Blockchain and large model-based dynamic game decision method, device and medium

Abstract

The application discloses a dynamic game decision method and device based on a blockchain and a large model and a medium. The method comprises the following steps: receiving local data obtained through off-chain calculation submitted by a participating node, and writing the local data into a blockchain through a blockchain layer after verifying that the local data is passed; based on a game decision layer, using a global large model to generate corresponding game parameters according to the local data through an authorized node, obtaining an optimal equilibrium strategy, and writing the optimal equilibrium strategy into the blockchain; and based on the optimal equilibrium strategy, allocating resources and writing the allocation result into the blockchain. Through real-time situational awareness and strategy evolution deduction capability of the large model, a dynamic game decision is constructed, and dynamic modeling of multi-agent interaction strategies is realized.

Description

Technical Field

[0001] This application relates to the field of computer science, specifically to dynamic game decision-making methods, devices, and media based on blockchain and large-scale models. Background Technology

[0002] Currently, blockchain technology has been applied in data storage and decentralized applications, but it still has shortcomings in decision optimization and complex game theory problems, especially in fields such as finance, supply chain, and healthcare, where multi-party collaboration scenarios place higher demands on data privacy, dynamic game theory, and fairness. While blockchain possesses decentralized and trusted storage capabilities, it lacks dynamic decision optimization mechanisms. Traditional game theory models mainly rely on static assumptions, making it difficult to adapt to real-time data and dynamic adjustments in multi-party strategies.

[0003] Therefore, in traditional solutions, relying solely on blockchain or game theory models is insufficient to effectively address issues of data privacy protection, computational load balancing, and fairness in multi-party dynamic collaboration.

[0004] While large language models (LLMs) can provide efficient prediction and decision-making capabilities, their black-box nature makes it difficult to integrate them with distributed systems for transparent and reliable collaborative decision-making. Summary of the Invention

[0005] To address the aforementioned issues, this application proposes a dynamic game decision-making method based on blockchain and large-scale models, which is applied to a game decision-making system. The layered hybrid architecture of the game decision-making system includes an application layer, a game decision-making layer, and a blockchain layer. The method includes: Based on the application layer, participating nodes and authorized nodes are connected; The system receives local data submitted by the participating nodes, which is obtained through off-chain computation, and writes the local data into the blockchain through the blockchain layer after verifying the local data. Based on the game decision layer, through the authorized node, the global large model is used to generate corresponding game parameters based on the local data, and the optimal equilibrium strategy is obtained and written into the blockchain. Resource allocation is performed based on the optimal equilibrium strategy, and the allocation results are written into the blockchain.

[0006] In one example, the system receives local data submitted by the participating node, which is obtained through off-chain computation. After verifying the local data, the system writes the local data into the blockchain through the blockchain layer. Specifically, this includes: Receive local data submitted by the participating nodes, which is obtained through off-chain computation; wherein, the local data is obtained by the participating nodes, based on the raw data collected locally, using a local large model to extract features, and generating corresponding structured summaries and zero-knowledge proofs; The pre-compiled verification function is invoked to perform verification based on the structured digest and the zero-knowledge proof; If the verification is successful, the local data is written into the shared state database of the blockchain through the blockchain layer.

[0007] In one example, based on the game decision layer, and through the authorized node, a global large model is used to generate corresponding game parameters based on the local data, and the optimal equilibrium strategy is obtained, specifically including: Based on the game decision layer, the authorized node obtains a summary set consisting of structured summaries corresponding to the local data. Based on the summary set, the global large model generates the corresponding game parameters; The optimal equilibrium strategy is obtained based on the game parameters; wherein, the optimal equilibrium strategy is accompanied by a multi-party signature based on aggregate signature.

[0008] In one example, a global large model is used to generate corresponding game parameters based on the local data, and the optimal equilibrium strategy is obtained, specifically including: A game theory model is constructed, and a corresponding set of participating nodes and a corresponding strategy space are set; wherein, the strategy space corresponding to the participating node is the set of strategies selected by the participating node; The global large model is used to extract the corresponding environmental parameters from the local data, and a payoff matrix is ​​generated based on the environmental parameters as game parameters. Based on the game parameters, the optimal equilibrium strategy is obtained through Nash equilibrium.

[0009] In one example, a global large model is used to extract corresponding environmental parameters from the local data, and a payoff matrix is ​​generated based on the environmental parameters as game parameters, specifically including: The global model is used to extract corresponding environmental parameters from the local data, including the current market status and resource constraints. Based on the direct gains in the strategy space, the risks caused by the opponent's strategies, and the historical reputation score, a payoff matrix is ​​generated by weighted summation and used as game parameters.

[0010] In one example, the method further includes: In the process of calculating the Nash equilibrium, Monte Carlo tree search is used to search for potential policies, the policy space is reduced according to the potential policies, and the policy is updated using incremental gradient descent. If the calculation results show multiple equilibrium strategies, these strategies are written into the blockchain, and the optimal equilibrium strategy is selected through an on-chain voting mechanism based on the reputation and preference functions of the participating nodes.

[0011] In one example, the method further includes: Once the current round of the game has ended, the reputation score is updated based on the resource gains of each participating node: the reputation score is used to determine the gains in resource allocation; If a malicious policy is detected, a resource penalty mechanism will be triggered.

[0012] In one example, a payoff matrix is ​​generated based on the environmental parameters, serving as game parameters, specifically including: The participating nodes are identified as supplier nodes, logistics nodes, and bank nodes; A payoff matrix is ​​generated based on the environmental parameters and used as game parameters; wherein, for the supplier node, the payoff matrix is ​​used to represent optimizing goods quality; for the logistics node, the payoff matrix is ​​used to represent optimizing transportation speed; and for the bank node, the payoff matrix is ​​used to represent optimizing loan interest rate.

[0013] On the other hand, this application also proposes a dynamic game decision-making device based on blockchain and large models, including: At least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, which, when executed by the at least one processor, enables the at least one processor to: implement the dynamic game decision-making method based on blockchain and large models as described in any of the above examples.

[0014] On the other hand, this application also proposes a non-volatile computer storage medium storing computer-executable instructions, wherein the computer-executable instructions are configured as: the dynamic game decision-making method based on blockchain and large models as described in any of the above examples.

[0015] The dynamic game decision-making method based on blockchain and large models proposed in this application can bring the following benefits: 1. By leveraging the real-time context awareness and strategy evolution inference capabilities of large models, dynamic game decision-making is constructed to achieve dynamic modeling of multi-agent interaction strategies.

[0016] 2. Blockchain distributed ledgers enable the immutable storage of decision-making process data, and smart contracts are used to solidify game rules and constraints.

[0017] 3. Design a hybrid Nash equilibrium solver algorithm, which utilizes a large model-driven policy space compression and incremental gradient descent algorithm, as well as an on-chain voting mechanism to achieve fast convergence of Nash equilibrium solutions.

[0018] 4. Introduce a decentralized incentive-compatible system based on token economics, and ensure the fairness and sustainability of multi-party collaboration through an on-chain verifiable contribution evaluation function. Attached Figure Description

[0019] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a flowchart illustrating the dynamic game decision-making method based on blockchain and large models in the embodiments of this application; Figure 2 This is a schematic diagram of the architecture of a dynamic game decision-making system under one scenario in the embodiments of this application; Figure 3 This is a schematic diagram of the usage process of the dynamic game decision-making system under one scenario in the embodiments of this application; Figure 4 This is a schematic diagram of a dynamic game decision-making device based on blockchain and a large model, as described in this application embodiment. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0021] The technical solutions provided by the various embodiments of this application are described in detail below with reference to the accompanying drawings.

[0022] Traditional game theory models often suffer from the following shortcomings in decision-making: 1. Data silos and privacy issues: With multiple parties involved, data sharing carries the risk of privacy leaks and makes it difficult to build a trusted exchange mechanism.

[0023] 2. Centralization risk: Decision-making mechanisms based on a single central server are vulnerable to attacks or manipulation, and the decision-making process lacks fairness.

[0024] 3. Irrationality and dynamic adjustment in game theory: Traditional game theory models struggle to dynamically adjust strategies to achieve global optimum and cannot effectively balance situations where multiple parties have conflicting interests.

[0025] While blockchain technology provides a data storage mechanism, it lacks complex decision-making capabilities and struggles to handle multi-party game theory and decision optimization problems. Game theory models largely rely on static assumptions, making them ill-suited to real-time data and changes in multi-party strategies. The black-box nature of large models leads to an opaque decision-making process, making it difficult to guarantee fairness and verifiability.

[0026] With the development of blockchain, game theory, and large model technology, how to combine these three technologies to solve challenges such as data privacy, decentralization, and fairness in multi-party collaborative decision-making has become an urgent issue.

[0027] Based on this, such as Figure 1 As shown, this application provides a dynamic game decision-making method based on blockchain and large models, applied in a game decision-making system. The layered hybrid architecture of the game decision-making system includes an application layer, a game decision-making layer, and a blockchain layer.

[0028] like Figure 1 As shown, the method includes: S101: Based on the application layer, access the participating nodes and authorized nodes.

[0029] The application layer is mainly used for the access of participating nodes and authorized nodes. It can also be used for the access of related large models and the intelligent agents corresponding to the large models. It can also establish corresponding interactive interfaces for interaction with users.

[0030] S102: Receive the local data submitted by the participating node, which is obtained through off-chain computation, and after verifying the local data, write the local data into the blockchain through the blockchain layer.

[0031] Specifically, it receives local data submitted by participating nodes, which is obtained through off-chain computation. The local data is obtained by participating nodes using a local large model to extract features from the raw data collected locally, and then generating corresponding structured summaries and zero-knowledge proofs.

[0032] This process can also be called the off-chain preprocessing process of participating nodes. Participating nodes run lightweight large models (such as distilled LLMs) locally, and process the raw data through the large model. Feature extraction is performed, and a structured summary is generated. , This represents a local large model, in which zero-knowledge proofs (ZKP) are added. This is to prove the correctness of the calculation process and the validity of the data.

[0033] Its formula can be shown as in Formula 1: Formula 1; ZK-SNARK (Zero-Knowledge Succinct Non-Interactive Argument of Knowledge) is a zero-knowledge proof protocol. It also requires ensuring... and ValidDataset. ValidDataset is the validation set.

[0034] The pre-compiled verification function is invoked to perform verification based on the structured digest and zero-knowledge proof. If the verification passes, the local data is written to the blockchain's shared state database through the blockchain layer.

[0035] Smart contracts receive feature digests and zero-knowledge proof The data is used to call the pre-compiled ZKP verification function. If the verification passes, then... Write it into the shared state database of the blockchain.

[0036] The system includes key constraints that trigger the subsequent game decision-making process only when the conditions shown in Formula 2 are met: Formula 2; Among them, threshold The process is dynamically adjusted based on specific scenarios to ensure that a certain percentage of the data passes verification.

[0037] S103: Based on the game decision layer, through the authorized node, the global large model is used to generate corresponding game parameters according to the local data, and the optimal equilibrium strategy is obtained. The optimal equilibrium strategy is then written into the blockchain.

[0038] Specifically, based on the game-theoretic decision-making layer, authorized nodes obtain a summary set consisting of structured summaries corresponding to local data. Authorized nodes retrieve verified summary sets from the blockchain. .

[0039] The global large model generates corresponding game parameters based on the summary set. The summary set is input into the global large model to generate game parameters (e.g., payoff matrix). ).

[0040] The optimal equilibrium strategy is derived based on the game parameters; this optimal equilibrium strategy is accompanied by a multi-party signature based on aggregate signatures. Authorized nodes solve for the optimal equilibrium strategy based on these parameters. and the results After signing, it is submitted to the blockchain. A tamper-proof mechanism is also included, along with the optimal equilibrium strategy output by the large model. Includes multi-party signature based on BLS aggregate signature This is to ensure that the result cannot be forged or tampered with.

[0041] S104: Allocate resources based on the optimal equilibrium strategy and write the allocation results into the blockchain.

[0042] Smart contract verification After verifying the effectiveness, proceed according to the strategy. It automatically performs resource allocation and records the results on the blockchain, ensuring transparent and tamper-proof records.

[0043] 1. By leveraging the real-time context awareness and strategy evolution inference capabilities of large models, dynamic game decision-making is constructed to achieve dynamic modeling of multi-agent interaction strategies.

[0044] 2. Blockchain distributed ledgers enable the immutable storage of decision-making process data, and smart contracts are used to solidify game rules and constraints.

[0045] 3. Design a hybrid Nash equilibrium solver algorithm, which utilizes a large model-driven policy space compression and incremental gradient descent algorithm, as well as an on-chain voting mechanism to achieve fast convergence of Nash equilibrium solutions.

[0046] 4. Introduce a decentralized incentive-compatible system based on token economics, and ensure the fairness and sustainability of multi-party collaboration through an on-chain verifiable contribution evaluation function.

[0047] In one embodiment, during the dynamic game process, a game theory model can be first constructed, setting the corresponding set of participating nodes and the corresponding policy space; wherein, the policy space corresponding to the participating node is the set of policies chosen by the participating node. For example, let the set of participating nodes be... The strategy space is The policy space of each participating node It is the set of possible strategies it chooses.

[0048] A global model is used to extract corresponding environmental parameters from local data, and a payoff matrix is ​​generated based on these parameters as game parameters. These environmental parameters include the current market state and resource constraints.

[0049] At this point, a payoff matrix is ​​generated by weighted summation based on the direct payoffs in the strategy space, the risks caused by the opponent's strategy, and the historical reputation score, which serves as the game parameters.

[0050] The large model will analyze on-chain data in real time and extract the current market state. Resource constraints Given environmental factors, the returns can be quantified as shown in Formula 3: Formula 3; in, For the profit matrix, For strategy The direct benefits (such as trading profits) can be predicted by large models. Counter-strategy The resulting risk can be obtained by calculating the variance using Monte Carlo simulation. Historical reputation score, accumulated via the blockchain, reflects the credit and past performance of participating nodes. α, β, and γ are weighting coefficients, dynamically adjusted by the smart contract according to specific scenarios to control the relative importance of the three components.

[0051] Based on the game parameters, the optimal equilibrium strategy is obtained through Nash equilibrium. Solving for the Nash equilibrium. That is, a combination of strategies for all participating nodes i∈N such that no participating node can obtain a higher return by unilaterally changing its own strategy. Its formula can be shown as Equation 4: Formula 4.

[0052] In Phase 1, policy space reduction occurs. The large model generates an initial policy candidate set through reinforcement learning methods, reducing computational complexity. Monte Carlo tree search can be used to search for potential policies, and the policy space is reduced based on these potential policies.

[0053] In phase two, an equilibrium calculation is performed. At this stage, incremental gradient descent can be used to update the strategy, as shown in formula five: Formula 5; Among them, learning rate The algorithm adaptively adjusts based on historical data. Updates stop when the gradient satisfies the convergence condition ||∇U||<ϵ.

[0054] In Phase 3, an equilibrium selection is performed. If multiple equilibrium solutions exist, these strategies are written to the blockchain. An on-chain voting mechanism is then used to determine the equilibrium based on the reputation and preference functions of the participating nodes. Choose the final strategy As shown in Formula Six: Formula 6.

[0055] For the Nash equilibrium hybrid solver (referred to as the hybrid solver, used to obtain the optimal equilibrium policy based on the Nash equilibrium), its inputs include: the set of participating nodes players (including reputation and preference functions), the multidimensional payoff tensor payoffmatrix (the dimension is the Cartesian product of the policy space), the maximum number of iterations maxiter (default 1000), and the gradient convergence threshold ε (default 1e-4). The output is the Nash equilibrium (a dictionary of policy distributions for each participating node).

[0056] In Phase 1, the strategy space is reduced, the Monte Carlo Tree Search (MCTS) module is imported, the searcher is initialized, and the exploration parameter c=1.414 is set.

[0057] Parallelized policy pre-search is performed, with each participating node conducting 1000 Monte Carlo tree simulations based on its payoff tensor to explore potential policies. Policy pruning is then performed: the top 10 optimal policies for each participating node are extracted from the simulation results, forming a reduced policy space.

[0058] In Phase 2, incremental gradient descent initializes the hybrid policy, generating an initial hybrid policy probability distribution for each participating node in the reduced policy space.

[0059] Iterative optimization is performed (at most maxiter times), which involves calculating gradients, traversing all participating nodes, and calculating the expected reward gradients for each party based on the current policy distribution and reward tensor. It also involves dynamically adjusting the learning rate (AdaGrad mechanism), including calculating the global gradient norm, the square root of the sum of the squares of the L2 norms of the gradient vectors of all participating nodes; and the learning rate decaying with the number of iterations: lr = 0.1 / (1 + 0.01 * number of iterations); projected gradient descent update: for the policy of each participating node, it updates along the gradient direction according to the learning rate and projects it onto the simplex space using the project_simplex function to ensure the probability distribution is valid; and convergence checking: if the global gradient norm is less than the threshold ε, the iteration is terminated early.

[0060] In Phase 3, equilibrium selection involves screening candidate equilibria and finding all Nash equilibrium candidates that satisfy the approximate condition ε=1e-3 in the final policy distribution.

[0061] If there are multiple candidates, an on-chain voting mechanism is set up. Through reputation-weighted scoring, each participating node scores the candidate equilibrium based on its own reputation and preference function; and aggregated scoring, the scores of all participating nodes for each candidate are accumulated, and the candidate with the highest comprehensive score is selected as the final Nash equilibrium.

[0062] If only one candidate exists, it is directly used as the final result.

[0063] Finally, the selected Nash equilibrium policy distribution is returned.

[0064] In addition, a dynamic adjustment mechanism can be set up.

[0065] Specifically, once the current round of the game has ended, the reputation score is updated based on the resource gains of each participating node: the reputation score is used to determine the gains in resource allocation.

[0066] An evolutionary game feedback mechanism is set up. After each round of the game, the reputation score of each participating node is updated based on its gains, as shown in Formula 7: Formula 7; in, The average revenue for all participating nodes. The learning rate is used to incentivize participating nodes to cooperate and avoid extreme non-cooperative behavior.

[0067] If a malicious policy is detected, a resource penalty mechanism will be triggered.

[0068] Set up a penalty mechanism to detect malicious policies. This will trigger the token confiscation mechanism, as shown in Formula 8: Formula 8; in, This is a penalty coefficient, which is positively correlated with the severity of the malicious behavior.

[0069] In one embodiment, such as Figure 2 As shown, the game decision-making system adopts a layered hybrid architecture, including an application layer, a game decision-making layer, and a blockchain layer, with each layer having clearly defined responsibilities and coordinating with each other.

[0070] The application layer primarily enables users to submit requests or obtain decision results through DApp interfaces. Users access the system via APIs to submit data, request decisions, or receive feedback. The application layer processes user requests, calls the game decision layer's interfaces, and transmits the corresponding data or request information.

[0071] The game theory decision layer is primarily used to acquire data from off-chain computation and on-chain verification to generate strategies, solve multi-party game problems, and calculate the optimal equilibrium strategy. The game theory decision layer includes: a large-model strategy generator, a Nash equilibrium hybrid solver, and a smart contract execution engine.

[0072] The large-model strategy generator primarily analyzes on-chain data and real-time environmental parameters based on a large language model (hereinafter referred to as the large model) to generate the strategy space and payoff matrix required for game decision-making. Upon receiving data from the application layer, the large model analyzes the current environmental data (e.g., market demand, resource constraints) and generates the strategy space required for the game based on this data. The large model generates the strategy space, providing possible actions and corresponding payoff calculations.

[0073] The Nash equilibrium hybrid solver is primarily used to compute Nash equilibria or other types of optimal solutions in games, ensuring that parties can reach the optimal cooperation plan in situations of conflicting interests. It uses game theory algorithms (e.g., a combination of Monte Carlo tree search and gradient descent) to solve for equilibrium solutions in game models. If multiple equilibria exist in the game, the system can select the optimal solution based on specific rules (e.g., reputation score).

[0074] The smart contract execution engine is primarily used to execute decision schemes based on the output of the game solver and trigger subsequent blockchain operations, such as resource allocation or token settlement. Tokens can also be considered a type of resource, thus token settlement can also be seen as a form of resource allocation. Based on the solved strategy, the smart contract engine calculates specific decisions such as resource allocation and transaction execution. These decisions are packaged and submitted to the blockchain for further verification and execution.

[0075] The blockchain layer is primarily used for storing and verifying data, executing decentralized smart contracts, and ensuring the trustworthiness and immutability of data. This layer mainly includes: data storage functions, ZKP verification, and incentive mechanisms.

[0076] For data storage, blockchain provides sharded storage and management of data through private or consortium blockchains, ensuring transparency and data sharing among participating nodes. The blockchain layer receives data (such as data summaries and computation results) provided by the off-chain computing layer. Data is sharded and stored in the blockchain, ensuring decentralization and preventing single points of failure.

[0077] ZKP validator nodes are primarily used by Zero-Knowledge Proof (ZKP) validator nodes to verify the validity of off-chain submitted data, ensuring data privacy and authenticity. Upon receiving a request containing a data digest and a zero-knowledge proof, the ZKP validator node executes a verification process to ensure the submitted data has undergone legitimate computation and conforms to predefined computation rules. If the verification passes, the data is recorded in the blockchain; if the verification fails, the data submission is rejected.

[0078] A resource (token) management module can also be set up based on needs to manage the on-chain incentive mechanism and control the allocation, reward, and penalty mechanisms of resources. Based on the results of game theory decisions, the resource management module is responsible for automatically allocating resource rewards to compliant participating nodes and penalizing nodes engaging in malicious behavior. The distribution of resources is controlled by smart contracts to ensure the fairness and transparency of the incentive mechanism.

[0079] Based on this, the entire system's workflow integrates blockchain, game theory, and large-scale models through multiple layers: the application layer provides an interface for interaction with users, requesting decisions through API calls; the game decision layer is responsible for calculating the optimal decision based on the analysis of the large-scale model and game theory, and generating corresponding strategies; the blockchain layer ensures the credibility and privacy of data, and provides a decentralized incentive mechanism to verify the rationality of game decisions.

[0080] Through these levels of collaborative work, it is ensured that the collaborative decision-making process involving multiple parties is open, transparent, credible, and conforms to the optimal strategy in game theory, while also ensuring data privacy and fair incentives for participating nodes.

[0081] Based on this, such as Figure 3 As shown, the system's actual workflow includes off-chain computation and on-chain verification steps.

[0082] The application layer is used by users or intelligent agents to process data submitted through interfaces. The off-chain computation layer processes the data using a local large model, generates feature summaries, and attaches zero-knowledge proofs (ZKPs) to verify the correctness of the computations. The on-chain blockchain layer uses smart contracts to verify the ZKPs and stores the feature summaries on the blockchain, ensuring data trustworthiness. The off-chain game decision-making layer generates strategies based on the data on the blockchain and submits them to the blockchain. The blockchain layer (strategy verification and execution): smart contracts verify strategy signatures, execute resource allocation or token settlement, ensuring all operations are transparent and tamper-proof.

[0083] In summary, blockchain enables trusted data storage and rule solidification, combined with game theory models to dynamically adjust the strategies of participating nodes, and utilizes Large Language Models (LLM) for real-time data analysis and decision prediction, forming a complete "data-computation-incentive" closed loop.

[0084] Dynamic Game Chain (DGC) utilizes blockchain to achieve trusted storage of multi-party data. Its smart contracts incorporate various game models (such as Bayesian games and Stackelberg games) to adapt to different decision-making needs. Zero-knowledge proofs (ZKP) are used to verify the validity of data inputs, ensuring the trustworthiness of data and computation.

[0085] The large model-driven Nash equilibrium solver (LLM-based Equilibrium Solver) uses large language models (LLM) to analyze historical and real-time environmental data, generates a game strategy space and payoff matrix, and employs Monte Carlo Tree Search (MCTS) and gradient descent to quickly solve for the Nash equilibrium point, and adjusts the decisions of participating nodes according to the strategy space.

[0086] The decentralized incentive mechanism (Tokenized Incentive Mechanism) uses blockchain tokenization incentives, combined with game equilibrium theory, to ensure cooperation and fairness among participating nodes. Malicious nodes are penalized through a token staking penalty mechanism to prevent collusion and manipulation.

[0087] In the system architecture, data submission and verification are handled by multiple parties: participating nodes preprocess the data by running a large model locally, generating feature summaries, and then verifying them with zero-knowledge proofs (ZKP) to ensure data credibility. Smart contracts verify the validity of the data and reject invalid data.

[0088] Dynamic game model construction: Generate game models based on current environmental data (such as market prices, resource inventory, etc.), and generate payoff functions and policy constraints by dynamically selecting game models suitable for the scenario (such as Stackelberg game).

[0089] Equilibrium Solution and Policy Output: The large model combines historical data to predict the strategies of other participating nodes, and outputs the optimal strategy by solving the game equilibrium problem. If multiple equilibrium solutions exist, the smart contract determines the final strategy through on-chain voting.

[0090] On-chain execution and incentive feedback: Smart contracts automatically allocate resources or settle tokens based on the game's outcome, recording the results on the blockchain to ensure transparency and immutability. In long-term games, node reputation scores will influence future revenue weighting, promoting cooperation.

[0091] In one embodiment, the convergence and existence proof of Nash equilibrium are explained in detail to illustrate the effects and implementation process of the above embodiments.

[0092] The theorem for proving the existence of Nash equilibrium states that in a finite game, if the policy space of each participating node is non-empty and compactly convex, and the payoff function of each participating node is continuous and quasi-concave, then there exists at least one pure policy Nash equilibrium.

[0093] Based on this, the proof process includes: To prove the existence of Nash equilibrium, the Kakutani fixed-point theorem is used. This theorem provides the existence of game equilibrium under certain conditions. The proof process is as follows: 1. The strategy space is a compact convex set: Assume that the strategy space of each participating node in the game is a compact convex set. It is non-empty and compactly convex, which means that there is a connection path between any two policies in each policy space, and these policy spaces are bounded and continuous.

[0094] 2. The payoff function is continuous and quasi-concave: Assume that the payoff function for each participating node is... It is continuous and quasi-concave; quasi-concaveness means that for any two strategies... Payoff function The linear combination of strategies exhibits a monotonicity as shown in Equation 9: Formula Nine; in, Except The strategy of other participating nodes.

[0095] 3. Construct the Best Response Function (BR): For each participating node i, define its best response function. In order to make The most effective strategy is shown in Formula 10: Formula 10; Since the payoff function is quasi-concave, It is a monotonic mapping, and it is a mapping on a convex set.

[0096] 4. Application of Kakutani's Fixed-Point Theorem: Due to the policy space It is tightly convex, and each Since it is a convex mapping, according to the Kakutani fixed-point theorem, there exists a fixed point, i.e., there exists a policy combination. This enables the strategy of each participating node. It is its corresponding optimal response function Element.

[0097] Based on this, it is concluded that, according to the Kakutani fixed-point theorem, there exists at least one Nash equilibrium under the above conditions. That is, each participating node's strategy corresponds to its best response to other strategies.

[0098] The convergence proof for Nash equilibrium is as follows: In a multi-round game, if participating nodes continuously adjust their strategies to maximize their own gains, then under certain conditions, the game will converge to a Nash equilibrium.

[0099] Based on this, the proof process includes: 1. Strategy update rules: Assuming participating nodes The strategy is continuously adjusted by using a certain update rule, such as gradient descent or Monte Carlo tree search (MCTS). To maximize its returns, the rule for policy updates can be expressed as shown in Equation 11: Formula 11; in, It is the learning rate, which represents the step size for policy updates. It is about The gradient.

[0100] 2. Convergence Criteria: To ensure convergence, we need to meet the following conditions: Payoff function It is convex, which means that as the strategy is updated, the payoff function either increases or remains constant. Learning rate It is small enough to ensure that each update does not exceed the range of the optimal solution.

[0101] 3. Convergence of gradient descent: If gradient descent is used for policy updates and the payoff functions of all participating nodes are convex and smooth, then according to the convergence theorem of gradient descent, when the number of updates approaches infinity, the system will converge to a local optimum. In many cases, this local optimum is the global optimum, i.e., Nash equilibrium.

[0102] 4. Dynamically adjusted feedback mechanism: In multi-agent games, the policy adjustments of participating nodes depend not only on their own decisions but also on the policies of other participating nodes. Therefore, during the policy update process, each participating node's policy is dynamically adjusted, gradually converging towards a Nash equilibrium as the game progresses. Through learning from and receiving feedback from historical data, the system can find the optimal solution after multiple rounds of gameplay.

[0103] Based on this, we conclude that under specific conditions (e.g., the policy update rule is gradient descent, and the payoff function is convex), the policy updates of the participating nodes in the game will converge to a Nash equilibrium. Through the above proof, we can draw the following conclusion: Existence of Nash Equilibrium: In a finite game, if the strategy space is a non-empty compact convex set and the payoff function is continuous and quasi-concave, then there exists at least one pure strategy Nash equilibrium.

[0104] Convergence of Nash Equilibrium: When participating nodes in a game adjust their strategies through gradient descent or other optimization algorithms, the game will converge to Nash equilibrium if certain conditions are met (e.g., the payoff function is convex and the update rule is reasonable).

[0105] These two theories guarantee the effectiveness and rationality of the game decision-making layer in this invention, and ensure the stability and optimality of decision-making in the multi-party collaborative game process.

[0106] In one embodiment, based on the method mentioned in the above embodiments, corresponding experiments were conducted, and experimental data and performance verification were obtained.

[0107] Convergence speed comparison experiment: Compared with traditional algorithms (gradient descent and linear programming) and the hybrid solver proposed in the above embodiments, the results show that the algorithm in the embodiments of this application converges faster when the game scale is large.

[0108] Objective: To verify the convergence speed of hybrid solvers (including Gradient Descent (GD) and Monte Carlo Tree Search (MCTS)) in large-scale games.

[0109] Experimental design: Comparison of traditional pure gradient descent (GD) with linear programming (LP) methods.

[0110] The experimental results are shown in Table 1 below: Table 1 Experimental Results

[0111] Based on this, we conclude that the hybrid algorithm reduces the number of iterations by 57% and the total time by 38.7% by preprocessing the policy space using MCTS. The reduction in the number of iterations and the total time demonstrates the efficiency of this method in complex games.

[0112] In one embodiment, taking a supply chain finance scenario as an example, the corresponding decision-making process is illustrated.

[0113] In supply chain finance applications, the participating nodes include the supplier, the logistics provider, and the bank. In this case, the participating nodes are determined to be the supplier, the logistics provider, and the bank. The decision is optimized through the following steps: Scenario description: Suppliers, logistics providers, and banks jointly decide on the amount of goods-backed loans. Participating nodes upload historical transaction data (encrypted and stored on the blockchain). A large-scale model analyzes market risk and generates default probability curves for different loan amounts.

[0114] For game modeling and decision-making processes, a payoff matrix is ​​generated based on environmental parameters, which serves as game parameters. Specifically, for supplier nodes, the payoff matrix represents optimizing goods quality; for logistics nodes, the payoff matrix represents optimizing transportation speed; and for bank nodes, the payoff matrix represents optimizing loan interest rates.

[0115] In other words, suppliers maximize their profits by optimizing the quality of goods. Banks optimize loan interest rates. To maximize their own profits, logistics providers optimize transportation speed. To reduce costs.

[0116] Equilibrium strategy obtained through game theory The loan disbursement and transportation route allocation will be automatically executed by the smart contract.

[0117] Regarding the incentive mechanism, if the actual default rate deviates from the prediction, a token staking penalty mechanism will be triggered to ensure the fairness of the decision-making.

[0118] For parameter definition, participating nodes include: suppliers Logistics ,bank The policy space includes: Select cargo quality . Select transportation speed (corresponding to different transportation costs) ). Setting loan interest rates .

[0119] For the instantiation of the revenue function, the supplier's revenue is shown in Formula Twelve: Formula 12; in, Price per unit. For the loan amount, .

[0120] Bank returns are shown in Formula Thirteen: Formula Thirteen; Among them, default risk MarketData Predicted by a large model.

[0121] At this point, during the Nash equilibrium solution process, the large model analyzes historical on-chain data to generate default risk. The predicted surface. Construct a three-party non-cooperative game model and solve for the Nash equilibrium, including: supplier optimization. To maximize Bank optimization To maximize Logistics optimization To minimize transportation costs and ensure .

[0122] Smart contracts are based on the game equilibrium outcome Automatic loan disbursement and transportation assignment.

[0123] like Figure 4 As shown in the embodiments of this application, a dynamic game decision-making device based on blockchain and large models is also proposed, including: At least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, which, when executed by the at least one processor, enables the at least one processor to: perform the dynamic game decision-making method based on blockchain and large models as described in any of the above embodiments.

[0124] This application also proposes a non-volatile computer storage medium storing computer-executable instructions, wherein the computer-executable instructions are configured as: the dynamic game decision-making method based on blockchain and large models described in any of the above embodiments.

[0125] The various embodiments in this application are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the device and medium embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the description of the method embodiments.

[0126] The devices and media provided in this application are one-to-one with the methods. Therefore, the devices and media also have similar beneficial technical effects as their corresponding methods. Since the beneficial technical effects of the methods have been described in detail above, the beneficial technical effects of the devices and media will not be repeated here.

[0127] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0128] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0129] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0130] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0131] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0132] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0133] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0134] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0135] The above description is merely an embodiment of this application and is not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A dynamic game decision-making method based on blockchain and large-scale models, characterized in that, In a game-theoretic decision-making system, the layered hybrid architecture of the game-theoretic decision-making system includes an application layer, a game-theoretic decision-making layer, and a blockchain layer. The method includes: Based on the application layer, participating nodes and authorized nodes are connected; The system receives local data submitted by the participating nodes, which is obtained through off-chain computation, and writes the local data into the blockchain through the blockchain layer after verifying the local data. Based on the game decision-making layer, through the authorized node, the global large model is used to generate corresponding game parameters based on the local data, and the optimal equilibrium strategy is obtained and written into the blockchain; the game parameters include a payoff matrix. Resource allocation is performed based on the optimal equilibrium strategy, and the allocation results are written into the blockchain; The process of receiving local data submitted by participating nodes through off-chain computation, and writing the local data into the blockchain through the blockchain layer after verifying the local data, specifically includes: receiving local data submitted by participating nodes through off-chain computation; wherein, the local data is obtained by the participating nodes through feature extraction using a local large model based on locally collected raw data, and generating corresponding structured summaries and zero-knowledge proofs; calling a pre-compiled verification function to verify the data based on the structured summaries and zero-knowledge proofs; if the verification passes, writing the local data into the shared state database of the blockchain through the blockchain layer; The process, based on the game decision-making layer and through the authorized nodes, uses a global large model to generate corresponding game parameters based on the local data and obtain the optimal equilibrium strategy. Specifically, this includes: obtaining a summary set composed of structured summaries corresponding to the local data through the authorized nodes; generating corresponding game parameters based on the summary set using a global large model; wherein the optimal equilibrium strategy is accompanied by a multi-party signature based on aggregate signatures; the participating nodes include supplier nodes, logistics nodes, and bank nodes; wherein, for the supplier node, the payoff matrix represents optimizing goods quality; for the logistics node, the payoff matrix represents optimizing transportation speed; for the bank node, the payoff matrix represents optimizing loan interest rates; and obtaining the optimal equilibrium strategy through Nash equilibrium based on the game parameters. Alternatively, based on the game decision-making layer, through the authorized nodes, a global large model is used to generate corresponding game parameters based on the local data, and the optimal equilibrium strategy is obtained. Specifically, this includes: modeling a game theory model based on the game decision-making layer and through the authorized nodes, setting a corresponding set of participating nodes and a corresponding strategy space; wherein the strategy space corresponding to the participating nodes is the set of strategies selected by the participating nodes; extracting corresponding environmental parameters using the global large model based on the local data; determining that the participating nodes include supplier nodes, logistics nodes, and bank nodes; generating a payoff matrix based on the environmental parameters as game parameters; wherein, for the supplier nodes, the payoff matrix represents optimizing goods quality; for the logistics nodes, the payoff matrix represents optimizing transportation speed; for the bank nodes, the payoff matrix represents optimizing loan interest rates; and obtaining the optimal equilibrium strategy through Nash equilibrium based on the game parameters.

2. The method according to claim 1, characterized in that, The method further includes: In the process of calculating the Nash equilibrium, a Monte Carlo tree search is used to search for potential policies, the policy space is reduced according to the potential policies, and the policy is updated using incremental gradient descent. If the calculation results show multiple equilibrium strategies, these strategies are written into the blockchain, and the optimal equilibrium strategy is selected through an on-chain voting mechanism based on the reputation and preference functions of the participating nodes.

3. The method according to claim 1, characterized in that, The method further includes: Once the current round of the game has ended, the reputation score is updated based on the resource gains of each participating node: the reputation score is used to determine the gains in resource allocation; If a malicious policy is detected, a resource penalty mechanism will be triggered.

4. A dynamic game decision-making device based on blockchain and large-scale models, characterized in that, include: At least one processor; as well as, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, which, when executed by the at least one processor, causes the at least one processor to perform the dynamic game decision-making method based on blockchain and large models as described in any one of claims 1 to 3.

5. A non-volatile computer storage medium storing computer-executable instructions, characterized in that, The computer-executable instructions are configured to execute the dynamic game decision-making method based on blockchain and large models as described in any one of claims 1 to 3.

Images