Supply chain risk prediction method and system

Abstract

The invention belongs to the technical field of supply chain data analysis, and discloses a supply chain risk prediction method and system. The method comprises the following steps of collecting historical supply chain big data based on a cloud data center, and performing preprocessing; performing clustering processing and data dimension reduction; constructing a corresponding supply chain risk prediction model and a supply chain optimization model by using an artificial intelligence algorithm; collecting real-time supply chain data of a current supply chain node, obtaining a matched supply chain link, and performing data dimension reduction; performing supply chain risk prediction by using the matched supply chain risk prediction model to obtain a real-time supply chain risk prediction result; and performing matched supply chain optimization by using the matched supply chain optimization model to obtain a real-time supply chain optimization strategy, returning the real-time supply chain optimization strategy to the corresponding supply chain node, and executing the real-time supply chain optimization strategy. According to the method, the problems of difficulty in data collection and integration, poor risk prediction effect and inflexible response mechanism in the prior art are solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of supply chain data analysis, and particularly relates to a supply chain risk prediction method and system. Background Art

[0002] In today's globalized business environment, supply chain management has become one of the key factors in an enterprise's competitiveness. The stability of the supply chain directly affects the enterprise's production efficiency, cost control, and customer satisfaction. However, the supply chain faces various internal and external risks, which may come from multiple aspects such as market changes, political unrest, natural disasters, supplier problems, etc. The existing supply chain risk prediction technologies have the following limitations: 1) Difficult data collection and integration. Supply chain data is usually scattered in different systems and departments, and the data formats and standards are not unified, resulting in difficulties in data collection and integration.

[0003] 2) Poor risk prediction effect. Existing technologies mostly rely on qualitative risk assessment methods such as expert scoring and Delphi method. These methods are highly subjective, lack quantitative analysis, and are difficult to comprehensively cover the diversified risks of the supply chain. 3) Inflexible response mechanism. When risks occur, traditional supply chain risk management lacks a rapid response mechanism, resulting in lagging risk response measures. The supply chain lacks flexibility and adaptability and is difficult to adjust according to real-time situations. Summary of the Invention

[0004] In order to solve the problems of difficult data collection and integration, poor risk prediction effect, and inflexible response mechanism existing in the prior art, the purpose of the present invention is to provide a supply chain risk prediction method and system.

[0005] The technical solution adopted by the present invention is as follows: A supply chain risk prediction method, comprising the following steps: Based on a cloud data center, collect heterogeneous historical supply chain big data of historical supply chain data from different data sources and including different supply chain links, and preprocess the historical supply chain big data to obtain a number of homogeneous preprocessed historical supply chain data; Perform clustering processing on a number of preprocessed historical supply chain data to obtain a number of cluster centers and corresponding data clusters, and perform data dimensionality reduction on the number of data clusters to obtain a dimensionality-reduced historical supply chain data set and a corresponding key index set for each supply chain link; According to the dimensionality-reduced historical supply chain data set, use an artificial intelligence algorithm to construct a corresponding supply chain risk prediction model and a supply chain optimization model, and after traversing the dimensionality-reduced historical supply chain data sets of all supply chain links, obtain a supply chain risk prediction model and a supply chain optimization model for each supply chain link; Collect the real-time supply chain data of the current supply chain node, obtain the corresponding matching supply chain links according to the Euclidean distance between the real-time supply chain data and several clustering centers, and perform data dimensionality reduction on the real-time supply chain data according to the matching key index set corresponding to the matching supply chain links to obtain the real-time supply chain data after dimensionality reduction; Extract the matching supply chain risk prediction model and the matching supply chain optimization model corresponding to the matching supply chain links, and perform supply chain risk prediction using the matching supply chain risk prediction model according to the real-time supply chain data after dimensionality reduction to obtain the real-time supply chain risk prediction result; According to the real-time supply chain risk prediction result, use the matching supply chain optimization model to perform matching supply chain optimization to obtain the real-time supply chain optimization strategy, return the real-time supply chain optimization strategy to the corresponding supply chain node, execute the real-time supply chain optimization strategy, and return to the real-time supply chain data collection step.

[0006] Furthermore, based on the cloud data center, collect heterogeneous historical supply chain big data from different data sources and including different supply chain links, and preprocess the historical supply chain big data to obtain several homogeneous preprocessed historical supply chain data, including the following steps: Based on the cloud data center, collect heterogeneous historical supply chain big data from different data sources and including different supply chain links; Perform data screening on the historical supply chain big data to obtain several cleaned historical supply chain data; Construct a unified data model according to the data format of the cloud data center, and set the corresponding historical data mapping and conversion strategy for each data source according to the unified data model; Perform data format mapping and conversion on several heterogeneous cleaned historical supply chain data according to the historical data mapping and conversion strategy to obtain several homogeneous converted historical supply chain data.

[0007] Furthermore, perform clustering processing on several preprocessed historical supply chain data to obtain several clustering centers and corresponding data clusters, and perform data dimensionality reduction on several data clusters to obtain the historical supply chain data set after dimensionality reduction and the corresponding key index set for each supply chain link, including the following steps: Using the FCM clustering algorithm, perform clustering processing on several preprocessed historical supply chain data with the supply chain link as the clustering target and the data index difference as the clustering criterion to obtain several clustering centers; Divide the preprocessed historical supply chain data into data clusters corresponding to each clustering center according to the Euclidean distance between the preprocessed historical supply chain data and each clustering center; Convert each data cluster into a data matrix, and use the PCA method to reduce the dimension of the data matrix, obtaining the historical supply chain data sets after dimension reduction and the corresponding key index sets for each supply chain link.

[0008] Furthermore, based on the historical supply chain data sets after dimension reduction, use artificial intelligence algorithms to construct corresponding supply chain risk prediction models and supply chain optimization models. After traversing the historical supply chain data sets after dimension reduction for all supply chain links, obtain the supply chain risk prediction models and supply chain optimization models for each supply chain link, including the following steps: Based on the historical supply chain data sets after dimension reduction of the same supply chain link, use deep learning algorithms to construct corresponding supply chain risk prediction models and generate several historical supply chain risk prediction results; Based on several historical supply chain risk prediction results of the same supply chain link, use reinforcement learning algorithms to construct corresponding supply chain optimization models; Traverse the historical supply chain data sets after dimension reduction for all supply chain links to construct the supply chain risk prediction models and supply chain optimization models for each supply chain link; Use the supply chain link as the retrieval label for the supply chain risk prediction models and supply chain optimization models, and store several supply chain risk prediction models and several supply chain optimization models with the retrieval label set.

[0009] Furthermore, the supply chain risk prediction model is constructed based on the IFWA-DBN algorithm.

[0010] Furthermore, the supply chain optimization model is constructed based on the DQN algorithm.

[0011] Furthermore, collect the real-time supply chain data of the current supply chain node. According to the Euclidean distance between the real-time supply chain data and several cluster centers, obtain the corresponding matching supply chain link, and based on the matching key index set corresponding to the matching supply chain link, reduce the dimension of the real-time supply chain data to obtain the real-time supply chain data after dimension reduction, including the following steps: Collect the real-time supply chain data stored in the data server of the current supply chain node, and set the corresponding real-time data mapping and conversion strategy according to the data format of the current supply chain node and the unified data model; According to the real-time data mapping and conversion strategy, perform data mapping and conversion on the real-time supply chain data to obtain the real-time supply chain data after conversion; Obtain the Euclidean distance between the real-time supply chain data after conversion and several cluster centers, and use the supply chain link of the cluster center with the closest Euclidean distance as the matching supply chain link; Extract the matching key index set corresponding to the matching supply chain link, and based on the matching key index set, reduce the dimension of the real-time supply chain data after conversion to obtain the real-time supply chain data after dimension reduction.

[0012] Further, extract the matching supply chain risk prediction model and the matching supply chain optimization model corresponding to the matching supply chain link, and based on the dimension-reduced real-time supply chain data, use the matching supply chain risk prediction model to conduct supply chain risk prediction to obtain the real-time supply chain risk prediction result, including the following steps: Search in several retrieval tags according to the matching supply chain link to obtain the corresponding target retrieval tag; Take the supply chain risk prediction model and the supply chain optimization model corresponding to the target retrieval tag as the matching supply chain risk prediction model and the matching supply chain optimization model; Input the dimension-reduced real-time supply chain data into the matching supply chain risk prediction model, and use the matching supply chain risk prediction model to extract the real-time data features of the dimension-reduced real-time supply chain data; Conduct supply chain risk prediction based on the real-time data features to obtain the real-time supply chain risk prediction result.

[0013] Further, according to the real-time supply chain risk prediction result, use the matching supply chain optimization model to conduct matching supply chain optimization to obtain the real-time supply chain optimization strategy, return the real-time supply chain optimization strategy to the corresponding supply chain node, execute the real-time supply chain optimization strategy, and return to the real-time supply chain data collection step, including the following steps: Update the model parameters of the matching supply chain optimization model according to the real-time supply chain risk prediction result to obtain the updated model parameters; Based on the updated model parameters, use the matching supply chain optimization model to conduct matching supply chain optimization to obtain the real-time supply chain optimization strategy; Return the real-time supply chain optimization strategy to the corresponding supply chain node, and use the decision support system of the supply chain node to execute the real-time supply chain optimization strategy and return to the real-time supply chain data collection step.

[0014] A supply chain risk prediction system for implementing the supply chain risk prediction method. The system includes a cloud data center and several supply chain nodes. Several supply chain nodes are all communicatively connected to the cloud data center, and the supply chain nodes include a data server and a decision support system; The cloud data center includes a big data collection unit, a data processing unit, a model construction unit, a supply chain link classification unit, a supply chain risk prediction unit, and a supply chain optimization unit connected in sequence. The big data collection unit and the supply chain link classification unit are respectively communicatively connected to the data servers of several supply chain nodes, and the supply chain optimization unit is communicatively connected to the decision support systems of several supply chain nodes.

[0015] The beneficial effects of the present invention are: The present invention discloses a supply chain risk prediction method and system. By adopting a cloud data center, it can efficiently collect and process heterogeneous historical supply chain big data from different data sources, achieving data unification and standardization, and reducing the difficulty of data collection and integration. Through clustering processing and dimensionality reduction techniques, the present invention can extract key indicators from complex supply chain data, simplify the complexity of model construction, and improve the accuracy and effectiveness of prediction. By using artificial intelligence algorithms to construct a supply chain risk prediction model and an optimization model, the prediction accuracy and the effectiveness of the optimization strategy are significantly improved. By feeding back the supply chain optimization strategy to the supply chain nodes and executing it, a closed-loop management mechanism is formed, ensuring the continuity and effectiveness of supply chain risk prediction and management. By constructing corresponding supply chain risk prediction models and supply chain optimization models according to different supply chain links, the pertinence and practicability of risk prediction and optimization strategies are ensured. And when risks occur, a rapid response mechanism can avoid the lag of risk response measures, improve the flexibility and adaptability of the supply chain, and optimize the supply chain according to real-time situations. Through the cloud data center and artificial intelligence algorithms, a high degree of integration of supply chain risk management tools and systems is achieved, breaking information silos and improving overall management efficiency.

[0016] Other beneficial effects of the present invention will be further described in the specific implementation manners. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 is a flowchart of the supply chain risk prediction method in the present invention.

[0018] Figure 2 is a structural block diagram of the supply chain risk prediction system in the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0019] The present invention will be further explained below in conjunction with the drawings and specific embodiments.

[0020] Embodiment 1: As Figure 1 shown, this embodiment provides a supply chain risk prediction method, including the following steps: S1: Based on the cloud data center, collect heterogeneous historical supply chain big data of historical supply chain data from different data sources and including different supply chain links, and preprocess the historical supply chain big data to obtain a number of homogeneous preprocessed historical supply chain data, including the following steps: S1-1: Based on the cloud data center, collect heterogeneous historical supply chain big data of historical supply chain data from different data sources and including different supply chain links; The supply chain links include planning, procurement, production, delivery, transportation, etc. Each link involves different data indicators. The planning link is the preliminary planning process of supply chain activities, involving data indicators in aspects such as materials, procurement, production, warehousing, distribution, sales, fulfillment, inventory, etc. Good planning can guide and monitor the execution process of the supply chain to ensure the achievement of goals. The procurement link is based on the procurement plan and includes data indicators such as inquiry, supplier management, bargaining, contract signing, purchase order, purchase delivery, purchase receipt, inspection, and purchase settlement. The production link includes product production and order production. Product production is usually completed in factories or processing centers, manufacturing raw materials into finished products, including various data indicators of product production. Order production involves the packing and sending of customer orders by warehouses, distribution centers, or stores, including various data indicators of order production. The delivery link is the process of delivering products from upstream suppliers to downstream customers, including data indicators such as process management and timeliness control of logistics distribution. The distribution link includes data indicators involved in the process where goods go from suppliers through manufacturers, wholesalers, logistics, retailers, etc. and finally point to consumers. Therefore, the supply chain data structures of different links vary greatly and cannot be analyzed uniformly; S1-2: Screen the historical supply chain big data to obtain a number of cleaned historical supply chain data; Data screening includes duplicate data cleaning, error data removal, and data denoising to improve data quality and provide support for subsequent data analysis and processing, so as to improve the speed and effect of model training; S1-3: According to the data format of the cloud data center, construct a unified data model, and according to the unified data model, set corresponding historical data mapping and conversion strategies for each data source, including the following steps: S1-3-1: Analyze the historical supply chain data of different data sources to obtain a number of core data elements for each data source; S1-3-2: Set the data structure of the unified data model according to the number of core data elements of different data sources, such as tables, fields, relationships, etc.; S1-3-3: Set the data relationships between core data elements according to the data structure of the unified data model, including one-to-one, one-to-many, many-to-many, etc. relationships; S1-3-4: Set the data constraints of each core data element according to the core data elements of different data sources, the data structure of the unified data model, and the data relationships between core data elements, such as data type, value range, default value, etc.; S1-3-5: Construct the corresponding unified data model according to the core data elements of different data sources, the data structure of the unified data model, the data relationships between core data elements, and the data constraints of each core data element; S1-3-6: Set corresponding historical data mapping and conversion strategies for each data source according to the unified data model; The hardware systems adopted by supply chain nodes are different, and the stored data formats are different. Therefore, the big data collected in the supply chain includes several heterogeneous data, which pose obstacles to subsequent data analysis and processing. In this embodiment, the data mapping and conversion strategy includes a mapping strategy and a data conversion strategy. The mapping strategy is used to determine how to map heterogeneous data into the unified data model, and the data conversion strategy is used to implement the conversion of heterogeneous data into the data in the unified data model according to the mapping strategy based on a preset conversion algorithm or script. The data mapping and conversion strategy is used to handle data heterogeneity and solve the inconsistencies in data formats, data types, data semantics, etc. in different data sources; S1-4: Perform data format mapping and conversion on several heterogeneous cleaned historical supply chain data according to the historical data mapping and conversion strategy to obtain several homogeneous converted historical supply chain data; S2: Perform clustering processing on several preprocessed historical supply chain data to obtain several cluster centers and corresponding data clusters, and perform data dimensionality reduction on several data clusters to obtain the reduced-dimensional historical supply chain data sets and corresponding key index sets for each supply chain link, including the following steps: S2-1: Using the Fuzzy C-mean (FCM) clustering algorithm, perform clustering processing on several preprocessed historical supply chain data with the supply chain link as the clustering target and the data index difference as the clustering criterion to obtain several cluster centers, including the following steps: S2-1-1: Select clustering parameters, including the fuzzy factor and the total number of cluster centers, and perform clustering on several preprocessed historical supply chain data according to the clustering parameters using the Fuzzy C-mean (FCM) clustering algorithm to obtain several initial cluster centers; S2-1-2: Set corresponding fuzzy membership degrees for each initial cluster center according to the Euclidean distance between each preprocessed historical supply chain data and several initial cluster centers; The formula is:

[0021] In the formula, is the Euclidean distance between the i preprocessed historical supply chain data and the j cluster center; is the preprocessed historical supply chain data; is the j cluster center; is the data indication quantity; is the clustering center indicator quantity; S2-1-3: Update the clustering centers according to the fuzzy membership degrees to obtain a corresponding number of updated clustering centers; The formula for updating the fuzzy membership degrees is:

[0022] In the formula, is the historical supply chain data after preprocessing for the th; is the data indicator quantity; are all clustering center indicator quantities; is the total number of clustering centers; is the i distance from the historical supply chain data after preprocessing for the j th to the k th clustering center; is the i updated fuzzy membership degree of the historical supply chain data after preprocessing belonging to the th clustering center; The formula for updating the clustering centers is:

[0023] In the formula, is the j updated clustering center; is the fuzzy factor; is the data indicator quantity; is the total number of data; is the clustering center indicator quantity; is the historical supply chain data after preprocessing for the th; i is the fuzzy membership degree of the historical supply chain data after preprocessing belonging to the S2-1-4: Use the Lagrange multiplier method to calculate the merging function to obtain the merging function value and the merging function change value. The formula is:

[0024] In the formula, is the t th, t (t - 1) - th iteration's merging function value of the Lagrange multiplier method; is the corresponding change value; is the i characteristic parameter; t is the iteration number indicator quantity; is the i fuzzy membership degree of the historical supply chain data after preprocessing belonging to the Membership degree of the clustering center; is the fuzzy factor; is the data indicator; is the total number of data; is the clustering center indicator; is the total number of clustering centers; is the i Euclidean distance from the preprocessed historical supply chain data to the j clustering center; S2-1-5: If the merging function value is greater than the function threshold, or the change value of the merging function is greater than the change value threshold, then continue to update the clustering center; otherwise, take the current clustering center as the final clustering center; S2-2: According to the Euclidean distances between several preprocessed historical supply chain data and each clustering center, divide the preprocessed historical supply chain data into the data clusters corresponding to each clustering center; S2-3: Convert each data cluster into a data matrix, and use the Principal Component Analysis (PCA) method to reduce the dimension of the data matrix to obtain the reduced-dimensional historical supply chain data set and the corresponding key index set for each supply chain link, including the following steps: S2-3-1: Perform matrix conversion on the data cluster to obtain the corresponding data matrix , where is the p th row vector of the preprocessed historical supply chain data, p is the row vector indicator, n is the total number of preprocessed historical supply chain data; the initial row vector of the data matrix is the preprocessed historical supply chain data, and the initial column vector of the data matrix is the historical supply chain index data; S2-3-2: Perform standardization processing on the data matrix to obtain the corresponding standardized data matrix; The formula is:

[0025] In the formula, is the standardized data matrix; is the mean of the data matrix; is the variance of the data matrix; S2-3-3: Obtain the covariance matrix of the standardized data matrix, and based on the standardized data matrix and the covariance matrix, obtain the corresponding alternative principal component matrix; the alternative row vectors of the alternative principal component matrix are the preprocessed historical supply chain data, and the alternative column vectors of the alternative principal component matrix are the historical supply chain alternative index data; The formula is:

[0026] In the formula, is the covariance matrix of the data matrix after standardization; is the alternative principal component matrix; is the transformation matrix; is the unit eigenvector matrix; n is the total number of historical supply chain data after preprocessing;

[0027] In the formula, is the alternative principal component matrix; is the transformation matrix; is the data matrix after standardization; S2-3-4: Take several alternative column vectors with the cumulative contribution rate of variance of the first 90% in the alternative principal component matrix as the corresponding several principal component column vectors, and obtain the dimensionality-reduced data matrix composed of several principal component column vectors; The key row vectors of the dimensionality-reduced data matrix are the dimensionality-reduced historical supply chain data, and the key column vectors of the dimensionality-reduced data matrix are the key index data of the historical supply chain; ; The formula is: The formula is:

[0028] In the formula, is the cumulative contribution rate of variance; The th alternative principal component variance; is the alternative principal component indicator; is the total number of alternative principal components; is the total number of principal components; S2-3-5: Convert the dimensionality-reduced data matrix into the corresponding several dimensionality-reduced historical supply chain data sets, and integrate the supply chain key indicators corresponding to the key column vectors to obtain the corresponding key indicator set; S3: According to the dimensionality-reduced historical supply chain data sets, use artificial intelligence algorithms to construct the corresponding supply chain risk prediction model and supply chain optimization model, and after traversing the dimensionality-reduced historical supply chain data sets of all supply chain links, obtain the supply chain risk prediction model and supply chain optimization model for each supply chain link, including the following steps: S3-1: Based on the dimension-reduced historical supply chain dataset of the same supply chain link, use the Improved Fireworks Algorithm (IFWA)-Deep Belief Network (DBN) algorithm to construct a corresponding supply chain risk prediction model and generate several historical supply chain risk prediction results, including the following steps: S3-1-1: Add labels to the clustering centers of the dimension-reduced historical supply chain dataset of the same supply chain link, and spread the added labels to the corresponding data clusters to obtain a supply chain risk prediction sample set with true supply chain risk labels; S3-1-2: Divide the supply chain risk prediction sample set of the same supply chain link into a model training sample set and a model test sample set according to the ratio of 7:3; S3-1-3: Use several unlabeled dimension-reduced historical supply chain data to pre-train the DBN network and construct an initial supply chain risk prediction model; S3-1-4: Take the initial network parameters of the DBN network as the optimization target of the IFWA optimization algorithm, and use the IFWA optimization algorithm for optimization according to the optimization target, including the following steps: S3-1-4-1: Take the initial network parameters of the DBN network as the optimization target of the IFWA optimization algorithm; S3-1-4-3: Set the IFWA population parameters, maximum number of iterations, and fitness function of the IFWA optimization algorithm, and take the optimization target as the position of the IFWA individuals in the IFWA population; S3-1-4-3: Initialize the IFWA population using the Circle chaotic mapping sequence according to the IFWA population parameters to obtain an initialized IFWA population. The formula is:

[0029] In the formula, is the initial IFWA individual of the Circle chaotic mapping; is the randomly generated initial IFWA individual; is the indicator; S3-1-4-4: Calculate the fitness values of the IFWA individuals in the initialized IFWA population according to the fitness function. The formula is:

[0030] In the formula, is the fitness value of the initial IFWA individual ; is the prediction mean square error function; are the predicted value and the true value; S3-1-4-5: Obtain the explosion radius and the number of sparks of each initial firework individual in the initial firework set. The formula is:

[0031] In the formula, is the initial IFWA individual 's number of sparks; is a constant; is the maximum fitness value in the initialized IFWA population; is the initial IFWA individual 's fitness value; is an infinitesimal constant;

[0032] In the formula, is the initial IFWA individual 's explosion radius; is the explosion radius adjustment constant; is the minimum fitness value in the initialized IFWA population; S3-1-4-6: According to the explosion radius and the number of sparks of each initial IFWA individual in the initialized IFWA population, perform firework explosion to obtain the updated IFWA population. The formula is:

[0033] In the formula, is the updated IFWA individual; is a random number from -1 to 1; is the initial IFWA individual; S3-1-4-7: Use the Gaussian mutation algorithm to perform Gaussian mutation on the initialized IFWA population to generate the Gaussian mutated IFWA population. The formula is:

[0034] In the formula, is the Gaussian mutated IFWA individual; is a random number from a Gaussian distribution with both mean and variance equal to 1; S3-1-4-8: Use the dynamic reverse learning algorithm to perform dynamic reverse learning on the initialized IFWA population to generate the dynamically reversed IFWA population. The formula is:

[0035] In the formula, is the dynamically reversed IFWA individual; is the forward candidate optimal solution; γ is the decreasing inertia coefficient; are the maximum and minimum values of the vector space respectively; S3-1-4-9: Calculate the fitness values of all IFWA individuals in the updated IFWA population, the IFWA population with Gaussian mutation, and the IFWA population with dynamic reverse, and take the IFWA individual with the minimum fitness value as the optimal individual; S3-1-4-10: If the number of iterations reaches the threshold or the fitness value of the optimal individual meets the requirements, output the optimal solution corresponding to the current optimal individual to obtain the optimal initial network parameters of the DBN network; S3-1-5: According to the optimal initial network parameters of the DBN network, optimize the network structure of the initial supply chain risk prediction model, input the model training sample set, and perform optimization training to obtain the optimized supply chain risk prediction model; S3-1-6: Input the model test sample set, perform model testing on the optimized supply chain risk prediction model, and obtain the model test accuracy rate; S3-1-7: If the model test accuracy rate is greater than the model test accuracy rate threshold, output the optimal supply chain risk prediction model; otherwise, continue with the optimization training; S3-2: According to several historical supply chain risk prediction results of the same supply chain link, use the (Deep Q Network, DQN) algorithm to construct the corresponding supply chain optimization model, including the following steps: S3-2-1: Take the generation of the supply chain optimization plan as the simulation environment of the DQN algorithm, and construct the intelligent agent and the experience replay pool; S3-2-2: Define the state space of the DQN algorithm according to each supply chain state type corresponding to the historical supply chain risk prediction results, and the parameters of the state space correspond to each supply chain state. For example, the supply chain states in the planning link include the procurement plan state, the performance plan state, the inventory plan state, etc. The above planning states are used to characterize the state situations involved in the planning link in the supply chain data analysis process; S3-2-3: Define the action space of the DQN algorithm according to the actions that need to be output by the supply chain optimization strategy; for example, the actions in the planning link include adjusting the start time of the procurement plan in the procurement plan state, adjusting the completion time of the performance plan in the performance plan state, etc. Through the above actions, the adjustment of each state of the supply chain is realized, and thus the supply chain optimization is realized; S3-2-4: Define the reward function of the DQN algorithm according to the possible influence situations of each action in the action space, which is used to evaluate the quality or influence of the action; S3-2-5: Construct the input layer, several hidden layers, and output layer of the deep Q-network, connect the input layer to the state space, and connect the output layer to the action space; S3-2-6: Based on the state space, action space, and reward function, optimize and train the deep Q-network and the agent according to several historical supply chain risk prediction results, construct a supply chain optimization model, and store several generated supply chain optimization experiences in the experience replay pool; S3-3: Traverse the dimension-reduced historical supply chain data sets of all supply chain links, and construct a supply chain risk prediction model and a supply chain optimization model for each supply chain link; As described in step S1-1, the supply chain links include links such as planning, procurement, production, delivery, and transportation. The data indicators involved in each link are different, and the data structures vary greatly, making it impossible to perform unified data analysis. Existing artificial intelligence models require strict definition of the input feature quantities, resulting in the inability of the same supply chain risk prediction model and supply chain optimization model to analyze the supply chain data of different supply chain links. In this embodiment, corresponding supply chain risk prediction models and supply chain optimization models are trained for different supply chain links, improving the adaptability and matching degree to the data, and enhancing the efficiency and accuracy of data analysis; S3-4: Use the supply chain link as the retrieval label for the supply chain risk prediction model and the supply chain optimization model, and store several supply chain risk prediction models and several supply chain optimization models with the retrieval label set; S4: Collect the real-time supply chain data of the current supply chain node, obtain the corresponding matching supply chain link according to the Euclidean distance between the real-time supply chain data and several cluster centers, and perform data dimension reduction on the real-time supply chain data according to the matching key index set corresponding to the matching supply chain link to obtain the dimension-reduced real-time supply chain data, including the following steps: S4-1: Collect the real-time supply chain data stored in the data server of the current supply chain node, and set the corresponding real-time data mapping and conversion strategy according to the data format of the current supply chain node and the unified data model; S4-2: Perform data mapping and conversion on the real-time supply chain data according to the real-time data mapping and conversion strategy to obtain the converted real-time supply chain data; S4-3: Obtain the Euclidean distance between the converted real-time supply chain data and several cluster centers, and use the supply chain link of the cluster center with the closest Euclidean distance as the matching supply chain link; Pre-analyze the supply chain data to obtain its supply chain link information, providing support for subsequent model retrieval; S4-4: Extract the matching key index set corresponding to the matching supply chain link, and perform data dimension reduction on the converted real-time supply chain data according to the matching key index set to obtain the dimension-reduced real-time supply chain data; S5: Extract the matching supply chain risk prediction model and the matching supply chain optimization model corresponding to the matching supply chain link, and based on the dimension-reduced real-time supply chain data, use the matching supply chain risk prediction model to perform supply chain risk prediction to obtain the real-time supply chain risk prediction result, including the following steps: S5-1: Search among a number of retrieval tags according to the matching supply chain link to obtain the corresponding target retrieval tag; S5-2: Use the supply chain risk prediction model and the supply chain optimization model corresponding to the target retrieval tag as the matching supply chain risk prediction model and the matching supply chain optimization model; S5-3: Input the dimension-reduced real-time supply chain data into the matching supply chain risk prediction model, and use the matching supply chain risk prediction model to extract the real-time data features of the dimension-reduced real-time supply chain data; S5-4: Perform supply chain risk prediction based on the real-time data features to obtain the real-time supply chain risk prediction result; S6: According to the real-time supply chain risk prediction result, use the matching supply chain optimization model to perform matching supply chain optimization to obtain the real-time supply chain optimization strategy, return the real-time supply chain optimization strategy to the corresponding supply chain node, execute the real-time supply chain optimization strategy, and return to the real-time supply chain data collection step, including the following steps: S6-1: Update the model parameters of the matching supply chain optimization model according to the real-time supply chain risk prediction result to obtain the updated model parameters; The updated model parameters include the updated state space and the updated action space , where is the updated state value, is the state indicator, is the total number of state space dimensions, is the updated action value, is the action indicator; is the total number of action space dimensions; S6-2: Based on the updated model parameters, use the matching supply chain optimization model to perform matching supply chain optimization to obtain the real-time supply chain optimization strategy, including the following steps: S6-2-1: Use the updated state space as the input of the automatic control model, and use the deep Q network to generate the Q values of each possible action in the updated action space ; S6-2-2: Use the reward function to obtain the reward values of each possible action in the updated action space, and update the Q-values of the possible actions according to the reward values to obtain the updated Q-values of the possible actions; The formula is:

[0036] In the formula, is the updated state value and the updated action value corresponding updated Q-value; is the state value and the action value corresponding predicted Q-value; is the learning rate; is the highest predicted Q-value; S6-2-3: Repeat the above steps until the iteration number threshold is reached. Use the greedy strategy to take the possible action corresponding to the highest updated Q-value as the execution action, and output the execution action as the real-time supply chain optimization strategy; the real-time supply chain optimization strategy includes adjusting the start time of the procurement plan in the procurement plan adjustment state, adjusting the completion time of the fulfillment plan in the fulfillment plan adjustment state, etc.; S6-3: Return the real-time supply chain optimization strategy to the corresponding supply chain node, and use the decision support system of the supply chain node to execute the real-time supply chain optimization strategy and return to the real-time supply chain data collection step; the decision support system adjusts the start time of the procurement plan in the procurement plan adjustment state, adjusts the completion time of the fulfillment plan in the fulfillment plan adjustment state, etc. according to the detailed strategies involved in the real-time supply chain optimization strategy, generates the corresponding real-time supply chain optimization decision, adds parameters such as the procurement plan start time parameter, shortens the fulfillment plan completion time parameter, etc. Through the above optimization, the cost of the procurement plan in the start-up stage of the planning link can be increased, and the node time of the fulfillment plan in the planning link can be advanced, thereby solving the risks of high cost and low efficiency in the real-time supply chain risk prediction results.

[0037] Embodiment 2: As Figure 2 shown, this embodiment provides a supply chain risk prediction system for implementing the supply chain risk prediction method. The system includes a cloud data center and several supply chain nodes. Several supply chain nodes are all communicatively connected to the cloud data center, and the supply chain nodes include a data server and a decision support system; The data server is used to store historical / real-time supply chain data and provide data support for supply chain risk prediction and supply chain optimization; The decision support system is used to execute the real-time supply chain optimization strategy, make decision adjustments to the supply chain, and affect the collection and storage of real-time supply chain data; The cloud data center includes a big data collection unit, a data processing unit, a model construction unit, a supply chain link classification unit, a supply chain risk prediction unit, and a supply chain optimization unit that are connected in sequence. The big data collection unit and the supply chain link classification unit are respectively communicatively connected to the data servers of a number of supply chain nodes, and the supply chain optimization unit is communicatively connected to the decision support systems of a number of supply chain nodes; The big data collection unit is used to collect heterogeneous historical supply chain big data of historical supply chain data from different data sources and including different supply chain links, and preprocess the historical supply chain big data to obtain a number of preprocessed historical supply chain data of the same type; The data processing unit is used to perform clustering processing on a number of preprocessed historical supply chain data to obtain a number of cluster centers and corresponding data clusters, and perform data dimensionality reduction on the number of data clusters to obtain a dimensionality-reduced historical supply chain data set and a corresponding key index set for each supply chain link; The model construction unit is used to construct a corresponding supply chain risk prediction model and a supply chain optimization model using artificial intelligence algorithms based on the dimensionality-reduced historical supply chain data set, and obtain a supply chain risk prediction model and a supply chain optimization model for each supply chain link after traversing the dimensionality-reduced historical supply chain data sets of all supply chain links; The supply chain link classification unit is used to collect real-time supply chain data of the current supply chain node, obtain the corresponding matching supply chain link according to the Euclidean distance between the real-time supply chain data and a number of cluster centers, and perform data dimensionality reduction on the real-time supply chain data according to the matching key index set corresponding to the matching supply chain link to obtain dimensionality-reduced real-time supply chain data; The supply chain risk prediction unit is used to extract the matching supply chain risk prediction model and the matching supply chain optimization model corresponding to the matching supply chain link, and perform supply chain risk prediction using the matching supply chain risk prediction model according to the dimensionality-reduced real-time supply chain data to obtain a real-time supply chain risk prediction result; The supply chain optimization unit is used to perform matching supply chain optimization using the matching supply chain optimization model according to the real-time supply chain risk prediction result to obtain a real-time supply chain optimization strategy, and return the real-time supply chain optimization strategy to the corresponding supply chain node.

[0038] The present invention discloses a supply chain risk prediction method and system. By adopting a cloud data center, it can efficiently collect and process heterogeneous historical supply chain big data from different data sources, achieving data unification and standardization, and reducing the difficulty of data collection and integration. Through clustering processing and dimensionality reduction techniques, the present invention can extract key indicators from complex supply chain data, simplify the complexity of model construction, and improve the accuracy and effectiveness of prediction. By using artificial intelligence algorithms to construct a supply chain risk prediction model and an optimization model, the prediction accuracy and the effectiveness of the optimization strategy are significantly improved. By feeding back the supply chain optimization strategy to the supply chain nodes and executing it, a closed-loop management mechanism is formed, ensuring the continuity and effectiveness of supply chain risk prediction and management. By constructing corresponding supply chain risk prediction models and supply chain optimization models according to different supply chain links, the pertinence and practicability of risk prediction and optimization strategies are ensured. And when risks occur, a rapid response mechanism is adopted to avoid the lag of risk response measures, improve the flexibility and adaptability of the supply chain, and enable the supply chain to be optimized according to real-time situations. Through the cloud data center and artificial intelligence algorithms, a high degree of integration of supply chain risk management tools and systems is achieved, breaking the information silos and improving the overall management efficiency.

[0039] The present invention is not limited to the above optional embodiments, and any person can obtain other various forms of products under the inspiration of the present invention. The above specific embodiments should not be construed as limiting the protection scope of the present invention, and the protection scope of the present invention should be defined by the claims, and the specification can be used to interpret the claims.

Claims

1. A supply chain risk prediction method, characterized by: The steps include: Based on the cloud data center, the heterogeneous historical supply chain big data from different data sources and including historical supply chain data from different supply chain links are collected, and the historical supply chain big data is preprocessed to obtain a number of homogeneous preprocessed historical supply chain data; Perform clustering on several pre-processed historical supply chain data to obtain several cluster centers and corresponding data clusters, and perform data dimension reduction on several data clusters to obtain the dimension-reduced historical supply chain data set and the corresponding key indicator set for each supply chain link; Based on the dimension-reduced historical supply chain data set, use artificial intelligence algorithms to build corresponding supply chain risk prediction models and supply chain optimization models, and after traversing the dimension-reduced historical supply chain data sets of all supply chain links, obtain the supply chain risk prediction model and supply chain optimization model for each supply chain link; Collect the real-time supply chain data of the current supply chain node, obtain the corresponding matching supply chain link according to the Euclidean distance between the real-time supply chain data and several cluster centers, and perform data dimension reduction on the real-time supply chain data according to the matching key indicator set corresponding to the matching supply chain link to obtain the real-time supply chain data after dimension reduction; Extract the matching supply chain risk prediction model and matching supply chain optimization model corresponding to the matching supply chain link, and use the matching supply chain risk prediction model to perform supply chain risk prediction based on the real-time supply chain data after dimensionality reduction to obtain real-time supply chain risk prediction results; According to the real-time supply chain risk prediction results, the matching supply chain optimization model is used to perform matching supply chain optimization, obtain the real-time supply chain optimization strategy, return the real-time supply chain optimization strategy to the corresponding supply chain node, execute the real-time supply chain optimization strategy, and return to the real-time supply chain data collection step.

2. A supply chain risk prediction method according to claim 1, characterized in that: Based on the cloud data center, heterogeneous historical supply chain big data from different data sources and including historical supply chain data of different supply chain links are collected, and the historical supply chain big data are preprocessed to obtain a number of homogeneous preprocessed historical supply chain data, including the following steps: Based on the cloud data center, collect heterogeneous historical supply chain big data from different data sources and including historical supply chain data from different supply chain links; Screen the historical supply chain big data to obtain some cleaned historical supply chain data; Build a unified data model based on the data format of the cloud data center, and set corresponding historical data mapping and conversion strategies for each data source based on the unified data model; According to the historical data mapping and conversion strategy, data format mapping and conversion are performed on several heterogeneous cleaned historical supply chain data to obtain several homogeneous converted historical supply chain data.

3. A supply chain risk prediction method according to claim 1, characterized in that: Clustering is performed on several pre-processed historical supply chain data to obtain several cluster centers and corresponding data clusters, and data dimension reduction is performed on several data clusters to obtain the dimension-reduced historical supply chain data set and the corresponding key indicator set for each supply chain link, including the following steps: Taking the supply chain link as the clustering target and the data indicator difference as the clustering standard, the FCM clustering algorithm is used to cluster several pre-processed historical supply chain data to obtain several cluster centers. According to the Euclidean distance between several pre-processed historical supply chain data and each cluster center, the pre-processed historical supply chain data are divided into data clusters corresponding to each cluster center; Each data cluster is converted into a data matrix, and the PCA method is used to reduce the dimension of the data matrix to obtain the reduced dimension historical supply chain data set and the corresponding key indicator set for each supply chain link.

4. A supply chain risk prediction method according to claim 1, characterized in that: According to the dimension-reduced historical supply chain data set, the corresponding supply chain risk prediction model and supply chain optimization model are constructed using artificial intelligence algorithms. After traversing the dimension-reduced historical supply chain data set of all supply chain links, the supply chain risk prediction model and supply chain optimization model of each supply chain link are obtained, including the following steps: Based on the dimension-reduced historical supply chain data set of the same supply chain link, a corresponding supply chain risk prediction model is constructed using a deep learning algorithm, and several historical supply chain risk prediction results are generated; Based on several historical supply chain risk prediction results of the same supply chain link, a corresponding supply chain optimization model is constructed using a reinforcement learning algorithm; Traverse the dimension-reduced historical supply chain data sets of all supply chain links, and build supply chain risk prediction models and supply chain optimization models for each supply chain link; The supply chain links are used as retrieval tags for supply chain risk prediction models and supply chain optimization models, and several supply chain risk prediction models and several supply chain optimization models with retrieval tags are stored.

5. A supply chain risk prediction method according to claim 4, characterized in that: The supply chain risk prediction model is constructed based on the IFWA-DBN algorithm.

6. A supply chain risk prediction method according to claim 5, characterized in that: The supply chain optimization model is constructed based on the DQN algorithm.

7. A supply chain risk prediction method according to claim 2, characterized in that: Collect the real-time supply chain data of the current supply chain node, obtain the corresponding matching supply chain link according to the Euclidean distance between the real-time supply chain data and several cluster centers, and perform data dimension reduction on the real-time supply chain data according to the matching key indicator set corresponding to the matching supply chain link to obtain the real-time supply chain data after dimension reduction, including the following steps: Collect the real-time supply chain data stored in the data server of the current supply chain node, and set the corresponding real-time data mapping and conversion strategy according to the data format of the current supply chain node and the unified data model; According to the real-time data mapping and conversion strategy, the real-time supply chain data is mapped and converted to obtain the converted real-time supply chain data; Obtain the Euclidean distance between the converted real-time supply chain data and several cluster centers, and use the supply chain link of the cluster center with the closest Euclidean distance as the matching supply chain link; The matching key indicator set corresponding to the matching supply chain link is extracted, and the converted real-time supply chain data is subjected to data dimensionality reduction according to the matching key indicator set to obtain the real-time supply chain data after dimensionality reduction.

8. A supply chain risk prediction method according to claim 5, characterized in that: Extract the matching supply chain risk prediction model and matching supply chain optimization model corresponding to the matching supply chain link, and use the matching supply chain risk prediction model to perform supply chain risk prediction based on the real-time supply chain data after dimensionality reduction to obtain the real-time supply chain risk prediction results, including the following steps: According to the matching supply chain link, search among several search tags to obtain the corresponding target search tag; The supply chain risk prediction model and the supply chain optimization model corresponding to the target retrieval label are used as the matching supply chain risk prediction model and the matching supply chain optimization model; Input the reduced-dimensional real-time supply chain data into the matching supply chain risk prediction model, and use the matching supply chain risk prediction model to extract the real-time data features of the reduced-dimensional real-time supply chain data; Based on the real-time data characteristics, supply chain risk prediction is carried out to obtain real-time supply chain risk prediction results.

9. A supply chain risk prediction method according to claim 6, characterized in that: According to the real-time supply chain risk prediction results, the matching supply chain optimization model is used to match the supply chain optimization, obtain the real-time supply chain optimization strategy, return the real-time supply chain optimization strategy to the corresponding supply chain node, execute the real-time supply chain optimization strategy, and return the real-time supply chain data collection step, including the following steps: According to the real-time supply chain risk prediction results, the model parameters of the matching supply chain optimization model are updated to obtain updated model parameters; Based on the updated model parameters, the matching supply chain optimization model is used to perform matching supply chain optimization and obtain a real-time supply chain optimization strategy; The real-time supply chain optimization strategy is returned to the corresponding supply chain node, and the decision support system of the supply chain node is used to execute the real-time supply chain optimization strategy, and return to the real-time supply chain data collection step.

10. A supply chain risk prediction system, used to implement the supply chain risk prediction method according to any one of claims 1 to 9, characterized in that: The system includes a cloud data center and several supply chain nodes, the several supply chain nodes are all connected to the cloud data center in communication, and the supply chain nodes include a data server and a decision support system; The cloud data center includes a big data collection unit, a data processing unit, a model building unit, a supply chain link classification unit, a supply chain risk prediction unit and a supply chain optimization unit which are connected in sequence. The big data collection unit and the supply chain link classification unit are respectively communicated with the data servers of several supply chain nodes, and the supply chain optimization unit is respectively communicated with the decision support systems of several supply chain nodes.

Images