Industrial supply chain management coordination method based on cloud computing and big data analysis

Abstract

The invention discloses an industrial supply chain management coordination method based on cloud computing and big data analysis, and relates to the technical field of supply chain management, and the method comprises the steps: constructing a federated learning network, loading local private data through a supply chain node, inputting the local private data into the federated learning network, and generating an encryption gradient parameter; when the order fluctuation rate exceeds a preset fluctuation threshold value and a logistics delay event occurs, converting the encryption gradient parameter into a space-time correlation feature; aggregating the space-time correlation features based on an attention weight algorithm to generate an aggregated feature vector; updating the weight of the federated learning network through the market public opinion features and the aggregated feature vectors, and generating a global causal feature map; and analyzing the global causal feature map according to a causal discovery algorithm, constructing a multi-level intervention path, positioning a root cause node through anti-factual reasoning, and generating a root cause tracing report. According to the method, encryption collaboration of local data of supply chain nodes is realized by constructing the federated learning network, and the response efficiency of an interrupt event is improved from delay to a nearly real-time level.

Description

Technical Field

[0001] This invention relates to the field of supply chain management technology, and in particular to an industrial supply chain management coordination method based on cloud computing and big data analysis. Background Technology

[0002] In recent years, cloud computing and big data analytics technologies have continued to deepen their integration in the field of industrial supply chain management, gradually forming a distributed collaborative decision-making system. Cloud computing platforms achieve elastic scaling of computing power through virtualized resource pools, providing high-concurrency processing capabilities for massive supply chain data (such as order flow, logistics trajectory, and inventory status); while big data analytics technologies rely on machine learning models to mine the correlation patterns of multi-source heterogeneous data, supporting core scenarios such as supplier performance evaluation, demand forecasting, and risk warning.

[0003] Current industrial supply chain management methods have shortcomings. First, data collaboration in a cloud computing environment requires the centralized uploading of original sensitive information. Although encryption is used to ensure transmission security, encrypted data cannot directly participate in collaborative computing. Frequent decryption operations lead to response delays, making it difficult to meet the timeliness decision-making needs of sudden supply chain disruptions. Second, big data analysis solutions mostly rely on static historical models (such as clustering algorithms) to identify supplier risks, lacking the ability to build dynamic causal reasoning chains and ignoring the cross-node transmission effects of real-time factors such as market sentiment and logistics topology, making it difficult to implement coordination instructions. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides an industrial supply chain management coordination method based on cloud computing and big data analysis to solve the problems of real-time response delay and lack of dynamic response.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: This invention provides an industrial supply chain management coordination method based on cloud computing and big data analytics, comprising: A federated learning network is constructed, and local private data is loaded into the federated learning network through supply chain nodes to generate encrypted gradient parameters. When the order volatility exceeds a preset volatility threshold and a logistics delay event occurs, the encrypted gradient parameters are converted into spatiotemporal correlation features. The spatiotemporal correlation features are aggregated based on an attention weight algorithm to generate an aggregated feature vector. The weights of the federated learning network are updated through market sentiment features and the aggregated feature vector to generate a global causal feature map. The global causal feature map is analyzed according to a causal discovery algorithm to construct a three-level intervention path. The root cause node is located through counterfactual reasoning, and a root cause tracing report is generated. The root cause tracing report includes primary and secondary cause types, intervention paths, and confidence ratings. When the confidence rating exceeds a confidence threshold, a backup supplier switching ratio is generated based on the primary and secondary cause types, and a logistics path reallocation plan is generated based on the intervention path. The backup supplier switching ratio and the logistics path reallocation plan are integrated to generate a dynamic coordination instruction. The dynamic coordination instruction is distributed to supply chain nodes and procurement allocation adjustments, transportation path updates, and resource supply guarantees are executed to generate a management coordination report.

[0007] As a preferred embodiment of the industrial supply chain management coordination method based on cloud computing and big data analysis described in this invention, the supply chain nodes include supplier nodes, manufacturer nodes, and logistics provider nodes.

[0008] As a preferred embodiment of the industrial supply chain management coordination method based on cloud computing and big data analysis described in this invention, the specific steps for generating the encrypted gradient parameters are as follows: A federated learning network framework is built based on the hardware layer and protocol layer. The federated learning network framework is transformed into a federated learning network through node collaboration and gradient aggregation. Supply chain nodes load local private data through local data interfaces and input it into the federated learning network, using homomorphic encryption algorithms to generate encrypted gradient parameters.

[0009] As a preferred embodiment of the industrial supply chain management coordination method based on cloud computing and big data analysis described in this invention, the specific steps for converting the encrypted gradient parameters into spatiotemporal correlation features are as follows: Order volatility is obtained through supply chain management structure, and logistics delay events are determined through GPS trajectory analysis; when order volatility exceeds a preset volatility threshold and a logistics delay event occurs simultaneously, an activation signal is generated. The feature converter receives encrypted gradient parameters and activation signals, and converts them to generate spatiotemporally correlated features.

[0010] As a preferred embodiment of the industrial supply chain management coordination method based on cloud computing and big data analysis described in this invention, the specific steps for generating the aggregated feature vector are as follows: Calculate similarity scores between spatiotemporally related features; Attention weights are assigned based on similarity scores to generate aggregated feature vectors.

[0011] As a preferred embodiment of the industrial supply chain management coordination method based on cloud computing and big data analysis described in this invention, the specific steps for generating a global causal feature map are as follows: Market sentiment texts are obtained from public data sources, and market sentiment features are extracted through natural language processing. The market sentiment features and aggregated feature vectors are then concatenated into a fused feature vector. The fused feature vectors generate weight increments through dynamic weight field modulation, and the weights of the federated learning network are updated based on the weight increments. The updated federated learning network weights are input into the federated learning network, and the global causal feature map is output.

[0012] As a preferred embodiment of the industrial supply chain management coordination method based on cloud computing and big data analysis described in this invention, the specific steps for constructing multiple three-level intervention paths are as follows: Extract the causal relationship strength between supply chain nodes in the global causal feature map and calculate the node influence factor; Based on the node impact factor, a three-level intervention path is generated by screening key nodes and expanding them at multiple levels.

[0013] As a preferred embodiment of the industrial supply chain management coordination method based on cloud computing and big data analysis described in this invention, the specific steps for generating the root cause traceability report are as follows: Perform counterfactual reasoning on the three-level intervention path to generate a counterfactual path; The effect difference between the three-level intervention path and the counterfactual path is calculated, and the supply chain node with the largest effect difference is marked as the root cause node. The root cause node information is integrated to generate a root cause tracing report.

[0014] As a preferred embodiment of the industrial supply chain management coordination method based on cloud computing and big data analysis described in this invention, the specific steps for generating dynamic coordination instructions are as follows: When the confidence rating exceeds the preset confidence threshold, calculate the alternative supplier switching ratio in the primary and secondary cause types; Based on the intervention path, high-risk paths are identified and low-risk alternative paths are matched to generate logistics path redistribution schemes. The system integrates the backup supplier switching ratio and logistics route reallocation scheme, and generates dynamic coordination instructions through digital certificates and timing services.

[0015] As a preferred embodiment of the industrial supply chain management and coordination method based on cloud computing and big data analysis described in this invention, the specific steps for generating the management and coordination report are as follows: Dynamic coordination instructions are distributed to supply chain nodes through a blockchain peer-to-peer protocol, and the supply chain nodes verify the integrity and authenticity of the dynamic coordination instructions through digital signatures. Supplier nodes reallocate purchase orders based on the backup supplier switching ratio; Based on the logistics route redistribution scheme, logistics nodes terminate transportation tasks on high-risk routes, activate low-risk alternative routes, and schedule cargo volume according to the allocation weight. The manufacturer node adjusts raw material inventory thresholds and reorders production line priorities based on the updated procurement and logistics plan; The system aggregates the execution results from supplier nodes, manufacturer nodes, and logistics provider nodes to generate a management coordination report.

[0016] The beneficial effects of this invention are as follows: by constructing a federated learning network to achieve encrypted collaboration of local data at supply chain nodes, while strictly protecting the privacy of business secrets, the response efficiency for sudden interruption events is improved from traditional delays to near real-time levels; by combining causal graphs and counterfactual reasoning to dynamically generate root cause tracing reports, the primary and secondary responsible nodes are accurately distinguished and cross-node coordination instructions are automatically triggered, thus realizing closed-loop autonomy of the entire chain of procurement, logistics and production resources. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart of an industrial supply chain management coordination method based on cloud computing and big data analytics.

[0019] Figure 2 A flowchart for generating associated spatiotemporal features.

[0020] Figure 3 A flowchart for generating a root cause analysis report.

[0021] Figure 4 A flowchart for generating management coordination reports. Detailed Implementation

[0022] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0023] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0024] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0025] Reference Figures 1-4 This is one embodiment of the present invention, which provides an industrial supply chain management coordination method based on cloud computing and big data analysis, including the following steps: S1. Construct a federated learning network. Load local private data into the federated learning network through supply chain nodes to generate encrypted gradient parameters. When the order volatility exceeds the preset volatility threshold and a logistics delay event occurs, convert the encrypted gradient parameters into spatiotemporal correlation features. S1.1 Construct a federated learning network framework based on the hardware layer and protocol layer, and transform the federated learning network framework into a federated learning network through node collaboration and gradient aggregation; It should be noted that the hardware layer is constructed by applying for physical host resources through a cloud computing platform and configuring a virtual private cloud isolation environment, dividing the supplier, manufacturer, and logistics provider into independent security zones, allocating dedicated computing and storage resource groups to each independent security zone, pre-installing local data access interfaces for each supply chain node, establishing encrypted communication links between supply chain nodes, binding the physical host and supply chain node identities with digital certificate authentication mechanisms, configuring access control policies to restrict cross-regional data flow, dynamically scaling the computing and storage resource groups according to the supply chain node load, and monitoring supply chain link latency and bandwidth usage in real time. The federated average protocol defines the gradient aggregation cycle time, sets encrypted communication rules, and uses transport layer security protocols and digital certificates for two-way authentication; configures node collaborative training rules: each supply chain node calculates gradients locally, uses homomorphic encryption algorithms to encrypt gradient parameters, and aggregates encrypted gradients through protocols and updates global parameters to complete the construction of the protocol layer; The protocol layer configures an independent encrypted communication group for each independent security zone. Each encrypted communication group is bound to a unique digital certificate and a transport layer security protocol key pair. The network interface of the physical host in the hardware layer establishes a one-to-one connection with the virtual endpoint of the encrypted communication group in the protocol layer. Access requests from supply chain nodes within an independent security zone are transmitted to the corresponding encrypted communication group in the protocol layer through the physical link in the hardware layer. The protocol layer verifies the digital certificate and executes the encrypted communication rules. The protocol layer configures an independent channel for cross-regional communication. When supply chain nodes need to interact across regions, the protocol layer automatically calls the encrypted communication group key of the target independent security zone for secondary encryption, ensuring that data is transmitted under the dual protection of hardware layer isolation and protocol layer encryption. This completes the full-domain collaborative binding of physical resources and communication rules, enabling supply chain nodes to execute collaborative training tasks according to the protocol in a secure and isolated environment, thus completing the construction of the federated learning network framework. The hardware layer allocates physical host resources to load local private data from supply chain nodes. Each supply chain node executes local gradient calculation by calling the federated averaging protocol defined by the protocol layer within its independent secure area. After encrypting the gradient parameters using a homomorphic encryption algorithm, the parameters are transmitted to the protocol layer via an encrypted communication link. The protocol layer aggregates the encrypted gradient parameters of all supply chain nodes, decrypts them, updates the global parameters, and synchronizes them to each supply chain node. The local calculation, encrypted transmission, and aggregation update process is repeated until the predicted loss function values ​​of all supply chain nodes change by less than the convergence threshold (based on the definition of loss function stability, such as 0.1%) for three consecutive iterations, and the average gradient norm of the gradient aggregation result approaches zero. This is considered as the federated learning network converging, thus forming a federated learning network.

[0026] S1.2 Supply chain nodes load local private data through local data interfaces and input it into the federated learning network, using homomorphic encryption algorithms to generate encrypted gradient parameters; It should be noted that supply chain nodes include supplier nodes, manufacturer nodes, and logistics provider nodes; Supply chain nodes connect to the business database via local data interfaces, loading local private data into the memory buffer of the computing resource group; they initialize the federated learning network parameters and perform forward propagation to calculate predicted values. The difference between the predicted and actual values ​​(based on local private data) is compared using a loss function. The encryption gradient parameters of the federated learning network are calculated using backpropagation and homomorphic encryption algorithms, expressed as follows: ; ; ; ; in, Indicates the encryption gradient parameters; The original gradient parameters are the partial derivatives of the loss function with respect to the parameters of the federated learning network. Represents a random integer, with a range of values ​​of 100. This is dynamically generated for encryption purposes, designed to prevent cracking. Indicates the parameters of the transport layer security protocol key pair; Modulo operation refers to the remainder after dividing two integers; Indicates the parameters of the federated learning network; This represents the value of the loss function; This represents the total amount of local private data; This represents a local private data index, with a value of 1- ; Indicates the first Predicted values ​​for each data point; Indicates the first The actual value of each data point; This represents local private data; This indicates the forward computation function.

[0027] S1.3 Obtain order volatility through the supply chain management structure and determine logistics delay events through GPS trajectory analysis; when order volatility exceeds the preset volatility threshold and logistics delay events occur simultaneously, generate an activation signal; It should be noted that the current order volume data is obtained through the local data interface, the difference between the current order volume data and the average order volume of the past seven days is calculated, and the ratio of the difference to the average order volume of the past seven days is used as the order volatility. The GPS trajectory analysis structure analyzes the real-time location and planned route of the logistics vehicle. When the actual arrival time is more than 2 hours later than the promised arrival time (based on the definition of road freight transportation contract), it is judged as a logistics delay event. If the order volatility exceeds a preset volatility threshold (defined based on the correlation analysis of historical supply chain disruption events and order volatility, such as 15%) and a logistics delay event occurs simultaneously, an activation signal with a timestamp is generated.

[0028] S1.4 The feature converter receives the encrypted gradient parameters and activation signal, and converts them to generate spatiotemporal correlated features.

[0029] It should be noted that the original gradient parameters are restored using the homomorphic decryption algorithm, and the expression is: ; ; in, This represents an intermediate calculated value, which is a transitional variable used to solve for the original gradient parameters; This represents the component values ​​of the encryption gradient parameters, i.e., the values ​​of a single encryption gradient. This represents the security parameters of the private key, which determine the bit length of the key; the larger the bit length, the more difficult it is to crack. This represents a derived value, which is an intermediate value calculated during the private key generation process and used for subsequent decryption operations; Extract the timestamp and geographical coordinates (latitude and longitude of supplier nodes) from the activation signal; The Geohash encoding algorithm is used to convert two-dimensional latitude and longitude coordinates into geocoded strings. Specifically, the two-dimensional latitude and longitude coordinates are recursively divided into a grid on the Earth's surface using a binary search method. In each round of division, binary bits are allocated according to the position of the two-dimensional latitude and longitude coordinate points in the sub-grid. Every 5 bits of binary data are combined into a group, and the Base32 encoding table (a standard component of the Geohash algorithm and its built-in encoding table) is queried to map them into characters. This process is iterated until a preset precision is achieved (defined based on a balance between business scenario requirements and privacy protection, such as 6-bit characters with a precision of 1.2km), and then the geocoded string is output. The timestamp and the duration of logistics delay are concatenated to form a time feature vector; the original gradient parameters, geocoded strings, and time feature vector are concatenated to form a spatiotemporal correlation feature.

[0030] Existing technologies centrally collect raw data and plaintext gradients from supply chain nodes through a central server, and perform unified calculations and feature generation on the central server. While this can achieve basic collaborative training, it suffers from drawbacks such as high risk of raw data leakage, low transmission efficiency, and inability to associate with spatiotemporal context. This solution processes raw data locally at distributed supply chain nodes and generates encrypted gradient parameters using a homomorphic encryption algorithm. Combined with activation signal-driven spatiotemporal feature transformation, it achieves the fusion of encrypted gradient parameters with geographical coordinates and timestamps, achieving multi-dimensional feature collaboration under privacy protection. This solves the data security bottleneck and spatiotemporal feature loss problem of centralized architecture.

[0031] S2. Aggregate spatiotemporal correlation features based on the attention weight algorithm to generate an aggregated feature vector; update the weights of the federated learning network using market sentiment features and the aggregated feature vector to generate a global causal feature map; S2.1 Calculate the similarity scores between spatiotemporally related features, assign attention weights based on the similarity scores, and generate aggregated feature vectors; It should be noted that, based on a large-scale geocoded corpus (such as the Global Open Street Map dataset), the frequency of occurrence of characters in all geocoded strings is counted, and a geodetic dense vector for each character is initialized; the geodetic dense vector is adjusted through a character relationship optimization task so that the geodetic dense vector values ​​of characters that are adjacent in location or frequently co-occur are close, and the optimized geodetic dense vector is stored as a geo-embedding lookup table. Extract the original gradient parameters, geocoded strings, and time feature vectors from the spatiotemporal correlation features; segment the geocoded strings into character sequences, obtain the character geodetic dense vector corresponding to each character through a geo-embedding lookup table, and concatenate all the character geodetic dense vectors in order and compress them into a fixed-dimensional geo-embedding vector through a fully connected layer; The original gradient parameters, geographic embedding vector, and temporal feature vector are concatenated into a unified feature matrix; the similarity score between every two unified features in the unified feature matrix is ​​calculated, expressed as follows: ; in, This indicates the similarity score; This indicates a unified feature dimension of 128. This represents the dimension index, with values ​​ranging from 1 to... ; Represents the first in the unified characteristic matrix Any two unified feature components in dimension; The similarity scores are normalized into attention weights using the Softmax function, expressed as follows: ; in, Indicates the first Attention weights for each feature; Indicates the first Similarity scores for each feature; This represents the total number of uniform features in the uniform feature matrix; This represents an index in the unified feature matrix, with values ​​ranging from 1 to... , It is a fixed index, specifying the feature position where the weight is to be calculated, such as , indicating the calculation of the weight of the first feature. ; This represents a temporary index used during the summation process, used to iterate through all the common features and to calculate the denominator. Indicates the first Similarity scores for each feature; The aggregated feature vector is generated by weighted summation of each unified feature based on the attention weights, expressed as follows: ; in, Represents aggregated feature vectors; Indicates the first A unified characteristic.

[0032] S2.2 Extract market sentiment features through natural language processing, and concatenate the market sentiment features and aggregated feature vectors into a fused feature vector; It should be noted that market sentiment text is obtained in real time from public data sources (such as news websites or social media APIs), and the market sentiment text is cleaned through text preprocessing: special symbols and HTML / XML tags are removed from the market sentiment text, stop words in the language are deleted, continuous number sequences are replaced with uniform placeholders, the remaining text is segmented, all letters are converted to lowercase and extra spaces are removed, and the segmented sequence is output to complete the preprocessing and cleaning of market sentiment text. Collect market sentiment text data, extract high-frequency words and filter noise words through word segmentation, initialize the sentiment density vector of each word, and use the sentiment density vector optimization task to adjust the value of the sentiment density vector so that the sentiment density vectors of semantically similar words are close in distance. Store the optimized sentiment density vector as a sentiment embedding lookup table. The word embedding method is used to convert the word segmentation sequence into a numerical market sentiment feature vector. Specifically, the dense word vector corresponding to each word segmentation word in the word segmentation sequence is obtained through the sentiment embedding lookup table, and the average value of the dense word vectors of the entire word segmentation sequence is calculated as the basic text representation. The weight values ​​of words in the basic text are counted, and the weighted average of each dense word vector is calculated to generate a compressed market sentiment feature vector. The market sentiment feature vector is input into a preset sentiment weight matrix (a 64-row, 128-column numerical table based on the input and output dimensions). Matrix multiplication is performed, and the result is summed with the bias vector (a 128-dimensional numerical sequence optimized based on the initial all-zero sequence and historical sentiment dataset). The dimension-aligned market sentiment feature vector is then output. The dimension-aligned market sentiment feature vector is then concatenated with the aggregated feature vector according to the dimensions to generate a fused feature vector.

[0033] It should also be noted that both the sentiment embedding lookup table and the geographic embedding lookup table are constructed using the basic idea of ​​representation learning, that is, to represent discrete symbols in a distributed manner through dense vectors, and to use optimization tasks to adjust the values ​​of dense vectors so that the distance in the vector space reflects a certain correlation between the original symbols. However, the sentiment embedding lookup table deals with words in natural language, and the optimization goal is to make the distance between word vectors that are semantically similar or frequently co-occur in the same context close, thereby representing the semantic similarity between words. The geographic embedding lookup table deals with characters in geocoded strings, and the optimization goal is to make the distance between character vectors that are physically adjacent or frequently co-occur in the geocoded sequence close, thereby representing the spatial proximity and regional co-occurrence patterns implied between characters. The sentiment embedding lookup table serves the task of text semantic understanding, while the geographic embedding lookup table serves the task of spatial location relationship.

[0034] S2.3 The fused feature vector generates weight increments through dynamic weight field modulation, and updates the federated learning network weights based on the weight increments; It should be noted that scenario classification labels are defined based on supply chain node business scenarios (such as raw material shortages, logistics delays, and demand surges). For each scenario, real weight change data is extracted from the historical federated learning network weight update records. The arithmetic mean of multiple weight change data for the same business scenario is taken to generate the initial basic incremental vector. The fused feature vector is divided into 128 consecutive 2D numerical structures according to their dimensional indices. The first value of each 2D numerical structure is used as a scaling factor to adjust the magnitude of the initial base increment vector, and the second value is used as a translation factor to shift the overall value of the initial base increment vector relative to its numerical reference. Element-wise independent modulation is then performed on the corresponding initial base increment vector to generate weight increment components, expressed as follows: ; in, Indicates the weight increment component; Indicates the scaling factor; Represents the initial basic increment vector components; Indicates the translation coefficient; All independently modulated weight increment components are combined in dimensional index order to generate weight increments; the original gradient parameters are used as the original weight increments and superimposed on the previous federated learning network weight vector (a random initial federated learning network weight vector, such as 0, is used during the first training) to generate the current federated learning network weights; the product of the weight increment and the preset learning rate (based on gradient descent convergence theory and industrial scenario tuning specifications, such as 0.01) is calculated to scale the magnitude of the weight increment; the scaled weight increment and the current federated learning network weights are summed element-wise along the same dimensional position to generate the updated federated learning network weights.

[0035] S2.4 Input the updated federated learning network weights into the federated learning network and output the global causal feature map.

[0036] It should be noted that the updated federated learning network weights are input into the federated learning network. In the early stage of building the federated learning network, the node parameter segment indexing rules are preset: based on the business characteristics and data dimensions of supplier nodes, manufacturer nodes and logistics provider nodes, the federated learning network weights are logically divided into continuous and non-overlapping parameter segments, and a fixed vector interval index range is assigned to each type of node. When the updated federated learning network weights are input, the updated federated learning network weights are directly cut according to the node parameter segment index rules, and the corresponding dimension numerical subsequences are extracted as the node attribute vectors of each node parameter segment. The values ​​at specific positions in the node attribute vectors directly represent the business attributes. The attribute similarity scores between all node parameter segments are calculated using the cosine similarity score expression. The attribute similarity score is mapped to the causal association strength. Specifically, the attribute similarity score is increased by 1, the original attribute similarity score interval [−1,1] is shifted to the interval [0,2], and the shifted attribute similarity score interval is halved (interval [0,2]→[0,1]). This achieves complete coverage of attribute similarity scores from negative to positive correlation, and outputs the causal association strength value in the range of 0-1. The spatial distance between supply chain nodes is calculated based on their geographical coordinates. This involves obtaining the latitude and longitude coordinates of two nodes, calculating the ratio of pi to 180°, and multiplying this ratio by the latitude and longitude coordinates to obtain radians, thus converting latitude and longitude from degrees to radians. The latitude and longitude differences between the two points are then calculated. An intermediate value is calculated based on these differences, and finally, the large-rounded corner distance is calculated. The final expression for the spatial distance between supply chain nodes is: ; ; ; ; ; in, Indicates the spatial distance between nodes in the supply chain; Indicates the Earth's radius; Indicates the distance between large rounded corners; Indicates intermediate quantity; Indicates latitude difference; , Indicates the latitude and longitude of supply chain node 2; Indicates the difference in longitude; , Indicates the latitude and longitude of supply chain node 1; The inverse of the spatial distance between supply chain nodes and the strength of causal relationship are weighted and fused according to a preset fusion weight ratio (based on the analysis of historical supply chain interruption event data and industry experience consensus definition, such as the inverse of the spatial distance between supply chain nodes having a weight of 30% and the strength of causal relationship having a weight of 70%) to generate edge weights; the supply chain node attribute vectors and edge weights are integrated to construct a global causal feature map. S3. Analyze the global causal feature map based on the causal discovery algorithm, construct multi-level intervention paths, locate root cause nodes through counterfactual reasoning, and generate a root cause tracing report; the root cause tracing report includes primary and secondary cause types, intervention paths, and confidence ratings. S3.1 Extract the causal relationship strength between supply chain nodes in the global causal feature map and calculate the node influence factor; It should be noted that the causal correlation strength of all edges in the global causal feature graph is extracted, and the in-degree influence factor (sum of the causal correlation strength of inbound edges) and out-degree influence factor (sum of the causal correlation strength of outbound edges) of each supply chain node are calculated. The in-degree influence factor and out-degree influence factor are then weighted and summed according to a preset risk transmission weight ratio (based on regression analysis of historical data of supply chain disruption events and industry expert consensus, with an in-degree weight of 40% and an out-degree weight of 60%). The maximum value of the influence factor is selected from the in-degree influence factor and out-degree influence factor, and the ratio of the weighted sum to the maximum value of the influence factor is calculated and normalized to generate the node influence factor.

[0037] S3.2. Generate a three-level intervention path based on node impact factors; perform counterfactual reasoning on each intervention path to generate a counterfactual path; It should be noted that, based on all the node impact factors in the supply chain, supply chain nodes whose node impact factor values ​​exceed the critical node threshold (defined based on the correlation analysis between the node impact factor and the actual interruption probability, such as 0.7) are selected as critical nodes. A three-level intervention path is generated centered on the key node. Specifically, the direct neighbor nodes of the key node are extracted to construct the first-level intervention path; the first-order neighbor nodes and directly connected nodes of the key node are extended to construct the second-level intervention path; and the second-order neighbor nodes and directly connected nodes of the key node are extended to construct the third-level intervention path. All intervention paths must meet the requirement that the path length does not exceed the limit of the number of three-level hops. Perform counterfactual reasoning for each intervention path: remove a node from the intervention path and calculate the causal relationship strength after removal, calculate the difference between the causal relationship strength after removal and the causal relationship strength before removal, and generate a strength change value; integrate the original intervention path, the location of the removed node, the counterfactual intervention path topology (the new intervention path after removing the node), and the strength change value into a counterfactual path.

[0038] S3.3 Calculate the effect difference value between the intervention path and the counterfactual path, mark the supply chain nodes in the intervention path whose effect difference value exceeds the preset root cause determination threshold as root cause nodes, and integrate the root cause node information to generate a root cause tracing report.

[0039] It should be noted that the effect difference value between each intervention path and the counterfactual path is calculated using the following expression: ; in, Indicates the difference in effect values; Indicates the overall causal strength of the counterfactual path; Indicates the overall causal strength of the intervention path; Calculate the percentage mapping of effect differences to generate confidence ratings; the expression is as follows: ; in, Indicates the confidence level rating; The supply chain node with the largest effect difference value is selected. If the effect difference value exceeds the root cause determination threshold (based on the historical effect difference value distribution pattern of the root cause node and the consensus definition of industry experts, such as 30%), the corresponding supply chain node is marked as the root cause node. All root cause nodes are determined by primary and secondary cause classification thresholds to generate primary cause type and secondary cause type: the primary cause type threshold is defined based on the cumulative distribution mutation point analysis of effect difference value (such as 70%), and the root cause determination threshold (such as 30%) is directly used as the secondary cause type threshold; all root cause nodes with effect difference values ​​in the range of 30%-70% are defined as secondary cause type; all root cause nodes with effect difference values ​​exceeding 70% are defined as primary cause type. The report integrates the primary and secondary cause types, locations, effect differences, intervention pathways, and confidence ratings of all root cause nodes to generate a root cause tracing report.

[0040] S4. When the confidence rating exceeds the confidence threshold, generate the alternative supplier switching ratio based on the primary and secondary cause types, generate the logistics route reallocation plan based on the intervention path, and integrate the alternative supplier switching ratio and the logistics route reallocation plan to generate a dynamic coordination instruction. S4.1 When the confidence rating exceeds the preset confidence threshold, calculate the alternative supplier switching ratio in the primary and secondary cause types; It should be noted that when the confidence rating in the root cause tracing report exceeds the preset confidence threshold (based on the historical confidence rating distribution pattern and industry risk management standard definition, such as 70%), the primary and secondary cause types of the current root cause node are extracted. The optimal solution for the backup supplier switching ratio of primary and secondary cause nodes in the historical supply chain disruption events is calculated (e.g., the optimal backup supplier switching ratio of primary cause nodes is 80%, and the optimal backup supplier switching ratio of secondary cause nodes is 30%). The optimal solution for the backup supplier switching ratio of primary and secondary cause nodes is used as the primary cause baseline ratio and the secondary cause baseline ratio and is fixed into the parameter library. Based on the extracted primary and secondary cause types, the corresponding baseline ratio in the parameter library is called. If the root cause node is a primary cause type, the primary cause baseline ratio is output; if it is a secondary cause type, the secondary cause baseline ratio is output. The standby supplier switching ratio is calculated based on the confidence rating and the benchmark ratio, expressed as follows: ; in, Indicates the switching ratio of backup suppliers; This indicates the baseline ratio, with the primary factor baseline ratio being 0.8 and the secondary factor baseline ratio being 0.3. This indicates the upper limit threshold for scaling, preset based on the cost-benefit balance point analysis of supply chain backup switching.

[0041] S4.2. Based on the intervention path, identify high-risk paths and match low-risk alternative paths to generate a logistics path redistribution scheme; It should be noted that, based on historical logistics records of the supply chain and industry standard route plans, a low-risk alternative route library is pre-set. Available logistics routes between all supply chain nodes are collected, and the edge weights and historical interruption frequency data of each logistics route are extracted. Logistics routes with edge weights greater than the edge weight threshold (based on the definition of supply chain security standards, such as 0.8; edge weight and risk level are inversely proportional, the larger the edge weight value, the lower the risk level) and historical interruption frequency data lower than the frequency baseline (based on the definition of stability probability distribution inflection point analysis, such as 5%) are selected as candidate low-risk routes. The candidate low-risk routes are classified and stored according to the origin-end node pairs. Each candidate low-risk route is accompanied by an edge weight list, historical interruption frequency data, and logistics cost coefficient to form a low-risk alternative route library. Based on the intervention paths in the root cause tracing report, path segments with edge weights below the edge weight threshold are extracted and marked as high-risk paths. The low-risk alternative path library is queried based on the start and end nodes of the high-risk paths, and all low-risk alternative paths that meet the positional constraints of the start and end nodes are matched. The product of the sum of the reciprocals of all edge weights in the low-risk alternative paths and the logistics cost coefficient is used as the comprehensive risk value. The low-risk alternative path with the lowest comprehensive risk value is selected, and the cargo volume is allocated according to the path risk weight ratio (based on the definition of logistics network flow allocation theory, high-risk paths are allocated 30% weight, and low-risk alternative paths are allocated 70% weight) to generate a logistics path redistribution scheme.

[0042] S4.3 Integrate the backup supplier switching ratio and logistics route reallocation scheme, and generate dynamic coordination instructions through digital certificates and timing services.

[0043] It should be noted that the backup supplier switching ratio and logistics route reallocation scheme are directly integrated into a key-value pair data set; The digital certificate signing service is invoked. A hash value is calculated using a hash function with a dedicated key and a set of key-value pairs, and a cryptographic digital signature is generated. The expression is: ; ; in, Represents a hash value; This represents the input data, i.e., a collection of key-value pairs. This represents a cryptographic hash function that outputs a fixed-length 256-bit hash value. Indicates encrypted digital signature; This refers to the private key within the proprietary key; This refers to the public key within the private key; Obtain the accurate timestamp provided by the time synchronization center; merge the key-value pair data set, encrypted digital signature, and accurate timestamp into a dynamic coordination command.

[0044] S5. Distribute dynamic coordination instructions to supply chain nodes and execute procurement allocation adjustments, transportation route updates, and resource supply guarantees, generating management coordination reports.

[0045] S5.1. Dynamic coordination instructions are distributed to supply chain nodes through a blockchain peer-to-peer protocol. Supply chain nodes verify the integrity and authenticity of the dynamic coordination instructions through digital signatures. It should be noted that the dynamic coordination command is broadcast to all relevant supply chain nodes via a blockchain peer-to-peer transmission protocol. After receiving the command, each supply chain node verifies whether the timestamp has expired. If the deviation between the precise timestamp in the dynamic coordination command and the local clock is less than the time deviation baseline (defined based on industrial safety requirements and timing service requirements, such as 5 seconds), then the timestamp is determined not to have expired. The digital signature is decrypted using the public key to obtain hash value A. The received key-value pair data set is then recalculated to obtain hash value B (using the same hash function, such as SHA-256). Hash value A and hash value B are compared to see if they are completely consistent. If they are completely consistent, the dynamic coordination command is confirmed to be complete and authentic. The supply chain node executes the content of the dynamic coordination command and reports the execution result. If they are inconsistent or the timestamp has expired, the dynamic coordination command is discarded, and an alarm log is triggered and uploaded.

[0046] S5.2 The supplier node reallocates purchase orders based on the backup supplier switching ratio; It should be noted that the supplier node extracts the standby supplier switching ratio from the key-value pair database and queries the business database to obtain the current total purchase order volume, in-transit inventory, and safety stock; based on the standby supplier switching ratio, it calculates the actual switchable quantity, expressed as follows: ; in, Indicates the actual number of switches available; This indicates the total number of current purchase orders; Indicates the switching ratio of backup suppliers; Indicates the amount of inventory in transit; Indicates safety stock; Apply the actual switchable quantity to purchase order redistribution: use the difference between the original supplier order quantity and the actual switchable quantity as the new order quantity, and generate an equal amount of new orders for the backup supplier. Encapsulate the new order ID (a unique identifier automatically generated when creating a new order), execution timestamp, and node private key signature as the execution result of the new order.

[0047] S5.3. Based on the logistics route redistribution scheme, the logistics provider node terminates the transportation tasks of high-risk routes, activates low-risk alternative routes, and schedules cargo volume according to the allocation weight. It should be noted that the local logistics management database is queried in real time to obtain the sum of the current on-the-go cargo volume and the new cargo volume to be allocated on high-risk routes, which is used as the total cargo volume to be dispatched. The logistics provider node extracts the logistics route reallocation scheme from the key-value pair data set, sends a termination instruction to the transportation structure of the high-risk route, and loads the detailed parameters of the low-risk alternative route (such as route coordinates, starting point, and estimated time). The product of the allocation weight of the logistics route reallocation scheme and the total amount of cargo to be scheduled is used as the actual cargo scheduling value. The low-risk alternative route is then activated to execute the transportation task according to the actual cargo scheduling value, and the route scheduling result is generated.

[0048] S5.4 The manufacturer node calculates the raw material inventory threshold and rearranges the production line priority based on the updated procurement and logistics plan; it summarizes the execution results of each supply chain node and generates a management coordination report.

[0049] It should be noted that the manufacturer node calculates the raw material inventory threshold based on the new order execution results reported by the supplier node and the route scheduling results provided by the logistics provider node, combined with the historical raw material consumption rate (obtained from the production logs in the manufacturer node's local database) and the actual delivery time. The expression is as follows: ; ; in, Indicates the raw material inventory threshold; Indicates the actual delivery time; Indicates the historical consumption rate of raw materials; This represents the volatility coefficient, defined based on the variance of the historical consumption rate of raw materials. This represents buffer inventory, fixed at 20% of the historical consumption rate of raw materials (based on the theoretical benchmark definition of supply chain safety stock). Based on the urgency of raw material arrival (e.g., actual arrival time < 2 hours is marked as urgent, based on production cycle critical value and supply chain disruption risk definition) and order priority (e.g., customer level weight), a weighted scoring rule (60% weight for urgency and 40% weight for order priority) is used to sort all production line tasks. Urgent and high-priority orders are prioritized in the production sequence. The new order execution results of suppliers, the route scheduling results of logistics providers, raw material inventory thresholds and production line sequences are summarized to generate a management coordination report.

[0050] This embodiment also provides a computer device applicable to the industrial supply chain management coordination method based on cloud computing and big data analysis, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the industrial supply chain management coordination method based on cloud computing and big data analysis proposed in the above embodiment.

[0051] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0052] This embodiment also provides a storage medium storing a computer program. When executed by a processor, the program implements the industrial supply chain management coordination method based on cloud computing and big data analysis proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0053] In summary, this invention achieves encrypted collaboration of local data at supply chain nodes by constructing a federated learning network, which, while strictly protecting the privacy of trade secrets, improves the response efficiency of sudden interruption events from traditional delays to near real-time levels; and by combining causal graphs and counterfactual reasoning to dynamically generate root cause tracing reports, it accurately distinguishes primary and secondary responsible nodes and automatically triggers cross-node coordination instructions, thereby realizing closed-loop autonomy of the entire supply chain of procurement, logistics, and production resources.

[0054] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A cloud computing and big data analysis based industrial supply chain management coordination method, characterized in that: The application relates to a supply chain management method based on a supply chain management system. The application comprises the following steps: A federal learning network is constructed, local private data of a supply chain node is loaded into the federal learning network, and encrypted gradient parameters are generated; when order fluctuation exceeds a preset fluctuation threshold and a logistics delay event occurs, the encrypted gradient parameters are converted into spatiotemporal correlation features; The spatiotemporal correlation features are aggregated based on an attention weight algorithm to generate an aggregated feature vector; the federal learning network weight is updated based on market public opinion features and the aggregated feature vector to generate a global causal feature graph; The global causal feature graph is analyzed according to a causal discovery algorithm, three-level intervention paths are constructed, a root cause node is located through counterfactual reasoning, and a root cause tracing report is generated; the root cause tracing report comprises primary and secondary factor types, intervention paths and confidence ratings; When the confidence rating exceeds a confidence threshold, a backup supplier switching ratio is generated according to the primary and secondary factor types, a logistics path redistribution scheme is generated based on the intervention paths, the backup supplier switching ratio and the logistics path redistribution scheme are integrated to generate a dynamic coordination instruction; 2. The cloud computing and big data analytics based industrial supply chain management coordination method as claimed in claim 1, wherein: The dynamic coordination instruction is distributed to the supply chain node and used for performing purchase distribution adjustment, transportation path update and resource supply guarantee to generate a management coordination report.

3. The cloud computing and big data analytics based industrial supply chain management coordination method as claimed in claim 2, wherein: The supply chain node comprises a supplier node, a manufacturer node and a logistics merchant node. The encrypted gradient parameters are generated in the following specific steps: A federal learning network framework is constructed based on a hardware layer and a protocol layer, and the federal learning network framework is converted into the federal learning network through node cooperation and gradient aggregation; 4. The cloud computing and big data analytics based industrial supply chain management coordination method as claimed in claim 3, wherein: The supply chain node loads local private data through a local data interface and inputs the local private data into the federal learning network, and homomorphic encryption algorithm is adopted to generate encrypted gradient parameters. The encrypted gradient parameters are converted into spatiotemporal correlation features in the following specific steps: An order fluctuation rate is obtained through a supply chain management structure, and a logistics delay event is determined through GPS track analysis; when the order fluctuation rate exceeds a preset fluctuation threshold and the logistics delay event occurs simultaneously, an activation signal is generated; 5. The cloud computing and big data analytics based industrial supply chain management coordination method as claimed in claim 4, wherein: A feature converter receives the encrypted gradient parameters and the activation signal and converts the encrypted gradient parameters and the activation signal to generate spatiotemporal correlation features. The aggregated feature vector is generated in the following specific steps: Similarity scores among the spatiotemporal correlation features are calculated; 6. The cloud computing and big data analytics based industrial supply chain management coordination method as claimed in claim 5, wherein: Attention weights are allocated according to the similarity scores to generate the aggregated feature vector. The global causal feature graph is generated in the following specific steps: Market public opinion texts are obtained from public data sources, and market public opinion features are extracted through natural language processing; the market public opinion features and the aggregated feature vector are spliced into a fusion feature vector; The fusion feature vector generates a weight increment through dynamic weight field modulation, and the federal learning network weight is updated based on the weight increment; 7. The cloud computing and big data analytics based industrial supply chain management coordination method as claimed in claim 6, wherein: The updated federal learning network weight is input into the federal learning network, and the global causal feature graph is output. The multiple three-level intervention paths are constructed in the following specific steps: Causal correlation strengths among the supply chain nodes in the global causal feature graph are extracted, and node influence factors are calculated; 8. The cloud computing and big data analytics based industrial supply chain management coordination method as claimed in claim 7, wherein: The three-level intervention paths are generated based on the node influence factors through screening of key nodes and multilevel expansion. The root cause tracing report is generated in the following specific steps: Counterfactual reasoning is performed on the three-level intervention paths to generate counterfactual paths; The effect difference of the three-level intervention path and the counterfactual path is calculated, the supply chain node with the largest effect difference is marked as a root cause node, and a root cause tracing report is generated by integrating the root cause node information.

9. The cloud computing and big data analytics based industrial supply chain management coordination method as claimed in claim 8, wherein: The generated dynamic coordination instruction has the following specific steps, When the confidence rating exceeds the pre-set confidence threshold, the standby supplier switching ratio in the primary and secondary cause type is calculated; Based on the intervention path, high-risk paths are identified and matched with low-risk alternative paths to generate a logistics path redistribution scheme; The standby supplier switching ratio and the logistics path redistribution scheme are integrated, and the dynamic coordination instruction is generated through digital certificates and time service.

10. The cloud computing and big data analytics based industrial supply chain management coordination method as claimed in claim 9, wherein: The generated management coordination report has the following specific steps, The dynamic coordination instruction is distributed to the supply chain nodes through the block chain point-to-point protocol, and the supply chain nodes verify the integrity and authenticity of the dynamic coordination instruction through digital signature; The supplier node reallocates the purchase order according to the standby supplier switching ratio; Based on the logistics path redistribution scheme, the logistics merchant node terminates the high-risk path transportation task, enables the low-risk alternative path, and schedules the cargo volume according to the allocated weight; The manufacturer node adjusts the raw material inventory threshold and rearranges the production line priority according to the updated procurement and logistics plan; The execution results of the supplier node, the manufacturer node and the logistics merchant node are summarized to generate a management coordination report.

Images