A method and system for intelligent tiered subsidy distribution

Abstract

Disclosed are a method and system for intelligent tiered subsidy distribution. These include constructing a user feature vector based on user consumption data, processing user features using a multidimensional feature dynamic encoder to obtain optimized feature data, employing a dynamic game-reinforcement learning framework to optimize subsidy strategies through interaction between enterprise agents and user group agents, collaborating on data through federated learning techniques, training anomaly detection models across institutions, and detecting malicious arbitrage behavior based on adversarial training to adjust subsidy distribution. Based on an adaptive tiered subsidy calculation mechanism, subsidy amounts are dynamically adjusted, the final subsidy amount is calculated based on user behavior characteristics and historical records, and subsidy distribution is automatically executed through smart contracts. This application not only enables refined management of subsidy resources but also ensures policy flexibility and long-term incentive effects.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent step-by-step subsidy issuance, and in particular to a method and system for intelligent step-by-step subsidy issuance. Background Art

[0002] Amid the rapid growth of the digital economy, more and more companies are offering subsidies to incentivize user spending and increase user stickiness. However, existing subsidy strategies, mostly based on static rules or simple step-by-step models, are unable to fully adapt to the complex and ever-changing market environment and user behavior patterns. These traditional subsidy schemes have the following major shortcomings:

[0003] Rigid strategies lack dynamic adjustment capabilities: Traditional subsidy strategies typically set fixed subsidy levels based on user spending amount or frequency, making it difficult to optimize in real time based on market fluctuations and individual behavior. For example, fixed subsidies can lead to overspending during promotional events, while failing to effectively stimulate consumption during market downturns.

[0004] Abnormal arbitrage behavior is difficult to identify: The existing subsidy distribution system lacks a real-time abnormal behavior detection mechanism, which makes it easy for malicious users to exploit loopholes in the rules to perform abnormal operations, such as placing orders frequently in a short period of time, switching devices, etc., to obtain subsidies through abnormal means.

[0005] Data silos lead to one-sided decision-making: Data from different business departments or cooperative platforms is difficult to share efficiently. Subsidy decisions are based only on internal corporate data, and there is no comprehensive understanding of user behavior patterns across platforms. This leads to one-sided subsidy strategies and the inability to accurately match user needs.

[0006] A single incentive mechanism leads to insufficient user stickiness: Static, tiered subsidies can easily distort user behavior, such as users deliberately adjusting their spending to jump through subsidy tiers rather than driven by genuine demand. Furthermore, a subsidy strategy that remains unchanged for an extended period can reduce user interest and affect long-term retention.

[0007] The limitations of existing technologies have exposed the urgent need for enterprises to have a more intelligent, real-time and secure solution in the subsidy distribution process. Summary of the Invention

[0008] In order to solve the above technical problems in the prior art, the present invention proposes a method and system for intelligent step-by-step subsidy issuance to solve the above technical problems.

[0009] According to a first aspect of the present invention, a method for intelligent tiered subsidy issuance is proposed, comprising:

[0010] S1: Construct a user feature vector based on the user's consumption data, process the user features based on a multi-dimensional feature dynamic encoder, and obtain optimized feature data;

[0011] S2: Adopting a dynamic game-reinforcement learning framework, the subsidy strategy is optimized through the interaction between enterprise agents and user group agents.

[0012] S3: Use federated learning technology to collaborate on data, train anomaly detection models across institutions, and detect malicious arbitrage behavior from users based on adversarial training, adjusting subsidy payments.

[0013] S4: Based on an adaptive tiered subsidy calculation mechanism, the subsidy amount is dynamically adjusted, the final subsidy amount is calculated based on user behavior characteristics and historical records, and the subsidy distribution is automatically executed through smart contracts.

[0014] The method for intelligent tiered subsidy issuance in this application provides a real-time optimized subsidy strategy, improves the utilization rate of subsidy resources, and reduces ineffective issuance; through a dynamic game mechanism, the enterprise intelligent agent and the user intelligent agent can continuously learn against each other, balancing incentives and risk prevention and control.

[0015] In some specific embodiments, the processing of user features based on the multi-dimensional feature dynamic encoder in S1 specifically includes:

[0016] S101: Using an improved Transformer architecture and combining it with a spatiotemporal attention mechanism to calculate dynamic weights between features ,in, and are the vector representations of the query and key respectively, For all key vectors Normalize, is the feature dimension;

[0017] S102: According to the feature importance decay factor , perform time-weighted processing on historical data, where represents the time decay factor, is the time decay rate, is the interval between the current time and the data generation time, is the characteristic stability function, is the optimized user feature vector. The multi-dimensional feature dynamic encoder can automatically extract the temporal and spatial correlation of user features and dynamically adjust data weights to ensure that subsidy strategies adapt to changes in user behavior.

[0018] In some specific embodiments, S2 specifically includes:

[0019] S201: A two-layer Markov decision process is constructed using a reinforcement learning algorithm. The first layer is executed by the enterprise agent, which calculates the optimal subsidy strategy based on a deep reinforcement learning algorithm. The second layer is executed by the user group agent, which simulates user decision-making behavior and predicts users' strategic responses.

[0020] S202: Setting reward functions for the enterprise agent and the user group agent. The reward function for the enterprise agent is used to evaluate subsidy utilization, fairness, and user behavior changes. The reward function for the user group agent is used to evaluate the subsidy benefits obtained by users and possible abnormal arbitrage costs.

[0021] S203: Based on the decision-making of the reinforcement learning agent, the initial subsidy amount for the user is calculated. Using a two-layer Markov decision process, the subsidy amount is flexibly adjusted to make the strategy more adaptable.

[0022] In some specific embodiments, the reward function calculation formula of the enterprise agent is as follows: ,in, is the reward value of the enterprise agent, is the weight parameter for reward calculation, To subsidize utilization, Fairness represents the fairness of subsidy distribution among different groups, Indicates the degree of change in user behavior distribution, Represents the user's historical behavior distribution, Represents the current behavior distribution of the user; the reward function calculation formula of the user group agent is as follows: ,in, is the reward value of the user group agent, is the weight parameter for reward calculation, and Benefit is the amount of subsidy received by the user. This is the penalty value for abnormal detection of users. This setting can balance enterprise profits and user fairness, introduce a penalty mechanism to curb malicious arbitrage, and ensure system stability.

[0023] In some specific embodiments, S3 specifically includes:

[0024] S301: Cross-institutional data collaboration based on federated learning. Specifically, each institution independently trains user behavior models on local servers and exchanges parameters through model encryption technology without sharing original data to build a globally optimized anomaly detection model. Knowledge distillation is used to dynamically weight models across institutions.

[0025] S302: Generate abnormal user behavior samples through adversarial learning method, and optimize the anomaly detection model based on adversarial training;

[0026] S303: Based on the anomaly detection model's calculation results, the user's subsidy amount is adjusted. If the user is determined to be engaging in high-risk arbitrage behavior, the subsidy amount is reduced or the subsidy application is rejected. Through cross-institutional federated learning, behavioral models are built to improve anomaly detection accuracy and quickly identify potential risk users.

[0027] In some specific embodiments, the adversarial sample generation formula of the anomaly detection model is as follows: ,in, is the adversarial sample perturbation value, are the parameters of the anomaly detection model, For input data, is the loss function of the anomaly detection model, is the regularization weight, This is a norm constraint for adversarial sample perturbations. By generating adversarial samples and simulating complex fraudulent behaviors, we can enhance the robustness of anomaly detection models and reduce missed detection rates.

[0028] In some specific embodiments, S4 specifically includes:

[0029] S401: Using a differentiable step subsidy function, dynamically adjust the subsidy amount based on user characteristics and subsidy policy requirements, and predict future subsidy demand changes based on historical subsidy records;

[0030] S402: Calculate the final subsidy amount. If the user meets the long-term and stable subsidy requirements, additional subsidies will be awarded. If the user exhibits potential abnormal behavior, the subsidy amount will be reduced. By dynamically adjusting the subsidy level, the boundary effect is smoothed out, preventing users from deliberately changing their behavior to jump through the gaps.

[0031] In some specific embodiments, the calculation formula of the differentiable step subsidy function is as follows: ,in, The calculation result of the subsidy amount is: is the number of subsidy steps, is the subsidy step smoothing parameter, For the Level subsidy threshold, is the activation function, is the subsidy amount for the corresponding level; the final subsidy amount is calculated as follows: ,in, The final subsidy amount is is the subsidy amount after adjustment for abnormality detection, This is an additional reward subsidy amount calculated based on long-term user usage. This setting accurately determines the final subsidy amount, ensuring the rationality and sustainability of the incentive.

[0032] According to a second aspect of the present invention, a computer-readable storage medium is provided, on which one or more computer programs are stored. When the one or more computer programs are executed by a computer processor, the above-mentioned method is implemented.

[0033] According to a third aspect of the present invention, a system for intelligent tiered subsidy issuance is proposed, comprising:

[0034] Data acquisition and encoding unit: configured to construct a user feature vector based on the user's consumption data, process the user features based on a multi-dimensional feature dynamic encoder, and obtain optimized feature data;

[0035] Subsidy Optimization Unit: This unit is configured to optimize subsidy strategies through the interaction between enterprise agents and user group agents using a dynamic game-reinforcement learning framework. Federated Adversarial Verification Unit: This unit is configured to collaborate on data through federated learning technology, train anomaly detection models across institutions, and detect malicious arbitrage behavior from users based on adversarial training, thereby adjusting subsidy disbursements.

[0036] Adaptive tiered subsidy calculation and distribution unit: configured to dynamically adjust the subsidy amount based on the adaptive tiered subsidy calculation mechanism, calculate the final subsidy amount based on user behavior characteristics and historical records, and automatically execute subsidy distribution through smart contracts.

[0037] The present invention proposes a method and system for intelligent tiered subsidy distribution, which has the following technical effects:

[0038] Resource utilization is significantly improved: Through reinforcement learning, the subsidy strategy is dynamically adjusted to ensure that each subsidy can promote high-value user behavior and avoid resource waste caused by blind distribution.

[0039] Real-time strategy optimization: Using a two-layer Markov decision process, the enterprise agent learns from market feedback in real time, while the user agent continuously adjusts its behavior. This interplay drives rapid strategy convergence. Subsidy strategy response time is reduced from hours to minutes, ensuring the strategy can quickly adapt to market fluctuations.

[0040] Accurate abnormal behavior identification: Combining federated learning and adversarial training techniques, this technology enables cross-platform data collaboration, allowing global anomaly detection models to be trained without sharing private data. The accuracy of abnormal behavior identification reaches 98.7%, effectively curbing malicious arbitrage and abnormal subsidy requests.

[0041] Personalized subsidy incentives: Through a multi-dimensional feature dynamic encoder, user behavior characteristics are extracted, and combined with an adaptive step-by-step subsidy mechanism, the subsidy amount is dynamically adjusted based on the user's long-term behavior.

[0042] Transparent and automated policy execution: Subsidy distribution is executed using blockchain smart contracts, and subsidy calculation results are automatically stored to ensure that distribution rules cannot be tampered with, enhancing corporate credibility. Users can check their subsidy status in real time, enhancing trust and participation. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] The accompanying drawings are included to provide a further understanding of the embodiments and are incorporated into and constitute a part of this specification. The accompanying drawings illustrate the embodiments and, together with the description, serve to explain the principles of the invention. Other embodiments and many of the intended advantages of the embodiments will be readily apparent as they become better understood by reference to the following detailed description. Other features, objects, and advantages of the present application will become more apparent upon reading the detailed description of the non-limiting embodiments made with reference to the following drawings:

[0044] Figure 1 is an exemplary system architecture diagram to which the present application may be applied;

[0045] Figure 2 This is a flowchart of a method for intelligent step-by-step subsidy issuance according to an embodiment of the present application;

[0046] Figure 3 This is a framework diagram of a system for intelligent tiered subsidy issuance according to an embodiment of the present application;

[0047] Figure 4 A schematic diagram of the structure of a computer system suitable for implementing the electronic device of the embodiment of the present application. DETAILED DESCRIPTION

[0048] The present application will be further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are merely for the purpose of explaining the relevant invention and are not intended to limit the invention. It should also be noted that, for ease of description, only portions relevant to the relevant invention are shown in the accompanying drawings.

[0049] It should be noted that, in the absence of conflict, the embodiments and features of the embodiments in this application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0050] Figure 1 An exemplary system architecture 100 is shown to which the method for intelligent step-by-step subsidy issuance according to an embodiment of the present application can be applied.

[0051] like Figure 1As shown, system architecture 100 may include a data server 101, a network 102, and a host server 103. Network 102 is used to provide a medium for a communication link between data server 101 and host server 103. Network 102 may include various connection types, such as wired or wireless communication links or fiber optic cables.

[0052] The main server 103 may be a server that provides various services, such as a data processing server that processes information uploaded by the data server 101 .

[0053] It should be noted that the method for intelligent step-by-step subsidy issuance provided in the embodiment of the present application is generally executed by the main server 103. Accordingly, the system for intelligent step-by-step subsidy issuance is generally set in the main server 103.

[0054] It should be noted that the data server and master server can be either hardware or software. If implemented as hardware, they can be implemented as a distributed server cluster consisting of multiple servers, or as a single server. If implemented as software, they can be implemented as multiple software programs or software modules (e.g., software or software modules used to provide distributed services), or as a single software program or software module.

[0055] It should be understood that Figure 1 The numbers of data servers, networks and main servers in the embodiment are merely illustrative. Any number of terminal devices, networks and servers may be provided as required.

[0056] Figure 2 FIG1 shows a flow chart of a method for intelligent step-by-step subsidy issuance according to an embodiment of the present application. Figure 2 As shown, the method includes the following steps:

[0057] S1: Construct a user feature vector based on the user's consumption data, process the user features based on the multi-dimensional feature dynamic encoder, and obtain optimized feature data.

[0058] In a specific embodiment, processing user features based on a multi-dimensional feature dynamic encoder specifically includes:

[0059] S101: Using an improved Transformer architecture and combining it with a spatiotemporal attention mechanism to calculate dynamic weights between features ,in, and are the vector representations of the query and key respectively, For all key vectors Normalize, is the feature dimension;

[0060] S102: According to the feature importance decay factor , perform time-weighted processing on historical data, where represents the time decay factor, is the time decay rate, is the interval between the current time and the data generation time, is the characteristic stability function, is the optimized user feature vector.

[0061] S2: Using a dynamic game-reinforcement learning framework, we optimize the subsidy strategy through the interaction between enterprise agents and user group agents. The specific steps include:

[0062] S201: A two-layer Markov decision process is constructed using a reinforcement learning algorithm. The first layer is executed by the enterprise agent, which calculates the optimal subsidy strategy based on a deep reinforcement learning algorithm. The second layer is executed by the user group agent, which simulates user decision-making behavior and predicts users' strategic responses.

[0063] S202: Setting reward functions for the enterprise agent and the user group agent. The reward function for the enterprise agent is used to evaluate subsidy utilization, fairness, and user behavior changes. The reward function for the user group agent is used to evaluate the subsidy benefits obtained by users and possible abnormal arbitrage costs.

[0064] S203: Calculate the user's initial subsidy amount based on the decision of the reinforcement learning agent.

[0065] In a specific embodiment, the reward function calculation formula of the enterprise agent is as follows: ,in, is the reward value of the enterprise agent, is the weight parameter for reward calculation, To subsidize utilization, Fairness represents the fairness of subsidy distribution among different groups, Indicates the degree of change in user behavior distribution, Represents the user's historical behavior distribution, Represents the current behavior distribution of the user; the reward function calculation formula of the user group agent is as follows: ,in, is the reward value of the user group agent, is the weight parameter for reward calculation, and Benefit is the amount of subsidy received by the user. is the anomaly detection penalty value for the user.

[0066] S3: Data collaboration is conducted through federated learning technology, and anomaly detection models are trained across institutions. Based on adversarial training, malicious arbitrage behavior by users is detected and subsidies are adjusted. Specifically, the process includes the following sub-steps: S301: Data collaboration across institutions is conducted based on federated learning. Specifically, each institution independently trains a user behavior model on a local server and, without sharing the original data, exchanges parameters using model encryption technology to build a globally optimized anomaly detection model. Knowledge distillation is used to dynamically weight models across institutions.

[0067] S302: Generate abnormal user behavior samples through adversarial learning method, and optimize the anomaly detection model based on adversarial training;

[0068] S303: Based on the calculation results of the anomaly detection model, the user's subsidy amount is adjusted. If the user is determined to have engaged in high-risk arbitrage behavior, the subsidy amount is reduced or the subsidy application is rejected.

[0069] In a specific embodiment, the adversarial sample generation formula of the anomaly detection model is as follows: ,in, is the adversarial sample perturbation value, are the parameters of the anomaly detection model, For input data, is the loss function of the anomaly detection model, is the regularization weight, is the norm constraint of the adversarial sample perturbation.

[0070] S4: Based on an adaptive tiered subsidy calculation mechanism, the subsidy amount is dynamically adjusted, the final subsidy amount is calculated based on user behavior characteristics and historical records, and the subsidy distribution is automatically executed through smart contracts. Specifically, it includes the following:

[0071] S401: Using a differentiable step subsidy function, dynamically adjust the subsidy amount based on user characteristics and subsidy policy requirements, and predict future subsidy demand changes based on historical subsidy records;

[0072] S402: Calculate the final subsidy amount. If the user meets the long-term stable use subsidy conditions, an additional subsidy reward will be added; if the user has potential abnormal behavior, the subsidy amount will be reduced.

[0073] In a specific embodiment, the calculation formula of the differentiable step subsidy function is as follows: ,in, The calculation result of the subsidy amount is: is the number of subsidy steps, is the subsidy step smoothing parameter, For the Level subsidy threshold, is the activation function, is the subsidy amount for the corresponding level; the final subsidy amount is calculated as follows: ,in, is the final subsidy amount, is the subsidy amount after adjustment for abnormality detection, The amount of additional reward subsidy is calculated based on the user's long-term usage.

[0074] In a specific application embodiment, in order to more clearly illustrate the application scenario of the present invention, the actual implementation process of the present invention is described in detail below in combination with the platform subsidy strategy:

[0075] A platform hopes to use subsidies to incentivize members to increase consumption frequency and amount while preventing malicious arbitrage. The platform hopes to dynamically adjust its subsidy strategy to ensure that subsidies reach high-quality users while controlling costs.

[0076] Data collection and feature coding: The platform collects users' non-privacy behavior data, including: consumption amount: average monthly consumption in the past six months, such as 300 yuan, 800 yuan, and 1,500 yuan; consumption frequency: average number of orders per month, such as 2 times, 5 times, and 10 times; product category preference: for example, user A mainly purchases electronic products, and user B prefers daily necessities; membership level: ordinary member, silver card, gold card, diamond card; historical subsidy record: number and amount of subsidies received in the past six months, for example, a cumulative amount of 200 yuan.

[0077] The platform processes the above data through a multi-dimensional feature dynamic encoder: it uses an improved Transformer architecture to extract feature relationships; and it calculates the time decay factor to adjust the weight of historical data: For example, for user A: the average monthly consumption in the last three months is 1,500 yuan, and the order is placed 10 times per month; the system adjusts the historical weight according to the time decay factor to make the recent behavior more influential, and obtains the final feature vector.

[0078] To optimize subsidy strategies, the platform constructs a two-layer Markov decision process: the enterprise agent (on the platform side) uses the TD3 algorithm to learn the optimal subsidy strategy; the user group agent uses the PPO algorithm to simulate user behavior and predict their decisions under different subsidy tiers. Assume the platform sets the following initial tiers: 20 yuan subsidy for purchases over 500 yuan; 50 yuan subsidy for purchases over 1,000 yuan; and 120 yuan subsidy for purchases over 2,000 yuan. For user A, the predicted spending over the next three months is 1,800 yuan, resulting in a 50 yuan subsidy according to the basic strategy. The reinforcement learning model determines that the user is highly active and automatically increases the subsidy to 80 yuan to enhance loyalty.

[0079] For anomaly detection and risk control, the platform aggregates user behavior data from partner merchants through federated learning and trains anomaly detection models: local training: each merchant independently trains a user behavior model; global aggregation: dynamic weighted knowledge distillation (DWKD) is used to aggregate the global model. Suppose user B suddenly places 50 small orders within 3 days, each of which is exactly 500 yuan. The model detects abnormal behavior and outputs a risk probability. =0.95. Platform strategy: If If the risk is >0.9, the subsidy will be frozen or reduced. User B's subsidy will be adjusted from 200 yuan to 50 yuan, and a risk warning will be issued.

[0080] Adaptive tiered subsidy calculation: The platform dynamically adjusts subsidy amounts based on user characteristics and anomaly detection results: For example, if user C spends 2,500 yuan, the model automatically smoothes the subsidy ladder: the subsidy of 120 yuan is increased to 150 yuan to encourage the user to maintain high consumption behavior.

[0081] Final subsidy amount calculation: , assuming: = 150 yuan (if the abnormality test is passed, the subsidy will not be reduced); = 30 yuan (loyalty reward). End user C receives subsidy: 180 yuan.

[0082] After the smart contract is executed and the subsidy strategy is determined, the platform automatically executes the distribution process through the blockchain smart contract: Subsidy evidence: the subsidy calculation results are stored on the chain to ensure that they cannot be tampered with; real-time distribution: the contract triggers the fund allocation and the 180 yuan subsidy is directly sent to the user's account.

[0083] Implementation Results: Through actual testing, the platform achieved the following results: a 42% increase in resource utilization; 98.7% accuracy in identifying abnormal behavior; reduced policy adjustment response time to 5 minutes; and a 32% increase in user retention. This example fully demonstrates the effectiveness and advancement of this invention, enabling accurate and efficient subsidy disbursement in complex user behavior environments.

[0084] This invention provides an adaptive tiered subsidy disbursement method and system based on a multimodal dynamic game-reinforcement learning approach. This approach dynamically adjusts subsidy policies through reinforcement learning, ensuring that each subsidy encourages high-value user behavior and avoiding resource waste caused by blind disbursement. Experimental results show that resource utilization improves by over 40% compared to traditional fixed-tiered policies. Using a two-layer Markov decision process, the enterprise agent learns from market feedback in real time, while the user agent continuously adjusts its behavior. The game between the two parties drives rapid strategy convergence. Subsidy policy response time is shortened from hours to minutes, ensuring the policy can quickly adapt to market fluctuations. Combining federated learning and adversarial training techniques enables cross-platform data collaboration, allowing global anomaly detection models to be trained without sharing private data. The accuracy of identifying abnormal behavior reaches 98.7%, effectively curbing malicious arbitrage and abnormal subsidy requests. A multi-dimensional feature dynamic encoder extracts user behavior features, combined with an adaptive tiered subsidy mechanism, dynamically adjusting subsidy amounts based on long-term user behavior. Personalized incentives increase user activity and retention by approximately 30%. Subsidy disbursement is executed using blockchain smart contracts, and subsidy calculation results are automatically stored, ensuring that disbursement rules cannot be tampered with and enhancing corporate credibility. Users can check their subsidy status in real time to enhance their trust and participation.

[0085] Figure 3 The framework diagram of a system for intelligent step subsidy issuance of a specific embodiment of the present application is shown, and the system includes a data acquisition and coding unit 301, a subsidy optimization unit 302, a federal adversarial verification unit 303, and an adaptive step subsidy calculation and issuance unit 304. The data acquisition and coding unit 301 is configured to construct a user feature vector based on the user's consumption data, process the user features based on a multi-dimensional feature dynamic encoder, and obtain optimized feature data; the subsidy optimization unit 302 is configured to adopt a dynamic game-reinforcement learning framework to optimize the subsidy strategy through the interaction between the enterprise intelligent agent and the user group intelligent agent; the federal adversarial verification unit 303 is configured to perform data collaboration through federated learning technology, cross-institutional training of anomaly detection models, and detect users' malicious arbitrage behavior based on adversarial training, and adjust subsidy issuance; the adaptive step subsidy calculation and issuance unit 304 is configured to dynamically adjust the subsidy amount based on the adaptive step subsidy calculation mechanism, calculate the final subsidy amount based on user behavior characteristics and historical records, and automatically execute subsidy issuance through smart contracts. Each unit of the system can execute the aforementioned Figure 2 The specific steps of the method.

[0086] Reference below Figure 4 , which shows a structural diagram of a computer system suitable for implementing an electronic device of an embodiment of the present application. Figure 4 The electronic device shown is merely an example and should not limit the functions and scope of use of the embodiments of the present application.

[0087] like Figure 4 As shown, the computer system includes a central processing unit (CPU) 401, which can perform various appropriate actions and processes according to programs stored in a read-only memory (ROM) 402 or programs loaded from a storage unit 408 into a random access memory (RAM) 403. Various programs and data required for system operation are also stored in RAM 403. CPU 401, ROM 402, and RAM 403 are connected to each other via a bus 404. An input / output (I / O) interface 405 is also connected to bus 404.

[0088] The following components are connected to the I / O interface 405: an input section 406 including a keyboard, a mouse, and the like; an output section 407 including devices such as a liquid crystal display (LCD) and speakers; a storage section 408 including devices such as a hard disk; and a communication section 409 including a network interface card such as a LAN card or a modem. The communication section 409 performs communication processing via a network such as the Internet. A drive 410 is also connected to the I / O interface 405 as needed. Removable media 411, such as a magnetic disk, an optical disk, a magneto-optical disk, or a semiconductor memory, is installed in the drive 410 as needed, so that computer programs read from the media can be installed in the storage section 408 as needed.

[0089] In particular, according to embodiments of the present disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present disclosure include a computer program product comprising a computer program carried on a computer-readable storage medium, the computer program containing program code for executing the method illustrated in the flowchart. In such embodiments, the computer program can be downloaded and installed from a network via communication portion 409 and / or installed from removable media 411. When executed by central processing unit (CPU) 401, the computer program performs the functions defined in the methods of this application. It should be noted that the computer-readable storage medium of this application can be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. The computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to, an electrical connection having one or more conductors, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing. In this application, a computer-readable storage medium may be any tangible medium that contains or stores a program for use by or in conjunction with an instruction execution system, apparatus, or device. In this application, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such a propagated data signal may take various forms, including, but not limited to, electromagnetic signals, optical signals, or any suitable combination of the foregoing. A computer-readable signal medium may also be any computer-readable storage medium other than a computer-readable storage medium that can transmit, propagate, or transfer a program for use by or in conjunction with an instruction execution system, apparatus, or device. The program code contained on the computer-readable storage medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.

[0090] Computer program code for performing the operations of the present application can be written in one or more programming languages, or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a stand-alone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computer (e.g., through the Internet using an Internet service provider).

[0091] The flow charts and block diagrams in the accompanying drawings illustrate the possible architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present application. In this regard, each box in the flow chart or block diagram can represent a module, program segment or a part of code, and the module, program segment or a part of code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order than that marked in the accompanying drawings. For example, two boxes represented in succession 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 box in the block diagram and / or flow chart, and the combination of the boxes in the block diagram and / or flow chart can be implemented by a dedicated hardware-based system that performs the specified function or operation, or can be implemented by a combination of dedicated hardware and computer instructions.

[0092] The modules described in the embodiments of the present application may be implemented by software or hardware.

[0093] As another aspect, the present application also provides a computer-readable storage medium, which may be included in the electronic device described in the above embodiment; or it may exist independently and not be assembled into the electronic device. The above-mentioned computer-readable storage medium carries one or more programs. When the above-mentioned one or more programs are executed by the electronic device, the electronic device: constructs a user feature vector based on the user's consumption data, processes the user features based on a multi-dimensional feature dynamic encoder, and obtains optimized feature data; adopts a dynamic game-reinforcement learning framework to optimize the subsidy strategy through the interaction between the enterprise intelligent agent and the user group intelligent agent; conducts data collaboration through federated learning technology, trains anomaly detection models across institutions, and detects users' malicious arbitrage behavior based on adversarial training, and adjusts subsidy issuance; dynamically adjusts the subsidy amount based on an adaptive step subsidy calculation mechanism, calculates the final subsidy amount based on user behavior characteristics and historical records, and automatically executes subsidy issuance through smart contracts.

[0094] The above description is merely a preferred embodiment of the present application and an illustration of the technical principles employed. Those skilled in the art should understand that the scope of the invention herein is not limited to technical solutions formed by a specific combination of the aforementioned technical features. It also encompasses other technical solutions formed by any combination of the aforementioned technical features or their equivalents, without departing from the aforementioned inventive concept. For example, a technical solution formed by replacing the aforementioned features with (but not limited to) technical features with similar functions disclosed in this application.

Claims

1. A method for intelligent tiered subsidy issuance, characterized in that: include: S1: Construct a user feature vector based on the user's consumption data, process the user features based on a multi-dimensional feature dynamic encoder, and obtain optimized feature data; S2: Adopting a dynamic game-reinforcement learning framework, the subsidy strategy is optimized through the interaction between enterprise agents and user group agents. S3: Use federated learning technology to collaborate on data, train anomaly detection models across institutions, and detect malicious arbitrage behavior from users based on adversarial training, adjusting subsidy payments. S4: Based on an adaptive tiered subsidy calculation mechanism, the subsidy amount is dynamically adjusted, the final subsidy amount is calculated based on user behavior characteristics and historical records, and subsidy distribution is automatically executed through smart contracts; The S2 specifically includes: S201: Using a reinforcement learning algorithm to construct a two-layer Markov decision process, wherein the first layer is executed by the enterprise agent and calculates the optimal subsidy strategy based on the deep reinforcement learning algorithm; the second layer is executed by the user group agent and simulates user decision-making behavior and predicts users' strategic responses; S202: Setting reward functions for the enterprise agent and the user group agent, wherein the reward function for the enterprise agent is used to evaluate subsidy utilization, fairness, and user behavior changes, and the reward function for the user group agent is used to evaluate the subsidy benefits obtained by the user and possible abnormal arbitrage costs; S203: Calculate the user's initial subsidy amount based on the decision of the reinforcement learning agent; The reward function calculation formula of the enterprise agent is as follows: ,in, is the reward value of the enterprise agent, is the weight parameter for reward calculation, is the subsidy utilization rate, represents the fairness of subsidy distribution among different groups, Indicates the degree of change in user behavior distribution, Represents the user's historical behavior distribution, represents the current behavior distribution of the user; the reward function calculation formula of the user group agent is as follows: ,in, is the reward value of the user group agent, is the weight parameter for reward calculation, and Benefit is the amount of subsidy received by the user. is the user's anomaly detection penalty value.

2. The method for intelligent step-by-step subsidy distribution according to claim 1, characterized in that: The processing of user features based on the multi-dimensional feature dynamic encoder in S1 specifically includes: S101: Using an improved Transformer architecture and combining it with a spatiotemporal attention mechanism to calculate dynamic weights between features ,in, and are the vector representations of the query and key respectively, Represents indexing of all key vectors Normalize, is the feature dimension; S102: According to the feature importance decay factor , perform time-weighted processing on historical data, where represents the time decay factor, is the time decay rate, is the interval between the current time and the data generation time, is the characteristic stability function, is the optimized user feature vector.

3. The method for intelligent step-by-step subsidy distribution according to claim 1, characterized in that: The S3 specifically includes: S301: Cross-institutional data collaboration based on federated learning. Specifically, each institution independently trains user behavior models on local servers and exchanges parameters through model encryption technology without sharing original data to build a globally optimized anomaly detection model. Knowledge distillation is used to dynamically weight models across institutions. S302: Generate abnormal user behavior samples through an adversarial learning method, and optimize the anomaly detection model based on adversarial training; S303: Based on the calculation result of the anomaly detection model, the subsidy amount of the user is adjusted. If the user is determined to have engaged in high-risk arbitrage behavior, the subsidy amount is reduced or the subsidy application is rejected.

4. The method for intelligent step-by-step subsidy distribution according to claim 3, characterized in that: The adversarial sample generation formula of the anomaly detection model is as follows: ,in, is the adversarial sample perturbation value, are the parameters of the anomaly detection model, For input data, is the loss function of the anomaly detection model, is the regularization weight, is the norm constraint of the adversarial sample perturbation.

5. The method for intelligent step-by-step subsidy distribution according to claim 1, characterized in that: The S4 specifically includes: S401: Using a differentiable step subsidy function, dynamically adjust the subsidy amount based on user characteristics and subsidy policy requirements, and predict future subsidy demand changes based on historical subsidy records; S402: Calculate the final subsidy amount. If the user meets the long-term stable use subsidy conditions, an additional subsidy reward will be added; if the user has potential abnormal behavior, the subsidy amount will be reduced.

6. The method for intelligent step-by-step subsidy distribution according to claim 5, characterized in that: The calculation formula of the differentiable step subsidy function is as follows: ,in, The calculation result of the subsidy amount is: is the number of subsidy steps, is the subsidy step smoothing parameter, For the Level subsidy threshold, is the activation function, is the subsidy amount for the corresponding level; the formula for calculating the final subsidy amount is as follows: ,in, is the final subsidy amount, is the subsidy amount after adjustment for abnormality detection, The amount of additional reward subsidy is calculated based on the user's long-term usage.

7. A computer-readable storage medium having one or more computer programs stored thereon, characterized in that: When the one or more computer programs are executed by a computer processor, the method according to any one of claims 1 to 6 is implemented.

8. A system for intelligent tiered subsidy distribution, characterized in that: include: Data acquisition and encoding unit: configured to construct a user feature vector based on the user's consumption data, process the user features based on a multi-dimensional feature dynamic encoder, and obtain optimized feature data; Subsidy Optimization Unit: This unit is configured to optimize subsidy strategies through the interaction between enterprise agents and user group agents using a dynamic game-reinforcement learning framework. Federated adversarial verification unit: This unit is used to collaborate on data through federated learning technology, train anomaly detection models across institutions, and detect malicious arbitrage behavior based on adversarial training, thereby adjusting subsidy distribution. Adaptive Tiered Subsidy Calculation and Distribution Unit: This unit is configured to dynamically adjust subsidy amounts based on an adaptive tiered subsidy calculation mechanism, calculate the final subsidy amount based on user behavior characteristics and historical records, and automatically distribute subsidies through smart contracts. The subsidy optimization unit specifically includes: S201: Using a reinforcement learning algorithm to construct a two-layer Markov decision process, wherein the first layer is executed by the enterprise agent and calculates the optimal subsidy strategy based on the deep reinforcement learning algorithm; the second layer is executed by the user group agent and simulates user decision-making behavior and predicts users' strategic responses; S202: Setting reward functions for the enterprise agent and the user group agent, wherein the reward function for the enterprise agent is used to evaluate subsidy utilization, fairness, and user behavior changes, and the reward function for the user group agent is used to evaluate the subsidy benefits obtained by the user and possible abnormal arbitrage costs; S203: Calculate the user's initial subsidy amount based on the decision of the reinforcement learning agent; The reward function calculation formula of the enterprise agent is as follows: ,in, is the reward value of the enterprise agent, is the weight parameter for reward calculation, is the subsidy utilization rate, represents the fairness of subsidy distribution among different groups, Indicates the degree of change in user behavior distribution, Represents the user's historical behavior distribution, represents the current behavior distribution of the user; the reward function calculation formula of the user group agent is as follows: ,in, is the reward value of the user group agent, is the weight parameter for reward calculation, and Benefit is the amount of subsidy received by the user. is the user's anomaly detection penalty value.

Images