Low-code ERP construction method for connecting upstream and downstream businesses in the supply chain

Abstract

This invention provides a low-code ERP construction method for connecting upstream and downstream businesses in the supply chain, relating to the field of enterprise management software technology. The method includes: acquiring business process data, extracting key node information to generate a configuration interface and data collection forms, determining node priorities based on time-series memory operations and detecting anomalies using an isolated forest algorithm, generating business rule configuration parameters, performing data desensitization and encryption processing and performing multi-party secure computation, applying business collaboration optimization strategies and updating configuration parameters, thereby enabling flexible configuration, secure collaboration, and intelligent optimization of supply chain businesses.

Description

Technical Field

[0001] This invention relates to the field of enterprise management software technology, and in particular to a low-code ERP construction method for connecting upstream and downstream businesses in the supply chain. Background Technology

[0002] As supply chain networks become increasingly complex, the demand for business collaboration and data sharing among enterprises continues to grow. Traditional ERP systems face significant challenges in connecting upstream and downstream enterprises in the supply chain. Low-code development platforms, due to their flexibility and efficiency, are gradually becoming an important tool for enterprises to build customized business systems. In the field of supply chain management, low-code ERP simplifies the development process, enabling enterprises to quickly adapt to market changes and business needs.

[0003] Currently, low-code ERP systems are widely used in various supply chain links such as procurement management, inventory control, order processing, and production planning, helping enterprises to standardize and automate business processes. However, existing low-code ERP construction methods lack the ability to intelligently analyze and optimize supply chain business processes, cannot automatically identify key business nodes and prioritize them, often rely on manual experience when configuring business rules, making it difficult to cope with complex and ever-changing supply chain business scenarios, have relatively simple exception handling mechanisms, usually use preset rules for judgment and processing, lack intelligent exception detection and processing capabilities, cannot effectively deal with various abnormal situations in the supply chain, and have difficulty achieving efficient collaborative computing while ensuring data privacy and security. Summary of the Invention

[0004] This invention provides a low-code ERP construction method for connecting upstream and downstream businesses in the supply chain, which can at least solve some of the problems existing in the prior art.

[0005] A first aspect of this invention provides a low-code ERP construction method for connecting upstream and downstream businesses in a supply chain, comprising:

[0006] Acquire business process data from upstream and downstream enterprises in the supply chain, extract key business node information, and generate business component configuration interfaces and business data collection forms.

[0007] Based on the business data in the business data collection form, feature encoding is performed on the key business node information to extract business process feature data. Temporal memory operation is performed on the business process feature data to obtain temporal dependency features and calculate the importance weight coefficient of business nodes to obtain a business node priority list. Anomaly detection is performed on the temporal dependency features based on the isolated forest algorithm to determine the anomaly type list and generate an anomaly handling action sequence through reinforcement learning algorithm. A set of business rule configuration parameters is generated based on the business node priority list and the anomaly handling action sequence.

[0008] Configure business functions based on the business component configuration interface and the set of business rule configuration parameters, and generate business rule execution instructions and data processing flow;

[0009] According to the business rules, the execution instructions and data processing flow are carried out. The data in the business data collection form is desensitized to obtain local calculation parameters. The local calculation parameters are encrypted to generate an encrypted parameter package and perform multi-party secure calculation to obtain global calculation parameters. Based on sensitivity analysis and adaptive parameter allocation algorithm, the global calculation parameters are allocated to the enterprise's local server to form a business collaborative optimization strategy.

[0010] The business collaboration optimization strategy is applied to the business rule execution instructions, the business execution status is monitored, the execution data is collected, and the set of business rule configuration parameters is updated.

[0011] In one alternative implementation,

[0012] Acquire business process data from upstream and downstream enterprises in the supply chain, extract key business node information, and generate business component configuration interfaces and business data collection forms, including:

[0013] Business process data is obtained by collecting sales order processing records, procurement plan execution records, and warehouse inventory change records from upstream and downstream enterprises in the supply chain and combining them. The business process data is then parsed to identify the data input types, business processing rules, and status change triggering conditions to obtain key business node information.

[0014] Based on the key business node information, a list of component types is constructed. Input specifications, business rules, and status change control attributes are configured for each component type in the business component configuration interface. A business data collection form is generated based on the key business node information. Data input validation procedures, business rule setting procedures, and status change control functions are configured in the business data collection form interface.

[0015] In one alternative implementation,

[0016] Based on the business data in the business data collection form, feature encoding is performed on the key business node information to extract business process feature data. Temporal memory operations are then performed on the business process feature data to obtain temporal dependency features, and the importance weight coefficients of business nodes are calculated to obtain a priority list of business nodes, including:

[0017] Based on the business data in the business data collection form, feature encoding is performed on the key business node information to obtain business process feature data. The business process feature data is constructed into a graph structure. The node feature similarity is obtained by calculating the overlap of business processing rules between nodes and the similarity of state change triggering conditions. The temporal dependency strength is obtained based on the temporal sequence of data input and output between nodes and the triggering response time interval of state change. The weighted sum of the node feature similarity and temporal dependency strength is used as the edge weight between nodes.

[0018] Feature extraction is performed on the graph structure based on a pre-set graph convolutional network. Local feature representations for each node are obtained by weighted aggregation of the features of the neighbor node set corresponding to each node. Temporal dependency features are obtained by selectively retaining and updating the local feature representations and historical temporal state information through a long short-term memory algorithm. The local feature representations and temporal dependency features are then weighted and fused to obtain the fused features of the nodes.

[0019] A random perturbation is added to the fusion feature, and the expected difference between the fusion features of other nodes before and after the perturbation is calculated to obtain the direct causal effect between nodes. The importance weight coefficient of the business node is determined based on the direct causal effect. The priority score of the business node is calculated based on the importance weight coefficient of the business node and the pre-set time series influence factor. The business nodes are sorted in descending order according to the priority score to obtain the priority list of business nodes.

[0020] In one alternative implementation,

[0021] Anomaly detection is performed on temporal dependency features based on the isolation forest algorithm to determine an anomaly type list. An anomaly handling action sequence is generated using a reinforcement learning algorithm. Based on the business node priority list and the anomaly handling action sequence, a set of business rule configuration parameters is generated, including:

[0022] Anomaly detection of temporal dependency features is performed using the isolated forest algorithm. Anomaly scores are calculated based on the ratio of the average path length of the temporal dependency features to the normalization factor. Temporal dependency features with anomaly scores greater than a preset anomaly threshold are identified as anomalous features. Based on cluster analysis, the anomalous features are divided into multiple anomaly types and an anomaly type list is constructed.

[0023] For each exception type in the exception type list, determine the policy parameters that include the action space, state space and reward function, and calculate the initial policy function corresponding to each exception type based on the policy parameters;

[0024] The propagation probability between different anomaly types in the anomaly type list is calculated by using a pre-set correlation evaluation function and a sigmoid activation function. Based on the propagation probability, the optimized policy function is obtained by calculating the KL divergence weighted sum between the pre-set policy expected reward and the initial policy function.

[0025] Obtain the priority list of business nodes, calculate the attention of each anomaly type by applying attention to the current state and hidden state, calculate the fusion weight coefficient of the anomaly type based on the attention, and obtain the fusion strategy function by weighting the optimization strategy function corresponding to each anomaly type with the fusion weight coefficient.

[0026] An exception handling action sequence is generated based on the fusion strategy function, and the exception handling action sequence is mapped to a predefined rule template to obtain a set of business rule configuration parameters.

[0027] In one alternative implementation,

[0028] Based on the business component configuration interface and the set of business rule configuration parameters, configure business functions, and generate business rule execution instructions and data processing flows, including:

[0029] The configuration parameters of the business component configuration interface are collected. The configuration parameters include component identifier, component parameters and connection relationships between components.

[0030] The configuration parameters are matched and mapped with the set of business rule configuration parameters. Component instantiation configuration and component calling order are generated based on preset mapping rules. Data processing flow is constructed based on the component instantiation configuration. Business rule execution instructions are generated according to the component calling order.

[0031] In one alternative implementation,

[0032] Based on business rules, instructions and data processing flows are executed. Data in the business data collection form is anonymized to obtain local calculation parameters. These local calculation parameters are then encrypted to generate an encrypted parameter package, which is then subjected to multi-party secure computation to obtain global calculation parameters, including:

[0033] According to the business rules, the execution instructions extract enterprise operating data and transaction amount data from the business data collection form as sensitive fields, calculate the sensitivity and privacy budget parameters corresponding to the sensitive fields, and classify the sensitive fields into categorical data and numerical data according to the data format;

[0034] The privacy budget parameter and the number of categories in the categorical data are used as input parameters of a pre-set random response function to obtain the first de-identified data. Laplace distribution noise is calculated based on sensitivity and privacy budget parameter. Laplace distribution noise is added to the numerical data to obtain the second de-identified data. The first de-identified data and the second de-identified data are combined to obtain the local calculation parameter.

[0035] Obtain the public and private keys generated based on additive homomorphic encryption, encrypt the local computation parameters based on the public key to obtain the encrypted parameter packet, generate a parameter confusion matrix based on the encrypted parameter packet, fill the encrypted parameter packet in the diagonal positions of the parameter confusion matrix, and fill the off-diagonal positions with the encrypted result of random confusion values;

[0036] The encrypted parameter packets in the parameter confusion matrix are fragmented and sent to multiple computing nodes. A secure summation operation is performed to obtain the encrypted parameter share. The multi-party computation result is obtained by decryption using the private key. Based on the multi-party computation result, the sample mean and variance parameters and Euclidean distance are calculated to determine the parameter aggregation weight. The multi-party computation result is weighted and summed based on the parameter aggregation weight to obtain the global computation parameter.

[0037] In one alternative implementation,

[0038] Based on sensitivity analysis and adaptive parameter allocation algorithms, global computational parameters are distributed to local enterprise servers to form a business collaborative optimization strategy, including:

[0039] Business data samples are collected from the enterprise's local server. Feature extraction is performed on the business data samples to obtain sample feature vectors and the corresponding gradient sensitivity is calculated. A sensitivity matrix is ​​constructed based on the gradient sensitivity.

[0040] The average sensitivity is obtained by summing the absolute values ​​of each element in the sensitivity matrix. The fluctuation level is obtained by calculating the square root of the sum of squares of each element in the sensitivity matrix. The average sensitivity and the fluctuation level are weighted to obtain the parameter importance index. The global calculated parameters are sorted in descending order according to the parameter importance index and grouped into multiple parameter groups by combining the preset importance threshold. The mean vector of the parameter vector in each parameter group is calculated as the feature center of the current parameter group.

[0041] The system obtains the performance metrics of the enterprise's local server and constructs a performance feature vector. It calculates the cosine similarity between the performance feature vector and each parameter group, as well as the Euclidean distance between the performance feature vector and the feature center. The matching coefficient is obtained by summing the cosine similarity and the Euclidean distance after exponential decay.

[0042] Based on the matching degree coefficient, a softmax operation is performed to obtain the parameter allocation probability. Based on the parameter allocation probability, the globally calculated parameters are allocated to the enterprise's local server to obtain a business collaboration optimization strategy.

[0043] A second aspect of the present invention provides a low-code ERP building system for connecting upstream and downstream businesses in a supply chain, comprising:

[0044] The first unit is used to acquire business process data of upstream and downstream enterprises in the supply chain, extract key business node information, and generate business component configuration interfaces and business data collection forms.

[0045] The second unit is used to extract business process feature data by feature encoding of key business node information based on business data in the business data collection form, perform temporal memory operation on the business process feature data to obtain temporal dependency features and calculate the importance weight coefficient of business nodes to obtain a business node priority list, perform anomaly detection on the temporal dependency features based on the isolated forest algorithm, determine the anomaly type list and generate anomaly handling action sequence through reinforcement learning algorithm, and generate a set of business rule configuration parameters based on the business node priority list and anomaly handling action sequence.

[0046] The third unit is used to configure business functions based on the business component configuration interface and the set of business rule configuration parameters, and to generate business rule execution instructions and data processing flow.

[0047] The fourth unit is used to execute instructions and data processing flow according to business rules, perform desensitization processing on the data in the business data collection form to obtain local calculation parameters, encrypt the local calculation parameters, generate encrypted parameter packages and perform multi-party secure calculations to obtain global calculation parameters, and allocate the global calculation parameters to the enterprise local server based on sensitivity analysis and adaptive parameter allocation algorithm to form a business collaborative optimization strategy.

[0048] The fifth unit is used to apply business collaboration optimization strategies to business rule execution instructions, monitor business execution status, collect execution data, and update the set of business rule configuration parameters.

[0049] A third aspect of the present invention provides an electronic device, comprising:

[0050] A processor and a memory for storing processor-executable instructions, wherein the processor is configured to invoke instructions stored in the memory to perform the aforementioned method.

[0051] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0052] This invention lowers the technical barrier to system integration between supply chain enterprises by automatically extracting key business node information and generating configuration interfaces and data collection forms. This allows non-technical personnel to quickly complete business system configuration, improving the integration efficiency and accuracy of cross-enterprise business processes. Temporal memory computation and the isolated forest algorithm are used for feature analysis and anomaly detection of business processes. Combined with reinforcement learning to generate anomaly handling solutions, intelligent configuration of business rules and automatic anomaly handling are achieved, significantly enhancing the stability and resilience of supply chain collaboration. Data anonymization and multi-party secure computation technologies ensure the security of data sharing between enterprises. Simultaneously, sensitivity analysis and adaptive parameter allocation algorithms optimize resource allocation, promoting efficient collaboration between upstream and downstream businesses while protecting core enterprise data, thus improving the overall operational efficiency and resilience of the supply chain. Attached Figure Description

[0053] Figure 1 This is a flowchart illustrating a low-code ERP construction method for connecting upstream and downstream businesses in the supply chain, as described in an embodiment of the present invention.

[0054] Figure 2 This is a flowchart illustrating the privacy-preserving multi-party secure computation process of a low-code ERP construction method for connecting upstream and downstream businesses in the supply chain, as described in an embodiment of the present invention. Detailed Implementation

[0055] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0056] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0057] Figure 1 This is a flowchart illustrating a low-code ERP construction method for connecting upstream and downstream businesses in the supply chain, as described in an embodiment of the present invention. Figure 1 As shown, the method includes:

[0058] Acquire business process data from upstream and downstream enterprises in the supply chain, extract key business node information, and generate business component configuration interfaces and business data collection forms.

[0059] Based on the business data in the business data collection form, feature encoding is performed on the key business node information to extract business process feature data. Temporal memory operation is performed on the business process feature data to obtain temporal dependency features and calculate the importance weight coefficient of business nodes to obtain a business node priority list. Anomaly detection is performed on the temporal dependency features based on the isolated forest algorithm to determine the anomaly type list and generate an anomaly handling action sequence through reinforcement learning algorithm. A set of business rule configuration parameters is generated based on the business node priority list and the anomaly handling action sequence.

[0060] Configure business functions based on the business component configuration interface and the set of business rule configuration parameters, and generate business rule execution instructions and data processing flow;

[0061] According to the business rules, the execution instructions and data processing flow are carried out. The data in the business data collection form is desensitized to obtain local calculation parameters. The local calculation parameters are encrypted to generate an encrypted parameter package and perform multi-party secure calculation to obtain global calculation parameters. Based on sensitivity analysis and adaptive parameter allocation algorithm, the global calculation parameters are allocated to the enterprise's local server to form a business collaborative optimization strategy.

[0062] The business collaboration optimization strategy is applied to the business rule execution instructions, the business execution status is monitored, the execution data is collected, and the set of business rule configuration parameters is updated.

[0063] In one alternative implementation,

[0064] Acquire business process data from upstream and downstream enterprises in the supply chain, extract key business node information, and generate business component configuration interfaces and business data collection forms, including:

[0065] Business process data is obtained by collecting sales order processing records, procurement plan execution records, and warehouse inventory change records from upstream and downstream enterprises in the supply chain and combining them. The business process data is then parsed to identify the data input types, business processing rules, and status change triggering conditions to obtain key business node information.

[0066] Based on the key business node information, a list of component types is constructed. Input specifications, business rules, and status change control attributes are configured for each component type in the business component configuration interface. A business data collection form is generated based on the key business node information. Data input validation procedures, business rule setting procedures, and status change control functions are configured in the business data collection form interface.

[0067] Sales order processing records, procurement plan execution records, and warehouse inventory change records are collected from upstream and downstream enterprises in the supply chain. Data collection methods include data interface integration, file import, or manual entry. For example, sales order data, including order number, customer information, product details, quantity, amount, and delivery date, can be extracted from the seller's ERP system via API; procurement plan data, including plan number, supplier information, material requirements, procurement quantity, and expected delivery date, can be extracted from the buyer's system; and inventory change records, including inbound and outbound order numbers, material codes, changed quantities, storage location information, and operation time, can be extracted from the warehouse management system.

[0068] After data collection, the collected data is combined and processed to form complete business process data. During the combination and processing, data matching is performed based on business association identifiers. For example, a sales order number is associated with a purchase plan, and a purchase order number is associated with an inbound record. For instance, after an electronics manufacturer receives a sales order A001, it generates a purchase plan P001. After the purchase is completed, an inbound record W001 is generated, and an outbound record D001 is generated when the goods are shipped out. These records form a chain of association through document numbers.

[0069] The combined business process data is analyzed in depth to identify data input types, business processing rules, and status change triggering conditions. Data input type analysis identifies field types, such as text, numeric, date, and selection types, by statistically analyzing the value range and data format characteristics of data fields. For example, it identifies product codes as fixed-length strings, prices as two decimal places, and delivery dates as date data. Business processing rule analysis extracts rules by analyzing the logical relationships between fields, such as total amount = unit price × quantity × (1 - discount rate). Status change triggering condition analysis identifies triggering conditions by analyzing changes in other field values ​​before and after a status field change, such as when the received quantity reaches the order quantity, the order status changes from "partial receipt" to "full receipt".

[0070] By parsing the key node information, a component type list is constructed. The component type list includes three main categories: data input components, business rule components, and state control components. Data input components include text input boxes, numeric input boxes, date pickers, dropdown selection boxes, etc.; business rule components include calculation rule components, validation rule components, query rule components, etc.; and state control components include state transition components, process control components, etc.

[0071] In the business component configuration interface, input specifications, business rules, and status change control attributes are configured for each component type. For text input components, attributes such as maximum length, required fields, and format validation regular expressions are configured; for numeric input components, attributes such as minimum value, maximum value, decimal places, and unit of measurement are configured; and for date selection components, attributes such as date range and format are configured. Business rule component configuration includes rule triggering conditions, rule calculation logic, and rule execution priority. Status control component configuration includes status definitions, status transition conditions, and status change permissions.

[0072] For example, for the purchase order component, the purchase number is configured to be required and conform to the format "PO-Year-Month-Serial Number"; the purchase quantity is an integer greater than zero; the purchase amount is calculated as the unit price multiplied by the quantity; the purchase status includes "Draft", "Submitted", "Approved", "Partial Receipt", "Fully Receipt", "Closed" and other statuses, and the transition conditions between each status are defined.

[0073] Business data collection forms are generated based on key business node information. The form generation process includes field layout design, input control selection, and relationship configuration. The field layout adopts a responsive design to adapt to different screen sizes; input controls are intelligently matched according to data type; and relationship configuration enables data transfer and linkage between forms. For example, a purchase order form is generated, containing a basic information area (order number, supplier, date, etc.) and a detailed information area (material, quantity, unit price, etc.), with appropriate field arrangement and grouping methods set.

[0074] Configure a data input validation procedure in the business data acquisition form interface. Data validation includes format validation, range validation, uniqueness validation, and correlation validation. Format validation verifies whether the input data conforms to a predefined format; range validation verifies whether the value is within the allowed range; uniqueness validation verifies the uniqueness of key fields in the system; and correlation validation verifies the validity of related field values. For example, configure a material code validation procedure for purchase orders to verify whether the input code exists in the material master data.

[0075] The configuration program sets up business rules to define calculation relationships, constraints, and business logic between business data. It uses a visual rule editor for configuration and supports operations such as conditional judgment, numerical calculation, and data querying. For example, configuring purchase order approval rules can automatically determine the approval level based on the purchase amount: orders exceeding 100,000 yuan require general manager approval; otherwise, department manager approval is sufficient.

[0076] Configure status change control functions to achieve status management of business processes, including status definition, status transition condition setting, and status change permission control. For example, configure the status flow of a purchase order, which can be manually changed from "draft" to "submitted" with automatic checks for required fields; after approval, it can be changed from "submitted" to "approved"; and after the goods receipt record is generated, it can automatically determine whether it is "partial receipt" or "full receipt" based on the quantity received.

[0077] In this embodiment, by collecting and combining sales order processing records, procurement plan execution records, and warehouse inventory change records from upstream and downstream enterprises in the supply chain, complete business process data can be formed. By parsing the business process data, identifying data input types, business processing rules, and status change triggering conditions, accurate basis is provided for the subsequent construction of business components. Based on key node information, a business data collection form is generated, and data input verification, business rule settings, and status change control functions are configured in the form interface. This effectively improves the accuracy of data input, the compliance of business processing, and the traceability of status flow, realizing the structured management and standardized configuration of supply chain business process data, and improving the automation level and overall operating efficiency of the business execution process.

[0078] In one alternative implementation,

[0079] Based on the business data in the business data collection form, feature encoding is performed on the key business node information to extract business process feature data. Temporal memory operations are then performed on the business process feature data to obtain temporal dependency features, and the importance weight coefficients of business nodes are calculated to obtain a priority list of business nodes, including:

[0080] Based on the business data in the business data collection form, feature encoding is performed on the key business node information to obtain business process feature data. The business process feature data is constructed into a graph structure. The node feature similarity is obtained by calculating the overlap of business processing rules between nodes and the similarity of state change triggering conditions. The temporal dependency strength is obtained based on the temporal sequence of data input and output between nodes and the triggering response time interval of state change. The weighted sum of the node feature similarity and temporal dependency strength is used as the edge weight between nodes.

[0081] Feature extraction is performed on the graph structure based on a pre-set graph convolutional network. Local feature representations for each node are obtained by weighted aggregation of the features of the neighbor node set corresponding to each node. Temporal dependency features are obtained by selectively retaining and updating the local feature representations and historical temporal state information through a long short-term memory algorithm. The local feature representations and temporal dependency features are then weighted and fused to obtain the fused features of the nodes.

[0082] A random perturbation is added to the fusion feature, and the expected difference between the fusion features of other nodes before and after the perturbation is calculated to obtain the direct causal effect between nodes. The importance weight coefficient of the business node is determined based on the direct causal effect. The priority score of the business node is calculated based on the importance weight coefficient of the business node and the pre-set time series influence factor. The business nodes are sorted in descending order according to the priority score to obtain the priority list of business nodes.

[0083] During feature encoding extraction, node type encoding, data attribute encoding, and business rule encoding are extracted for business data such as sales order processing, procurement plan execution, and warehouse inventory changes. Node type encoding uses one-hot encoding to convert different types of nodes, such as sales orders, procurement plans, inbound orders, and outbound orders, into feature vectors. Data attribute encoding vectorizes the data field features contained in the nodes, such as field type, mandatory fields, and length limits. Business rule encoding converts the business processing rules and status change conditions associated with the nodes into feature representations. For example, in an electronic component supply chain, the sales order node contains attributes such as customer information, product details, quantity, and delivery date, while the procurement plan node contains attributes such as supplier information, material requirements, and procurement quantity. Feature encoding converts these attributes into fixed-dimensional feature vectors.

[0084] The business process feature data is constructed as a graph structure, with business nodes as vertices and relationships between nodes as edges. Node feature similarity is obtained by calculating the overlap of business processing rules and the similarity of state change triggering conditions between nodes. Business processing rule overlap is calculated by comparing the ratio of the intersection to the union of the rule sets of two nodes; state change triggering condition similarity is calculated by comparing the field matching degree and logical structure similarity of the triggering conditions. Temporal dependency strength is obtained based on the temporal sequence of data input and output between nodes and the trigger response time interval for state changes. The temporal sequence is calculated by recording the average time difference of state changes between two nodes; the trigger response time interval is calculated by statistically analyzing the temporal correlation of state changes between two nodes in historical data. The weighted sum of node feature similarity and temporal dependency strength is used as the edge weight between nodes, where the weight coefficient can be adjusted according to the business scenario. For example, in the supply chain graph structure of an electronics manufacturing company, the edge weight between the sales order node and the procurement plan node is 0.85, indicating that their business logic is closely related and there is a strong temporal dependency.

[0085] Feature extraction is performed on graph structures based on a pre-defined graph convolutional network. The graph convolutional network contains multiple convolutional layers, each layer transferring and aggregating node features. Within each convolutional layer, a weighted aggregation operation is performed on the features of each node's neighboring node set to obtain the local feature representation of each node. During weighted aggregation, the weights of neighboring nodes are proportional to the edge weights, indicating that nodes with higher business relevance have a greater impact on the current node. The aggregation operation uses average pooling, merging the weighted average of neighboring node features with the central node's features. For example, for an electronics manufacturing company, the local feature representation of a sales order node will incorporate information from related nodes such as procurement plans and production schedules to form a more comprehensive business semantic representation.

[0086] The Long Short-Term Memory (LSTM) algorithm selectively retains and updates local feature representations and historical temporal state information to obtain temporal dependency features. The LSM algorithm consists of an input gate, a forget gate, and an output gate. The input gate controls the proportion of new information input at the current moment; the forget gate controls the proportion of historical information retained; and the output gate controls the proportion of information output. Based on the historical state change sequence of business nodes and the current local feature representation, the parameters of each gating unit are dynamically adjusted to generate temporal dependency features that integrate historical and current information. For example, for a procurement plan node, during the change process from "draft" to "submitted" and then to "approved," the LSM algorithm can capture the temporal dependency patterns of state transitions and retain these patterns in the temporal dependency features.

[0087] The fused features of a node are obtained by weighted fusion of local feature representations and temporal dependency features. During the fusion process, the weights of the local feature representations reflect the importance of the node's spatial structure information, while the weights of the temporal dependency features reflect the importance of the node's temporal evolution information. The weight coefficients are obtained through training and optimization using historical data. Different weight configurations can be adopted for different types of business nodes. For example, for an electronics manufacturing company, the temporal dependency features have a higher weight in the fused features of the warehousing and inventory node because inventory changes have a clear temporal pattern. Conversely, the local feature representations have a higher weight in the fused features of the procurement approval node because approval decisions rely more on the current business state than on historical changes.

[0088] By adding random perturbations to the fused features and calculating the expected difference between the fused features of other nodes before and after the perturbation, the direct causal effect between nodes can be obtained. The random perturbation is achieved by adding normally distributed noise to the fused features, with the perturbation amplitude controlled within 10% of the original feature standard deviation. Through multiple perturbation experiments, the impact of each perturbation on the features of other nodes is statistically analyzed, and the direct causal effect matrix is ​​obtained by calculating the expected difference. For example, after perturbing the sales order node feature, the change in the procurement plan node feature is 0.72, indicating that sales orders have a strong causal influence on procurement plans.

[0089] The importance weight coefficient of business nodes is determined based on direct causal effects. This coefficient comprehensively considers both out-degree and in-degree causal effects. Out-degree causal effects represent the degree of influence a node has on other nodes, while in-degree causal effects represent the degree of influence a node receives from other nodes. The weight coefficient is calculated by weighting the sum of out-degree and in-degree causal effects. A larger weight coefficient indicates greater importance of the node in the business process. For example, for an electronics manufacturing company, the importance weight coefficient for the sales order node is 0.89, for the procurement planning node it is 0.76, and for the inventory management node it is 0.65, indicating that sales orders play a key driving role in the entire supply chain.

[0090] The priority score of a business node is calculated based on its importance weight coefficient and a pre-set time-series influence factor. The time-series influence factor reflects the node's priority on the business process timeline; the earlier the node occurs, the higher its time-series influence factor. The priority score is calculated by multiplying the importance weight coefficient by the time-series influence factor; a higher score indicates a higher node priority. The business nodes are then sorted in descending order based on their priority scores to obtain a priority list. This priority list guides the development sequence and resource allocation of the low-code ERP system. For example, for an electronics manufacturing company, the business node priorities from highest to lowest are: sales order management, customer demand forecasting, procurement planning management, supplier collaboration, production scheduling, quality control, inventory management, and logistics distribution. Development will proceed in this order to ensure that core business processes are implemented first.

[0091] In this embodiment, by extracting features from the business data in the business data collection form and constructing a graph structure in conjunction with key business node information, the complex logical relationships of the business process can be transformed into a graph computation problem. This enables quantitative analysis of the overlap of business rules between nodes, the similarity of state change triggering conditions, and the strength of temporal dependencies. By introducing a graph convolutional network, the features of nodes and their neighboring nodes are weighted and aggregated, effectively extracting the local structural features of business nodes. By weighted fusion of local features and temporal features, the resulting node fusion features can more comprehensively reflect the multi-dimensional information of business nodes. By adding random perturbations and calculating the difference between node features before and after the perturbation to characterize direct causal effects, the actual impact of each business node on the overall process operation can be effectively identified. This achieves accurate evaluation of the value and priority of nodes in complex business processes, thereby improving the intelligence level of process optimization and decision support.

[0092] In one alternative implementation,

[0093] Anomaly detection is performed on temporal dependency features based on the isolation forest algorithm to determine an anomaly type list. An anomaly handling action sequence is generated using a reinforcement learning algorithm. Based on the business node priority list and the anomaly handling action sequence, a set of business rule configuration parameters is generated, including:

[0094] Anomaly detection of temporal dependency features is performed using the isolated forest algorithm. Anomaly scores are calculated based on the ratio of the average path length of the temporal dependency features to the normalization factor. Temporal dependency features with anomaly scores greater than a preset anomaly threshold are identified as anomalous features. Based on cluster analysis, the anomalous features are divided into multiple anomaly types and an anomaly type list is constructed.

[0095] For each exception type in the exception type list, determine the policy parameters that include the action space, state space and reward function, and calculate the initial policy function corresponding to each exception type based on the policy parameters;

[0096] The propagation probability between different anomaly types in the anomaly type list is calculated by using a pre-set correlation evaluation function and a sigmoid activation function. Based on the propagation probability, the optimized policy function is obtained by calculating the KL divergence weighted sum between the pre-set policy expected reward and the initial policy function.

[0097] Obtain the priority list of business nodes, calculate the attention of each anomaly type by applying attention to the current state and hidden state, calculate the fusion weight coefficient of the anomaly type based on the attention, and obtain the fusion strategy function by weighting the optimization strategy function corresponding to each anomaly type with the fusion weight coefficient.

[0098] An exception handling action sequence is generated based on the fusion strategy function, and the exception handling action sequence is mapped to a predefined rule template to obtain a set of business rule configuration parameters.

[0099] Anomaly detection is performed on time-dependent features by constructing an isolated forest model. The isolated forest model consists of 100 isolated trees, with a subsampling size of 256 for each tree, and the split points for each feature dimension are randomly selected. For the input set of time-dependent feature vectors, each feature vector has a dimension of 20 and includes features such as timestamps, numerical indicators, and state changes. A decision tree is constructed by randomly selecting feature dimensions and splitting thresholds, and the average path length of each sample in the tree is calculated. For a feature vector x, its average path length h(x) in the forest is calculated, and the normalization factor c(n) is set to 2(ln(n-1)+0.5772156649)-2(n-1) / n, where n is the number of samples. The anomaly score s(x) is calculated as 2 raised to the power of negative h(x) / c(n). When s(x) is greater than 0.65, the feature vector is marked as an anomalous feature. For example, for the time series [10.2, 11.5, 10.8, 50.7, 10.9], the calculated anomaly scores are [0.32, 0.41, 0.38, 0.89, 0.40], and the fourth value is judged as an anomaly.

[0100] Cluster analysis was performed on the detected anomalous features using the K-means algorithm. The number of clusters k was set to 5, the maximum number of iterations was set to 100, and the initial centroids were selected using the k-means++ method. Similarity was calculated using the Euclidean distance between feature vectors, and the cluster centroids were iteratively updated until convergence or the maximum number of iterations was reached. After clustering, based on the centroid features and feature distribution characteristics of each cluster, the anomaly types were labeled as "numerical mutation type," "trend anomaly type," "periodic anomaly type," "missing value type," and "noise interference type," forming a list of anomaly types.

[0101] For each of the constructed anomaly types, policy parameters are determined. Taking the "numerical mutation" anomaly as an example, the action space includes five operations: "ignore," "smooth processing," "replace with the mean," "mark as invalid," and "trigger alarm." The state space is represented by a 20-dimensional feature vector, containing information such as anomaly score, duration, and impact range. The reward function is designed based on the degree of improvement in system stability after processing, with a score of +10 for normal processing, -5 for error processing, and -2 for delayed processing. The initial policy function adopts a deep neural network structure, containing two hidden layers, each with 64 nodes, using the ReLU activation function, and the output layer uses the softmax function to generate the action probability distribution.

[0102] The propagation probability between anomaly types is calculated using the Pearson correlation coefficient, with values ​​ranging from -1 to 1. The correlation coefficient is then mapped to the interval [0, 1] using the sigmoid activation function as the propagation probability. For example, the correlation coefficient between "numerical mutation" and "trend anomaly" is 0.72, and the propagation probability after processing with the sigmoid function is 0.67. An optimization objective is constructed based on the propagation probability. When optimizing the policy function, the KL divergence weighting coefficient between the expected policy reward and the initial policy function is set to 0.01. The policy gradient method is iterated 500 times with a learning rate of 0.001 to obtain the optimized policy function.

[0103] When obtaining the priority list of business nodes, they are sorted according to their criticality. For example, the "Payment Processing" node has a priority of 5, the "Order Creation" node has a priority of 4, and the "Inventory Check" node has a priority of 3. When calculating the attention score for anomaly types, a multi-head attention mechanism is used with 8 heads and a hidden layer dimension of 256. The current state is represented by a 20-dimensional vector, with a hidden state dimension of 256. Attention calculations yield an attention score of 0.38 for "numerical mutation" anomalies, 0.25 for "trend anomalies," and 0.18, 0.12, and 0.07 for other types, respectively. After attention score normalization, a fusion weight coefficient is obtained, which is then weighted with the optimization strategy function to obtain the fusion strategy function.

[0104] Based on the fusion strategy function, an anomaly handling action sequence is generated by sampling the input state vector. For example, for a "numerical mutation" anomaly detected on the "payment processing" node, the generated action sequence is ["mark as invalid", "trigger alarm", "replace with average"]. This action sequence is mapped to a predefined rule template, with the rule template format being "IF condition THEN action". For the above action sequence, the generated business rule configuration parameter set is: {condition: "payment amount is abnormal and anomaly score > 0.8", action: "mark transaction as invalid; send system alarm; replace with 30-minute average", priority: 5, timeout: 60 seconds, number of retries: 2}.

[0105] In this embodiment, by utilizing the isolated forest algorithm to detect anomalies in time-dependent features, potential anomalies can be automatically identified in complex business processes. Cluster analysis is used to classify these anomalies and generate a list of anomaly types, facilitating systematic management and differentiated handling of different anomalies. Configuring action spaces, state spaces, and reward functions for each anomaly type allows for the formation of targeted initial policy functions, providing a customized policy foundation for subsequent anomaly handling. By introducing a correlation evaluation function and a sigmoid activation function to calculate the propagation probability between anomaly types, and combining this with a KL divergence-weighted optimization policy function, the mutual influence relationships between anomalies can be effectively reflected, improving the rationality and stability of policy adjustments. Based on the fusion policy function, anomaly handling action sequences are generated and mapped to predefined rule templates, automatically generating a set of business rule configuration parameters. This enables intelligent detection, classification, and optimization decision-making for anomalies in complex business processes, improving the accuracy, flexibility, and automation level of anomaly handling, and effectively enhancing the robustness and continuity of business processes.

[0106] In one alternative implementation,

[0107] Based on the business component configuration interface and the set of business rule configuration parameters, configure business functions, and generate business rule execution instructions and data processing flows, including:

[0108] The configuration parameters of the business component configuration interface are collected. The configuration parameters include component identifier, component parameters and connection relationships between components.

[0109] The configuration parameters are matched and mapped with the set of business rule configuration parameters. Component instantiation configuration and component calling order are generated based on preset mapping rules. Data processing flow is constructed based on the component instantiation configuration. Business rule execution instructions are generated according to the component calling order.

[0110] The configuration parameters collected from the business component configuration interface include component identifiers, component parameters, and inter-component connections. Each business component is assigned a unique identifier, such as "PO_COMPONENT_001" for the purchase order component, "SO_COMPONENT_002" for the sales order component, and "INV_COMPONENT_003" for the inventory management component. Component parameters describe the component's attribute settings, including data field definitions, display attributes, validation rules, and calculation logic. For example, the purchase order component's parameters include fields such as "Order Number" (required, string type, maximum length 20), "Supplier" (required, foreign key reference), "Purchase Date" (required, date type), and "Total Amount" (read-only, numeric type, precision to 2 decimal places). Inter-component connections define the interaction methods between components, including data flow, triggering conditions, and response actions. For example, the connection between the purchase order component and the inventory management component is defined as "When the purchase order status changes to 'Received,' the inventory management component is triggered to perform an 'Inbound' operation."

[0111] Configuration parameters can be collected in various ways, including user interface interaction recording, configuration file import, and API calls. In the user interface interaction recording method, the sequence of user operations—dragging and dropping components, setting properties, and establishing connections—is recorded on the low-code platform. In the configuration file import method, JSON or XML configuration files are parsed to extract parameter information. In the API call method, structured configuration data is received from external systems. The collected configuration parameters are stored in a unified data structure, forming a configuration parameter set. For example, in the supply chain management scenario of an electronic component manufacturing company, the collected configuration parameters may include configuration information for more than 10 business components such as sales order processing, procurement plan management, supplier collaboration, and inventory management. Each component contains an average of 15-20 parameter items, and approximately 25 connections are established between components, forming a complete business process network.

[0112] The configuration parameters are matched and mapped to a set of business rule configuration parameters. This set originates from the rules generated by the aforementioned exception handling and includes categories such as exception detection rules, early warning rules, and automatic processing rules. The matching and mapping process is based on parameter semantic similarity calculation. A vector space model is used to calculate the cosine similarity between the configuration parameters and rule parameters. A match is considered successful if the similarity exceeds a preset threshold (usually set to 0.75). For successfully matched parameter pairs, parameter values ​​are transformed and supplemented according to the mapping table. For example, the exception handling rule "Automatically trigger an early warning if supplier delivery delay exceeds 3 days" is matched with the "Arrival Reminder" function parameter of the purchase order component. The "3 days" in the rule is converted to the "reminderDays: 3" configuration in the component parameters. Parameters that fail to match automatically are recorded in the exception mapping list for manual processing.

[0113] The component instantiation configuration and component call order are generated based on preset mapping rules. These rules define the conversion methods between configuration parameters and instantiation code templates, including field mapping rules, attribute conversion rules, and logical expression conversion rules. Field mapping rules define the correspondence between data fields and database table columns; attribute conversion rules define the conversion methods between UI attributes and front-end code attributes; and logical expression conversion rules define the generation methods for business logic and execution code. Component instantiation configurations are described in JSON format, including component type, instance identifier, attribute settings, event handling, and other information. For the purchase order component, the generated instantiation configuration includes form layout definitions, field validation rules, calculation formulas, and state transition logic. The component call order is determined based on business process analysis, considering data dependencies between components and the sequential relationships of business logic, and generating a directed acyclic call sequence using a topological sorting algorithm. In the supply chain business process, a typical call order is: "Sales order creation → Purchase requirement generation → Purchase order creation → Supplier confirmation → Material arrival → Quality inspection warehousing → Production material requisition → Product warehousing → Sales outbound → Logistics distribution".

[0114] The data processing workflow is built based on component instantiation configuration and includes five stages: data input, data validation, business calculation, data storage, and data transmission. In the data input stage, a data collection form is generated according to the field definitions in the component instantiation configuration, supporting manual input, batch import, and automatic collection. In the data validation stage, data format checks, range validation, and correlation validation are performed according to field validation rules. In the business calculation stage, data transformation, summary statistics, and logical judgments are performed according to the configured calculation formulas. In the data storage stage, the processed data is persisted to the database according to a predefined data model. In the data transmission stage, the processing results are transmitted to downstream components based on the connection relationships between components. For example, in the purchase order processing workflow of the electronic component supply chain, the data processing workflow includes steps such as receiving purchase requirement data, validating material codes, calculating purchase amounts, saving purchase order records, and notifying suppliers for confirmation. The constructed data processing workflow is executed through a workflow engine, supporting parallel processing, conditional branching, and exception handling mechanisms.

[0115] Business rule execution instructions are generated based on the component call order. These instructions are described using a rule language recognizable by the rule engine, including conditional expressions, action instructions, and execution priorities. Conditional expressions define the rule trigger conditions, such as "Purchase order status = 'Approved' AND Supplier confirmation = 'Yes'". Action instructions define the operations performed after a rule is triggered, such as "Set estimated delivery date = current date + supplier's average delivery days". Execution priorities define the processing order when multiple rules conflict. The rule execution instructions are organized according to an event-driven model, triggering corresponding rules in response to business events. For example, in the electronic component supply chain case, the generated rule execution instructions include automatic purchase approval rules (automatic approval of purchase orders less than 10,000), inventory warning rules (triggering purchase recommendations when inventory is below safety stock), and delivery risk warning rules (triggering risk warnings when the supplier confirms a delivery date later than the customer's required date). The business rule execution instructions are parsed and executed by the rule engine, supporting hot rule updates, allowing adjustments to business logic without code modification.

[0116] In this embodiment, by collecting the configuration parameters from the business component configuration interface and matching and mapping them with the business rule configuration parameter set, the automatic association between component configuration and business rules can be achieved. This avoids the inefficiency and error-proneness caused by manual matching. Based on preset mapping rules, component instantiation configuration and component calling order are generated, enabling business components to automatically complete combination and calling according to predetermined logic. This ensures the integrity and consistency of the data processing flow, improves the automation level of component configuration and rule execution, reduces manual intervention, and enhances the accuracy, standardization, and operational efficiency of business process execution.

[0117] In one alternative implementation,

[0118] Based on business rules, instructions and data processing flows are executed. Data in the business data collection form is anonymized to obtain local calculation parameters. These local calculation parameters are then encrypted to generate an encrypted parameter package, which is then subjected to multi-party secure computation to obtain global calculation parameters, including:

[0119] According to the business rules, the execution instructions extract enterprise operating data and transaction amount data from the business data collection form as sensitive fields, calculate the sensitivity and privacy budget parameters corresponding to the sensitive fields, and classify the sensitive fields into categorical data and numerical data according to the data format;

[0120] The privacy budget parameter and the number of categories in the categorical data are used as input parameters of a pre-set random response function to obtain the first de-identified data. Laplace distribution noise is calculated based on sensitivity and privacy budget parameter. Laplace distribution noise is added to the numerical data to obtain the second de-identified data. The first de-identified data and the second de-identified data are combined to obtain the local calculation parameter.

[0121] Obtain the public and private keys generated based on additive homomorphic encryption, encrypt the local computation parameters based on the public key to obtain the encrypted parameter packet, generate a parameter confusion matrix based on the encrypted parameter packet, fill the encrypted parameter packet in the diagonal positions of the parameter confusion matrix, and fill the off-diagonal positions with the encrypted result of random confusion values;

[0122] The encrypted parameter packets in the parameter confusion matrix are fragmented and sent to multiple computing nodes. A secure summation operation is performed to obtain the encrypted parameter share. The multi-party computation result is obtained by decryption using the private key. Based on the multi-party computation result, the sample mean and variance parameters and Euclidean distance are calculated to determine the parameter aggregation weight. The multi-party computation result is weighted and summed based on the parameter aggregation weight to obtain the global computation parameter.

[0123] According to business rules, the system extracts enterprise operating data and transaction amount data from the business data collection form as sensitive fields. Sensitive field identification employs a combination of rule matching and semantic analysis. Rule matching is based on predefined sensitive data identifiers, such as keywords like "customer ID," "supplier code," "transaction amount," and "profit margin." Natural language processing technology analyzes field descriptions and contextual relationships to assess the sensitivity of each field. For example, in the electronic component supply chain, typical sensitive fields include supplier price lists, customer discount rates, inventory costs, and gross profit margins. Sensitivity and privacy budget parameters are calculated for the identified sensitive fields. Sensitivity calculation considers two dimensions: the scope of data impact and data importance. The scope of impact represents the number of business entities potentially affected by a data breach, while data importance represents the significance of the data to the enterprise's operations. Sensitivity is calculated as a weighted average of the scope of impact and importance, ranging from 0 to 1, with values ​​closer to 1 indicating higher sensitivity. For example, the sensitivity of the supplier price list is 0.85, and the sensitivity of the customer discount rate is 0.78. The privacy budget parameter controls the strength of privacy protection; a smaller value indicates stricter protection but lower data availability. The privacy budget parameter is determined based on business needs and data sensitivity, typically ranging from 0.1 to 1.0. For example, the privacy budget parameter for a supplier price list might be set to 0.3, and the customer discount rate to 0.4.

[0124] Sensitive fields are categorized into categorical and numerical data based on their data format. Categorical data includes discrete data with limited options, such as supplier level, product category, and quality inspection results; numerical data includes continuous data that can be arithmetically calculated, such as transaction amount, quantity, and price. The data type classification is based on the field's data type definition and value characteristics, using heuristic rules for judgment. For categorical data, the privacy budget parameter and the number of categories are used as input parameters to a pre-set random response function to obtain the first desensitized data. The random response function is implemented based on probability flipping, keeping the original data unchanged with a certain probability and randomly replacing it with other category values ​​with a certain probability. Specifically, a retention probability p and a replacement probability q are set, where p is positively correlated with the privacy budget parameter, and q is related to the number of categories and the privacy budget parameter. For example, in the electronic component supply chain, when processing supplier level data (divided into four levels: A, B, C, and D), with a privacy budget parameter of 0.4, the retention probability p is 0.7, and the replacement probability q for each other category is 0.1. After random response processing, suppliers whose original rating is A have a 70% probability of still being displayed as A, and a 30% probability of being randomly displayed as one of B, C, or D.

[0125] For numerical data, Laplace distribution noise is calculated based on sensitivity and privacy budget parameters. Adding Laplace distribution noise to the numerical data yields the second desensitized data. The scale parameter of the Laplace noise is directly proportional to the sensitivity and inversely proportional to the privacy budget parameter, reflecting a balance between protection strength and data sensitivity. During noise generation, uniformly distributed random numbers are first generated and then mapped to a Laplace distribution using a transformation function. For example, in the electronic component supply chain, noise is added to transaction amount data (sensitivity 0.85, privacy budget parameter 0.3). A transaction record with an original amount of 10,000 yuan may become random values ​​such as 10,247 yuan or 9,856 yuan after adding noise, maintaining the data distribution characteristics but masking the precise value. Combining the first and second desensitized data yields local calculation parameters. These local calculation parameters retain the statistical characteristics of the original data but mask individual precise information, and can be used for subsequent secure calculations.

[0126] The process involves obtaining public and private keys generated using additive homomorphic encryption. The Paillier encryption system is used, which exhibits additive homomorphism, meaning that performing operations on ciphertext is equivalent to performing corresponding operations on plaintext before encryption. The key generation process includes selecting two large prime numbers to generate a public and private key pair. The public key is used for encryption, and the private key is used for decryption. In multi-party collaboration scenarios within a supply chain, keys are typically generated by a trusted third party or consortium blockchain network and securely distributed to participants. Local computation parameters are encrypted using the public key to obtain an encrypted parameter package. The encryption process maps local computation parameters to values ​​in the ciphertext space, ensuring data security during transmission and computation. For local computation parameters containing multiple fields, each field is encrypted to form a structured encrypted parameter package. For example, in an electronic component supply chain, a manufacturer's anonymized procurement data is encrypted using Paillier and converted into a fixed-length ciphertext string; the original data cannot be directly obtained from the ciphertext.

[0127] A parameter obfuscation matrix is ​​generated based on the encrypted parameter packets. The parameter obfuscation matrix is ​​a two-dimensional matrix structure used to further enhance data privacy protection. The matrix dimension is determined by the number of nodes participating in the computation; for N nodes, an N×N matrix is ​​generated. Encrypted parameter packets are filled into the diagonal positions of the parameter obfuscation matrix, while the off-diagonal positions are filled with the encrypted results of random obfuscated values. The random obfuscated values ​​are generated using a random number generator and undergo the same encryption process as the original parameters to obtain the encrypted results. For example, in a collaborative supply chain scenario in an electronics manufacturing industry, three companies participate in the computation, generating a 3×3 parameter obfuscation matrix. Each company's encrypted parameters are located on the diagonal of the matrix, and the remaining positions are filled with encrypted random data.

[0128] The encrypted parameter packets in the parameter obfuscation matrix are fragmented and sent to multiple computing nodes. The fragmentation strategy is based on row or column partitioning, with each computing node receiving one row or one column of data from the matrix. In row-based fragmentation, the i-th node receives the i-th row of data; in column-based fragmentation, the i-th node receives the i-th column of data. Fragmented transmission employs a secure communication protocol to ensure the confidentiality and integrity of data transmission. For example, in supply chain collaborative computing, a row-based fragmentation strategy is used, with each enterprise receiving a single row of data containing its own encrypted parameters and other random obfuscated values. Each computing node performs a secure summation operation to obtain its share of encrypted parameters. The secure summation operation utilizes homomorphic encryption to directly add the received encrypted data without decrypting intermediate results. The calculation result remains encrypted, maintaining data privacy. After completing the calculation, each node returns its encrypted parameter share to the coordinating node. The multi-party calculation result is obtained by decryption using a private key. The coordinating node collects all encrypted parameter shares, decrypts them using its private key, and obtains the aggregated result of the data from all participants.

[0129] The parameter aggregation weights are determined by calculating the sample mean, variance, and Euclidean distance based on the multi-party computation results. The sample mean represents the average level of the data from each participant, while the variance reflects the dispersion of the data distribution. The Euclidean distance measures the deviation of each participant's data from the overall mean and is used to assess data quality and representativeness. The parameter aggregation weights are calculated by normalizing the inverse of the Euclidean distance; the smaller the distance, the greater the weight, reflecting the greater contribution of high-quality data to the results. For example, in the electronic component supply chain, the aggregation weights calculated from the purchase price data of three companies are 0.45, 0.35, and 0.2, respectively, reflecting their different influences in the global computation. The global computation parameters are obtained by weighted summation of the multi-party computation results based on the parameter aggregation weights. The global computation parameters integrate the data characteristics of each participant without disclosing any participant's original sensitive data. They can be used for supply chain collaborative decision-making and analysis, and can be used to construct pricing strategies that consider multi-party data, improving overall forecast accuracy while protecting the commercial secrets of all parties.

[0130] In this embodiment, sensitive fields such as enterprise operating data and transaction amount data are extracted from the business data collection form and classified in combination with sensitivity and privacy budget parameters. This enables differentiated privacy protection while ensuring data type characteristics. By constructing a parameter obfuscation matrix, encrypted parameter packets are reasonably mixed with random obfuscation values, improving anti-attack capabilities and avoiding the risk of single-point leakage. By performing mean, variance, and Euclidean distance analysis on the multi-party calculation results, the parameter aggregation weights are reasonably determined and the weighted aggregation of global calculation parameters is completed. This achieves an organic combination of data privacy protection and cross-node collaborative computing. While ensuring the security of sensitive data, it enables efficient and secure computing with multi-party participation, improving the reliability of global calculation results and the level of privacy protection.

[0131] Figure 2 This is a flowchart illustrating the privacy-preserving multi-party secure computation process of a low-code ERP construction method for connecting upstream and downstream businesses in the supply chain, as described in an embodiment of the present invention.

[0132] In one alternative implementation,

[0133] Based on sensitivity analysis and adaptive parameter allocation algorithms, global computational parameters are distributed to local enterprise servers to form a business collaborative optimization strategy, including:

[0134] Business data samples are collected from the enterprise's local server. Feature extraction is performed on the business data samples to obtain sample feature vectors and the corresponding gradient sensitivity is calculated. A sensitivity matrix is ​​constructed based on the gradient sensitivity.

[0135] The average sensitivity is obtained by summing the absolute values ​​of each element in the sensitivity matrix. The fluctuation level is obtained by calculating the square root of the sum of squares of each element in the sensitivity matrix. The average sensitivity and the fluctuation level are weighted to obtain the parameter importance index. The global calculated parameters are sorted in descending order according to the parameter importance index and grouped into multiple parameter groups by combining the preset importance threshold. The mean vector of the parameter vector in each parameter group is calculated as the feature center of the current parameter group.

[0136] The system obtains the performance metrics of the enterprise's local server and constructs a performance feature vector. It calculates the cosine similarity between the performance feature vector and each parameter group, as well as the Euclidean distance between the performance feature vector and the feature center. The matching coefficient is obtained by summing the cosine similarity and the Euclidean distance after exponential decay.

[0137] Based on the matching degree coefficient, a softmax operation is performed to obtain the parameter allocation probability. Based on the parameter allocation probability, the globally calculated parameters are allocated to the enterprise's local server to obtain a business collaboration optimization strategy.

[0138] Business data samples are collected from the enterprise's local server. The collection process is based on a preset data interface specification, using API calls to obtain business data from the enterprise's ERP system. The collected data samples include various business types such as order processing records, inventory change data, procurement execution status, and sales fulfillment data. Data collection supports incremental methods, acquiring only data added or changed since the last collection, reducing data transmission volume. For a certain electronic component manufacturing company, the collected business data samples include 5,000 order records, 12,000 inventory change records, and 3,500 procurement execution records from the past three months, with a total data volume of approximately 500MB. Feature extraction is performed on the business data samples to obtain sample feature vectors. Feature extraction is achieved using a combination of methods, including statistical feature extraction, time-series feature extraction, and correlation feature extraction. Statistical feature extraction calculates basic statistics of the data, such as mean, standard deviation, and quantiles; time-series feature extraction analyzes the time-varying patterns of the data, including trends, periodicity, and volatility; correlation feature extraction analyzes the correlation and dependency relationships between data. The procurement data of electronic component manufacturers extracted features including approximately 50 dimensions such as monthly procurement volume, supplier delivery cycle, price volatility, and quality pass rate.

[0139] The gradient sensitivity corresponding to the sample feature vector is calculated, representing the degree of impact of feature value changes on business results. The calculation process uses the finite difference method, applying small perturbations to each feature and observing the degree of change in business indicators. Sensitivity calculation considers multiple business indicators, including order completion rate, inventory turnover rate, and on-time delivery rate. For example, for an electronic component manufacturing company, the sensitivity of the supplier delivery cycle feature is 0.85, indicating that changes in this feature have a significant impact on business results; while the sensitivity of the packaging type feature is only 0.12, indicating a smaller impact. A sensitivity matrix is ​​constructed based on the gradient sensitivity, where rows represent feature dimensions, columns represent business indicators, and matrix element values ​​represent the sensitivity of a specific feature to a specific indicator. For the electronic component manufacturing company, the constructed sensitivity matrix has a dimension of 50×8, reflecting the degree of impact of 50 features on 8 key business indicators.

[0140] The average sensitivity is obtained by summing the absolute values ​​of each element in the sensitivity matrix. The summation process first takes the absolute value of each element to eliminate the offsetting effect of positive and negative influences, then sums the values ​​row-wise and divides by the number of columns to obtain the average sensitivity of each feature, representing the overall impact of the feature on all business indicators. The square root of the sum of squares of each element in the sensitivity matrix is ​​calculated to obtain the volatility, which reflects the dispersion of the feature's influence; a larger value indicates a greater difference in the feature's impact on different business indicators. The average sensitivity and volatility are weighted to obtain the parameter importance index. Adaptive weights are used, dynamically adjusted according to the characteristics of the business scenario. Generally, the average sensitivity weight is 0.7, and the volatility weight is 0.3. For example, for an electronic component manufacturing company, the supplier delivery cycle feature has a parameter importance index of 0.78, ranking first; the price volatility feature has an importance index of 0.72, ranking second.

[0141] The globally calculated parameters are sorted in descending order based on their importance index, and then grouped into multiple parameter groups based on a preset importance threshold. The sorting process is directly based on the parameter importance index values, arranged from largest to smallest. The grouping process uses an interval sampling method, with preset importance thresholds typically set at 3-5 levels, such as 0.7, 0.5, 0.3, and 0.1, dividing the parameters into high importance, medium-high importance, medium-low importance, and low importance groups. For the globally calculated parameters of electronic component manufacturing companies, the high importance group contains 12 parameters, the medium-high importance group contains 18 parameters, the medium-low importance group contains 15 parameters, and the low importance group contains 5 parameters. The mean vector of the parameter vectors within each parameter group is calculated as the feature center of the current parameter group. The mean vector calculation uses an arithmetic mean method, averaging each dimension within the parameter group. The feature center represents the typical features of the parameter group and is used for subsequent matching calculations.

[0142] The performance metrics of the enterprise's local server are acquired and a performance feature vector is constructed. Performance metrics are collected through system monitoring tools, including hardware metrics such as CPU utilization, memory utilization, disk I / O rate, and network bandwidth utilization, as well as software performance metrics such as concurrent connections, response time, and throughput. The performance feature vector is constructed through standardization, converting metrics with different dimensions into a unified scale. For example, for the server of an electronic component manufacturing company, the collected performance metrics show an average CPU utilization of 45%, memory utilization of 60%, disk I / O rate of 120MB / s, and network bandwidth utilization of 35%, resulting in a performance feature vector of [0.45, 0.6, 0.6, 0.35]. The cosine similarity between the performance feature vector and each parameter group, as well as the Euclidean distance between the performance feature vector and the feature center, are calculated. Cosine similarity is calculated by taking the cosine of the angle between two vectors, reflecting the degree of directional similarity. Euclidean distance is calculated by taking the straight-line distance between two points in Euclidean space, reflecting the degree of absolute difference. For example, for an electronic component manufacturing company, the cosine similarity between the server performance feature vector and the high-importance parameter group is 0.82, and the Euclidean distance between the feature vector and the feature center is 0.35.

[0143] The matching coefficient is obtained by summing the cosine similarity and Euclidean distance after exponential decay. Exponential decay converts cosine similarity into a similarity gain and Euclidean distance into a distance loss; the sum of these two yields the overall matching degree. Specifically, the similarity gain equals the cosine similarity multiplied by a positive weighting factor (usually 1.0); the distance loss equals the negative weighting factor (usually -0.5) multiplied by the exponential function value of the Euclidean distance. For example, for an electronic component manufacturing company, the matching coefficient for its server is 0.65 with the high-importance parameter group, 0.58 with the medium-high importance group, 0.42 with the medium-low importance group, and 0.35 with the low importance group. A softmax operation is then performed based on the matching coefficient to obtain the parameter assignment probabilities. The softmax operation converts the matching coefficient into a probability distribution, ensuring that the sum of the assignment probabilities for each parameter group is 1. The calculation results reflect the probability that each parameter group is suitable for execution on the current server. For example, for an electronic component manufacturing company, the allocation probability of the high importance parameter group is 0.4, the medium-high importance group is 0.3, the medium-low importance group is 0.2, and the low importance group is 0.1.

[0144] Based on parameter allocation probabilities, global computational parameters are distributed to local enterprise servers to obtain a business collaboration optimization strategy. Parameter allocation employs a probabilistic proportional allocation method, determining the deployment ratio of each parameter group on the current server according to the allocation probability. Simultaneously, the dependencies between parameters are considered, ensuring that interdependent parameters are allocated to the same server. For example, the server is allocated 10 out of 12 parameters in the high-importance parameter group, 12 out of 18 parameters in the medium-high importance group, 6 out of 15 parameters in the medium-low importance group, and 2 out of 5 parameters in the low importance group. The final business collaboration optimization strategy includes a parameter deployment plan, a computational resource allocation strategy, and business process optimization suggestions. The deployment plan specifies the deployment location and operating environment for each parameter; the resource allocation strategy specifies the computational resource quota available to each parameter group; and the business process optimization suggestions provide directions for process improvement based on parameter analysis.

[0145] In this embodiment, global calculation parameters are sorted and grouped, and feature centers of each group are calculated, making parameter management more hierarchical and structured. A comprehensive matching coefficient is obtained through an exponential decay mechanism, which can accurately reflect the degree of matching between server performance and parameter group requirements. The matching coefficient is converted into parameter allocation probability, realizing adaptive allocation of global calculation parameters in a multi-server environment. While ensuring that parameters of importance are used first, the performance differences of local enterprise servers are fully considered, thereby improving the rationality of parameter allocation and the efficiency of computing resource utilization, achieving collaborative optimization of business processes and improvement of overall operating performance.

[0146] A second aspect of the present invention provides a low-code ERP building system for connecting upstream and downstream businesses in a supply chain, comprising:

[0147] The first unit is used to acquire business process data of upstream and downstream enterprises in the supply chain, extract key business node information, and generate business component configuration interfaces and business data collection forms.

[0148] The second unit is used to extract business process feature data by feature encoding of key business node information based on business data in the business data collection form, perform temporal memory operation on the business process feature data to obtain temporal dependency features and calculate the importance weight coefficient of business nodes to obtain a business node priority list, perform anomaly detection on the temporal dependency features based on the isolated forest algorithm, determine the anomaly type list and generate anomaly handling action sequence through reinforcement learning algorithm, and generate a set of business rule configuration parameters based on the business node priority list and anomaly handling action sequence.

[0149] The third unit is used to configure business functions based on the business component configuration interface and the set of business rule configuration parameters, and to generate business rule execution instructions and data processing flow.

[0150] The fourth unit is used to execute instructions and data processing flow according to business rules, perform desensitization processing on the data in the business data collection form to obtain local calculation parameters, encrypt the local calculation parameters, generate encrypted parameter packages and perform multi-party secure calculations to obtain global calculation parameters, and allocate the global calculation parameters to the enterprise local server based on sensitivity analysis and adaptive parameter allocation algorithm to form a business collaborative optimization strategy.

[0151] The fifth unit is used to apply business collaboration optimization strategies to business rule execution instructions, monitor business execution status, collect execution data, and update the set of business rule configuration parameters.

[0152] A third aspect of the present invention provides an electronic device, comprising:

[0153] A processor and a memory for storing processor-executable instructions, wherein the processor is configured to invoke instructions stored in the memory to perform the aforementioned method.

[0154] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0155] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0156] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A low-code ERP construction method for connecting upstream and downstream businesses in the supply chain, characterized in that, include: The system acquires business process data from upstream and downstream enterprises in the supply chain, extracts key business node information, and generates a business component configuration interface and a business data collection form. The business process data includes sales order data, procurement plan data, and inventory change records. The sales order data includes order number, customer information, product details, quantity, amount, and delivery date. The procurement plan data includes plan number, supplier information, material requirements, procurement quantity, and expected delivery date. The inventory change records include inbound order number, outbound order number, material code, changed quantity, storage location information, and operation time. Based on the business data in the business data collection form, feature encoding is performed on the key business node information to extract business process feature data. Temporal memory operation is performed on the business process feature data to obtain temporal dependency features and calculate the importance weight coefficient of business nodes to obtain a business node priority list. Anomaly detection is performed on the temporal dependency features based on the isolated forest algorithm to determine the anomaly type list and generate an anomaly handling action sequence through reinforcement learning algorithm. A set of business rule configuration parameters is generated based on the business node priority list and the anomaly handling action sequence. Configure business functions based on the business component configuration interface and the set of business rule configuration parameters, and generate business rule execution instructions and data processing flow; According to the business rules, the execution instructions and data processing flow are carried out. The data in the business data collection form is desensitized to obtain local calculation parameters. The local calculation parameters are encrypted to generate an encrypted parameter package and perform multi-party secure calculation to obtain global calculation parameters. Based on sensitivity analysis and adaptive parameter allocation algorithm, the global calculation parameters are allocated to the enterprise's local server to form a business collaborative optimization strategy. The business collaboration optimization strategy is applied to the business rule execution instructions, the business execution status is monitored, the execution data is collected, and the set of business rule configuration parameters is updated.

2. The method according to claim 1, characterized in that, Acquire business process data from upstream and downstream enterprises in the supply chain, extract key business node information, and generate business component configuration interfaces and business data collection forms, including: Business process data is obtained by collecting sales order processing records, procurement plan execution records, and warehouse inventory change records from upstream and downstream enterprises in the supply chain and combining them. The business process data is then parsed to identify the data input types, business processing rules, and status change triggering conditions to obtain key business node information. Based on the key business node information, a list of component types is constructed. Input specifications, business rules, and status change control attributes are configured for each component type in the business component configuration interface. A business data collection form is generated based on the key business node information. Data input validation procedures, business rule setting procedures, and status change control functions are configured in the business data collection form interface.

3. The method according to claim 1, characterized in that, Based on the business data in the business data collection form, feature encoding is performed on the key business node information to extract business process feature data. Temporal memory operations are then performed on the business process feature data to obtain temporal dependency features, and the importance weight coefficients of business nodes are calculated to obtain a priority list of business nodes, including: Based on the business data in the business data collection form, feature encoding is performed on the key business node information to obtain business process feature data. The business process feature data is constructed into a graph structure. The node feature similarity is obtained by calculating the overlap of business processing rules between nodes and the similarity of state change triggering conditions. The temporal dependency strength is obtained based on the temporal sequence of data input and output between nodes and the triggering response time interval of state change. The weighted sum of the node feature similarity and temporal dependency strength is used as the edge weight between nodes. Feature extraction is performed on the graph structure based on a pre-set graph convolutional network. Local feature representations for each node are obtained by weighted aggregation of the features of the neighbor node set corresponding to each node. Temporal dependency features are obtained by selectively retaining and updating the local feature representations and historical temporal state information through a long short-term memory algorithm. The local feature representations and temporal dependency features are then weighted and fused to obtain the fused features of the nodes. A random perturbation is added to the fusion feature, and the expected difference between the fusion features of other nodes before and after the perturbation is calculated to obtain the direct causal effect between nodes. The importance weight coefficient of the business node is determined based on the direct causal effect. The priority score of the business node is calculated based on the importance weight coefficient of the business node and the pre-set time series influence factor. The business nodes are sorted in descending order according to the priority score to obtain the priority list of business nodes.

4. The method according to claim 1, characterized in that, Anomaly detection is performed on temporal dependency features based on the isolation forest algorithm to determine an anomaly type list. An anomaly handling action sequence is generated using a reinforcement learning algorithm. Based on the business node priority list and the anomaly handling action sequence, a set of business rule configuration parameters is generated, including: Anomaly detection of temporal dependency features is performed using the isolated forest algorithm. Anomaly scores are calculated based on the ratio of the average path length of the temporal dependency features to the normalization factor. Temporal dependency features with anomaly scores greater than a preset anomaly threshold are identified as anomalous features. Based on cluster analysis, the anomalous features are divided into multiple anomaly types and an anomaly type list is constructed. For each exception type in the exception type list, determine the policy parameters that include the action space, state space and reward function, and calculate the initial policy function corresponding to each exception type based on the policy parameters; The propagation probability between different anomaly types in the anomaly type list is calculated by using a pre-set correlation evaluation function and a sigmoid activation function. Based on the propagation probability, the optimized policy function is obtained by calculating the KL divergence weighted sum between the pre-set policy expected reward and the initial policy function. Obtain the priority list of business nodes, calculate the attention of each anomaly type by applying attention to the current state and hidden state, calculate the fusion weight coefficient of the anomaly type based on the attention, and obtain the fusion strategy function by weighting the optimization strategy function corresponding to each anomaly type with the fusion weight coefficient. An exception handling action sequence is generated based on the fusion strategy function, and the exception handling action sequence is mapped to a predefined rule template to obtain a set of business rule configuration parameters.

5. The method according to claim 1, characterized in that, Based on the business component configuration interface and the set of business rule configuration parameters, configure business functions, and generate business rule execution instructions and data processing flows, including: The configuration parameters of the business component configuration interface are collected. The configuration parameters include component identifier, component parameters and connection relationships between components. The configuration parameters are matched and mapped with the set of business rule configuration parameters. Component instantiation configuration and component calling order are generated based on preset mapping rules. Data processing flow is constructed based on the component instantiation configuration. Business rule execution instructions are generated according to the component calling order.

6. The method according to claim 1, characterized in that, Based on business rules, instructions and data processing flows are executed. Data in the business data collection form is anonymized to obtain local calculation parameters. These local calculation parameters are then encrypted to generate an encrypted parameter package, which is then subjected to multi-party secure computation to obtain global calculation parameters, including: According to the business rules, the execution instructions extract enterprise operating data and transaction amount data from the business data collection form as sensitive fields, calculate the sensitivity and privacy budget parameters corresponding to the sensitive fields, and classify the sensitive fields into categorical data and numerical data according to the data format; The privacy budget parameter and the number of categories in the categorical data are used as input parameters of a pre-set random response function to obtain the first de-identified data. Laplace distribution noise is calculated based on sensitivity and privacy budget parameter. Laplace distribution noise is added to the numerical data to obtain the second de-identified data. The first de-identified data and the second de-identified data are combined to obtain the local calculation parameter. Obtain the public and private keys generated based on additive homomorphic encryption, encrypt the local computation parameters based on the public key to obtain the encrypted parameter packet, generate a parameter confusion matrix based on the encrypted parameter packet, fill the encrypted parameter packet in the diagonal positions of the parameter confusion matrix, and fill the off-diagonal positions with the encrypted result of random confusion values; The encrypted parameter packets in the parameter confusion matrix are fragmented and sent to multiple computing nodes. A secure summation operation is performed to obtain the encrypted parameter share. The multi-party computation result is obtained by decryption using the private key. Based on the multi-party computation result, the sample mean and variance parameters and Euclidean distance are calculated to determine the parameter aggregation weight. The multi-party computation result is weighted and summed based on the parameter aggregation weight to obtain the global computation parameter.

7. The method according to claim 1, characterized in that, Based on sensitivity analysis and adaptive parameter allocation algorithms, global computational parameters are distributed to local enterprise servers to form a business collaborative optimization strategy, including: Business data samples are collected from the enterprise's local server. Feature extraction is performed on the business data samples to obtain sample feature vectors and the corresponding gradient sensitivity is calculated. A sensitivity matrix is ​​constructed based on the gradient sensitivity. The average sensitivity is obtained by summing the absolute values ​​of each element in the sensitivity matrix. The fluctuation level is obtained by calculating the square root of the sum of squares of each element in the sensitivity matrix. The average sensitivity and the fluctuation level are weighted to obtain the parameter importance index. The global calculated parameters are sorted in descending order according to the parameter importance index and grouped into multiple parameter groups by combining the preset importance threshold. The mean vector of the parameter vector in each parameter group is calculated as the feature center of the current parameter group. The system obtains the performance metrics of the enterprise's local server and constructs a performance feature vector. It calculates the cosine similarity between the performance feature vector and each parameter group, as well as the Euclidean distance between the performance feature vector and the feature center. The matching coefficient is obtained by summing the cosine similarity and the Euclidean distance after exponential decay. Based on the matching degree coefficient, a softmax operation is performed to obtain the parameter allocation probability. Based on the parameter allocation probability, the globally calculated parameters are allocated to the enterprise's local server to obtain a business collaboration optimization strategy.

8. A low-code ERP building system for connecting upstream and downstream businesses in the supply chain, used to implement the method of any one of claims 1-7, characterized in that, include: The first unit is used to acquire business process data of upstream and downstream enterprises in the supply chain, extract key business node information, and generate business component configuration interfaces and business data collection forms. The second unit is used to extract business process feature data by feature encoding of key business node information based on business data in the business data collection form, perform temporal memory operation on the business process feature data to obtain temporal dependency features and calculate the importance weight coefficient of business nodes to obtain a business node priority list, perform anomaly detection on the temporal dependency features based on the isolated forest algorithm, determine the anomaly type list and generate anomaly handling action sequence through reinforcement learning algorithm, and generate a set of business rule configuration parameters based on the business node priority list and anomaly handling action sequence. The third unit is used to configure business functions based on the business component configuration interface and the set of business rule configuration parameters, and to generate business rule execution instructions and data processing flow. The fourth unit is used to execute instructions and data processing flow according to business rules, perform desensitization processing on the data in the business data collection form to obtain local calculation parameters, encrypt the local calculation parameters, generate encrypted parameter packages and perform multi-party secure calculations to obtain global calculation parameters, and allocate the global calculation parameters to the enterprise local server based on sensitivity analysis and adaptive parameter allocation algorithm to form a business collaborative optimization strategy. The fifth unit is used to apply business collaboration optimization strategies to business rule execution instructions, monitor business execution status, collect execution data, and update the set of business rule configuration parameters.

9. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 7.

Images