Strategy dynamic optimization system and method based on big data analysis prediction

Abstract

The application discloses a strategy dynamic optimization system and method based on big data analysis prediction, relates to the technical field of data analysis and processing, and comprises the following layers: a data acquisition layer for establishing a data source; a data processing layer for performing data feature extraction after data cleaning of the data source; an extreme scenario simulation layer for executing coping strategy training under an antagonistic environment; a competitive optimization layer for configuring a competitive intelligent agent, performing adaptive evaluation of strategy quality, updating an adaptive strategy set, performing game optimization, and generating a game optimization result; and a strategy dynamic optimization layer for receiving the game optimization result and establishing a dynamic response strategy. The application solves the technical problems of poor data quality leading to analysis deviation, static strategies being difficult to adapt to complex and changeable market environments and being unable to effectively cope with dynamic game of different competitive roles, and resulting in insufficient strategy adaptability and flexibility, and achieves the technical effect of improving decision flexibility and adaptability in volatile market environments.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data analysis processing, and particularly relates to a strategy dynamic optimization system and method based on big data analysis prediction. BACKGROUND

[0002] The market environment changes rapidly, various types of transaction activities, information dissemination and interaction methods have changed greatly, with the popularization of the Internet, transaction data has shown explosive growth, covering various online and offline transaction scenarios, and contains key information such as market supply and demand, price fluctuations, market hotspots. For example, news data reflects the macroeconomic situation, industry dynamics and unexpected events in real time, which has a profound impact on market trends; social data shows the preferences, emotions and group behavior trends of consumers, providing a new perspective for understanding market demand; search data can intuitively reflect the user's focus and potential demand. However, the massive data has the problem of uneven data quality, such as data missing, errors, duplication, etc., and directly using the original data will lead to biased analysis results; at the same time, in many high-volatility markets (such as cryptocurrencies, luxury goods, NFTs, creative content, futures trading, etc.), traditional time series models cannot be stably predicted, and the strategy formulation method is difficult to adapt to the complex and changing market environment. Static strategies cannot cope with the changing dynamic game environment, resulting in insufficient flexibility and adaptability of strategies.

[0003] Therefore, in the related art at the present stage, there is a technical problem of insufficient flexibility and adaptability of strategies due to poor data quality leading to biased analysis, static strategies being difficult to adapt to complex and changing market environments and being unable to effectively cope with dynamic games of different competitive roles. SUMMARY

[0004] The present application provides a strategy dynamic optimization system and method based on big data analysis prediction, which solves the technical problem of insufficient flexibility and adaptability of strategies due to poor data quality leading to biased analysis, static strategies being difficult to adapt to complex and changing market environments and being unable to effectively cope with dynamic games of different competitive roles in the prior art, and achieves the technical effect of improving the decision flexibility and adaptability of volatile market environments.

[0005] This application provides a strategy dynamic optimization system based on big data analysis and prediction. The system includes: a data acquisition layer for collecting transaction data, news data, social data, search data, public data, and private domain user retention data to establish a data source; a data processing layer for cleaning the data source, extracting data features, identifying abnormal patterns based on the extracted features, and constructing a strategy training environment; an extreme scenario simulation layer for injecting random noise into the strategy training environment and training coping strategies under adversarial conditions to establish an adaptive strategy set; a competitive optimization layer for configuring competitive agents based on the data source, including leaders, followers, and adversaries, updating the adaptive strategy set after evaluating its quality, and using the updated adaptive strategy set and the competitive agents to perform game optimization and generate game optimization results; and a strategy dynamic optimization layer for receiving the game optimization results and establishing a dynamic response strategy.

[0006] In a possible implementation, the strategy dynamic optimization system based on big data analysis and prediction further performs the following processes: establishing an extreme scenario dataset and expanding the extreme scenario dataset using a generative adversarial network to generate an updated extreme scenario dataset; performing a self-check on the injection of random noise within a preset period, evaluating scenario coverage based on the injection self-check results and the updated extreme scenario dataset, and establishing random noise injection constraints; configuring time-continuous perturbation constraints and establishing risk factors, including liquidity risk and credit risk; and injecting random noise into the strategy training environment based on the random noise injection constraints, the time-continuous perturbation constraints, and the risk factors.

[0007] In a possible implementation, the strategy dynamic optimization system based on big data analysis and prediction further performs the following processing: monitoring the indicators of the strategy optimization subject and establishing a set of key indicators, including return rate, risk exposure, liquidity, and resource utilization efficiency; establishing the current state of the strategy optimization subject based on the set of key indicators; and using the current state and the updated adaptive strategy set as input data to perform game optimization with the competing intelligent agent.

[0008] In a possible implementation, the strategy dynamic optimization system based on big data analysis and prediction also performs the following processing: recording the payoff data, risk data, and decision performance data of all roles during the game, and establishing feedback signals; using the updated adaptive strategy set as the initial strategy set, and using the feedback signals as optimization constraints, performing iterative updates of the initial strategy set; and completing game optimization based on the iterative update results.

[0009] In a possible implementation, the strategy dynamic optimization system based on big data analysis prediction further performs the following processing: calculating fitness values in the initial strategy set, performing initial strategy set screening by using the fitness values, and establishing excellent individual identification; performing individual selection of the initial strategy set according to the reconfigured iteration probability based on the excellent individual identification; performing gene exchange according to the individual selection result, and updating the parameter contribution degree of the individual selection result by using the optimization constraint; and completing one iteration update after randomly disturbing the parameters of the individual selection result, where the current individual selection result is the individual selection result after the gene exchange and the adjustment of the parameter contribution degree.

[0010] In a possible implementation, the strategy dynamic optimization system based on big data analysis prediction further performs the following processing: configuring a quality screening threshold, performing adaptive strategy screening of the adaptive evaluation result according to the quality screening threshold; and removing adaptive strategies that do not meet the quality screening threshold, and updating the adaptive strategy set by using the remaining adaptive strategies.

[0011] In a possible implementation, the strategy dynamic optimization system based on big data analysis prediction further performs the following processing: extracting the data source after data cleaning into structured data and unstructured data; performing cross-modal feature fusion of the structured data and the unstructured data by using a deep learning model; and performing self-adaptive feature selection on the cross-modal feature fusion result to complete data feature extraction.

[0012] In a possible implementation, the strategy dynamic optimization system based on big data analysis prediction further performs the following processing: obtaining a behavior mode database of the competitive intelligent agent, performing reverse reasoning according to the behavior mode database, and predicting a potential coping strategy of the competitive intelligent agent; establishing a mapping game strategy according to the potential coping strategy; and performing compensation on the game optimization result according to the mapping game strategy.

[0013] In a possible implementation, the strategy dynamic optimization system based on big data analysis prediction further performs the following processing: a warning layer, configured to analyze the dynamic response strategy, generate a risk warning signal, and report the risk warning signal for risk warning.

[0014] The application also provides a strategy dynamic optimization method based on big data analysis prediction, comprising the following steps: collecting transaction data, news data, social data, search data, public data, private domain user retention data, and establishing a data source; performing data feature extraction after data cleaning of the data source, identifying abnormal patterns according to the data feature extraction result, and constructing a strategy training environment; injecting random noise into the strategy training environment, performing coping strategy training in the adversarial environment, and establishing an adaptive strategy set; configuring a competitive agent according to the data source, the competitive agent comprising a leader, a follower and an opponent, updating the adaptive strategy set after adaptive strategy quality evaluation, performing game optimization by using the updated adaptive strategy set and the competitive agent, and generating a game optimization result; and receiving the game optimization result and establishing a dynamic response strategy.

[0015] The strategy dynamic optimization system and method based on big data analysis prediction provided by the application comprise a data acquisition layer for establishing a data source; a data processing layer for performing data feature extraction after data cleaning of the data source; an extreme scenario simulation layer for performing coping strategy training in an adversarial environment; a competitive optimization layer for configuring a competitive agent, performing adaptive strategy quality evaluation, updating an adaptive strategy set, performing game optimization, and generating a game optimization result; and a strategy dynamic optimization layer for receiving the game optimization result and establishing a dynamic response strategy. The technical problems of poor data quality leading to analysis deviation, static strategies being difficult to adapt to complex and changeable market environments and being unable to effectively cope with dynamic game of different competitive roles, and resulting in insufficient strategy adaptability and flexibility are solved, and the technical effects of improving decision flexibility and adaptability in volatile market environments are achieved. BRIEF DESCRIPTION OF DRAWINGS

[0016] In order to more clearly illustrate the technical solutions of the embodiments of the present disclosure, the drawings of the embodiments of the present disclosure will be briefly introduced below. The flowcharts are used to illustrate the operations performed by the system according to the embodiments of the present application. It should be understood that the foregoing or the following operations are not necessarily performed in sequence. On the contrary, various steps can be processed in reverse order or simultaneously according to needs. Meanwhile, other operations can be added to these processes, or one or more steps can be removed from these processes.

[0017] Figure 1 The strategy dynamic optimization system based on big data analysis prediction provided by the embodiments of the present application is shown in the structural schematic diagram.

[0018] Figure 2 The strategy dynamic optimization method based on big data analysis prediction provided by the embodiments of the present application is shown in the flowchart.

[0019] Explanation of reference signs: data collection layer 10, data processing layer 20, extreme scenario simulation layer 30, competition optimization layer 40, strategy dynamic optimization layer 50. DETAILED DESCRIPTION

[0020] The above description is only a summary of the technical scheme of the present application. In order to make the technical means of the present application more clear, the following specific embodiments of the present application are described in accordance with the content of the specification, and in order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the following specific embodiments of the present application are described.

[0021] In order to make the purposes, technical schemes and advantages of the present application more clear, the following will further describe the present application with reference to the accompanying drawings. The described embodiments should not be regarded as limiting the present application. All other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0022] In the following description, "some embodiments" are referred to, which describe a subset of all possible embodiments, but it can be understood that "some embodiments" can be the same subset or different subset of all possible embodiments, and can be combined with each other without conflict. The term "first\second" referred to is only to distinguish similar objects, and does not represent the specific order of the objects. The terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, system, product or server including a series of steps or units does not have to be limited to those steps or units clearly listed, but can include other steps or modules not clearly listed or inherent to these processes, products or devices. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as understood by those skilled in the art to which the present application belongs. The terms used herein are only for the purpose of describing the embodiments of the present application.

[0023] The embodiments of the present application provide a strategy dynamic optimization system based on big data analysis prediction, as shown in Figure 1 The system comprises:

[0024] The data collection layer 10 is used to collect transaction data, news data, social data, search data, public data, private domain user retention data, and establish data sources.

[0025] Preferably, the data collection layer serves as the front end of the data processing system and is mainly used for collecting various types of data and establishing data sources, including transaction data, news data, social data, search data, public data, and private user retention data. Specifically, transaction data refers to data generated in various business transactions, covering various online and offline transaction scenarios, which may include transaction time, transaction amount, transaction commodity or service information, transaction party identity information, transaction location, etc. For example, order data on e-commerce platforms contains the name, price, quantity, payment method, and merchant information of the goods purchased by consumers. Offline transaction data of physical stores is recorded through a cash register system, including transaction time, item details, and payment method. By analyzing transaction data, one can understand market supply and demand, price fluctuation trends, consumer purchasing preferences and consumption capacity, etc. News data refers to news information published by various news media, including text, pictures, videos, and other forms, which reflect real-time dynamic information in various fields. It mainly includes news titles, texts, publication times, news sources, and keywords, etc. For example, economic news may contain information such as macroeconomic policy adjustments, industry development trends, and enterprise mergers and acquisitions; technology news may report on breakthroughs in new technology research and development and the rise of emerging industries, etc., which helps enterprises and investors stay informed about macroeconomic trends, industry dynamics, and the impact of unexpected events on the market.

[0026] Preferably, social data is various data generated by users on social platforms, mainly including user profiles, published content (such as text, pictures, videos, etc.), social relationships (friend lists, follow and followed relationships, etc.), interaction behaviors (likes, comments, shares), and other behavior data, as well as user location, device used, login time, etc. For example, user reviews on product use experience, opinions on a certain brand, discussions on hot topics, etc. on social media platforms can help enterprises deeply understand consumer preferences, needs, emotions, and social group behavior trends. Search data is the keyword entered by the user in the search engine or other search tools, search time, search result click behavior, etc. related data, mainly including user input query words, search result page views, user clicked link information, etc. For example, when the user inputs "smartphone recommendation" in the search engine, the search engine will record this query word and the subsequent related search result pages clicked by the user, which all belong to search data, which can reflect the user's focus and potential needs, help enterprises understand market hotspots and consumer information search habits, and thus optimize product information display and adjust advertising placement strategies. Public data is data published by public institutions or other public channels, such as economic data (such as GDP, inflation rate, unemployment rate, etc.), industry reports, population census data, etc. Public data is authoritative and macroscopic, which helps overall market analysis. Private domain user retention data refers to user data in the private domain traffic pool owned by enterprises or organizations, including user registration information, login data, browsing records, consumption history, retention time, user feedback, etc. on enterprise platforms (such as official websites, APPs). Private domain user retention data helps understand the characteristics and behaviors of enterprise user groups. Finally, the collected various data is integrated to establish a data source to ensure efficient storage, fast query and safe management of data.

[0027] The data processing layer 20 is configured to perform data feature extraction after data cleaning of the data source, identify abnormal patterns according to the data feature extraction result, and build a strategy training environment.

[0028] Preferably, the data processing layer mainly further processes the collected data sources, including data cleaning and feature extraction on the data sources to obtain data feature extraction results. Specifically, since the collected data may have problems such as incompleteness, errors, duplication or inconsistency, data cleaning aims to improve data quality and ensure the accuracy and reliability of subsequent analysis results, usually including processing missing values, such as selecting a method of deleting records with missing values, filling in mean values or median values, etc. according to the characteristics of the data; removing duplicate data to ensure the uniqueness of the data; detecting and correcting error data, for example, checking the rationality of the data through data verification rules, correcting or marking the obviously erroneous data; processing inconsistent data, unifying the format, coding and unit of the data, etc. to make the data consistent and comparable.

[0029] Preferably, then the cleaned data is subjected to feature extraction to obtain key feature information reflecting the characteristics and rules of the data, and to convert the original data into a form that is easier to understand and analyze. Specifically, different feature extraction methods are used for different types of data. For transaction data, statistical features of transaction frequency and transaction amount (such as mean, standard deviation, maximum value, minimum value, etc.), commodity category distribution, etc. may be extracted; for news data, keyword frequency, news theme category, sentiment orientation (judging the positive, negative or neutral sentiment of the news content through sentiment analysis technology), etc. may be extracted; for social data, user social activity (such as the frequency of posting content, the number of interactions, etc.), social influence (measured by the number of followers, likes and comments, etc.), topic preference, etc. may be extracted; for search data, search keyword frequency, search time distribution, search result click rate, etc. may be extracted; and these features are taken as data feature extraction results.

[0030] Preferably, the abnormal patterns are identified according to the data feature extraction results, that is, abnormal data points or data sets different from the normal patterns in the data are found, and then abnormal fluctuations in the market, abnormal behaviors of users, etc. are found in time. Specifically, based on the results of data feature extraction, various anomaly detection algorithms and models are used to identify abnormal patterns. For example, a threshold is set using a statistical method, and data points that exceed the threshold of the normal range are considered abnormal; or a clustering algorithm in machine learning is used to divide the data into different clusters, and data points in isolated clusters may be considered as abnormal points; a deep learning model can also be used to learn the patterns of normal data and perform anomaly detection on data that deviates greatly from the learned patterns. Finally, a strategy training environment is constructed, which can simulate real scenarios to adapt to various situations in actual applications. Specifically, according to data features and business requirements, various parameters and rules of the environment are determined, for example, in a financial trading strategy training environment, parameters such as asset price change range, transaction cost, risk indicators are set according to historical transaction data and market fluctuations; in a marketing decision strategy training environment, different marketing strategy options (such as advertising channels, promotion activity forms, etc.) and corresponding effect evaluation indicators are set according to consumer behavior characteristics and market competition conditions; at the same time, the cleaned data and extracted features are used as input data of the environment for training to achieve the expected goal, and finally a strategy training environment is obtained.

[0031] The extreme scenario simulation layer 30 is used to perform coping strategy training in an adversarial environment after injecting random noise into the strategy training environment, and to establish an adaptive strategy set.

[0032] Preferably, the extreme scenario simulation layer is mainly used to simulate extreme situations that may occur in reality, such as many high-volatility markets (such as cryptocurrencies, luxury goods, NFTs, creative content, futures trading, etc.). By injecting random noise into the strategy training environment to construct an adversarial environment, an adaptive strategy set is trained and established. Specifically, there are various uncertainties and unexpected situations in real scenarios. By injecting random noise into the strategy training environment to simulate unpredictable factors, the trained strategy can cope with various complex and volatile market environments. Specifically, according to different application scenarios and data characteristics, the type and distribution of noise are determined. For example, in financial market simulation, random noise conforming to a certain probability distribution (such as normal distribution, t distribution, etc.) can be generated according to the volatility of historical data, and added to asset prices, transaction volumes, etc. In user behavior simulation, random noise can be added to user click frequency, purchase time interval, etc. to simulate the uncertainty of user behavior. The strength and range of the noise need to be adjusted according to the actual situation, that is, it should be ensured that the extreme situation can be simulated, and the noise should not be too large to make the data lose its authenticity.

[0033] Preferably, a coping strategy training in an adversarial environment is then performed, that is, in an adversarial environment with noise, the model or strategy is trained to be able to cope with various possible challenges and changes, specifically, using reinforcement learning, game theory, etc., the agent (such as a trading strategy model, a marketing decision model, etc.) continuously carries out trial and error learning in the adversarial environment, for example, in the financial trading scenario, the agent makes buy and sell decisions according to the market data with noise, and obtains reward or punishment feedback through interaction with the environment, and then adjusts the strategy to maximize the long-term cumulative reward, so as to learn to make the optimal decision in the adversarial environment. In the training process, multiple agents can also be introduced for competition or cooperation to further enhance the adaptability and robustness of the strategy, for example, in a simulated market competition scenario, different agents represent different market participants, and through mutual competition and learning, they continuously optimize their market strategies; and then multiple adaptive strategies are obtained and the strategies are analyzed and screened to remove the strategies with poor effects and retain the strategies that perform well in various extreme scenarios to form an adaptive strategy set. For example, in marketing decision, the adaptive strategy set may include multiple marketing strategy combinations for different market competition intensities and different consumer demand fluctuations, such as a strategy of taking low-price promotion combined with precise advertisement placement when the market competition is fierce, and a strategy of increasing product added value and intensifying brand promotion when consumer demand is strong, etc.

[0034] The competition optimization layer 40 is configured to configure a competition agent according to the data source, the competition agent including a leader, a follower, and an adversary. After the adaptive strategy set is evaluated for strategy quality adaptation, the adaptive strategy set is updated. The updated adaptive strategy set and the competition agent are used for game optimization to generate a game optimization result.

[0035] Preferably, the competition optimization layer is mainly used to evaluate and optimize the adaptive strategy set by configuring competition agents to generate better game optimization results. Specifically, competition agents are configured according to data sources to simulate the behavior and decision-making of different roles in the competition environment, i.e., according to the information contained in the data sources, such as market dynamics, user behavior patterns, competitor data, etc., different competition agents are set with corresponding attributes and behavior rules. Among them, competition agents are virtual roles with autonomous decision-making and behavior in the competition environment, including leaders, followers, and antagonists. Leaders usually have strong market influence and resource advantages and may adopt proactive innovation and market trend leading strategies. Followers tend to observe the actions of leaders and adjust their strategies according to their decisions to gain market share. Antagonists challenge leaders or other competitors and adopt aggressive strategies to try to break the existing market pattern. For example, in the e-commerce market, large e-commerce platforms can be leaders, attracting users by launching new service models or large-scale promotional activities. Small e-commerce platforms, as followers, will imitate some successful practices of large platforms and conduct fine operations targeting specific user groups. Emerging e-commerce platforms may act as antagonists by offering unique products or lower prices to compete for market share.

[0036] Preferably, the adaptive strategy set is evaluated for strategy quality to determine its effectiveness and adaptability in different competition scenarios and find areas for improvement and optimization. Specifically, multiple evaluation indicators are used to evaluate each strategy in the adaptive strategy set, including strategy profitability (such as expected profit, market share growth, etc.), risk (such as the probability of strategy failure, potential loss, etc.), adaptability (ability to adapt to different market conditions and competitor behavior), etc. The strategy is applied to a simulated competition scenario, its effect is observed and compared with preset targets and other strategies, for example, for a marketing promotion strategy, its sales growth under different market competition intensity, customer acquisition cost, and impact on brand image are evaluated to judge the quality and adaptability of the strategy. Then, based on the evaluation results, the strategies in the adaptive strategy set are adjusted and optimized to better cope with various competition situations, including modifying or eliminating poorly performing strategies in the adaptive strategy set based on the results of strategy quality adaptation evaluation, while introducing new strategies or improving existing strategies, for example, if a marketing strategy is found to be ineffective in the face of fierce competition, it may be adjusted, such as changing the advertising channel, adjusting the product pricing strategy, etc.; or based on new trends in the market and new developments of competitors, new marketing methods such as social media marketing and word-of-mouth marketing are introduced to enrich and optimize the adaptive strategy set.

[0037] Preferably, the updated adaptive strategy set and the competitive agents are used for game optimization, that is, through a simulated competitive scenario, different role competitive agents interact according to the strategy to find the optimal strategy combination. Specifically, each competitive agent (leader, follower, and opponent) selects a suitable strategy according to its own target, current environmental information, and the updated adaptive strategy set, and then interacts between competitive agents. For example, after the leader launches a new product, it may affect the market share of the follower and the opponent. The follower may reduce the price, which may cause the leader and the opponent to adjust the price. Each competitive agent obtains feedback according to the interaction results, such as the increase or decrease of income, the change of market share, etc. Based on the feedback information, the competitive agent learns and judges the effectiveness of the current strategy, and then adjusts the strategy according to the learning result. After multiple games and strategy adjustments, the competitive agent gradually converges to a strategy combination, which is the game optimization result, to achieve better results in the competitive environment.

[0038] The strategy dynamic optimization layer 50 is used to receive the game optimization result and establish a dynamic response strategy.

[0039] Preferably, the strategy dynamic optimization layer establishes a dynamic response strategy based on the game optimization result according to the changing environment and competitive situation to achieve better adaptability and competitiveness. Specifically, the strategy dynamic optimization layer collects the game optimization result, including the relatively optimal strategy combination formed by different competitive agents (leader, follower, and opponent) after the game, and related evaluation indexes and feedback information, such as the income of each agent, the change of market share, the stability of the strategy, etc. Then, the game optimization result is analyzed and evaluated to understand the performance and potential problems of the current strategy in different aspects, for example, the effectiveness and adaptability of the strategy in response to market demand fluctuations, technological changes, etc. are analyzed to find out possible weaknesses or areas for improvement. Then, the changes in the external environment are continuously monitored, including market trends, technological development, changes in consumer preferences, etc. According to the environmental changes and the analysis of the game optimization result, the existing strategy is dynamically adjusted and optimized, which may include reconfiguring the strategies of different competitive agents, such as adjusting the innovation direction of the leader, the imitation speed and range of the follower, and the attack intensity and key areas of the opponent, etc. Multiple alternative strategy schemes are developed, and the corresponding trigger mechanism is established to quickly switch to the appropriate strategy when specific environmental conditions occur, ensuring that the dynamic response strategy has sufficient flexibility and adaptability to quickly respond to various emergencies and uncertainties. Further, the strategy can be continuously optimized as the environment changes to maintain a dominant position in the competitive environment.

[0040] Further, the specific configuration of the extreme scenario simulation layer 30 also includes establishing an extreme scenario data set and expanding the extreme scenario data set using a generative adversarial network to generate an updated extreme scenario data set; performing self-checking on the injection of random noise within a preset period, performing scenario coverage evaluation according to the self-checking result and the updated extreme scenario data set, establishing an injection constraint of random noise; configuring a time-continuous disturbance constraint and establishing a risk factor, the risk factor including liquidity risk and credit risk; and performing random noise injection of the strategy training environment according to the injection constraint of random noise, the time-continuous disturbance constraint, and the risk factor.

[0041] Preferably, from historical event records and databases, data in various extreme situations are collected and sorted, such as extreme fluctuation situations in financial markets (such as stock market collapse, large exchange rate fluctuations), to constitute an initial extreme scenario data set, and then the initial extreme scenario data set is expanded using a generative adversarial network, that is, by learning the characteristics and distribution rules of the original data, new extreme scenario data similar but not completely the same is generated, the diversity and size of the data set are increased, and an updated extreme scenario data set is obtained, so as to better cover various possible extreme situations. Then, according to a preset time period, the process of injecting random noise into the strategy training environment is checked, which can include checking whether the generation of noise conforms to the predetermined rules and distribution, whether the strength of the noise is within a reasonable range, and whether the injection process is normal, whether there are errors or abnormal situations, etc., to obtain an injection self-checking result.

[0042] Preferably, in combination with the self-checking result of random noise injection and the updated extreme scenario data set after expansion, it is evaluated whether the current random noise injection can sufficiently cover various extreme scenarios, including checking whether the training environment after injecting noise can simulate various extreme situations contained in the updated extreme scenario data set, and whether there are some scenarios that are not effectively covered. Through scenario coverage evaluation, the effect and deficiency of the current random noise injection are understood, and based on the scenario coverage evaluation result, the constraint condition of random noise injection is determined. If it is found that some extreme scenarios are not well covered, the generation method, strength range or injection frequency of noise needs to be adjusted to ensure that the injection of random noise can more comprehensively cover various extreme scenarios, so that the strategy training environment can more realistically simulate the actual situation under extreme situations.

[0043] Preferably, the time continuous disturbance constraint is configured, that is, when random noise is injected into the strategy training environment, it is ensured that the noise is continuously changed in time, which conforms to the dynamic change law in the actual situation. When random noise is injected into the strategy training environment, it is ensured that the noise is continuously changed in time, which conforms to the dynamic change law, for example, the change of market conditions or environmental factors often occurs gradually, and through the time continuous disturbance constraint, the injected random noise can more realistically simulate the characteristics of such continuous change; determine the risk factors related to strategy training, specifically including liquidity risk and credit risk, wherein the liquidity risk reflects the difficulty of realizing assets at a reasonable price in the short term, and the credit risk refers to the possibility of counterparty default or credit deterioration; consider the influence of risk factors on the strategy, so that the trained strategy can better cope with various risks in the actual scene.

[0044] Preferably, according to the random noise injection constraint, the time continuous disturbance constraint and the risk factor, the random noise of the strategy training environment is injected, and in the injection process, it is ensured that the generation and injection of noise meet the requirements of the injection constraint, guarantee the continuity in time, and can reflect the influence of factors such as liquidity risk and credit risk. The strategy training environment can simulate more complex and realistic scenarios, so that the trained strategy has better adaptability when facing various uncertainties and risks. By continuously injecting random noise and training the strategy, the model can better learn how to make optimal decisions in different extreme scenarios and risk conditions.

[0045] Further, the specific configuration of the competition optimization layer 40 also includes monitoring the indicators of the strategy optimization subject, establishing a key indicator set, the key indicator set including yield, risk exposure, liquidity, resource utilization efficiency; according to the key indicator set, the current state of the strategy optimization subject is established; using the current state and the updated adaptive strategy set as input data, the game optimization with the competition agent is executed.

[0046] Preferably, the relevant data of the strategy optimization subject (such as portfolio, enterprise operation strategy, etc.) is monitored by indicators, that is, by collecting and analyzing a large amount of monitoring data, obtaining performance information of the strategy optimization subject in different aspects, selecting indicators that are crucial to the evaluation and decision of the strategy optimization subject, forming a key indicator set, including yield, risk exposure, liquidity, resource utilization efficiency. Specifically, the yield measures the profit level of the strategy optimization subject within a certain period of time, such as investment yield, asset return rate, etc.; the risk exposure assesses the degree of various risks faced by the strategy optimization subject, such as market risk, credit risk, etc.; the liquidity represents the realization ability of the assets of the strategy optimization subject and the ability to meet the demand for funds, such as the turnover speed of assets, cash reserve situation, etc.; the resource utilization efficiency investigates the utilization of resources (such as funds, manpower, equipment, etc.) by the strategy optimization subject, such as the efficiency of fund use, the work efficiency of employees, etc., to determine whether the resources have been reasonably allocated.

[0047] Preferably, based on the selected key indicator set, each indicator is quantified and analyzed to establish the overall state of the strategy optimization subject at the current time, for example, the evaluation results of yield, risk exposure, liquidity and resource utilization efficiency are integrated to form a description that can fully reflect the current operation status of the strategy optimization subject; the current state data of the strategy optimization subject and the updated adaptive strategy set are provided as input data to the game optimization model, and the strategy optimization subject and the competitive agent (leader, follower, opponent, etc.) perform game, specifically, the competitive agent selects a strategy to interact based on the input current state of the strategy optimization subject and the adaptive strategy set according to its own target and understanding of the environment, through multiple strategy selection and game processes, the competitive agent continuously adjusts its strategy to achieve the optimal game result, for example, in a financial investment scenario, the strategy optimization subject (such as an investment institution) adjusts its investment strategy through continuous adjustment to achieve optimal allocation of assets and maximize returns in a competitive environment.

[0048] Further, the specific configuration of the competition optimization layer 40 further includes recording the income data, risk data, decision performance data of all roles in the game process, establishing a feedback signal; taking the updated adaptive strategy set as the initial strategy set, taking the feedback signal as the optimization constraint, performing iterative update of the initial strategy set; completing game optimization according to the iterative update result.

[0049] Preferably, during the game between the strategy optimization subject and competing intelligent agents (dominant, follower, adversary, etc.), detailed data on all participating roles is recorded, including payoff data, risk data, and decision performance data. Payoff data includes the gains each role receives at each stage of the game, such as sales revenue and profits for companies in business competition games, and investment returns for investors in financial investment games. Risk data includes the risk situation faced by each role, such as market risk and credit risk, for example, the Value at Risk (VaR) of an investment portfolio and the probability of debt default for a company. Decision performance data includes the decisions made by each role during the game and the actual effects of these decisions, such as the timing of the decision, the strategy chosen, and the resulting changes in market share and product sales. Integrating the recorded payoff data, risk data, and decision performance data to obtain feedback signals comprehensively reflects the overall performance of each role in the game and the overall game situation, such as determining which strategies are effective, which decisions are problematic, and the gap between the current game state and the expected goal.

[0050] Preferably, the updated adaptive strategy set is used as the initial strategy set, and the established feedback signal is used as a constraint condition in the iterative update process to guide the optimization direction of the strategy. Specifically, the initial strategy set is iteratively updated using optimization algorithms (such as genetic algorithms, reinforcement learning algorithms, etc.). In each iteration, the strategies in the strategy set are adjusted, mutated, or combined according to the feedback signal to generate new strategies. The newly generated strategies are then evaluated to determine whether they are superior to the strategies in the original strategy set. If the new strategy is superior, it is included in the updated strategy set; otherwise, the original strategy is retained. Through multiple iterations, the strategy set is continuously optimized to obtain the iterative update result. Based on the iterative update result, game optimization is completed. That is, the strategy combination most suitable for the current game environment and the goals of each role is selected from the evaluated strategy set, which enables each role to maximize its own interests or achieve other predetermined goals in the competitive or cooperative game process.

[0051] Furthermore, the specific configuration of the competitive optimization layer 40 also includes: calculating the fitness value in the initial strategy set; using the fitness value to screen the initial strategy set and establish a superior individual identifier; reconstructing the iteration probability based on the superior individual identifier and then performing individual selection of the initial strategy set; performing gene exchange based on the individual selection result and updating the parameter contribution of the individual selection result using the optimization constraint; and completing an iteration update after randomly perturbing the parameters of the individual selection result, wherein the current individual selection result is the individual selection result after gene exchange and parameter contribution adjustment.

[0052] Preferably, the fitness value in the initial strategy set is calculated, that is, each evaluation index is quantitatively processed according to the profit data, risk data, decision performance data, etc., and the weight is allocated according to the importance, and then the advantages and disadvantages of each strategy in the initial strategy set are calculated by weighting, that is, the fitness value of each strategy is obtained; then the initial strategy set is screened according to the fitness value, the strategies with higher fitness value are retained, and the strategies with lower fitness value are eliminated, so as to remove the strategies that obviously do not meet the optimization target, reduce the range of the strategy set, and then the strategies (excellent individuals) retained after screening are identified; then the probability of each strategy being selected in the subsequent iteration process is reconstructed according to the excellent individual identification. Generally speaking, the excellent individuals with higher fitness value are given higher iteration probability; based on the reconstructed iteration probability, individuals (i.e. strategies) are selected from the initial strategy set by random sampling, a certain number of strategies are selected as the basis for the next round of iteration.

[0053] Preferably, the strategies in the individual selection result are subjected to gene exchange operation, that is, some gene fragments of different strategies are exchanged and combined to generate new strategies, wherein the gene refers to the parameter in the strategy, which is similar to gene recombination in biological evolution, and the purpose is to create better strategies by combining the advantages of different strategies, for example, in two investment strategies, one focuses on short-term returns and the other focuses on risk control, through gene exchange, a new strategy is generated which considers both short-term returns and risk control; then the contribution degree of each parameter in the individual selection result is updated by using the optimization constraint (i.e. the constraint condition based on the feedback signal), wherein the parameter contribution degree reflects the importance of each parameter in the strategy for achieving the optimization target. By analyzing the performance of the strategy under the optimization constraint, the contribution degree of each parameter is adjusted so that in the subsequent iteration, the parameters that contribute more to the optimization target can be adjusted more. Finally, the parameters of the individual selection result (i.e. the new strategy) after gene exchange and adjustment of the parameter contribution degree are subjected to random disturbance, that is, by randomly changing the parameter values of the strategy within a certain range, more possibilities of the strategy space are explored to avoid the algorithm falling into a local optimal solution, for example, the investment proportion parameter in the investment strategy is randomly increased or decreased, and the performance of the strategy under the new parameter is observed, and then one iteration update of the initial strategy set is completed. Compared with the initial strategy set, the fitness and the degree of approaching the optimal solution are improved.

[0054] Further, the specific configuration of the competition optimization layer 40 also includes configuring a quality screening threshold, and performing adaptive strategy screening on the adaptive evaluation result according to the quality screening threshold; the adaptive strategies that do not meet the quality screening threshold are removed, and the remaining adaptive strategies are used to update the adaptive strategy set.

[0055] Preferably, the quality screening threshold is a standard set according to actual business needs, used to measure the quality of adaptive strategies, and each adaptive strategy is evaluated by comparing its indicators with the pre-set quality screening threshold. For example, for an investment strategy, if the threshold of annual yield is set at 10% and the threshold of risk volatility is set at 15%, then the actual annual yield and risk volatility of each strategy are checked. If the annual yield of a strategy is lower than 10% or the risk volatility is higher than 15%, the strategy is considered to not meet the quality screening threshold. After comparison with the threshold, the adaptive strategies that do not meet the requirements are removed from the original strategy set, and only those strategies whose indicators meet or exceed the quality screening threshold are retained to form an updated adaptive strategy set, thereby improving the quality and effectiveness of the overall strategy.

[0056] Further, the specific configuration of the competition optimization layer 40 also includes obtaining a behavior pattern database of the competition agent, performing reverse reasoning according to the behavior pattern database to predict potential countermeasures of the competition agent, establishing a mapping game strategy according to the potential countermeasures, and performing compensation on the game optimization result according to the mapping game strategy.

[0057] Preferably, the behavior pattern database of the competition agent is established by collecting and organizing various behavior data of the competition agent in previous game processes, including decision-making choices, action methods, reaction times, and other information in different situations. Reverse reasoning is performed according to the behavior pattern database, that is, taking the past behavior results of the competition agent as the starting point, using the data in the behavior pattern database to analyze the motivations, goals, and decision-making rules behind the behavior. For example, if it is found that the competitor suddenly reduces the product price in a certain market area and obtains a higher market share, it can be inferred from the analysis of the relevant data in the database that the possible motivation is to exclude other competitors, expand market share, or clear inventory. In combination with the current game situation and information, the countermeasures that the competition agent may take in the future are predicted.

[0058] Preferably, the predicted potential response strategies of the competing agent are analyzed in depth to understand the strategy goals, advantages, disadvantages and possible impact on oneself, and then combined with the game goals and resources of oneself according to the potential response strategy analysis results, a mapping game strategy is established, that is, for each potential response strategy, what game strategy should be taken to respond, for example, if the competitor adopts a low price strategy, oneself can choose to respond by optimizing product cost, providing differentiated value-added services, strengthening brand marketing and other strategies to maintain market share and brand image while maintaining profits, and then determine the specific actions and decision-making schemes oneself should take in different competitive situations in response to different potential response strategies of the competitor; finally, the game optimization result is compensated according to the mapping game strategy, which may include adjusting the strategy, optimizing resource allocation, strengthening actions in some aspects, etc. Through continuous evaluation and compensation, the game result can be gradually optimized, and the competitiveness and benefits in the game can be improved.

[0059] Further, the specific configuration of the data processing layer 20 also includes extracting the data source after data cleaning into structured data and unstructured data; using a deep learning model to perform cross-modal feature fusion of structured data and unstructured data; and performing adaptive feature selection on the cross-modal feature fusion result to complete data feature extraction.

[0060] Preferably, the data source is cleaned, for example, missing values are handled, incorrect data is corrected, and duplicate data is removed, to improve data quality, and then the cleaned data source is divided into structured data and unstructured data according to data format and structure characteristics. Structured data refers to data with a clear structure and fixed format, such as table data in a database, each column has a specific data type and meaning; unstructured data has no fixed structure, such as text, image, audio, video, etc. Then the structured data and unstructured data are input into corresponding deep learning models for feature extraction, and the extracted features of different modalities are fused. The deep learning model (such as convolutional neural network, recurrent neural network or long short-term memory network) can automatically extract valuable features from a large amount of data, for example, for a data set containing images and text descriptions, first use a convolutional neural network (CNN) to extract visual features of the image, such as color, texture, shape, etc. Then use a long short-term memory network (LSTM) to extract semantic features of the text, such as the meaning of words, the structure of sentences, etc. The two feature vectors are spliced to form cross-modal features. Finally, according to the characteristics of the data and the task requirements, the most representative data features are automatically selected, which can better reflect the essential information of the original data and have higher quality and efficiency.

[0061] Further, the strategy dynamic optimization system based on big data analysis prediction further comprises a pre-warning layer configured to analyze the dynamic response strategy, generate a risk pre-warning signal, and report the risk pre-warning signal.

[0062] Preferably, the pre-warning layer analyzes the dynamic response strategy, that is, understands various rules, conditions and corresponding operation instructions contained in the strategy, and generates a corresponding risk pre-warning signal according to the current actual operation situation, which is used to prompt potential risks and contains key information about the risks, such as risk type, risk level, possible impact range, etc. Finally, the risk pre-warning signal is reported to obtain risk information in time and take corresponding measures to reduce the loss caused by the risk, so as to ensure that the strategy becomes stronger in an extreme fluctuation environment.

[0063] In the foregoing, with reference to Figure 1 The strategy dynamic optimization system based on big data analysis prediction according to the embodiment of the present application is described in detail. Next, the strategy dynamic optimization method based on big data analysis prediction according to the embodiment of the present application will be described with reference to Figure 2 The strategy dynamic optimization method based on big data analysis prediction, as shown in Figure 2 includes: collecting transaction data, news data, social data, search data, public data, and private user retention data to establish a data source; performing data feature extraction after data cleaning of the data source, identifying abnormal patterns according to the data feature extraction result, and constructing a strategy training environment; performing coping strategy training in an adversarial environment after injecting random noise into the strategy training environment, establishing an adaptive strategy set; configuring a competitive agent according to the data source, the competitive agent including a leader, a follower, and an opponent, updating the adaptive strategy set after adaptive strategy quality evaluation of the adaptive strategy set, performing game optimization using the updated adaptive strategy set and the competitive agent to generate a game optimization result; and receiving the game optimization result to establish a dynamic response strategy.

[0064] In a possible implementation manner, the strategy dynamic optimization method based on big data analysis prediction further comprises: establishing an extreme scenario data set and expanding the extreme scenario data set using a generative adversarial network to generate an updated extreme scenario data set; performing self-checking on injection of the random noise within a preset period, performing scenario coverage evaluation according to the self-checking result and the updated extreme scenario data set to establish an injection constraint of the random noise; configuring a time continuous disturbance constraint and establishing a risk factor, the risk factor including liquidity risk and credit risk; and performing random noise injection of the strategy training environment according to the injection constraint of the random noise, the time continuous disturbance constraint, and the risk factor.

[0065] In a possible implementation, the strategy dynamic optimization method based on big data analysis prediction further includes: monitoring indexes of the strategy optimization subject, establishing a key index set, the key index set including yield, risk exposure, liquidity, and resource utilization efficiency; establishing a current state of the strategy optimization subject according to the key index set; and using the current state and the updated adaptive strategy set as input data to perform game optimization with the competitive agent.

[0066] In a possible implementation, the strategy dynamic optimization method based on big data analysis prediction further includes: recording income data, risk data, and decision-making performance data of all roles in the game process to establish a feedback signal; using the updated adaptive strategy set as an initial strategy set, and using the feedback signal as an optimization constraint to perform iterative updating of the initial strategy set; and completing game optimization according to an iterative updating result.

[0067] In a possible implementation, the strategy dynamic optimization method based on big data analysis prediction further includes: calculating fitness values in the initial strategy set, using the fitness values to perform initial strategy set screening, and establishing an excellent individual identifier; performing individual selection of the initial strategy set after the excellent individual identifier is reconstructed according to an iteration probability; performing gene exchange according to the individual selection result, and updating parameter contribution degrees of the individual selection result using the optimization constraint; and completing one iteration update after parameters of the individual selection result are randomly disturbed, where the current individual selection result is the individual selection result after the gene exchange and the adjustment of the parameter contribution degrees.

[0068] In a possible implementation, the strategy dynamic optimization method based on big data analysis prediction further includes: configuring a quality screening threshold, performing adaptive strategy screening of the adaptation evaluation result according to the quality screening threshold; and removing adaptive strategies that do not meet the quality screening threshold, and updating the adaptive strategy set using the remaining adaptive strategies.

[0069] In a possible implementation, the strategy dynamic optimization method based on big data analysis prediction further includes: extracting a data source after data cleaning into structured data and unstructured data; using a deep learning model to perform cross-modal feature fusion of the structured data and the unstructured data; and performing self-adaptive feature selection on a cross-modal feature fusion result to complete data feature extraction.

[0070] In a possible implementation, the strategy dynamic optimization method based on big data analysis prediction further includes: obtaining a behavior pattern database of the competitive agent, performing reverse reasoning according to the behavior pattern database to predict a potential coping strategy of the competitive agent; establishing a mapping game strategy according to the potential coping strategy; and performing compensation on the game optimization result according to the mapping game strategy.

[0071] In a possible implementation, the strategy dynamic optimization method based on big data analysis prediction further comprises: a pre-warning layer configured to analyze the dynamic response strategy, generate a risk pre-warning signal, and report the risk pre-warning signal.

[0072] The strategy dynamic optimization system based on big data analysis prediction provided in the embodiments of the present application can execute the strategy dynamic optimization method based on big data analysis prediction provided in any of the embodiments of the present application, and has the corresponding function modules and beneficial effects of the execution method.

[0073] Although the present application makes various references to certain modules in the system according to the embodiments of the present application, however, any number of different modules can be used and run on the user terminal and / or server, and the various units and modules are only divided according to the functional logic, but are not limited to the above division, as long as the corresponding functions can be realized; in addition, the specific names of the functional units are only for the convenience of mutual differentiation, and do not limit the protection scope of the present application.

[0074] The above specific embodiments do not constitute a limitation on the protection scope of the present application. Those skilled in the art should understand that various modifications, combinations and substitutions can be made according to design requirements and other factors. Any modifications, equivalent replacements and improvements made within the spirit and principles of the present application should be included in the protection scope of the present application.

Claims

1. A strategy dynamic optimization system based on big data analysis and prediction, characterized in that, The system includes: The data acquisition layer is used to collect transaction data, news data, social data, search data, public data, and private domain user retention data to establish data sources; The data processing layer is used to clean the data from the data source, extract data features, identify abnormal patterns based on the data feature extraction results, and build a strategy training environment. An extreme scenario simulation layer is used to perform adversarial strategy training and establish an adaptive strategy set after random noise is injected into the strategy training environment. The competitive optimization layer is used to configure competitive agents according to the data source. The competitive agents include leaders, followers, and adversaries. After evaluating the adaptation quality of the adaptive strategy set, the adaptive strategy set is updated. The updated adaptive strategy set and the competitive agents are used to perform game optimization and generate game optimization results. The strategy dynamic optimization layer is used to receive the game optimization results and establish a dynamic response strategy. The extreme scenario simulation layer is also used for: An extreme scenario dataset is established, and the extreme scenario dataset is expanded using a generative adversarial network to generate an updated extreme scenario dataset. The random noise injection self-check is performed within a preset period. Based on the injection self-check results and the updated extreme scenario dataset, scene coverage evaluation is performed, and random noise injection constraints are established. Configure time-continuous disturbance constraints and establish risk factors, including liquidity risk and credit risk; Random noise is injected into the policy training environment based on the random noise injection constraints, the time-continuous perturbation constraints, and the risk factors. In the competitive optimization layer, feedback signals are used as optimization constraints to perform iterative updates of the initial policy set, including: Calculate the fitness value in the initial strategy set, use the fitness value to screen the initial strategy set, and establish the identifier of superior individuals; After reconstructing the iterative probabilities based on the superior individual identifiers, the individual selection of the initial strategy set is performed; Gene exchange is performed based on the individual selection results, and the parameter contribution of the individual selection results is updated using the optimization constraints. After randomly perturbing the parameters of the individual selection results, an iterative update is completed, where the current individual selection results are the individual selection results after gene exchange and parameter contribution adjustment; In the data processing layer, after cleaning the data source, data feature extraction is performed, including: The cleaned data source is extracted into structured data and unstructured data; Using deep learning models to fuse cross-modal features of structured and unstructured data; Adaptive feature selection is performed on the cross-modal feature fusion results to complete data feature extraction; The competition optimization layer is also used for: Monitor the main body of strategy optimization by indicators and establish a set of key indicators, which includes return on investment, risk exposure, liquidity, and resource utilization efficiency. The current state of the strategy optimization subject is established based on the set of key indicators; Using the current state and the updated set of adaptation strategies as input data, game optimization is performed against the competing agent.

2. The strategy dynamic optimization system based on big data analysis and prediction as described in claim 1, characterized in that, In the competitive optimization layer, using the current state and the updated set of adaptation strategies as input data, the game optimization with the competing agent includes: Record the payout data, risk data, and decision-making performance data of all roles during the game, and establish feedback signals; The updated adaptation strategy set is used as the initial strategy set, and the feedback signal is used as the optimization constraint to perform iterative updates of the initial strategy set. Game optimization is completed based on the iterative update results.

3. The strategy dynamic optimization system based on big data analysis and prediction as described in claim 1, characterized in that, In the competitive optimization layer, after evaluating the adaptation quality of the adaptation strategy set, the adaptation strategy set is updated, including: Configure a quality screening threshold, and then perform adaptation strategy screening based on the adaptation evaluation results according to the quality screening threshold; After removing adaptation strategies that do not meet the quality screening threshold, the set of adaptation strategies is updated with the retained adaptation strategies.

4. The strategy dynamic optimization system based on big data analysis and prediction as described in claim 1, characterized in that, The competition optimization layer is also used for: Obtain the behavior pattern database of the competing intelligent agent, and perform reverse reasoning based on the behavior pattern database to predict the potential response strategies of the competing intelligent agent; Establish a mapping game strategy based on the potential coping strategies; The game optimization result is compensated according to the mapping game strategy.

5. The strategy dynamic optimization system based on big data analysis and prediction as described in claim 1, characterized in that, The system also includes: The early warning layer is used to analyze the dynamic response strategy, generate risk warning signals, and issue risk warnings based on the risk warning signals.

6. A strategy dynamic optimization method based on big data analysis and prediction, characterized in that, The method is applied to the strategy dynamic optimization system based on big data analysis and prediction as described in any one of claims 1-5, and the method includes: Collect transaction data, news data, social data, search data, public data, and private domain user retention data to establish data sources; After cleaning the data source, perform data feature extraction, identify abnormal patterns based on the data feature extraction results, and build a strategy training environment. After injecting random noise into the strategy training environment, the coping strategy training in the adversarial environment is performed to establish an adaptive strategy set; The competitive agents are configured according to the data source. The competitive agents include a leader, a follower, and an adversary. After the adaptation quality of the adaptive strategy set is evaluated, the adaptive strategy set is updated. The game optimization is performed using the updated adaptive strategy set and the competitive agents to generate the game optimization result. Receive the game optimization results and establish a dynamic response strategy.

Images