Agent-based financial risk decision-making method, device and equipment and storage medium

Abstract

The application discloses an agent-based financial risk decision method and device, equipment and a storage medium, relates to the field of artificial intelligence, is applied to a terminal carrying a target risk decision agent, and comprises the following steps: obtaining original multi-source heterogeneous data from each target financial data source and performing a data preprocessing operation to obtain corresponding preprocessed data; performing feature extraction processing on the preprocessed data based on a pre-set feature extraction algorithm to obtain corresponding target data, determining a target risk type and a corresponding risk measurement value based on the target data by using the target risk decision agent, determining target knowledge from a financial knowledge data source based on the target risk type and the risk measurement value corresponding to the target data, and fusing all the target knowledge by using a pre-determined target data fusion algorithm to generate a corresponding target decision strategy. The application improves the analysis capability of financial related data, thereby improving the financial risk assessment capability and decision efficiency.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence, and in particular to a method, apparatus, device, and storage medium for financial risk decision-making based on intelligent agents. Background Technology

[0002] Currently, organizations and individuals are facing an ever-increasing volume of data, with intertwined and rapidly changing risk factors. In financial risk analysis applications, traditional manual decision-making or decision-making methods based on single models and limited data are no longer sufficient. These traditional methods often face challenges such as information overload, slow processing speed, strong decision-making subjectivity, difficulty in comprehensively assessing risks, and inability to adapt to rapidly changing environments, resulting in low decision-making efficiency, insufficient accuracy, and even the potential for significant losses.

[0003] In conclusion, improving the ability to analyze financial data in order to enhance financial risk assessment capabilities and decision-making efficiency is an urgent problem to be solved. Summary of the Invention

[0004] In view of this, the purpose of this invention is to provide a method, apparatus, device, and storage medium for financial risk decision-making based on intelligent agents, which improves the ability to analyze financial-related data, thereby enhancing financial risk assessment capabilities and decision-making efficiency. The specific solution is as follows:

[0005] Firstly, this application discloses an agent-based financial risk decision-making method, applied to a terminal equipped with a target risk decision-making agent, comprising:

[0006] The system obtains raw, multi-source heterogeneous data from various target financial data sources through a target data interface, and performs data preprocessing operations on all the multi-source heterogeneous data to obtain corresponding preprocessed data; the target financial data sources include preset private data sources and preset public data sources;

[0007] The preprocessed data is processed by a preset feature extraction algorithm to obtain the corresponding target data, and the target risk decision-making agent is used to determine the target risk type and the corresponding risk measurement value based on the target data.

[0008] The target risk decision-making agent uses the target risk type and risk metric value corresponding to the target data to determine target knowledge from financial knowledge data sources, and uses a pre-determined target data fusion algorithm to fuse all the target knowledge to generate a corresponding target decision-making strategy.

[0009] Optionally, performing data preprocessing operations on all the multi-source heterogeneous data to obtain corresponding preprocessed data includes:

[0010] The multi-source heterogeneous data is cleaned according to preset data cleaning rules to obtain the first processed data;

[0011] Based on the target financial data source and data format corresponding to each of the multi-source heterogeneous data, a preset data integration algorithm is used to perform data transformation on the corresponding first processed data to obtain second processed data;

[0012] The second processing data is standardized based on a preset data standardization algorithm to obtain the corresponding preprocessed data.

[0013] Optionally, the step of performing feature extraction processing on the preprocessed data based on a preset feature extraction algorithm to obtain the corresponding target data includes:

[0014] Based on the data source information features corresponding to each of the target financial data sources, a preset feature extraction algorithm is determined for each of the target financial data sources.

[0015] Based on the preset feature extraction algorithm corresponding to each of the target financial data sources, feature extraction is performed on the corresponding preprocessed data to obtain the initial feature dataset corresponding to each of the target financial data sources;

[0016] The target weights of each initial feature dataset are determined based on the current decision-making context using a target multimodal data fusion model, and a data fusion operation is performed on all the initial feature datasets based on the target weights to obtain the corresponding target data.

[0017] Optionally, determining the target risk type and corresponding risk metric based on the target data includes:

[0018] Based on a preset text analysis algorithm, the unstructured data in the target data is analyzed to determine the target risk type corresponding to the target data;

[0019] Based on the target risk type, a corresponding target risk prediction model is determined, and the target risk prediction model is used to determine the corresponding risk metric value based on the structured data in the target data.

[0020] Optionally, before fusing all the target knowledge using a pre-determined target data fusion algorithm to generate a corresponding target decision strategy, the method further includes:

[0021] Using the target risk decision-making agent, a target data fusion algorithm is determined based on pre-acquired decision data through reinforcement learning;

[0022] The decision data includes decision objectives, decision space, and environmental dynamics; the decision objectives include maximizing expected returns, minimizing expected risks, and balancing returns and risks.

[0023] Optionally, after fusing all the target knowledge using a pre-determined target data fusion algorithm to generate a corresponding target decision strategy, the method further includes:

[0024] The target decision strategy is sent to the strategy receiving end, and the corresponding strategy execution result is obtained;

[0025] A preset strategy evaluation mechanism is used to evaluate the target decision strategy based on the strategy execution result, so as to generate a feedback signal corresponding to the strategy execution result;

[0026] Based on the target decision-making strategy, the corresponding strategy execution results, and the feedback signal, the target risk decision-making agent is updated to obtain the updated target risk decision-making agent.

[0027] Optionally, the agent-based financial risk decision-making method further includes:

[0028] The multi-source heterogeneous data, the target decision strategy corresponding to the multi-source heterogeneous data, the strategy execution result corresponding to the target decision strategy, and the generation process of the target decision strategy are determined as decision process data. The real-time acquired decision process data is sent to the target visualization platform so that the decision process data can be processed by the target visualization platform based on a preset data processing method to obtain processed decision process data. The processed decision process data is then presented based on a preset data presentation format.

[0029] The preset data presentation formats include statistical charts, dashboards, and heat maps.

[0030] Secondly, this application discloses a financial risk decision-making device based on an intelligent agent, applied to a terminal equipped with a target risk decision-making intelligent agent, comprising:

[0031] The data preprocessing module is used to obtain raw multi-source heterogeneous data from various target financial data sources through the target data interface, and to perform data preprocessing operations on all the multi-source heterogeneous data to obtain corresponding preprocessed data; the target financial data sources include preset private data sources and preset public data sources;

[0032] The risk analysis module is used to perform feature extraction processing on the preprocessed data based on a preset feature extraction algorithm to obtain the corresponding target data, and to use the target risk decision-making agent to determine the target risk type and the corresponding risk measurement value based on the target data.

[0033] The strategy generation module is used to utilize the target risk decision-making agent to determine target knowledge from the financial knowledge data source based on the target risk type and the risk measurement value corresponding to the target data, and to use a pre-determined target data fusion algorithm to fuse all the target knowledge to generate a corresponding target decision-making strategy.

[0034] Thirdly, this application discloses an electronic device, including:

[0035] Memory, used to store computer programs;

[0036] A processor is used to execute the computer program to implement the aforementioned agent-based financial risk decision-making method.

[0037] Fourthly, this application discloses a computer-readable storage medium for storing a computer program, wherein the computer program, when executed by a processor, implements the aforementioned agent-based financial risk decision-making method.

[0038] In this application, when making financial risk decisions, a terminal equipped with a target risk decision-making intelligent agent acquires raw, multi-source heterogeneous data from various target financial data sources through a target data interface, and performs data preprocessing operations on all the multi-source heterogeneous data to obtain corresponding preprocessed data. The target financial data sources include preset private data sources and preset public data sources. Based on a preset feature extraction algorithm, feature extraction processing is performed on the preprocessed data to obtain corresponding target data. The target risk decision-making intelligent agent then determines the target risk type and corresponding risk measurement value based on the target data. Based on the target risk type and risk measurement value corresponding to the target data, the target risk decision-making intelligent agent determines target knowledge from financial knowledge data sources, and a pre-determined target data fusion algorithm is used to fuse all the target knowledge to generate a corresponding target decision-making strategy. Therefore, this application can efficiently and accurately acquire information from massive and diverse internal and external data sources, and perform deep cleaning, integration, and feature engineering, laying a high-quality data foundation for subsequent risk modeling. Based on this, the target risk decision-making agent is used to analyze the obtained target data, enabling accurate identification, multi-dimensional modeling, and quantitative assessment of various complex risks. This transforms abstract risk concepts into actionable numerical values, providing a solid quantitative basis for decision-making. Furthermore, based on the target risk type and risk measurement value corresponding to the target data, the corresponding target knowledge is determined from financial data sources, providing a clear explanation for the target decision-making strategy generated based on this target knowledge. This improves the transparency, reliability, and traceability of decision-making and promotes effective human-machine collaboration. Attached Figure Description

[0039] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0040] Figure 1 This is a flowchart of a financial risk decision-making method based on intelligent agents disclosed in this application;

[0041] Figure 2 This is a schematic diagram of the structure of a financial risk decision-making device based on an intelligent agent disclosed in this application;

[0042] Figure 3 This is a structural diagram of an electronic device disclosed in this application. Detailed Implementation

[0043] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0044] Currently, organizations and individuals face an ever-increasing volume of data, with intertwined and rapidly changing risk factors. In financial risk analysis applications, traditional manual decision-making or decision-making methods based on single models and limited data are no longer sufficient. These traditional methods often face challenges such as information overload, slow processing speed, strong decision-making subjectivity, difficulty in comprehensively assessing risks, and inability to adapt to rapidly changing environments, leading to low decision-making efficiency, insufficient accuracy, and potentially significant losses. To address these technical problems, this application discloses an agent-based financial risk decision-making method that can improve the ability to analyze financial data, thereby enhancing financial risk assessment capabilities and decision-making efficiency.

[0045] See Figure 1 As shown, this embodiment of the invention discloses a financial risk decision-making method based on an intelligent agent, applied to a terminal equipped with a target risk decision-making intelligent agent, including:

[0046] Step S11: Obtain raw multi-source heterogeneous data from each target financial data source through the target data interface, and perform data preprocessing operations on all the multi-source heterogeneous data to obtain corresponding preprocessed data; the target financial data source includes a preset private data source and a preset public data source.

[0047] In this embodiment, the terminal equipped with the target risk decision-making intelligent agent can efficiently and accurately acquire various types of data from complex and massive amounts of information. Whether this data is structured tables, semi-structured logs, or unstructured text and multimedia content, it can effectively incorporate it. Specifically, it acquires raw, multi-source heterogeneous data from various target financial data sources through target data interfaces. These target financial data sources include preset private data sources and preset public data sources. Preset private data sources can be internal company systems such as transaction databases, customer relationship management systems, and operational logs. Preset public data sources can be external public resources such as financial market data interfaces, news media APIs (Application Programming Interfaces), social media crawlers, and macroeconomic databases. In one specific implementation, a terminal equipped with a target risk decision-making intelligent agent can collect data in real time or in batches from diverse data sources such as relational databases, NoSQL databases, data lakes, file systems, real-time data streams (such as message queues, API interfaces), web crawlers, and third-party data services, in various heterogeneous formats such as structured, semi-structured, and unstructured data. These data sources include financial market data (stock prices, exchange rates, interest rates, commodity prices), macroeconomic indicators (GDP (Gross Domestic Product), CPI (Consumer Price Index), employment data), enterprise operation data (sales, inventory, production logs, equipment status), customer behavior data (transaction records, browsing behavior, complaint records), social media sentiment (user comments, posts, news reports), geographic location information, meteorological data, supply chain upstream and downstream data, and legal and regulatory documents.

[0048] In this embodiment, since the original multi-source heterogeneous data usually contains problems such as missing values, outliers, duplicates, and inconsistent formats, after collecting the multi-source data, it is necessary to perform data preprocessing operations on all multi-source heterogeneous data to obtain corresponding preprocessed data. Specifically, this may include: cleaning the multi-source heterogeneous data based on preset data cleaning rules to obtain first processed data; transforming the corresponding first processed data using a preset data integration algorithm based on the target financial data source and data format corresponding to each multi-source heterogeneous data to obtain second processed data; and standardizing the second processed data based on a preset data standardization algorithm to obtain corresponding preprocessed data.

[0049] In one specific implementation, the aforementioned data preprocessing process may include: firstly, identifying and correcting or deleting missing values, outliers, duplicate records, and erroneous data through rule-based or machine learning methods to clean multi-source heterogeneous data. Then, unifying the data from different sources and formats, converting the raw data into a format suitable for model processing, eliminating data conflicts and redundancy, and ensuring data consistency and integrity, thereby achieving the integration of cleaned multi-source heterogeneous data. For example, entity alignment technology can be used to identify and merge representations of the same entity in different systems, and text data can be segmented, stop words removed, and word vectorized (e.g., Word2Vec, BERT (Bidirectional Encoder Representations from Transformers) embedding). Feature extraction can be performed on image data, and time series data can be smoothed, sampled, or feature derived (e.g., moving average, volatility, trend). Finally, pre-defined data standardization algorithms such as Min-Max normalization or Z-score normalization can be used to normalize and standardize the data obtained after data transformation, thereby scaling data of different dimensions to a specific range and eliminating the impact of dimensional differences on model training.

[0050] Step S12: Perform feature extraction processing on the preprocessed data based on a preset feature extraction algorithm to obtain the corresponding target data, and use the target risk decision-making agent to determine the target risk type and the corresponding risk measurement value based on the target data.

[0051] In this embodiment, after obtaining the preprocessed data, feature extraction processing is performed on the preprocessed data based on a preset feature extraction algorithm to obtain target data with greater representational power and predictive value. In a specific implementation, parallel and targeted feature extraction is first performed on the raw data from different sources to transform the low-value-density raw data into a structured initial feature set. Specifically, performing feature extraction processing on the preprocessed data based on the preset feature extraction algorithm to obtain the corresponding target data may include: determining the preset feature extraction algorithm corresponding to each target financial data source based on the data source information features corresponding to each target financial data source; performing feature extraction on the corresponding preprocessed data based on the preset feature extraction algorithm corresponding to each target financial data source to obtain the initial feature dataset corresponding to each target financial data source; using a target multimodal data fusion model to determine the target weight of each initial feature dataset based on the current decision context, and performing data fusion operation on all initial feature datasets based on each target weight to obtain the corresponding target data.

[0052] Specifically, in the parallel extraction of multimodal features and the construction of the initial feature dataset, for the original transaction data, not only are basic statistical features such as transaction frequency and capital flow extracted, but also dynamic time-series features such as moving averages and volatility of key indicators are constructed; for the original text data, natural language processing technology is used to extract the sentiment tendency of the text, key entities (such as companies and people) and their interrelationships, and transform them into semantic vectors; for historical event data, the impact intensity of specific events (such as policy changes and industry news) is quantified, and event features that take into account the decay of influence over time are constructed. Then, after obtaining the initial feature datasets corresponding to each target financial data source, the target multimodal data fusion model is used to determine the target weights of each initial feature dataset based on the current decision-making context. Based on each target weight, data fusion operations are performed on all initial feature datasets to obtain the corresponding target data. For example, the target multimodal data fusion model can learn a unified, compact feature representation that removes redundant information. The target multimodal data fusion model can dynamically assign fusion weights to features from different sources (such as text being more important than transaction importance) according to the current decision-making context, so as to achieve effective fusion of multi-source heterogeneous data and make the target data generated by fusion sufficient to fully reflect the risk status.

[0053] Understandably, to ensure the efficiency of feature extraction and the timeliness of target data, a collaborative approach using parallel processing, batch processing, and streaming processing mechanisms can be employed to handle these multi-source heterogeneous data. Specifically, batch processing can be used to perform initial training and periodic full updates of the target multimodal data fusion model using a complete dataset within a time window; streaming processing can be used to handle real-time decision-making scenarios, performing instantaneous, incremental feature computation on newly acquired data streams; and parallel processing serves as the underlying computing architecture, providing high-speed computational support for batch and streaming processing tasks to ensure overall efficiency. For example, data such as trading frequency, trading volume, fund flows, and position concentration can be derived from raw trading data; sentiment and topic distribution can be extracted from text data; and the frequency and intensity of events can be extracted from historical events. By employing parallel processing, streaming processing, and batch processing techniques, the timeliness and accuracy of data are ensured, effectively improving the quality of target data used for subsequent analysis.

[0054] In this embodiment, after acquiring the target data, advanced statistical, econometric, machine learning, and operations research methods are applied to utilize a target risk decision-making intelligent agent to identify, model, and quantify the multidimensional risks of the target data. This means determining the target risk type and corresponding risk metric based on the target data, achieving a deep understanding and assessment of the risk. The analysis of the target data involves three stages: first, risk identification, where unstructured data is analyzed using a pre-defined text analysis algorithm to determine the corresponding target risk type; second, risk modeling, where a corresponding target risk prediction model is determined based on the target risk type; and finally, risk quantification, where the target risk prediction model is used to determine the corresponding risk metric based on the structured data within the target data. It is understood that the target risk decision-making intelligent agent can not only assess individual risks but also capture the interactions and transmission mechanisms between risks, providing a multi-layered, multi-granular view of risk. This comprehensive data processing and in-depth risk insight capability allows the target risk decision-making intelligent agent to examine the decision-making environment from a broader perspective, capturing potential subtle signals, thereby significantly improving the accuracy and foresight of risk assessment and avoiding decision-making errors caused by insufficient information or one-sided analysis.

[0055] In one specific implementation, during the risk identification phase, text mining and natural language processing techniques are used to extract potential risk events (such as policy changes, natural disasters, and rumors of market manipulation) from unstructured data. This is combined with anomaly pattern detection in structured data (such as credit score anomalies and sudden changes in trading frequency) to identify various explicit and implicit risks, including operational risk, credit risk, market risk, liquidity risk, compliance risk, reputational risk, and technological risk. During the risk modeling phase, multi-level and multi-granularity risk prediction models are constructed for different risk types. For example, for market risk, historical simulation, parametric methods (such as VaR (Value at Risk) and Expected Shortfall calculations), volatility models (GARCH series models), and Monte Carlo simulations are used to predict asset price fluctuations and potential losses. For credit risk, credit scoring models based on logistic regression, support vector machines, neural networks, or gradient boosting trees are constructed to predict default probabilities. For operational risk, causal models based on event trees, fault tree analysis, or Bayesian networks are established. For supply chain risk, network flow models and queuing theory models are constructed to analyze bottleneck and disruption risks. In the risk quantification phase, this module transforms abstract concepts such as the probability of risk events, potential loss amounts, and risk exposure into concrete, quantifiable values. It calculates metrics such as expected loss (EL), unexpected loss (UEL), risk-adjusted return on capital (RAROC), loss under stress testing, and sensitivity analysis indicators. It can also assess the correlations and interactions between different risk factors, for example, by modeling nonlinear dependencies between multiple variables using copula functions, thus providing a comprehensive view of risk and accurate quantitative results. It is understood that this embodiment allows for dynamic updates to the risk prediction model. When new data flows in or the environment changes, the risk prediction model can be quickly recalibrated or retrained to ensure the real-time nature and accuracy of risk assessment.

[0056] Step S13: Using the target risk decision-making agent, based on the target risk type and risk measurement value corresponding to the target data, determine the target knowledge from the financial knowledge data source, and use a pre-determined target data fusion algorithm to fuse all the target knowledge to generate a corresponding target decision-making strategy.

[0057] In this embodiment, the knowledge graph in the financial knowledge data source systematically integrates structured knowledge within the domain, massive amounts of unstructured text information, and valuable implicit experience from human experts. This provides the intelligent decision-making engine of the target risk decision-making agent with rich background knowledge, logical constraints, and experiential guidance, thereby significantly improving the accuracy, rationality, and interpretability of decisions. In one specific implementation, the knowledge graph is first constructed, organizing entities, relationships, and attributes in a graph structure. This process may include: knowledge extraction, using natural language processing technology to identify entities and relationships from texts such as industry reports, news, and regulations; knowledge fusion and alignment, resolving conflicts and redundancies between knowledge from different sources to ensure consistency; and knowledge representation and storage in a graph database, supporting efficient querying and reasoning. Through knowledge reasoning, the target risk decision-making agent can even discover new hidden knowledge from known facts. It is understood that the financial knowledge data source also includes a pre-defined rule base and financial knowledge that has been acquired but not yet updated in the knowledge graph, such as newly acquired expert feedback suggestions on previous decision-making strategies. It should be noted that the knowledge data source in this embodiment also includes feedback information obtained by the target risk decision-making agent. In other words, the target risk decision-making agent supports human-machine collaborative feedback, allowing experts to correct the decisions made by the target risk decision-making agent and feed these corrections back to the target risk decision-making agent as new knowledge or rules. This allows for fine-tuning of the target risk decision-making agent or updating of the knowledge graph using heuristic methods or rule bases. Through the deep integration of knowledge graphs and expert experience, the target risk decision-making agent can not only learn from data but also from expert experience and human thinking patterns when dealing with similar problems. In scenarios with sparse data or requiring compliance judgments, it can make more comprehensive, logical, and explainable decisions. With the assistance of knowledge graphs, the target risk decision-making agent can provide clear explanations for decisions, such as the basis for the decision, the risk factors involved, and related knowledge points, thereby improving the transparency, reliability, and traceability of decisions, solving the "black box" problem of traditional AI models, and promoting effective human-machine collaboration.

[0058] In this embodiment, the intelligent decision engine of the target risk decision-making agent integrates multiple algorithms, including reinforcement learning, multi-objective optimization, and game theory. Based on the target risk type and risk metric values ​​corresponding to the target data obtained in the aforementioned process, it can identify and handle complex, dynamic, and multi-objective conflict decision scenarios, and determine corresponding target knowledge from financial knowledge data sources to automatically generate and optimize target decision strategies. Specifically, before fusing all the target knowledge using a pre-determined target data fusion algorithm to generate the corresponding target decision strategy, the process further includes: using the target risk decision-making agent to determine the target data fusion algorithm based on pre-acquired decision data through reinforcement learning; wherein, the decision data includes decision objectives, decision space, and environmental dynamics; the decision objectives include maximizing expected returns, minimizing expected risks, and balancing returns and risks. In other words, in this embodiment, the target risk decision-making agent can dynamically orchestrate and manage multiple algorithms within the engine according to specific decision objectives (such as maximizing returns, minimizing risks, or balancing multiple objectives), decision space, and environmental dynamics, based on the characteristics of the current task, thereby intelligently selecting or fusing suitable algorithms as target data fusion algorithms. The strategy of multi-algorithm fusion enables the target risk decision-making agent to flexibly select or combine the optimal decision-making algorithm according to the specific decision-making objectives, environmental dynamics and risk characteristics.

[0059] In one specific implementation, the nature of the current task is analyzed before each decision, primarily evaluated from three aspects: clarifying the decision objective, assessing the decision difficulty, and judging the environmental state. First, the core objective of the task is determined, such as pursuing maximum profit, prioritizing risk control, or achieving a balance among multiple objectives. Then, the complexity and constraints of the available options are analyzed. Finally, the stability of the external environment is assessed, such as whether current market data is stable or experiencing significant fluctuations, thus completing the decision-making scenario analysis. Based on the results of the scenario analysis, the most suitable algorithm is selected from the algorithm library using built-in logical rules based on preset strategies or empirical learning. For example, when determining the algorithm based on preset strategies, if the environment is judged to be "significantly volatile," a highly adaptive reinforcement learning algorithm is automatically prioritized. If the algorithm selection method is based on empirical learning, it can learn from a large number of historical decisions to summarize which algorithm has the highest success rate in which situations. During decision-making, the target risk decision intelligence makes the optimal choice based on this experience. When multiple algorithms need to work collaboratively, the target risk decision intelligence is responsible for fusing their outputs to form a unified decision recommendation. There are two main fusion methods: weighted fusion and dominant selection. Weighted fusion assigns different "importance" weights to the suggestions of each algorithm based on the current context, and then combines them into a final solution. For example, in a stable market, the suggestions of the predictive model have a higher weight; during market upheavals, the weight of the risk control model may be increased. In the case of a dominant selection method, under certain circumstances, the target risk decision-making AI will determine that there is one and only one most professional "dominant algorithm" in the current scenario, and use its output as the final decision, while the outputs of other algorithms are used as references or verifications.

[0060] In this embodiment, reinforcement learning algorithms are particularly suitable for sequential decision-making and dynamic environments. The target risk decision-making agent learns how to take actions to maximize long-term cumulative rewards through interaction with the environment, such as dynamically adjusting an investment portfolio in financial transactions. The application of reinforcement learning endows the target risk decision-making agent with the ability to make sequential decisions and learn autonomously in dynamic and uncertain environments, enabling it to continuously optimize its decision-making strategy through interaction with the environment to maximize long-term benefits. Furthermore, when multiple conflicting objectives exist, multi-objective optimization algorithms (such as NSGA-II, Non-dominated Sorting Genetic Algorithm II) are used to find the optimal trade-off. Simultaneously, the target risk decision-making agent can also integrate multiple decision results from deep learning, predictive analytics (using time-series models to predict future trends to assist decision-making), and rule-based reasoning expert systems (for handling explicit, rule-based decisions) to determine the final target decision-making strategy. Finally, the target risk decision-making agent provides a decision explanation, explaining the reasons for choosing this option as the target decision-making strategy, the underlying key risk factors, and the possible consequences of implementing the target decision-making strategy, thereby enhancing user trust. By integrating and coordinating these advanced decision-making algorithms, a multidimensional target risk decision-making agent is endowed with the powerful ability to learn autonomously, adapt, and make high-quality decisions in complex and dynamic environments.

[0061] In this embodiment, after fusing all target knowledge using a pre-determined target data fusion algorithm to generate corresponding target decision-making strategies, the process may further include: sending the target decision-making strategies to a strategy receiving end and obtaining the corresponding strategy execution results; using a pre-defined strategy evaluation mechanism to evaluate the target decision-making strategies based on the strategy execution results to generate feedback signals corresponding to the strategy execution results; and updating the target risk decision-making agent based on the target decision-making strategies, the corresponding strategy execution results, and the feedback signals to obtain the updated target risk decision-making agent. This approach constructs a dynamic, closed-loop system, ensuring that the target risk decision-making agent can continuously learn from the practice of its own target decision-making strategies, constantly optimize its performance, and maintain a high degree of adaptability to environmental changes, ensuring efficient decision-making capabilities even when facing new risk types or drastic environmental changes. The key to implementing this closed loop lies in rigorously evaluating each target decision-making strategy generated by the target risk decision-making agent and, based on the evaluation results, driving iterative updates and optimizations of various components within the target risk decision-making agent (especially components used to implement the intelligent decision engine and risk modeling) through a feedback mechanism. The evaluation dimensions are diverse, measuring the actual effectiveness of the target decision-making strategies from multiple perspectives. Simultaneously, the efficiency metrics (such as response time and resource consumption) and compliance metrics of the target decision-making strategy will be evaluated. The evaluation process includes real-time monitoring of changes in key metrics, periodic debriefing (summarizing and attributing the decision-making effectiveness of the target strategy over a period of time), comparative analysis (comparing the agent's decisions with human decisions or baselines), and error analysis. This feedback mechanism is a crucial bridge connecting evaluation and learning.

[0062] It should be noted that in this embodiment, the evaluation results are not simply reported, but are transformed into structured feedback signals, such as "rewards" or "punishments," which are directly input into the agent's learning mechanism. For example, if the agent's investment decisions yield unexpected returns, a positive reward signal will be generated; conversely, if they lead to unnecessary losses, a negative punishment signal will be generated. Furthermore, the target risk decision-making agent can also perceive environmental changes. When significant changes occur in the external environment, feedback signals are triggered, prompting the target risk decision-making agent to learn and adapt. Feedback learning and optimization are the core processes for the target risk decision-making agent to achieve evolution. Based on these feedback signals, the target risk decision-making agent can be driven to perform a series of self-improvements, including: updating model parameters, such as adjusting the weights of the neural network through backpropagation or optimizing the parameters of other machine learning models through gradient descent; optimizing policies, in reinforcement learning, adjusting decision-making policies based on accumulated rewards so that they can take better actions when facing similar situations in the future; revising the knowledge graph and rule base, making corresponding additions, deletions, and modifications if feedback indicates that a fact in the knowledge graph is no longer accurate or a rule in the rule base is no longer applicable; and sometimes even adaptively adjusting the algorithm, switching or combining different decision-making algorithms to achieve the best results based on current performance and feedback. Through this continuous cycle of evaluation, feedback, and learning, the target risk decision-making agent can continuously learn from historical policy execution results, correct its cognitive biases and decision-making deficiencies, and thus maintain the foresight, accuracy, and robustness of its decision-making capabilities in a constantly changing and uncertain environment, achieving continuous self-optimization and evolution.

[0063] In this example, multi-source heterogeneous data, the target decision-making strategies corresponding to the multi-source heterogeneous data, the execution results of the target decision-making strategies, and the generation process of the target decision-making strategies can be defined as decision process data. The real-time acquired decision process data is sent to a target visualization platform. The platform processes this data using preset data processing methods to obtain processed decision process data, which is then presented in preset data presentation formats, including statistical charts, dashboards, and heatmaps. This approach presents the complex operating mechanisms, sophisticated data analysis results, abstract risk assessment reports, and final decision recommendations of the target risk decision-making agent to users in an intuitive, clear, and easy-to-understand manner. Furthermore, the target visualization platform provides a series of flexible human-computer interaction functions, including but not limited to allowing users to adjust risk preferences, set decision target weights, review, intervene in, or correct decisions recommended by the agent, and receive real-time system alerts. This enables users to deeply understand, fully trust, and effectively intervene in and optimize the agent's decision-making process, achieving true human-computer collaboration.

[0064] In one specific implementation, the target visualization platform can display complex raw and preprocessed data using carefully designed charts, dashboards, heatmaps, and other visual elements, helping users quickly capture the trends, distributions, and potential problems behind the data. The visualization of risk analysis results clearly presents the output of risk modeling and quantitative analysis through dynamically updated dashboards and risk maps. By presenting complex internal data analysis results, risk assessment reports, decision-making processes, and decision recommendations in intuitive and easy-to-understand charts and dashboards, users can quickly grasp the overall risk picture and decision-making logic. Simultaneously, the target visualization platform supports intuitive comparisons of scenario analysis and stress test results and can decompose risk factors, explaining which factors contribute most to the overall risk, helping users deeply understand the root causes of risk. When the target risk decision-making agent provides a target decision-making strategy as a decision recommendation, the target visualization platform simultaneously displays the basis and rationale for the decision. If the decision-making process is complex, the target visualization platform can visualize the decision path. Users can directly submit feedback on the target visualization platform and receive corresponding notifications in a timely manner. Through a highly visualized and flexible interactive design, the intelligent agent becomes a transparent, controllable, and easy-to-understand intelligent partner, greatly enhancing users' trust in and adoption of the target risk decision-making intelligent agent, and achieving efficient human-machine collaboration.

[0065] In this embodiment, the above process integrates six closely collaborative components: multi-source heterogeneous data acquisition and preprocessing, risk modeling and quantitative analysis, intelligent decision-making, knowledge graph and expert experience fusion, decision evaluation and feedback learning, and visualization and interaction. This constructs a comprehensive, efficient, and adaptive multi-dimensional risk decision-making agent, ensuring the breadth, depth, and quality of data required for decision-making. It breaks through the limitations of traditional decision-making relying on single or limited data, enabling the target risk decision-making agent to capture subtle risk signals and opportunities from massive amounts of information, thus improving the comprehensiveness and information richness of decisions. By integrating data, models, algorithms, knowledge, feedback, and human-computer interaction, this solution achieves a seamless connection from data to insight to intelligent decision-making, endowing the target risk decision-making agent with powerful risk management capabilities and significantly improving the efficiency, accuracy, adaptability, and intelligence of risk decision-making in complex and uncertain environments.

[0066] As can be seen, this application can efficiently and accurately acquire information from massive and diverse internal and external data sources, and perform deep cleaning, integration, and feature engineering, laying a high-quality data foundation for subsequent risk modeling. Based on this, the target risk decision-making agent analyzes the obtained target data, achieving accurate identification, multi-dimensional modeling, and quantitative assessment of various complex risks. It transforms abstract risk concepts into actionable numerical values, providing a solid quantitative basis for decision-making. Furthermore, based on the target risk type and risk measurement value corresponding to the target data, it determines the corresponding target knowledge from financial data sources, providing a clear explanation for the target decision-making strategy ultimately generated based on this target knowledge. This improves the transparency, reliability, and traceability of decision-making, promotes effective human-machine collaboration, and ensures that the target risk decision-making agent can better serve actual business needs.

[0067] Based on the above, this application discloses a financial risk decision-making method based on intelligent agents, applied to terminals equipped with target risk decision-making intelligent agents. The following section uses the real-time portfolio risk management of a large financial asset management company as an example to illustrate the specific financial risk decision-making process.

[0068] The operation of the intelligent agent begins with data acquisition and preprocessing: real-time batch acquisition of high-frequency structured data from global financial markets (such as stock and bond quotes and trading volumes), macroeconomic indicators, and capture of unstructured text data such as financial news; it also connects with the company's internal trading and holding systems; subsequently, the heterogeneous raw data is cleaned, integrated, transformed, and feature-engineered (such as calculating technical indicators, extracting fundamental indicators, and conducting public opinion analysis). The processed data is stored in a high-speed time-series database or data lake, laying the foundation for subsequent modeling.

[0069] Once the data is ready, the core risk modeling and quantitative analysis phase begins: For market risk, historical simulation and parametric methods are used to calculate risk value and expected loss, predict asset volatility, and conduct stress tests; for credit risk, a machine learning credit scoring model is built to predict the probability of default; for liquidity risk, the asset liquidity is assessed; and at the same time, compliance risk assessments are conducted in real time to check whether the risk complies with regulations and internal policies. The risk quantification results are dynamically updated, and an early warning is triggered when a threshold is reached.

[0070] Once risk insights are generated, the intelligent decision engine begins to function: it combines risk quantification results, market conditions, and investment objectives, relying on reinforcement learning to learn interactively in a simulated market environment. It dynamically adjusts asset allocation ratios to maximize long-term risk-adjusted returns. When encountering sudden high-risk events (such as negative news about a listed company), it can quickly assess the impact, provide stop-loss and strategy adjustment suggestions, and even execute contingency plans. For example, if negative news breaks after the market closes on a trading day that Company A was ordered to remove its product due to a security vulnerability, the system instantly detects this and triggers an alarm. The decision engine quickly analyzes the impact of the event, generates emergency stop-loss suggestions, and, if a pre-set contingency plan exists, prepares to execute trading instructions at the opening of the next trading day.

[0071] The knowledge graph and expert experience integration module provides in-depth support for decision-making: it constructs a knowledge graph covering entities such as financial products, companies, and regulatory laws, encodes the experience of investment managers and risk control experts into an expert rule base, and also stores internal investment cases and risk events in a structured manner. This allows intelligent agents to not only learn from data, but also apply domain common sense and rules to make more reasonable decisions in scenarios where data is sparse or compliance considerations are required.

[0072] The decision evaluation and feedback learning module ensures the continuous evolution of the agent: it monitors the actual performance of decisions in real time and compares it with expectations; it performs backtracking and attribution when major market fluctuations or risk events occur; it transforms the results into signals to optimize the reinforcement learning model; it also records the investment manager's manual corrections to decisions as supervisory signals for model retraining; the agent can also simulate trading in a sandbox environment and optimize strategies through A / B testing to form a closed-loop learning mechanism.

[0073] Finally, the visualization and interactive interface module presents the internal operations to investment managers: it provides a real-time risk dashboard to display core risk indicators, presents portfolio holdings and profit and loss, pops up trading suggestions and decision-making basis, and also supports investment managers to adjust risk preferences and revise suggestions. All intervention behaviors will serve as a learning sample feedback system to help humans and machines collaborate efficiently.

[0074] Through the above process, the intelligent agent can safely and efficiently process massive amounts of heterogeneous data, deeply understand and quantify multi-dimensional risks, and autonomously learn to optimize decision-making strategies. While providing accurate suggestions, it ensures the interpretability of decisions and the flexibility of human-machine collaboration, ultimately significantly improving the risk management capabilities, return on investment, and operational intelligence level of asset management companies. Its multi-layered integrated architecture and adaptive learning capabilities make it an ideal solution for dealing with the uncertainties of the financial market.

[0075] See Figure 2 As shown, this application discloses a financial risk decision-making device based on an intelligent agent, applied to a terminal equipped with a target risk decision-making intelligent agent, comprising:

[0076] The data preprocessing module 11 is used to obtain raw multi-source heterogeneous data from various target financial data sources through the target data interface, and to perform data preprocessing operations on all the multi-source heterogeneous data to obtain corresponding preprocessed data; the target financial data sources include preset private data sources and preset public data sources;

[0077] The risk analysis module 12 is used to perform feature extraction processing on the preprocessed data based on a preset feature extraction algorithm to obtain corresponding target data, and to use the target risk decision-making agent to determine the target risk type and corresponding risk measurement value based on the target data.

[0078] The strategy generation module 13 is used to utilize the target risk decision-making agent to determine target knowledge from the financial knowledge data source based on the target risk type and the risk measurement value corresponding to the target data, and to use a pre-determined target data fusion algorithm to fuse all the target knowledge to generate a corresponding target decision-making strategy.

[0079] As can be seen, this application can efficiently and accurately acquire information from massive and diverse internal and external data sources, and perform deep cleaning, integration, and feature engineering, laying a high-quality data foundation for subsequent risk modeling. Based on this, the target risk decision-making agent analyzes the obtained target data, achieving accurate identification, multi-dimensional modeling, and quantitative assessment of various complex risks. It transforms abstract risk concepts into actionable numerical values, providing a solid quantitative basis for decision-making. Furthermore, based on the target risk type and risk measurement value corresponding to the target data, it determines the corresponding target knowledge from financial data sources, providing a clear explanation for the final target decision-making strategy generated based on this target knowledge. This improves the transparency, reliability, and traceability of decision-making, and promotes effective human-machine collaboration.

[0080] In one specific embodiment, the data preprocessing module 11 may include:

[0081] A data cleaning unit is used to clean the multi-source heterogeneous data based on preset data cleaning rules to obtain first processed data.

[0082] The data conversion unit is used to convert the corresponding first processed data based on the target financial data source and data format corresponding to each of the multi-source heterogeneous data, using a preset data integration algorithm to obtain the second processed data.

[0083] The data standardization unit is used to standardize the second processed data based on a preset data standardization algorithm to obtain corresponding preprocessed data.

[0084] In one specific implementation, the risk analysis module 12 may include:

[0085] The target algorithm determination unit is used to determine the preset feature extraction algorithm corresponding to each target financial data source based on the data source information features corresponding to each target financial data source.

[0086] The feature extraction unit is used to extract features from the preprocessed data based on the preset feature extraction algorithm corresponding to each of the target financial data sources, so as to obtain the initial feature dataset corresponding to each of the target financial data sources.

[0087] The data fusion unit is used to determine the target weights of each of the initial feature datasets based on the current decision-making context using the target multimodal data fusion model, and to perform data fusion operations on all the initial feature datasets based on the target weights to obtain the corresponding target data.

[0088] In one specific implementation, the risk analysis module 12 may include:

[0089] The risk type determination unit is used to analyze the unstructured data in the target data based on a preset text analysis algorithm to determine the target risk type corresponding to the target data.

[0090] The risk metric determination unit is used to determine the corresponding target risk prediction model based on the target risk type, and to determine the corresponding risk metric based on the structured data in the target data using the target risk prediction model.

[0091] In one specific embodiment, the device may further include:

[0092] The fusion algorithm determination module is used to determine the target data fusion algorithm based on pre-acquired decision data using the target risk decision-making agent through reinforcement learning;

[0093] The decision data includes decision objectives, decision space, and environmental dynamics; the decision objectives include maximizing expected returns, minimizing expected risks, and balancing returns and risks.

[0094] In one specific embodiment, the device may further include:

[0095] The execution result acquisition module is used to send the target decision strategy to the strategy receiving end and acquire the corresponding strategy execution result;

[0096] The feedback signal generation module is used to evaluate the target decision strategy based on the strategy execution result using a preset strategy evaluation mechanism, so as to generate a feedback signal corresponding to the strategy execution result.

[0097] The agent update module is used to update the target risk decision agent based on the target decision strategy, the corresponding strategy execution result, and the feedback signal, so as to obtain the updated target risk decision agent.

[0098] In one specific embodiment, the device may further include:

[0099] The process data display module is used to determine the multi-source heterogeneous data, the target decision strategy corresponding to the multi-source heterogeneous data, the strategy execution result corresponding to the target decision strategy, and the generation process of the target decision strategy as decision process data, and send the real-time acquired decision process data to the target visualization platform, so that the target visualization platform can process the decision process data according to a preset data processing method to obtain processed decision process data, and present the processed decision process data according to a preset data presentation format;

[0100] The preset data presentation formats include statistical charts, dashboards, and heat maps.

[0101] Furthermore, embodiments of this application also disclose an electronic device, Figure 3 This is a structural diagram of an electronic device 20 according to an exemplary embodiment. The content of the diagram should not be construed as limiting the scope of this application.

[0102] Figure 3This is a schematic diagram of the structure of an electronic device 20 provided in an embodiment of this application. Specifically, the electronic device 20 may include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input / output interface 25, and a communication bus 26. The memory 22 stores a computer program, which is loaded and executed by the processor 21 to implement the relevant steps in the agent-based financial risk decision-making method disclosed in any of the foregoing embodiments. Alternatively, the electronic device 20 in this embodiment may specifically be an electronic computer.

[0103] In this embodiment, the power supply 23 is used to provide operating voltage for each hardware device on the electronic device 20; the communication interface 24 can create a data transmission channel between the electronic device 20 and external devices, and the communication protocol it follows can be any communication protocol applicable to the technical solution of this application, and is not specifically limited here; the input / output interface 25 is used to acquire external input data or output data to the outside world, and its specific interface type can be selected according to specific application needs, and is not specifically limited here.

[0104] In addition, the memory 22, as a carrier for resource storage, can be a read-only memory, random access memory, disk, or optical disk, etc. The resources stored thereon can include an operating system 221, computer programs 222, etc., and the storage method can be temporary storage or permanent storage.

[0105] The operating system 221 is used to manage and control the various hardware devices on the electronic device 20 and the computer program 222, which may be Windows Server, Netware, Unix, Linux, etc. In addition to including a computer program capable of performing the agent-based financial risk decision-making method executed by the electronic device 20 as disclosed in any of the foregoing embodiments, the computer program 222 may further include computer programs capable of performing other specific tasks.

[0106] Furthermore, this application also discloses a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, it implements the aforementioned agent-based financial risk decision-making method. Specific steps of this method can be found in the corresponding content disclosed in the foregoing embodiments, and will not be repeated here.

[0107] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to in the method section.

[0108] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0109] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly by hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.

[0110] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0111] The technical solutions provided in this application have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A financial risk decision-making method based on intelligent agents, characterized in that, Applications include terminals equipped with target risk decision-making intelligent agents, including: The system obtains raw, multi-source heterogeneous data from various target financial data sources through a target data interface, and performs data preprocessing operations on all the multi-source heterogeneous data to obtain corresponding preprocessed data; the target financial data sources include preset private data sources and preset public data sources; The preprocessed data is processed by a preset feature extraction algorithm to obtain the corresponding target data, and the target risk decision-making agent is used to determine the target risk type and the corresponding risk measurement value based on the target data. The target risk decision-making agent uses the target risk type and risk measurement value corresponding to the target data to determine target knowledge from financial knowledge data sources, and uses a pre-determined target data fusion algorithm to fuse all the target knowledge to generate a corresponding target decision-making strategy. The step of performing data preprocessing operations on all the multi-source heterogeneous data to obtain corresponding preprocessed data includes: The multi-source heterogeneous data is cleaned according to preset data cleaning rules to obtain the first processed data; Based on the target financial data source and data format corresponding to each of the multi-source heterogeneous data, a preset data integration algorithm is used to perform data transformation on the corresponding first processed data to obtain second processed data; The second processing data is standardized based on a preset data standardization algorithm to obtain corresponding preprocessed data; The step of performing feature extraction processing on the preprocessed data based on a preset feature extraction algorithm to obtain the corresponding target data includes: Based on the data source information features corresponding to each of the target financial data sources, a preset feature extraction algorithm is determined for each of the target financial data sources. Based on the preset feature extraction algorithm corresponding to each of the target financial data sources, feature extraction is performed on the corresponding preprocessed data to obtain the initial feature dataset corresponding to each of the target financial data sources; The target weights of each initial feature dataset are determined based on the current decision-making context using a target multimodal data fusion model, and a data fusion operation is performed on all the initial feature datasets based on the target weights to obtain the corresponding target data. The process of determining the target risk type and corresponding risk metric based on the target data includes: Based on a preset text analysis algorithm, the unstructured data in the target data is analyzed to determine the target risk type corresponding to the target data; Based on the target risk type, a corresponding target risk prediction model is determined, and the target risk prediction model is used to determine the corresponding risk metric value based on the structured data in the target data.

2. The agent-based financial risk decision-making method according to claim 1, characterized in that, Before fusing all the target knowledge using a pre-determined target data fusion algorithm to generate the corresponding target decision strategy, the method further includes: Using the target risk decision-making agent, a target data fusion algorithm is determined based on pre-acquired decision data through reinforcement learning; The decision data includes decision objectives, decision space, and environmental dynamics; the decision objectives include maximizing expected returns, minimizing expected risks, and balancing returns and risks.

3. The agent-based financial risk decision-making method according to claim 1 or 2, characterized in that, After fusing all the target knowledge using a pre-determined target data fusion algorithm to generate a corresponding target decision strategy, the method further includes: The target decision strategy is sent to the strategy receiving end, and the corresponding strategy execution result is obtained; A preset strategy evaluation mechanism is used to evaluate the target decision strategy based on the strategy execution result, so as to generate a feedback signal corresponding to the strategy execution result; Based on the target decision-making strategy, the corresponding strategy execution results, and the feedback signal, the target risk decision-making agent is updated to obtain the updated target risk decision-making agent.

4. The agent-based financial risk decision-making method according to claim 3, characterized in that, Also includes: The multi-source heterogeneous data, the target decision strategy corresponding to the multi-source heterogeneous data, the strategy execution result corresponding to the target decision strategy, and the generation process of the target decision strategy are determined as decision process data. The real-time acquired decision process data is sent to the target visualization platform so that the decision process data can be processed by the target visualization platform based on a preset data processing method to obtain processed decision process data. The processed decision process data is then presented based on a preset data presentation format. The preset data presentation formats include statistical charts, dashboards, and heat maps.

5. A financial risk decision-making device based on intelligent agents, characterized in that, Applications include terminals equipped with target risk decision-making intelligent agents, including: The data preprocessing module is used to obtain raw multi-source heterogeneous data from various target financial data sources through the target data interface, and to perform data preprocessing operations on all the multi-source heterogeneous data to obtain corresponding preprocessed data; the target financial data sources include preset private data sources and preset public data sources; The risk analysis module is used to perform feature extraction processing on the preprocessed data based on a preset feature extraction algorithm to obtain the corresponding target data, and to use the target risk decision-making agent to determine the target risk type and the corresponding risk measurement value based on the target data. The strategy generation module is used to utilize the target risk decision-making agent to determine target knowledge from the financial knowledge data source based on the target risk type and the risk measurement value corresponding to the target data, and to use a pre-determined target data fusion algorithm to fuse all the target knowledge to generate a corresponding target decision-making strategy. The data preprocessing module specifically includes: A data cleaning unit is used to clean the multi-source heterogeneous data based on preset data cleaning rules to obtain first processed data. The data conversion unit is used to convert the corresponding first processed data based on the target financial data source and data format corresponding to each of the multi-source heterogeneous data, using a preset data integration algorithm to obtain the second processed data. The data standardization unit is used to standardize the second processed data based on a preset data standardization algorithm to obtain corresponding preprocessed data. The risk analysis module specifically includes: The target algorithm determination unit is used to determine the preset feature extraction algorithm corresponding to each target financial data source based on the data source information features corresponding to each target financial data source. The feature extraction unit is used to extract features from the preprocessed data based on the preset feature extraction algorithm corresponding to each of the target financial data sources, so as to obtain the initial feature dataset corresponding to each of the target financial data sources. The data fusion unit is used to determine the target weights of each of the initial feature datasets based on the current decision-making context using the target multimodal data fusion model, and to perform data fusion operations on all the initial feature datasets based on the target weights to obtain the corresponding target data. The risk analysis module specifically includes: The risk type determination unit is used to analyze the unstructured data in the target data based on a preset text analysis algorithm to determine the target risk type corresponding to the target data. The risk metric determination unit is used to determine the corresponding target risk prediction model based on the target risk type, and to determine the corresponding risk metric based on the structured data in the target data using the target risk prediction model.

6. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor for executing the computer program to implement the agent-based financial risk decision-making method as described in any one of claims 1 to 4.

7. A computer-readable storage medium, characterized in that, Used to store a computer program, wherein the computer program, when executed by a processor, implements the agent-based financial risk decision-making method as described in any one of claims 1 to 4.

Images