Intelligent strategy deduction method and device for machine self-game

Abstract

The embodiment of the invention provides an intelligent strategy deduction method and device for machine self-game, and the method comprises the steps: constructing a game event agent, and extracting deduction elements of a to-be-deduced event based on the game event agent; constructing a decision preference matrix added with an interference coefficient; an intelligent agent self-game model is constructed based on the deduction elements and the decision preference matrix, multiple groups of game scenes and game paths are generated based on the intelligent agent self-game model, and the intelligent agent self-game model comprises a large model based on a Transformer architecture and a small model based on a deep Q network; and constructing a strategic evaluation model, performing risk evaluation and trend prediction on the multiple groups of game scenes and game paths based on the strategic evaluation model, and generating a deduction report.

Description

Technical Field

[0001] The embodiments of the present disclosure relate to the field of artificial intelligence technology, and in particular, to a method and device for intelligent strategy deduction of machine self-game. Background Art

[0002] Strategic deduction is an important support tool for enterprise development planning and strategic decision-making. It helps decision makers evaluate and optimize strategic choices when facing uncertainty and complexity by simulating and analyzing complex situations. Whether in the military or business fields, strategic deduction can help organizations improve the quality of decision-making, enhance their ability to respond to emergencies, and continuously optimize strategic plans in practice.

[0003] The shortcomings of the traditional strategic game model are mainly reflected in the time-consuming preparation, high professional threshold, reliance on expert wisdom, and difficulty in ensuring the effectiveness of the results. In the face of these challenges, finding and implementing innovative game methods to improve the efficiency of games, reduce excessive reliance on experts, and enhance the effectiveness of games has become an urgent problem that decision makers and game organizers need to solve.

[0004] Currently, artificial intelligence has been integrated into strategic deduction design. For example, large language models (LLMs) are being widely used in multi-agent systems. LLMs can understand and generate natural language texts, making communication between agents more natural and efficient. This ability is essential for strategy formulation and execution in complex game environments. LLMs are pre-trained with massive data and have extensive knowledge and reasoning capabilities. They can quickly learn and generate effective strategies during the game. In addition, LLMs can also adjust strategies in real time to respond to changes in opponent behavior through interactive learning. Therefore, the application of artificial intelligence technology provides new directions and approaches for improving the efficiency and optimizing processes of strategic deduction. Summary of the invention

[0005] The embodiments described herein provide a method and apparatus for intelligent strategy deduction for machine self-game, and a computer-readable storage medium storing a computer program.

[0006] According to a first aspect of the present disclosure, there is provided a method for intelligent strategic deduction of machine self-game, comprising: constructing a game event agent, and extracting deduction elements of the event to be deduced based on the game event agent; constructing a decision preference matrix with an interference coefficient added; constructing an intelligent agent self-game model based on the deduction elements and the decision preference matrix, and generating multiple groups of game scenarios and game paths based on the intelligent agent self-game model, wherein the intelligent agent self-game model comprises a large model based on a Transformer architecture and a small model based on a deep Q network; and constructing a strategic evaluation model, and performing risk evaluation and trend prediction on multiple groups of game scenarios and game paths based on the strategic evaluation model to generate a deduction report.

[0007] In some embodiments of the present disclosure, constructing a game event agent includes: collecting raw text data from multiple data sources, and performing data cleaning and preprocessing on the collected raw text data; labeling key elements of the preprocessed data to obtain a training data set, wherein the key elements include event type, event background, participants, and decision-making style; training a large language model based on the training data set and a strategy optimization algorithm, introducing new training data to the large language model during the training process, and adjusting the parameters of the model to obtain a trained large language model; and encapsulating the trained large language model into an agent, and integrating multiple algorithm processing logics in the agent.

[0008] In some embodiments of the present disclosure, data sources include news media, academic literature, policy documents, social media, and historical cases, and deduction elements include game name, theme, deduction background, deduction purpose, deduction party, initial value setting of deduction party indicators, and initial situation events.

[0009] In some embodiments of the present disclosure, constructing a decision preference matrix with interference coefficients added includes: encoding the characteristic vector and preference vector of the enterprise, and the kth element of the preference vector is expressed as:

[0010] In the formula, w ij is the weight coefficient of the strategy, P ij represents the enterprise's preference for strategy j in market scenario i, m is the total number of strategies, and n is the number of market scenarios; the interference matrix R is constructed using the Monte Carlo method ij , R ij = Random(0.5, 1); before each game, according to P k ×R ij Determine the decision preference vectors of all parties in this game and obtain the decision preference matrix with the interference coefficient added.

[0011] In some embodiments of the present disclosure, the characteristic vector of an enterprise includes enterprise size, financial status, technological level and market share, and the preference vector is the degree of inclination of the enterprise towards different strategic directions, including expansion, contraction, innovation and cost control.

[0012] In some embodiments of the present disclosure, an intelligent agent self-game model is constructed based on deduction elements and decision preference matrix, and multiple sets of game scenarios and game paths are generated based on the intelligent agent self-game model, including: inputting feature vectors and decision preference vectors into a large model based on Transformer architecture, capturing the relationship between feature vectors and decision preference vectors through a self-attention mechanism, and generating a preliminary strategic framework for the enterprise. The self-attention mechanism is Where Q, K, and V are query, key, and value vectors respectively,k The key vector dimension is used; the preliminary strategic framework is input into a small model based on a deep Q network, and the parameters of the deep Q network model are fine-tuned through the back-propagation algorithm in combination with the historical strategic data of the enterprise and the current market sample data, so as to refine the initial strategic framework; the external environment of the enterprise is continuously monitored, and the external environmental factors are input into the large model, so that the large model can predict market changes or policy trends based on the input external environmental factors, provide strategic adjustment suggestions, and the small model can update the Q value based on the strategic adjustment suggestions and the current market status and competitive environment; the large model based on the Transformer architecture and the small model based on the deep Q network are constructed to obtain an intelligent agent self-game model; and the intelligent agent self-game model is used to automatically conduct multiple rounds of game deductions to generate game scenarios and game paths covering a variety of strategy combinations.

[0013] In some embodiments of the present disclosure, the deep Q network includes multiple convolutional layers and fully connected layers. The convolutional layers are used to extract market competition situation features, and the fully connected layers are used to output strategic decisions. The loss function of the deep Q network is: where y i is the target Q value, s i , a j is the state-action pair, θ is the model parameter, and after each strategic decision is implemented, the Q value is updated according to the new market state and the reward obtained. The formula is:

[0014] Q(s,a)=Q(s,a)+α[r+γmax a′ Q(s′, a′)-Q(s, a)],

[0015] Among them, α is the learning rate, γ is the discount factor, r is the reward obtained, (s, a) is the current state-action pair, and (s′, a′) is the new state-action pair.

[0016] In some embodiments of the present disclosure, a strategic evaluation model is constructed, and risk assessment and trend prediction are performed on multiple groups of game scenarios and game paths based on the strategic evaluation model, and a deduction report is generated, including: calculating the immediate benefits of each game path, evaluating the execution difficulty, required resources, reputation risk, opponent reaction strategy and long-term effects of each strategy after execution, and screening out the strategy combination with the greatest benefits under different decision-making styles; analyzing the historical behavior of the game parties and the current game path, predicting the game parties' strategic direction, resource allocation preference and decision-making style evolution trend; establishing a dynamic event model to simulate market changes or emergencies, and predicting the behavior of the game parties in different scenarios based on the current game path; and automatically generating a standardized game deduction report according to a predetermined format, the content of which includes: game path and decision analysis; benefit, risk and long-term impact evaluation of each strategy; strategic tendency analysis of each game party; trend prediction of the dynamic event model; response plan and risk mitigation measures.

[0017] According to the second aspect of the present disclosure, an intelligent strategic deduction device for machine self-game is provided. The device includes at least one processor; and at least one memory storing a computer program. When the computer program is executed by at least one processor, the device: constructs a game event agent, and extracts the deduction elements of the event to be deduced based on the game event agent; constructs a decision preference matrix with an interference coefficient added; constructs an intelligent agent self-game model based on the deduction elements and the decision preference matrix, and generates multiple groups of game scenarios and game paths based on the intelligent agent self-game model, and the intelligent agent self-game model includes a large model based on the Transformer architecture and a small model based on the deep Q network; and constructs a strategic evaluation model, and performs risk assessment and trend prediction on multiple groups of game scenarios and game paths based on the strategic evaluation model, and generates a deduction report.

[0018] In some embodiments of the present disclosure, when the computer program is executed by at least one processor, the device constructs a gaming event agent by:

[0019] Collect raw text data from multiple data sources, and perform data cleaning and preprocessing on the collected raw text data; annotate the key elements of the preprocessed data to obtain a training data set, where the key elements include event type, event background, participants, and decision-making style; train a large language model based on the training data set and a strategy optimization algorithm, introduce new training data to the large language model during the training process, and adjust the parameters of the model to obtain a trained large language model; and encapsulate the trained large language model into an intelligent agent, and integrate multiple algorithm processing logics into the intelligent agent.

[0020] In some embodiments of the present disclosure, when the computer program is executed by at least one processor, the device constructs a decision preference matrix adding interference coefficients by performing the following operations:

[0021] Encode the enterprise's characteristic vector and preference vector, and the kth element of the preference vector is expressed as:

[0022] In the formula, w ij is the weight coefficient of the strategy, P ij represents the enterprise's preference for strategy j in market scenario i, m is the total number of strategies, and n is the number of market scenarios; the interference matrix R is constructed using the Monte Carlo method ij , R ij = Random(0.5, 1); before each game, according to P k ×R ij Determine the decision preference vectors of all parties in this game and obtain the decision preference matrix with the interference coefficient added.

[0023] In some embodiments of the present disclosure, when the computer program is executed by at least one processor, the device generates multiple sets of game scenarios and game paths by performing the following operations: inputting the feature vector and the decision preference vector into a large model based on the Transformer architecture, capturing the relationship between the feature vector and the decision preference vector through a self-attention mechanism, and generating a preliminary strategic framework for the enterprise. The self-attention mechanism is Where Q, K, and V are query, key, and value vectors respectively, k is the key vector dimension;

[0024] The preliminary strategic framework is input into a small model based on a deep Q network, and the parameters of the deep Q network model are fine-tuned through the back-propagation algorithm in combination with the company's historical strategic data and current market sample data, so as to refine the initial strategic framework; the external environment of the company is continuously monitored, and external environmental factors are input into the large model, so that the large model can predict market changes or policy trends based on the input external environmental factors, provide strategic adjustment suggestions, and the small model can update the Q value based on the strategic adjustment suggestions and the current market status and competitive environment; the large model based on the Transformer architecture and the small model based on the deep Q network are constructed to obtain an intelligent agent self-game model; and the intelligent agent self-game model is used to automatically conduct multiple rounds of game deductions to generate game scenarios and game paths covering a variety of strategy combinations.

[0025] In some embodiments of the present disclosure, when the computer program is executed by at least one processor, the device generates a deduction report by performing the following operations:

[0026] Calculate the immediate benefits of each game path, evaluate the execution difficulty, required resources, reputation risk, opponent response strategy and long-term effects of each strategy after execution, and screen out the strategy combination with the highest benefits under different decision-making styles; analyze the historical behavior and current game path of the game parties, predict the game parties' strategic direction, resource allocation preference and decision-making style evolution trend; establish a dynamic event model to simulate market changes or emergencies, and predict the behavior of the game parties in different scenarios based on the current game path; and automatically generate a standardized game deduction report according to a predetermined format. The game deduction report includes: game path and decision analysis; benefit, risk and long-term impact assessment of each strategy; strategic tendency analysis of each game party; trend prediction of dynamic event model; response plan and risk mitigation measures.

[0027] According to a third aspect of the present disclosure, a computer-readable storage medium storing a computer program is provided, wherein the computer program, when executed by a processor, implements the steps of the intelligent strategy deduction method for machine self-game according to the first aspect of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] In order to more clearly illustrate the technical solutions of the embodiments of the present disclosure, the drawings of the embodiments will be briefly described below. It should be noted that the drawings described below only relate to some embodiments of the present disclosure, but are not intended to limit the present disclosure, wherein:

[0029] Figure 1 is an exemplary flow chart of intelligent strategy deduction of machine self-game according to an embodiment of the present disclosure;

[0030] Figure 2 An exemplary flow chart showing an intelligent strategy deduction method 200 for machine self-game according to an embodiment of the present disclosure;

[0031] Figure 3 A flowchart of constructing an intelligent agent self-game model according to an embodiment of the present disclosure is shown;

[0032] Figure 4 It is a schematic block diagram of an intelligent strategy deduction device 400 for machine self-game according to an embodiment of the present disclosure.

[0033] It should be noted that the elements in the drawings are schematic and not drawn to scale. DETAILED DESCRIPTION

[0034] In order to make the purpose, technical solution and advantages of the embodiments of the present disclosure clearer, the technical solution of the embodiments of the present disclosure will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present disclosure, rather than all of the embodiments. Based on the described embodiments of the present disclosure, all other embodiments obtained by those skilled in the art without creative work also fall within the scope of protection of the present disclosure.

[0035] Unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by those skilled in the art to which the subject matter of the present disclosure belongs. It will be further understood that terms such as those defined in commonly used dictionaries should be interpreted as having a meaning consistent with their meanings in the context of the specification and the relevant art, and will not be interpreted in an idealized or overly formal form unless otherwise explicitly defined herein. In addition, terms such as "first" and "second" are only used to distinguish one component (or a portion of a component) from another component (or another portion of a component).

[0036] The use of artificial intelligence technology for decision-making simulation and optimization, especially the innovation of generative artificial intelligence and collaborative scheduling technology of large and small models, provides new directions and ways to improve the efficiency of strategic deduction.

[0037] In response to the problems of strategic deduction and decision optimization in complex environments, the present invention proposes an intelligent strategic deduction method and device for machine self-game. Figure 1 The overall process design diagram of the intelligent strategic deduction method according to the embodiment of the present disclosure is shown. Figure 1 As shown, first, create the game.

[0038] Through the game event agent, hot events are deeply analyzed at multiple levels and angles. The game event agent uses natural language processing, semantic analysis and information extraction technologies to extract the core elements of the event from massive data, including the deduction background, participants, involved fields, themes, etc., to ensure the accuracy and pertinence of subsequent game simulations.

[0039] Construct a decision preference matrix. Through deep learning and historical data analysis, the decision preferences of each participant are quantified, and then their strategic intentions and policy propositions in different situations are generated to construct a decision preference matrix. The matrix mainly reflects the decision-making tendencies of the participants and provides personalized and diversified strategy inputs for subsequent game simulations.

[0040] Then, based on the decision preference matrix and event background, the intelligent strategic self-game model generates game agents representing each participant (Enterprise A, Enterprise B, and Enterprise C), and uses the collaborative scheduling of large and small models to conduct automatic games, simulate the dynamic development of events, explore the optimal strategy path, and achieve self-iteration and optimization of the strategy.

[0041] Finally, the strategic assessment model deeply integrates and analyzes the results of multiple rounds of games. The strategic assessment model uses advanced data analysis, trend forecasting and risk assessment technology to extract the development trends, potential risks and response strategies of hot events, and automatically generates detailed analysis reports and trend analysis reports to provide decision makers with fast and accurate decision-making references.

[0042] Figure 2 An exemplary flow chart of an intelligent strategy deduction method 200 for machine self-game according to an embodiment of the present disclosure is shown.

[0043] exist Figure 2 In box S202, a game event agent is constructed, and deduction elements of the event to be deduced are extracted based on the game event agent.

[0044] The process of building a game event intelligent agent is a systematic, data-driven task, covering the entire process from data collection to model training and then to application. The purpose is to accurately analyze complex game events through the integration and intelligent analysis of multi-source data, and provide a scientific basis for decision support and game simulation.

[0045] In order to enable the constructed intelligent agent to fully understand the game events, according to an embodiment of the present disclosure, raw text data is collected from multiple data sources, and the collected raw text data is cleaned and preprocessed. These raw text data can cover different types of game events and their background information. For example, relevant texts about current affairs, hot events, economic competition, etc. are obtained from mainstream news websites, online news platforms and various special reports. These texts are usually highly timely and comprehensive. Content is collected from academic journals, research papers, conference records, etc., especially research results in the fields of social sciences, game theory, economics, international relations, etc., in order to provide theoretical support and in-depth analysis for the model. In addition, policy documents, government reports, various official announcements, etc. are also key sources for obtaining decision-making backgrounds, stakeholders and policy analysis. Collecting user-generated content through social platforms such as Twitter and Weibo can help the model understand public sentiment, differences of opinion and its response to game events. And, through historical case analysis of the deduction model of similar events, understand the long-term evolution and impact of the behavior of all parties in the event.

[0046] Then, perform data cleaning and preprocessing on multi-source data to ensure that the obtained text data is suitable for model training. For example, remove repeated, irrelevant, and redundant texts, unify text formats, and solve format inconsistencies in different data sources. Use technologies such as word segmentation and part-of-speech tagging to preprocess the text in preparation for subsequent in-depth analysis and annotation. Automatically identify key information in the text, such as the time, location, participants, and behavior of the event. Through natural language processing technology (such as sentiment analysis, semantic role tagging, etc.), extract deep information from the text and understand the potential game patterns and decision logic involved in the text.

[0047] Data annotation can be used to annotate the basic components of the game, such as the game event type, event background, key participants, decision-making style, and initial conditions, based on the characteristics of the game event, to provide clear learning goals and labels for the large model. For example, event types include economic games, etc. Event backgrounds include historical backgrounds, economic conditions, etc. Key participants in the event are annotated, such as countries, enterprises, organizations, individuals, etc., and the interests, positions, resources, etc. of each participant are further analyzed. The decision-making styles of each party include cooperation, confrontation, compromise, etc., as well as possible strategies (such as cooperation strategies and betrayal strategies in games). Key event factors include the cause of the conflict, resource allocation, and transaction terms. Initial conditions are set for each participant, including: resource quantity (such as fiscal funds, natural resources, etc.), economic strength (such as GDP, market share, etc.), public support (such as public support, public opinion orientation), etc. Define the specific situation at the beginning of the game, including all observable variables and initial states, which are the basis for game deduction.

[0048] In order to ensure the quality of the training corpus, the validity and accuracy of the data can also be ensured by combining expert review with automation. After the data is integrated and labeled, the large language model can be trained and optimized based on the training data set and the strategy optimization algorithm, so that the trained model can parse the key elements of the game event. For example, using pre-trained language models such as Transformer and BERT, training is carried out in game theory, situation analysis, strategy selection, etc. to help the model better understand the deep logic of the game. Among them, the key elements include the name of the game, theme, deduction background, deduction purpose, deduction party, deduction party indicator initial value setting, and initial situation events. With the continuous emergence of new data, the model is regularly updated through incremental learning, that is, new training data is introduced to the large language model during the training process, so that it always maintains sensitivity to new game events, and the model parameters are adjusted online to obtain a trained large language model. Finally, the trained large model is encapsulated into an intelligent agent, and a variety of algorithm processing logics are integrated, so that the intelligent agent can comprehensively apply different algorithm processing logics when facing complex game situations.

[0049] Reference Figure 2 As shown, at block S204, a decision preference matrix with interference coefficients added is constructed.

[0050] When constructing a decision preference matrix, the decision support model is mainly formed by encoding the characteristic vector and preference vector of the enterprise. Among them, the characteristic vector of the enterprise includes enterprise size, financial status, technological level, market share, etc. For example, enterprise size is usually quantified based on the number of employees, turnover or total assets. Financial status can be expressed by indicators such as profit margin, debt ratio, and capital structure. The technological level can be measured by the proportion of R&D investment and the number of technical patents. Market share reflects the market share of an enterprise in the industry, usually as a percentage.

[0051] These characteristic vectors are standardized to convert data of different dimensions into dimensionless data so that they can be input into the model for subsequent analysis. Commonly used methods of standardization include Z-score standardization and Min-Max standardization. After standardization, the characteristic vectors of each enterprise can be directly compared with other enterprises.

[0052] The preference matrix is ​​used to represent the degree of inclination of enterprises towards different strategic directions in different market situations and convert it into a preference vector Assume that an enterprise may face different market scenarios. For example, market scenario 1 is a highly competitive market, market scenario 2 is a market with great growth potential, and market scenario 3 is a mature market. In each market scenario, the enterprise may have different strategic choices, such as expansion, contraction, innovation, and cost control. On this basis, construct an n×m preference matrix P, where each element P ijrepresents the firm's preference for strategy j in market situation i. The preference can be determined from sources such as the firm's historical data, industry analysis, and strategic goals. For example, P 11 In Scenario 1, the company’s preference for the “expansion” strategy, P 12 Indicates the degree to which the enterprise prefers the "contraction" strategy in scenario 1.

[0053] In order to more accurately quantify the preference for each strategy, the preference vector needs to be further refined and combined with the focus of the enterprise strategy. Assume that a certain weight coefficient is assigned to the strategic direction in each market scenario. These weights reflect the importance of each strategy in different scenarios. The kth element of can be calculated as w ij is the weight coefficient of the strategy, reflecting the importance of the strategy, m is the total number of strategies, and n is the number of market scenarios. Through this calculation method, the preference vector in each market scenario can reflect the strategic tendency of the enterprise.

[0054] In multi-party games, in addition to the decision preference matrix, the strategic choices of enterprises will also be affected by various external factors and uncertainties (such as market fluctuations, policy changes, technological innovation, etc.). Therefore, complex scenario interference coefficients are added to the decision model. In order to simulate these uncertainties, the Monte Carlo method can be used to construct the interference matrix, and the interference coefficient can be added during the game process to enhance the reality and adaptability of the model.

[0055] According to an embodiment of the present disclosure, the interference coefficient usually fluctuates within a specific range, such as 0.5 to 1, which means that the interference coefficient will reduce the original decision preference degree (less than 1) or remain unchanged (equal to 1). Using the Monte Carlo method, the interference matrix R is constructed. n×m , R ij Indicates a random number (0.5~1), R ij = Random(0.5, 1), this random number can be generated by uniform distribution, or distribution fitting based on historical data. Before each game, according to P k ×R ij Determine the decision preference vectors of all parties in this game and obtain the decision preference matrix with interference coefficient added As the input precondition of the game model. During the game, each party will choose the most appropriate strategy based on these adjusted preference vectors and conduct game analysis. In this way, the intelligent agent can make more accurate and realistic strategic choices based on actual market uncertainties.

[0056] In various market scenarios, the intelligent agent can provide reasonable strategic selection suggestions for enterprises based on the preference of different strategies and the characteristics of enterprises. For example, when the market scenario changes, enterprises can adjust their strategic choices based on the preference matrix, and the decision preferences of enterprises will be adjusted according to the uncertainty before each game, thereby improving the actual effect and accuracy of decision-making.

[0057] Then in box S206, an intelligent agent self-game model is constructed based on the deduction factors and the decision preference matrix, and multiple sets of game scenarios and game paths are generated based on the intelligent agent self-game model. The intelligent agent self-game model includes a large model based on the Transformer architecture and a small model based on the deep Q network.

[0058] Enterprise-level strategic deduction usually involves multiple factors, such as market environment, competitor dynamics, enterprise's own resources and capabilities, etc. In order to deal with these complex influencing factors, building an intelligent agent self-game model can help enterprises make dynamic adjustments and decisions according to actual situations. This process is a synergy of large and small models to jointly build a framework for enterprise strategic deduction and optimization.

[0059] Figure 3 FIG. 1 shows a flow chart of constructing an intelligent agent self-game model according to an embodiment of the present disclosure. Figure 3 As shown, first, an initial strategic framework is generated based on the big model.

[0060] In some embodiments of the present disclosure, the feature vector and the decision preference vector are input into a large model based on the Transformer architecture, and the relationship between the feature vector and the decision preference vector is captured through the self-attention mechanism to generate a preliminary strategic framework for the enterprise. The self-attention mechanism is Where Q, K, and V are query, key, and value vectors respectively, k is the key vector dimension.

[0061] Specifically, the big model receives the enterprise's feature vector (such as market size, competitive situation, enterprise resources, etc.) and preference vector (such as corporate strategic preferences, market goals, etc.). Through the self-attention mechanism ( Where Q, K, and V are query, key, and value vectors respectively, kThe key vector dimension captures the relationship between input features and decision preference vectors, thereby identifying the importance and mutual influence of various factors. Specifically, the query vector (Q) represents the needs or problems of the current strategy. The key vector (K) represents the potential strategic direction or market information. The value vector (V) is the specific content extracted from the market and strategic information. The model generates a preliminary strategic framework through these relationships and outputs the market areas that the company should focus on, the direction of cultivating core competitiveness, etc. For example, decide whether to increase R&D investment and expand into emerging markets.

[0062] Subsequently, the initial strategic framework is refined based on the small model. At this point, the preliminary strategic framework is input into the small model based on the deep Q network, combined with the company's historical strategic data and current market sample data, and the deep Q network model parameters are fine-tuned through the back propagation algorithm to refine the initial strategic framework. The specific steps are as follows:

[0063] The DQN model includes multiple levels of neural networks: the convolutional layer is used to extract features such as market competition and identify the main factors affecting market changes. The fully connected layer is used to output specific strategic decisions based on the extracted features. The strategic framework generated by the large model is used as the initial input of the DQN model. The parameters of the DQN model are adjusted through the back propagation algorithm using historical data and sample data to optimize strategic decisions. The loss function of DQN is:

[0064] where y i is the target Q value, s i , a j is a state-action pair (such as market state and corresponding strategic action), θ is a DQN parameter, and the loss function is optimized using stochastic gradient descent (SGD). After multiple training and adjustments, the DQN model can refine the company's strategic decisions to make them more in line with market demand and corporate goals.

[0065] In the process of collaborative calling of the large model and the small model, the strategic parameters are adjusted based on environmental detection. The large model not only provides a preliminary strategic framework, but also continuously monitors the external environment of the enterprise, including market trends, changes in policies and regulations, and technological innovations, and timely provides macro-directional suggestions for strategic adjustments to the DQN small model. The large model encodes these external environmental factors into vector inputs by analyzing information such as industry news, government policies, and competitor announcements. The large model predicts possible market changes or policy trends based on the input environmental factors. For example, when the market is predicted to undergo a transformation, the large model will recommend that the enterprise adjust its strategy, such as turning to new technologies or market areas. The large model provides macro-strategic adjustment suggestions to the DQN model, such as suggesting reallocation of resources when the market turns. Based on these suggestions, the DQN small model combines its current strategic status and its perception of the local market competition situation (such as recent actions of competitors, the performance of its own products in the market segments, etc.) to further optimize strategic decisions through SGD. After each strategic decision is implemented, the Q value is updated according to the new market status s′ and the reward r obtained (such as market share growth, profit improvement, etc.), and the formula is:

[0066] Q(s,a)=Q(s,a)+α[r+γmax a′ Q(s′, a′)-Q(s, a)], where α is the learning rate, γ is the discount factor, r is the reward obtained, (s, a) is the current state-action pair, and (s, a′) is the new state-action pair.

[0067] Through the above steps, a large model based on the Transformer architecture and a small model based on the deep Q network are constructed to obtain an intelligent agent self-game model for strategic game deduction. Through the synergy of the large model and the small model, it is possible to simulate the possible behavior of competitors and make corresponding strategic adjustments. The large model analyzes text information such as industry news, competitor announcements, and financial reports to explore the potential intentions behind the opponent's market actions. The small model (DQN) predicts the opponent's possible actions based on historical data and current market dynamics. For example, if it is predicted that the opponent will reduce prices in a certain product area, the DQN model can decide whether to follow up with price cuts or respond through differentiation strategies such as increasing product added value.

[0068] Through the intelligent agent self-game model, multiple rounds of game deductions are automatically carried out to generate game scenarios and game paths covering multiple strategy combinations. The intelligent agent makes multiple rounds of decisions during the game process, not only considering its own strategy, but also predicting the opponent's possible actions and adjusting its own strategic response strategy. For example, the intelligent agent can adjust its own resource allocation and marketing strategy in a timely manner according to the opponent's pricing strategy, market promotion and other behaviors. Through the self-game mechanism, the intelligent agent simulates the game path under different scenarios and explores the optimal response strategy. This process can generate multiple strategy combinations and provide comprehensive analysis results for the subsequent strategic evaluation model. Through multiple rounds of game deductions, the intelligent agent generates multiple possible game paths, providing diversified and comprehensive evaluation results for corporate strategic decisions, and helping companies make the best decisions in a complex and uncertain market environment.

[0069] Finally, in box S208, a strategic evaluation model is constructed, and risk assessment and trend prediction are performed on multiple sets of game scenarios and game paths based on the strategic evaluation model to generate a deduction report.

[0070] According to an embodiment of the present disclosure, after multiple rounds of machine game deduction, a series of strategic results are obtained by deeply analyzing the game path, evaluating risks and benefits, and identifying the strategic tendencies of all parties, providing decision makers with detailed response plans and forecast reports.

[0071] The specific analysis elements of multiple game paths include: evaluating the direct economic benefits of each plan based on current market conditions, product pricing, marketing strategies and other factors; considering the long-term effects after the implementation of the strategy, such as the improvement of brand value and the growth of market share; and including synergistic effects in terms of market competition, resource consumption, and technological innovation.

[0072] It can calculate the immediate benefits of each game path, evaluate the execution difficulty, required resources, reputation risk, opponent reaction strategy and long-term effects of each strategy after execution, and screen out the strategy combination with the highest benefits under different decision-making styles. Analyze the historical behavior and current game path of the game parties, and predict the strategic direction, resource allocation preference and decision-making style evolution trend of the game parties. Establish a dynamic event model to simulate market changes or emergencies, and predict the behavior of the game parties in different situations based on the current game path. Finally, according to the predetermined format, a standardized game deduction report is automatically generated. The game deduction report includes: game path and decision analysis; evaluation of the benefits, risks and long-term impact of each strategy; analysis of the strategic tendency of each game party; trend prediction of the dynamic event model; response plan and risk mitigation measures.

[0073] Figure 4 4 is a schematic block diagram of an intelligent strategy deduction device 400 for machine self-game according to an embodiment of the present disclosure. Figure 4As shown, the device 400 may include a processor 410 and a memory 420 storing a computer program. When the computer program is executed by the processor 410, the device 400 may perform the following operations: Figure 2 The steps of method 200 shown. In one example, the device 400 can be a computer device or a cloud computing node. The device 400 can construct a game event agent, and extract the deduction elements of the event to be deduced based on the game event agent. The device 400 can construct a decision preference matrix with an interference coefficient added. In response to the construction of the game event agent and the decision preference matrix, the device 400 can construct an intelligent agent self-game model based on the deduction elements and the decision preference matrix, and generate multiple groups of game scenarios and game paths based on the intelligent agent self-game model. The intelligent agent self-game model includes a large model based on the Transformer architecture and a small model based on the deep Q network. In response to the intelligent agent self-game model generating multiple groups of game scenarios and game paths, the device 400 can construct a strategic evaluation model, and perform risk assessment and trend prediction on multiple groups of game scenarios and game paths based on the strategic evaluation model to generate a deduction report.

[0074] In some embodiments of the present disclosure, the device 400 can collect raw text data from multiple data sources, perform data cleaning and preprocessing on the collected raw text data; label key elements of the preprocessed data to obtain a training data set, where the key elements include event type, event background, participants, and decision-making style; train a large language model based on the training data set and a strategy optimization algorithm, introduce new training data to the large language model during the training process, adjust the parameters of the model to obtain a trained large language model; and encapsulate the trained large language model into an intelligent agent, and integrate multiple algorithm processing logics in the intelligent agent.

[0075] In some embodiments of the present disclosure, the apparatus 400 may encode the characteristic vector and preference vector of the enterprise, and the kth element of the preference vector is expressed as:

[0076] In the formula, w ij is the weight coefficient of the strategy, P ij It represents the firm’s preference for strategy j in market scenario i, m is the total number of strategies, and n is the number of market scenarios;

[0077] The interference matrix R is constructed using the Monte Carlo method. ij , R ij = Random(0.5, 1); before each game, according to P k ×R ij Determine the decision preference vectors of all parties in this game and obtain the decision preference matrix with the interference coefficient added.

[0078] In some embodiments of the present disclosure, the device 400 may input the feature vector and the decision preference vector into a large model based on the Transformer architecture, capture the relationship between the feature vector and the decision preference vector through the self-attention mechanism, and generate a preliminary strategic framework for the enterprise. The self-attention mechanism is Where Q, K, and V are query, key, and value vectors respectively, k The key vector dimension is used; the preliminary strategic framework is input into a small model based on a deep Q network, and the parameters of the deep Q network model are fine-tuned through the back-propagation algorithm in combination with the historical strategic data of the enterprise and the current market sample data, so as to refine the initial strategic framework; the external environment of the enterprise is continuously monitored, and the external environmental factors are input into the large model, so that the large model can predict market changes or policy trends based on the input external environmental factors, provide strategic adjustment suggestions, and the small model can update the Q value based on the strategic adjustment suggestions and the current market status and competitive environment; the large model based on the Transformer architecture and the small model based on the deep Q network are constructed to obtain an intelligent agent self-game model; and the intelligent agent self-game model is used to automatically conduct multiple rounds of game deductions to generate game scenarios and game paths covering a variety of strategy combinations.

[0079] In some embodiments of the present disclosure, the device 400 can calculate the immediate benefits of each game path, evaluate the execution difficulty, required resources, reputation risk, opponent reaction strategy and long-term effects of each strategy after execution, and screen out the strategy combination with the highest benefits under different decision-making styles; analyze the historical behavior and current game path of the game parties, predict the game parties' strategic direction, resource allocation preference and decision-making style evolution trend; establish a dynamic event model to simulate market changes or emergencies, and predict the behavior of the game parties in different scenarios based on the current game path; and automatically generate a standardized game deduction report according to a predetermined format, the game deduction report including: game path and decision analysis; benefit, risk and long-term impact assessment of each strategy; strategic tendency analysis of each game party; trend prediction of dynamic event model; response plan and risk mitigation measures.

[0080] In some embodiments of the present disclosure, the device 400 may be in the embodiments of the present disclosure, the processor 410 may be, for example, a central processing unit (CPU), a microprocessor, a digital signal processor (DSP), a processor based on a multi-core processor architecture, etc. The memory 420 may be any type of memory implemented using data storage technology, including but not limited to random access memory, read-only memory, semiconductor-based memory, flash memory, disk storage, etc.

[0081] In addition, in the embodiment of the present disclosure, the apparatus 400 may also include an input device 430, such as a keyboard, a mouse, etc., for inputting hot events, training data sets from multiple data sources, etc. In addition, the apparatus 400 may also include an output device 440, such as a display, etc., for outputting a simulation report, etc.

[0082] In other embodiments of the present disclosure, a computer-readable storage medium storing a computer program is further provided, wherein the computer program can achieve the following when executed by a processor: Figure 2 The steps of the intelligent strategy deduction method for machine self-game are shown.

[0083] In summary, according to the intelligent strategy deduction method and device for machine self-game according to the embodiments of the present disclosure, by constructing a game event agent, it is possible to conduct comprehensive simulation and deduction based on the deduction elements of actual events. The agent can extract multidimensional elements related to the game according to specific scenarios, form a variety of game paths, and simulate dynamic changes in the game to ensure accurate reproduction of complex game scenarios. The decision preference matrix with the addition of interference coefficients can dynamically adjust the decision preferences of the decision-makers to reflect the uncertainty factors or external interference factors in the game environment. This mechanism makes the deduction results more in line with the actual dynamic environment and improves the accuracy and practicality of decision deduction.

[0084] The large model is based on the Transformer architecture and can handle high-dimensional and long-time-step game paths, optimizing long-term dependencies in game deduction; the small model is based on the deep Q network and can efficiently perform reinforcement learning and quickly adapt to the real-time decisions of the game players. This collaborative optimization solution of the large model and the small model makes the deduction results more accurate and flexible.

[0085] By analyzing multiple groups of game paths, the strategic assessment model can identify the possible strategic tendencies of different game players, predict the next development of the game, and generate response plans based on the prediction results. This can help decision makers prepare in advance for emergencies or complex market changes and reduce the risks of strategic implementation.

[0086] Therefore, through machine game deduction and risk assessment, this solution can quantify decision-making risks and provide specific optimization suggestions to help decision makers make more informed decisions and reduce risks caused by incomplete or uncertain information. It can be applied to various complex game scenarios.

[0087] The flow chart and block diagram in the accompanying drawings show the possible architecture, function and operation of the device and method according to multiple embodiments of the present disclosure. In this regard, each square box in the flow chart or block diagram can represent a part of a module, a program segment or an instruction, and a part of the module, a program segment or an instruction includes one or more executable instructions for realizing the specified logical function. In some alternative implementations, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two continuous square boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or action, or can be implemented with a combination of special hardware and computer instructions.

[0088] Unless the context clearly indicates otherwise, the singular form of the words used herein and in the appended claims includes the plural and vice versa. Thus, when referring to the singular, the plural form of the corresponding term is generally included. Similarly, the words "comprise" and "include" are to be interpreted as inclusive rather than exclusive. Likewise, the terms "include" and "or" should be interpreted as inclusive unless such interpretation is expressly prohibited herein. Where the term "example" is used herein, particularly when it is located after a group of terms, the "example" is merely exemplary and illustrative and should not be considered exclusive or comprehensive.

[0089] Further aspects and scopes of adaptability become apparent from the description provided herein. It should be understood that various aspects of the present application can be implemented individually or in combination with one or more other aspects. It should also be understood that the description and specific embodiments herein are intended for purposes of illustration only and are not intended to limit the scope of the present application.

[0090] Several embodiments of the present disclosure are described in detail above, but it is obvious that those skilled in the art can make various modifications and variations to the embodiments of the present disclosure without departing from the spirit and scope of the present disclosure. The protection scope of the present disclosure is defined by the attached claims.

Claims

1. A method for intelligent strategy deduction of machine self-game, characterized in that: include: Constructing a game event agent, and extracting deduction elements of the event to be deduced based on the game event agent; Construct a decision preference matrix with interference coefficient added; Building an intelligent agent self-game model based on the deduction elements and the decision preference matrix, and generating multiple sets of game scenarios and game paths based on the intelligent agent self-game model, wherein the intelligent agent self-game model includes a large model based on a Transformer architecture and a small model based on a deep Q network; and Construct a strategic assessment model, and based on the strategic assessment model, perform risk assessment and trend prediction on the multiple groups of game scenarios and game paths, and generate a deduction report.

2. The intelligent strategy deduction method for machine self-game according to claim 1, characterized in that: The constructing of the game event agent comprises: Collecting raw text data from multiple data sources, and performing data cleaning and preprocessing on the collected raw text data; Labeling the key elements of the preprocessed data to obtain a training data set, wherein the key elements include event type, event background, participants, and decision-making style; Training a large language model based on the training data set and the strategy optimization algorithm, introducing new training data into the large language model during the training process, and adjusting the parameters of the model to obtain a trained large language model; and The trained large language model is encapsulated into an intelligent agent, and multiple algorithm processing logics are integrated into the intelligent agent.

3. The intelligent strategy deduction method for machine self-game according to claim 1 or 2, characterized in that: The data sources include news media, academic literature, policy documents, social media and historical cases, and the simulation elements include game name, theme, simulation background, simulation purpose, simulation party, initial value setting of simulation party indicators and initial situation events.

4. The intelligent strategy deduction method for machine self-game according to claim 1, characterized in that: The construction of the decision preference matrix with interference coefficient added includes: The characteristic vector and preference vector of the enterprise are encoded, and the kth element of the preference vector is expressed as: In the formula, w ij is the weight coefficient of the strategy, P ij It represents the firm’s preference for strategy j in market scenario i, m is the total number of strategies, and n is the number of market scenarios; The Monte Carlo method is used to construct the interference matrix Rij, Rjj = Random (0.5, 1); Before each game, according to P k ×R ij Determine the decision preference vectors of all parties in this game and obtain the decision preference matrix with the interference coefficient added.

5. The intelligent strategy deduction method for machine self-game according to claim 4, characterized in that: The characteristic vector of the enterprise includes enterprise scale, financial status, technological level and market share, and the preference vector is the degree of inclination of the enterprise towards different strategic directions, which include expansion, contraction, innovation and cost control.

6. The intelligent strategy deduction method for machine self-game according to claim 4, characterized in that: The intelligent agent self-game model is constructed based on the deduction elements and the decision preference matrix, and multiple sets of game scenarios and game paths are generated based on the intelligent agent self-game model. The intelligent agent self-game model includes a large model based on the Transformer architecture and a small model based on the deep Q network, including: The feature vector and decision preference vector are input into a large model based on the Transformer architecture, and the relationship between the feature vector and the decision preference vector is captured through the self-attention mechanism to generate a preliminary strategic framework for the enterprise. The self-attention mechanism is Where Q, K, and V are query, key, and value vectors respectively, k is the key vector dimension; Input the preliminary strategic framework into a small model based on a deep Q network, combine the historical strategic data of the enterprise and the current market sample data, and fine-tune the parameters of the deep Q network model through a back-propagation algorithm to refine the initial strategic framework; Continuously monitor the external environment of the enterprise and input external environmental factors into the large model so that the large model predicts market changes or policy trends based on the input external environmental factors and provides strategic adjustment suggestions, and the small model updates the Q value based on the strategic adjustment suggestions and in combination with the current market status and competitive environment; An intelligent agent self-game model is constructed based on the large model based on the Transformer architecture and the small model based on the deep Q network; and The intelligent agent self-game model automatically conducts multiple rounds of game deductions to generate game scenarios and game paths covering a variety of strategy combinations.

7. The intelligent strategy deduction method for machine self-game according to claim 6, characterized in that: The deep Q network includes multiple convolutional layers and fully connected layers. The convolutional layers are used to extract market competition situation features, and the fully connected layers are used to output strategic decisions. The loss function of the deep Q network is: where y i is the target Q value, s i , a j is the state-action pair, θ is the model parameter, and after each strategic decision is implemented, the Q value is updated according to the new market state and the reward obtained. The formula is: Q(s,a)=Q(s,a)+α[r+γmax a′ Q(s′,a′)-Q(s,a)], Among them, α is the learning rate, γ is the discount factor, r is the reward obtained, (s, a) is the current state-action pair, and (s′, a′) is the new state-action pair.

8. The intelligent strategy deduction method for machine self-game according to claim 1, characterized in that: The constructing of the strategic assessment model, and performing risk assessment and trend prediction on the multiple groups of game scenarios and game paths based on the strategic assessment model, and generating a deduction report include: Calculate the immediate benefits of each game path, evaluate the execution difficulty, required resources, reputation risk, opponent reaction strategy and long-term effect of each strategy after execution, and select the strategy combination with the highest benefits under different decision-making styles; Analyze the historical behavior and current game path of the game players, and predict the game players' strategic direction, resource allocation preferences, and decision-making style evolution trends; Establish dynamic event models to simulate market changes or emergencies, and predict the behavior of game players in different scenarios based on the current game path; and A standardized game simulation report is automatically generated according to a predetermined format, and the content of the game simulation report includes: game path and decision analysis; benefit, risk and long-term impact assessment of each strategy; strategic tendency analysis of each game player; trend prediction of dynamic event model; response plan and risk mitigation measures.

9. An intelligent strategy deduction device for machine self-game, characterized in that: The device comprises: at least one processor; and at least one memory storing a computer program; Wherein, when the computer program is executed by the at least one processor, the device executes the steps of the intelligent strategy deduction method for machine self-game according to any one of claims 1 to 8.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the computer program implements the steps of the intelligent strategy deduction method for machine self-game according to any one of claims 1 to 8.

Images