Financial data generation method and system based on rule configuration engine

Abstract

The invention discloses a financial data generation method and system based on a rule configuration engine, and the method comprises the steps: converting a business demand into structured rule description through a semantic analysis engine according to the financial business demand, dynamically configuring a financial data processing rule through a context-aware rule template library, and generating financial data according to the financial data processing rule. Generating an executable rule configuration set; extracting related data from a plurality of heterogeneous financial data sources according to the rule configuration set, and mapping features of different data sources to a unified high-dimensional vector space by adopting a multi-source data fusion algorithm to generate a fused financial feature matrix; and inputting the fused financial feature matrix into a rule configuration engine, adopting a rule execution optimization model, dynamically adjusting a rule execution sequence and parameters, generating financial data, and performing noise addition processing on the financial data to ensure data privacy security. According to the embodiment of the invention, the financial data can be flexibly and efficiently generated, and the safety and efficiency of overall data processing are improved.

Description

Technical Field

[0001] The present invention belongs to the field of financial technology, and particularly relates to a method and system for generating financial data based on a rule configuration engine. Background Art

[0002] With the development of information technology and the explosion of data volume, enterprises face complex business requirements and challenges in financial data processing. Traditional static rule configuration methods are inefficient and lack flexibility, making it difficult to adapt to a rapidly changing environment. Moreover, existing natural language processing capabilities have limitations in understanding user intentions, resulting in possible information loss when converted into executable rules. In addition, the effective integration of multi-source heterogeneous data and the adaptive ability of dynamic rule execution also need to be urgently addressed. At the same time, with the strict implementation of data privacy protection laws, how to improve the efficiency and security of financial data processing while ensuring data compliance has become an urgent need. Summary of the Invention

[0003] The purpose of the present invention is to provide a method and system for generating financial data based on a rule configuration engine to solve the deficiencies in the prior art, capable of flexibly and efficiently generating financial data and enhancing the security and efficiency of overall data processing.

[0004] An embodiment of the present application provides a method for generating financial data based on a rule configuration engine, the method comprising:

[0005] According to the financial business requirements input by the user, using a semantic parsing engine based on natural language processing, converting the business requirements into a structured rule description, and dynamically configuring financial data processing rules through a context-aware rule template library to generate an executable rule configuration set;

[0006] According to the rule configuration set, extracting relevant data from multiple heterogeneous financial data sources, and using a multi-source data fusion algorithm based on graph embedding to map the features of different data sources to a unified high-dimensional vector space to generate a fused financial feature matrix;

[0007] Inputting the fused financial feature matrix into a rule configuration engine, using a rule execution optimization model based on reinforcement learning to dynamically adjust the rule execution order and parameters to generate financial data, and performing noise addition processing on the financial data through a differential privacy protection mechanism to ensure data privacy and security.

[0008] Optionally, the step of according to the financial business requirements input by the user, using a semantic parsing engine based on natural language processing, converting the business requirements into a structured rule description, and dynamically configuring financial data processing rules through a context-aware rule template library to generate an executable rule configuration set includes:

[0009] According to the financial business requirement text input by the user, a semantic parsing engine based on a pre-trained language model is used to tokenize the text, identify entities, and perform semantic role labeling, extract key business elements, where the key business elements include financial accounts, calculation logics, time ranges, and constraint conditions, and generate a preliminary structured semantic representation;

[0010] The preliminary structured semantic representation is input into a context-aware rule template library, and a template matching algorithm based on an attention mechanism is used to dynamically match the most relevant rule templates in combination with historical rule configuration records and the current business scenario, generating a candidate rule set;

[0011] For the candidate rule set, a rule logic optimization method based on formal verification is adopted. Through temporal logic and a constraint solver, the consistency and completeness of the rules are verified, rule conflicts and redundancies are eliminated, and an optimized rule logic description is generated;

[0012] The optimized rule logic description is transformed into an executable rule configuration set, and a rule description framework based on a domain-specific language is used to generate an executable rule configuration set including trigger conditions, execution actions, and exception handling mechanisms.

[0013] Optionally, according to the rule configuration set, relevant data is extracted from multiple heterogeneous financial data sources, and a multi-source data fusion algorithm based on graph embedding is used to map the features of different data sources to a unified high-dimensional vector space, generating a fused financial feature matrix, including:

[0014] According to the financial accounts, time ranges, and data source requirements defined in the rule configuration set, relevant data is extracted from multiple heterogeneous financial data sources, and a data alignment algorithm based on dynamic time warping is used to eliminate the problem of inconsistent timestamps and perform interpolation filling for missing values, obtaining a time-synchronized multi-source data set;

[0015] The time-synchronized multi-source data set is transformed into a graph structure representation, where the nodes of the graph structure represent financial entities, and the edges of the graph structure represent the association relationships between entities. A construction method based on an attribute graph model is used to generate a heterogeneous financial data graph including node attributes and edge weights;

[0016] For the heterogeneous financial data graph, a graph embedding algorithm based on a graph attention network is used. Through a multi-layer attention mechanism, complex dependencies between nodes are captured, and each node is mapped to a unified high-dimensional vector space, obtaining a node-level embedded feature representation;

[0017] The node-level embedded feature representations are aggregated according to financial entity categories, and a feature fusion method based on a multi-head attention mechanism is used to perform weighted fusion of the features of different data sources, generating a fused financial feature matrix including time dimension, account dimension, and entity dimension.

[0018] Optionally, when inputting the fused financial feature matrix into the rule configuration engine, a rule execution optimization model based on reinforcement learning is adopted to dynamically adjust the rule execution order and parameters, generate financial data, and perform noise addition processing on the financial data through a differential privacy protection mechanism to ensure data privacy and security, including:

[0019] Input the fused financial feature matrix into the rule configuration engine, and adopt a reinforcement learning model based on a deep Q-network to dynamically adjust the rule execution order according to the historical feedback data of rule execution, and generate an optimal rule execution sequence;

[0020] For each rule in the optimal rule execution sequence, adopt a parameter adjustment method based on Bayesian optimization to dynamically optimize the rule parameters according to the distribution characteristics of the current financial feature matrix, and generate preliminary financial data;

[0021] For the preliminarily generated financial data, adopt a differential privacy protection method based on the Laplace mechanism to dynamically calculate the noise addition amount according to the data sensitivity level, and perform noise addition processing on key financial indicators to generate intermediate financial data that meets the privacy protection requirements;

[0022] For the intermediate financial data after noise addition, adopt a consistency check model based on a constraint satisfaction problem to correct and adjust the data according to financial rules and business logic constraints, and generate final high-quality financial data.

[0023] Another embodiment of the present application provides a financial data generation system based on a rule configuration engine, and the system includes:

[0024] A conversion module, configured to convert business requirements into structured rule descriptions by adopting a semantic parsing engine based on natural language processing according to the financial business requirements input by a user, and dynamically configure financial data processing rules through a context-aware rule template library to generate an executable rule configuration set;

[0025] A fusion module, configured to extract relevant data from multiple heterogeneous financial data sources according to the rule configuration set, and adopt a multi-source data fusion algorithm based on graph embedding to map the features of different data sources to a unified high-dimensional vector space to generate a fused financial feature matrix;

[0026] A generation module, configured to input the fused financial feature matrix into the rule configuration engine, adopt a rule execution optimization model based on reinforcement learning to dynamically adjust the rule execution order and parameters, generate financial data, and perform noise addition processing on the financial data through a differential privacy protection mechanism to ensure data privacy and security.

[0027] Another embodiment of the present application provides a storage medium, in which a computer program is stored, and the computer program is configured to execute the method described in any one of the above when running.

[0028] Another embodiment of the present application provides an electronic device, including a memory and a processor. A computer program is stored in the memory, and the processor is configured to run the computer program to execute the method described in any one of the above.

[0029] Compared with the prior art, a financial data generation method based on a rule configuration engine provided by the present invention, according to financial business requirements, uses a semantic parsing engine to convert business requirements into structured rule descriptions, and through a context-aware rule template library, dynamically configures financial data processing rules to generate an executable rule configuration set; according to the rule configuration set, relevant data is extracted from multiple heterogeneous financial data sources, and a multi-source data fusion algorithm is used to map the features of different data sources to a unified high-dimensional vector space to generate a fused financial feature matrix; the fused financial feature matrix is input into the rule configuration engine, and a rule execution optimization model is used to dynamically adjust the rule execution order and parameters to generate financial data, and noise addition processing is performed on the financial data to ensure data privacy and security, so that financial data can be generated flexibly and efficiently, and the security and efficiency of overall data processing are improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 It is a hardware structure block diagram of a computer terminal for a financial data generation method based on a rule configuration engine provided by an embodiment of the present invention;

[0031] Figure 2 It is a schematic flow chart of a financial data generation method based on a rule configuration engine provided by an embodiment of the present invention;

[0032] Figure 3 It is a schematic structural diagram of a financial data generation system based on a rule configuration engine provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0033] The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0034] An embodiment of the present invention first provides a financial data generation method based on a rule configuration engine. This method can be applied to an electronic device, such as a computer terminal, specifically, such as an ordinary computer, etc.

[0035] The following takes running on a computer terminal as an example to describe it in detail. Figure 1The following is a block diagram of the hardware structure of a computer terminal for a financial data generation method based on a rule configuration engine provided by an embodiment of the present invention. As Figure 1 shown, the computer device includes a processor, a memory, and a network interface connected through a system bus. Among them, the memory may include a non-volatile storage medium and an internal memory.

[0036] The non-volatile storage medium can store an operating system and a computer program. The computer program includes program instructions. When the program instructions are executed, the processor can execute any financial data generation method based on a rule configuration engine.

[0037] The processor is used to provide computing and control capabilities to support the operation of the entire computer device.

[0038] The internal memory provides an environment for the operation of the computer program in the non-volatile storage medium. When the computer program is executed by the processor, the processor can execute any financial data generation method based on a rule configuration engine.

[0039] The network interface is used for network communication, such as sending assigned tasks, etc. Those skilled in the art can understand that Figure 1 the structure shown in is only a block diagram of some structures related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine some components, or have different component arrangements.

[0040] It should be understood that the processor may be a central processing unit (CPU), and the processor may also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0041] See Figure 2 , an embodiment of the present invention provides a financial data generation method based on a rule configuration engine, which may include the following steps:

[0042] S201. According to the financial business requirements input by the user, a semantic parsing engine based on natural language processing is adopted to transform the business requirements into structured rule descriptions. Through a context-aware rule template library, financial data processing rules are dynamically configured to generate an executable set of rule configurations.

[0043] According to the financial business requirements input by the user, a semantic parsing engine based on natural language processing is adopted to transform the business requirements into structured rule descriptions. Through dynamic configuration with the help of a context-aware rule template library, an executable set of rule configurations is generated. With natural language processing technology, the system can identify key financial elements in the text provided by the user, such as financial accounts, calculation logics, and time ranges. Through in-depth understanding of these elements and grasping of the context, the semantic parsing engine transforms the user's natural language requirements into a formal rule structure to ensure that these rules can be effectively executed by the subsequent data processing engine.

[0044] The significance of this method lies in that it greatly improves the flexibility and efficiency of financial data configuration. Without the need to deeply understand the underlying technical details, users can directly express business requirements in natural language, and the system will intelligently parse and automatically generate executable rules. This not only reduces the learning cost of users but also shortens the time from requirement to execution, thereby improving the response speed and accuracy of financial data processing, enabling enterprises to adapt to market changes and business adjustments faster.

[0045] Specifically, according to the financial business requirement text input by the user, a semantic parsing engine based on a pre-trained language model is adopted to perform word segmentation, entity recognition, and semantic role labeling on the text, extract key business elements, where the key business elements include financial accounts, calculation logics, time ranges, and constraint conditions, and generate a preliminary structured semantic representation.

[0046] In this step, the system uses a pre-trained language model and inputs the financial business requirement text provided by the user. The model first performs word segmentation on the text to identify words and their combinations. Next, the system conducts entity recognition to find terms with specific meanings in the text, such as financial accounts like "income" and "expense", and time information like "the first quarter of 2023". At the same time, semantic role labeling is used to identify the role relationships in the text, such as who is performing what operation, so as to extract calculation logics and constraint conditions. These processes will form a preliminary structured semantic representation for subsequent rule configuration.

[0047] The importance of this process lies in transforming vague natural language into clear and recognizable structured information, ensuring that the system can understand the user's true needs. By accurately extracting key business elements, the system can provide the necessary data foundation for subsequent rule configuration, thereby reducing the risk of misunderstandings and improving the accuracy of rule generation. This ability enables non-professional users to interact with the system conveniently, promoting the automation and intelligence of financial data processing.

[0048] First, the system will receive the natural language text input by the user, which may cover various financial transactions, such as report generation, budget approval, etc. Using a pre-trained language model, the system will perform detailed word segmentation on the text. For example, if the text input by the user is "Calculate the sales revenue of each product in the first quarter of 2023", the system will break it down into multiple meaningful words and phrases, such as "Calculate", "the first quarter of 2023", "each product", "sales revenue". This word segmentation not only helps the system identify the independent meaning of each word but also lays the foundation for subsequent processing.

[0049] Next, the system will perform entity recognition to extract the key business elements in the text. In the above example, the system can identify "the first quarter of 2023" as the time range and "sales revenue" as the financial item. At the same time, based on semantic role labeling, the system will analyze who is the subject and what the action is, so as to confirm that the action of "Calculate" is initiated by the user and the target is "the sales revenue of each product". Through this series of processes, the system can form a preliminary structured semantic representation, such as in the form of a dictionary, which contains key-value pairs of various key elements.

[0050] Finally, the generated preliminary structured semantic representation will provide the basic data for subsequent steps. For example, the system may output a structured representation, such as: json{"Time range": "the first quarter of 2023", "Action": "Calculate", "Financial item": "Sales revenue", "Object": "Each product"}. This data structure provides a clear and understandable framework for subsequent rule matching and execution, enabling the system to understand the user's specific needs.

[0051] Input the preliminary structured semantic representation into the context-aware rule template library, adopt a template matching algorithm based on the attention mechanism, and combine historical rule configuration records and the current business scenario to dynamically match the most relevant rule templates to generate a set of candidate rules;

[0052] In this step, the system inputs the preliminary structured semantic representation into the context-aware rule template library and uses an attention-based template matching algorithm for dynamic selection of candidate rules. The attention mechanism helps the system focus on the templates most relevant to the input semantic description and combines historical rule configuration records to better match the current business scenario. This ensures that the generated set of candidate rules has the maximum relevance to user requirements and adapts to changes in the current business environment.

[0053] By leveraging context awareness and the attention mechanism, the system can filter out the most relevant candidate rules from a large number of possible rule templates. This process improves the accuracy and flexibility of rule matching, making the generated rules more in line with the actual needs of users. In addition, this dynamic matching method also helps to learn from historical experience, reduce the workload of manual configuration, and thus accelerate the efficiency of financial data processing.

[0054] The system passes the already generated preliminary structured semantic representation to the context-aware rule template library. At this time, using the attention-based template matching algorithm, the system will automatically match the most suitable rule template for the current business requirements. For example, in the case where the user wishes to calculate the sales revenue of each product in the first quarter of 2023, the system will evaluate multiple rules related to this request, such as the "quarterly report generation rule" and the "sales revenue calculation rule", through the algorithm and determine which rules are the most relevant in the current context.

[0055] The working mechanism of this algorithm is based on attention allocation. It analyzes the current input structured semantics and calculates the relevance scores for each rule template. The system will pay attention to the patterns learned in model training and historical data, such as analyzing the rules used in past similar requests, so as to improve the accuracy of matching. For example, if historical data shows that the "quarterly report generation rule" is frequently used in similar scenarios, the system will give priority to including it in the set of candidate rules.

[0056] Finally, after dynamic matching, the system will generate a set of candidate rules that cover different dimensions and processing methods required by user needs. This may include formulas for calculating sales revenue, corresponding output format definitions, etc. Suppose the candidate rules finally matched by the system include "calculate the sales revenue of all products" and "output the quarterly sales report", which will form the basis for rule optimization in subsequent steps.

[0057] For the set of candidate rules, adopt a rule logic optimization method based on formal verification. Through temporal logic and constraint solvers, verify the consistency and completeness of the rules, eliminate rule conflicts and redundancies, and generate an optimized rule logic description;

[0058] In this step, the system performs formal verification on the generated set of candidate rules to ensure the consistency and completeness among the selected rules. Through temporal logic analysis, the system checks whether the execution order and conditions of the rules meet the logical requirements. Meanwhile, by using a constraint solver, the system can identify and eliminate possible rule conflicts and redundancies, ensuring that the generated rule logic is efficient and adaptable.

[0059] This optimization process greatly improves the quality and reliability of the generated rules. Through a strict verification mechanism, logical errors caused by rule conflicts can be avoided, and the efficiency of rule execution can be enhanced. The finally generated rule logic description will have high consistency and integrity, and can effectively reduce errors during actual financial data processing, improving the stability of the entire system.

[0060] The system performs formal verification on the set of candidate rules to ensure that the generated rules are not only valid, but also logically compatible and complete with each other. First, the system uses temporal logic to analyze the execution order of the rules to ensure that no logical contradictions occur during rule execution. For example, if one rule states that "if the sales revenue is greater than 0, then the condition is met", and another rule states that "if the sales revenue is negative, then record it as an error", the system will check whether these two rules can coexist. If there are redundancies or conflicts, the system must take measures to make adjustments.

[0061] Secondly, the system will apply a constraint solver, which can verify the interrelationships among the rules and eliminate unnecessary conflicts. When multiple rules in the candidate rules involve similar conditions, the system will evaluate the constraint conditions of these rules to ensure that no rule will conflict with other rules under specific circumstances. For example, if one rule requires that "the sales revenue cannot exceed 1 million yuan", and another rule requires that "the sales revenue must be below 2 million yuan", the system will mark it as having a potential conflict and make necessary optimizations.

[0062] Finally, after the above verification process, the system will generate an optimized rule logic description based on logical judgment and constraint conditions. This logic description not only ensures the validity of the rules, but also improves the execution efficiency of the rules. Suppose that after optimization, the rule generated by the system is "if the time is the first quarter of 2023 and the sales revenue is greater than 0, then generate a sales report". This optimized description will lay a solid foundation for the executability configuration in the last step, ensuring smooth execution in actual operations.

[0063] Convert the optimized rule logic description into an executable set of rule configurations, adopt a rule description framework based on a domain-specific language, and generate an executable set of rule configurations including trigger conditions, execution actions, and exception handling mechanisms.

[0064] In this final step, the optimized rule logic is transformed into a set of executable rule configurations. With the help of a domain-specific language (DSL), the system can describe the triggering conditions, execution actions, and corresponding exception handling mechanisms of each rule, making the rules operable. This transformation ensures that the generated rules not only meet the business requirements but can also be effectively recognized and executed by the system's execution engine.

[0065] The finally generated set of executable rule configurations provides specific guidance for subsequent financial data processing. By clearly defining the triggering conditions and execution actions, the system can automatically trigger the rules and execute relevant actions when receiving corresponding inputs, thus achieving efficient financial data generation. In addition, the addition of an exception handling mechanism also enhances the system's response ability and stability in the face of unexpected situations, further ensuring the integrity and accuracy of data processing.

[0066] The system transforms the optimized rule logic description into a set of rule configurations that can be executed by a machine. First, the system will adopt a domain-specific language (DSL), which is specifically used to describe the business logic of a specific domain, making the definition of rules more intuitive and easy to understand. For example, assuming that the optimized rule logic description is "If the time is the first quarter of 2023 and the sales revenue is greater than 0, then generate a sales report", the system will convert it into the DSL format, defining the triggering conditions, execution actions, and corresponding exception handling.

[0067] Next, the specific rule configurations will detail the triggering conditions, such as "time = the first quarter of 2023" and "sales revenue > 0". The system will define the corresponding execution actions, such as "perform the operation of generating a sales report", when generating the rules. In addition, an exception handling mechanism is also essential, such as "if the sales revenue is negative, record the error information", this refined processing ensures that the system can make appropriate responses when facing situations that do not meet the conditions.

[0068] Finally, the system will generate a complete set of executable rule configurations, including all necessary elements, to ensure smooth operation during execution. For example, the final rule configuration may include the following: json{"triggering conditions":{"time":"the first quarter of 2023","sales revenue":">0"},"execution action":"generate a sales report","exception handling":"if the sales revenue < 0, record the error information"}. Such a configuration is not only clear but also in actual execution, the system can automatically respond according to the preset conditions, greatly improving the automation level and accuracy of financial data processing.

[0069] S202. Extract relevant data from multiple heterogeneous financial data sources according to the rule configuration set, and use a multi-source data fusion algorithm based on graph embedding to map the features of different data sources into a unified high-dimensional vector space to generate a fused financial feature matrix;

[0070] In this step, first, according to the requirements such as financial accounts and time ranges defined in the rule configuration set, it is clear which data sources need to be extracted. For example, it may be necessary to extract data from a sales database, an inventory management system, and financial statements. These data sources may have different structures and formats. The system will use the dynamic time warping algorithm to uniformly process the timestamps in each data source to solve the data alignment problem caused by inconsistent time records in different data sources.

[0071] On this basis, the system will convert the obtained datasets into a graph structure data model, representing financial entities and their relationships in the form of nodes and edges. Nodes represent different financial entities (such as customers, products, transactions, etc.), while edges represent the associations between these entities. Next, use the graph attention network algorithm to embed these heterogeneous data graphs, map the nodes into a unified high-dimensional vector space, so as to effectively integrate multi-source data and generate a fused financial feature matrix.

[0072] By extracting and fusing relevant data from multiple heterogeneous financial data sources, the system can integrate financial information from different sources, form a comprehensive financial view, and provide data support for subsequent analysis and decision-making. This fusion not only improves the integrity and consistency of data, but also strengthens the modeling ability of complex relationships through the form of graph structure, enabling the system to more accurately reflect the upstream and downstream relationships between financial data. The finally generated fused financial feature matrix will be used as the basis for subsequent rule execution, ensuring that the system can make financial decisions based on accurate and efficient data, greatly improving the efficiency of financial data processing.

[0073] Specifically, relevant data can be extracted from multiple heterogeneous financial data sources according to the financial accounts, time ranges, and data source requirements defined in the rule configuration set, and a data alignment algorithm based on dynamic time warping is used to eliminate the problem of inconsistent timestamps and interpolate and fill missing values to obtain a time-synchronized multi-source data set;

[0074] In this step, the system identifies the required data sources and related financial accounts based on the previously generated rule configuration set. The system first extracts this data from multiple heterogeneous sources, such as accounting systems, sales databases, and customer relationship management systems. To achieve data integration, the system employs the Dynamic Time Warping (DTW) algorithm, which can effectively handle the irregularities and variations in time series data, ensuring that the timestamps from different data sources are aligned. This process is crucial because in financial data, the consistency of time helps ensure the comparability and accuracy of various metrics.

[0075] After completing the time alignment, the system checks for missing values in the extracted data set. If missing data is found, the system repairs the missing values through interpolation filling methods. For example, linear interpolation or time series prediction models are used to complete the missing values, ensuring the continuity and integrity of the data. Ultimately, this series of operations will result in a time-synchronized multi-source data set that meets the requirements of subsequent processing.

[0076] The accurate implementation of this process is crucial for improving data quality. The time-synchronized multi-source data set not only ensures the comparability between different data sources but also provides a solid data foundation for subsequent analysis. This high-quality data integration can significantly enhance the effectiveness of subsequent decision-making, ensuring that the generated financial data is more reliable and accurate, thus better supporting the operation and decision-making of enterprises in practical applications.

[0077] Taking a specific example, assume that an enterprise needs to extract data for the "first quarter of 2023" from its sales, inventory, and financial systems. The system first identifies these data sources and extracts relevant data such as sales amounts, inventory quantities, and financial expenses. Then, the system checks the timestamps of this data. Assume the timestamp format of the sales system is "YYYY-MM-DD", while the financial system records it as "YYYY / MM / DD". Through the Dynamic Time Warping algorithm, the system can convert all timestamps into a unified format to ensure daily comparability.

[0078] During this process, the system discovers that the sales data for some days is missing. For example, the sales amount on March 15th is not recorded. To solve this problem, the system can utilize the sales data at the adjacent time points and calculate the value for March 15th through linear interpolation and fill it into the data set. The final result of this series of steps is a time-consistent and multi-source aligned data set, laying a solid foundation for subsequent data processing.

[0079] Convert the time - synchronized multi - source data set into a graph - structure representation. Among them, the nodes of the graph structure represent financial entities, and the edges of the graph structure represent the association relationships between entities. Adopt a construction method based on the property graph model to generate a heterogeneous financial data graph containing node attributes and edge weights.

[0080] In this step, the system converts the time - synchronized data set into a graph - structure representation to better capture the complex relationships between financial entities and among them. First, the system represents each financial entity (such as customers, products, transactions, etc.) in the data set as a node in the graph, while the relationships between entities (such as transaction behaviors like purchases, returns, etc.) are represented as edges in the graph. This graph structure can intuitively reflect the relevance between financial entities and provide richer information for subsequent data processing.

[0081] To construct this graph structure, the system adopts a construction method based on the property graph model. Each node, in addition to having its identifier, also has certain attributes, such as the name of the customer, the transaction amount, etc.; for edges, certain weights are assigned according to the nature of the relationship. For example, if the edge between two nodes represents "Customer A purchases Product B", then the weight of this edge can depend on the purchase amount or the transaction frequency. Such edge - weight design will contribute to subsequent graph - based analysis and mining.

[0082] Converting the time - synchronized data set into a graph - structure representation not only effectively enhances the expressiveness of the data but also provides convenience for subsequent analysis. This structured representation method can help the system better understand and analyze the complex relationships in financial data. For example, it can identify the most important customers, popular products, etc., and then achieve more accurate financial forecasting and decision - making support. In addition, the flexibility of the graph structure makes it smoother to adopt graph algorithms for in - depth analysis in the future. For example, graph traversal algorithms can be used to find key financial relationships or patterns.

[0083] In the specific implementation process, assume that the system has obtained time - synchronized sales data, customer data, and product data. First, the system defines each customer, each product, and each transaction as a node in the graph. For example, node A corresponds to the customer "Zhang San", node B corresponds to the product "laptop", and edge C represents that Zhang San purchased this product.

[0084] Next, in the process of constructing edges, the system will check the transaction records to determine the relationship strength. For example, if in the transaction records, the purchase amount of customer Zhang San for laptops this month is 1000 yuan, while the purchase amount for another product is only 100 yuan, the system can set the weight of edge C to 10 (weight = ratio of purchase amounts). In this way, the generated heterogeneous financial data graph will contain rich node attributes and edge weights, constructing a graph structure that is helpful for subsequent analysis.

[0085] For the heterogeneous financial data graph, a graph embedding algorithm based on a graph attention network is adopted to capture the complex dependencies between nodes through a multi-layer attention mechanism, and each node is mapped to a unified high-dimensional vector space to obtain an embedded feature representation at the node level;

[0086] In this step, the system processes the constructed heterogeneous financial data graph and maps the nodes to a unified high-dimensional vector space using an algorithm based on the Graph Attention Network (GAT). The core idea of ​​the Graph Attention Network is to dynamically weight the information of adjacent nodes in the graph structure through multiple levels of attention mechanisms, thereby capturing the complex dependencies between nodes. For example, when processing a customer node, the system not only considers the customer's direct purchase record, but also the purchase behavior of the customers around him (friends or related customers), thereby enhancing the understanding of the customer's behavior.

[0087] Specifically, the graph embedding algorithm first assigns an initial feature vector to each node, which represents the attribute information of the node. The subsequent multi-layer attention mechanism will calculate the mutual influence between nodes and dynamically update the feature vector through the training model. This process not only allows the system to weight the influence of neighboring nodes when summarizing local information, but also improves the flexibility and adaptability of graph data processing, ensuring that the nonlinear relationship between nodes can be captured.

[0088] The node-level embedded feature representation generated by this step can provide strong support for subsequent data analysis, modeling, and prediction. The embedded features retain the basic attribute information of the nodes and the complex relationships implicit in the graph structure, thereby promoting deep learning and intelligent analysis of financial data. Ultimately, the embedded features of the nodes can be used for a variety of tasks, such as customer segmentation, risk assessment, or financial early warning, further improving the quality of corporate decision-making.

[0089] Taking a specific case as an example, suppose the system processes a heterogeneous data graph containing multiple customers and their purchasing behaviors. In the initial stage, the system generates randomly initialized feature vectors for each customer node and product node. For example, the initial features of customer A may be [0.2, 0.5], and the feature vector of product B may be [0.9, 0.1]. Then, the graph attention network begins to perform its multi-layer processing. In the first layer, the system calculates the correlation between each customer node and its adjacent product nodes, and gives different weights to adjacent nodes through the attention mechanism. Suppose the relationship weight between customer A and product B is 0.8, and the weight with product C is 0.2.

[0090] After multiple iterations, the system will output the embedded feature vectors of each node. In this way, all nodes are successfully mapped to a unified high-dimensional vector space. In the subsequent processing stage, these embedded features will be used for subsequent financial data analysis and decision support to ensure that the system can quickly adapt to the changing financial environment.

[0091] Aggregate the embedded feature representations at the node level by financial entity category. Adopt a feature fusion method based on the multi-head attention mechanism to perform weighted fusion on the features from different data sources, and generate a fused financial feature matrix that includes time dimension, account dimension, and entity dimension.

[0092] In this step, the system aggregates the previously generated node-level embedded feature representations and classifies and integrates them by financial entity category. By adopting a feature fusion method based on the multi-head attention mechanism, the system can effectively capture the feature correlations between different types of financial entities, thereby improving the quality of feature fusion. Each type of financial entity (such as customers, products, transactions) may have different feature vectors, and these vectors need to be weighted and fused under specific categories to generate a unified fused financial feature matrix.

[0093] The use of the multi-head attention mechanism enables the system to assign different weights to the features from different data sources during feature fusion. For example, in customer features, if a certain customer has a very significant impact on a specific product, the system will assign it a higher weight, thereby highlighting its importance during fusion. In this way, the system will better understand and analyze how different categories of financial entities influence each other and generate a feature matrix containing rich information.

[0094] This feature fusion process provides a more comprehensive perspective for subsequent data analysis, enabling the system to analyze financial data from multiple dimensions. The generated fused financial feature matrix not only retains the time dimension but also covers multiple dimensions of financial accounts and entity categories. This multi-dimensional feature representation can provide all-round information support during data mining, modeling, and analysis. For example, the system can perform more accurate financial forecasts, trend analysis, and decision-making based on these comprehensive features, improving the overall financial management effect of the enterprise.

[0095] In specific implementation, assume that the system has generated the embedded feature vectors of different customers with capital inflows and outflows. The system will first classify these features. For example, it will classify the embedded vectors of all customers into three categories: "high-value customers", "medium-value customers", and "low-value customers". Next, using the multi-head attention mechanism, the system will assign different weights to each category of customers. Assume that the feature vector of high-value customers becomes [1.0, 0.5] after weighting, and the feature vector of medium-value customers becomes [0.5, 0.3] after weighting.

[0096] Subsequently, the system will aggregate all customers and their characteristics to form a new integrated financial feature matrix, which includes a time dimension (e.g., the first quarter of 2023), an account dimension (such as "revenue", "expenses"), and an entity dimension (such as "customer type", "product category"). Ultimately, this integrated feature matrix will provide the basic data support for the generation of financial data and decision-making, ensuring that the system can conduct financial analysis from a global perspective.

[0097] S203. Input the integrated financial feature matrix into the rule configuration engine, adopt a rule execution optimization model based on reinforcement learning to dynamically adjust the rule execution order and parameters, generate financial data, and perform noise addition processing on the financial data through a differential privacy protection mechanism to ensure data privacy and security.

[0098] First, the integrated financial feature matrix is passed as input to the rule configuration engine. At this stage, the system will use a reinforcement learning model based on a deep Q-network to analyze the historical rule execution feedback data, and then adjust the rule execution order. The reinforcement learning model continuously updates and optimizes its strategy through feedback learning, and selects the optimal execution path in a dynamic environment to ensure efficiency and effectiveness.

[0099] During the execution process, the system will calculate the optimal rule execution sequence based on the current input features. This dynamic adjustment enables the system to flexibly adapt to data changes, ensuring that the generated financial data best meets the user's needs. At the same time, to protect sensitive information, the system will adopt a differential privacy protection mechanism to perform noise addition processing on the generated financial data. This process will dynamically calculate the amount of noise added according to the sensitivity of each financial indicator to ensure that data privacy is properly protected.

[0100] The implementation of this process is extremely important. First of all, it ensures that the generated financial data not only meets the user's needs but also realizes intelligent optimization during the rule execution process. Through the dynamic adjustment of reinforcement learning, the system can continuously improve the rule execution efficiency and flexibility, thereby enhancing the overall quality of financial data generation. In addition, the introduction of the differential privacy protection mechanism ensures that user privacy is protected when dealing with sensitive financial data, which is a key factor that must be considered first when dealing with any personal or corporate data operations. Through the combination of the two, the system can protect user privacy while meeting business requirements, enhancing the usability and reliability of the system.

[0101] Specifically, the integrated financial feature matrix can be input into the rule configuration engine, adopt a reinforcement learning model based on a deep Q-network, and dynamically adjust the rule execution order according to the historical feedback data of rule execution to generate the optimal rule execution sequence;

[0102] At this stage, the system first passes the integrated financial feature matrix as input data to the rule configuration engine. Then, the reinforcement learning model based on the deep Q-network analyzes the rule feedback data collected during the historical execution process. By learning from the past execution results, the model can continuously adjust and optimize the rule execution strategy. Specifically, through continuous trial and error, the deep Q-network learns the effects of different rule execution orders to determine the most beneficial execution path for generating financial data under the current feature matrix.

[0103] Each rule will have a corresponding reward feedback when executed, and the system will adjust according to these rewards, gradually converging to the optimal execution order. For example, when a certain rule is executed and the generated data has a high accuracy, the model will increase the execution probability of this rule; conversely, if the execution effect of a certain rule is not good, the system will reduce the usage frequency of this rule, thus forming a more efficient rule execution sequence.

[0104] The core function of this step is to improve the efficiency and accuracy of financial data generation. Through the dynamic optimization of the reinforcement learning model, the system can flexibly adjust the priority of rules according to the changes in each execution environment, thereby generating higher-quality data. This intelligent execution strategy can not only effectively reduce the need for manual intervention, but also provide personalized data generation solutions for different business scenarios, greatly improving the adaptability and flexibility of the system.

[0105] When specifically implemented, assume that the system receives a new integrated financial feature matrix during operation. The model first saves the execution history of all current rules, such as the results of executing a certain rule before and the corresponding feedback. Subsequently, the system calculates the "Q-values" of each rule through the deep Q-network, and these "Q-values" represent the expected benefits of different rule combinations in terms of generating financial data effects.

[0106] For example, assume that there are three rules to be executed. The system will calculate the expected effects of each rule under the current feature matrix in turn, generating corresponding Q-values. The model determines the rule combination to be executed first by comparing these Q-values and formulates the optimal execution strategy for the subsequent financial data generation. Such an implementation method ensures the efficiency and high accuracy of the financial data generation process.

[0107] For each rule in the optimal rule execution sequence, adopt a parameter adjustment method based on Bayesian optimization. According to the distribution characteristics of the current financial feature matrix, dynamically optimize the rule parameters to generate preliminary financial data;

[0108] In this step, the system extracts each rule in the optimal rule execution sequence and dynamically adjusts its parameters using the method based on Bayesian optimization. This method aims to optimize the parameters in the rule execution process through statistical learning, leveraging the distribution characteristics of historical data and the current financial feature matrix. The core of Bayesian optimization lies in establishing a probability model between parameters and the target result, and dynamically updating the parameters through this model to achieve the goal of generating the best financial data.

[0109] When the system executes a certain rule, it first evaluates the distribution of each parameter by combining the current financial feature matrix with the previously successful parameter settings. The system will automatically generate a prior distribution and perform posterior updates based on the execution effect, enabling continuous adjustment of parameters to adapt to new data characteristics and obtain better output results in subsequent executions.

[0110] Using Bayesian optimization can effectively address the problem of improper parameter settings that may occur during rule execution. By leveraging the statistical information of historical data and current features, the system can significantly improve the accuracy and effectiveness of the initial financial data generation, avoiding incorrect outputs caused by improper parameter selection. At the same time, this dynamic optimization process enables the system to have the ability of self-learning, continuously adjusting to adapt to future changes and enhancing the robustness and flexibility of the overall financial data generation.

[0111] Specifically, assume that a certain rule needs to adjust the calculation parameters of a financial indicator according to the financial feature matrix. The system first searches for the historical parameter settings related to this indicator and their corresponding output results. By analyzing these historical data, a prior distribution model of the parameters is established. For example, the calculation parameters of a financial indicator may follow a normal distribution.

[0112] During execution, the system inputs the current feature matrix into the model and calculates the corresponding posterior distribution. Assume that in the case of the current feature matrix, the optimal setting of a certain parameter is 0.8. The system will automatically adjust this parameter and generate the initial financial data. For example, through the analysis of sales data, the finally generated initial financial data may show that the sales amount in the first quarter of 2023 reached 1 million yuan. Through this specific implementation method, the system ensures higher data quality and meets business requirements.

[0113] For the initially generated financial data, a differential privacy protection method based on the Laplace mechanism is adopted. The noise addition amount is dynamically calculated according to the data sensitivity level, and noise is added to the key financial indicators to generate intermediate financial data that meets the privacy protection requirements;

[0114] At this stage, the system will perform privacy protection on the preliminarily generated financial data, adopting the differential privacy protection method based on the Laplace mechanism. Differential privacy is an important technology for protecting user data privacy. By adding noise to the data, it ensures that external observers cannot obtain specific information about a particular user when accessing the data. In this process, the system first needs to evaluate the sensitivity level of each financial indicator to determine the amount of noise to be added for privacy protection.

[0115] The Laplace mechanism calculates the amount of noise based on the sensitivity and privacy budget (also known as ε). This is achieved by creating a Laplace distribution and adding it to the real data. The higher the sensitivity, the greater the amount of noise required, ensuring that the change of any single data will not significantly affect the overall data output. In this way, even if an attacker obtains the processed data, they cannot obtain the accurate financial information of a specific user, thus protecting the privacy of the data.

[0116] By using the differential privacy protection method based on the Laplace mechanism, the system can effectively protect the privacy of users' financial data and enhance users' trust in the system. This process is particularly important for the financial industry because financial data often involves sensitive information such as users' income and expenditure. In this way, it is ensured that the generated intermediate financial data complies with privacy protection standards, laying a secure foundation for subsequent data use. Ultimately, this step can help enterprises make full use of their financial data for analysis and decision-making while complying with regulatory requirements.

[0117] In specific implementation, assume that the system has generated a set of preliminary financial data, including multiple financial indicators such as income, expenditure, and profit. First, the system will evaluate the sensitivity of each indicator. For example, the sensitivity of income is relatively high, while the sensitivity of some expenses can be relatively low.

[0118] Next, the system will determine the amount of noise to be added to each indicator according to the Laplace mechanism. For example, if the sensitivity of income is 0.5 and the set privacy budget (ε) is 1, the calculated noise range may be between -0.5 and 0.5. The system will randomly generate a noise value within this range and add it to the income indicator to generate a new income value. This specific process ensures that the finally generated intermediate financial data not only retains the business value but also maximally protects user privacy.

[0119] For the intermediate financial data after adding noise, a consistency verification model based on constraint satisfaction problems is adopted to correct and adjust the data according to financial rules and business logic constraints to generate the final high-quality financial data.

[0120] In the last step, the system performs a consistency check on the intermediate financial data protected by differential privacy. Using a model based on Constraint Satisfaction Problem (CSP), the system checks and adjusts the data according to preset financial rules and business logics. The basic idea of the Constraint Satisfaction Problem is to ensure that the finally generated data is both logically and financially reasonable by setting a series of constraints.

[0121] The system conducts checks from multiple dimensions. For example, it compares whether the generated financial data is within a reasonable range, whether it complies with financial standards, and whether there are inconsistencies among various indicators. This process not only guarantees the mutual consistency of the data but also ensures that the final data can reflect the real business situation. For instance, if the system detects that the expenditure is greater than the income, it will trigger the constraint conditions and make corresponding corrections.

[0122] Through this consistency check process, the system can ensure that the quality of the generated data reaches a high standard, not only meeting the requirements of financial compliance but also truly providing a basis for business decisions. After such corrections and adjustments, the accuracy and reliability of the data are guaranteed, ultimately solving the problems that may be caused by adding noise during the data processing process and enhancing the application value of the system in actual business.

[0123] In practical applications, assume that the intermediate financial data after differential privacy processing contains key indicators such as income, expenditure, and profit. The system first compares this data with the pre-defined financial rules. For example, it ensures that the expenditure does not exceed a certain proportion of the income. If it is found that the expenditure exceeds the income, the system will automatically trigger the constraint correction mechanism.

[0124] For example, if the initially generated expenditure is 90 million yuan while the income is 80 million yuan, after the system detects this non-compliant situation, it will try to adjust the expenditure data. Through logical reasoning, it will lower the expenditure to below 80 million yuan while maintaining the consistency of the overall financial data. The finally generated high-quality financial data will meet all business rules and logical requirements, providing a reliable data foundation for subsequent business analysis and financial decisions.

[0125] It can be seen that according to the financial business requirements, a semantic parsing engine is adopted to transform the business requirements into structured rule descriptions. Through a context-aware rule template library, the financial data processing rules are dynamically configured to generate an executable rule configuration set. According to the rule configuration set, relevant data is extracted from multiple heterogeneous financial data sources, and a multi-source data fusion algorithm is used to map the features of different data sources to a unified high-dimensional vector space to generate a fused financial feature matrix. The fused financial feature matrix is input into a rule configuration engine, and a rule execution optimization model is used to dynamically adjust the rule execution order and parameters to generate financial data, and noise addition processing is performed on the financial data to ensure data privacy and security, so that financial data can be generated flexibly and efficiently, improving the security and efficiency of the overall data processing.

[0126] Another embodiment of the present invention provides a financial data generation system based on a rule configuration engine. Refer to Figure 3 , the system may include:

[0127] A transformation module 301, configured to, according to the financial business requirements input by a user, adopt a semantic parsing engine based on natural language processing to transform the business requirements into structured rule descriptions, and through a context-aware rule template library, dynamically configure the financial data processing rules to generate an executable rule configuration set;

[0128] A fusion module 302, configured to, according to the rule configuration set, extract relevant data from multiple heterogeneous financial data sources, and adopt a multi-source data fusion algorithm based on graph embedding to map the features of different data sources to a unified high-dimensional vector space to generate a fused financial feature matrix;

[0129] A generation module 303, configured to input the fused financial feature matrix into a rule configuration engine, adopt a rule execution optimization model based on reinforcement learning to dynamically adjust the rule execution order and parameters to generate financial data, and perform noise addition processing on the financial data through a differential privacy protection mechanism to ensure data privacy and security.

[0130] It can be seen that according to the financial business requirements, a semantic parsing engine is adopted to transform the business requirements into structured rule descriptions. Through a context-aware rule template library, the financial data processing rules are dynamically configured to generate an executable rule configuration set. According to the rule configuration set, relevant data is extracted from multiple heterogeneous financial data sources, and a multi-source data fusion algorithm is used to map the features of different data sources to a unified high-dimensional vector space to generate a fused financial feature matrix. The fused financial feature matrix is input into a rule configuration engine, and a rule execution optimization model is used to dynamically adjust the rule execution order and parameters to generate financial data, and noise addition processing is performed on the financial data to ensure data privacy and security, so that financial data can be generated flexibly and efficiently, improving the security and efficiency of the overall data processing.

[0131] An embodiment of the present invention further provides a storage medium, in which a computer program is stored, and the computer program is configured to execute the steps in any one of the above method embodiments when running.

[0132] Specifically, in this embodiment, the above storage medium may be configured to store a computer program for executing the following steps:

[0133] S201, according to the financial business requirements input by the user, using a semantic parsing engine based on natural language processing, convert the business requirements into a structured rule description, through a context-aware rule template library, dynamically configure financial data processing rules, and generate an executable rule configuration set;

[0134] S202, according to the rule configuration set, extract relevant data from multiple heterogeneous financial data sources, and use a multi-source data fusion algorithm based on graph embedding to map the features of different data sources to a unified high-dimensional vector space, and generate a fused financial feature matrix;

[0135] S203, input the fused financial feature matrix into a rule configuration engine, use a rule execution optimization model based on reinforcement learning to dynamically adjust the rule execution order and parameters, generate financial data, and perform noise addition processing on the financial data through a differential privacy protection mechanism to ensure data privacy and security.

[0136] It can be seen that according to the financial business requirements, using a semantic parsing engine, convert the business requirements into a structured rule description, through a context-aware rule template library, dynamically configure financial data processing rules, and generate an executable rule configuration set; according to the rule configuration set, extract relevant data from multiple heterogeneous financial data sources, use a multi-source data fusion algorithm to map the features of different data sources to a unified high-dimensional vector space, and generate a fused financial feature matrix; input the fused financial feature matrix into a rule configuration engine, use a rule execution optimization model to dynamically adjust the rule execution order and parameters, generate financial data, and perform noise addition processing on the financial data to ensure data privacy and security, so as to be able to flexibly and efficiently generate financial data and improve the security and efficiency of overall data processing.

[0137] An embodiment of the present invention further provides an electronic device, including a memory and a processor, a computer program is stored in the memory, and the processor is configured to run the computer program to execute the steps in any one of the above method embodiments.

[0138] Specifically, the above electronic device may further include a transmission device and an input / output device, wherein the transmission device is connected to the above processor, and the input / output device is connected to the above processor.

[0139] Specifically, in this embodiment, the above-mentioned processor can be set to execute the following steps through a computer program:

[0140] S201, according to the financial business requirements input by the user, using a semantic parsing engine based on natural language processing, convert the business requirements into a structured rule description, and through a context-aware rule template library, dynamically configure financial data processing rules to generate an executable rule configuration set;

[0141] S202, according to the rule configuration set, extract relevant data from multiple heterogeneous financial data sources, and use a multi-source data fusion algorithm based on graph embedding to map the features of different data sources to a unified high-dimensional vector space to generate a fused financial feature matrix;

[0142] S203, input the fused financial feature matrix into a rule configuration engine, use a rule execution optimization model based on reinforcement learning to dynamically adjust the rule execution order and parameters to generate financial data, and through a differential privacy protection mechanism, perform noise addition processing on the financial data to ensure data privacy and security.

[0143] It can be seen that according to the financial business requirements, using a semantic parsing engine, convert the business requirements into a structured rule description, and through a context-aware rule template library, dynamically configure financial data processing rules to generate an executable rule configuration set; according to the rule configuration set, extract relevant data from multiple heterogeneous financial data sources, use a multi-source data fusion algorithm to map the features of different data sources to a unified high-dimensional vector space to generate a fused financial feature matrix; input the fused financial feature matrix into a rule configuration engine, use a rule execution optimization model to dynamically adjust the rule execution order and parameters to generate financial data, and perform noise addition processing on the financial data to ensure data privacy and security, so as to be able to flexibly and efficiently generate financial data and improve the security and efficiency of overall data processing.

[0144] The above has described in detail the structure, features and effects of the present invention according to the illustrated embodiments. The above is only the preferred embodiment of the present invention, but the present invention is not limited to the scope defined by the drawings. Any changes made according to the concept of the present invention, or modified into equivalent embodiments with equivalent changes, still within the spirit covered by the specification and drawings, should be within the protection scope of the present invention.

Claims

1. A financial data generation method based on a rule configuration engine, characterized in that The method includes: According to the financial business requirements input by the user, using a semantic parsing engine based on natural language processing, transforming the business requirements into a structured rule description, and through a context-aware rule template library, dynamically configuring financial data processing rules to generate an executable rule configuration set; According to the rule configuration set, extracting relevant data from multiple heterogeneous financial data sources, and using a multi-source data fusion algorithm based on graph embedding to map the features of different data sources to a unified high-dimensional vector space to generate a fused financial feature matrix; Inputting the fused financial feature matrix into a rule configuration engine, using a rule execution optimization model based on reinforcement learning to dynamically adjust the rule execution order and parameters to generate financial data, and through a differential privacy protection mechanism, adding noise to the financial data to ensure data privacy and security.

2. The method according to claim 1, characterized in that, The step of, according to the financial business requirements input by the user, using a semantic parsing engine based on natural language processing, transforming the business requirements into a structured rule description, and through a context-aware rule template library, dynamically configuring financial data processing rules to generate an executable rule configuration set, includes: According to the financial business requirement text input by the user, using a semantic parsing engine based on a pre-trained language model to perform word segmentation, entity recognition, and semantic role annotation on the text, extracting key business elements, where the key business elements include financial accounts, calculation logics, time ranges, and constraint conditions, to generate a preliminary structured semantic representation; Inputting the preliminary structured semantic representation into a context-aware rule template library, using a template matching algorithm based on an attention mechanism, and combining historical rule configuration records and the current business scenario, dynamically matching the most relevant rule templates to generate a candidate rule set; For the candidate rule set, using a rule logic optimization method based on formal verification, and through temporal logic and a constraint solver, verifying the consistency and completeness of the rules, eliminating rule conflicts and redundancies, to generate an optimized rule logic description; Converting the optimized rule logic description into an executable rule configuration set, and using a rule description framework based on a domain-specific language to generate an executable rule configuration set including trigger conditions, execution actions, and exception handling mechanisms.

3. The method according to claim 2, wherein The step of, according to the rule configuration set, extracting relevant data from multiple heterogeneous financial data sources, and using a multi-source data fusion algorithm based on graph embedding to map the features of different data sources to a unified high-dimensional vector space to generate a fused financial feature matrix, includes: According to the financial accounts, time ranges, and data source requirements defined in the rule configuration set, extracting relevant data from multiple heterogeneous financial data sources, and using a data alignment algorithm based on dynamic time warping to eliminate the problem of inconsistent timestamps and perform interpolation filling on missing values to obtain a time-synchronized multi-source data set; Converting the time-synchronized multi-source data set into a graph structure representation, where the nodes of the graph structure represent financial entities, and the edges of the graph structure represent the association relationships between entities, and using a construction method based on an attribute graph model to generate a heterogeneous financial data graph including node attributes and edge weights. For the heterogeneous financial data graph, a graph embedding algorithm based on a graph attention network is adopted. Through a multi-layer attention mechanism, the complex dependencies between nodes are captured, and each node is mapped to a unified high-dimensional vector space to obtain a node-level embedded feature representation; The node-level embedded feature representations are aggregated according to the financial entity categories. A feature fusion method based on a multi-head attention mechanism is used to perform weighted fusion on the features of different data sources, generating a fused financial feature matrix that includes time dimension, subject dimension, and entity dimension.

4. The method according to claim 3, wherein Inputting the fused financial feature matrix into a rule configuration engine, a rule execution optimization model based on reinforcement learning is adopted to dynamically adjust the rule execution order and parameters, generating financial data. And through a differential privacy protection mechanism, noise addition processing is performed on the financial data to ensure data privacy and security, including: Inputting the fused financial feature matrix into a rule configuration engine, a reinforcement learning model based on a deep Q-network is adopted. According to the historical feedback data of rule execution, the rule execution order is dynamically adjusted to generate an optimal rule execution sequence; For each rule in the optimal rule execution sequence, a parameter adjustment method based on Bayesian optimization is adopted. According to the distribution characteristics of the current financial feature matrix, the rule parameters are dynamically optimized to generate preliminary financial data; For the preliminarily generated financial data, a differential privacy protection method based on the Laplace mechanism is adopted. According to the data sensitivity level, the noise addition amount is dynamically calculated, and noise addition processing is performed on key financial indicators to generate intermediate financial data that meets the privacy protection requirements; For the intermediate financial data after noise addition, a consistency verification model based on a constraint satisfaction problem is adopted. According to financial rules and business logic constraints, the data is corrected and adjusted to generate final high-quality financial data.

5. A financial data generation system based on a rule configuration engine, characterized in that, The system includes: A transformation module, which is used to convert the business requirements into a structured rule description according to the financial business requirements input by the user, adopting a semantic parsing engine based on natural language processing. Through a context-aware rule template library, the financial data processing rules are dynamically configured to generate an executable rule configuration set; A fusion module, which is used to extract relevant data from multiple heterogeneous financial data sources according to the rule configuration set, adopting a multi-source data fusion algorithm based on graph embedding to map the features of different data sources to a unified high-dimensional vector space, generating a fused financial feature matrix; A generation module, which is used to input the fused financial feature matrix into a rule configuration engine, adopting a rule execution optimization model based on reinforcement learning to dynamically adjust the rule execution order and parameters, generating financial data. And through a differential privacy protection mechanism, noise addition processing is performed on the financial data to ensure data privacy and security.

6. The system according to claim 5, characterized in that The transformation module is specifically used for: According to the financial business requirement text input by the user, adopting a semantic parsing engine based on a pre-trained language model to perform word segmentation, entity recognition, and semantic role annotation on the text, extracting key business elements, where the key business elements include financial subjects, calculation logics, time ranges, and constraint conditions, generating a preliminary structured semantic representation; Input the preliminary structured semantic representation into the context-aware rule template library, and use the template matching algorithm based on the attention mechanism. Combine the historical rule configuration records and the current business scenario to dynamically match the most relevant rule templates and generate a candidate rule set; For the candidate rule set, adopt the rule logic optimization method based on formal verification. Through temporal logic and constraint solvers, verify the consistency and completeness of the rules, eliminate rule conflicts and redundancies, and generate an optimized rule logic description; Convert the optimized rule logic description into an executable rule configuration set, and use the rule description framework based on the domain-specific language to generate an executable rule configuration set including trigger conditions, execution actions, and exception handling mechanisms.

7. The system according to claim 6, characterized in that, The fusion module is specifically used for: According to the financial accounts, time range, and data source requirements defined in the rule configuration set, extract relevant data from multiple heterogeneous financial data sources. Adopt the data alignment algorithm based on dynamic time warping to eliminate the problem of inconsistent timestamps and perform interpolation filling for missing values to obtain a time-synchronized multi-source data set; Convert the time-synchronized multi-source data set into a graph structure representation. Among them, the nodes of the graph structure represent financial entities, and the edges of the graph structure represent the association relationships between entities. Adopt the construction method based on the attribute graph model to generate a heterogeneous financial data graph including node attributes and edge weights; For the heterogeneous financial data graph, adopt the graph embedding algorithm based on the graph attention network. Through the multi-layer attention mechanism, capture the complex dependency relationships between nodes, and map each node to a unified high-dimensional vector space to obtain the node-level embedding feature representation; Aggregate the node-level embedding feature representations according to the financial entity categories, and adopt the feature fusion method based on the multi-head attention mechanism to perform weighted fusion on the features of different data sources to generate a fused financial feature matrix including the time dimension, subject dimension, and entity dimension.

8. The system according to claim 7, characterized in that, The generation module is specifically used for: Input the fused financial feature matrix into the rule configuration engine, and use the reinforcement learning model based on the deep Q network. According to the historical feedback data of rule execution, dynamically adjust the rule execution order to generate an optimal rule execution sequence; For each rule in the optimal rule execution sequence, adopt the parameter adjustment method based on Bayesian optimization. According to the distribution characteristics of the current financial feature matrix, dynamically optimize the rule parameters to generate preliminary financial data; For the preliminarily generated financial data, adopt the differential privacy protection method based on the Laplace mechanism. Dynamically calculate the noise addition amount according to the data sensitivity level, and perform noise addition processing on the key financial indicators to generate intermediate financial data that meets the privacy protection requirements; For the intermediate financial data after adding noise, adopt the consistency verification model based on the constraint satisfaction problem. According to the financial rules and business logic constraints, correct and adjust the data to generate the final high-quality financial data.

9. A storage medium, characterized in that, The computer program is stored in the storage medium, wherein the computer program is set to execute the method according to any one of claims 1-4 when running.

10. An electronic device, comprising a memory and a processor, characterized in that, A computer program is stored in the memory, and the processor is configured to run the computer program to execute the method according to any one of claims 1-4.

Images