealing with complex multi-dimensional evaluations of the supply chain. The static rule method lacks flexibility and is difficult to adapt to the dynamic changes of data; the statistical analysis method is computationally complex and difficult to handle the non-linear relationships between data; a single deep learning model, due to its fixed network structure, cannot adaptively adjust the influence weights of different types of data, affecting the generalization ability of the model. Facing the complexity of multi-dimensional indexes and the heterogeneity of data sources in supply chain management, traditional methods are difficult to meet the needs of real-time optimization and precise decision-making. Therefore, a method that can adaptively and dynamically adjust data weights is needed to better optimize the resource allocation of the supply chain, improve the accuracy of multi-dimensional index evaluation, and achieve the intelligent upgrade of supply chain management. The dynamic weight allocation method based on the attention mechanism can flexibly capture the dynamic relationships between different data features, achieve accurate modeling of the complex interactions between various links of the supply chain, thereby improving the accuracy of resource allocation and the intelligent level of supply chain optimization. Summary of the Invention

[0003] The purpose of the present invention is to propose an intelligent decision-making support system and method based on dynamic weight allocation for application in the supply chain multi-dimensional index evaluation and resource allocation system.

[0004] The technical solution for achieving the purpose of the present invention is: an intelligent decision-making support method based on dynamic weight value allocation for supply chain multi-dimensional index evaluation and resource allocation, including:

[0005] Construct a supply chain data source integration module for collecting and integrating supply chain data from different sources;

[0006] Build a multi-source data preprocessing module to construct an information hierarchy system from five dimensions: delivery cycle, supply stability, transportation cost, quality pass rate, and supply risk, and perform standardized processing on supply chain data, including positive normalization and normalization, to ensure that all indicators can be compared on the same scale;

[0007] Build a dynamic weight allocation module to calculate the importance of each decision object on different indicators based on the attention mechanism by constructing a query matrix and a key matrix, and perform dynamic weight allocation on multi-dimensional data in the supply chain;

[0008] Build an intelligent decision-making analysis and monitoring module to identify the most critical indicators, and combine the dynamic weight calculation results to help managers make intelligent decisions in supplier selection, warehousing, and scheduling.

[0009] Furthermore, the supply chain data source integration module includes a supply chain data streaming access model, a supply chain data standardization model, and a supply chain data storage and access model, where:

[0010] The supply chain data streaming access model includes a streaming data accessor, a data preprocessor, a data monitor, and a data distributor to achieve real-time collection, quality control, and distribution of supply chain data, where: The streaming data accessor establishes a multi-source supply chain data stream channel based on the message queue and API gateway to achieve real-time communication with the ERP system, warehouse management system, supplier database, and transportation management system, and obtain supply chain data from the database, including inventory data, transportation data, order data, order remarks, order status, inventory changes, supplier evaluations, contract documents, historical inventory, and transportation costs. The data type depends on the number of data tables in the system; The data preprocessor uses the streaming computing framework Flink to clean, deduplicate, fill in missing values, and perform format conversion on the accessed original supply chain data to ensure data integrity and consistency; The data monitor monitors the quality of the data stream. If data anomalies are found, including sudden changes in inventory levels and duplicate orders, an automatic alarm or correction mechanism will be triggered. The monitored data is standardized and encapsulated by the data distributor and sent to the supply chain data standardization model in the <STREAM,TOPIC,DATA> format to ensure efficient data transmission and processing;

[0011] The supply chain data standardization model includes a data type recognizer and a data conversion engine, which are responsible for converting the incoming supply chain data into standardized numerical and text - type metric data. The data type recognizer automatically identifies the data type according to the data source and format, and distributes it to the numerical data converter and the text data converter. The numerical data converter standardizes numerical data, including inventory data and transportation data, ensures the uniformity of data format, and stores it in the <KEY,VALUE,TIMESTAMP> format to support subsequent calculations and analyses. The text data converter uses natural language processing (NLP) technology to parse text data, which also includes supplier evaluations and order remarks in the database, extracts key information and generates text summaries, and finally stores it in the <KEY,TEXT,TIMESTAMP> format.

[0012] The supply chain data storage and access model realizes the efficient persistent storage and access of supply chain data through a data storage router and a hierarchical data storage engine. The data storage router intelligently allocates data storage paths according to the data type and access requirements to optimize data management efficiency and query performance. For frequently accessed data, including order status and inventory changes, the hierarchical data storage engine stores them in the Redis in - memory database to ensure low - latency queries and real - time calculations. For structured data, including historical inventory and transportation costs, it stores them in the MySQL relational database and establishes an index based on KEY_TIMESTAMP to improve query performance and data retrieval efficiency. For unstructured or semi - structured data, including supplier evaluations and contract documents, it stores them in the NoSQL database to support full - text retrieval, fuzzy matching, and complex queries.

[0013] Furthermore, the multi - source data pre - processing module standardizes each supply chain database table, including positive normalization and normalization processing, to ensure that all metrics can be compared on the same scale. The specific method is as follows:

[0014] Define the initial data matrix X as follows:

[0015] X = {x1,x2…x n}

[0016] Determine the minimum and maximum values of each metric. The normalized data matrix Z is as follows:

[0017] Z = {z1,z2…z n}

[0018] The normalization formula is as follows:

[0019]

[0020] where x minis the minimum value of the index column, x max is the maximum value of the index column, x n represents each element of the index column.

[0021] Furthermore, the dynamic weight allocation module, based on the attention mechanism, calculates the importance of each decision object on different indicators by constructing a query matrix and a key matrix, and dynamically allocates weights to the multi-dimensional data in the supply chain. The specific method is as follows:

[0022] Extract the database of the supply chain nodes that are the focus of attention from the normalized data matrix Z to form the query matrix Q; at the same time, extract the databases of other supply chain nodes to form the key matrix K to calculate the similarity of the target supply chain node relative to other nodes;

[0023]

[0024] Among them, Attention is the similarity of the target supply chain node relative to other nodes. Each query matrix Q represents the indicators of the supply chain nodes that are the focus of attention, the key matrix K represents the corresponding data of other supply chain nodes, and the initial weight matrix V is used to measure the relative importance of each indicator in supply chain decision-making;

[0025] After the initial attention calculation, an iterative optimization mechanism is adopted to continuously adjust the weights of each indicator in the supply chain, so that it gradually converges to the optimal allocation strategy. Specifically, it is updated according to the weighted result calculated in the previous round and used as the query matrix for the next round:

[0026] Q t+1 = Attention·V.

[0027] Furthermore, the intelligent decision-making analysis and monitoring module identifies the most critical indicators, combines the dynamic weight calculation results, and helps managers make intelligent decisions in supplier selection, warehousing, and scheduling. The specific method is as follows:

[0028] In the iterative process of the attention mechanism in each round, calculate and monitor the change rate of the attention distribution. When the change range of the weight is lower than the set threshold, it is considered that the weight matrix has tended to be stable, and the iteration is stopped and the final supply chain optimization strategy, that is, the weights of each indicator, is output;

[0029] The finally generated weight matrix is directly applied to the optimization of resource allocation in the supply chain. In terms of supplier selection, based on the dynamically calculated weight results, suppliers with higher comprehensive weights are preferentially selected for long-term cooperation to ensure that the delivery cycle, supply stability, and quality qualification rate meet the supply chain requirements; in terms of warehousing management, adjust the inventory strategy in combination with the dynamic changes in the inventory level to ensure that the inventory level is consistent with the market demand fluctuations, thereby optimizing the inventory turnover rate and reducing the risks of inventory backlog or supply shortage.

[0030] An intelligent decision-making support system based on dynamic weight allocation, implementing the intelligent decision-making support method based on dynamic weight allocation to achieve intelligent decision-making support based on dynamic weight allocation, including:

[0031] A supply chain data source integration module for collecting and integrating supply chain data from different sources;

[0032] A multi-source data preprocessing module for constructing an information hierarchy system from five dimensions of delivery cycle, supply stability, transportation cost, quality qualification rate, and supply risk, and performing standardized processing on supply chain data, including positive normalization and normalization processing, to ensure that all indicators can be compared on the same scale;

[0033] A dynamic weight allocation module for calculating the importance of each decision object on different indicators based on the attention mechanism by constructing a query matrix and a key matrix, and performing dynamic weight allocation on multi-dimensional data in the supply chain;

[0034] An intelligent decision-making analysis and monitoring module for identifying the most critical indicators, and combining with the dynamic weight calculation results to help managers make intelligent decisions in supplier selection, warehousing, and scheduling.

[0035] A computer device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, it implements the intelligent decision-making support method based on dynamic weight allocation to achieve intelligent decision-making support based on dynamic weight allocation.

[0036] A computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements the intelligent decision-making support method based on dynamic weight allocation to achieve intelligent decision-making support based on dynamic weight allocation.

[0037] Compared with the prior art, the significant advantages of the present invention are: 1) Adaptive dynamic weight allocation, which can adaptively adjust weights through the attention mechanism in a complex supply chain environment to improve decision-making accuracy; 2) Efficient iterative optimization, gradually approaching the optimal solution through multiple rounds of calculations, providing a systematic and scalable calculation framework; 3) Compatibility with multi-source heterogeneous data, which can integrate various data from different types to improve data integration capabilities;

[0038] 4) Interpretability of decision-making analysis, intuitively showing the importance of each indicator through the attention distribution to enhance decision-making credibility; 5) Strong applicability and scalability, applicable to various industrial scenarios such as supply chain, manufacturing scheduling, and resource scheduling, with broad application prospects. Description of the Drawings

[0039] Figure 1 It is the architecture diagram of the multi-dimensional index evaluation and resource allocation system of the present invention for the supply chain.

[0040] Figure 2 It is the processing flowchart of the intelligent decision-making support method based on dynamic weight allocation.

[0041] Figure 3 It is the result of the intelligent decision-making support method based on dynamic weight allocation. Detailed implementation manners

[0042] In order to make the objectives, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0043] The intelligent decision-making support method of the present invention based on dynamic weight allocation has an overall architecture as Figure 1 shown, including a multi-source data preprocessing module, a dynamic weight allocation module, and an intelligent decision analysis module, where:

[0044] Supply chain data source integration module: used to collect and integrate supply chain databases from different nodes. Since there are multiple nodes in the supply chain, each node is stored in a different supply chain database, and there are multiple nodes in the supply chain system, each node has a database, and each database contains: inventory data, transportation data and order data, supplier evaluation, order remarks, order status, inventory changes, contract documents, historical inventory and transportation costs, etc. Convert the multi-source data into processable data according to different formats;

[0045] Multi-source data preprocessing module: used to collect and integrate supply chain data from different sources, including core business indicators such as supplier delivery cycle, transportation cost, and quality qualification rate. This module performs positive normalization on the original data (all data tables in the supply chain node corresponding database) to ensure that all indicators are compared on the same scale and provide clear input data for subsequent dynamic weight allocation. Through standardization processing, the weight deviation caused by different dimensions is avoided;

[0046] Dynamic weight allocation module: used to perform dynamic weight allocation on multi-dimensional data in different supply chain nodes based on the attention mechanism. This module constructs a query matrix and a key matrix, calculates the importance of different indicators in the supply chain node database, and adjusts the weight allocation in real time according to the data in other supply chain node databases. This module supports adaptive adjustment of weights to ensure that supply chain decisions can respond accurately according to environmental changes and provide a continuously updated weight distribution;

[0047] The intelligent decision-making analysis and monitoring module is used to conduct supply chain decision-making analysis and optimization based on the dynamic weight calculation results. By analyzing the final attention output results, it identifies the most critical decision-making factors to help managers make intelligent decisions in aspects such as supplier selection, warehousing, and scheduling. This module supports priority ranking, provides specific action recommendations based on weight allocation, and updates decisions in combination with real-time changes. At the same time, this module continuously monitors the dynamic changes in the supply chain decision-making process, real-time tracks the weight distribution of each indicator, and helps managers intuitively understand the change trends of different indicators through a visual interface. It also has an intelligent feedback mechanism that can adjust the decision-making model in a timely manner to ensure that the decisions are synchronized with the latest business status.

[0048] (1) Supply chain data source integration module

[0049] The supply chain data source integration module includes: a supply chain data streaming access model, a supply chain data standardization model, and a supply chain data storage and access model, where:

[0050] The supply chain data streaming access model realizes the real-time collection, quality control, and distribution of supply chain data through four key components: a streaming data accessor, a data pre-processor, a data monitor, and a data distributor. First, the streaming data accessor establishes a multi-source supply chain data stream channel based on the message queue and API gateway to achieve real-time communication with data sources such as the ERP system, warehouse management system, supplier database, and transportation management system. The data pre-processor uses the streaming computing framework Flink to clean, deduplicate, fill in missing values, and convert the format of the data in the accessed original supply chain database to ensure data integrity and consistency. For example, check whether there are abnormal missing values in the inventory data or whether the transportation data contains invalid timestamps. The data monitor conducts quality monitoring on the data stream. If data anomalies are found (such as sudden changes in inventory levels, duplicate orders, etc.), it will trigger an automatic alarm or correction mechanism. The monitored data is standardized and encapsulated by the data distributor and sent to the supply chain data pipeline in the <STREAM,TOPIC,DATA> format to ensure efficient data transmission and processing;

[0051] The supply chain data standardization model includes a data type recognizer and a data conversion engine, which are responsible for converting the incoming raw supply chain data into standardized numerical and text-based metric data. The data type recognizer automatically identifies the data type based on the data source and format, and distributes it to the numerical data converter and the text data converter. The numerical data converter standardizes numerical data such as inventory levels and transportation costs, ensures a unified data format, and stores it in the <KEY,VALUE,TIMESTAMP> format to support subsequent calculations and analyses. The text data converter uses natural language processing (NLP) techniques to parse text data such as supplier evaluations and order remarks, extracts key information, and generates text summaries, which are ultimately stored in the <KEY,TEXT,TIMESTAMP> format. This process ensures that all data can be stored in a structured manner, improves the data's computational capabilities and readability, and enables it to better meet the requirements of supply chain intelligent analysis and optimization.

[0052] Finally, the supply chain data storage and access model realizes the efficient persistent storage and access of supply chain data through a data storage router and a hierarchical data storage engine. The data storage router intelligently allocates data storage paths based on the data type and access requirements to optimize data management efficiency and query performance. For frequently accessed data such as order status and inventory changes, it is stored in a Redis in-memory database to ensure low-latency queries and real-time calculations. Structured data such as historical inventory and transportation costs are stored in a MySQL relational database, and indexes are established based on KEY_TIMESTAMP to improve query performance and data retrieval efficiency. For unstructured or semi-structured data such as supplier evaluations and contract documents, they are stored in a NoSQL database to support full-text retrieval, fuzzy matching, and complex queries. This multi-level data storage solution not only ensures the real-time availability of data but also enhances the scalability and management flexibility of the data, enabling the supply chain system to dynamically adjust the data storage strategy according to different business requirements, thereby supporting intelligent decision-making and resource optimization.

[0053] (2) Multi-source data preprocessing module

[0054] The multi-source data preprocessing module, because there are multiple nodes in the supply chain, each node is stored in a different supply chain database, and there are multiple nodes in the supply chain system, each node has a database, and each database contains: inventory data, transportation data, and order data, supplier evaluations, order remarks, order status, inventory changes, supplier evaluations, contract documents, historical inventory, transportation costs, etc. Core metrics are used to build an information hierarchy system to formalize the structured metrics and text summary data output by the supply chain data source integration module, which is described as follows:

[0055] In the multi-dimensional index evaluation of the supply chain system, it is necessary to standardize the core indicators (such as delivery cycle, supply stability, transportation cost, quality qualification rate, and supply risk, etc.) to ensure the comparability between different indicators. First, the system will determine the minimum and maximum values of each indicator. For example, in the delivery cycle indicator, calculate the shortest delivery time (such as 3 days) and the longest delivery time (such as 30 days) among all suppliers, warehousing centers, or transportation routes, providing a basis for subsequent normalization. Similarly, for the transportation cost indicator, it is necessary to determine the minimum transportation cost (such as 1000 yuan) and the maximum transportation cost (such as 3000 yuan) to ensure that all data can be calculated and analyzed on the same scale.

[0056] During the normalization process, the data of all supply chain indicators will be mapped to the standard range of [0,1]. For example, if a supplier's delivery cycle is 12 days, after Min-Max normalization, this value will be adjusted to a relative value between 0 and 1. Suppliers with a higher quality qualification rate (such as 98.5%) will be close to 1 after normalization, while a higher supply risk (such as 4.1 points) will be close to 0. Through such a normalization mapping, indicators with different dimensions can be adjusted to the same scale, making the subsequent weight calculation more reasonable and avoiding some indicators with larger values from having an excessive impact on the final decision.

[0057] In addition, to avoid the influence of extreme values on the normalization calculation. In Min-Max normalization, all values will be compressed into the interval [0,1], where the minimum value is mapped to 0 and the maximum value is mapped to 1. For example, if the transportation cost of a certain supply chain node is extremely high or low, this value will be close to 1 or 0 respectively during the normalization process to ensure that extreme values will not cause serious deviations in the overall data distribution. This standardization method can improve the stability of supply chain data, making subsequent intelligent analysis and weight calculation more accurate, and providing a scientific basis for optimizing resource allocation and supply chain management.

[0058] Define the initial data matrix X as follows:

[0059] X = {x1, x2…x n}

[0060] The normalized data matrix Z is as follows:

[0061] Z = {z1, z2…z n}

[0062] The quantile normalization formula is as follows:

[0063]

[0064] where x min is the minimum value of the indicator column, x maxis the maximum value of the index column.

[0065] (3) Dynamic Weight Allocation Module

[0066] The dynamic weight allocation module realizes the intelligent optimization of the multi-dimensional index evaluation and resource allocation of the supply chain by constructing a query matrix Q and a key matrix K. First, extract the database related to specific supply chain nodes from the normalized data matrix Z to form the query matrix Q; at the same time, extract the corresponding data of other supply chain nodes to form the key matrix K to calculate the similarity of a specific supply chain node relative to other nodes.

[0067]

[0068] Among them, Attention is the similarity of the target supply chain node relative to other nodes. Each query matrix Q represents the various indicators of the supply chain nodes that are the focus of attention, the key matrix K represents the corresponding data of other supply chain nodes, and the initial weight matrix V is used to measure the relative importance of each indicator in supply chain decision-making; through this calculation, the characteristic correlation between supply chain nodes can be dynamically evaluated, providing an accurate weight calculation basis for resource allocation.

[0069] After the initial attention calculation, the dynamic weight allocation module adopts an iterative optimization mechanism to continuously adjust the weights of the various indicators of the supply chain, making it gradually converge to the optimal allocation strategy. Specifically, it will be updated according to the weighted result of the previous round of calculation and used as the query matrix for the next round:

[0070] Q t+1 = Attention·V

[0071] This optimization process can adapt to the dynamic changes of the supply chain environment. For example, when the transportation cost rises significantly, it will automatically increase the weight of transportation cost in supply chain decision-making, and when the supply chain security faces challenges, it will increase the weights of supply stability and supply risk indicators. Through this method, the supply chain resource allocation strategy can be adjusted in real time with the changes of the market environment, thereby improving the accuracy of decision-making and the intelligent level of supply chain management.

[0072] Furthermore, during each iteration of the attention mechanism, the change rate of the attention distribution will be calculated and monitored. When the weight change amplitude is lower than the set threshold, it is considered that the weight matrix has stabilized, and the iteration is stopped and the final supply chain optimization strategy is output. This mechanism enables the system to adapt to the changes of the supply chain environment, dynamically adjust the weights of each key indicator, and thus ensure that the resource allocation reaches the optimal solution.

[0073] (4) Intelligent Decision Analysis and Monitoring Module

[0074] The intelligent decision-making analysis and monitoring module continuously collects data and uses visualization means to track the dynamic changes of key indicators in the supply chain nodes in real time. Combining with the weight distribution generated by the attention mechanism, it can timely detect abnormal fluctuations and issue warnings. When the change in the weight distribution exceeds the preset threshold, it will feedback warning information to the manager, prompting them to re-evaluate the supply chain resource allocation strategy or perform manual intervention to optimize the process. According to the manager's adjustment decision, the system updates the input data again and conducts a new round of weight calculation and iteration. In this way, it can ensure that the supply chain resource allocation is continuously optimized and balanced in dynamic changes, forming a closed-loop intelligent decision-making support.

[0075] In terms of supplier selection, based on the dynamically calculated weight results, suppliers with higher comprehensive weights are preferentially selected for long-term cooperation to ensure that key indicators such as delivery cycle, supply stability, and quality qualification rate meet the supply chain requirements. For warehouse management, the system adjusts the inventory strategy in combination with the dynamic changes in inventory levels to ensure that the inventory levels are consistent with the market demand fluctuations, thereby optimizing the inventory turnover rate and reducing the risks of inventory backlog or supply shortage.

[0076] An intelligent decision-making support method based on dynamic weight allocation realizes intelligent decision-making support based on dynamic weight allocation through the described intelligent decision-making support system based on dynamic weight allocation. The specific method is as follows:

[0077] The supply chain data extraction model is responsible for the streaming access and structured encapsulation of the original supply chain data. It is composed of a streaming data accessor, a data pre-processor, a data monitor, and a data distributor in series. The specific implementation steps are as follows:

[0078] 1) The streaming data accessor builds a multi-source supply chain data stream channel and establishes real-time communication with supply chain data sources such as the ERP system, warehouse management system, supplier database, and transportation management system based on the message queue and API gateway to ensure the efficient and stable access of supply chain data.

[0079] 2) The data pre-processor uses the streaming computing framework Flink to clean, format-check, and convert the accessed original data to ensure data consistency and integrity. For example, filling in missing values in inventory data or removing invalid timestamps in transportation data to ensure the reliability of the data.

[0080] 3) The data that passes the consistency check is subjected to real-time quality detection by the data monitor and encapsulated into standardized streaming data messages, which are then sent to the distributed message queue to achieve asynchronous decoupling from the data distributor, improving the data throughput capacity and streaming processing efficiency.

[0081] 4) The data distributor receives the data stream from the distributed message queue and further encapsulates it into<STREAM,TOPIC,DATA> format to ensure the standardization and traceability of data structure to support efficient processing of downstream data standardization and storage access models.

[0082] The multi-source data preprocessing module builds an information hierarchy system based on the structured indicators and text summary data output by the supply chain data source integration module, from core indicators such as delivery cycle, supply stability, transportation cost, quality qualification rate, supply risk, etc., to achieve standardized processing of supply chain data, so as to improve the comparability between indicators and optimize the subsequent intelligent analysis and decision-making process. The specific implementation steps are as follows:

[0083] (1) Multi-source data analysis and structured processing

[0084] After the supply chain data flows through the data standardization model and the streaming data preprocessing pipeline, it first enters the data type identifier, which parses the input data into numerical indicator data and text description data according to the data source, data mode, and field type. Among them, numerical data includes key indicators for supply chain performance evaluation (such as delivery cycle, supply stability, transportation cost, quality qualification rate, supply risk, etc.), while text data mainly involves unstructured information such as supplier evaluation and order notes. The classified data enters the data cleaning and consistency check process, and the system will eliminate abnormal data, fill in missing values, and correct the data format. For example, for transportation data, if a missing timestamp is found, reasonable time data can be filled in through historical trends to ensure data integrity.

[0085] (2) Normalized calculation of core indicators

[0086] In the process of multidimensional indicator evaluation, in order to ensure that different indicators are calculated and analyzed on the same scale, the core indicators need to be normalized. First, the minimum and maximum values ​​of each indicator are calculated to provide a basis for subsequent normalization calculations. For example, in the delivery cycle indicator, the shortest delivery time (such as 3 days) and the longest delivery time (such as 30 days) are calculated from the data set of all suppliers, storage centers or transportation routes; in the transportation cost indicator, the minimum transportation cost (such as 1,000 yuan) and the maximum transportation cost (such as 3,000 yuan) are determined to determine the value range of the indicator. On this basis, the Min-Max normalization method is used to map all indicators, and the values ​​are standardized to the [0,1] interval, so that indicators of different dimensions are comparable. The normalization calculation formula is as follows:

[0087] Define the initial data matrix X as follows:

[0088] X={x1,x2…x n}

[0089]

[0090] Among them, x min is the minimum value of the index column, and x max is the maximum value of the index column. After normalization, a shorter delivery cycle or a higher quality pass rate will approach 1, while a higher supply risk or transportation cost will approach 0, thus avoiding calculation biases caused by mismatches in data dimensions.

[0091] (3) Construction of the standardized data matrix

[0092] The normalized supply chain data constitutes the final standardized data matrix Z:

[0093] Z = {z1, z2…z n}

[0094] Among them, z n is the standardized supply chain data point. Each data point contains the normalized values of all core indicators and is used for subsequent weight calculation, indicator fusion, and intelligent decision-making analysis. For the supply chain system, this data matrix can provide highly consistent and comparable structured information, thereby improving the accuracy and reliability of data analysis.

[0095] From Figure 2 the method flow chart, it can be seen that the dynamic weight allocation module realizes the intelligent optimization of supply chain multi-dimensional index evaluation and resource allocation by constructing a query matrix Q and a key matrix K. Based on the normalized data matrix Z, this module calculates the similarity between supply chain nodes using the attention mechanism and dynamically adjusts the weights of each indicator through an iterative optimization mechanism, enabling the resource allocation strategy to adapt to the real-time changes in the supply chain environment. The specific implementation steps are as follows:

[0096] (1) Extraction and matrix construction of supply chain node data

[0097] Taking the normalized data matrix Z output by the supply chain data standardization model as the input, the system first extracts the core indicators related to specific supply chain nodes (such as suppliers, warehousing centers) from Z, including delivery cycle, supply stability, transportation cost, quality pass rate, supply risk, etc., to form the query matrix Q. At the same time, the corresponding data of other supply chain nodes are extracted to form the key matrix K, which is used to calculate the similarity between supply chain nodes.

[0098] (2) Attention calculation and initial weight allocation

[0099] The attention mechanism is used to calculate the feature correlation between the target supply chain node and other nodes to measure its relative importance in supply chain resource allocation. The calculation formula is as follows:

[0100]

[0101] Among them, Q represents the various indicators of the current supply chain node, K represents the corresponding data of other supply chain nodes, and the initial weight matrix V is used to measure the relative influence of each indicator in supply chain decision-making. Through softmax normalization, the weight values are distributed between 0 and 1 to ensure that the influence of each supply chain indicator is calculated on a unified scale.

[0102] (3) Weight Iterative Optimization and Adaptive Adjustment

[0103] After completing the initial attention calculation, the system adopts an iterative optimization mechanism to continuously adjust the weights of various supply chain indicators, gradually converging them to the optimal allocation strategy. Specifically, the system updates based on the weighted results calculated in the previous round and uses them as the query matrix for the next round:

[0104] Q t+1 = Attention·V

[0105] This optimization process can adapt to the dynamic changes in the supply chain environment. For example, when the market environment causes the transportation cost to rise, the system will automatically increase the weight of the transportation cost indicator, making resource allocation pay more attention to logistics efficiency; when the supply chain faces security risks (such as unstable suppliers), the system will dynamically increase the weights of supply stability and supply risk to ensure the overall stability of the supply chain. Figure 3 The results of the intelligent decision-making support method based on dynamic weight allocation can be seen.

[0106] The intelligent decision-making analysis and monitoring module realizes the intelligent optimization of supply chain resource allocation through continuous data collection, real-time weight calculation, dynamic adjustment, and early warning mechanisms. This module not only ensures that the core indicators of the supply chain (such as delivery cycle, transportation cost, inventory level, supply stability, and supply risk, etc.) maintain the optimal allocation in the face of market environment changes, but also ensures the intelligence, accuracy, and visualization of supply chain management decisions through a closed-loop monitoring and feedback system. The specific implementation steps are as follows:

[0107] (1) In each round of iteration, the change rate of the attention distribution is calculated and monitored to ensure the stability of weight adjustment. Specifically, the change amplitude between the current weight matrix and the previous weight matrix is calculated after each round of iteration. When the change amplitude is lower than the set threshold, it is determined that the weight matrix has stabilized, and the iteration is stopped, and the final supply chain optimization strategy is output. This mechanism ensures that the system reduces unnecessary calculation costs while ensuring accuracy and improves decision-making efficiency.

[0108] (2) To ensure that the allocation of supply chain resources is always in an optimal state, a real-time monitoring and early warning mechanism is adopted to continuously track the dynamic changes of the core indicators of the supply chain. The fluctuation trends of each core indicator are displayed in real time through a visual dashboard, and combined with the weight distribution generated by the attention mechanism, early warnings are issued for abnormal situations. When the change in the weight distribution exceeds the preset threshold, an alarm is triggered and a warning message is sent to the manager.

[0109] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope recorded in this specification.

[0110] The above-described embodiments merely represent several implementation manners of the present application. Their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of the present application. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present application, several modifications and improvements can still be made, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the appended claims.

Claims

1. An intelligent decision support method based on dynamic weight allocation, characterized in that: Used for supply chain multi-dimensional indicator evaluation and resource allocation, including: Build a supply chain data source integration module to collect and integrate supply chain data from different sources; Construct a multi-source data preprocessing module to build an information hierarchy system from five dimensions: delivery cycle, supply stability, transportation cost, quality qualification rate and supply risk, and standardize the supply chain data, including forward and normalization processing, to ensure that all indicators can be compared on the same scale; Build a dynamic weight allocation module to dynamically allocate weights to multi-dimensional data in the supply chain by building query matrices and key matrices based on the attention mechanism to calculate the importance of each decision object on different indicators; Build an intelligent decision-making analysis and monitoring module to identify the most critical indicators and combine the dynamic weight calculation results to help managers make intelligent decisions in supplier selection, warehousing and scheduling.

2. The intelligent decision support method based on dynamic weight allocation according to claim 1 is characterized in that: The supply chain data source integration module includes the supply chain data streaming access model, the supply chain data standardization model, and the supply chain data storage and access model, among which: The supply chain data streaming access model includes a streaming data accessor, a data preprocessor, a data monitor and a data distributor to realize the real-time collection, quality control and distribution of supply chain data. Among them: the streaming data accessor is based on the message queue and API gateway to establish a multi-source supply chain data stream channel to realize real-time communication with the ERP system, warehouse management system, supplier database and transportation management system, and obtain supply chain data from the database, including inventory data, transportation data, order data, order notes, order status, inventory changes, supplier evaluation, contract documents, historical inventory and transportation costs. The data type depends on the number of data tables in the system; the data preprocessor uses the streaming computing framework Flink to clean and deduplicate the raw supply chain data, fill in missing values, and perform format conversion to ensure data integrity and consistency; the data monitor monitors the quality of the data stream. If data anomalies are found, including sudden changes in inventory and repeated orders, an automatic alarm or correction mechanism will be triggered. The monitored data is standardized and packaged by the data distributor and used as a<STREAM,TOPIC,DATA> The format is sent to the supply chain data standardization model to ensure efficient data transmission and processing; The supply chain data standardization model includes a data type identifier and a data conversion engine, which are responsible for converting the incoming supply chain data into standardized numerical and text indicator data. The data type identifier automatically identifies the data type according to the source and format of the data, and distributes it to the numerical data converter and the text data converter; the numerical data converter standardizes the numerical data, including inventory data and transportation data, to ensure the uniformity of the data format, and adopts<KEY,VALUE,TIMESTAMP> format to support subsequent calculation and analysis; the text data converter uses natural language processing (NLP) technology to parse text data. The database also includes supplier evaluations and order notes, extracts key information and generates text summaries, and finally<KEY,TEXT,TIMESTAMP> Format storage; The supply chain data storage and access model achieves efficient and persistent storage and access to supply chain data through a data storage router and a hierarchical data storage engine. The data storage router intelligently allocates data storage paths based on data types and access requirements to optimize data management efficiency and query performance. The hierarchical data storage engine stores frequently accessed data, including order status and inventory changes, in a Redis memory database to ensure low-latency queries and real-time calculations. Structured data, including historical inventory and transportation costs, are stored in a MySQL relational database and indexed based on KEY_TIMESTAMP to improve query performance and data retrieval efficiency. Unstructured or semi-structured data, including supplier evaluations and contract documents, are stored in a NoSQL database to support full-text retrieval, fuzzy matching, and complex queries.

3. The intelligent decision support method based on dynamic weight allocation according to claim 1 is characterized in that: The multi-source data preprocessing module standardizes each supply chain database table, including forward and normalization, to ensure that all indicators can be compared on the same scale. The specific methods are as follows: Define the initial data matrix X as follows: X={x1,x2…x n } Determine the minimum and maximum values ​​of each indicator, and the normalized data matrix Z is as follows: Z={z1,z2…z n } The normalization formula is as follows: where x min is the minimum value of the indicator column, x max is the maximum value of the indicator column, x n Represents each element of the indicator column.

4. The intelligent decision support method based on dynamic weight allocation according to claim 1 is characterized in that: The dynamic weight allocation module, based on the attention mechanism, constructs query matrix and key matrix, calculates the importance of each decision object on different indicators, and dynamically allocates weights to multi-dimensional data in the supply chain. The specific method is as follows: Extract the database of supply chain nodes of focus from the normalized data matrix Z to form the query matrix Q; at the same time, extract the database of other supply chain nodes to form the key matrix K to calculate the similarity of the target supply chain node relative to other nodes; Among them, Attention is the similarity of the target supply chain node relative to other nodes, each query matrix Q represents the indicators of the supply chain node that is focused on, the key matrix K represents the corresponding data of other supply chain nodes, and the initial weight matrix V is used to measure the relative importance of each indicator in the supply chain decision; After the initial attention calculation, an iterative optimization mechanism is used to continuously adjust the weights of various supply chain indicators so that they gradually converge to the optimal allocation strategy. Specifically, the weighted results of the previous round of calculations are updated and used as the query matrix for the next round: Q t+1 =Attention·V。 5. The intelligent decision support method based on dynamic weight allocation according to claim 1 is characterized in that: The intelligent decision-making analysis and monitoring module identifies the most critical indicators and combines the dynamic weight calculation results to help managers make intelligent decisions in supplier selection, warehousing and scheduling. The specific methods are as follows: In each round of attention mechanism iteration, the change rate of attention distribution is calculated and monitored. When the weight change is lower than the set threshold, the weight matrix is ​​considered to be stable, so the iteration is stopped and the final supply chain optimization strategy, that is, the weight of each indicator, is output; The final generated weight matrix is ​​directly applied to the optimization of supply chain resource allocation. In terms of supplier selection, based on the dynamically calculated weight results, suppliers with higher comprehensive weights are given priority for long-term cooperation to ensure that the delivery cycle, supply stability and quality qualification rate meet the supply chain requirements; In terms of warehouse management, inventory strategies are adjusted based on dynamic changes in inventory levels to ensure that inventory levels are consistent with market demand fluctuations, thereby optimizing inventory turnover and reducing the risk of inventory backlogs or supply shortages.

6. An intelligent decision support system based on dynamic weight allocation, characterized in that: Implementing the intelligent decision support method based on dynamic weight allocation as described in any one of claims 1 to 5 to realize intelligent decision support based on dynamic weight allocation includes: Supply chain data source integration module, used to collect and integrate supply chain data from different sources; The multi-source data preprocessing module is used to build an information hierarchy system from five dimensions: delivery cycle, supply stability, transportation cost, quality qualification rate and supply risk, and standardize the supply chain data, including positive and normalization processing, to ensure that all indicators can be compared on the same scale; The dynamic weight allocation module is used to dynamically allocate weights to multi-dimensional data in the supply chain by constructing query matrices and key matrices based on the attention mechanism to calculate the importance of each decision object on different indicators; The intelligent decision-making analysis and monitoring module is used to identify the most critical indicators and, combined with the dynamic weight calculation results, help managers make intelligent decisions in supplier selection, warehousing and scheduling.

7. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the intelligent decision support method based on dynamic weight allocation as described in any one of claims 1 to 5 is implemented to realize intelligent decision support based on dynamic weight allocation.

8. A computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the intelligent decision support method based on dynamic weight allocation according to any one of claims 1 to 5 is implemented to realize intelligent decision support based on dynamic weight allocation.

Citation Information

Cited By

  • E-commerce intelligent customer service AI dynamic duty and automatic recovery system

    CN120782456A

  • Configurable multi-information device source selection method, system and device

    CN121396791A