Farmer monitoring assistance method and system, storage medium and program product

Abstract

The invention provides a peasant household monitoring assistance method and system, a storage medium and a program product, and the method comprises the steps: obtaining the basic information of a peasant household and the information of an industry department, carrying out the deep analysis, and constructing the structured personal information of the peasant household; calculating and converting the personal information of the peasant households to form portraits taking the households as units, and storing the portraits in a database and regularly updating the portraits; carrying out data preprocessing, data binning and feature engineering by taking historical farmer monitoring data as labels on the basis of portraits taking households as units, obtaining risk coefficients by utilizing an integrated model, determining risk thresholds and setting risk early warning; and carrying out vectorization coding on the peasant household personal information triggering risk early warning by using a natural language processing technology, matching similar historical assistance measures, retrieving and recommending assistance policies, and integrating and outputting the assistance policies. According to the invention, the condition of the farmer is dynamically monitored, and intelligent assistance recommendation is timely and accurately provided.

Description

Technical Field

[0001] This invention relates to the field of data monitoring and early warning technology, and in particular to a method, system, storage medium, and program product for monitoring and assisting farmers. Background Technology

[0002] In recent years, my country has achieved remarkable results in its efforts to assist farmers. To consolidate these achievements, the government has successively issued a series of relevant policy documents, explicitly emphasizing the importance of strengthening dynamic monitoring of farmers and requiring the full implementation of a farmer monitoring and assistance mechanism. In the implementation process, farmer monitoring and assistance are considered two crucial links. However, due to complex factors such as vast geographical areas and significant cultural differences, farmer monitoring work in various regions faces numerous challenges.

[0003] Currently, monitoring efforts in many regions remain at a stage of no monitoring or static monitoring, making it difficult to identify unstable households and those facing sudden and severe difficulties in a timely and accurate manner, thus hindering effective early warning and assistance. The limitation of this static monitoring model is that it cannot comprehensively and dynamically reflect farmers' risks, resulting in assistance measures often lagging behind the actual occurrence of problems.

[0004] Overcoming the limitations of existing monitoring models and achieving dynamic monitoring and intelligent assistance recommendations for farmers has become a pressing technical challenge. Therefore, a new technical solution is urgently needed. This solution utilizes advanced artificial intelligence to build an intelligent data platform, aiming to achieve dynamic monitoring of farmers' conditions and provide timely and accurate intelligent assistance recommendations. This would overcome monitoring difficulties caused by geographical and cultural factors, improve the efficiency and accuracy of assistance efforts, and provide strong technical support for consolidating and expanding the results of assistance programs. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a method, system, storage medium, and program product for monitoring and assisting farmers. It aims to utilize artificial intelligence technology to build an intelligent data platform, enabling dynamic monitoring of farmers' conditions and providing timely and accurate intelligent assistance recommendations. This overcomes monitoring challenges caused by geographical and cultural factors, improves the efficiency and accuracy of assistance work, and provides strong technical support for consolidating and expanding assistance achievements.

[0006] To make the technical description of this patent clearer, the following explanation is provided: The term "inclusion / elimination" in this patent means: inclusion in monitoring / elimination of risk; where "inclusion in monitoring" means entering the scope of risk warning, that is, indicating the existence of risk.

[0007] In a first aspect, the present invention provides a method for monitoring and assisting farmers, comprising the following steps:

[0008] We acquire basic information about farmers and industry sectors, and use natural language processing and data mining techniques to conduct in-depth analysis to construct structured personal information about farmers.

[0009] The personal information of farmers is calculated and converted to form a profile of each household, which is stored in a database and updated regularly.

[0010] Based on household-based profiles, using historical farmer monitoring data as annotations, data preprocessing, data binning, and feature engineering are performed. Using an integrated model, risk coefficients are obtained, risk thresholds are determined, and risk warnings are set.

[0011] For farmers' personal information that triggers risk warnings, natural language processing technology is used to vectorize and encode it, match it with similar historical assistance measures, retrieve and recommend assistance policies, and integrate and output them using a large language model.

[0012] As a further improvement of the present invention, the deep analysis using natural language processing and data mining techniques includes:

[0013] The types of basic information about farmers and industry sectors include text information and image information. Artificial intelligence models are used to extract elements and summarize the text information to obtain key information and overall overview. Artificial intelligence visual algorithms are used to first screen the effectiveness of the image information, and then identify and analyze the image category and content.

[0014] As a further improvement of the present invention, the construction of structured farmer personal information includes:

[0015] The basic information of farmers and industry departments after in-depth analysis are summarized, the cross-referenced information is verified and merged, and an automatically updated workflow is formed to obtain structured personal information of farmers.

[0016] As a further improvement of the present invention, the calculation and conversion of farmers' personal information to form a profile at the household level includes:

[0017] Farmers' personal information includes directly computable data, data requiring processing, and custom indicators;

[0018] For data that can be directly calculated, the per capita value per household is calculated on a household basis to obtain the per capita data for each piece of information.

[0019] Data that requires processing is transformed to obtain categorized data;

[0020] Custom metrics are indicators that are manually customized to uncover deeper information in the data and are tailored to practical business experience. These include indicators such as the average household labor force coefficient and income diversity.

[0021] As a further improvement of the present invention, the average household labor force coefficient is obtained as follows:

[0022] Based on factors such as the farmer's health status, disability level, age, and gender, a comprehensive score is assigned to each farmer's labor capacity coefficient. The conversion method is as follows:

[0023] The initial workforce is 1:

[0024] Health status assessment: Healthy: X*1 = Y; Serious illness: X * Serious illness retention coefficient = Y; Chronic disease: X * Chronic disease retention coefficient = Y;

[0025] Disability level and type determination: Y * disability labor retention coefficient = Z; No disability: Y * 1 = Z;

[0026] Age determination: Z * age group retention coefficient = H;

[0027] Gender: H * Gender retention coefficient = T;

[0028] The final labor force coefficient was set as T;

[0029] The household labor force coefficient T is obtained by combining multiple factors, and the average household labor force coefficient is calculated.

[0030] As a further improvement of the present invention, the income diversity index is obtained by statistically converting various income items of farmers, calculating the proportion of each income item, and statistically analyzing income categories in which a single income item accounts for more than 15% of the total income.

[0031] As a further improvement of the present invention, the process of creating a household-based profile, using historical farmer monitoring data as annotation, performing data preprocessing, data binning, feature engineering, and utilizing an integrated model to obtain a risk coefficient, determine a risk threshold, and set a risk warning includes:

[0032] Acquire historical farmer monitoring data and combine it with household-based profiles to construct a dataset;

[0033] Data cleaning was performed on the dataset, using methods such as box plots, outlier detection, and missing value prediction and imputation to clean and screen the data, resulting in cleaned and screened data.

[0034] The cleaned and screened data is first binned, and then transformed using whether it is included or removed as a label to obtain the transformed data.

[0035] For the transformed data, correlation analysis and stepwise regression algorithm are used to select relevant features and remove features that are irrelevant to the target result.

[0036] Using inclusion / exclusion as a label, supervised learning of the ensemble model is performed using artificial intelligence algorithms to output a risk coefficient;

[0037] Based on the risk coefficient and compared with historical inclusion / elimination results, the risk threshold is determined;

[0038] The inclusion / elimination criteria were used as markers to calculate the risk warning threshold and the risk elimination threshold.

[0039] The calculated results are stored, and information on farmers whose risk levels exceed the risk warning threshold and fall below the risk elimination threshold is distributed to the intelligent assistance module, and the feedback data is stored.

[0040] Based on the feedback data, a scheduled task is assigned to periodically optimize and update the model, iterating repeatedly to improve recognition accuracy.

[0041] As a further improvement of the present invention, the data after cleaning and screening is first subjected to binning processing, including:

[0042] For numerical indicators in the cleaned and screened data, the optimal segmentation node is obtained by using the decision tree method based on the maximum information gain criterion, and the data is transformed from numerical data into segmented interval data.

[0043] If the categorized data in the cleaned and screened data has more than 5 categories, it will be merged manually.

[0044] As a further improvement of the present invention, the conversion of the data after cleaning and screening includes:

[0045] By using WOE transformation, the data is converted into numerical values ​​between 0 and 1, which are of the same dimension. This transformation has a positive selection significance for labeling.

[0046] For each variable feature, the WOE calculation method is as follows:

[0047]

[0048] WOE i If the result is positive, it indicates that the feature interval has a positive impact on the final result.

[0049] As a further improvement of the present invention, the step of using whether to include / eliminate as a label, and performing supervised learning using artificial intelligence algorithms to output a risk coefficient includes:

[0050] The Logistic Regression algorithm, XGBoost algorithm, and LightGBM algorithm were used for training. The primary goal was to comprehensively identify the personnel who needed to be included in the monitoring, and to assign higher weights to the data of the personnel who needed to be included in the monitoring, so that the algorithm could learn.

[0051] The Stacking ensemble learning method integrates the results of various models and outputs them to obtain a risk coefficient for each farmer in the range of 0-1, which represents the level of risk. The closer the coefficient is to 1, the greater the risk.

[0052] The formula for calculating the risk coefficient is as follows:

[0053] risk = w1 * p logistic +w2*p xgboost +w3*p lightgbm ,

[0054] Where w1 + w2 + w3 = 1, 0 ≤ w i ≤1, the parameter is calculated by the ensemble learning algorithm.

[0055] As a further improvement of the present invention, the risk threshold adopts the FPR and TPR indices, and the formula is as follows:

[0056]

[0057] Among them, TP indicates that farmers' forecasts were included in the monitoring and were actually included in the monitoring; FP indicates that farmers' forecasts were included in the monitoring but were actually not included in the monitoring; FN indicates that farmers' forecasts were not included in the monitoring but were actually included in the monitoring; and TN indicates that farmers' forecasts were not included in the monitoring and were actually not included in the monitoring.

[0058] The TPR-FPR result is calculated by modifying the risk threshold. The optimal risk threshold is the one with the largest TPR-FPR difference.

[0059] As a further improvement of the present invention, the personal information of farmers that trigger risk warnings is vectorized and encoded using natural language processing technology, matched with similar historical assistance measures, and recommended assistance policies are retrieved and integrated using a large language model, including:

[0060] By querying the database for profile information, we can find the personal information of farmers who have received assistance and the historical assistance measures in the farmers' areas;

[0061] Extract elements from historical farmers' personal information, farmers' personal information that triggered risk warnings, and historical assistance measures, break them down into individual items, encode them using an Embedding model, input text, output vectors, and store both text and vectors in the database;

[0062] By calculating the cosine similarity between the historical farmer personal information vector and the personal information vector of the farmer that triggered the risk warning, similar historical farmer personal information can be obtained, and corresponding historical assistance measures can be queried for alternatives;

[0063] Recommended assistance measures are obtained by calculating the cosine similarity between the vector of historical assistance measures and the vector of farmers' personal information that triggered risk warnings;

[0064] By using a large language model, historical and recommended assistance measures are integrated, distributed to relevant departments in the farmers' locations, and supplementary suggestions are provided. The verification results and adopted measures are stored and recorded.

[0065] In a second aspect, the present invention provides a computer system for implementing the steps of the method described in the first aspect, including a data aggregation module, a household profile construction module, an intelligent early warning calculation module, and an intelligent assistance module;

[0066] The data aggregation module is used to obtain basic information about farmers and industry sectors, and to conduct in-depth analysis using natural language processing and data mining techniques to construct structured personal information about farmers.

[0067] The household profile construction module is used to calculate and convert farmers' personal information to form a profile of each household, which is stored in the database and updated regularly.

[0068] The intelligent early warning calculation module is used to perform data preprocessing, data binning, and feature engineering based on household-based profiles and historical farmer monitoring data as annotations. It then uses an integrated model to obtain risk coefficients, determine risk thresholds, and set risk warnings.

[0069] The intelligent assistance module is used to vectorize and encode the personal information of farmers that trigger risk warnings using natural language processing technology, match similar historical assistance measures, retrieve and recommend assistance policies, and integrate and output the results using a large language model.

[0070] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in the first aspect.

[0071] Fourthly, the present invention provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the method described in the first aspect.

[0072] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0073] This invention provides a method, system, storage medium, and program product for monitoring and assisting farmers. By utilizing advanced artificial intelligence technology to build an intelligent data platform, it enables dynamic monitoring of farmers' conditions and provides timely and accurate intelligent assistance recommendations. This overcomes the monitoring difficulties caused by factors such as region and culture, improves the efficiency and accuracy of assistance work, and provides strong technical support for consolidating and expanding assistance results. Attached Figure Description

[0074] Figure 1 A flowchart of the farmer monitoring and support method and system provided in the embodiments of the present invention;

[0075] Figure 2 A flowchart illustrating the process of constructing farmers' personal information provided in this embodiment of the invention;

[0076] Figure 3 A flowchart for constructing a user profile provided in an embodiment of the present invention;

[0077] Figure 4 This is a flowchart of the intelligent early warning calculation provided in an embodiment of the present invention;

[0078] Figure 5 A flowchart of intelligent assistance provided for embodiments of the present invention. Detailed Implementation

[0079] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Steps S1, S2… in the described embodiments of the present invention do not limit the unique execution steps of the present invention; the various models described in the present invention are not intended to limit the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0080] In this invention, computer device / equipment / system refers to a related entity applied to a computer, such as hardware, a combination of hardware and software, software, or software in execution. More specifically, for example, software includes, but is not limited to, a process running on a processor, a processor, an object, executable software, an execution thread, a program, and / or a computer. Furthermore, an application program or script running on a server, and the server itself, can also be software. One or more software programs may be in an execution process and / or thread, and the software may be localized on one computer and / or distributed across two or more computers, and may be run on various computer-readable media.

[0081] For example, the system of the present invention can rely on the Django framework for algorithm development, the model can be based on the PyTorch framework, structured data can be uniformly stored in MySQL 5.7 or later, and text feature generation (encoding) of the text feature generation module can be stored in a MongoDB database. Django is an open-source web application framework developed based on Python, using the MTV framework model, which can quickly develop a highly secure and maintainable system platform. Currently, the programming language used for artificial intelligence algorithm models is mainly Python. Image data preprocessing and deep neural network algorithms can also be completed using some Python modules. For example, image preprocessing can be completed using modules such as OpenCV, Pillow, and Scikit-image, while deep neural network algorithms can be quickly constructed using modules such as Tensorflow, PyTorch, and Keras. The information system and algorithm modules involved in this invention can all be implemented using the Python programming language.

[0082] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other.

[0083] Firstly, this invention provides a method for monitoring and supporting farmers, such as... Figure 1 As shown, the specific steps can be as follows:

[0084] S1: Obtain basic information about farmers and industry sectors, and use natural language processing and data mining techniques for in-depth analysis to construct structured personal information about farmers;

[0085] In one embodiment of the present invention, as Figure 1 As shown, structured farmer personal information can be a farmer information summary table.

[0086] Preferably, the structured farmer personal information can be constructed not only based on basic farmer information and industry sector information, but also based on basic farmer information and industry sector information combined with other information other than basic farmer information and industry sector information. For example, other information may include information voluntarily reported by farmers other than basic farmer information and industry sector information.

[0087] Preferably, basic information about farmers, information about industry sectors, and other information can be collected and obtained regularly, and structured personal information about farmers can be kept up-to-date.

[0088] In one embodiment of the present invention, as Figure 2 As shown, for example, the basic information of farmers includes identity information, health and security status, income and expenditure information, and other basic information.

[0089] The information includes: identity information such as name, member ID number, relationship to the head of household, gender, age, political affiliation, and education; health and security information such as whether the individual has major medical insurance, minimum living allowance, pension insurance, commercial supplementary medical insurance, personal accident insurance, employee medical insurance, chronic diseases, and disability status; income and expenditure information such as wage income, operating income / expenditure, property income, large fixed expenditures, and transfer income, with income and expenditure data divided into quarterly and annual data based on update frequency; and other basic information such as housing area, cultivated land area, forest land area, water surface area, area of ​​land converted from farmland to forest, whether there are drinking water safety issues, whether electricity is safe, whether there is housing security, and whether there is access to radio and television.

[0090] In one embodiment of the present invention, as Figure 2 As shown, for example, the industry sector information includes information from the Medical Security Bureau, the Health Commission, the Civil Affairs Bureau, the Provincial Federation of Disabled Persons, the Public Security Bureau, the Provincial Emergency Management Department, and the Provincial Education Department.

[0091] The information includes: information from the Medical Security Bureau and the Health Commission, such as medical insurance consumption information, disease treatment cost information, lists of newly added people with serious and chronic diseases, reimbursement and enrollment information; information from the Civil Affairs Bureau, such as employment monitoring information, student and agricultural assistance subsidies, employment information for returning laborers, family member changes, and assistance policies; information from the Provincial Disabled Persons' Federation, such as information on newly added and existing disabled persons; information from the Public Security Department, such as vehicle information; information from the Provincial Emergency Management Department, such as disaster emergency information; and information from the Provincial Education Department, such as dropout and withdrawal information.

[0092] Preferably, the deep analysis using natural language processing and data mining techniques includes:

[0093] like Figure 2 As shown, the types of information include text information and image information;

[0094] Element extraction or text content understanding of text information: Using artificial intelligence models to extract elements, summarize or understand the content of the text information to obtain key information and overall overview of the text;

[0095] Feature engineering or image content understanding of image information: First, the image information is screened for validity using artificial intelligence visual algorithms, and then the image category and content are identified, understood and analyzed.

[0096] In one embodiment of the present invention, the information voluntarily reported by farmers is, by way of example, as follows:

[0097] "A household in Village A, consisting of 5 people, has the following family information:"

[0098] Wang Yi, head of household, 55 years old, suffers from chronic illness, stroke, hospitalization, subsequent medical expenses of 6,000 yuan, is not the main breadwinner of the family, has an impact on family income, has no labor capacity, and is not a student.

[0099] Wang Er's son, 30 years old, is suffering from a serious illness, a brain tumor, and is hospitalized. The subsequent medical expenses are 80,000 yuan. He is the main breadwinner of the family, which affects the family income. He is not a student and has no other means of working.

[0100] Li San's daughter-in-law, 27 years old, healthy, ordinary labor force, not working;

[0101] Wang Si's grandson, 6 years old, healthy, unable to work, attending the First Nursery School in District B, preschool education, tuition fee 8000 yuan;

[0102] Wang Wu's granddaughter, 11 years old, healthy, unable to work, attends Bing Elementary School, tuition fee 4000 yuan; the whole family participates in urban and rural residents' medical insurance (5 people), owns their own house, and there are no problems with drinking water in the house;

[0103] First, the data is extracted using a large language model. After the text data is processed by the large language model, appropriate prompts such as "Help me extract the basic information of each family member in the following text and classify the elements according to medical care, education, income and expenditure, and family situation" are used to obtain output such as "Wang Er, son of the head of household, has a brain tumor with medical expenses of 80,000 yuan, which belongs to medical expenses".

[0104] In another embodiment of the present invention, exemplarily, the information voluntarily reported by farmers is as follows:

[0105] "Wang Liu reported that the Wuhe River in Ding Village flooded, causing his house to collapse and his water supply to be cut off, making life difficult." Using an artificial intelligence model, element extraction and summarization were performed on this text to obtain key information and an overall overview, concluding that "Wang Liu's household faces a hidden danger to their housing and drinking water supply."

[0106] Preferably, the construction of structured farmer personal information includes:

[0107] The basic information of farmers and industry departments after in-depth analysis are summarized, the cross-referenced information is verified and merged, and an automatically updated workflow is formed to obtain structured personal information of farmers.

[0108] S2: Calculate and convert farmers' personal information to form a profile of each household, store it in a database, and update it regularly;

[0109] In one embodiment of the present invention, as Figure 1 As shown, a household-based portrait can be a household portrait table.

[0110] Preferably, the step of calculating and converting farmers' personal information to form a household-based profile includes:

[0111] Farmers' personal information, such as Figure 3 As shown, it includes directly calculable data, data requiring processing, and custom indicators;

[0112] For data that can be directly calculated, the per capita value per household is calculated on a household basis to obtain the per capita data for each piece of information.

[0113] Data that requires processing is transformed to obtain categorized data;

[0114] Custom metrics are indicators that are manually customized to uncover deeper information in the data and are tailored to practical business experience. These include indicators such as the average household labor force coefficient and income diversity.

[0115] In one embodiment of the present invention, as Figure 3 As shown, for example, directly calculable data may include medical expenses, income, living area, loan amount, etc.; per capita data for each type of information includes per capita medical expenses, per capita income, per capita living area, per capita loan amount, etc.

[0116] The method for obtaining the per capita medical expenditure of a household is as follows: based on the information from the Medical Security Bureau and the Health Commission obtained in step S1, the medical expenditure data of household members in each period are summed to obtain the average medical expenditure of family members. The calculation formula is: per capita medical expenditure of household in each period = total medical expenditure of household in each period / number of family members.

[0117] The method for obtaining per capita household income is as follows: based on the income and expenditure property information collected in step S1, the corresponding per capita household income is calculated, and the data period includes quarters and years.

[0118] The formula for calculating the per capita living area per household is: total living area per household / number of people per household.

[0119] In one embodiment of the present invention, as Figure 3 As shown, for example, the data to be processed may include vehicle information pushed by the public security bureau, farmers' serious illnesses, chronic diseases, farmers' education level or school status, etc.

[0120] Among them, the vehicle information pushed by the public security is converted into farmers' fixed assets through corresponding value conversion, and converted through vehicle coefficients. For example, the vehicle coefficient can be 0.3 for ordinary two-wheeled motorcycles, 1 for small cars, 3.5 for medium-sized ordinary trucks, etc., totaling 62 categories.

[0121] The farmers' major and chronic diseases are converted into the farmers' health status categories, including: healthy, long-term chronic disease, suffering from major disease, disabled, and disabled and suffering from major disease;

[0122] Similarly, farmers' educational background or school status are converted into corresponding categorized data.

[0123] Preferably, the average household labor force coefficient is obtained in the following way:

[0124] Based on factors such as the farmer's health status, disability level, age, and gender, a comprehensive score is assigned to each farmer's labor capacity coefficient. The conversion method is as follows:

[0125] The initial workforce is 1:

[0126] Health status assessment: Healthy: X*1 = Y; Serious illness: X * Serious illness retention coefficient = Y; Chronic disease: X * Chronic disease retention coefficient = Y;

[0127] Disability level and type determination: Y * disability labor retention coefficient = Z; No disability: Y * 1 = Z;

[0128] Age determination: Z * age group retention coefficient = H;

[0129] Gender: H * Gender retention coefficient = T;

[0130] The final labor force coefficient was set as T;

[0131] The household labor force coefficient T is obtained by combining multiple factors, and the average household labor force coefficient is calculated.

[0132] In one embodiment of the present invention, the conversion method for the average household labor force coefficient is exemplarily as follows:

[0133] The initial workforce is 1:

[0134] a. Assessment of health status:

[0135] Health: X*1 = Y; Serious illness: X*0.7 = Y; Chronic illness: X*0.8 = Y;

[0136] b. Determination of disability level and type:

[0137] Y*Disabled labor force retention coefficient = Z; No disability: Y*1 = Z;

[0138] c. Age determination:

[0139] 16-59: Z*1=H; 60-70: Z*0.8=H; 70-80: Z*0.6=H; above 80: Z*0.3=H;

[0140] d. Gender:

[0141] Male: H*1 = T; Female: H*0.9 = T;

[0142] The final labor force coefficient was set as T;

[0143] The household labor force coefficient T is obtained by combining multiple factors, and the average household labor force coefficient is calculated.

[0144] The retention coefficients for disabled laborers are shown in Table 1: Level 1 visual impairment 0.2, Level 2 visual impairment 0.25, Level 3 visual impairment 0.3, Level 4 visual impairment 0.35; Level 1 hearing impairment 0.5, Level 2 hearing impairment 0.55, Level 3 hearing impairment 0.6, Level 4 hearing impairment 0.65; Level 1 speech impairment 0.6, Level 2 speech impairment 0.65, Level 3 speech impairment 0.7, Level 4 speech impairment 0.75; Level 1 physical disability 0.5, Level 2 physical disability 0.55, Level 3 physical disability 0.6, Level 4 physical disability 0.65; Level 1 intellectual disability 0.1, Level 2 intellectual disability 0.15, Level 3 intellectual disability 0.2, Level 4 intellectual disability 0.25; Level 1 multiple disability 0, Level 2 multiple disability 0.05, Level 3 multiple disability 0.2, Level 4 multiple disability 0.25.

[0145] Table 1:

[0146]

[0147] Preferably, the income diversity index is obtained by statistically converting various income items of farmers, calculating the proportion of each income item, and statistically analyzing income categories in which a single income item accounts for more than 15% of the total income.

[0148] Among them, an indicator of the proportion of transfer income is constructed to characterize the subsidies received by farmers.

[0149] S3: Based on household-based profiles, using historical farmer monitoring data as annotations, data preprocessing, data binning, and feature engineering are performed. Using an integrated model, risk coefficients are obtained, risk thresholds are determined, and risk warnings are set.

[0150] Preferably, such as Figure 4 As shown, the process involves creating household-based profiles, using historical farmer monitoring data as annotations, performing data preprocessing, data binning, feature engineering, and utilizing an ensemble model to obtain risk coefficients, determine risk thresholds, and set risk warnings. This includes:

[0151] S31: Obtain historical farmer monitoring data, combine it with household-based profiles, and construct a dataset;

[0152] In one embodiment of the present invention, for example, historical farmer monitoring data includes farmer application records, verification by grassroots personnel, final inclusion / exclusion decisions, etc., combined with farmer profile information obtained from the household profile, to obtain a jointly constructed training-test dataset, with approximately 800,000 pre-training data entries.

[0153] S32: Perform data cleaning on the dataset, using methods such as box plots, outlier detection, and missing value prediction and imputation to clean and screen the data, and obtain the cleaned and screened data.

[0154] S33: The cleaned and screened data is first binned, and then the data is transformed using whether it is included or removed as a label to obtain the transformed data for modeling.

[0155] Among them, data binning is a process of discretizing the indicator features using decision tree binning. This operation can enhance robustness, improve the model's resistance to data anomalies, and enhance the model's robustness.

[0156] Preferably, the data after cleaning and screening is first binned, including:

[0157] For numerical indicators in the cleaned and screened data, the optimal segmentation node is obtained by using the decision tree method based on the maximum information gain criterion, and the data is transformed from numerical data into segmented interval data.

[0158] If the categorized data in the cleaned and screened data has more than 5 categories, it will be merged manually.

[0159] In one embodiment of the present invention, numerical indicators such as income, labor force, and loan amount are transformed from numerical data into segmented interval data by using a decision tree method based on the maximum information gain criterion to obtain the optimal segmentation node of the data; categorical data such as disability level and number of family members are processed manually to merge data if there are more than 5 categories.

[0160] Preferably, the conversion of the data after cleaning and screening includes:

[0161] By using WOE transformation, the data is converted into numerical values ​​between 0 and 1, which are of the same dimension. This transformation has a positive selection significance for labeling.

[0162] For each variable feature, the WOE calculation method is as follows:

[0163]

[0164] WOE i If the result is positive, it indicates that the feature interval has a positive impact on the final result.

[0165] S34: For the transformed data, through correlation analysis and stepwise regression algorithm, relevant features are selected and features that are irrelevant to the target result are removed;

[0166] The importance of variables output by the model can also be used to comprehensively consider the selection of features.

[0167] S35: Using inclusion / exclusion as a label, supervised learning of the ensemble model is performed using artificial intelligence algorithms to output a risk coefficient;

[0168] Preferably, the step of using inclusion / exclusion as a label, and employing artificial intelligence algorithms for supervised learning to output a risk coefficient, includes:

[0169] To increase model stability and interpretability, Logistic regression, XGboost, and LightGBM algorithms were used for training. Since the primary goal of this system is to "include all those who should be included," that is, to identify as many people as possible who need to be included in the monitoring, higher weights were assigned to the data of those who need to be included in the monitoring, allowing the algorithm to learn.

[0170] The Stacking ensemble learning method integrates the results of various models and outputs them to obtain a risk coefficient for each farmer in the range of 0-1, which represents the level of risk. The closer the coefficient is to 1, the greater the risk.

[0171] The formula for calculating the risk coefficient is as follows:

[0172] risk = w1 * p logistic +w2*p xgboost +w3*p lightgbm ,

[0173] Where w1 + w2 + w3 = 1, 0 ≤ w i ≤1, the parameter is calculated by the ensemble learning algorithm.

[0174] S36: Determine the risk threshold based on the risk coefficient and by comparing it with historical inclusion / elimination results;

[0175] Preferably, the risk threshold uses FPR and TPR indicators, and the formula is:

[0176]

[0177] Among them, TP indicates that the farmer's prediction was included in the monitoring and was actually included in the monitoring; FP indicates that the farmer's prediction was included in the monitoring but was actually not included in the monitoring; FN indicates that the farmer's prediction was not included in the monitoring but was actually included in the monitoring; and TN indicates that the farmer's prediction was not included in the monitoring and was actually not included in the monitoring, as shown in Table 2.

[0178] Table 2:

[0179] The actual data was included in the monitoring. The actual cases were not included in the testing. Predictions were incorporated into monitoring. TP FP The forecast was not included in the monitoring. FN TN

[0180] The farmer's risk coefficient is a value between 0 and 1, and the risk threshold is p. i That is, the classification result is determined by adjusting the risk threshold p. i Calculating TPR i -FPR i The result is that p is the largest when the difference is maximized. i This can be considered the optimal risk threshold, at which the relative accuracy of the classification is highest.

[0181] S37: Combining expert experience and the requirements of actual business scenarios, the risk coefficient of farmers is divided into rules, and corresponding threshold assessments are conducted on whether to include them in monitoring and whether to eliminate risks, so as to obtain risk warning thresholds and risk elimination thresholds respectively.

[0182] It should be noted that the risk warning threshold and risk elimination threshold obtained from the threshold assessment in step S37 are dynamically adjusted based on national and local policies and expert experience in actual business scenarios. The risk threshold in step S36 is determined based on technology and history, and this risk threshold provides a reference for the threshold assessment in step S37.

[0183] In one embodiment of the present invention, for example, the obtained risk warning threshold is 0.7 and the risk elimination threshold is 0.2.

[0184] S38: Store the calculated results, distribute information on farmers whose risk warning threshold is greater than the risk elimination threshold and whose risk elimination threshold is less than the risk warning threshold, enter the intelligent assistance module, and store the feedback data;

[0185] like Figure 1 As shown, information on farmers whose risk warning threshold is greater than the threshold indicates that they need to be identified and assisted, while information on farmers whose risk elimination threshold is less than the threshold indicates that they need to have their risks eliminated.

[0186] S39: Based on the feedback data, specify a scheduled task to periodically optimize and update the model, and iterate repeatedly to improve recognition accuracy.

[0187] S4: For farmers' personal information that triggers risk warnings, use natural language processing technology to vectorize and encode it, match similar historical assistance measures, retrieve and recommend assistance policies, and integrate and output them using a large language model.

[0188] Preferably, the personal information of farmers that trigger risk warnings is vectorized using natural language processing technology, matched with similar historical assistance measures, and recommended assistance policies are retrieved. The results are then integrated and output using a large language model, including:

[0189] S41: Retrieve profile information from the database to find the personal information and historical assistance measures of farmers who have received assistance in the historical data of the farmers' areas;

[0190] S42: Extract elements from historical farmers' personal information, farmers' personal information that triggered risk warnings, and historical assistance measures, split them into items, and encode them through the Embedding model. Input text, output vector, and store the text and vector together in the database.

[0191] In one embodiment of the present invention, for example, historical farmers' personal information, farmers' personal information that triggered risk warnings, and historical assistance measures are extracted as elements, split into individual items, and encoded using a BGE serialization model. The input length is approximately 512 tokens, which is sufficient for text encoding tasks of most events; if it exceeds this length, it is truncated. The output is a 768-dimensional vector, and both the text and the vector are stored in the database.

[0192] It should be noted that, in addition to using the bge series model described in the embodiments of this invention, other deep learning models, such as the embedding function of Chinese models like m3e and simcse, can also be used to encode historical farmers' personal information, farmers' personal information that triggers risk warnings, and historical assistance measures.

[0193] S43: By calculating the cosine similarity between the historical farmer personal information vector and the farmer personal information vector that triggered the risk warning, similar historical farmer personal information is obtained, and corresponding historical assistance measures are queried for alternatives;

[0194] Cosine similarity assesses the similarity between two non-zero vectors by measuring the cosine of the angle between them. Its value ranges from -1 to 1, where 1 indicates that they are in exactly the same direction, 0 indicates that they are orthogonal, and -1 indicates that they are in completely opposite directions. In text retrieval, cosine similarity is often used to assess the similarity between text vectors.

[0195] Preferably, the top 5 historical farmers' personal information data with the cosine similarity between the historical farmers' personal information vector and the farmers' personal information vector that triggered the risk warning can be retrieved, and the corresponding historical assistance measures can be found.

[0196] S44: Recommended assistance measures are obtained by calculating the cosine similarity between the historical assistance measure vector and the personal information vector of farmers that triggered the risk warning;

[0197] Preferably, the top 5 most similarity-ranked assistance measures can be retrieved as recommended assistance measures by calculating the cosine similarity between the vector of historical assistance measures and the vector of farmers' personal information that triggered the risk warning.

[0198] S45: Using a large language model, integrate the historical assistance measures described in step S43 and the recommended assistance measures described in step S44, distribute them to relevant departments in the farmers' locations, provide auxiliary suggestions, and store and record the verification results and adopted measures.

[0199] In one embodiment of the present invention, as Figure 5 As shown, historical assistance information refers to the personal information of farmers in history, early warning information refers to the personal information of farmers who triggered risk warnings, and assistance policies refer to historical assistance measures. The three are vectorized to obtain historical farmer information codes, early warning farmer information codes, and assistance policy codes, respectively. Among them, the historical farmer information code is a vector of historical farmer personal information, the early warning farmer information code is a vector of farmer personal information that triggered risk warnings, and the assistance policy code is a vector of historical assistance measures.

[0200] In one embodiment of the present invention, exemplarily, the farmer is Wang Qi, a general farmer with only one family member and an annual net income of about 7,000 yuan. Due to surgery for cancer, he triggered a medical insurance warning, of which he had to pay 29,134 yuan out of pocket, leading to financial hardship.

[0201] After the industry department pushed the data, the intelligent early warning system calculated its risk coefficient to be as high as 0.87, exceeding the assessment threshold. Therefore, it was assessed as requiring a risk warning and dispatched to the intelligent assistance module. The intelligent assistance module retrieved information based on the farmer's profile and, with the assistance measures knowledge base, provided suggested assistance measures.

[0202] 1. This household is experiencing financial hardship due to cancer; it is recommended that they apply to the village committee for minimum living allowance.

[0203] 2. It is recommended that reimbursement be made in a timely manner in accordance with medical insurance policies.

[0204] In a second aspect, this invention provides a computer system for implementing the steps of the method described in the first aspect, including a data aggregation module, a household profile construction module, an intelligent early warning calculation module, and an intelligent assistance module.

[0205] The data aggregation module is used to obtain basic information about farmers and industry sectors, and to conduct in-depth analysis using natural language processing and data mining techniques to construct structured personal information about farmers.

[0206] The household profile construction module is used to calculate and convert farmers' personal information to form a profile of each household, which is stored in the database and updated regularly.

[0207] The intelligent early warning calculation module is used to perform data preprocessing, data binning, and feature engineering based on household-based profiles and historical farmer monitoring data as annotations. It then uses an integrated model to obtain risk coefficients, determine risk thresholds, and set risk warnings.

[0208] The intelligent assistance module is used to vectorize and encode the personal information of farmers that trigger risk warnings using natural language processing technology, match similar historical assistance measures, retrieve and recommend assistance policies, and integrate and output the results using a large language model.

[0209] Thirdly, embodiments of the present invention provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in the first aspect.

[0210] Fourthly, embodiments of the present invention provide a computer program product, including a computer program that, when executed by a processor, implements the steps of the method described in the first aspect.

[0211] This invention provides a method, system, storage medium, and program product for monitoring and assisting farmers. By utilizing advanced artificial intelligence technology to build an intelligent data platform, it enables dynamic monitoring of farmers' conditions and provides timely and accurate intelligent assistance recommendations. This overcomes the monitoring difficulties caused by factors such as region and culture, improves the efficiency and accuracy of assistance work, and provides strong technical support for consolidating and expanding assistance results.

Claims

1. A method for monitoring and supporting farmers, characterized in that, Includes the following steps: We acquire basic information about farmers and industry sectors, and use natural language processing and data mining techniques to conduct in-depth analysis to construct structured personal information about farmers. The personal information of farmers is calculated and converted to form a profile of each household, which is stored in a database and updated regularly. Based on household-based profiles, using historical farmer monitoring data as annotations, data preprocessing, data binning, and feature engineering are performed. Using an integrated model, risk coefficients are obtained, risk thresholds are determined, and risk warnings are set. For farmers' personal information that triggers risk warnings, natural language processing technology is used to vectorize and encode it, match it with similar historical assistance measures, retrieve and recommend assistance policies, and integrate and output them using a large language model.

2. The farmer monitoring and support method according to claim 1, characterized in that, The deep analysis employing natural language processing and data mining techniques includes: The types of basic information about farmers and industry sectors include text information and image information. Artificial intelligence models are used to extract elements and summarize the text information to obtain key information and overall overview. Artificial intelligence visual algorithms are used to first screen the effectiveness of the image information, and then identify and analyze the image category and content.

3. The farmer monitoring and support method according to claim 1, characterized in that, The structured personal information of farmers includes: The basic information of farmers and industry departments after in-depth analysis are summarized, the cross-referenced information is verified and merged, and an automatically updated workflow is formed to obtain structured personal information of farmers.

4. The farmer monitoring and support method according to claim 1, characterized in that, The process of calculating and converting farmers' personal information to create a household-based profile includes: Farmers' personal information includes directly computable data, data requiring processing, and custom indicators; For data that can be directly calculated, the per capita value per household is calculated on a household basis to obtain the per capita data for each piece of information. Data that requires processing is transformed to obtain categorized data; Custom metrics are indicators that are manually customized to uncover deeper information in the data and are tailored to practical business experience. These include indicators such as the average household labor force coefficient and income diversity.

5. The farmer monitoring and support method according to claim 4, characterized in that, The method for obtaining the average household labor force coefficient is as follows: Based on factors such as the farmer's health status, disability level, age, and gender, a comprehensive score is assigned to each farmer's labor capacity coefficient. The conversion method is as follows: The initial workforce is 1: Health status assessment: Healthy: X*1 = Y; Serious illness: X * Serious illness retention coefficient = Y; Chronic disease: X * Chronic disease retention coefficient = Y; Disability level and type determination: Y * disability labor retention coefficient = Z; No disability: Y * 1 = Z; Age determination: Z * age group retention coefficient = H; Gender: H * Gender retention coefficient = T; The final labor force coefficient was set as T; The household labor force coefficient T is obtained by combining multiple factors, and the average household labor force coefficient is calculated.

6. The farmer monitoring and support method according to claim 4, characterized in that, The income diversity index is calculated by statistically converting various income items of farmers, determining the proportion of each income item, and identifying income categories where a single income item accounts for more than 15% of the total income.

7. The farmer monitoring and support method according to claim 1, characterized in that, The process, based on household-level profiling and using historical farmer monitoring data as annotations, involves data preprocessing, binning, and feature engineering. An ensemble model is then used to obtain risk coefficients, determine risk thresholds, and set risk warnings. This includes: Acquire historical farmer monitoring data and combine it with household-based profiles to construct a dataset; Data cleaning was performed on the dataset, using methods such as box plots, outlier detection, and missing value prediction and imputation to clean and screen the data, resulting in cleaned and screened data. The cleaned and screened data is first binned, and then transformed using whether it is included or removed as a label to obtain the transformed data. For the transformed data, correlation analysis and stepwise regression algorithm are used to select relevant features and remove features that are irrelevant to the target result. Using inclusion / exclusion as a label, supervised learning of the ensemble model is performed using artificial intelligence algorithms to output a risk coefficient; Based on the risk coefficient and compared with historical inclusion / elimination results, the risk threshold is determined; Based on expert experience and the requirements of actual business scenarios, the risk coefficient of farmers is divided into rules, and corresponding threshold assessments are conducted for whether to include them in monitoring and whether to eliminate risks, resulting in risk warning thresholds and risk elimination thresholds. The threshold assessment results are stored, and information on farmers whose thresholds are greater than the risk warning threshold and less than the risk elimination threshold is distributed to the intelligent assistance module, and the feedback data is stored. Based on the feedback data, a scheduled task is assigned to periodically optimize and update the model, iterating repeatedly to improve recognition accuracy.

8. The farmer monitoring and support method according to claim 7, characterized in that, The data after cleaning and screening is first binned, including: For numerical indicators in the cleaned and screened data, the optimal segmentation node is obtained by using the decision tree method based on the maximum information gain criterion, and the data is transformed from numerical data into segmented interval data. If the categorized data in the cleaned and screened data has more than 5 categories, it will be merged manually.

9. The farmer monitoring and support method according to claim 7, characterized in that, The transformation of the data after cleaning and screening includes: By using WOE transformation, the data is converted into numerical values ​​between 0 and 1, which are of the same dimension. This transformation has a positive selection significance for labeling. For each variable feature, the WOE calculation method is as follows: WOE i If the result is positive, it indicates that the feature interval has a positive impact on the final result.

10. The farmer monitoring and support method according to claim 7, characterized in that, The method of using inclusion / exclusion as a label, employing supervised learning with artificial intelligence algorithms, and outputting a risk coefficient includes: The Logistic Regression algorithm, XGBoost algorithm, and LightGBM algorithm were used for training. The primary goal was to comprehensively identify the personnel who needed to be included in the monitoring, and to assign higher weights to the data of the personnel who needed to be included in the monitoring, so that the algorithm could learn. The Stacking ensemble learning method integrates the results of various models and outputs them to obtain a risk coefficient for each farmer in the range of 0-1, which represents the level of risk. The closer the coefficient is to 1, the greater the risk. The formula for calculating the risk coefficient is as follows: risk=w1*p logistic +w2*p xgboost +w3*p lightgbm , Where w1 + w2 + w3 = 1, 0 ≤ w i ≤1, the parameter is calculated by the ensemble learning algorithm.

11. The farmer monitoring and support method according to claim 7, characterized in that, The risk threshold uses FPR and TPR indicators, and the formula is as follows: Among them, TP indicates that farmers' forecasts were included in the monitoring and were actually included in the monitoring; FP indicates that farmers' forecasts were included in the monitoring but were actually not included in the monitoring; FN indicates that farmers' forecasts were not included in the monitoring but were actually included in the monitoring; and TN indicates that farmers' forecasts were not included in the monitoring and were actually not included in the monitoring. The TPR-FPR result is calculated by modifying the risk threshold. The optimal risk threshold is the one with the largest TPR-FPR difference.

12. The farmer monitoring and support method according to claim 1, characterized in that, The personal information of farmers triggering risk warnings is vectorized using natural language processing technology, matched with similar historical assistance measures, and recommended assistance policies are retrieved. The results are then integrated and output using a large language model, including: By querying the database for profile information, we can find the personal information of farmers who have received assistance and the historical assistance measures in the farmers' areas; Extract elements from historical farmers' personal information, farmers' personal information that triggered risk warnings, and historical assistance measures, break them down into individual items, encode them using an Embedding model, input text, output vectors, and store both text and vectors in the database; By calculating the cosine similarity between the historical farmer personal information vector and the personal information vector of the farmer that triggered the risk warning, similar historical farmer personal information can be obtained, and corresponding historical assistance measures can be queried for alternatives; Recommended assistance measures are obtained by calculating the cosine similarity between the vector of historical assistance measures and the vector of farmers' personal information that triggered risk warnings; By using a large language model, historical and recommended assistance measures are integrated, distributed to relevant departments in the farmers' locations, and supplementary suggestions are provided. The verification results and adopted measures are stored and recorded.

13. A computer system, characterized in that, The computer system is used to implement the steps of the method according to any one of claims 1-12, including a data aggregation module, a household profile construction module, an intelligent early warning calculation module, and an intelligent assistance module; The data aggregation module is used to obtain basic information about farmers and industry sectors, and to conduct in-depth analysis using natural language processing and data mining techniques to construct structured personal information about farmers. The household profile construction module is used to calculate and convert farmers' personal information to form a profile of each household, which is stored in the database and updated regularly. The intelligent early warning calculation module is used to perform data preprocessing, data binning, and feature engineering based on household-based profiles and historical farmer monitoring data as annotations. It then uses an integrated model to obtain risk coefficients, determine risk thresholds, and set risk warnings. The intelligent assistance module is used to vectorize and encode the personal information of farmers that trigger risk warnings using natural language processing technology, match similar historical assistance measures, retrieve and recommend assistance policies, and integrate and output the results using a large language model.

14. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the steps of the method described in any one of claims 1-12.

15. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the steps of the method described in any one of claims 1-12.

Images