ERP system risk management system and method based on artificial intelligence

Abstract

The invention discloses an ERP system risk management system and method based on artificial intelligence, and relates to the technical field of system risk monitoring, and the risk management method comprises the following steps: collecting multi-source data related to a supply chain; the collected multi-source data are preprocessed; constructing a dynamic supply chain network by using the preprocessed data, and carrying out risk conduction path analysis; according to the result of the risk conduction path analysis, establishing a comprehensive risk assessment formula, and performing risk assessment on nodes and sub-networks in the dynamic supply chain network; and according to the risk assessment result, comparing the risk assessment result with a preset first risk threshold and a preset second risk threshold, classifying the risks, and performing management according to a classification result. According to the invention, through the comprehensive risk assessment formula, risk assessment is carried out on the nodes and the sub-networks in the dynamic network, the risk levels of different nodes and sub-networks can be quantified, and accurate risk scores are provided.

Description

Technical Field

[0001] The present invention relates to the technical field of system risk monitoring, and specifically to an ERP system risk management system and method based on artificial intelligence. Background Art

[0002] With the increasing complexity of the global supply chain, the Enterprise Resource Planning (ERP) system is more and more widely used in enterprise supply chain management. However, due to the large amount and diverse sources of data in links such as suppliers, logistics, warehousing, and customers involved in the supply chain, traditional ERP systems have limitations in risk management. The integration ability of multi-source data in existing ERP systems is limited, and it is difficult to effectively obtain and process data from multiple systems such as procurement, logistics, warehousing, and finance; existing methods usually conduct static risk assessments and cannot dynamically analyze the risk propagation paths and conduction intensities in the supply chain, resulting in low accuracy of risk early warnings; there is a lack of automated management and it is impossible to build a dynamic model according to the time changes of the supply chain network, making it difficult to adapt to the real-time and dynamic characteristics of risks. Summary of the Invention

[0003] The purpose of the present invention is to provide an ERP system risk management system and method based on artificial intelligence to solve the problems raised in the prior art.

[0004] To achieve the above purpose, the present invention provides the following technical solution: An ERP system risk management method based on artificial intelligence, the risk management method includes the following steps: Collect multi-source data related to the supply chain; Preprocess the collected multi-source data; Use the preprocessed data to build a dynamic supply chain network and conduct risk propagation path analysis; According to the results of the risk propagation path analysis, establish a comprehensive risk assessment formula and conduct risk assessments on the nodes and sub-networks in the dynamic supply chain network respectively; According to the risk assessment results, compare with the preset first risk threshold and second risk threshold, classify the risks, and manage according to the classification results; Collecting multi-source data related to the supply chain specifically includes: Search for purchase orders and purchase contract documents in the procurement department archives, and collect information such as supplier names, supplied products and services, transaction data, and transaction dates; Among them, the transaction data includes transaction amount, transaction frequency, and transaction failure frequency; Collect the transportation data of the supplier, and the transportation data includes information such as shipping time, estimated arrival time, actual arrival time, transportation method, and transportation cost; Regularly update the data of the warehouse, including the update of the warehouse location and the update of inventory consumption data, obtain the location information of the warehouse, and get the distances, transportation tools, transportation times, and transportation frequencies between each warehouse and its adjacent warehouses; By checking the payment history of customers, collect information on overdue payments and arrears of customers; Preprocess the collected multi-source data, specifically: Use the Python data processing library NumPy for data cleaning and feature extraction; For high-frequency noise, use Fourier transform to denoise; For continuous data, use moving average filtering to denoise; The continuous data includes inventory consumption and transportation duration; For specific tasks (such as missing value filling, standardization), combine the preprocessing tools provided by Scikit-learn for automated processing; Use the preprocessed data to construct a dynamic supply chain network, specifically: Each node in the supply chain network represents an entity in the supply chain, including suppliers, logistics nodes, and customers; The features of the supplier node include the historical on-time delivery rate, quality qualification rate, and order amount; The features of the logistics node include the transportation duration distribution, delay rate, and transportation cost; The features of the customer node include the order volume and transaction score; The specific calculation method of the transaction score is as follows: ; Among them, S represents the transaction score; f represents the transaction frequency of the supplier; r represents the transaction failure frequency of the supplier; f min and f max respectively represent the minimum and maximum values of the transaction frequency, set based on historical data; r min and r max represent the minimum and maximum values of the transaction failure frequency, set based on historical data; a1 and a2 respectively represent the weights of the transaction frequency and the transaction failure frequency, a1 + a2 = 1; Among them, the transaction score is a digital feature used to judge the customer transaction success rate; The edges in the supply chain network represent the relationships between nodes, including supply-demand relationships, logistics paths, and risk conduction weights; The risk conduction weight is represented by the edge weight to indicate the transmission intensity of risk between nodes; The edge weight is calculated based on the transaction frequency and the logistics path cost; Among them, suppliers: provide raw materials and components; Logistics nodes: responsible for transportation and warehousing; Customers: End - consumers and enterprises; Supply - demand relationship: The transaction relationship between suppliers and customers; Logistics path: The connection relationship between logistics nodes, suppliers, and customers; Model through a Temporal Graph, update the structure and characteristics of the supply - chain network at each time step to obtain a dynamic supply - chain network; Among them, the supply - chain network is dynamic. The characteristics of nodes and edges change over time. New suppliers join or a certain supplier withdraws, and the edge weights are adjusted due to seasonal fluctuations or market changes. Therefore, use a Temporal Graph to model and update the structure and characteristics of the graph at each time step; Specifically: The Graph Neural Network (GNN) realizes the representation and analysis of the entire network by recursively aggregating the neighborhood information of nodes; The GNN model structure includes message passing and risk - conduction path analysis; The specific message passing is as follows: In each iteration, each node receives information from its neighbors and updates its own state; Node update formula: ; where h i (t) represents the feature vector of node i at the t - th round; i and j represent nodes; h i (t+1) represents the feature vector of node i at the (t + 1)-th round; e ij represents the feature vector from node i to node j; AGGREGATE represents the information aggregation function; N(i) represents the set of neighbor nodes of node i; σ represents the activation function; W represents the learned weight matrix; Full - graph embedding. After multiple iterations, generate the final embedding representation h i (T) of each node, and generate the embedding representation h G of the entire graph through a pooling operation; Among them, the pooling operation includes global sum pooling, global average pooling, and global max pooling; The specific risk - conduction path analysis is as follows: Input: Initial nodes and their characteristics (such as suppliers with delivery failures); Output: The propagation path of risk in the graph; Propagation rule: According to the edge weights and node characteristics, the risk value spreads step by step along the edges until it stabilizes.

[0005] Based on the results of risk conduction path analysis, a comprehensive risk assessment formula is established to evaluate the risks of nodes and sub-networks in the dynamic supply chain network respectively, specifically as follows: To comprehensively evaluate the risk levels of nodes and sub-networks in the supply chain, a comprehensive risk assessment formula is constructed considering the following factors. The formula takes into account key indicators such as direct risk, conduction risk, redundancy, etc.; The comprehensive risk assessment formula includes a node risk assessment formula and a sub-network risk assessment formula; The specific form of the node risk assessment formula is as follows: For the risk score R of node i i , specifically: ; where D i represents the direct risk of node i; T i represents the conduction risk of node i; where the conduction risk represents the comprehensive impact of risk transfer from neighbor nodes to node i, which is calculated by combining the conduction probability and the risks of neighbor nodes; R i red represents the redundancy coefficient; α, β, and γ represent weight coefficients, and α + β + γ = 1; where α, β, and γ are used to control the contributions of each part to the comprehensive risk and are adjusted according to actual needs; where the redundancy coefficient is used to measure the redundancy degree of node i in the supply chain network (including the ratio of the number of available alternative suppliers to the total number of suppliers, and the ratio of the number of alternative logistics paths to the total number of paths), and its value range is [0, 1]; the larger the value, the higher the redundancy and the lower the risk; ; w ij represents the edge weight; D j represents the direct risk of node j; where the direct risk includes the delayed delivery rate, the probability of insufficient inventory, etc.; where w ij represents the risk conduction weight from node j to node i; The specific form of the sub-network risk assessment formula is as follows: For the risk score R of sub-network C C , based on the weighted average risk values of all included nodes, the formula is as follows: ; where I i represents the influence factor of node i; R i represents the risk score of node i.

[0006] Among them, the influence factor represents the influence of a node in the sub-network, including the proportion of transaction amount, logistics throughput, etc.

[0007] According to the risk assessment results, compare with the preset first risk threshold and second risk threshold, classify the risks, and manage according to the classification results. Specifically: Set the first risk threshold. When the risk score is greater than the first risk threshold, it is determined as a high risk; Set the second risk threshold. When the risk score is greater than the second risk threshold and less than or equal to the first risk threshold, it is determined as a medium risk; When the risk score is less than or equal to the second risk threshold, it is determined as a low risk.

[0008] Among them, the setting of the first risk threshold and the second risk threshold is determined by historical experience and business decisions, and the score value that seriously affects the stability of the supply chain is used as the threshold; Among them, the K-Means clustering method can also be used to group the risk scores and set the thresholds. Specifically: The nodes are divided into three categories according to the risk scores: high risk, medium risk, and low risk; Use the between-group boundary of the clustering results as the threshold. Specifically: The boundary where the central mean of high risk approaches medium risk is used as the first threshold; The boundary where the central mean of medium risk approaches low risk is used as the second threshold; Among them, for high-risk nodes: Negotiate with the supplier to sign an improvement agreement to improve service quality and stability; Introduce alternative suppliers and logistics service providers to reduce dependence; Increase inventory buffer at high-risk nodes to reduce the conduction impact; Medium-risk nodes: Regularly monitor data fluctuations to prevent risk deterioration; Provide appropriate support, such as adjusting the order size or optimizing the transportation plan; Low-risk nodes: Continue to maintain cooperation and communicate regularly to ensure a low-risk state.

[0009] Among them, for high-risk sub-networks: Increase the number of suppliers or logistics service providers, establish a multi-source supply chain model, and disperse risks; Through simulation analysis, test the breakpoint effect of different sub-networks to ensure that alternative nodes can take over quickly; Set safety inventory levels at high-risk sub-network nodes and use inventory buffers to cope with supply interruptions; Adjust the inventory strategy according to historical demand fluctuations and risk scores; The critical nodes in the high-risk sub-network can be changed to a localized supply mode to reduce cross-regional logistics risks; Adopt a near-shore supplier and customer matching mechanism to reduce the possibility of cross-network transmission; Among them, for the medium-risk sub-network: Collect real-time data on the critical nodes and edges of the medium-risk sub-network to improve the risk update frequency; use the stream processing framework Apache Flink to achieve efficient processing and monitoring of data streams; If the score of the medium-risk network continues to approach the high-risk threshold, promptly adjust it to a high-risk management mode and take more stringent measures; Prioritize the allocation of resources (such as spare inventory, temporary logistics support) to the medium-risk sub-network to avoid further escalation to high risk; Initiate contracts with reserve suppliers or partners to provide additional support; Establish buffer inventory at the critical warehouse nodes of the medium-risk sub-network to cope with possible demand surges or supply interruptions; Among them, for the low-risk sub-network: Regularly evaluate the score of the low-risk sub-network to ensure that it remains stable at a low-risk level; Establish a moderate redundant supply chain link (such as adding reserve suppliers or transportation channels) in the low-risk sub-network to cope with potential sudden risks; An ERP system risk management system based on artificial intelligence, the risk management system includes a multi-source data collection module, a data preprocessing module, a dynamic supply chain network construction module, a graph neural network analysis module, a risk assessment and classification module, and a visualization management module; The multi-source data collection module is used to collect multi-source data related to the supply chain; The data preprocessing module is used to clean and optimize the collected multi-source data and provide input for subsequent analysis; The dynamic supply chain network construction module is used to model suppliers, logistics nodes, and customers as nodes according to the preprocessed data, set edge weights, and construct a dynamic supply chain network; and update the network using Temporal Graph technology; the edge weights are calculated based on transaction frequency and logistics path cost; The graph neural network analysis module is used to analyze the risk conduction path through the graph neural network GNN and generate the embedding representation of nodes and the whole graph; The risk assessment and classification module is used to construct a comprehensive risk assessment formula based on the analysis of the risk conduction path, calculate the risk score according to the formula, and classify the risk based on the risk score and the preset threshold; The visualization management module is used to provide a user visualization interface for viewing and managing risk assessment, classification, and response measures. The multi-source data collection module provides different database interfaces, supports REST API and database connections, and directly obtains data from the database.

[0010] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. By constructing a multi-source data collection module, the present invention supports real-time data acquisition from multiple systems such as procurement, logistics, warehousing, and finance, enhancing data integrity and real-time performance. 2. The present invention uses Temporal Graph technology to construct a dynamic supply chain network, and analyzes risk conduction paths and node characteristics through Graph Neural Network (GNN), effectively capturing the time dynamics and spatial propagation characteristics of risks. 3. Through a comprehensive risk assessment formula, the present invention respectively conducts risk assessment on nodes and sub-networks in the dynamic network, can quantify the risk levels of different nodes and sub-networks, and provides accurate risk scores. 4. By setting the first and second risk thresholds, the present invention conducts classification management for high, medium, and low risks, improving the efficiency and accuracy of supply chain risk management. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 It is a schematic diagram of the steps of a risk management method for an ERP system based on artificial intelligence according to the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0012] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0013] Embodiment: As Figure 1 shown, the present invention provides a technical solution, a risk management method for an ERP system based on artificial intelligence, and the risk management method includes the following steps: Collect multi-source data related to the supply chain; Preprocess the collected multi-source data; Use the preprocessed data to construct a dynamic supply chain network and conduct risk conduction path analysis; According to the results of the risk conduction path analysis, establish a comprehensive risk assessment formula, and respectively conduct risk assessment on nodes and sub-networks in the dynamic supply chain network; According to the risk assessment results, compare them with the preset first risk threshold and second risk threshold, classify the risks, and manage them according to the classification results; Collect multi-source data related to the supply chain, specifically: Search for purchase orders and purchase contract documents in the procurement department's archives, and collect information on supplier names, supplied products and services, transaction data, and transaction dates; Among them, the transaction data includes transaction amount, transaction frequency, and transaction failure frequency; Collect the transportation data of the supplier, and the transportation data includes shipping time, estimated arrival time, actual arrival time, transportation mode, and transportation cost information; Regularly update the data of the warehouse, including warehouse location update and inventory consumption data update, obtain the location information of the warehouse, and get the distances, transportation tools, transportation times, and transportation frequencies between each warehouse and its adjacent warehouses; Collect customer overdue payment and arrears information by checking the customer's payment history.

[0014] Preprocess the collected multi-source data, specifically: Use the Python data processing library NumPy for data cleaning and feature extraction; For high-frequency noise, use Fourier transform to denoise; Use moving average filtering to denoise continuous data; the continuous data includes inventory consumption and transportation duration.

[0015] For specific tasks (such as missing value filling, standardization), combine the preprocessing tools provided by Scikit-learn for automated processing.

[0016] Use the preprocessed data to construct a dynamic supply chain network, specifically: Each node in the supply chain network represents an entity in the supply chain, including suppliers, logistics nodes, and customers; The features of the supplier node include historical delivery on-time rate, quality qualification rate, and order amount; The features of the logistics node include transportation duration distribution, delay rate, and transportation cost; The features of the customer node include order volume and transaction score; The specific calculation method of the transaction score is as follows: ; Among them, S represents the transaction score; f represents the transaction frequency of the supplier; r represents the transaction failure frequency of the supplier; f min and f max respectively represent the minimum and maximum values of the transaction frequency, set based on historical data; r min and rmax Represents the minimum and maximum values of the transaction failure frequency, set based on historical data; a1 and a2 respectively represent the weights of the transaction frequency and the transaction failure frequency, and a1 + a2 = 1; Among them, the transaction score is a digital feature used to judge the customer transaction success rate.

[0017] The edges in the supply chain network represent the relationships between nodes, including supply-demand relationships, logistics paths, and risk conduction weights; The risk conduction weight represents the transmission intensity of risks between nodes using edge weights; The edge weights are calculated based on the transaction frequency and the logistics path cost; Among them, suppliers: provide raw materials and components; Logistics nodes: responsible for transportation and warehousing; Customers: the ultimate consumer groups and enterprises; Supply-demand relationship: the transaction relationship between suppliers and customers; Logistics path: the connection relationship between logistics nodes and suppliers and customers; Modeled by the Temporal Graph, and update the structure and features of the supply chain network according to time steps to obtain a dynamic supply chain network.

[0018] Among them, the supply chain network is dynamic, and the features of nodes and edges will change over time. New suppliers join or a certain supplier withdraws, and the edge weights are adjusted due to seasonal fluctuations or market changes. Therefore, use the Temporal Graph to model and update the structure and features of the graph according to time steps; Specifically: The Graph Neural Network (GNN) realizes the representation and analysis of the entire network by recursively aggregating the neighborhood information of nodes; The GNN model structure includes message passing and risk conduction path analysis; The specific message passing is as follows: In each round of iteration, each node receives information from its neighbors and updates its own state; the node update formula: ; Among them, h i (t) represents the feature vector of node i at the t-th round; i and j represent nodes; h i (t+1) represents the feature vector of node i at the (t + 1)-th round; e ij represents the feature vector from node i to node j; AGGREGATE represents the information aggregation function; N(i) represents the set of neighbor nodes of node i; σ represents the activation function; W represents the learned weight matrix; Full graph embedding. After multiple rounds of iteration, the final embedding representation h of each node is generated i (T) , and the embedding representation h of the entire graph is generated through a pooling operation G ; Among them, the pooling operation includes global sum pooling, global average pooling, and global max pooling; The specific risk conduction path analysis is as follows: Input: Initial node and its features (such as suppliers with delivery failures); Output: The propagation path of risk in the graph; Propagation rule: According to the edge weight and node features, the risk value gradually diffuses along the edge until it stabilizes.

[0019] According to the results of the risk conduction path analysis, a comprehensive risk assessment formula is established to evaluate the risks of nodes and sub-networks in the dynamic supply chain network respectively, specifically: To comprehensively evaluate the risk levels of nodes and sub-networks in the supply chain, a comprehensive risk assessment formula is constructed considering the following factors. The formula takes into account key indicators such as direct risk, conduction risk, redundancy, etc.; The comprehensive risk assessment formula includes a node risk assessment formula and a sub-network risk assessment formula; The specific node risk assessment formula is: For the risk score R of node i i , specifically: ; Among them, D i represents the direct risk of node i; T i represents the conduction risk of node i; Among them, the conduction risk represents the comprehensive impact of risk transfer from neighbor nodes to node i, which is comprehensively calculated by the conduction probability and the risks of neighbor nodes; R i red represents the redundancy coefficient; α, β, and γ represent weight coefficients, and α + β + γ = 1; Among them, α, β, and γ are used to control the contributions of each part to the comprehensive risk and are adjusted according to actual needs; Among them, the redundancy coefficient is used to measure the redundancy degree of node i in the supply chain network (including the ratio of the number of available alternative suppliers to the total number of suppliers, the ratio of the number of alternative logistics paths to the total number of paths), and the value range is [0, 1]; the larger the value, the higher the redundancy and the lower the risk; ; w ij represents the edge weight; D j represents the direct risk of node j; Among them, the direct risks include the delivery delay rate, the probability of insufficient inventory, etc.; Among them, w ij represents the risk conduction weight from node j to node i; The sub-network risk assessment formula is specifically as follows: For the risk score R C of sub-network C, based on the weighted average risk values of all included nodes, the formula is as follows: ; Among them, I i represents the influence factor of node i; R i represents the risk score of node i; Among them, the influence factor represents the influence of the node in the sub-network, including the proportion of transaction amount, logistics throughput, etc.; Direct risk D i : includes the risk characteristics of the node itself, If the inventory shortage rate of a warehouse reaches 80%, then D i has a greater contribution to its overall risk; Conduction risk T i : The risk propagated through the graph structure takes into account how the risk is transmitted from other nodes to the current node; The delay risk of a certain logistics node is transmitted through the edge to multiple downstream nodes, and the risk T i of downstream customers will be affected; Redundancy coefficient R i red : reflects the flexibility and buffering capacity of the supply chain; If a node depends on a single supplier, the redundancy R i red is close to 0, and its risk score is higher; Sub-network risk R C : aggregates the risk values of all nodes in the sub-network and focuses on measuring the overall impact of the entire sub-network.

[0020] Node i: A certain supplier, the inventory shortage rate D i = 0.6, the conduction risk T i received from the upstream = 0.4, the redundancy coefficient R i red = 0.3; Weight coefficients: α = 0.4; β = 0.4; γ = 0.2; Node risk score calculation: R i = 0.54; Sub-network C consists of three nodes with node importance I = [1, 2, 1], and the node risk scores are in sequence: [0.54, 0.6, 0.5]; Risk score of the sub-network: R C = 0.56; According to the risk assessment results, compare with the preset first risk threshold and second risk threshold, classify the risks, and manage according to the classification results. Specifically: Set the first risk threshold. When the risk score is greater than the first risk threshold, it is determined as high risk; Set the second risk threshold. When the risk score is greater than the second risk threshold and less than or equal to the first risk threshold, it is determined as medium risk; When the risk score is less than or equal to the second risk threshold, it is determined as low risk; Among them, the setting of the first risk threshold and the second risk threshold is determined by historical experience and business decisions, and the score value that seriously affects the stability of the supply chain is used as the threshold; Among them, the K-Means clustering method can also be used to group the risk scores and set the thresholds. Specifically: The nodes are divided into three categories according to the risk scores: high risk, medium risk, and low risk; Use the between-group boundary of the clustering results as the threshold. Specifically: The boundary where the central mean of high risk approaches medium risk is used as the first threshold; The boundary where the central mean of medium risk approaches low risk is used as the second threshold; Among them, for high-risk nodes: Negotiate with the supplier to sign an improvement agreement to improve service quality and stability; Introduce alternative suppliers and logistics service providers to reduce dependence; Increase inventory buffer at high-risk nodes to reduce the conduction impact; Medium-risk nodes: Regularly monitor data fluctuations to prevent risk deterioration; Provide appropriate support, such as adjusting the order size or optimizing the transportation plan; Low-risk nodes: Continue to maintain cooperation and communicate regularly to ensure a low-risk state.

[0021] Among them, for high-risk sub-networks: Increase the number of suppliers or logistics service providers, establish a multi-source supply chain model to disperse risks; Through simulation analysis, test the breakpoint effect of different sub-networks to ensure that alternative nodes can take over quickly; Set safety inventory levels at high-risk sub-network nodes and use inventory buffers to cope with supply interruptions; Adjust the inventory strategy according to historical demand fluctuations and risk scores; The key nodes in the high-risk sub-network can be changed to a localized supply mode to reduce cross-regional logistics risks; Adopt a mechanism for matching near-source suppliers and customers to reduce the possibility of cross-network conduction.

[0022] Among them, for the medium-risk sub-network: Collect real-time data on the key nodes and edges of the medium-risk sub-network to improve the risk update frequency; use the stream processing framework Apache Flink to achieve efficient processing and monitoring of data streams; If the score of the medium-risk network continues to approach the high-risk threshold, promptly adjust it to a high-risk management mode and take more stringent measures; Prioritize the allocation of resources (such as spare inventory, temporary logistics support) to the medium-risk sub-network to avoid further escalation to high risk; Initiate contracts with reserve suppliers or partners to provide additional support; Establish buffer inventory at the key warehouse nodes of the medium-risk sub-network to cope with possible demand surges or supply interruptions.

[0023] Among them, for the low-risk sub-network: Regularly evaluate the scores of the low-risk sub-network to ensure that it remains stable at a low-risk level; Evaluate key indicators such as transaction frequency and transportation delay rate quarterly; Establish appropriate redundant supply chain links (such as adding backup suppliers or transportation channels) in the low-risk sub-network to cope with potential sudden risks; An ERP system risk management system based on artificial intelligence, the risk management system includes a multi-source data collection module, a data preprocessing module, a dynamic supply chain network construction module, a graph neural network analysis module, a risk assessment and classification module, and a visualization management module; The multi-source data collection module is used to collect multi-source data related to the supply chain; The data preprocessing module is used to clean and optimize the collected multi-source data to provide input for subsequent analysis; The dynamic supply chain network construction module is used to model suppliers, logistics nodes, and customers as nodes according to the preprocessed data, set edge weights, and construct a dynamic supply chain network; and use Temporal Graph technology to update the network; the edge weights are calculated based on transaction frequency and logistics path costs; The graph neural network analysis module is used to analyze the risk conduction path through the graph neural network GNN and generate the embedding representations of nodes and the whole graph; The risk assessment and classification module is used to construct a comprehensive risk assessment formula according to the analysis of the risk conduction path, calculate the risk score according to the formula, and classify the risks based on the risk score and the preset threshold; The visualization management module is used to provide a user visualization interface to view and manage risk assessment, classification, and response measures; The multi-source data collection module provides different database interfaces, supports REST API and database connections, and directly obtains data from the database.

[0024] For those skilled in the art, it is obvious that the present invention is not limited to the details of the above-described exemplary embodiments, and can be implemented in other specific forms without departing from the spirit or basic characteristics of the present invention. Therefore, from any point of view, the embodiments should be regarded as exemplary and non-limiting. The scope of the present invention is defined by the appended claims rather than the above description. Therefore, all changes falling within the meaning and scope of the equivalent elements of the claims are intended to be included in the present invention. Any reference signs in the claims should not be construed as limiting the claimed claim.

Claims

1. An ERP system risk management method based on artificial intelligence, characterized in that: The risk management method includes the following steps: Collect multi-source data related to the supply chain; Preprocess the collected multi-source data; Use the preprocessed data to construct a dynamic supply chain network and conduct risk conduction path analysis; According to the results of the risk conduction path analysis, establish a comprehensive risk assessment formula, and conduct risk assessments on the nodes and sub-networks in the dynamic supply chain network respectively; According to the risk assessment results, compare with the preset first risk threshold and second risk threshold, classify the risks, and manage according to the classification results.

2. The risk management method of an ERP system based on artificial intelligence according to claim 1, wherein: Collect multi-source data related to the supply chain, specifically: Search for purchase orders and purchase contract documents in the procurement department archives, and collect supplier names, supplied products and services, transaction data, and transaction date information; Among them, the transaction data includes transaction amount, transaction frequency, and transaction failure frequency; Collect the transportation data of the supplier, and the transportation data includes shipping time, estimated arrival time, actual arrival time, transportation mode, and transportation cost information; Regularly update the data of the warehouse, including warehouse location update and inventory consumption data update, obtain the location information of the warehouse, and get the distances, transportation tools, transportation times, and transportation frequencies between each warehouse and its adjacent warehouses; Collect customer overdue payment and arrears information by viewing the customer's payment history.

3. The risk management method of an ERP system based on artificial intelligence according to claim 2, characterized in that: Preprocess the collected multi-source data, specifically: Use the Python data processing library NumPy for data cleaning and feature extraction; For high-frequency noise, use Fourier transform for denoising; Use moving average filtering for denoising of continuous data; The continuous data includes inventory consumption and transportation duration.

4. A risk management method for an ERP system based on artificial intelligence according to claim 3, characterized in that: Use the preprocessed data to construct a dynamic supply chain network, specifically: Each node in the supply chain network represents an entity in the supply chain, including suppliers, logistics nodes, and customers; The features of the supplier node include historical delivery on-time rate, quality qualification rate, and order amount; The features of the logistics node include transportation duration distribution, delay rate, and transportation cost; The features of the customer node include order volume and transaction score; The transaction score is obtained by fusing and calculating the transaction frequency and transaction failure frequency of the supplier in history; The edges in the supply chain network represent the relationships between nodes, including supply-demand relationships, logistics paths, and risk conduction weights; The risk conduction weight represents the transmission intensity of the risk between nodes with edge weights; The edge weight is calculated based on the transaction frequency and logistics path cost; Model through the Temporal Graph, and update the structure and features of the supply chain network at each time step to obtain a dynamic supply chain network.

5. A risk management method for an ERP system based on artificial intelligence according to claim 4, characterized in that: Specifically: The Graph Neural Network (GNN) realizes the representation and analysis of the entire dynamic supply chain network by recursively aggregating the neighborhood information of nodes; The GNN model structure includes message passing and risk conduction path analysis; The risk conduction path analysis is specifically: Input: Initial node and its features; Output: The propagation path of the risk in the graph; Propagation rule: According to the edge weight and node features, the risk value gradually spreads along the edge until all nodes are covered.

6. A risk management method for an ERP system based on artificial intelligence according to claim 5, characterized in that: According to the results of risk conduction path analysis, a comprehensive risk assessment formula is established to conduct risk assessments on the nodes and sub-networks in the dynamic supply chain network respectively, specifically as follows: The comprehensive risk assessment formula includes a node risk assessment formula and a sub-network risk assessment formula; The node risk assessment formula is jointly composed of the direct risk of the node, the conduction risk of the node, and the redundancy of the node; The sub-network risk assessment formula is based on the weighted average risk values of all nodes included.

7. The risk management method of an ERP system based on artificial intelligence according to claim 6, characterized in that: According to the risk assessment results, compare with the preset first risk threshold and second risk threshold, classify the risks, and manage according to the classification results, specifically as follows: Set the first risk threshold. When the risk score is greater than the first risk threshold, it is determined as a high risk; Set the second risk threshold. When the risk score is greater than the second risk threshold and less than or equal to the first risk threshold, it is determined as a medium risk; When the risk score is less than or equal to the second risk threshold, it is determined as a low risk.

8. An ERP system risk management system based on artificial intelligence, which applies a risk management method for an ERP system based on artificial intelligence as described in any one of claims 1-7, characterized in that: The risk management system includes a multi-source data collection module, a data preprocessing module, a dynamic supply chain network construction module, a graph neural network analysis module, a risk assessment and classification module, and a visualization management module; The multi-source data collection module is used to collect multi-source data related to the supply chain; The data preprocessing module is used to clean and optimize the collected multi-source data to provide input for subsequent analysis; The dynamic supply chain network construction module is used to model suppliers, logistics nodes, and customers as nodes according to the preprocessed data, set edge weights, and construct a dynamic supply chain network; And use Temporal Graph technology to update the network; the edge weights are calculated based on transaction frequencies and logistics path costs; The graph neural network analysis module is used to analyze the risk conduction path through the graph neural network GNN and generate the embedding representations of nodes and the whole graph; The risk assessment and classification module is used to construct a comprehensive risk assessment formula according to the risk conduction path analysis, calculate the risk score according to the formula, and classify the risks according to the risk score and the preset threshold; The visualization management module is used to provide a user visualization interface to view and manage risk assessments, classifications, and countermeasures.

9. The risk management system of an AI-based ERP system according to claim 8, wherein: The multi-source data collection module provides different database interfaces, supports REST API and database connections, and directly obtains data from the database.

Images