A supply chain demand prediction method and system based on multi-source data fusion

Abstract

This application provides a supply chain demand forecasting method and system based on multi-source data fusion, relating to the field of data analysis. The method includes: acquiring multi-source supply chain data for machinery and equipment; constructing a heterogeneous supply chain relationship graph based on the multi-source supply chain data; extracting demand feature vectors from the multi-source supply chain data, performing cluster analysis on the feature demand vectors using a data field fuzzy clustering method to obtain a first importance weight; constructing a supply chain network structure graph based on the supply chain relationship heterogeneous graph and the first importance weight, generating a first adjacency matrix; the first adjacency matrix is ​​used to characterize the relationships between supply chain nodes; extracting spatial correlation features and demand time series features from the supply chain network structure graph, the first adjacency matrix, and the demand feature vectors, and performing basic demand forecasting to obtain the basic demand forecasting result. This application, used in the supply chain demand forecasting process, solves the technical problems of low accuracy and stability in existing demand forecasting technologies.

Description

Technical Field

[0001] This application relates to the field of data analysis, and in particular to a supply chain demand forecasting method and system based on multi-source data fusion. Background Technology

[0002] With the continuous improvement of digitalization and intelligence in the industrial manufacturing sector, supply chain demand forecasting technology plays an increasingly important role in machinery and equipment production planning, inventory management, and supply chain collaborative optimization. Current technologies typically predict future demand by statistically analyzing historical sales, order, or inventory data, using regression analysis or traditional machine learning algorithms. However, the machinery and equipment supply chain is characterized by multiple layers and demand significantly influenced by market conditions and logistics. Traditional forecasting methods often focus on a single data source or a single dimension, making it difficult to effectively characterize the structural relationships between different supply chain nodes and the synergistic effects between multi-source data. This results in insufficient utilization of overall supply chain structural information in demand forecasting, leading to technical problems such as low accuracy and stability in demand forecasting. Summary of the Invention

[0003] This application provides a supply chain demand forecasting method and system based on multi-source data fusion, which solves the technical problems of low accuracy and stability of demand forecasting in existing technologies.

[0004] To achieve the above objectives, this application adopts the following technical solution: Firstly, this paper provides a supply chain demand forecasting method based on multi-source data fusion, including: acquiring multi-source supply chain data for machinery and equipment; the multi-source supply chain data includes: internal supply chain data and external market environment data; the internal supply chain data includes: sales data, order data, inventory data, logistics data, and price data; constructing a supply chain relationship heterogeneous graph based on the multi-source supply chain data; extracting demand feature vectors from the multi-source supply chain data, and performing cluster analysis on the feature demand vectors using data field fuzzy clustering to obtain the first importance weight; the first importance weight refers to the demand importance weight of each machinery and equipment category in different supply chain nodes; constructing a supply chain network structure graph based on the supply chain relationship heterogeneous graph and the first importance weight, and generating a first adjacency matrix; the first adjacency matrix is ​​used to represent the correlation between supply chain nodes; extracting the spatial correlation features and demand time series features of the supply chain network structure graph, the first adjacency matrix, and the demand feature vectors, and performing basic demand forecasting to obtain the basic demand forecasting results.

[0005] In conjunction with the first aspect mentioned above, in one possible implementation, after obtaining the basic demand forecast result, the method further includes: acquiring multi-source supply chain data of historical machinery and equipment, and constructing a demand fluctuation risk assessment vector; the multi-source supply chain data of historical machinery and equipment includes: historical demand data, historical inventory levels, historical logistics delay information, and historical market fluctuation indicators; based on the demand fluctuation risk assessment vector, assessing the demand fluctuation risk of the basic demand forecast result to obtain a demand risk assessment level; and based on the risk assessment level, collaboratively correcting the basic demand forecast result based on the supply chain relationship heterogeneity diagram to obtain the final supply chain demand forecast result.

[0006] In conjunction with the first aspect mentioned above, one possible implementation involves constructing a heterogeneous supply chain relationship graph based on multi-source supply chain data. This includes: identifying the main supply chain nodes from the multi-source supply chain data, defining supplier nodes, machinery and equipment manufacturing nodes, parts manufacturing nodes, regional distribution nodes, and end-customer nodes as the node set in the supply chain graph; determining the transaction relationships and logistics flow relationships between supply chain nodes based on sales data, order data, and logistics data from the multi-source supply chain data, and determining the functional association relationships based on the matching relationships between machinery and equipment and parts, thus obtaining a set of edge relationships between nodes; and constructing a heterogeneous supply chain relationship graph structure based on the node set and the edge relationship set.

[0007] In conjunction with the first aspect mentioned above, one possible implementation involves extracting demand feature vectors from multi-source supply chain data and performing cluster analysis on these feature demand vectors using a data field fuzzy clustering method to obtain the first importance weight. This includes: extracting feature information related to the demand for machinery and equipment from multi-source supply chain data and constructing demand feature vectors, whereby the feature information includes sales volume, order volume, inventory volume, and logistics timeliness information; initializing a membership matrix based on the demand feature vectors and setting the number of cluster categories and clustering iteration parameters; calculating the cluster centers corresponding to each cluster category based on the membership matrix and iteratively updating the membership matrix based on the cluster centers to obtain iteratively updated membership degrees; and when the membership matrix converges, iteratively updating the membership degrees of the demand feature vectors in each cluster category, introducing time decay weights, and calculating the demand importance weights of each machinery and equipment category in different supply chain nodes to obtain the first importance weight.

[0008] In conjunction with the first aspect mentioned above, in one possible implementation, the cluster center... Satisfy the following formula:

[0009] Where j represents the cluster category number, and i represents the sample index of a certain type of mechanical equipment. This represents the membership degree of the i-th sample to the j-th class. This represents the demand characteristic value of the i-th mechanical equipment.

[0010] In conjunction with the first aspect mentioned above, one possible implementation involves iteratively updating the membership degree. Satisfy the following formula:

[0011] Where C represents the total number of clusters, and k is the traversal index. This represents the demand characteristic value of the i-th piece of machinery. For the j-th cluster center, For the k-th cluster center, Here, m represents the Euclidean distance, and m is the fuzzy weighting exponent. First Importance Weight Satisfy the following formula:

[0012] in, To iteratively update the membership degree; For time decay weight, , For time decay coefficient, This is the current data collection point in time. The reference time point is i, which represents the sample index of a certain type of mechanical equipment.

[0013] In conjunction with the first aspect mentioned above, in one possible implementation, a supply chain network structure diagram is constructed based on the supply chain relationship heterogeneity diagram and the first importance weight, and a first adjacency matrix is ​​generated. This includes: determining the connection relationships between supply chain nodes based on the supply chain relationship heterogeneity diagram, and extracting the transaction relationships, logistics flow relationships, and functional association relationships between each node; determining the demand impact weight of each supply chain node based on the first importance weight, and using the demand impact weight as the node weight; weighting the connection relationships between supply chain nodes based on the node weight to construct a weighted supply chain network structure diagram; and generating a first adjacency matrix based on the weighted supply chain network structure diagram, wherein the elements in the first adjacency matrix are used to represent the connection relationships between supply chain nodes and their demand impact weights.

[0014] In conjunction with the first aspect mentioned above, one possible implementation involves extracting spatial correlation features and demand time series features from the supply chain structure diagram, the first adjacency matrix, and the demand feature vector, and performing basic demand forecasting to obtain the basic demand forecasting result. This includes: extracting spatial features of the relationships between supply chain nodes using a graph convolutional network based on the supply chain network structure diagram and the first adjacency matrix to obtain supply chain spatial correlation features; constructing a mechanical equipment demand time series based on the demand feature vector, and extracting time features of the demand time series using a recurrent neural network to obtain demand time series features; fusing the supply chain spatial correlation features and the demand time series features, and using a demand forecasting model for forecasting to obtain the basic demand forecasting result.

[0015] In conjunction with the first aspect mentioned above, in one possible implementation, a demand fluctuation risk assessment is performed on the basic demand forecast results based on the demand fluctuation risk assessment vector to obtain a demand risk assessment level. This includes: inputting the demand fluctuation risk assessment vector into a BP neural network model, performing feature mapping on the correlation between various risk indicators through hidden layer neurons to obtain the hidden layer output; performing risk prediction calculation based on the hidden layer output to obtain the risk assessment level; and using the risk assessment level to characterize the demand fluctuation risk level of the machinery and equipment supply chain.

[0016] Secondly, a supply chain demand forecasting system based on multi-source data fusion is provided. The system includes: a data acquisition device and an electronic device; wherein, the data acquisition device is used to acquire multi-source supply chain data of machinery and equipment; the multi-source supply chain data includes: internal supply chain data and external market environment data; the internal supply chain data includes: sales data, order data, inventory data, logistics data, and price data; the electronic device is used to construct a heterogeneous supply chain relationship diagram based on the multi-source supply chain data; extract the demand feature vectors of the multi-source supply chain data, and perform cluster analysis on the feature demand vectors using the data field fuzzy clustering method to obtain the first importance weight; the first importance weight refers to the demand importance weight of each category of machinery and equipment in different supply chain nodes; construct a supply chain network structure diagram based on the heterogeneous supply chain relationship diagram and the first importance weight, and generate a first adjacency matrix; the first adjacency matrix is ​​used to represent the correlation between supply chain nodes; extract the spatial correlation features and demand time series features of the supply chain structure diagram, the first adjacency matrix, and the demand feature vectors, and perform basic demand forecasting to obtain the basic demand forecasting results.

[0017] This application provides a supply chain demand forecasting method and system based on multi-source data fusion. By constructing a multi-source data system that integrates internal supply chain data and external market environment data, the overall expressive power of demand information is improved. Through fuzzy clustering analysis of demand feature vectors, the demand importance weights of each machinery and equipment category in different supply chain nodes are obtained, enabling a reasonable characterization of the demand influence relationships between supply chain nodes and improving the accuracy of supply chain network structure modeling. Furthermore, by combining supply chain spatial correlation characteristics and demand time series characteristics for basic demand forecasting, the forecasting model can simultaneously utilize supply chain structure information and demand change patterns, further improving the accuracy and stability of supply chain demand forecasting results. In addition, by constructing a demand fluctuation risk assessment mechanism, the basic demand forecasting results are risk-assessed and collaboratively corrected, making the forecasting results more consistent with actual market fluctuations, improving the overall operational efficiency of the machinery and equipment supply chain, and solving the technical problem of low accuracy and stability in existing supply chain demand forecasting technologies.

[0018] It should be understood that the descriptions of technical features, technical solutions, beneficial effects, or similar language in this application do not imply that all features and advantages can be achieved in any single embodiment. Rather, it is understood that the description of a feature or beneficial effect means that a specific technical feature, technical solution, or beneficial effect is included in at least one embodiment. Therefore, the descriptions of technical features, technical solutions, or beneficial effects in this specification do not necessarily refer to the same embodiment. Furthermore, the technical features, technical solutions, and beneficial effects described in this embodiment can be combined in any suitable manner. Those skilled in the art will understand that embodiments can be implemented without one or more specific technical features, technical solutions, or beneficial effects of a particular embodiment. In other embodiments, additional technical features and beneficial effects may be identified in specific embodiments that do not embody all embodiments. Attached Figure Description

[0019] Figure 1 A system architecture diagram of a supply chain demand forecasting system based on multi-source data fusion is provided for embodiments of this application; Figure 2 A flowchart illustrating a supply chain demand forecasting method based on multi-source data fusion, provided for an embodiment of this application; Figure 3 A flowchart illustrating another supply chain demand forecasting method based on multi-source data fusion provided in this application embodiment; Figure 4 A flowchart illustrating another supply chain demand forecasting method based on multi-source data fusion provided in this application embodiment; Figure 5A flowchart illustrating another supply chain demand forecasting method based on multi-source data fusion provided in this application embodiment. Detailed Implementation

[0020] In the description of this application, unless otherwise stated, " / " means "or," for example, A / B can mean A or B. The "and / or" in this document is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, and B alone. Furthermore, "at least one" means one or more, and "multiple" means two or more. The terms "first," "second," etc., do not limit the quantity or order of execution, and "first," "second," etc., do not necessarily imply differences.

[0021] It should be noted that, in this application, the terms "exemplary" or "for example" are used to indicate that something is being described as an example, illustration, or illustration. Any embodiment or design described as "exemplary" or "for example" in this application should not be construed as being more preferred or advantageous than other embodiments or design solutions. Specifically, the use of terms such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner.

[0022] The supply chain demand forecasting method based on multi-source data fusion provided in this application can be applied to, for example... Figure 1 In the supply chain demand forecasting system shown, based on multi-source data fusion, such as Figure 1 As shown, the system includes: a data acquisition device 101 and an electronic device 102; Among them, the data acquisition device 101 is used to acquire multi-source supply chain data of mechanical equipment; the multi-source supply chain data includes: internal supply chain data and external market environment data; the internal supply chain data includes: sales data, order data, inventory data, logistics data and price data.

[0023] Electronic device 102 is used to construct a heterogeneous supply chain relationship graph based on multi-source supply chain data; extract demand feature vectors from the multi-source supply chain data, and perform cluster analysis on the feature demand vectors using data field fuzzy clustering to obtain the first importance weight; the first importance weight refers to the demand importance weight of each category of machinery and equipment in different supply chain nodes; construct a supply chain network structure graph based on the heterogeneous supply chain relationship graph and the first importance weight, and generate a first adjacency matrix; the first adjacency matrix is ​​used to represent the correlation between supply chain nodes; extract the spatial correlation features and demand time series features of the supply chain structure graph, the first adjacency matrix, and the demand feature vectors, and perform basic demand forecasting to obtain the basic demand forecasting results.

[0024] To address the technical problems of low accuracy and stability in demand forecasting in existing technologies, embodiments of this application provide a supply chain demand forecasting method based on multi-source data fusion.

[0025] Figure 2 A flowchart illustrating the supply chain demand forecasting method based on multi-source data fusion provided in this application embodiment is shown below. Figure 2 As shown, the method includes: S201. Obtain multi-source supply chain data for machinery and equipment.

[0026] Multi-source supply chain data refers to a collection of data from different business systems or external information channels that reflects the operational status of the machinery and equipment supply chain and changes in market demand. This includes internal supply chain data and external market environment data. Internal supply chain data includes sales data, order data, inventory data, logistics data, and price data. External market environment data refers to external information that reflects the trend of market demand changes, such as industry prosperity index, construction commencement index, macroeconomic index, and raw material market price data.

[0027] In one possible implementation, sales and order records of machinery and equipment are obtained from the enterprise's ERP system through enterprise data interfaces or database access methods; inventory quantity and inventory turnover information are obtained from the warehouse management system; logistics transportation cycle and logistics node status information are obtained from the logistics management system; and price change data of machinery and equipment and key components are obtained from the price management system. Industry market indices, macroeconomic indicators, and raw material price fluctuation information are obtained through external data interfaces or public data platforms. The acquired multi-source data is then uniformly timestamped, cleaned, and outlier removed, and stored in a standardized manner according to a unified data structure, thereby forming a multi-source supply chain dataset that can be used for subsequent analysis.

[0028] It should be noted that during the data collection process, the time granularity of data from different sources should be uniformly processed. For example, daily, weekly, or monthly data from different systems should be converted into the same statistical period to avoid the impact of data at different time scales on subsequent analysis. Incomplete data can also be supplemented by missing value imputation or data smoothing to ensure data integrity.

[0029] S202. Construct a heterogeneous supply chain relationship diagram based on multi-source supply chain data.

[0030] Among them, the supply chain relationship heterogeneous graph refers to a graph structure model composed of multiple types of nodes and multiple types of relationships, which is used to describe the relationship between different participants in the supply chain network. In this graph structure, nodes are used to represent supply chain participants, such as suppliers, machinery and equipment manufacturers, parts manufacturers, regional distributors and end customers, while edges are used to represent business relationships between nodes, such as transaction relationships, logistics relationships or supporting relationships.

[0031] In one possible implementation, business entities in the supply chain are identified based on multi-source supply chain data, and different types of entities are abstracted as nodes in a graph structure. For example, node sets are established for component supplier nodes, machinery and equipment manufacturing nodes, regional distribution nodes, and end customer nodes, respectively. The transaction relationships between nodes are determined based on sales data and order data, the logistics flow relationships between nodes are determined based on logistics data, and the functional association relationships are determined based on the assembly or matching relationships between machinery and equipment and components. Finally, different types of nodes and their relationships are uniformly encoded, and a heterogeneous supply chain relationship graph is generated through graph structure modeling.

[0032] It should be noted that when constructing a heterogeneous supply chain relationship diagram, a unified node identifier should be set for different types of nodes, and relationship labels should be set for different types of relationships, so as to distinguish different business relationship types in the subsequent supply chain network modeling and feature extraction process.

[0033] As an example, in the machinery and equipment manufacturing supply chain, there is a raw material supply relationship between steel supplier nodes and parts manufacturing nodes, a component supply relationship between parts manufacturing nodes and complete machine manufacturing nodes, a sales relationship between complete machine manufacturing nodes and regional distribution nodes, and an equipment delivery relationship between regional distribution nodes and end customers. Through the above relationships, a complete supply chain relationship heterogeneous diagram can be formed.

[0034] Based on the above steps, this step constructs a heterogeneous graph structure that can reflect the types of supply chain nodes and their various business relationships, so that the business relationships in the supply chain network can be expressed intuitively, which is conducive to subsequent modeling and analysis of supply chain structure information.

[0035] S203. Extract the demand feature vector from the multi-source supply chain data, and perform cluster analysis on the feature demand vector using the data field fuzzy clustering method to obtain the first importance weight.

[0036] The demand feature vector refers to a set of features extracted from multi-source supply chain data that reflects changes in the demand for machinery and equipment. These features include indicators such as sales volume, order volume, inventory levels, and logistics timeliness. The vector represents the demand status of different machinery and equipment categories at different time stages. The first importance weight refers to the demand importance weight of each machinery and equipment category at different nodes in the supply chain.

[0037] In one possible implementation, demand-related indicator information is extracted from multi-source supply chain data, and a demand feature vector is constructed according to a unified data structure. The demand feature vectors of each category of machinery and equipment are input into a fuzzy clustering model of the data field. By calculating the similarity between different samples, a fuzzy membership relationship is formed, and the cluster center is continuously updated during the clustering process so that machinery and equipment samples with similar demand characteristics are clustered into the same category. When the clustering iteration reaches the convergence condition, the demand importance weight of each category of machinery and equipment in the supply chain node is calculated based on the membership degree of each sample in different categories, thereby obtaining the first importance weight.

[0038] Based on the above steps, this step uses fuzzy clustering analysis on demand characteristics to quantify the degree of demand impact of different mechanical equipment categories in the supply chain network, thereby improving the accuracy of subsequent supply chain network modeling.

[0039] S204. Based on the heterogeneous graph of supply chain relationships and the first importance weight, construct the supply chain network structure graph and generate the first adjacency matrix.

[0040] Among them, the supply chain network structure diagram refers to the network structure model formed by weighting the relationship between nodes by introducing demand importance weights on the basis of the supply chain relationship heterogeneous diagram, and the first adjacency matrix is ​​the matrix representation of the connection relationship between nodes in the network structure diagram.

[0041] In one possible implementation, the connection relationships between supply chain nodes are determined based on the supply chain relationship heterogeneity graph, and the transaction relationships, logistics flow relationships, and functional association relationships between each node are extracted; the demand impact weight of the supply chain node is determined based on the first importance weight, and the demand impact weight is used as the node weight; the connection relationships between supply chain nodes are weighted according to the node weight to construct a weighted supply chain network structure graph; a first adjacency matrix is ​​generated based on the weighted supply chain network structure graph, wherein the elements in the first adjacency matrix are used to represent the connection relationships between supply chain nodes and their demand impact weights.

[0042] As an example, in the supply chain scenario of construction machinery equipment, the heterogeneous supply chain relationship graph includes supplier node A, parts manufacturing node B, complete machine manufacturing node C, regional distribution node D, and end customer node E. These nodes have supply, assembly, and sales relationships. Simultaneously, step S203 obtains the first importance weight of the excavator equipment in each supply chain node; for example, the demand importance weight at complete machine manufacturing node C is 0.85, at regional distribution node D it is 0.92, and at end customer node E it is 0.95. Based on the node connection relationships in the heterogeneous supply chain relationship graph, the connections A→B, B→C, C→D, and D→E are determined, and the corresponding node demand importance weights are mapped to node weights. Then, the connection relationships between nodes are weighted using these node weights; for example, the connection weight between node C and node D is set to 0.92, and the connection weight between node D and node E is set to 0.95, thereby constructing a weighted supply chain network structure graph. Finally, a first adjacency matrix is ​​generated based on this network structure graph.

[0043] Based on the above steps, this step introduces the importance weight of demand into the supply chain network structure modeling process, so that the demand influence relationship between supply chain nodes can be expressed more accurately, thereby improving the structural expressiveness of the supply chain network model.

[0044] S205. Extract the spatial correlation features and time series features of the supply chain network structure diagram, the first adjacency matrix, and the demand feature vector, and perform basic demand forecasting to obtain the basic demand forecasting results.

[0045] Among them, spatial correlation features refer to the feature representation formed by the structural correlation information between different nodes in the supply chain network structure, while demand time series features refer to the sequence features formed by the changing patterns of mechanical equipment demand data in the time dimension.

[0046] In one possible implementation, the supply chain network input data is constructed based on the supply chain network structure diagram and the first adjacency matrix. Spatial features of the relationships between supply chain nodes are extracted using a graph structure feature extraction model to obtain spatial relationship features of the supply chain. A time series of mechanical equipment demand is constructed based on the demand feature vector, and the demand change trend is analyzed using a time series feature extraction model to obtain demand time series features. The spatial relationship features of the supply chain and the demand time series features are fused and processed, and then input into a demand forecasting model for prediction calculation to obtain the basic demand forecast results of the mechanical equipment supply chain.

[0047] Based on the above steps, this step utilizes both supply chain network structure information and demand time variation patterns for demand forecasting, enabling the forecasting model to comprehensively consider supply chain structure factors and market demand variation factors, thereby improving the accuracy and stability of the machinery and equipment supply chain demand forecasting results.

[0048] The embodiments of this application can not only reflect the structural relationships between nodes, but also reflect the impact of demand changes on the supply chain network. This enables the prediction model to comprehensively utilize supply chain structure information and demand change trends, thereby improving the accuracy and stability of demand prediction results for the machinery and equipment supply chain and solving the technical problem of low accuracy and stability of demand prediction in the prior art.

[0049] In one possible implementation of the embodiments of this application, combined with Figure 2 ,like Figure 3 As shown, the above S203 can be specifically implemented through the following S301 to S304, which are explained in detail below: S301. Extract feature information related to the demand for machinery and equipment from multi-source supply chain data and construct a demand feature vector.

[0050] The demand feature vector refers to a multi-dimensional data vector formed by combining multiple business indicators that reflect changes in the demand for machinery and equipment in a unified format. It is used to represent the demand status of a certain category of machinery and equipment within a specific time period. The feature information includes sales volume, order volume, inventory volume, and logistics timeliness information. Sales volume represents the actual number of sales within the statistical period, order volume represents the number of customer order demands, inventory volume represents the current inventory size, and logistics timeliness information represents the transportation cycle of equipment or parts in the supply chain.

[0051] In one possible implementation, data fields related to changes in machinery and equipment demand are filtered from multi-source supply chain data. For example, the number of equipment sold is extracted from sales data, the number of orders demanded is extracted from order data, the inventory level is extracted from inventory data, and the transportation cycle or delivery time is extracted from logistics data. The above data is then time-aligned according to a unified statistical period, and missing values ​​are filled in, for example, by using mean imputation or interpolation methods. Then, data of different dimensions are normalized, for example, by using the min-max normalization method to convert each indicator to a unified numerical range. Finally, the processed data are combined according to the feature dimension order to form a demand feature vector.

[0052] It should be noted that during the construction of demand feature vectors, all feature indicators should be constructed using a unified time statistical window, and data from different time periods should be avoided from being mixed. Otherwise, the demand feature vectors may lack comparability, which may affect the results of subsequent cluster analysis.

[0053] As an example, in the supply chain of construction machinery equipment, we can select a certain model of excavator with a sales volume of 120 units, an order volume of 150 units, an inventory of 60 units, and an average logistics transportation cycle of 4 days in a certain month. After normalization, we can construct the corresponding demand feature vector to represent the demand feature status of the equipment in that time period.

[0054] Based on the above steps, this step transforms key demand indicators from multi-source supply chain data into structured demand feature vectors, enabling the demand information for machinery and equipment to be expressed in a multi-dimensional feature form, thereby improving the completeness and computability of data expression in subsequent demand analysis.

[0055] S302. Initialize the membership matrix based on the requirement feature vector, and set the number of cluster categories and clustering iteration parameters.

[0056] Among them, the membership matrix is ​​a matrix used in the fuzzy clustering process to represent the degree to which each sample belongs to different cluster categories, and its element values ​​range from 0 to 1; the number of cluster categories represents the number of demand type categories into which the demand sample is divided; the clustering iteration parameter is used to control the number of iterations or convergence conditions in the clustering calculation process.

[0057] In one possible implementation, the number of samples N and the feature dimension d are determined based on the demand feature vector dataset; the number of cluster categories C is pre-set according to the distribution of demand features, for example, the samples are divided into several demand categories according to the changes in the demand features of mechanical equipment; then, an initial membership matrix U is randomly generated, and clustering iteration parameters are set, including parameters such as the maximum number of iterations and the fuzzy weighting index; after initialization, the demand feature vector dataset and the membership matrix are passed as input data to the next step for clustering calculation.

[0058] It should be noted that when initializing the membership matrix, the sum of the membership degrees of each sample in all cluster categories should be 1 to ensure the rationality of the fuzzy clustering calculation process. At the same time, the number of cluster categories should be set reasonably according to the sample size to avoid the clustering results being scattered due to too many categories or the need to distinguish differences due to too few categories.

[0059] As an example, suppose the demand feature dataset contains 100 mechanical equipment demand samples, and the number of cluster categories is set to 3, such as demand growth type, demand stability type, and demand fluctuation type; a 100×3 dimension membership matrix is ​​randomly generated, and the maximum number of iterations is set to 100 and the fuzzy weighting index is set to 2, thereby completing the initialization process of cluster calculation.

[0060] Based on the above steps, by constructing a membership matrix and setting cluster categories and iteration parameters, demand samples can participate in cluster analysis in the form of fuzzy membership relationships, thereby improving the flexibility of demand feature classification.

[0061] S303. Calculate the cluster center corresponding to each cluster category based on the membership matrix, and iteratively update the membership matrix based on the cluster centers to obtain the iteratively updated membership.

[0062] Here, cluster center refers to the representative center point of the demand feature vector in each cluster category, which is used to reflect the typical demand characteristics of the samples in that category; iterative membership update refers to the process of repeatedly updating the membership of a sample based on the distance relationship between the sample and the cluster center.

[0063] In one possible implementation, the cluster centers corresponding to each cluster category are calculated based on the membership matrix U, i.e., the cluster center vector is obtained by weighting the demand feature vectors of each sample and their membership degrees; the distance between each demand sample and each cluster center is calculated, for example, using Euclidean distance to calculate the degree of difference between the demand feature vector and the cluster center; then, the membership values ​​of each sample in different cluster categories are updated according to the distance relationship, so that samples that are closer have higher membership degrees and samples that are farther away have lower membership degrees; after the update is completed, the cluster centers are recalculated, and the above process is repeated for multiple iterations until the change in the membership matrix is ​​less than a preset threshold or the maximum number of iterations is reached, thereby obtaining the iteratively updated membership matrix.

[0064] It should be noted that during the clustering iteration process, a convergence threshold should be set for the membership update process. For example, the iteration can be stopped when the change in membership between two iterations is less than the set threshold, so as to avoid unnecessary computational consumption.

[0065] Preferred cluster centers Satisfy the following formula:

[0066] Where j represents the cluster category number, and i represents the sample index of a certain type of mechanical equipment. This represents the membership degree of the i-th sample to the j-th class. This represents the demand characteristic value of the i-th mechanical equipment.

[0067] Based on the above steps, by continuously updating the cluster centers and sample membership, the similarity relationships between demand features can be more accurately characterized, thereby improving the stability of demand classification results.

[0068] S304. When the membership matrix converges, update the membership degree according to the iterative update of the demand feature vector in each cluster category, introduce the time decay weight, calculate the demand importance weight of each mechanical equipment category in different supply chain nodes, and obtain the first importance weight.

[0069] Among them, the time decay weight refers to the weighting factor that adjusts the sample weight according to the time distance of the demand data, and is used to reflect the importance of recent demand data to the current demand analysis; the demand importance weight refers to the degree of demand influence of each category of machinery and equipment at different nodes in the supply chain.

[0070] In one possible implementation, the membership matrix is ​​iteratively updated based on the cluster centers, and the demand category distribution of each mechanical equipment sample in different cluster categories is determined based on its membership value. The time decay weight is calculated based on the time attribute of the sample data, for example, by calculating the weight of samples at different times using an exponential decay function, so that data closer to the current moment has a higher weight. The time decay weight is combined with the iteratively updated membership to calculate the decay weighted sum of each mechanical equipment sample, thereby obtaining the demand importance weight of each mechanical equipment category in different supply chain nodes, i.e., the first importance weight.

[0071] It should be noted that during the calculation of time decay weights, the time decay coefficient should be set reasonably to avoid the historical data weights decreasing too quickly or too slowly, thereby affecting the rationality of the calculation results of the demand importance weights.

[0072] Preferably, iteratively update the membership degree. Satisfy the following formula:

[0073] Where C represents the total number of clusters, and k is the traversal index. This represents the demand characteristic value of the i-th piece of machinery. For the j-th cluster center, For the k-th cluster center, Here, m represents the Euclidean distance, and m is the fuzzy weighting exponent. Preferred, first importance weight Satisfy the following formula:

[0074] in, To iteratively update the membership degree; For time decay weight, , For time decay coefficient, This is the current data collection point in time. The reference time point is i, which represents the sample index of a certain type of mechanical equipment.

[0075] Based on the above steps, by introducing a time decay factor in the demand weight calculation process, recent demand changes can be better reflected in the demand importance calculation, thereby improving the ability of demand importance weight to reflect actual market changes.

[0076] This application's embodiments enable the unified expression of machinery and equipment demand information in the form of multi-dimensional features, thereby improving the structuring degree of demand data during the analysis process. Simultaneously, by constructing a membership matrix and employing a data field fuzzy clustering method to cluster demand features, different machinery and equipment demand samples can participate in classification calculations with fuzzy membership relationships, thus more accurately reflecting the similarity between demand features. Furthermore, by continuously updating the cluster centers and membership matrix during the clustering iteration process, the demand feature classification results gradually stabilize, thereby improving the reliability of the demand feature analysis results. In addition, by introducing a time decay weight in the demand importance calculation process, recent demand changes are more effectively reflected in the demand importance weight calculation, thus enabling the obtained first importance weight to more realistically reflect the current demand changes in the machinery and equipment supply chain.

[0077] In one possible implementation of the embodiments of this application, combined with Figure 2 ,like Figure 4 As shown, the above S205 can be implemented through the following S401 to S403, which are explained in detail below: S401. Based on the supply chain network structure diagram and the first adjacency matrix, a graph convolutional network is used to extract spatial features of the relationship between supply chain nodes, thereby obtaining the spatial relationship features of the supply chain.

[0078] Among them, graph convolutional networks are a type of deep learning model used to process graph structure data. They extract structural association information between nodes by performing convolution operations on the features of nodes and their neighboring nodes. Supply chain spatial association features refer to the structural features extracted by analyzing the connection relationships and node attributes between different nodes in the supply chain network, which are used to characterize the influence relationships between supply chain nodes.

[0079] In one possible implementation, based on the supply chain network structure graph and the first adjacency matrix, the attributes of each node in the supply chain network (such as node demand feature vectors or node demand importance weights) are used as node feature input data. The first adjacency matrix is ​​normalized to obtain a standardized adjacency matrix, which is then input into the graph convolutional network model along with the node feature matrix. The node features are convolved through graph convolutional layers, and in each convolutional layer, the node's own features and the features of its neighboring nodes are weighted and aggregated to obtain an updated node feature representation. After multi-layer graph convolutional computation, a set of node feature vectors containing supply chain network structure relationship information is output, and this set is used as the supply chain spatial correlation feature output for subsequent demand prediction model input.

[0080] It should be noted that during graph convolution calculation, the number of convolutional layers and the dimension of hidden layers can be reasonably set according to the scale of the supply chain network to avoid excessive computational complexity due to too many model layers. At the same time, the stability of model training can be improved by normalizing node features or regularizing weights.

[0081] Based on the above steps, this step uses graph convolutional networks to extract features from the supply chain network structure, enabling the deep learning model to effectively capture the structural relationship information between supply chain nodes, thereby improving the ability to utilize supply chain network structure information in the demand forecasting process.

[0082] S402. Construct a mechanical equipment demand time series based on the demand feature vector, and use a recurrent neural network to extract time features from the demand time series to obtain demand time series features.

[0083] Among them, demand time series refers to the sequence of mechanical equipment demand data arranged in chronological order, used to reflect the pattern of equipment demand changing over time; recurrent neural network is a neural network structure that can process sequence data. It uses a recurrent connection structure to memorize and model historical sequence information, thereby extracting time dimension features.

[0084] In one possible implementation, the demand data for each category of machinery and equipment is arranged in chronological order based on the demand feature vector data, for example, forming a demand time series with monthly or quarterly time units; the demand time series data is input into a recurrent neural network model, and feature mapping is performed on the sequence data in the input layer; the historical demand data is calculated step by step through the hidden layer of the recurrent neural network, enabling the model to learn the time dependency of demand changes; after the cyclic calculation is completed, the time series feature vector is extracted through the output layer, and the vector is output as the demand time series feature.

[0085] It should be noted that during the construction of demand time series, the time series data should be continuous and the time intervals should be consistent. At the same time, during the model training phase, the demand series can be smoothed or outlier handled to reduce the adverse effects of abnormal demand data on the model learning process.

[0086] Based on the above steps, this step utilizes recurrent neural networks to extract features from the demand time series, enabling the model to capture the temporal patterns of changes in mechanical equipment demand, thereby improving the demand forecasting model's ability to identify changes in demand trends.

[0087] S403. Integrate the spatial correlation characteristics of the supply chain and the time series characteristics of demand, and make predictions through a demand forecasting model to obtain basic demand forecast results.

[0088] Among them, the demand forecasting model is a forecasting model used to calculate future demand values ​​based on input features, and can be implemented using neural network regression models or other machine learning forecasting models.

[0089] In one possible implementation, the spatial correlation characteristics of the supply chain and the time series characteristics of demand are processed in a unified dimension, for example, by mapping the feature dimensions through a fully connected layer; the two types of features are then fused, for example, by feature concatenation or weighted fusion to form a comprehensive feature vector; the fused comprehensive feature vector is then input into a demand forecasting model, and the model calculates the demand forecast values ​​for each category of machinery and equipment in the future time period; finally, the forecast results are output as the basic demand forecast results for the machinery and equipment supply chain.

[0090] It should be noted that during feature fusion, the influence of spatial and temporal features can be balanced through feature normalization or weight adjustment to avoid one type of feature having too high a weight in the prediction process and affecting the stability of the prediction results.

[0091] Based on the above steps, by integrating supply chain spatial structure information with demand temporal variation patterns into a model, the demand forecasting model can simultaneously utilize supply chain network relationships and demand temporal trend information, thereby improving the accuracy and stability of basic demand forecasting results.

[0092] In one possible implementation, combining Figure 2 ,like Figure 5 As shown, following S205, the supply chain demand forecasting method based on multi-source data fusion provided in this application embodiment further includes the following S501 to S503: S501. Obtain multi-source supply chain data of historical machinery and equipment, and construct a demand fluctuation risk assessment vector.

[0093] The multi-source supply chain data for historical machinery and equipment includes: historical demand data, historical inventory levels, historical logistics delay information, and historical market volatility indicators. The demand volatility risk assessment vector is a set of characteristic indicators used to characterize the degree of demand instability in the machinery and equipment supply chain. It is formed by extracting and combining indicators such as demand changes, inventory changes, and logistics delays from historical supply chain operation data, and is used to reflect the level of supply chain demand volatility risk.

[0094] In one possible implementation, historical multi-source supply chain data of machinery and equipment over a certain period of time is obtained from a supply chain database, such as data records from the past 12 or 24 months. Demand fluctuation-related indicators are extracted from the acquired historical data, including historical demand data, historical inventory levels, historical logistics delay information, and historical market fluctuation indicators. Historical demand data can be composed of historical sales volume or order volume, historical inventory levels can be represented by inventory quantity or inventory turnover rate for each time period, historical logistics delay information can be calculated from the difference between the logistics transportation cycle and the planned transportation cycle, and historical market fluctuation indicators can be represented by the change in industry index or raw material price index. Then, the above indicators are time-aligned, outlier removed, and standardized, and combined according to a unified dimension to form a demand fluctuation risk assessment vector.

[0095] It should be noted that when constructing a demand fluctuation risk assessment vector, all indicators should be sourced from the same statistical period, such as monthly or quarterly data, and data of different dimensions should be normalized to avoid the impact of differences in the numerical range of different indicators on the subsequent risk assessment results.

[0096] Based on the above steps, by comprehensively extracting and vectorizing demand changes and supply chain operation indicators from historical supply chain data, demand fluctuation risks can be expressed in the form of structured data, thereby improving the data expression capability in the demand risk assessment process.

[0097] S502. Based on the demand fluctuation risk assessment vector, conduct a demand fluctuation risk assessment on the basic demand forecast results to obtain the demand risk assessment level.

[0098] Demand fluctuation risk assessment refers to the process of quantitatively analyzing the degree of instability of supply chain demand based on the demand fluctuation risk assessment vector; the demand risk assessment level is used to indicate the degree of supply chain demand fluctuation risk, and can be divided into low risk, medium risk and high risk levels.

[0099] In one possible implementation, a demand fluctuation risk assessment vector is used as input data, and the basic demand forecast results are used as reference data to input into the demand risk assessment model. The demand fluctuation risk assessment vector is then input into a BP neural network model. In the input layer, features of indicators such as historical demand fluctuations, inventory changes, logistics delays, and market fluctuations are mapped, and the nonlinear correlations between the indicators are learned through hidden layer neurons. Subsequently, the output layer predicts and calculates the demand fluctuation risk and outputs the risk assessment result value. Finally, a risk level threshold is set based on the output value. For example, when the risk value is less than the first threshold, it is judged as low risk; when the risk value is between the first and second thresholds, it is judged as medium risk; and when the risk value is greater than the second threshold, it is judged as high risk, thereby obtaining the demand risk assessment level.

[0100] It should be noted that during the training of the BP neural network, historical supply chain demand fluctuation data can be used as training samples, and the network weight parameters can be continuously adjusted through the error backpropagation algorithm to improve the risk assessment model's ability to identify demand fluctuation patterns.

[0101] Based on the above steps, this step uses a neural network model to assess the risk of demand fluctuations, enabling the supply chain demand forecast results to be combined with historical supply chain operation data for risk identification, thereby improving the adaptability of the demand forecast results to market changes.

[0102] S503. Based on the risk assessment level, the basic demand forecast results are collaboratively corrected based on the supply chain relationship heterogeneity diagram to obtain the final supply chain demand forecast results.

[0103] Collaborative correction refers to the process of adjusting and optimizing the basic demand forecast results based on the demand risk assessment results, according to the supply chain network structure and the relationships between nodes; the final supply chain demand forecast result refers to the final demand forecast value obtained after comprehensively considering the demand forecast results and demand fluctuation risk factors.

[0104] In one possible implementation, a demand forecasting correction strategy is determined based on the risk assessment level. For example, the safety stock forecasting coefficient is increased in high-risk scenarios, moderate demand adjustments are made in medium-risk scenarios, and the original forecast results are maintained in low-risk scenarios. By combining the heterogeneous diagram of supply chain relationships, the correlation between each supply chain node is analyzed, and the demand forecast results are propagated and corrected according to the connection weight between nodes. For example, when the demand risk of a distribution node in a certain region is high, the demand forecast values ​​of upstream manufacturing nodes and parts supply nodes can be adjusted accordingly through the supply chain network relationship. Finally, the final supply chain demand forecast result is obtained through demand correction calculation.

[0105] It should be noted that during the process of demand coordination and adjustment, the connection weights between supply chain nodes and the degree of demand impact should be considered to avoid the excessive amplification of the overall supply chain forecast results due to changes in demand at a certain node.

[0106] As an example, in the machinery and equipment supply chain, if the demand forecast for regional distribution node D is 100 units and the risk level is high, the system can adjust its demand forecast proportionally according to the supply chain network structure, for example, to 115 units, and at the same time adjust the production plan of the upstream manufacturing node accordingly, thereby obtaining the final demand forecast result.

[0107] Based on the above steps, by combining the results of demand fluctuation risk assessment with the supply chain network structure to collaboratively correct the demand forecast results, the demand forecast results can simultaneously consider the demand change trend and the supply chain operation risk, thereby improving the reliability of the demand forecast results for the machinery and equipment supply chain.

[0108] Although this application has been described herein in conjunction with various embodiments, those skilled in the art, by reviewing the accompanying drawings and the appended claims, will understand and implement other variations of the disclosed embodiments in carrying out the claimed application. In the claims, the word "comprising" does not exclude other components or steps, and "a" or "an" does not exclude multiple components. A single processor or other unit can implement several functions listed in the claims. While different dependent claims may recite certain measures, this does not mean that these measures cannot be combined to produce good results.

[0109] Although this application has been described in conjunction with specific features and embodiments, it is obvious that various modifications and combinations can be made thereto without departing from the spirit and scope of this application. Accordingly, this specification and drawings are merely exemplary illustrations of this application as defined by the appended claims, and are considered to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from the spirit and scope of this application. Thus, if such modifications and modifications of this application fall within the scope of the claims of this application and their equivalents, this application is also intended to include such modifications and modifications.

Claims

1. A supply chain demand forecasting method based on multi-source data fusion, characterized in that, include: Acquire multi-source supply chain data for machinery and equipment; The multi-source supply chain data includes: internal supply chain data and external market environment data; the internal supply chain data includes: sales data, order data, inventory data, logistics data, and price data. Based on the aforementioned multi-source supply chain data, a heterogeneous supply chain relationship diagram is constructed; The demand feature vector of the multi-source supply chain data is extracted, and the feature demand vector is clustered using the data field fuzzy clustering method to obtain the first importance weight; the first importance weight refers to the demand importance weight of each category of machinery and equipment in different supply chain nodes. Based on the heterogeneous supply chain relationship graph and the first importance weight, a supply chain network structure graph is constructed, and a first adjacency matrix is ​​generated; the first adjacency matrix is ​​used to represent the relationship between supply chain nodes. Extract the spatial correlation features and demand time series features of the supply chain network structure diagram, the first adjacency matrix, and the demand feature vector, and perform basic demand forecasting to obtain the basic demand forecasting results.

2. The method of claim 1, wherein, After obtaining the basic demand forecast results, the method further includes: Acquire multi-source supply chain data of historical machinery and equipment, and construct a demand fluctuation risk assessment vector; the multi-source supply chain data of historical machinery and equipment includes: historical demand data, historical inventory levels, historical logistics delay information, and historical market fluctuation indicators; Based on the aforementioned demand fluctuation risk assessment vector, a demand fluctuation risk assessment is performed on the basic demand forecast results to obtain the demand risk assessment level. Based on the risk assessment level, the basic demand forecast results are collaboratively corrected using the supply chain relationship heterogeneity diagram to obtain the final supply chain demand forecast results.

3. The method of claim 1, wherein, The construction of a heterogeneous supply chain relationship diagram based on the multi-source supply chain data includes: Identify the main entities of supply chain nodes in multi-source supply chain data, and determine the node set in the supply chain diagram as supplier nodes, machinery and equipment manufacturing nodes, parts manufacturing nodes, regional distribution nodes, and end customer nodes; Based on sales data, order data, and logistics data from multi-source supply chain data, the transaction relationships and logistics flow relationships between supply chain nodes are determined, and the functional association relationships are determined based on the matching relationships between machinery and parts, thus obtaining the set of edge relationships between nodes; A heterogeneous graph structure of supply chain relationships is constructed based on the set of nodes and the set of edge relationships.

4. The method of claim 1, wherein, The process involves extracting the demand feature vector from the multi-source supply chain data and performing cluster analysis on the feature demand vector using a data field fuzzy clustering method to obtain the first importance weight, including: Extract feature information related to the demand for machinery and equipment from multi-source supply chain data, and construct a demand feature vector. The feature information includes sales volume, order volume, inventory volume, and logistics timeliness information. The membership matrix is ​​initialized based on the aforementioned requirement feature vector, and the number of cluster categories and clustering iteration parameters are set. The cluster center corresponding to each cluster category is calculated based on the membership matrix, and the membership matrix is ​​iteratively updated based on the cluster centers to obtain the iteratively updated membership. When the membership matrix converges, the membership degree is updated iteratively according to the demand feature vector in each cluster category. A time decay weight is introduced, and the demand importance weight of each mechanical equipment category in different supply chain nodes is calculated to obtain the first importance weight.

5. The method according to claim 4, characterized in that, The cluster center Satisfy the following formula: Where j represents the cluster category number, and i represents the sample index of a certain type of mechanical equipment. This represents the membership degree of the i-th sample to the j-th class. This represents the demand characteristic value of the i-th mechanical equipment.

6. The method according to claim 4, characterized in that, The iterative update of membership degree Satisfy the following formula: Where C represents the total number of clusters, and k is the traversal index. This represents the demand characteristic value of the i-th piece of machinery. For the j-th cluster center, For the k-th cluster center, Here, m represents the Euclidean distance, and m is the fuzzy weighting exponent. First importance weight Satisfy the following formula: in, To iteratively update the membership degree; For time decay weight, , For time decay coefficient, This is the current data collection point in time. The reference time point is i, which represents the sample index of a certain type of mechanical equipment.

7. The method according to claim 1, characterized in that, The step of constructing a supply chain network structure graph and generating a first adjacency matrix based on the supply chain relationship heterogeneity graph and the first importance weight includes: The connection relationships between supply chain nodes are determined based on the aforementioned supply chain relationship heterogeneity diagram, and the transaction relationships, logistics flow relationships, and functional association relationships between each node are extracted. The demand impact weight of each supply chain node is determined based on the first importance weight, and the demand impact weight is used as the node weight. The connection relationships between supply chain nodes are weighted according to the node weights to construct a weighted supply chain network structure diagram; A first adjacency matrix is ​​generated based on the weighted supply chain network structure diagram, wherein the elements in the first adjacency matrix are used to represent the connection relationship between supply chain nodes and their demand influence weights.

8. The method according to claim 1, characterized in that, The process of extracting spatial correlation features and demand time series features from the supply chain structure diagram, the first adjacency matrix, and the demand feature vector, and performing basic demand forecasting to obtain basic demand forecasting results includes: Based on the supply chain network structure diagram and the first adjacency matrix, a graph convolutional network is used to extract spatial features of the relationships between supply chain nodes, thus obtaining the spatial relationship features of the supply chain. A time series of mechanical equipment demand is constructed based on the demand feature vector, and a recurrent neural network is used to extract time features from the demand time series to obtain demand time series features; The supply chain spatial correlation characteristics and the demand time series characteristics are fused together and predicted using a demand forecasting model to obtain the basic demand forecasting results.

9. The method according to claim 2, characterized in that, The process of assessing demand fluctuation risk based on the aforementioned demand fluctuation risk assessment vector, and obtaining a demand risk assessment level, includes: Input the demand fluctuation risk assessment vector into the BP neural network model, and use the hidden layer neurons to perform feature mapping on the correlation between various risk indicators to obtain the hidden layer output; Risk prediction calculations are performed based on the output of the hidden layer to obtain a risk assessment level; the risk assessment level is used to characterize the risk level of demand fluctuations in the machinery and equipment supply chain.

10. A supply chain demand forecasting system based on multi-source data fusion, characterized in that, The system includes: a data acquisition device and an electronic device; The data acquisition device is used to acquire multi-source supply chain data of the machinery and equipment; the multi-source supply chain data includes: internal supply chain data and external market environment data; the internal supply chain data includes: sales data, order data, inventory data, logistics data and price data; The electronic device is used to construct a heterogeneous supply chain relationship graph based on the multi-source supply chain data; extract the demand feature vectors from the multi-source supply chain data, and perform cluster analysis on the feature demand vectors using data field fuzzy clustering to obtain a first importance weight; the first importance weight refers to the demand importance weight of each category of machinery and equipment in different supply chain nodes; construct a supply chain network structure graph based on the heterogeneous supply chain relationship graph and the first importance weight, and generate a first adjacency matrix; the first adjacency matrix is ​​used to characterize the correlation between supply chain nodes; extract the spatial correlation features and demand time series features of the supply chain structure graph, the first adjacency matrix, and the demand feature vectors, and perform basic demand forecasting to obtain basic demand forecasting results.

Images