Intelligent generation method and system for low-income population assistance strategy under rural revitalization background

Abstract

The application discloses a low-income population aid strategy intelligent generation method and system under the background of rural revitalization, relates to the technical field of rural digital governance, and comprises the following steps: obtaining multi-source data of low-income families from a rural digital governance platform and a related government information system, and performing data denoising and feature extraction processing on the edge side to form a standardized data vector; performing privacy protection type data fusion modeling on the standardized data vector, and constructing a multi-dimensional risk identification model to output a family risk label vector; constructing a reinforcement learning decision model based on the family risk label vector to generate an aid strategy combination, and dynamically optimizing and updating the aid strategy combination according to real-time feedback data. The method disclosed by the application realizes the intelligent generation of the low-income family aid strategy by introducing the integration of multi-source data, the privacy protection type data fusion modeling and the dynamic optimization and updating mechanism based on the reinforcement learning.

Description

Technical Field

[0001] This invention relates to the field of rural digital governance technology, specifically to a method and system for intelligently generating support strategies for low-income populations in the context of rural revitalization. Background Technology

[0002] With the advancement of rural revitalization, supporting low-income families has become a key focus of social attention. In recent years, with the rapid development of digital technology and information technology infrastructure, rural digital governance platforms and government information systems have provided crucial data support for targeted assistance to low-income families. These digital platforms integrate information from multiple channels, such as basic data on family size, income, education, and health, to identify families in need and provide targeted support. Furthermore, with the advancement of artificial intelligence (AI) and big data technologies, utilizing machine learning models to process and analyze this multi-source data has become an important means of improving the efficiency of assistance efforts. In particular, by denoising, extracting features, and modeling this data, family risks can be effectively predicted, leading to the development of scientific and personalized assistance strategies.

[0003] However, current methods for generating support strategies for low-income families still have some shortcomings. Although existing systems can collect multi-source data, the accuracy and real-time performance of data integration and analysis are insufficient, resulting in support strategies failing to respond promptly to dynamic changes in family needs. Traditional data processing and analysis methods often neglect data privacy protection, leading to significant privacy risks during information exchange and data sharing. Existing risk assessment models often focus only on a single dimension (such as income level or health status), lacking the ability for multi-dimensional risk assessment and comprehensive decision-making. Therefore, existing technologies cannot effectively combine data from various fields to precisely optimize support strategies for low-income families. Furthermore, existing decision-making models are mostly based on static rules or manually set strategies, lacking dynamic optimization and adaptive adjustment based on real-time data, resulting in limited effectiveness of support measures. To address these issues, this invention proposes a more comprehensive, intelligent, and dynamically optimized method for generating support strategies for low-income families by introducing advanced methods such as edge computing, privacy protection technology, federated learning, and reinforcement learning. Summary of the Invention

[0004] In view of the above-mentioned problems, the present invention is proposed.

[0005] Therefore, the technical problem solved by this invention is that existing methods for generating support strategies for low-income families lack real-time data integration and analysis, privacy protection mechanisms, and the ability to conduct multi-dimensional comprehensive evaluation and dynamic optimization. It also addresses the issue of how to generate personalized and real-time optimized support strategies while protecting data privacy.

[0006] To address the aforementioned technical problems, this invention provides the following technical solution: an intelligent generation method for support strategies for low-income populations under the background of rural revitalization, comprising: acquiring multi-source data of low-income families from rural digital governance platforms and related government information systems; performing data denoising and feature extraction processing at the edge to form standardized data vectors; performing privacy-preserving data fusion modeling on the standardized data vectors; constructing a multi-dimensional risk identification model to output family risk label vectors; constructing a reinforcement learning decision model based on the family risk label vectors to generate support strategy combinations; and dynamically optimizing and updating the model based on real-time feedback data.

[0007] As a preferred embodiment of the intelligent generation method for low-income population assistance strategies under the background of rural revitalization described in this invention, the acquisition of multi-source data of low-income families includes: acquiring basic information of villagers, family population structure information, employment and income information, education information, and various subsidy information through a rural digital village platform; acquiring village affairs disclosure records and villager meeting participation records through a village affairs management system; acquiring villager browsing behavior, application records, and feedback behavior data through a villager service mini-program; and connecting the multi-source data to an edge computing gateway through a data interface, and establishing a data registration mechanism to record data source identifiers, data types, collection cycles, and field mapping relationships.

[0008] As a preferred embodiment of the intelligent generation method for low-income population assistance strategies under the background of rural revitalization described in this invention, the step of performing data denoising and feature extraction processing on the edge side includes: smoothing the collected data using a sliding time window by weighting the data within the window; when missing data is detected, compensation is achieved by using effective data before and after the missing interval to estimate the linear trend; normalizing the processed data to map data of different dimensions to a unified numerical range; extracting a household income stability index by statistically analyzing the fluctuation of income time series to obtain a stability value; extracting a health risk fluctuation index by statistically analyzing the frequency of occurrence of abnormal health events within a unit of time to obtain a risk intensity value; extracting an education participation rate index by calculating the proportion of effective learning participation time to the monitoring period to obtain an education activity value; and forming a standardized data vector.

[0009] As a preferred embodiment of the intelligent generation method for low-income population assistance strategies under the background of rural revitalization described in this invention, the privacy-preserving data fusion modeling includes: encrypting and storing each standardized data vector using symmetric encryption, and performing irreversible hash mapping on family identity information to replace the original identity identifier; introducing a differential privacy mechanism during statistical analysis, injecting random perturbation values ​​related to query sensitivity into the statistical results to make it difficult to reverse-identify the contribution of individual family data; and adopting a federated learning architecture during multi-region model training, where each township node trains model parameters locally using the original data, only uploading the model parameters to the central server for aggregation and updating, and co-training the model without uploading the original data.

[0010] As a preferred embodiment of the intelligent generation method for low-income population assistance strategies under the background of rural revitalization described in this invention, the output family risk label vector includes: inputting a standardized data vector into a clustering model, continuously adjusting the position of the classification center to minimize the distance between the family sample and the center of its category, thereby completing the classification of family risk types; establishing a risk prediction model based on the classification results, and obtaining the probability value of future risk occurrence by weighted comprehensive calculation of income stability, health risk intensity, and education participation indicators; marking a high-risk level when the probability value of future risk occurrence exceeds a preset risk threshold; calculating a comprehensive health risk score based on historical abnormal frequency and current risk indicators; and finally generating a family risk label vector containing income risk level, health risk level, and education risk level.

[0011] As a preferred embodiment of the intelligent generation method for low-income population assistance strategies under the background of rural revitalization described in this invention, the generation of assistance strategy combinations includes: defining a family risk label vector and the current available resource set together as a state; defining executable assistance measures as an action set; recording the performance of different strategy combinations in historical execution by establishing a value assessment table between states and actions; calculating an immediate reward value based on the improvement of key family indicators after strategy execution, and iteratively updating the value of the current strategy in combination with future expected returns; outputting the corresponding assistance strategy combination by selecting the action with the highest value assessment, and writing the corresponding strategy into the execution queue.

[0012] As a preferred embodiment of the intelligent generation method for low-income population assistance strategies under the background of rural revitalization described in this invention, the dynamic optimization and updating includes: continuously collecting household indicator data after strategy execution to form a new indicator vector, and calculating the degree of difference between the new and old indicators as the strategy effect deviation value; when the deviation value exceeds a preset threshold, triggering the model retraining process to update the strategy value assessment table and risk weight parameters; simultaneously, generating access record summaries for data access behavior through blockchain nodes, and using smart contracts to verify access requests, allowing data decryption and access only when the preset permission rules are met.

[0013] As a preferred embodiment of the intelligent generation system for low-income population assistance strategies under the background of rural revitalization described in this invention, it includes: a standardization processing module, a risk identification module, and a strategy output module; the standardization processing module is used to obtain multi-source data of low-income families from the rural digital governance platform and related government information systems, and perform data denoising and feature extraction processing at the edge to form a standardized data vector; the risk identification module is used to perform privacy-preserving data fusion modeling on the standardized data vector, and construct a multi-dimensional risk identification model to output a family risk label vector; the strategy output module is used to construct a reinforcement learning decision model based on the family risk label vector to generate a combination of assistance strategies, and dynamically optimize and update it based on real-time feedback data.

[0014] A computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program as a step in realizing an intelligent generation method for assisting low-income populations in the context of rural revitalization.

[0015] A computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the steps of an intelligent generation method for support strategies for low-income populations in the context of rural revitalization are disclosed.

[0016] The beneficial effects of this invention are as follows: The intelligent generation method for low-income population assistance strategies under the background of rural revitalization provided by this invention achieves intelligent generation of assistance strategies for low-income families by introducing the integration of multi-source data, privacy-preserving data fusion modeling, and a dynamic optimization and update mechanism based on reinforcement learning. By acquiring multi-source data from rural digital governance platforms and related government systems, and performing noise reduction and feature extraction processing through edge computing, the high quality and real-time nature of the data are ensured. The use of privacy-preserving technologies, such as symmetric encryption, differential privacy, and federated learning, ensures that the personal information of low-income families is effectively protected when analyzing and sharing data, avoiding the risk of privacy leakage. Simultaneously, by combining multi-dimensional risk assessments such as income stability, health risk fluctuations, and educational participation, the comprehensiveness and accuracy of risk identification are further improved. Finally, based on a reinforcement learning-based decision-making model, assistance strategies can be dynamically adjusted according to real-time feedback, achieving personalized and real-time optimized strategy combinations, avoiding the limitations of traditional static rules. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is an overall flowchart of the intelligent generation method for low-income population assistance strategies under the background of rural revitalization provided in Embodiment 1 of the present invention.

[0019] Figure 2 This is a schematic diagram of the risk prediction model for the intelligent generation method of support strategies for low-income populations under the background of rural revitalization provided in Embodiment 1 of the present invention. Detailed Implementation

[0020] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0021] Example 1, referring to Figures 1-2 As an embodiment of the present invention, a method for intelligently generating support strategies for low-income populations in the context of rural revitalization is provided, comprising:

[0022] S1: Obtain multi-source data on low-income families from rural digital governance platforms and related government information systems, and perform data denoising and feature extraction processing at the edge to form a standardized data vector.

[0023] Furthermore, acquiring multi-source data on low-income families includes obtaining basic information about villagers, family population structure, employment and income, education, and various subsidies through the rural digital village platform; obtaining village affairs disclosure records and villager meeting participation records through the village affairs management system; obtaining villager browsing behavior, application records, and feedback behavior data through the villager service mini-program; and connecting the multi-source data to the edge computing gateway through the data interface, and establishing a data registration mechanism to record data source identifiers, data types, collection cycles, and field mapping relationships.

[0024] It should also be noted that a preferred scheme for obtaining multi-source data on low-income families specifically includes using the family as the basic processing unit, and assigning a family number as... ,in , This represents the total number of households included in the system management; let the data source number be... ,in , This indicates the total number of system data sources connected. Data sources must include at least the Rural Digital Village Platform, the Village Affairs Management System, and the Villager Service Mini Program. For any given data source... Establish a registration item in the edge computing gateway , represented as:

[0025]

[0026] in, Indicates the data source identifier. Represents data type, This indicates the collection period of the data source. This indicates the field mapping relationship. Its function is to map fields with different names but the same meaning in different platforms to a unified set of fields. For example, the total number of people in a family in the village affairs platform, the number of living members in the civil affairs platform, and the number of family members in the mini program are all mapped to the family population structure field, thereby ensuring that data from different systems can enter the same computing link.

[0027] It should also be noted that a preferred approach for multi-source data also includes obtaining business data related to farmers' production and life through interfaces of relevant business systems, including agricultural production management systems, social security systems, medical security systems, or education management systems, to supplement household behavioral data in terms of production and operation, social security, and public services; it also includes obtaining environmental and behavioral perception data by accessing IoT devices deployed on the farmer's side or in village-level public areas, including electricity consumption data collection devices, water consumption data collection devices, environmental monitoring devices, or agricultural production monitoring devices, to reflect the household's production and life status and environmental change information.

[0028] For families Data source and unified fields Let the original record be ,in Indicates the timestamp at which the original record was generated. , This indicates the total number of standardized fields maintained by the system in the edge gateway. To address the issue of inconsistent data collection periods across different systems, a standardized time step is pre-set in the edge gateway. And set the unified time index as ,in This represents the total number of time steps within the current observation window. The edge gateway performs time alignment and field aggregation on the raw data according to a uniform time step, obtaining the aligned field values. , is represented as:

[0029]

[0030] in, Indicates that fields can be provided. The collection of data sources Representation field The corresponding aggregate function; when the field When it is a numeric statistical field, It can take the sum, mean, or most recent value when the field is selected. When it is a behavior event field, Retrievable event counts or event occurrence markers. Data from villager basic information, family population structure information, employment and income information, education information, subsidy information, village affairs disclosure records, meeting participation records, browsing behavior, application records, and feedback behavior are uniformly mapped to... The time series field set is represented.

[0031] It should be noted that the data denoising and feature extraction processing performed on the edge side includes: smoothing the collected data using a sliding time window by weighting the data within the window; when missing data is detected, compensation is achieved by using effective data before and after the missing interval to estimate the linear trend; normalizing the processed data to map data of different dimensions to a unified numerical range; extracting a household income stability index by statistically analyzing the volatility of income time series to obtain a stability value; extracting a health risk volatility index by statistically analyzing the frequency of occurrence of abnormal health events within a unit of time to obtain a risk intensity value; extracting an education participation rate index by calculating the proportion of effective learning participation time to the monitoring period to obtain an education activity value; and forming a standardized data vector.

[0032] It should also be noted that a preferred approach for performing data denoising and feature extraction processing at the edge to form a standardized data vector specifically includes, after obtaining the aligned multi-source field sequence, the edge gateway first performs denoising processing. For the fields... Let its smooth window length be... Window weight is The time step index within the sliding time window And satisfy For fields At time step The smoothing result is defined as:

[0033]

[0034] in, This represents the field value after smoothing within a sliding time window. Samples closer to the current time step can have higher weights, thus suppressing occasional fluctuations while preserving recent trends. For high-frequency discrete fields such as browsing behavior and application requests, It can be set to a relatively short duration; for low-frequency fields such as income and subsidy receipts, A longer statistical window can be set to enhance stability. When a field has missing values, the edge gateway performs trend compensation for the missing values. For this purpose, a missing value indicator variable is defined. Represented as:

[0035]

[0036] like In the same field Finding distance in time series Most recent forward effective time step With backward effective time step The missing values ​​are estimated according to a linear trend to obtain the compensation results:

[0037]

[0038] when At that time, directly command:

[0039]

[0040] in, This represents the field value after denoising and missing value compensation. and Representing fields respectively In the family The nearest preceding and following valid time steps in the sequence. For fields with no data for a long time, the edge gateway can set a maximum compensation span, when... When the maximum compensation span is exceeded, only the missing markers are retained without mandatory compensation, in order to avoid the spread of errors caused by extrapolation over long intervals.

[0041] To eliminate the differences in units between different fields, normalization is performed on the compensated field values. Let the fields be... The effective lower bound is The effective upper bound is ,in The normalized field value Defined as:

[0042]

[0043] when When it is truncated to 0, This is truncated to 1, thus mapping all fields uniformly to the numerical range [0, 1]. Here, and It can be pre-set based on business rules, platform field definitions, or historical sample statistical results. For example, the subsidy amount field can adopt the local policy range, and the village meeting participation record field can adopt the experience upper limit of event frequency. With this structure, subsequent numerical values ​​from different sources and at different scales can be directly incorporated into a unified feature calculation process.

[0044] In the feature extraction stage, this embodiment preferably extracts indicators of household income stability, health risk fluctuation, and education participation rate. First, for income stability, let the set of income-related fields be... ,in At least include fields for employment and income information, work records, and subsidy receipt; for families At time step Income observations are defined as

[0045]

[0046] in, family The comprehensive income observation sequence, Representation field Weighting in income observation. Let the window length for income stability calculation be... Then the mean income and the intensity of income volatility within this window are respectively:

[0047]

[0048]

[0049] Furthermore, the stability of household income Defined as:

[0050]

[0051] in, Indicates recent Average income level within a time step Indicates the intensity of income fluctuations within the same window. To prevent extremely small positive numbers with a denominator of zero, according to this definition, if a family's income level is high and fluctuates little, then... The value is greater; if income fluctuates frequently, then The value will decrease accordingly. For households with incomplete income fields, an approximate income proxy can be constructed from employment records, subsidy payment records, and application records, but the weights of the corresponding fields will be reduced. The edge gateway needs to adjust the settings based on data reliability.

[0052] Secondly, regarding health risk fluctuation indicators, this embodiment specifically defines abnormal health events as health risk-related events that can be recorded by the platform. Let the set of health risk event types be... This includes at least the following event types: medical assistance applications, new or renewed disability subsidies, triggering of senior citizen subsidies, registration of serious illnesses, and feedback on health assistance requests; for families At time step The The count of health risk events is recorded as And set the risk weight of this event type as . Then the family At time step Health risk intensity Defined as:

[0053]

[0054] Let the length of the health risk statistics window be... The health risk fluctuation index is defined as follows:

[0055]

[0056] in, family In recent The average health risk intensity within a time step You can use a binary flag indicating whether an event has occurred, or you can use the number of times the event occurred within the same time step. This method is used to differentiate the importance of different health risk events to subsequent assistance decisions. After this processing, health risk-related information that was originally scattered across subsidies, relief, application processes, and feedback is transformed into quantifiable and comparable time-series risk intensity.

[0057] Secondly, for the education participation rate indicator, this embodiment uses the equivalent learning participation time method for calculation. Let the set of education participation event types be... This includes at least the following event types: valid school status, updated student registration information, application for education subsidies, browsing of education policies, and feedback on education matters; for families At time step The The number of educational participation events is recorded as follows And pre-set an equivalent duration coefficient for each type of event. Then the family At time step Equivalent learning engagement time Defined as:

[0058]

[0059] Let the length of the statistical window for education participation be... The education participation rate is defined as:

[0060]

[0061] in, Indicates recent Family within a time step Educational activity To unify the time step, therefore This represents the total observation duration corresponding to the statistical window for educational participation. Equivalent duration coefficient. This method is used to convert different types of educational behaviors to a uniform duration scale. For example, effective school attendance corresponds to continuous participation, educational subsidy applications correspond to phased participation in educational services, and browsing and providing feedback on educational policies correspond to interactive participation in educational services. Through this approach, even if the original system does not directly record actual online learning time, the level of educational participation can be extracted from available government and platform behavioral data.

[0062] After calculating the three core metrics, the edge gateway further generates standardized data vectors. (Assuming a household...) The standardized data vector at the current moment for:

[0063]

[0064] in, As an indicator of income stability, As an indicator of health risk fluctuations, As an indicator of education participation rate, For data quality coefficient, This refers to the confidence coefficient of the indicator. Preferably, the data quality coefficient can be defined as:

[0065]

[0066] in, This represents the set of fields involved in the calculation of key performance indicators. This indicates the number of fields in the field set. Since the indicator variable is missing, therefore It reflects the family The overall percentage of valid data in key fields. Indicator confidence coefficient. This can be determined by a combination of factors including the coverage of effective samples, the reliability of field sources, and the compensation ratio, and in conjunction with... The data is output along with the standardized data vector. Thus, the edge side not only outputs numerical features that can be directly used for modeling, but also outputs data quality and confidence information, enabling the subsequent risk identification stage to distinguish between high-confidence and low-confidence household samples, thereby improving the feasibility and robustness of the overall model.

[0067] It should also be noted that in the generation of support strategies for low-income families, existing technologies typically rely on traditional data sources for data collection and integration, and often lack effective denoising and feature extraction mechanisms during data processing. This leads to low data quality, making it difficult to extract meaningful information from multi-source data, thus affecting the accuracy of subsequent analysis and decision-making. This invention collects multi-source data on low-income families from multiple platforms (such as rural digital governance platforms and village affairs management systems), and combines edge computing for data denoising and feature extraction, improving data quality and real-time performance. In particular, the use of a unified time step solves the problem of inconsistent timestamps between different data sources, thereby ensuring the synchronization and consistency of multi-source data. This not only improves the accuracy of data integration but also ensures data stability and integrity through techniques such as sliding time window smoothing and missing data compensation. Furthermore, a unified field mapping mechanism standardizes identical semantic fields from different platforms, avoiding errors caused by inconsistent data fields.

[0068] S2: Perform privacy-preserving data fusion modeling on standardized data vectors and construct a multi-dimensional risk identification model to output family risk label vectors.

[0069] Furthermore, privacy-preserving data fusion modeling includes: using symmetric encryption to encrypt and store each standardized data vector; and performing irreversible hash mapping on family identity information to replace the original identity identifier; introducing a differential privacy mechanism during statistical analysis by injecting random perturbation values ​​related to query sensitivity into the statistical results, making it difficult to reverse-identify the contribution of individual family data; and adopting a federated learning architecture during multi-region model training, where each township node trains model parameters locally using the original data, and only uploads the model parameters to the central server for aggregation and updates, thus enabling collaborative model training without uploading the original data.

[0070] It should also be noted that a preferred scheme for privacy-preserving data fusion modeling specifically includes using symmetric encryption technology to encrypt and store each standardized data vector to ensure user privacy is not compromised. Let the encrypted standardized data vector be... For the family The encrypted storage operation can be represented as:

[0071]

[0072] in, This represents a symmetric encryption algorithm. For encryption key, This is a standardized data vector. The encrypted data can only be accessed by the holder of the key. The decryption is performed by authorized visitors. Simultaneously, to protect family identity information, the original identity identifier is replaced with an irreversible hash mapping. Let the original identity identifier be... The hashed identity identifier is ,but:

[0073]

[0074] in, This is a hash function. Through hashing, family identity information is anonymized, preventing the leakage of original identities. Furthermore, a differential privacy mechanism is introduced during statistical analysis to prevent the reverse identification of individual family data contributions. Random perturbation values ​​are injected into the statistical results to ensure privacy protection. Let the statistical results be... The disturbance value is Differential privacy mechanism can be represented as:

[0075]

[0076] in, It is related to query sensitivity Disturbance values ​​related to household information. To meet differential privacy requirements, that is, statistical results should be protected under the premise of privacy protection. This makes it impossible to deduce the individual data contributions of households. To ensure the privacy of data from different regions, multi-regional collaborative modeling employs a federated learning architecture for model training. Under this architecture, each township node trains its model parameters locally using raw data, and then uploads the model parameters to the central server for aggregation and updates. Let the local model parameters of each township node be... The local model training process can be represented as follows:

[0077]

[0078] in, This indicates the local training process. Township nodes The original dataset is uploaded. Each township node only uploads the model parameters trained locally. Instead of uploading the original data, the central server receives the model parameters from all nodes and updates the global model parameters through aggregation operations.

[0079]

[0080] in, Represents aggregate functions, This represents the model parameters after global aggregation. This indicates the number of nodes in the region, which is consistent with the total number of system data sources connected.

[0081] It should be noted that the output family risk label vector includes: inputting standardized data vectors into a clustering model, continuously adjusting the location of the classification centers to minimize the distance between family samples and their respective category centers, thus completing the classification of family risk types; establishing a risk prediction model based on the classification results, and obtaining the probability value of future risk occurrence by weighted comprehensive calculation of indicators such as income stability, health risk intensity, and education participation; marking a high-risk level when the probability value of future risk occurrence exceeds a preset risk threshold; calculating a comprehensive health risk score based on historical anomaly frequency and current risk indicators; and finally generating a family risk label vector containing income risk level, health risk level, and education risk level.

[0082] It should also be noted that a preferred scheme for constructing a multidimensional risk identification model to output family risk label vectors specifically includes, after completing privacy-preserving data fusion modeling, classifying the standardized data vectors into risk types using a clustering model. Let the set of risk types be... Clustering algorithms adjust the classification centers iteratively. Make each family sample Minimize the distance from the center of its category; specifically, the optimization objective is:

[0083]

[0084] in, For category Cluster centers To belong to category All families For family The standardized data vector is used. By minimizing this objective function, low-income families can be automatically classified into different risk types. Based on the classification results, a risk prediction model is further established. Let the weighted composite coefficients of income stability, health risk intensity, and education participation be respectively... Then the family Probability of future risks It can be obtained through the following weighted calculation:

[0085]

[0086] in, Each family Indicators such as income stability, health risk intensity, and education participation. These are the corresponding weighting coefficients. After calculating the probability of future risk occurrence, when... Exceeding the preset risk threshold At that time, family It will be marked as high-risk.

[0087] Meanwhile, to further improve the accuracy of health risk assessment, a comprehensive health risk score is calculated by comprehensively considering the frequency of historical abnormalities and current risk indicators.

[0088]

[0089] in, and These are the weighting coefficients. For family The frequency of abnormal health events within a historical time window. Finally, the output family risk label vector includes income risk level, health risk level, and education risk level, in the following format:

[0090]

[0091] in, , , These represent the risk levels for income, health, and education, respectively.

[0092] It should also be noted that traditional data analysis often lacks sufficient privacy protection measures, especially when it involves the personal information of low-income families, which carries a significant risk of data leakage. Existing technologies also lack a multi-dimensional risk assessment model to ensure accurate identification of various risks faced by low-income families and to provide a basis for targeted assistance. This invention employs symmetric encryption and differential privacy mechanisms to ensure the security of family personal information during data processing and analysis. Regarding multi-dimensional risk identification, this invention combines multiple dimensions such as income stability, health risk fluctuations, and education participation rates for risk assessment, and accurately reflects the different risks faced by families by aggregating the weights of each dimension. This not only effectively protects family privacy but also improves the comprehensiveness and accuracy of risk identification through the combination of multi-dimensional data. In particular, the introduction of a federated learning architecture for model training, where each regional node trains the model locally and uploads parameters instead of data, further enhances data privacy protection. This addresses the shortcomings of existing technologies in privacy protection and multi-dimensional risk identification, ensuring the privacy of low-income families is not leaked, while simultaneously improving the accuracy and effectiveness of the risk assessment model.

[0093] S3: Based on family risk label vectors, a reinforcement learning decision model is constructed to generate a combination of support strategies, and dynamic optimization and updates are performed based on real-time feedback data.

[0094] Furthermore, the generation of support strategy combinations includes defining the family risk label vector and the current available resource set together as a state, defining the executable support measures as an action set, recording the performance of different strategy combinations in historical execution by establishing a value assessment table between states and actions; after the strategy is executed, calculating the immediate reward value based on the improvement of key family indicators, and iteratively updating the value of the current strategy in combination with future expected returns; outputting the corresponding support strategy combination by selecting the action with the highest value assessment, and writing the corresponding strategy into the execution queue.

[0095] It should also be noted that a preferred scheme for constructing a reinforcement learning decision-making model to generate a combination of support strategies specifically includes, firstly, defining the state space, the state of the family. For family At time step The risk label vector and the set of available resources. Let's assume a family... Risk label vector for:

[0096]

[0097] in, They represent families. At time step Income, health, and education risk levels. The set of available resources for the household. This includes available resources, such as poverty alleviation funds, job opportunities, and skills training opportunities, denoted as:

[0098]

[0099] in, It is family At time step Specific resources that are available. Therefore, the family. The state vector can be represented as:

[0100]

[0101] Next, define actionable support measures. For family At time step The set of all available assistance strategies that can be adopted. Assume each assistance measure... This represents a specific assistance program, such as financial aid, vocational training, medical assistance, or educational subsidies; therefore, the scope of action is:

[0102]

[0103] in, This indicates different assistance measures. This represents the total number of support measures.

[0104] By using the Q-learning method in reinforcement learning, a state is established. With action Value assessment table between This table is used to record the value of taking different actions in each state. Assume... Indicates the state Take action below The expected return (i.e., the total future reward). The formula for calculating the updated Q value is:

[0105]

[0106] in, It is family At time step Execute action The immediate reward afterwards It is a discount factor. It's the learning rate. Indicate the next state The maximum Q value.

[0107] Calculate immediate rewards based on the improvement in key family indicators. Assuming a family At time step Key Indicators (Such as income, health, education) are represented as:

[0108]

[0109] in, , , Corresponding to families At time step Income, health, and education status, then family At time step Relative to time step Improvement in key indicators Defined as:

[0110]

[0111] Instant rewards The calculation is performed as follows:

[0112]

[0113] in, These are the weighting coefficients for each indicator. , , They represent families. Between two consecutive time steps ( and The reward calculation method dynamically adjusts the weight of each indicator based on the family's different needs and social background, taking into account changes in key indicators such as income, health, and education.

[0114] After executing the strategy, the action with the highest value is selected based on the updated Q-value. This refers to the optimal assistance strategy under the current circumstances. The formula for selecting the strategy is:

[0115]

[0116] After selecting the optimal action, the system writes the selected strategy into the execution queue, ready for execution.

[0117] It should be noted that dynamic optimization and updates include continuously collecting household indicator data after strategy execution to form new indicator vectors, and calculating the degree of difference between the old and new indicators as the strategy effect deviation value; when the deviation value exceeds a preset threshold, the model retraining process is triggered to update the strategy value assessment table and risk weight parameters; at the same time, access behavior is recorded by generating access record summaries through blockchain nodes, and access requests are verified by smart contracts, allowing data decryption and access only when the preset permission rules are met.

[0118] It should also be noted that a preferred approach for dynamic optimization and updating based on real-time feedback data specifically includes, in order to optimize the effectiveness of the strategy, collecting data on key household indicators in real time to form new indicator vectors. And calculate the degree of difference between the old and new indicators. This serves as a deviation value for the strategy's effectiveness. If the deviation value exceeds a preset threshold... This triggers the model retraining process, updating the policy value evaluation table and risk weight parameters. The specific calculation is as follows:

[0119]

[0120] when If this happens, the model retraining process is triggered, and the Q-value is recalculated and updated.

[0121] It should also be noted that traditional assistance strategies often rely on static rules and manual settings, lacking flexible adaptability and difficulty in adjusting to the real-time needs of families. Furthermore, existing technologies suffer from lag in post-implementation optimization and updates, failing to respond in real-time to changes in the circumstances of low-income families, leading to inefficient assistance. This invention introduces a reinforcement learning decision model to dynamically generate assistance strategy combinations based on the family's multi-dimensional risk label vector and available resources. This model can design personalized strategies according to the specific risks and needs of families and dynamically optimize and update them using real-time feedback data. Specifically, strategy selection and updates not only depend on the family's current risk label but also incorporate iterative optimization based on immediate reward calculations and future expected returns. This solves the problem of traditional static rules' inability to adaptively adjust strategies, enabling assistance strategies to be optimized in real-time according to changes in families. By continuously collecting post-implementation data and calculating the deviation of strategy effects, model retraining is automatically triggered when the deviation exceeds a preset threshold, readjusting strategy evaluation and risk weight parameters. In addition, access control based on blockchain technology ensures data access security, further improving credibility and stability. This not only enhances the intelligence and personalization of assistance strategies, but also ensures the effectiveness of strategy implementation through dynamic optimization, making assistance to low-income families more efficient and precise.

[0122] Example 2, an embodiment of the present invention, provides an intelligent generation system for support strategies for low-income populations in the context of rural revitalization, including a standardization processing module, a risk identification module, and a strategy output module.

[0123] The standardization processing module is used to obtain multi-source data on low-income families from the rural digital governance platform and related government information systems, and perform data denoising and feature extraction processing at the edge to form a standardized data vector; the risk identification module is used to perform privacy-preserving data fusion modeling on the standardized data vector, and construct a multi-dimensional risk identification model to output a family risk label vector; the strategy output module is used to construct a reinforcement learning decision model based on the family risk label vector to generate a combination of assistance strategies, and dynamically optimize and update it based on real-time feedback data.

Claims

1. A method for intelligently generating support strategies for low-income populations under the background of rural revitalization, characterized in that, include: Multi-source data on low-income families are obtained from rural digital governance platforms and related government information systems, and data denoising and feature extraction are performed at the edge to form standardized data vectors. Privacy-preserving data fusion modeling is performed on standardized data vectors, and a multi-dimensional risk identification model is constructed to output family risk label vectors; A reinforcement learning decision-making model is constructed based on family risk label vectors to generate a combination of support strategies, and the model is dynamically optimized and updated based on real-time feedback data.

2. The intelligent generation method for low-income population assistance strategies under the background of rural revitalization as described in claim 1, characterized in that: The acquisition of multi-source data on low-income families includes... The rural digital village platform obtains basic information about villagers, family population structure, employment and income, education, and various subsidies. Obtain village affairs disclosure records and villager meeting participation records through the village affairs management system; Data on villagers' browsing behavior, application records, and feedback behavior are obtained through the village service mini-program. Multi-source data is connected to the edge computing gateway through a data interface, and a data registration mechanism is established to record the data source identifier, data type, collection period, and field mapping relationship.

3. The intelligent generation method for low-income population assistance strategies under the background of rural revitalization as described in claim 2, characterized in that: The data denoising and feature extraction processing performed at the edge includes, The collected data is smoothed using a sliding time window by weighted averaging of the data within the window. When missing data is detected, compensation is achieved by using valid data before and after the missing interval to estimate the linear trend. The processed data is normalized to map data of different dimensions to a unified numerical range; Extract the household income stability index and obtain the stability value by statistically analyzing the degree of fluctuation in the income time series; Extract health risk fluctuation indicators and obtain risk intensity values ​​by statistically analyzing the frequency of occurrence of abnormal health events within a unit of time. The education participation rate indicator is extracted, and the education activity value is obtained by calculating the proportion of effective learning participation time to the monitoring period; a standardized data vector is formed.

4. The intelligent generation method for low-income population assistance strategies under the background of rural revitalization as described in claim 3, characterized in that: The privacy-preserving data fusion modeling includes, Each standardized data vector is encrypted and stored using symmetric encryption, and the family identity information is irreversibly hashed to replace the original identity identifier. When conducting statistical analysis, a differential privacy mechanism is introduced. By injecting random perturbation values ​​related to query sensitivity into the statistical results, the contribution of individual family data is difficult to be reverse-identified. In the multi-region model training process, a federated learning architecture is adopted. Each township node trains the model parameters locally using the raw data, and only uploads the model parameters to the central server for aggregation and update, so as to train the model collaboratively without uploading the raw data.

5. The intelligent generation method for low-income population assistance strategies under the background of rural revitalization as described in claim 4, characterized in that: The output family risk label vector includes, Standardized data vectors are input into a clustering model, and the distance between family samples and their respective category centers is minimized by continuously adjusting the location of the classification centers, thereby completing the classification of family risk types. Based on the classification results, a risk prediction model is established. By weighting and comprehensively calculating the indicators of income stability, health risk intensity, and education participation, the probability value of future risk occurrence is obtained. When the probability of a future risk occurring exceeds a preset risk threshold, it is marked as a high-risk level. Simultaneously, a comprehensive health risk score is calculated based on historical anomaly frequencies and current risk indicators; The final result is a family risk label vector containing income risk level, health risk level, and education risk level.

6. The intelligent generation method for low-income population assistance strategies under the background of rural revitalization as described in claim 5, characterized in that: The generated support strategy combination includes, The family risk label vector and the current set of available resources are defined together as the state, and the executable assistance measures are defined as the set of actions. By establishing a value assessment table between the state and the actions, the effectiveness of different strategy combinations in historical implementation is recorded. After the strategy is implemented, an immediate reward value is calculated based on the improvement of key family indicators, and the value of the current strategy is iteratively updated in combination with expected future returns. The action with the highest value assessment is selected to output the corresponding support strategy combination, and the corresponding strategy is written into the execution queue.

7. The intelligent generation method for low-income population assistance strategies under the background of rural revitalization as described in claim 6, characterized in that: The dynamic optimization update includes, After the strategy is implemented, household indicator data is continuously collected to form a new indicator vector, and the degree of difference between the old and new indicators is calculated as the strategy effect deviation value. When the deviation value exceeds the preset threshold, the model retraining process is triggered, and the strategy value assessment table and risk weight parameters are updated again. Meanwhile, the blockchain nodes generate access record summaries for data access behavior, and smart contracts are used to verify access requests, allowing data decryption and access only when the preset permission rules are met.

8. A system for intelligently generating strategies to support low-income populations under the background of rural revitalization, employing the method for intelligently generating strategies to support low-income populations under the background of rural revitalization as described in any one of claims 1 to 7, characterized in that: It includes a standardized processing module, a risk identification module, and a strategy output module; The standardized processing module is used to collect multi-source data from IoT collection terminals deployed in low-income families, and to perform data denoising and feature extraction processing at the edge to form a standardized data vector. The risk identification module is used to perform privacy-preserving data fusion modeling on standardized data vectors and construct a multi-dimensional risk identification model to output a family risk label vector. The strategy output module is used to build a reinforcement learning decision model based on family risk label vectors to generate a combination of support strategies, and to dynamically optimize and update the strategy based on real-time feedback data.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the intelligent generation method for low-income population assistance strategies under the background of rural revitalization, as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the intelligent generation method for low-income population assistance strategies under the background of rural revitalization, as described in any one of claims 1 to 7.

Images