CN117709446A

CN117709446A - Method for constructing dynamic financial credit risk model based on rule engine

Info

Publication number: CN117709446A
Application number: CN202311722808.6A
Authority: CN
Inventors: 方园; 毛继恩; 方深田
Original assignee: China Securities Pengyuan Credit Rating Co ltd
Current assignee: China Securities Pengyuan Credit Rating Co ltd
Priority date: 2023-12-14
Filing date: 2023-12-14
Publication date: 2024-03-15

Abstract

The invention relates to the technical field of model construction, in particular to a method for constructing a dynamic financial credit risk model based on a rule engine. The method comprises the following steps: acquiring knowledge data in the financial field; carrying out quantum bit information organization on the financial field knowledge data according to a preset graph structure, thereby obtaining a quantum financial knowledge graph; embedding the financial domain knowledge data into a quantum financial knowledge graph for data aggregation to generate quantum associated financial knowledge aggregation data; carrying out graph structure optimization on the quantum associated financial knowledge aggregate data to generate a quantum financial knowledge graph; based on quantum financial knowledge graph, carrying out multi-mode financial data acquisition so as to obtain financial information extraction data; according to the invention, the dynamic, accuracy and adaptability of the financial credit risk model construction are improved by carrying out rule network self-organization and virtual environment simulation on the multi-mode financial data.

Description

Method for constructing dynamic financial credit risk model based on rule engine

Technical Field

The invention relates to the technical field of model construction, in particular to a method for constructing a dynamic financial credit risk model based on a rule engine.

Background

Financial credit risk management is one of the core challenges in the financial arts. The conventional approach is struggled in the face of rapidly changing and increasing complexity financial markets. With the progress of technology and the continuous evolution of application scenes, a dynamic financial credit risk model construction method based on a rule engine gradually reaches a brand-new angle. Past financial risk models often rely on static rules and traditional statistical methods to capture real-time changing financial market dynamics. With the rapid development of computer science and data science, artificial intelligence and machine learning techniques have provided new solutions to identify risks and patterns by learning and analyzing massive amounts of data. However, the current dynamic financial credit risk model still depends on the decision of credit risk through the acquisition of real-time data, and the analysis of the behavior and risk of the financial system is not accurate and detailed enough, resulting in lower dynamics, accuracy and adaptability of the financial credit risk model.

Disclosure of Invention

Based on this, it is necessary to provide a method for constructing a dynamic financial credit risk model based on a rule engine, so as to solve at least one of the above technical problems.

In order to achieve the above object, a method for constructing a dynamic financial credit risk model based on a rule engine, the method comprises the following steps:

step S1: acquiring knowledge data in the financial field; carrying out quantum bit information organization on the financial field knowledge data according to a preset graph structure, thereby obtaining a quantum financial knowledge graph; embedding the financial domain knowledge data into a quantum financial knowledge graph for data aggregation to generate quantum associated financial knowledge aggregation data; carrying out graph structure optimization on the quantum associated financial knowledge aggregate data to generate a quantum financial knowledge graph;

step S2: based on quantum financial knowledge graph, carrying out multi-mode financial data acquisition so as to obtain financial information extraction data; carrying out multi-modal information alignment on the financial information extraction data to generate multi-modal fusion data; heterogeneous multi-mode information fusion optimization is carried out on the multi-mode fusion data, and multi-mode fusion optimization data are generated;

step S3: performing rule code conversion on the multi-mode fusion optimization data to generate encoded financial rule data; performing self-organizing network training according to the encoded financial rule data to generate a self-organizing map network; performing meta learning model training on the encoded financial rule data by using a meta learning technology to generate a meta learning training model; carrying out model network evolution on the self-organizing map network and the meta-learning training model, thereby generating an evolving financial rule network; performing adaptability evaluation on the evolution financial rule network to generate rule adaptability evaluation data;

Step S4: constructing a virtual environment for the rule adaptability evaluation data to generate virtual financial environment frame data; performing virtual cross-translocation analysis on the virtual financial environment frame data to obtain a virtual financial ecological analysis data set; performing environment optimization on the virtual financial environment frame data according to the virtual financial ecological analysis data set, so as to generate virtual financial world data;

step S5: constructing distributed nodes for the multi-mode fusion optimization data to obtain financial risk distributed nodes; constructing a dynamic rule engine for the financial risk distributed node to generate a dynamic financial rule engine; the dynamic financial rule engine is dynamically updated to generate a real-time optimized financial rule engine;

step S6: sampling the virtual transaction environment of the virtual financial world data to generate financial transaction environment state data; carrying out decision strategy optimization on the financial transaction environment state data based on a support vector machine algorithm to generate optimized decision data; and carrying out credit risk decision on the real-time optimized financial rule engine by using the optimized decision data so as to generate a dynamic financial credit risk model.

According to the invention, the data aggregation operation is executed in the quantum financial knowledge graph, and related knowledge elements are combined, so that quantum associated financial knowledge aggregation data can be generated by quantum superposition and other modes. This helps to discover potential associations and interactions between different knowledge elements, and performs graph structure optimization on quantum associated financial knowledge aggregate data, possibly including removing redundant information, optimizing association weights, and so on. This helps to improve the efficiency and expressive power of the knowledge graph. By using the advantages of quantum computation, the quantum financial knowledge graph may have higher information processing capability, and better capture the complex relationship of the knowledge in the financial field, so that the effects of financial decision, risk assessment and the like are expected to be improved. Through multi-mode fusion, more comprehensive and various financial information is obtained, financial markets and related factors are better understood, the multi-mode information is combined with quantum financial knowledge patterns, so that the knowledge patterns are richer, tasks such as financial decision, risk assessment and the like can be better supported, the advantages of quantum computing in heterogeneous multi-mode information fusion are utilized, the data processing efficiency is improved, and the effect of the whole flow is enhanced. Through meta learning and self-organizing networks, the system can learn and adapt to continuously evolving financial rules better, accuracy and flexibility of decision making are improved, the self-organizing networks and meta learning models are evolved to enable the whole network to be more complex and high in adaptability, so that dynamic changes of financial markets are adapted better, further optimization and improvement of the system can be guided through rule adaptability evaluation data, and good performance of the system in a real financial environment is ensured. By simulating the analysis of the cross-over surface, various transaction activities in the virtual environment can be known, including the buying and selling directions, the transaction frequency, the transaction amount and the like, and by analyzing the virtual cross-over surface, a data set is formed in a summarizing way, and can be used for deeper financial ecological analysis. Helping to build better market efficiency, more robust trading systems, more accurate rule adaptability assessment, etc. This facilitates a deep understanding of the financial ecosystem. The comprehensive multiple data types are helpful for comprehensively knowing financial risks, the data processing speed and efficiency can be improved, the rule engine is dynamically updated, so that the rule engine can timely cope with new financial risk conditions, the real-time optimized financial rule engine can be generated, different financial risk conditions can be flexibly dealt with, new risk factors can be timely dealt with, and the effectiveness of the rule engine is maintained. Through the simulation environment, experiments and optimization can be performed under the condition that the real market is not affected, the SVM is a powerful machine learning algorithm, complex data can be effectively processed, the accuracy of a decision strategy is improved, the changing market condition can be more flexibly adapted through updating the financial rule engine in real time, and a dynamic model is established, so that the method is beneficial to better capturing and coping with the change of the financial market. Therefore, the invention improves the dynamics, accuracy and adaptability of the financial credit risk model construction by carrying out rule network self-organization and virtual environment simulation on the multi-mode financial data.

The method has the beneficial effects that knowledge data in the financial field is collected through various channels (possibly including documents, news, market data and the like), the collected knowledge data in the financial field is organized according to a preset graph structure to form a quantum financial knowledge graph, and the knowledge data in the financial field is embedded into the quantum financial knowledge graph to carry out data aggregation. And then carrying out graph structure optimization on the quantum associated financial knowledge aggregate data to generate a final quantum financial knowledge graph. And acquiring multi-mode financial data by utilizing the quantum financial knowledge graph, acquiring various types of financial information extraction data, carrying out multi-mode information alignment on the financial information extraction data to generate multi-mode fusion data, optimizing the multi-mode fusion data, ensuring that heterogeneous information can be effectively fused, and generating multi-mode fusion optimization data. And performing rule code conversion on the multi-mode fusion optimization data, then performing self-organizing network training by utilizing the encoded financial rule data to generate a self-organizing map network, training the encoded financial rule data by utilizing a meta-learning technology to generate a meta-learning training model, and evolving the self-organizing map network and the meta-learning model to generate an evolving financial rule network. And carrying out adaptability evaluation on the evolution financial rule network to generate rule adaptability evaluation data. And constructing virtual financial environment frame data by utilizing the rule adaptability evaluation data, performing virtual cross-plane analysis on the virtual financial environment frame data to obtain a virtual financial ecological analysis data set, performing environment optimization on the virtual financial environment frame data according to the virtual financial ecological analysis data set, and generating virtual financial world data. And constructing distributed nodes for the multi-mode fusion optimization data to form financial risk distributed nodes, and constructing a dynamic rule engine for the financial risk distributed nodes to realize real-time optimization of the financial rule engine. More accurate and real-time financial decision support can be provided, which is helpful for managing financial risks, optimizing investment portfolios and improving decision-making efficiency. Therefore, the invention improves the dynamics, accuracy and adaptability of the financial credit risk model construction by carrying out rule network self-organization and virtual environment simulation on the multi-mode financial data.

Drawings

FIG. 1 is a flow chart of steps of a method for constructing a dynamic financial credit risk model based on a rule engine;

FIG. 2 is a flowchart illustrating the detailed implementation of step S2 in FIG. 1;

FIG. 3 is a flowchart illustrating the detailed implementation of step S3 in FIG. 1;

FIG. 4 is a flowchart illustrating the detailed implementation of step S4 in FIG. 1;

the achievement of the objects, functional features and advantages of the present invention will be further described with reference to the accompanying drawings, in conjunction with the embodiments.

Detailed Description

The following is a clear and complete description of the technical method of the present patent in conjunction with the accompanying drawings, and it is evident that the described embodiments are some, but not all, embodiments of the present invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the present invention without making any inventive effort, are intended to fall within the scope of the present invention.

Furthermore, the drawings are merely schematic illustrations of the present invention and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and thus a repetitive description thereof will be omitted. Some of the block diagrams shown in the figures are functional entities and do not necessarily correspond to physically or logically separate entities. The functional entities may be implemented in software or in one or more hardware modules or integrated circuits or in different networks and/or processor methods and/or microcontroller methods.

It will be understood that, although the terms "first," "second," etc. may be used herein to describe various elements, these elements should not be limited by these terms. These terms are only used to distinguish one element from another element. For example, a first element could be termed a second element, and, similarly, a second element could be termed a first element, without departing from the scope of example embodiments. The term "and/or" as used herein includes any and all combinations of one or more of the associated listed items.

To achieve the above objective, referring to fig. 1 to 4, a method for constructing a dynamic financial credit risk model based on a rule engine, the method comprises the following steps:

In the embodiment of the present invention, as described with reference to fig. 1, a flow chart of steps of a method for constructing a dynamic financial credit risk model based on a rule engine according to the present invention is shown, and in this example, the method for constructing a dynamic financial credit risk model based on a rule engine includes the following steps:

in embodiments of the present invention, data is cleaned and processed by collecting data from various financial fields, including, but not limited to, market data, corporate financial data, macro-economic data, news, papers, etc., to ensure accuracy and consistency of the data. The financial domain knowledge data is mapped to the qubits. This may involve encoding specific information into qubits, which are organized into a pattern according to a preset pattern. This may be a directed graph, undirected graph, or other graph suitable for representing financial relationships. The financial domain knowledge data is embedded into a quantum financial knowledge graph. This may include a process of mapping data to qubits, with the properties of quantum computing, aggregating data in a quantum financial knowledge graph. This may include summing, average calculation, etc. And generating quantum associated financial knowledge aggregate data according to the embedded and aggregated results, and optimizing the generated data by using a graph algorithm. This may include operations of node merging, pruning, weight adjustment, etc., to improve the quality and interpretability of the graph.

in the embodiment of the invention, knowledge in the field of quantum finance, including information on financial tools, market behaviors, quantum computation and the like, is collected and arranged, and is represented in a graph form by using a knowledge graph modeling tool, wherein nodes represent entities (such as financial products, concepts, quantum algorithms and the like) and edges represent the relationship between the entities. Multimodal data, including text, image, video, etc., is collected from different sources (financial markets, news, social media, etc.), and data crawling is performed using tools such as APIs, web crawlers, etc., and the quality and consistency of the data is ensured. Text data is extracted using Natural Language Processing (NLP) techniques, entities, relationships, and events therein are identified, image and video data is processed, relevant information therein is extracted, and techniques such as computer vision may be required. The alignment of the multi-modal data obtained from different sources ensures consistent representation of the same entity or event in different modalities, and the feature fusion and alignment can be performed using a deep learning model, such as a multi-modal neural network. Handling the heterogeneity between different modalities may require feature engineering or the use of special fusion techniques, considering the potential applications of quantum computing in information fusion, e.g., information fusion and optimization using quantum neural networks. And saving the data subjected to the information alignment and fusion optimization as a new multi-mode fusion optimization data set.

in the embodiment of the invention, the multimode fusion optimization data is subjected to rule code conversion, is expressed into the data in a specific form conforming to the rule, and is designed with proper coding rules, so that the coded data can be understood by a follow-up self-organizing network and a meta-learning model. The encoded financial rule data is trained using a Self-Organizing map (SOM) algorithm, such as Self-Organizing Maps. The self-organizing network can form a topological structure through unsupervised learning, and is helpful for discovering patterns and rules in data. The encoded financial rule data is trained using Meta-learning techniques, such as Meta-learning neural networks (Meta-Learning Neural Networks), which enable the Meta-learning model to adapt to new tasks faster by learning the behavior on different tasks. Model network evolution is carried out on the self-organizing map network and the meta-learning training model, and adjustment on network structure or parameters is possibly involved, wherein the method can be an evolutionary algorithm, a genetic algorithm and the like, and the network structure and parameters which are more suitable for financial rule data are generated through iterative evolution. By combining the evolution results of the self-organizing map network and the meta-learning model, a comprehensive evolution financial rule network is generated, and the network can better understand and adapt to rules and modes in the financial field. The adaptation of the evolving financial rule network is evaluated by using the real or synthetic financial data set to verify its performance, including accuracy of the model, generalization ability, and adaptation to new data.

in the embodiment of the invention, the virtual financial market is created by setting basic parameters of the virtual financial environment, including market structure, financial products, participant types and the like, including simulating a stock exchange, trade participants and related infrastructure, introducing randomness and variability, and simulating uncertainty and volatility in the real financial market. Making a virtual transaction rule, simulating various transaction behaviors, including buying and selling, investment decision making and the like, recording transaction data in a virtual environment, including transaction amount, price, participant behaviors and the like, and performing cross-translocation surface analysis by using the simulation data to obtain data about market dynamics and participant behaviors. Integrating the bit-plane analysis data with other data in the virtual environment to form a complete virtual financial ecological analysis data set, extracting meaningful features from the data set, such as market trends, transaction patterns, asset correlations and the like, and labeling the data set for subsequent environment optimization and meta-learning training. According to the virtual finance ecological analysis data set, parameters of the virtual environment are adjusted to be closer to the characteristics of the real market, and according to the analysis result, virtual transaction rules are adjusted to improve the reality and complexity of the simulation environment. New factors, such as macro economic indicators, policy changes, etc., are introduced into the virtual environment to increase the diversity of the environment. And simulating possible future financial market scenes by using the optimized virtual environment, and generating a large amount of virtual financial data in the simulated environment to construct a complete virtual financial world data set.

in the embodiment of the invention, the financial risk distributed nodes are established by using a distributed computing framework such as Apache Spark, so that large-scale data are effectively processed, task scheduling and coordination mechanisms are designed, the cooperative work of all the distributed nodes is ensured, and the comprehensive analysis of multi-mode data is completed. Defining financial risk rules, including various risk indexes and thresholds, constructing a dynamic rule engine by using rule engine technology (such as Drools, jess and the like) so that the dynamic rule engine can process and execute the defined rules, and inputting multi-mode fusion optimization data into the rule engine for real-time risk assessment. The method has the advantages that the streaming processing technology is utilized to process the data stream generated in real time, the engine can respond to new data in time, a rule dynamic update mechanism is designed, the rule engine can dynamically load, modify and delete rules in running, a machine learning model is integrated, the engine can learn from historical data, and the rules are dynamically updated to adapt to changing market conditions. The monitoring system is deployed to monitor the performance of the rule engine, ensure that the rule engine can keep high efficiency in real-time processing, and design a feedback mechanism so that the engine can optimize and adjust rules according to real-time results to improve accuracy and adaptability. And a fault-tolerant mechanism is introduced to ensure that the operation can be recovered and continued in time when faults or errors occur.

In the embodiment of the invention, the data source of the virtual financial world can be determined as a simulation transaction system, a simulation platform or a synthetic data generator, a virtual transaction environment data sample is obtained from the selected data source, and the financial transaction environment state data is generated by simulating or recording different states of the virtual transaction, such as market fluctuation, transaction behavior, price change and the like. Relevant features are extracted from the financial transaction environment state data, and preprocessing and feature engineering are performed so as to input a support vector machine model. And (3) establishing a model by using a support vector machine algorithm, and training and optimizing the financial transaction environment state data. The model is evaluated and parameters are adjusted by using cross-validation and other techniques to improve the performance and generalization ability of the model. And predicting and optimizing the transaction environment state data based on the SVM model to generate optimized decision data or transaction strategies. Integrating the optimized decision data into a real-time optimized financial rule engine as an important reference for decision making, adjusting rules and logic of the financial rule engine according to the optimized decision data so as to improve accuracy and instantaneity of credit risk decision, and carrying out real-time credit risk decision by using the dynamically updated financial rule engine and combining the optimized decision data. And verifying and evaluating the accuracy and the robustness of the dynamic financial credit risk model by using actual data, iteratively updating the model according to a verification result and real-time application feedback so as to adapt to a continuously-changing financial environment, deploying the generated dynamic financial credit risk model into the actual application, and establishing a monitoring mechanism to ensure the effectiveness and the stability of the dynamic financial credit risk model in a production environment.

Preferably, step S1 comprises the steps of:

step S11: acquiring knowledge data in the financial field;

step S12: carrying out quantum bit information decomposition on the financial domain knowledge data to generate financial quantum bit data; organizing the financial quantum bit data according to a preset graph structure, so as to obtain a quantum financial knowledge graph;

step S13: embedding the financial domain knowledge data into a quantum financial knowledge graph for quantum state superposition to generate a quantum coding financial knowledge embedded data set; carrying out data aggregation on the quantum-encoded financial knowledge embedded data set to generate quantum-associated financial knowledge aggregation data;

step S14: and carrying out graph structure optimization on the quantum associated financial knowledge aggregate data through a gradient descent optimization algorithm to generate a quantum financial knowledge graph.

According to the invention, the complexity and the information density of the data can be improved by decomposing the information of the quantum bit of the knowledge data in the financial field. This may enable better capture of complex relationships and features of financial data as it is processed. The preset graph structure is helpful for organizing financial quantum bit data, so as to form a quantum financial knowledge graph. The structure may better reflect the relationship between various elements in the financial field, providing a better basis for subsequent analysis and processing. By embedding the financial knowledge data into a quantum financial knowledge graph for quantum state superposition, more complex data processing can be performed in the framework of quantum computing. Quantum encoding can help to improve the efficiency and capacity of data encoding. Aggregating quantum-encoded financial knowledge embedded data facilitates generating quantum-associated financial knowledge aggregate data. This may improve the integrity and integrity of the data, providing more comprehensive information for subsequent analysis. And the graph structure of the quantum associated financial knowledge aggregate data is optimized through a gradient descent optimization algorithm, so that the quality and accuracy of the knowledge graph are expected to be improved. This helps to better understand the complex relationships between financial data. The final beneficial effect is that a quantum financial knowledge graph is generated. The atlas may provide a more powerful tool for analysis, prediction and decision making in the financial field, and can better understand and utilize complex financial knowledge systems.

In embodiments of the present invention, data in the financial domain, including financial market data, corporate financial data, economic indicators, transaction data, etc., is obtained through the use of data collection tools or from a variety of sources. And decomposing the obtained knowledge data in the financial field by quantum bit information. This may require techniques for quantum computation or quantum information processing to convert the data into qubit metadata. Designing a preset graph structure, and organizing the quantum bit data into a quantum financial knowledge graph according to the structure. Embedding the financial domain knowledge data into the quantum financial knowledge graph may utilize quantum state superposition methods. This may require quantum encoding techniques to embed the data into the qubit. The data aggregation is carried out on the quantum-encoded financial knowledge embedded data, and a special quantum computing method can be adopted for integrating and processing the data to form quantum-associated financial knowledge aggregation data. And carrying out graph structure optimization on the quantum associated financial knowledge aggregate data by using a gradient descent optimization algorithm. This may include adjusting graph structures, node connections, weights, etc. to optimize morphology and relevance of the graph. Finally, a quantum financial knowledge graph is generated, which is a graphical representation reflecting the knowledge complex relationship in the financial field, and can be displayed in the form of a chart or a graph to show the relationship and connection among the elements in the financial field.

Preferably, step S2 comprises the steps of:

step S21: based on quantum financial knowledge graph, carrying out multi-mode financial data acquisition so as to obtain financial information extraction data;

step S22: data cleaning is carried out on the financial information extraction data, and financial information cleaning data is generated; filling the data missing value of the financial information cleaning data to generate financial information filling data; data standardization is carried out on the financial information filling data, and standard financial information extraction data is generated;

step S23: performing data dimension reduction on the standard financial information extraction data to generate financial information extraction dimension reduction data; extracting the reduced data of the financial information to perform vector conversion to generate financial information vector conversion data; embedding multi-modal information into the financial information vector conversion data by utilizing the embeddable deep learning model to obtain embedded multi-modal information data;

step S24: carrying out multi-modal information alignment on the embedded multi-modal information data to generate embedded multi-modal information alignment data; carrying out data fusion on the embedded multi-modal information alignment data to generate multi-modal fusion data;

step S25: deep learning feature data extraction is carried out on the multi-mode fusion data, and deep learning feature data is generated; heterogeneous multi-modal information fusion optimization is carried out on the deep learning characteristic data through a migration learning method, and multi-modal fusion optimization data are generated.

According to the invention, through data acquisition based on quantum financial knowledge graph, comprehensive financial information is ensured to be acquired from a plurality of modes (such as text, images, time sequences and the like). This helps build a more comprehensive, multi-element dataset, providing a more global view. The quality of the data is ensured by the cleaning operation, the missing values are filled, the data in different modes have consistent scales by standardization, and the stability and the accuracy of subsequent processing are improved. The standard financial information extraction data is subjected to dimension reduction, vector conversion and multi-mode information embedding, so that key features and modes in the data can be extracted, and the embedded multi-mode information data which is more compact and has information quantity can be formed. The embedded multi-mode information is aligned and fused, so that the information of different modes can be ensured to work cooperatively, the consistency and the integrity of the data are improved, and the association relation in the data can be better understood. By deep learning extracted features, the system can automatically learn and capture complex patterns in the data. The heterogeneous multi-modal information fusion optimization further improves the performance of the model, so that the model can better understand and utilize the relationship of different modal information, and the understanding and analysis capability of financial information is improved.

As an example of the present invention, referring to fig. 2, the step S2 in this example includes:

in the embodiment of the invention, the relation among financial entities (such as companies, transactions, indexes and the like) is determined by determining the range of financial fields needing to be covered, including stocks, bonds, futures, market indexes, company financial information and the like, and a map structure is established, and can be represented by adopting a map database or a map network model. Determining reliable sources needing to collect data, such as financial news, official reports, exchange data, financial reports and the like, establishing connection, acquiring information from data sources of different sources and integrating the information into a knowledge graph, and possibly capturing and integrating the data by using technologies such as API (application program interface), web crawlers and the like. Financial related text information is collected from sources such as news, corporate announcements, forums, or social media, visual data such as charts, images, or financial statements are collected through image processing or data visualization techniques, time series data such as stock prices, volume of deals, etc. are obtained from stock exchanges or financial databases, other data types that may exist such as audio data (e.g., financial teleconference recordings), etc. are identified and collected, and financial information extraction data is obtained.

in the embodiment of the invention, the repeated items are identified and deleted, the data with wrong format is repaired or deleted, the abnormal value is detected and processed, the data can be identified and processed by using a statistical method or domain knowledge, and the data is unified to the same format and unit, so that the consistency is ensured. Determining the condition of missing values in data, analyzing the reasons of the missing values, selecting a proper filling method according to the data types and characteristics, wherein the filling method can be mean, median and mode filling, or performing predictive filling by using a machine learning-based method, executing a selected filling strategy, filling the missing values and ensuring the integrity of a data set. The data is normalized or standardized, so that the similarity of numerical ranges among different features is ensured, the excessive influence of certain features on the model is avoided, and the classification variable is encoded and converted into a numerical form so as to facilitate algorithm processing. If time series data is involved, it may be necessary to smooth or translate it into an appropriate time window to accommodate the needs of the model.

in the embodiment of the invention, the most suitable algorithm is selected according to the data characteristics and the dimension reduction requirement by selecting the suitable dimension reduction algorithm, such as Principal Component Analysis (PCA), t-distribution neighborhood embedding (t-SNE) and the like. Parameters of the dimension reduction algorithm, such as the number of principal components, the learning rate of t-SNE and the like, are set according to actual conditions. And applying a selected dimension reduction algorithm to the standard financial information extraction data to generate financial information extraction dimension reduction data. The financial information extraction reduced-dimension data may be converted into a vector form using techniques such as Word embedding. This may be achieved by a pre-trained embedded model or a custom vectorization method. And applying the selected vector conversion method to extract and convert the reduced data of the financial information into a vector form to obtain the financial information vector conversion data. An appropriate embeddable deep learning model, such as Autoencoders, variational Autoencoders (VAE), etc., is selected to be suitable for the model that handles the multi-modal information embedding task. And the financial information vector conversion data is used as input data, so that the matching of the data format and the model requirement is ensured. A deep learning model is constructed and trained using the financial information vector conversion data. This may include encoder and decoder structures for learning the representation and restoration of data. The embedded multi-modal information data is obtained through the encoder portion of the model, representing a low-dimensional embedded space of the input data.

in the embodiment of the invention, by selecting an appropriate multi-modal information alignment method, the method may involve unified representation of information from different modalities (such as text, images, numerical values and the like). Common methods include Canonical Correlation Analysis (CCA), and the like. And taking the embedded multi-mode information data as input data to ensure that the data format is matched with the requirements of an alignment method. And aligning the embedded multi-modal information data by applying the selected multi-modal information alignment method to generate embedded multi-modal information alignment data. Selecting an appropriate data fusion method may involve integrating data from different sources into a consistent data set. The method may include simple stitching, weighted fusion, etc. And taking the embedded multi-mode information alignment data as input data to ensure that the data format is matched with the requirements of a fusion method. And fusing the embedded multi-mode information alignment data by applying the selected data fusion method to generate multi-mode fusion data.

In the embodiment of the invention, the data format, standardization and preprocessing work are ensured to be correctly processed by preparing the multi-mode fusion data set so as to input a deep learning model, and the proper deep learning model, such as a Convolutional Neural Network (CNN), a cyclic neural network (RNN), an automatic encoder (Autoencoder) and the like, is selected according to the task requirement so as to extract the characteristics, and the multi-mode fusion data is input into the selected deep learning model to extract the characteristics. These features may be activation values of the model middle layer, feature maps of the convolutional layer output, etc. And applying the features extracted from the deep learning model to heterogeneous multi-modal information fusion optimization by using a transfer learning method. The goal of transfer learning is to transfer knowledge from one domain to another to improve performance or speed up learning. And selecting a proper heterogeneous multi-mode information fusion optimization method. This may include methods of joint training, multimodal feature fusion, attention mechanisms, etc. for fusing feature information from different sources. And constructing an optimized model, and initializing the model or adjusting partial parameters by using the characteristics obtained by transfer learning. And then, fusion optimization is carried out on the deep learning characteristic data by using an optimization method. And evaluating the optimized model to check whether the performance of the model is improved. The model architecture or optimization method can be tuned to further improve performance if desired.

Preferably, step S3 comprises the steps of:

step S31: rule extraction is carried out on the multimode fusion optimization data, and multimode fusion rule data are generated; performing code conversion on the multi-mode fusion rule data based on UTF codes to generate coded financial rule data;

step S32: initializing a self-organizing network according to the encoded financial rule data to generate self-organizing network structure data and self-organizing initial parameter data; performing network mapping on the self-organizing network structure data through self-organizing initial parameter data to generate initial mapping network data; performing network training on the initial mapping network data based on a self-organizing learning algorithm to generate a self-organizing mapping network;

step S33: performing meta learning model training on the encoded financial rule data by using a meta learning technology to generate a meta learning training model; carrying out model network evolution on the self-organizing map network and the meta-learning training model, thereby generating an evolving financial rule network;

step S34: carrying out adaptability evaluation on the evolution financial rule network by utilizing a dynamic adaptability evaluation formula to generate an adaptability evaluation coefficient; comparing the adaptability evaluation coefficient with a preset adaptability evaluation threshold, and generating rule adaptability evaluation data when the adaptability evaluation coefficient is larger than or equal to the adaptability evaluation threshold; and when the adaptability evaluation coefficient is smaller than the adaptability evaluation threshold value, performing performance optimization on the evolution financial rule network until rule adaptability evaluation data are output, wherein the performance optimization comprises data resampling, network structure adjustment and continuous learning iteration.

The invention extracts rules from the multi-mode fusion optimization data. This may involve extracting rules from patterns in the data using machine learning techniques, rules engines, or other methods. And carrying out UTF coding on the rule data to ensure correct coding of the rule data. Encoding the financial rule data may be to meet specific data formats or system requirements, and may include converting the rule data to a specific encoding format. The ad hoc network is initialized using the encoded financial rule data. This may include determining the structure of the network, setting initial parameters, etc., mapping the ad hoc network structure with ad hoc initial parameter data, and then training the initial mapped network data using an ad hoc learning algorithm to generate an ad hoc mapped network. Self-organizing learning is commonly used for unsupervised learning, allowing the network to learn the structure and pattern of data itself. And training the encoded financial rule data by using a meta-learning technology to generate a meta-learning training model. Meta-learning aims to enable a model to adapt quickly to new tasks, based on previously learned knowledge. And evolving the self-organizing map network and the meta-learning model to generate an evolving financial rule network. This may involve combining features or parameters of both to get a more powerful model. And evaluating the evolution financial rule network by using a dynamic adaptability evaluation formula to generate an adaptability evaluation coefficient. This evaluation may involve the performance of the network on certain tasks or data. And if the adaptability evaluation coefficient is lower than the threshold value, performing performance optimization on the evolution financial rule network. This includes data resampling, network structure adjustment, and continuous learning iterations to improve the adaptability and performance of the network.

As an example of the present invention, referring to fig. 3, the step S3 in this example includes:

in the embodiment of the invention, the data of different types, such as text, image, audio and the like, are ensured to be contained by collecting the multi-mode fusion optimization data. Cleaning data to remove noise, missing values or abnormal values, formulating rule extraction algorithms for different data types by using Natural Language Processing (NLP) technology, image processing technology and the like, performing supervised learning by using tagged data, automatically extracting rules by using a training model, determining to use UTF codes as a standard, usually UTF-8, and using corresponding programming languages or tools to ensure that multi-mode fusion rule data is correctly encoded into UTF-8. According to the requirements of a financial system or application, the multi-mode fusion rule data are converted into financial rule data conforming to specific codes, so that the converted data are ensured to be correct in format, and key information is not lost.

In the embodiment of the invention, the structure of the self-organizing network is defined by using the encoded financial rule data generated in the previous step, including the number of nodes, the topological structure and the like, and the weight and other parameters of the self-organizing network are initialized, so that a random value or other heuristic methods can be used. And inputting the encoded financial rule data into the self-organizing network, and mapping through the initial weight to obtain initial mapping network data. A neighborhood function is designed that determines which nodes are affected in each training iteration, a learning rate function is designed that determines the degree of update of each node weight, and node weights are iteratively adjusted using a self-organizing learning algorithm, such as SOM, to map adjacent rules to adjacent nodes while maintaining topology, creating a self-organizing map network.

in the embodiment of the invention, a proper Meta-Learning algorithm, such as Model-independent Meta-Learning (MAML), reptile, meta-SGD and the like, is selected according to task requirements, a Meta-Learning Model framework suitable for tasks is designed and constructed, and the encoded financial rule data is used for training of a Meta-Learning Model in consideration of a network structure, a loss function and the like. The meta-learning aims at learning the rapid adaptability to different financial rule changes, and combines a self-organizing map network and a meta-learning model which are already trained. This may involve connecting two models or establishing an additional federated network between them. And utilizing the integrated model to carry out evolution or adaptability adjustment on the financial rule data so as to generate an evolution financial rule network, and training or fine-tuning the integrated network so as to adapt to new financial rule change or task requirements.

In the embodiment of the invention, by designing a dynamic adaptability evaluation formula, the formula should consider indexes in the aspects of performance, accuracy, generalization capability and the like of the evolution financial rule network. The formula may be based on the specific problem and the characteristics of the data set, and the defined adaptive evaluation formula is used to evaluate the evolving financial rule network to generate an adaptive evaluation coefficient. An adaptive evaluation threshold is preset, which may be determined according to task requirements, performance criteria, etc., and the generated adaptive evaluation coefficients are compared with the preset adaptive evaluation threshold. If the fitness evaluation coefficient is greater than or equal to the fitness evaluation threshold, rule fitness evaluation data is generated. If the adaptability evaluation coefficient is smaller than the threshold value, resampling the data can be considered, new and more representative data can be introduced, the structure of the evolution finance rule network can be considered to be adjusted, the adjustment of the layer number, the node number, the activation function and the like can be possibly included, a continuous learning mechanism is introduced, and iterative learning is performed on the network to adapt to the new data and rules.

Preferably, the dynamic adaptability evaluation formula in step S34 is specifically as follows:

wherein A is represented as an adaptability evaluation coefficient, t is represented as an adaptability evaluation time point, t ₁ Expressed as the initial time of the fitness evaluation, t ₂ Expressed as the end time of the fitness evaluation, x _i (t) represents the value of the ith feature of the evolution financial rule network at time t, n represents the number of evaluation features of the evolution financial rule network, mu _i Represented as the mean, σ, of the ith feature in the evolving financial rule network _i Expressed as the ith standard deviation parameter, alpha, in the evolving financial rule network _i Index parameter, beta, expressed as the ith feature in evolving financial rule network _i Weight parameter, gamma, expressed as the ith feature in evolving financial rule network _i Scaling factor, delta, expressed as the ith feature in evolving financial rule network _i Expressed as the ith decay rate in the evolving financial rule network, μ is expressed as the dynamic fitness evaluation anomaly adjustment value.

The invention constructs a dynamic adaptability evaluation formula by using a time period [ t ] ₁ ,t ₂ ]And comprehensively evaluating the characteristic value change in the model, and calculating an adaptability evaluation coefficient A. For each feature x _i (t) by standardAnd (3) carrying out combination of parameters such as a mean value subtracting and dividing by a standard deviation, exponential transformation, weight adjustment, scaling factors, attenuation rates and the like, carrying out weighting treatment on the parameters, finally obtaining A through integral summation, and adding the dynamic adaptability evaluation abnormal adjustment value mu to obtain a final adaptability evaluation coefficient. The correlation between the time point and the parameters is evaluated according to the adaptability to form a functional relation:

The comprehensive consideration of the network is ensured by adjusting the value of the ith feature of the evolution financial rule network in the formula at time t and comprehensively evaluating all the features. The number of the evaluation features of the evolution financial rule network covers a plurality of features through adaptive evaluation, so that the evaluation accuracy of the overall performance of the network is improved. The mean value of the ith feature in the evolution financial rule network is subtracted, so that the integral deviation of the feature is eliminated, and the evaluation is more accurate. The standard deviation of the ith feature in the evolution financial rule network enables the influence of features of different orders on evaluation to be relatively balanced by normalizing the feature values. And carrying out nonlinear adjustment on the characteristic value by index transformation on the index parameter of the ith characteristic in the evolution financial rule network so as to adapt to different sensitivity degrees of different characteristics to adaptability evaluation. The weight parameters of the ith feature in the evolution financial rule network are subjected to weight adjustment to weight importance of different features so as to ensure more accurate evaluation of the important features. The scaling factors of the ith feature in the evolution financial rule network enable the ith feature to have similar magnitude in the evaluation process by scaling the feature values, so that the excessive influence of some features on the evaluation result due to magnitude difference is avoided. The attenuation rate of the ith feature in the evolving financial rule network gradually weakens the contribution of the past feature value to the evaluation through the attenuation feature value, and the latest feature value is more focused to adapt to the dynamic change of the network performance. The outlier mu is evaluated by dynamic adaptation for correcting errors and deviations due to the complexity and non-idealities of the actual system. The method can correct the difference between theoretical assumption in the formula and an actual system, improves the accuracy and reliability of dynamic adaptability evaluation, generates an adaptability evaluation coefficient A more accurately, and simultaneously adjusts parameters such as the evaluation feature quantity of an evolution financial rule network in a rating process in the formula, the mean value of an ith feature in the evolution financial rule network and the like according to actual conditions, thereby adapting to different dynamic adaptability evaluation scenes and improving the applicability and flexibility of the algorithm. When the dynamic adaptability evaluation formula conventional in the art is used, the adaptability evaluation coefficient can be obtained, and the adaptability evaluation coefficient can be calculated more accurately by applying the dynamic adaptability evaluation formula provided by the invention. The formula utilizes the combined action of a plurality of parameters to comprehensively evaluate the characteristics of the evolution financial rule network. Through operations such as standardization, nonlinear transformation, weight adjustment, scaling, attenuation and the like, the evaluation is more comprehensive and accurate, and the method can adapt to the differences and dynamic changes of different characteristics. The setting of each parameter can be adjusted according to specific application scenes and requirements so as to obtain more accurate and reliable adaptability evaluation results.

Preferably, step S4 comprises the steps of:

step S41: constructing a virtual environment for the rule adaptability evaluation data by utilizing a virtual reality technology, and generating virtual financial environment frame data, wherein the virtual financial environment frame data comprises simulated market data, transaction scene data and virtual financial asset data;

step S42: performing transaction simulation generation by using the simulated market data and the virtual financial asset data to generate virtual transaction order data; integrating the transaction scene data and the virtual transaction order data in a virtual environment, thereby generating simulated transaction data;

step S43: embedding a preset intelligent contract into virtual financial environment frame data for automatic rule execution to generate intelligent contract integration data;

step S44: performing risk event simulation on the simulated transaction data through the intelligent contract integrated data to generate risk event simulation data; performing financial behavior pattern analysis on the intelligent contract integrated data to generate financial system behavior analysis data; performing financial risk assessment on the risk event simulation data to generate financial system risk analysis data; carrying out transaction efficiency analysis on the simulated transaction data to generate financial system transaction analysis data;

Step S45: data integration is carried out on the financial system behavior analysis data, the financial system risk analysis data and the financial system transaction analysis data, and a virtual financial ecological analysis data set is generated; and carrying out environment optimization on the virtual financial environment framework data according to the virtual financial ecological analysis data set so as to generate virtual financial world data.

The invention can simulate the financial market environment more truly through the construction of the virtual environment, comprising market data, transaction scenes and various virtual financial assets, and improves the authenticity and fidelity of rule adaptability evaluation. By simulating the generation of transaction data, the transaction performance of the evolving financial rule network in the virtual environment can be evaluated, and data support is provided for the actual adaptability of the rules. The intelligent contracts are embedded into the virtual financial environment, so that automatic rule execution is realized, intelligent contract integrated data are generated, and automatic execution efficiency of the rules is improved. The risk event simulation is carried out on the simulated transaction data through the intelligent contract integrated data, so that the risk level of the financial system in different environments can be estimated, the intelligent contract integrated data is analyzed, different behavioral modes in the financial system can be insight, and the decision making process of participants and the evolution of markets can be understood. Efficiency analysis of the simulated transaction data may evaluate transaction efficiency and performance in the financial system. And integrating the different analysis data into a virtual financial ecological analysis data set to provide comprehensive evaluation for the overall performance of the system. And (3) performing environment optimization according to the data set, so that the virtual financial environment is closer to the actual market, and the reliability of evaluation is improved. In combination with the virtual financial ecological analysis dataset, virtual financial world data is generated, which is a complete description of the entire virtual financial ecological system, contributing to deeper system understanding and decision support.

As an example of the present invention, referring to fig. 4, the step S4 includes, in this example:

in embodiments of the present invention, by selecting an appropriate virtual reality technology, such as a Virtual Reality (VR) or Augmented Reality (AR) technology. This will depend on specific requirements, such as whether a fully immersed virtual environment is required. Historical data of the financial market is utilized or simulation data conforming to actual market trends is generated. This may involve simulations in terms of price, volume of transactions, etc. Different transaction scenarios are simulated, including situations of normal transaction, abnormal transaction, market fluctuation, and the like. This helps to assess the responsiveness and adaptability of the system in various situations. Virtual assets are created, including stocks, bonds, derivatives, and the like. The characteristics of these assets should be consistent with the characteristics of the real financial market. The generated simulated market data, transaction scenario data, and virtual financial asset data are integrated into a virtual financial environment framework. Consistency and correlation between data are ensured to create an organic virtual financial ecosystem. If virtual reality technology is employed, the user interface and interaction means are designed to ensure that the user is able to interact and view the simulated financial data effectively in the virtual environment. By utilizing the advantages of the virtual reality technology, an effective data visualization mode is designed, so that a user can intuitively understand various aspects of the virtual financial environment, including market trends, transaction scenes and the like.

in the embodiment of the invention, the transaction behavior is simulated by selecting or designing a proper transaction algorithm. This may include algorithms based on technical analytics, basic analytics, or other financial policies, utilizing simulated market data and virtual financial asset data, generating a virtual trade order using selected trade algorithms. The order should include information on the direction of the purchase, the amount traded, the price of the trade, etc. The transaction scenario data generated previously is integrated into the virtual environment. This may include data in terms of market fluctuations, news events, trading activity, etc., combining the generated virtual trade order data with the trade scene data to simulate the occurrence of an actual trade in a virtual environment. Ensuring that the generation and execution of orders matches the simulated market context. The execution process of the order is simulated in the virtual environment, including the situations of bargain, partial bargain or no bargain, and the like, and the state of the virtual market is updated according to the executed virtual bargain order, including price change, transaction amount change and the like. Recording detailed information of each simulated transaction, including transaction time, price, quantity, transaction state and the like, creating a simulated transaction log, and recording key events and data in the whole transaction process. Ensuring that the generated simulated transaction data is consistent with expectations and conforms to the settings of the virtual financial environment. Parameters of the simulation environment are adjusted as needed to improve the realism and fidelity of the simulation.

in the embodiment of the invention, by defining the purpose of intelligent contracts, such as executing specific financial transactions, automatic settlement and the like, the intelligent contract platform (such as Ethernet, hyperledger Fabric and the like) is used for writing contract codes, so that the contracts are ensured to contain necessary rules and logic, and the virtual financial environment framework compatible with the intelligent contract platform is ensured to be selected for smooth integration. Embedding the written intelligent contract into the data model of the virtual financial environment framework ensures that interaction with the simulated transaction data can be performed, and determining the triggering condition of the intelligent contract, namely when to start intelligent contract execution. This may involve specific market conditions, transaction events, or other predefined trigger conditions, when met, automatically executing embedded smart contract rules. This may include generating transaction instructions, updating account balances, triggering settlement, and the like. In the intelligent contract executing process, the executing condition of each step is recorded in detail, wherein the executing condition comprises input data, executing results and the like, and intelligent contract integrated data comprising contract executing time, executing results, triggering conditions and the like is generated according to the executing process.

in the embodiment of the invention, by determining risk event scenes to be simulated, including market fluctuation, technical faults, transaction errors and the like, the execution condition of intelligent contracts under different risk events is simulated by utilizing intelligent contract integration data, the influence of each simulation event, including the execution result of the intelligent contracts, the change of transaction instructions and the like, is recorded, and the simulation result is organized into risk event simulation data for subsequent financial risk assessment. Key features, such as transaction frequency, transaction amount, execution time, etc., are extracted from the smart contract integrated data, which is behavioral pattern analyzed using data analysis and machine learning algorithms to identify potential financial behavioral patterns. And (3) sorting the analysis result into financial system behavior analysis data, including information such as identified behavior patterns, abnormal behaviors and the like. Integrating the risk event simulation data with the financial behavior analysis data to comprehensively consider financial behavior modes under different risk events, evaluating the integrated data by using a risk evaluation model, quantifying the influence of different risks on a financial system, and sorting the evaluation result into financial system risk analysis data including indexes such as risk level, potential loss and the like. Key features such as transaction execution time, transaction cost, etc. are extracted from the simulated transaction data, the simulated transaction data is efficiently analyzed using an appropriate model (e.g., cost-benefit analysis), and the analysis results are consolidated into financial system transaction analysis data, including information on transaction execution efficiency, cost-benefit, etc.

In the embodiment of the invention, the financial system behavior analysis data, the financial system risk analysis data and the financial system transaction analysis data are imported into a unified data platform or database to ensure that the formats of different data sources are consistent, and necessary data cleaning and standardization are performed so as to facilitate subsequent integration and analysis, and the association relationship between the data is established to ensure that information with different dimensionalities can be comprehensively considered in subsequent analysis. The method comprises the steps of integrating behavior analysis data, risk analysis data and transaction analysis data to form a comprehensive virtual financial ecological analysis data set, defining association rules among the data, ensuring that the data set is reasonable and reflects characteristics of a real financial system, introducing a certain degree of randomness to simulate uncertainty and change in a real financial market, determining an optimization target of virtual financial environment frame data, such as improving transaction efficiency, reducing risk level and the like, adjusting parameters and rules in the virtual financial environment frame according to the result of the analysis data set to enable the parameters and rules to be more in line with the optimization target, simulating financial system behaviors and risk conditions in an optimized environment by using the virtual financial ecological analysis data set, integrating the optimized virtual financial environment frame data with the virtual financial ecological analysis data set to form virtual financial world data, recording the financial system behaviors, risk level, transaction efficiency and other data in the optimized environment, and finishing simulation results into final virtual financial world data, wherein the virtual financial world data can be used for further research, testing and decision support.

Preferably, step S42 includes the steps of:

step S421: generating a trade order based on the simulated market data and the virtual financial asset data to obtain virtual trade order data;

step S422: extracting order features of the virtual trade order data to generate trade order feature data; carrying out transaction order identification detection on the virtual transaction order data through the transaction order feature data, and carrying out abnormal blocking on the initial transaction order data when the transaction order identification detection result is false; when the transaction order identification detection result is determined to be true, carrying out transaction matching simulation on the initial transaction order data to generate transaction matching data;

step S423: price calibration is carried out on transaction matching data through a quantum financial knowledge graph, and calibrated transaction price data is generated; performing order simulation execution based on the initial transaction order data and the calibration transaction price data to generate transaction execution simulation data; carrying out transaction stability coefficient analysis on the transaction scene data by utilizing a virtual transaction stability detection formula to generate a virtual transaction stability coefficient;

step S424: and carrying out virtual environment integration on the transaction execution simulation data according to the virtual transaction stability coefficient, thereby generating simulation transaction data.

The invention can simulate various market conditions including price fluctuation, supply and demand change and the like by using the simulated market data and the virtual financial asset data, provide reality and diversity for subsequent trade order generation, generate virtual trade orders based on the simulated data, reflect the trade behaviors of investors in different market situations and provide basic data for subsequent analysis. By extracting the characteristics of the virtual trade order, important parameters of the order, such as trade amount, price and the like, can be known, key information is provided for subsequent analysis and simulation, and abnormal orders can be identified by detecting order identification, so that the safety of a trade system is improved. And matching simulation is carried out on legal orders, transaction matching data are generated, and the transaction process in the real market is simulated, so that the situation of transaction execution is known. And carrying out price calibration on transaction matching data by utilizing a quantum financial knowledge graph, improving the accuracy and the authenticity of simulation data, carrying out order simulation execution based on the calibrated price data, generating transaction execution simulation data, helping to know the possible situation in the order execution process, analyzing transaction scene data by utilizing a virtual transaction stability detection formula, generating a virtual transaction stability coefficient, and helping to evaluate the stability of a transaction system. And carrying out virtual environment integration on the transaction execution simulation data according to the virtual transaction stability coefficient, ensuring the rationality and the authenticity of the simulation environment, and finally generating simulation transaction data which can be used for system testing, algorithm optimization and research and development and verification of transaction strategies.

In the embodiment of the invention, the simulation market data and the virtual financial asset data are generated by using a proper simulation algorithm and model, factors such as market trend, volatility, asset types and the like are considered, and based on the simulation data, the virtual transaction order is created by using a generation algorithm and comprises information such as order type, price, quantity and the like. The method comprises the steps of extracting key features such as price, quantity and transaction direction from virtual transaction order data by using a feature extraction algorithm, detecting order identification by using transaction order feature data, judging whether an order is legal or not by using a corresponding algorithm, taking corresponding measures such as abnormal blocking of the order when an order identification detection result is false, ensuring system safety, and simulating an order matching process by using a transaction matching algorithm when the order identification detection result is true, so as to generate transaction matching data. And carrying out price calibration on the transaction matching data by utilizing the quantum financial knowledge graph so as to improve the accuracy of the price data, and simulating an order execution process by using the initial transaction order data and the calibrated price data through a transaction execution simulation algorithm to generate transaction execution simulation data. And analyzing the transaction scene data by using a virtual transaction stability detection formula, and calculating a virtual transaction stability coefficient. According to the virtual transaction stability coefficient, the transaction execution simulation data are integrated in a virtual environment, so that the stability and the authenticity of the simulation environment are ensured, and the simulation transaction data are generated according to the integrated data, so that the system test, algorithm optimization and transaction strategy verification can be realized.

Preferably, the virtual transaction stability detection formula in step S423 is specifically as follows:

wherein V is expressed as a virtual transaction stability factor, T ₁ Expressed as time of transactionStart time, T ₂ Expressed as the end time of the transaction time, n expressed as the number of assets in the transaction, w _k Weights expressed as kth asset, P _k Expressed as the purchase price of the kth asset, Q _k Represented as the sell price of the kth asset, B _j Weights expressed as j-th index, R _j Buying value expressed as j-th index, S _j The sell value expressed as the j-th index, c as the decay factor of time, g as the number of operations, y as the index in operation, T as the virtual trade time point,represented as virtual transaction stability detection anomaly correction.

The invention constructs a virtual transaction stability detection formula by applying a virtual transaction stability detection formula to a transaction time period [ T ] ₁ ,T ₂ ]And (5) comprehensively evaluating the price and index value of the assets and time in the virtual transaction system, and calculating a virtual transaction stability coefficient V. Each asset and index is weighted by a combination of weights and buy/sell values. And finally, obtaining a final virtual transaction stability coefficient V through integration, exponential transformation and addition of an abnormal correction quantity. The correlation between the starting time of the transaction time and the parameters forms a functional relation:

By considering the purchase price and the sell price of multiple assets, the assessment of the entire trade combination is achieved. The weight of the kth asset is adjusted by the weight, the importance of different assets is weighted, the important assets are accurately evaluated, the buying price of the kth asset is considered, and the income condition and market performance of the asset can be evaluated. The sell price of the kth asset can evaluate the value change and market performance of the asset by considering the sell price of the asset, and the trade stability by considering the weights of the multiple indicators and the buy/sell values.The weight of the j-th index. And the importance of different indexes is weighted through weight adjustment, so that the more accurate evaluation of the important indexes is ensured. The buying value of the j-th index can evaluate the market performance and trade signal of the index through the buying value of the index. The sales value of the j-th index, through which the market performance and trading signal of the index can be evaluated. By time decay, the contribution of past transaction data to stability assessment is gradually weakened, and the latest transaction data is more focused. By taking the number of operations into consideration, the transaction frequency and the stability of the transaction strategy can be evaluated. By considering the index, the income situation and market performance of the trading strategy can be evaluated. By taking into account the transaction time, the transaction stability at different points in time can be assessed. Stable detection of abnormal correction through virtual transaction For correcting errors and deviations due to the complexity and non-idealities of the actual system. The method can correct the difference between theoretical assumption in the formula and an actual system, improves the accuracy and reliability of virtual transaction stability detection, generates a virtual transaction stability coefficient V more accurately, and simultaneously adjusts the parameters such as the weight of the jth index, the buying value of the jth index and the like in the formula according to actual conditions, thereby adapting to different virtual transaction stability detection scenes and improving the applicability and flexibility of the algorithm. When the virtual transaction stability detection formula conventional in the art is used, the virtual transaction stability coefficient can be obtained, and the virtual transaction stability coefficient can be calculated more accurately by applying the virtual transaction stability detection formula provided by the invention.

Preferably, step S5 comprises the steps of:

step S51: performing distributed node construction on the multi-mode fusion optimization data by utilizing edge calculation to obtain financial risk distributed nodes;

step S52: autonomous learning rule distribution is carried out on the financial risk distribution nodes based on the virtual financial world data, and node rule learning data is generated;

step S53: the node rule learning data are subjected to rule fusion communication through the financial risk distribution nodes, and a financial risk learning rule base is generated;

Step S54: constructing a dynamic rule engine according to the financial risk learning library to generate a dynamic financial rule engine; monitoring data acquisition is carried out on the dynamic financial rule engine to obtain financial rule engine monitoring data; and utilizing the financial rule engine monitoring data to dynamically update the dynamic financial rule engine to generate the real-time optimized financial rule engine.

According to the invention, the financial risk node can be closer to the data source through edge calculation, the data transmission delay is reduced, the performance is improved, the edge calculation resources are utilized, the calculation resources can be more effectively managed and utilized, and the efficiency of the whole system is improved. The autonomous learning rule distribution can generate node rule learning data in a personalized way according to the virtual financial world data, so that the adaptability and learning capacity of the system are improved, and the learning rule distribution based on the actual data can be fed back in time and adapt to the change of the financial market. The rule learning data are subjected to rule fusion communication through the distributed nodes, so that knowledge of a plurality of nodes can be integrated, and a more comprehensive and accurate financial risk learning rule base is generated. The dynamic rule engine can make real-time decisions according to real-time data, the coping capability of the system to the rapidly-changing financial market is enhanced, the engine automatically adjusts rules according to the content of the financial risk learning library, the self-adaptability and flexibility of the system are maintained, the rule engine is monitored in real time by utilizing monitoring data, performance bottlenecks can be found and optimized, efficient operation of the engine is ensured, the financial rule engine can be updated in real time to cope with new financial risks and market changes, and the stability and safety of the system are improved. Through the distributed nodes and the learning rule base, the system can more comprehensively manage financial risks and reduce potential risk exposure. The dynamic rule engine and the real-time optimization mechanism make the system more efficient and can quickly make decisions in a changing financial environment.

In the embodiment of the invention, by designing the edge computing architecture, determining how to execute tasks on edge nodes, ensuring communication and cooperation among the nodes, developing an algorithm or a method, fusing data of different modes to acquire more comprehensive and comprehensive financial risk information, and deploying financial risk distributed nodes on the edge computing nodes, so as to ensure that the nodes can effectively process and analyze multi-mode data. Virtual financial world data is created, real financial environments including various trade, market fluctuation and the like are simulated, and algorithms are designed, so that financial risk distribution nodes can learn rules autonomously. This may include supervised learning, reinforcement learning, or other machine learning techniques, developing mechanisms that ensure that learned rules can be efficiently distributed to the various distributed nodes so that they can be executed locally. The algorithm is developed, so that the financial risk distributed nodes can communicate and merge the learned rules respectively to generate a comprehensive rule base, and the distributed communication protocol is designed to ensure that the learned rules can be communicated between the nodes efficiently. Based on the rule base, the dynamic rule engine is designed so that the dynamic rule engine can make decisions according to real-time data, a monitoring system is deployed, and the performance and decision data of the dynamic financial rule engine are collected. And updating the dynamic financial rule engine in real time according to the monitoring data so as to adapt to new market conditions and risks.

Preferably, step S6 comprises the steps of:

step S61: sampling the virtual transaction environment of the virtual financial world data to generate financial transaction environment state data; intelligent decision exploration is carried out on the financial transaction environment state data based on a support vector machine algorithm, and decision exploration strategy data is generated;

step S62: carrying out decision simulation on the decision exploration strategy data to obtain decision exploration simulation data; carrying out decision strategy optimization on the decision exploration simulation data by utilizing the decision exploration simulation data to generate optimized decision data; carrying out population initialization coding on the optimized decision data by utilizing an evolution financial rule network to generate initial population data;

step S63: carrying out evolutionary iteration on the initial population data through a genetic algorithm to generate iterative population data; performing cross selection on the iterative population data by utilizing the rule adaptability evaluation data to generate evolution population data; and importing the evolved population data into a real-time optimized financial rule engine to carry out credit risk decision, thereby generating a dynamic financial credit risk model.

According to the invention, by sampling virtual financial world data, different transaction scenes are simulated to generate diversified financial transaction environment state data, and a support vector machine (Support Vector Machine, SVM) algorithm is applied to analyze the financial transaction environment state data to generate intelligent decision exploration strategy data. The SVM can be used for classification and regression analysis, is suitable for processing complex nonlinear relations, utilizes the generated decision exploration strategy data to simulate the decision process under different situations to obtain decision exploration simulated data, optimizes the decision exploration simulated data, can adopt various optimization algorithms such as genetic algorithm, simulated annealing and the like to obtain more effective decision strategies, utilizes an evolution algorithm, possibly a genetic algorithm, and performs population initialization coding on the optimized decision data to generate initial population data. This can be seen as a method of heuristic searching, evolving to find better decision strategies. The method comprises the steps of carrying out evolutionary iteration on initial population data by using a genetic algorithm to gradually improve decision strategies, generating iterative population data, carrying out rule adaptability evaluation on the iterative population data, evaluating the iterative population data possibly through historical data or simulation data to select individuals which are most suitable for the current financial environment, selecting individuals with better adaptability through a cross selection mechanism to form evolutionary population data, and importing the evolutionary population data into a real-time optimized financial rule engine for actual credit risk decision. This may include applying the generated dynamic financial credit risk model to the actual transaction environment.

In the embodiment of the invention, by simulating and sampling the data of the virtual financial world, historical data, simulated data or simulation data can be utilized to acquire state information of various financial transaction environments, and a support vector machine algorithm is used for analyzing and mining the state data of the financial transaction environments to find patterns, associations or rules and generate intelligent decision exploration strategy data. This may help the model better understand and predict future financial environments. And performing simulation by using the generated decision exploration strategy data. This may involve modeling different decision paths, transaction scenarios, or risk management strategies to generate decision exploration modeling data. Based on the simulation data, an optimization algorithm is performed to improve the decision strategy, for example, using an evolution algorithm (e.g., genetic algorithm) to optimize to obtain a decision scheme that better meets the intended objectives. The initial population data is iterated and evolved according to a genetic algorithm to generate new populations which may represent better decision rules or strategies, the iterated populations are evaluated, individuals adapting to the current financial environment are selected, and rules or individuals with higher adaptability are reserved and exchanged through cross selection to form evolved population data. And importing the population data subjected to evolution optimization into a real-time optimization financial rule engine for actual credit risk decision. This can be used to predict, evaluate and manage risk situations in actual financial transactions, thereby generating a dynamic financial credit risk model.

The present embodiments are, therefore, to be considered in all respects as illustrative and not restrictive, the scope of the invention being indicated by the appended claims rather than by the foregoing description, and all changes which come within the meaning and range of equivalency of the claims are therefore intended to be embraced therein.

The foregoing is only a specific embodiment of the invention to enable those skilled in the art to understand or practice the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the invention. Thus, the present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. The method for constructing the dynamic financial credit risk model based on the rule engine is characterized by comprising the following steps of:

2. The method for constructing a dynamic financial credit risk model based on a rule engine according to claim 1, wherein the step S1 includes the steps of:

step S11: acquiring knowledge data in the financial field;

3. The method for constructing a dynamic financial credit risk model based on a rule engine according to claim 1, wherein the step S2 includes the steps of:

4. The method for constructing a dynamic financial credit risk model based on a rule engine according to claim 1, wherein the step S3 includes the steps of:

5. The method of claim 4, wherein the dynamic adaptability assessment formula in step S34 is as follows:

6. The method for constructing a dynamic financial credit risk model based on a rule engine according to claim 1, wherein the step S4 includes the steps of:

7. The method of claim 6, wherein the step S42 includes the steps of:

8. The method of claim 7, wherein the virtual transaction stability detection formula in step S423 is as follows:

wherein V is expressed as a virtual transaction stability factor, T ₁ Expressed as the start time of the transaction time, T ₂ Expressed as the end time of the transaction time, n expressed as the number of assets in the transaction, w _k Weights expressed as kth asset, P _k Expressed as the purchase price of the kth asset, Q _k Represented as the sell price of the kth asset, B _j Weights expressed as j-th index, R _j Buying value expressed as j-th index, S _j The sell value expressed as the j-th index, c as the decay factor of time, g as the number of operations, y as the index in operation, T as the virtual trade time point,represented as virtual transaction stability detection anomaly correctionAmount of the components.

9. The method for constructing a dynamic financial credit risk model based on a rule engine according to claim 1, wherein step S5 includes the steps of:

10. The method for constructing a dynamic financial credit risk model based on a rule engine according to claim 1, wherein step S6 includes the steps of: