Rural credit credit investigation system based on multi-source heterogeneous data fusion

Abstract

This invention relates to the field of agricultural finance technology, specifically to a rural credit reporting system based on multi-source heterogeneous data fusion. The system includes a data acquisition module, a yield prediction module, a price prediction module, an income distribution generation module, a pressure coefficient calculation module, a credit decision module, a repayment plan generation module, and a post-loan monitoring module. The system acquires multi-source data such as soil moisture, weather forecasts, and historical prices, generates probability distributions of yield and price, convolves these to obtain the total income probability distribution, and extracts quantiles. Based on this, it calculates a price cycle pressure coefficient. This coefficient is compared with a preset threshold, and, considering the bottom income support situation, outputs tiered credit instructions, matching flexible repayment plans such as normal repayment, income-ratio repayment, or price threshold-triggered deferral. Post-loan, the repayment amount is dynamically adjusted based on real-time sales revenue and market prices, and a price stabilization account is used to achieve cross-cycle risk hedging. This invention improves the accuracy and adaptability of rural credit risk assessment.

Description

Technical Field

[0001] This invention relates to the field of agricultural finance technology, specifically to a rural credit reporting system based on the fusion of multi-source heterogeneous data. Background Technology

[0002] Rural credit reporting systems are a crucial infrastructure for serving agricultural production and management entities. Early rural credit relied primarily on manual investigations and joint guarantee systems. In recent years, with the development of big data technology, multi-source data such as land title confirmation, agricultural insurance, and meteorological remote sensing have been gradually introduced to achieve online credit assessment. The current trend is to integrate agricultural production data across the entire chain with financial data to explore flexible credit granting mechanisms based on expected returns.

[0003] In existing technologies, some agricultural credit products have taken into account the matching problem between agricultural production cycles and repayment plans, setting the repayment window after the harvest season and adopting a "revenue-based expenditure" model to avoid farmers facing repayment pressure during the growing season. This model is a significant improvement over the fixed monthly payment model of urban commercial loans and reduces the risk of non-malicious default caused by cash flow mismatch.

[0004] However, the existing technologies still have the following shortcomings: For farmers growing agricultural products with volatile prices, such as lychees, garlic, and ginger, their income depends not only on yield but also heavily on market prices at the time of harvest. In the alternating years of high and low yields, bumper harvests may result in increased production but not increased income due to price crashes, while poor harvests may lead to a situation where there is demand but no supply. Current credit reporting systems only set repayment windows based on fixed harvest times, failing to differentiate between income differences in bumper, average, and poor harvest years, and even less able to automatically adjust repayment plans during years of low prices. When farmers encounter price troughs, the system still collects principal and interest according to the original schedule, forcing them to sell at a loss or borrow new money to repay old debts. This repayment mechanism is severely mismatched with the price fluctuation patterns of agricultural products, a major reason for the current high default rate in rural credit. Therefore, there is an urgent need for a system that can incorporate price fluctuation cycles into the credit assessment framework and dynamically match credit granting and repayment plans. Summary of the Invention

[0005] In view of the above-mentioned shortcomings of the existing technology, the present invention provides a rural credit investigation system based on multi-source heterogeneous data fusion, which can effectively solve the problems mentioned in the existing technology.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] This invention provides a rural credit reporting system based on multi-source heterogeneous data fusion, comprising:

[0008] The data acquisition module is used to acquire real-time soil moisture data of the land operated by the target farmers, weather forecast data for the future predetermined period, historical market price time series of the target agricultural products, planting area and expected market time declared by the target farmers, annual rigid expenditure data of the target farmers, and total principal and interest of the target farmers' loans.

[0009] The yield prediction module is used to generate a probability distribution of the target farmer's seasonal yield based on the real-time soil moisture data, the weather forecast data, and the planting area. The probability distribution of the yield includes the expected yield per acre and the standard deviation of the yield.

[0010] The price prediction module is used to generate the price probability distribution of the target agricultural product at the expected listing time based on the historical market price time series using a mixed-frequency volatility model.

[0011] The income distribution generation module is used to perform a convolution operation on the probability distribution of the output and the probability distribution of the price to generate the probability distribution of the total income of the target farmer in the current season, and to extract a first predetermined number income, a second predetermined number income and a third predetermined number income from the total income probability distribution, wherein the first predetermined number income is lower than the second predetermined number income and the second predetermined number income is lower than the third predetermined number income.

[0012] The pressure coefficient calculation module is used to calculate the price cycle pressure coefficient by dividing the difference between the second predetermined income and the annual rigid expenditure data by the total loan principal and interest.

[0013] The credit decision module is used to compare the price cycle pressure coefficient with a first preset threshold and a second preset threshold, wherein the first preset threshold is greater than the second preset threshold; when the price cycle pressure coefficient is greater than or equal to the first preset threshold, a full credit instruction is output; when the price cycle pressure coefficient is between the second preset threshold and the first preset threshold, based on the comparison result of the first predetermined income and the total principal and interest of the loan, a reduced credit instruction or a full credit instruction with a flexible repayment indicator is output; when the price cycle pressure coefficient is less than the second preset threshold, a restricted credit instruction with a risk hedging indicator is output.

[0014] The repayment plan generation module is used to generate corresponding repayment plan parameters in response to different instructions output by the credit decision module. The repayment plan parameters include repayment mode identifier, repayment ratio, and price trigger threshold.

[0015] The post-loan monitoring module is used to continuously acquire real-time agricultural product sales revenue data and real-time market price data of the target farmers after the loan is issued, dynamically adjust the amount due for the current period according to the repayment plan parameters, and automatically trigger the principal deferral operation when the real-time market price data is lower than the price trigger threshold for a predetermined period of time.

[0016] Furthermore, the steps performed by the production prediction module include:

[0017] The historical average yield per mu of the target crop is obtained as the prior expected yield per mu, and the corresponding prior standard deviation is obtained.

[0018] The deviation of surface water content is calculated based on the real-time soil moisture data, and the deviation of accumulated temperature and the number of extreme weather warning days are calculated based on the meteorological forecast data.

[0019] The yield correction factor is calculated based on the deviation of surface moisture content, the deviation of accumulated temperature, and the number of days of extreme weather warning.

[0020] Multiply the prior expected yield per acre by the yield correction factor to obtain the posterior expected yield per acre;

[0021] The posterior standard deviation is obtained by multiplying the prior standard deviation by the absolute value of the difference between the yield correction factor and 1, plus 1.

[0022] A probability distribution of the yield in the form of a normal distribution is constructed using the posterior expected yield per acre and the posterior standard deviation.

[0023] Furthermore, the steps performed by the price prediction module include:

[0024] Obtain the historical price time series of high-quality fruit and ordinary fruit of the target agricultural product;

[0025] The probability distribution of high-quality fruit prices and the probability distribution of ordinary fruit prices were predicted using the mixed-frequency GARCH-MIDAS model, respectively.

[0026] Obtain the historical management level score of the target farmer, and determine the weight of high-quality fruit based on the historical management level score;

[0027] The price probability distribution is obtained by multiplying the price probability distribution of high-quality fruit by the weight of high-quality fruit and the price probability distribution of ordinary fruit by one minus the weight of high-quality fruit.

[0028] Furthermore, the steps performed by the income distribution generation module include:

[0029] Obtain the correlation coefficient between the historical yield and historical price of the target crop;

[0030] The Monte Carlo simulation method is used to extract a preset number of sample pairs from the probability distribution of output and the probability distribution of price. Each time a sample is extracted, a negative correlation between output and price is introduced based on the correlation coefficient.

[0031] Multiply the yield sample extracted each time by the planting area, and then multiply by the price sample to obtain the total income for a single simulation.

[0032] Arrange the simulated total revenue of the preset number of times in ascending order, take the 10th percentile as the first predetermined percentile revenue, take the 50th percentile as the second predetermined percentile revenue, and take the 90th percentile as the third predetermined percentile revenue.

[0033] Furthermore, the steps performed by the credit decision module include:

[0034] When the price cycle pressure coefficient is greater than or equal to the first preset threshold, a full credit authorization instruction is output, and the credit authorization ratio corresponding to the full credit authorization instruction is 100%.

[0035] When the price cycle pressure coefficient is less than the second preset threshold, a restricted credit instruction is output. The credit ratio corresponding to the restricted credit instruction is 50% and is accompanied by a risk hedging indicator.

[0036] When the price cycle pressure coefficient is between the second preset threshold and the first preset threshold, the first predetermined income is further compared with the total loan principal and interest:

[0037] If the income of the first predetermined number is greater than or equal to the total principal and interest of the loan, then a full credit line instruction is output with a flexible repayment indicator.

[0038] If the income of the first predetermined number is less than the total principal and interest of the loan, a credit compression instruction is output. The credit compression instruction corresponds to a credit ratio of 80% and is accompanied by a flexible repayment indicator.

[0039] Furthermore, the steps performed by the repayment plan generation module include:

[0040] When the full credit instruction is received without the flexible repayment indicator, the repayment mode in the generated repayment plan parameters is the normal equal principal and interest repayment mode.

[0041] When the full credit instruction or the compressed credit instruction is received and accompanied by a flexible repayment identifier, the repayment mode identifier in the generated repayment plan parameters is the income ratio repayment mode, and the repayment ratio is set to a first preset ratio or a second preset ratio, wherein the first preset ratio is greater than the second preset ratio.

[0042] When the restricted credit instruction is received along with a risk hedging identifier, the repayment mode identifier in the generated repayment plan parameters is a price threshold triggered deferred mode, and the price trigger threshold is set to the third predetermined number of the historical market price time series.

[0043] Furthermore, the steps performed by the post-loan monitoring module include:

[0044] When the repayment mode is identified as the income ratio repayment mode, the cumulative agricultural product sales revenue of the target farmer in the previous repayment cycle is obtained at the end of each repayment cycle.

[0045] Multiply the cumulative agricultural product sales revenue by the repayment ratio to obtain the amount due for the current period.

[0046] If the cumulative sales revenue of agricultural products is zero, then the amount due for the current period is zero, no default is recorded, and only deferred interest is recorded;

[0047] The amount due for the current period will be automatically deducted from the settlement account linked to the target farmer until the loan principal and interest are fully repaid.

[0048] Furthermore, the steps performed by the post-loan monitoring module also include:

[0049] When the repayment mode is identified as a price threshold-triggered deferred mode, the real-time market price data is obtained daily.

[0050] Determine whether the real-time market price data has been continuously below the price trigger threshold for a predetermined duration;

[0051] If so, the principal deferral operation will be automatically triggered: the current principal due will be moved to the end of the next harvest season for repayment, and only interest will be charged during the deferral period. If the target farmer is found to hold valid price insurance, the interest during the deferral period will be deducted from the insurance claim first.

[0052] Furthermore, the steps performed by the post-loan monitoring module also include:

[0053] After the loan is disbursed, the actual sales revenue of agricultural products of the target farmers will be continuously monitored;

[0054] When the actual sales revenue of agricultural products is higher than the third predetermined income level, the excess portion is calculated, and a predetermined percentage of the excess portion is allocated to the price stabilization account of the target farmer. The price stabilization account is managed by a bank and cannot be freely withdrawn by the target farmer.

[0055] When the principal deferral operation is triggered, funds are automatically released from the price stabilization account to repay part of the principal and interest;

[0056] Once the loan principal and interest have been fully repaid, the remaining balance in the price stabilization account will be returned to the target farmer.

[0057] Furthermore, the steps performed by the post-loan monitoring module also include:

[0058] In the post-loan stage, the real-time updated value of the price cycle pressure coefficient is recalculated based on the real-time agricultural product sales revenue and the real-time market price data;

[0059] When the real-time updated value of the price cycle pressure coefficient decreases by more than a preset change threshold within a preset monitoring window, a first warning signal is generated and pushed to the bank customer manager terminal.

[0060] When the real-time market price data is lower than the price trigger threshold for a continuous predetermined period of time, a second early warning signal is generated and pushed to the target farmer's mobile terminal.

[0061] The technical solution provided by this invention has the following advantages compared with the known prior art:

[0062] This invention breaks through the traditional risk assessment model of agricultural credit that relies on static collateral or single financial indicators. For the first time, it deeply integrates real-time agricultural production data, such as soil moisture and weather forecasts, with agricultural product market price fluctuation data to construct a joint dynamic model of yield probability distribution and price probability distribution, thereby generating a probability distribution of farmers' total seasonal income. By extracting multiple quantile incomes from the income distribution and calculating the price cycle pressure coefficient as a core quantitative indicator for credit decisions, banks can accurately characterize the debt repayment capacity boundaries of farmers under different market scenarios. This significantly improves the scientific rigor and foresight of agricultural credit risk assessment, providing a systematic technical solution to the dilemma of information asymmetry and difficulty in quantifying risk in rural finance.

[0063] This invention addresses the inherent contradiction between drastic fluctuations in agricultural product prices and the seasonal mismatch of cash flow in agricultural production. It designs a differentiated intelligent repayment plan generation mechanism: for medium-risk farmers, an income-ratio repayment model is used, dynamically linking repayment amounts to actual sales revenue to eliminate liquidity squeezes caused by rigid repayments; for high-risk farmers, a price threshold-triggered deferral model is used, automatically triggering principal deferral when market prices consistently fall below a preset threshold, preventing farmers from falling into technical default due to short-term price depressions; simultaneously, a bank-managed price stabilization account is established, extracting excess profits as risk reserves during bumper years with high prices, and automatically releasing them to repay principal and interest during years of disaster or price troughs. This upgrades post-loan management from fixed contract execution to real-time adaptive adjustment based on market signals, effectively reducing the probability of farmer defaults and the bank's credit loss rate. Attached Figure Description

[0064] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0065] Figure 1 This is a schematic diagram of the system structure of the present invention. Detailed Implementation

[0066] 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 below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of 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.

[0067] The present invention will be further described below with reference to embodiments.

[0068] Example:

[0069] Reference Figure 1 A rural credit reporting system based on the fusion of multi-source heterogeneous data includes:

[0070] The data acquisition module is configured to: acquire real-time soil moisture data of the land plots operated by the target farmers, weather forecast data for the future predetermined period, historical market price time series of the target agricultural products, planting area and expected market launch time declared by the target farmers, annual rigid expenditure data of the target farmers, and total principal and interest of the target farmers' loans.

[0071] Specifically, the data acquisition module serves as the system's data entry point. It connects to soil moisture sensors (such as FDR-type moisture sensors) deployed in the fields via an IoT gateway, collecting data on surface and deep soil volumetric water content, electrical conductivity, and soil temperature every 24 hours. Weather forecast data is obtained through API calls from the National Meteorological Administration or commercial meteorological service providers, including daily precipitation, temperature, and extreme weather warnings for the next 90 days. Historical market price time series data is obtained from national or regional agricultural wholesale market data platforms, containing at least the daily price data for the past 5 years. Information such as planting area, variety, and expected market entry time declared by farmers is entered via a mobile app or village committee service point. Annual rigid expenditure data is automatically estimated by the system based on bank loan records and publicly available statistical data, calculated using the following formula:

[0072] ;

[0073] in, This represents the total annual rigid expenditure of farmers (in yuan / year). The per capita minimum living standard for rural residents in the local area (yuan / year·person) is taken from the minimum living standard guarantee for rural residents published by the National Bureau of Statistics in that year; N is the total number of people in the rural household (persons), which is taken from the loan application file; The average annual education cost (in yuan / year) for each stage of education a child is attending is calculated based on the standards published by the local education department. For example, it is approximately 2,000 yuan for primary school, 4,000 yuan for junior high school, 8,000 yuan for senior high school, and 15,000 yuan for university. The average annual medical expenditure per capita for rural residents in the area (yuan / year / person) is taken from the annual statistical bulletin of the local health commission; r is the individual coefficient of medical expenditure (dimensionless), determined based on medical reimbursement records in the loan file: 1.0 for no history of chronic diseases, 1.5 for one chronic disease, and 2.0 for two or more chronic diseases or a history of serious illness. The total principal and interest of the loan are retrieved from the bank's internal system. All the above data are used legally within the framework of the "Personal Information Authorization Letter" signed by farmers when applying for loans, and comply with the relevant provisions of the "Personal Information Protection Law".

[0074] The yield prediction module is configured to generate a probability distribution of the target farmer's seasonal yield based on real-time soil moisture data, weather forecast data, and planting area. The probability distribution of yield includes the expected yield per acre and the standard deviation of yield.

[0075] Specifically, the core of the yield prediction module is to construct a yield probability model based on Bayesian updates. This module first obtains the historical average yield per acre of the target crop as prior information, then calculates correction factors using current season soil moisture and meteorological data, and finally outputs a normally distributed yield probability distribution. The advantage of this approach is that it can quantify the uncertainty of yield prediction, providing probabilistic input for subsequent income distribution generation.

[0076] In a preferred embodiment, the production forecasting module specifically performs the following steps:

[0077] Obtain the historical average yield per mu of the target crop as the prior expected yield per mu, and obtain the corresponding prior standard deviation;

[0078] The deviation of surface water content is calculated based on real-time soil moisture data, and the deviation of accumulated temperature and the number of days of extreme weather warnings are calculated based on meteorological forecast data.

[0079] The yield correction factor is calculated based on the deviation of surface moisture content, the deviation of accumulated temperature, and the number of days of extreme weather warnings.

[0080] Multiply the prior expected yield per acre by the yield correction factor to obtain the posterior expected yield per acre;

[0081] The posterior standard deviation is obtained by multiplying the prior standard deviation by the absolute value of the difference between the yield correction factor and 1, plus 1.

[0082] The probability distribution of yield in the form of a normal distribution is constructed using the posterior expected yield per acre and the posterior standard deviation.

[0083] The formula for calculating the yield correction factor involved in the above steps is as follows:

[0084] ;

[0085] in, This is a yield correction factor (dimensionless). The weighting factor for moisture content deviation (dimensionless). ), This refers to the volumetric water content of the topsoil. The optimal surface water content for crops The accumulated temperature from now until the expected harvest date, Accumulated temperature for crops This refers to the number of days under extreme weather warnings. This represents the absolute deviation of soil moisture content from its optimum value. When soil moisture content is too high (waterlogged) or too low (drought), this term is always positive, thus affecting... A negative correction is generated; This is the accumulated temperature deviation weighting coefficient, typically ranging from 0.4 to 1.0. This is a penalty coefficient for extreme weather, expressed in days. -1 ,make The overall value is dimensionless, with a typical value of approximately 0.008 days. -1 (Taking Apple as an example); , , All results were obtained through multiple linear regression analysis of historical yield data and meteorological data of the same variety in the region over the past five years.

[0086] Accumulated temperature from now until the expected harvest date The calculation formula is as follows:

[0087] ;

[0088] in, For the expected harvest date, For the current date, Let be the daily average temperature on day t. Base temperature, usually taken .

[0089] Posterior expected yield per acre for:

[0090] ;

[0091] in, The prior expected yield per acre (historical average yield per acre). For the prior standard deviation, This is a production correction factor.

[0092] The posterior standard deviation σ_post is calculated using a piecewise asymmetric formula to reflect the difference in the impact of uncertainty under scenarios of increased and decreased production:

[0093] when When production increases or remains the same:

[0094] ;

[0095] when (During production cuts):

[0096] ;

[0097] in, α is the prior standard deviation (kg / mu), β is the statistical standard deviation of the historical yield of the target crop, reflecting the natural fluctuation range of yield caused by climate change over many years; α is the uncertainty amplification factor of the yield increase scenario (dimensionless), and it is recommended to take the value α=0.6, which can be determined by regression fitting of historical data; β is the uncertainty amplification factor of the yield decrease scenario.

[0098] The physical meaning of the above segmented design is: when λ=1 (the current seasonal conditions are consistent with the historical average), both formulas degenerate into... The prior uncertainty is retained as a baseline. When production decreases (λ<1), farmers' actual income is lower than the forecast, increasing repayment pressure, thus assigning a higher uncertainty amplification factor (β>α), reflecting the asymmetry of credit risk. When production increases (λ>1), repayment ability is positively guaranteed, and the uncertainty amplification is relatively small. The a posteriori expected yield per acre is then used. and posterior standard deviation Construct a probability distribution of output in the form of a normal distribution.

[0099] The price prediction module is configured to generate the price probability distribution of the target agricultural product at the expected market launch time using a mixed-frequency volatility model based on historical market price time series.

[0100] Specifically, the price forecasting module is one of the core innovations of this invention. Traditional methods only use the average of historical prices or simple trend extrapolation, which cannot capture the cyclical fluctuations of agricultural product prices and the differences in quality premiums. This module adopts a mixed-frequency GARCH-MIDAS model, which can combine daily price data with monthly weather indices to separate long-term cyclical fluctuations and short-term shocks, thereby more accurately predicting the price distribution during the harvest season.

[0101] In a preferred embodiment, the price prediction module is further configured as follows:

[0102] Obtain the historical price time series of high-quality fruits and ordinary fruits of the target agricultural product;

[0103] The probability distribution of high-quality fruit prices and the probability distribution of ordinary fruit prices were predicted using the mixed-frequency GARCH-MIDAS model, respectively.

[0104] Obtain the historical management level score of the target farmers, and determine the weight of high-quality fruit based on the historical management level score;

[0105] The price probability distribution is obtained by multiplying the probability distribution of high-quality fruit prices by the weight of high-quality fruit and the probability distribution of ordinary fruit prices by one minus the weight of high-quality fruit.

[0106] Specifically, the mixed-frequency GARCH-MIDAS model will use price-return ratios. The variance is decomposed into short-term components. and long-term components The product of:

[0107] ;

[0108] in, It follows the GARCH(1,1) procedure. Driven by monthly weather indices. Model parameters, determined through maximum likelihood estimation, are used to predict future market launches. The price distribution f(p) is obtained. Since prices typically do not follow a normal distribution, a kernel density estimation method is used to obtain a nonparametric distribution. After modeling high-quality fruit and ordinary fruit separately, the weight of high-quality fruit is determined based on the farmer's historical management level score w (0≤w≤1). The quantification method of w is as follows:

[0109] The system retrieves records of farmers' actual refined management measures over the past three years from the "Annual Verification Form of Agricultural Production" in the bank loan files (entered after on-site verification by agricultural technology extension station staff). Scores are awarded annually according to the following rules: 3 points for water-saving irrigation (drip irrigation or subsurface irrigation); 2 points for applying organic fertilizer; 2 points for professional green pest control; 1 point for fruit bagging; 1 point for scientific pruning; 1 point for using soil testing data to guide fertilization; and a maximum score for each year. The score is 10 points. The normalized average score over three years is the management level score.

[0110] ;

[0111] in, This represents the actual score (points) in year i. We assign a score of 10, where w ∈ [0,1]. The closer w is to 1, the higher the level of intensive management by the farmers, the greater the probability of producing high-quality fruit, and the higher the weight of high-quality fruit. Conversely, the closer w is to 0, the more extensive the management by the farmers, and the lower the weight of high-quality fruit. The final price distribution combines the price probability density functions of high-quality fruit and ordinary fruit.

[0112] ;

[0113] in, Let w be the probability density function of the overall price, i.e., the price distribution, and w be the weight of high-quality fruit, i.e., the farmer's historical management level score. Let be the probability density function of the price of high-quality fruit. Let w be the probability density function of the price of ordinary fruit, and w be the weight of high-quality fruit, which is the farmer's historical management level score.

[0114] The income distribution generation module is configured to: perform a convolution operation on the probability distribution of output and the probability distribution of price to generate the probability distribution of the target farmer's total income for the current season, and extract the first predetermined number income, the second predetermined number income, and the third predetermined number income from the total income probability distribution, wherein the first predetermined number income is lower than the second predetermined number income, and the second predetermined number income is lower than the third predetermined number income.

[0115] Specifically, total income is the product of output and price. At the regional supply and demand level, in years of bumper harvests of the same crop, the regional total output increases, and the average market price often falls accordingly, showing a negative correlation between the two. This module uses the Monte Carlo simulation method, introducing the output-price correlation coefficient calculated from regional macroeconomic data, to reflect this supply and demand linkage in the income distribution generation process, generating a probability distribution of total income, and extracting the 10%, 50%, and 90% quantiles as the basis for subsequent judgments.

[0116] The specific implementation steps are as follows:

[0117] Step 1: Obtain the historical correlation coefficient between regional yield and price. Obtain the time series of annual regional total yield for the past ten years from the national agricultural information dispatch system of the Ministry of Agriculture and Rural Affairs. Obtain the corresponding annual average market price time series from the national agricultural product wholesale market price monitoring database. Calculate the Pearson correlation coefficient:

[0118] ;

[0119] in The Pearson correlation coefficient (dimensionless) between the region's total annual output and average annual price reflects the pattern of price declines in bumper years driven by macroeconomic supply and demand; for agricultural products with cyclical prices... It is typically between -0.3 and -0.6.

[0120] Step 2: Generate correlated random sample pairs using the Cholesky decomposition method. Construct a 2×2 correlation matrix. Then, Cholesky decomposition is performed to obtain the lower triangular matrix L; independent random vectors Z are generated from the standard normal distribution, and X = L·Z is set so that the two components carry a correlation coefficient. ; thereby sampling the output distribution and price distribution respectively, to obtain the output sample for the i-th simulation. and price samples The number of simulations is N=10,000.

[0121] Step 3: Calculate the total revenue for each simulation:

[0122] ;

[0123] in, The total revenue (in yuan) for the i-th simulation. Let A be the yield sample (kg / mu) from the i-th sampling; A is the planting area (mu). Let be the price sample (yuan / kg) drawn for the i-th time.

[0124] Step 4: Sort the total income from the 10,000 simulations in ascending order and take the 10th percentile as the first predetermined percentile income. The 50th percentile is taken as the second predetermined percentile of income. The 90th percentile is taken as the third predetermined percentile income. .

[0125] The pressure coefficient calculation module is configured to calculate the price cycle pressure coefficient by dividing the difference between the second predetermined income and the annual rigid expenditure data by the total loan principal and interest.

[0126] Specifically, the price cycle pressure coefficient P is the core indicator of this invention. This coefficient represents the ratio of disposable income (after deducting basic living expenses) to the total principal and interest of loans at a moderate probability (50th percentile) income level. The calculation formula is as follows:

[0127] ;

[0128] in, For the second pre-determined number of revenue, This covers farmers' annual essential expenditures (including basic living expenses, children's education, and medical care). This represents the total principal and interest of the loan. When P ≥ 1.2, it indicates a sufficient safety margin; when P < 0.8, it indicates a higher risk.

[0129] The credit decision module is configured as follows: it compares the price cycle pressure coefficient with a first preset threshold and a second preset threshold, where the first preset threshold is greater than the second preset threshold; when the price cycle pressure coefficient is greater than or equal to the first preset threshold, it outputs a full credit instruction; when the price cycle pressure coefficient is between the second preset threshold and the first preset threshold, it outputs a reduced credit instruction or a full credit instruction with a flexible repayment flag based on the comparison result of the first preset income and the total loan principal and interest; when the price cycle pressure coefficient is less than the second preset threshold, it outputs a restricted credit instruction with a risk hedging flag.

[0130] Specifically, this module compares the price cycle pressure coefficient P with preset thresholds (the first preset threshold is 1.2, and the second preset threshold is 0.8), and further examines the worst-case scenario. Based on the debt repayment capacity, multi-level credit decision-making can be achieved.

[0131] In a preferred embodiment, the credit decision module specifically performs the following:

[0132] When the price cycle pressure coefficient is greater than or equal to the first preset threshold, a full credit authorization instruction is output, and the credit authorization ratio corresponding to the full credit authorization instruction is 100%.

[0133] When the price cycle pressure coefficient is less than the second preset threshold, a restricted credit instruction is output. The credit ratio corresponding to the restricted credit instruction is 50% and is accompanied by a risk hedging indicator.

[0134] When the price cycle pressure coefficient is between the second preset threshold and the first preset threshold, further compare the first predetermined income with the total loan principal and interest:

[0135] If the income of the first predetermined number is greater than or equal to the total principal and interest of the loan, a full credit line instruction will be issued with a flexible repayment indicator.

[0136] If the income of the first predetermined number is less than the total principal and interest of the loan, a credit compression instruction will be issued. The credit compression instruction corresponds to a credit ratio of 80% and is accompanied by a flexible repayment indicator.

[0137] The above logic can be summarized as follows: In the safe zone (P≥1.2), full credit is granted; in the high-risk zone (P<0.8), only 50% credit is granted with mandatory insurance; in the gray zone (0.8≤P<1.2), further credit is granted based on bottom support levels (…). Whether the credit line covers principal and interest determines whether to grant the full amount or reduce it by 20%.

[0138] The repayment plan generation module is configured to generate corresponding repayment plan parameters in response to different instructions output by the credit decision module. The repayment plan parameters include repayment mode identifier, repayment ratio, and price trigger threshold.

[0139] Specifically, this module matches different repayment plan templates based on the instruction type output by the credit decision module.

[0140] In a preferred embodiment, the repayment plan generation module specifically performs the following:

[0141] When a full credit line instruction is received without a flexible repayment option, the repayment mode in the generated repayment plan parameters will be set to the normal equal principal and interest repayment mode.

[0142] When a full credit line instruction or a compressed credit line instruction is received and a flexible repayment indicator is attached, the repayment mode in the generated repayment plan parameters is identified as income ratio repayment mode, and the repayment ratio is set to a first preset ratio or a second preset ratio, wherein the first preset ratio is greater than the second preset ratio.

[0143] When a restricted credit instruction is received with a risk hedging flag, the repayment mode in the generated repayment plan parameters is identified as a price threshold triggered deferred mode, and the price trigger threshold is set to the third predetermined number of the historical market price time series.

[0144] The first preset ratio can be set to 30%, corresponding to full credit line plus flexible repayment; the second preset ratio can be set to 25%, corresponding to reduced credit line plus flexible repayment. The price trigger threshold is the 20th percentile of the historical price series, indicating that the deferral is triggered when the market price falls to a relatively low level in the past 5 years.

[0145] The post-loan monitoring module is configured to continuously acquire real-time agricultural product sales revenue data and real-time market price data of target farmers after the loan is issued, dynamically adjust the amount due for the current period according to the repayment plan parameters, and automatically trigger the principal deferral operation when the real-time market price data is lower than the price trigger threshold for a predetermined period of time.

[0146] Specifically, the post-loan monitoring module is key to the system's dynamic adjustment. It continuously connects to agricultural product sales platforms (such as e-commerce platforms and cooperative purchase records) and price monitoring systems to obtain sales revenue and market prices in real time. Then, it automatically executes repayment based on income ratio or triggers deferral based on price thresholds according to repayment plan parameters.

[0147] In a preferred embodiment, the post-loan monitoring module is further configured (corresponding to the income ratio repayment mode):

[0148] When the repayment mode is identified as the income ratio repayment mode, the cumulative agricultural product sales revenue of the target farmer in the previous repayment cycle is obtained at the end of each repayment cycle.

[0149] Multiply the cumulative sales revenue of agricultural products by the repayment ratio to obtain the amount due for the current period.

[0150] If the cumulative sales revenue of agricultural products is zero, the amount due for the current period is zero, no default is recorded, and only deferred interest is recorded;

[0151] The loan repayment amount will be automatically deducted from the settlement account linked to the target farmer until the loan principal and interest are fully repaid.

[0152] In another preferred embodiment, the post-loan monitoring module is also configured to (correspond to a price threshold triggering a deferred mode):

[0153] When the repayment mode is identified as a price threshold-triggered deferred mode, real-time market price data is obtained daily.

[0154] Determine whether the real-time market price data has been below the price trigger threshold for a continuous pre-order duration (e.g., 14 consecutive days).

[0155] If so, the principal deferral operation will be automatically triggered: the current principal due will be moved to the end of the next harvest season for repayment, and only interest will be charged during the deferral period. If the target farmer is found to have valid price insurance, the interest during the deferral period will be deducted from the insurance claim first.

[0156] In addition, the post-loan monitoring module is also equipped with price stabilization account management functions:

[0157] After the loan is issued, the actual sales revenue of agricultural products of the target farmers will be continuously monitored;

[0158] When the actual sales revenue of agricultural products exceeds the third predetermined threshold revenue, the excess portion is calculated, and a predetermined percentage (e.g., 30%) of the excess portion is allocated to the price stabilization account of the target farmers. The price stabilization account is managed by a bank and cannot be freely withdrawn by the target farmers.

[0159] When the principal deferral operation is triggered, funds are automatically released from the price stabilization account to repay part of the principal and interest;

[0160] Once the loan principal and interest have been fully repaid, the remaining balance in the price stabilization account will be returned to the target farmers.

[0161] Specifically, the price stabilization account mechanism uses the excess profits in bumper years to establish a risk buffer pool, which is automatically released as compensation in lean years, thus realizing cross-cycle risk sharing between banks and farmers.

[0162] Finally, the post-loan monitoring module is also equipped with an early warning function:

[0163] In the post-loan phase, the real-time updated value of the price cycle pressure coefficient is recalculated based on real-time agricultural product sales revenue and real-time market price data;

[0164] When the real-time updated value of the price cycle pressure coefficient decreases by more than the preset change threshold (e.g., 0.3) within a preset monitoring window (e.g., 30 days), a first warning signal (orange warning) is generated and pushed to the bank's customer manager terminal.

[0165] When real-time market price data remains below the price trigger threshold for a predetermined duration (e.g., 7 days), a second warning signal (yellow warning) is generated and pushed to the target farmer's mobile terminal.

[0166] Through the above process, this system achieves adaptive credit granting and repayment matching based on agricultural product price fluctuation cycles, effectively reducing the default risk of rural credit.

[0167] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions will not cause the essence of the corresponding technical solutions to deviate from the protection scope of the technical solutions of the embodiments of the present invention.

Claims

1. A rural credit credit investigation system based on multi-source heterogeneous data fusion, characterized in that, include: The data acquisition module is used to acquire real-time soil moisture data of the land operated by the target farmers, weather forecast data for the future predetermined period, historical market price time series of the target agricultural products, planting area and expected market time declared by the target farmers, annual rigid expenditure data of the target farmers, and total principal and interest of the target farmers' loans. The yield prediction module is used to generate a probability distribution of the target farmer's seasonal yield based on the real-time soil moisture data, the weather forecast data, and the planting area. The probability distribution of the yield includes the expected yield per acre and the standard deviation of the yield. The price prediction module is used to generate the price probability distribution of the target agricultural product at the expected listing time based on the historical market price time series using a mixed-frequency volatility model. The income distribution generation module is used to perform a convolution operation on the probability distribution of the output and the probability distribution of the price to generate the probability distribution of the total income of the target farmer in the current season, and to extract a first predetermined number income, a second predetermined number income and a third predetermined number income from the total income probability distribution, wherein the first predetermined number income is lower than the second predetermined number income and the second predetermined number income is lower than the third predetermined number income. The pressure coefficient calculation module is used to calculate the price cycle pressure coefficient by dividing the difference between the second predetermined income and the annual rigid expenditure data by the total loan principal and interest. The credit decision module is used to compare the price cycle pressure coefficient with a first preset threshold and a second preset threshold, wherein the first preset threshold is greater than the second preset threshold; when the price cycle pressure coefficient is greater than or equal to the first preset threshold, a full credit instruction is output; when the price cycle pressure coefficient is between the second preset threshold and the first preset threshold, based on the comparison result of the first predetermined income and the total principal and interest of the loan, a reduced credit instruction or a full credit instruction with a flexible repayment indicator is output; when the price cycle pressure coefficient is less than the second preset threshold, a restricted credit instruction with a risk hedging indicator is output. The repayment plan generation module is used to generate corresponding repayment plan parameters in response to different instructions output by the credit decision module. The repayment plan parameters include repayment mode identifier, repayment ratio, and price trigger threshold. The post-loan monitoring module is used to continuously acquire real-time agricultural product sales revenue data and real-time market price data of the target farmers after the loan is issued, dynamically adjust the amount due for the current period according to the repayment plan parameters, and automatically trigger the principal deferral operation when the real-time market price data is lower than the price trigger threshold for a predetermined period of time. 2.The rural credit credit investigation system based on multi-source heterogeneous data fusion according to claim 1, characterized in that, The steps performed by the production prediction module include: The historical average yield per mu of the target crop is obtained as the prior expected yield per mu, and the corresponding prior standard deviation is obtained. The deviation of surface water content is calculated based on the real-time soil moisture data, and the deviation of accumulated temperature and the number of extreme weather warning days are calculated based on the meteorological forecast data. The yield correction factor is calculated based on the deviation of surface moisture content, the deviation of accumulated temperature, and the number of days of extreme weather warning. Multiply the prior expected yield per acre by the yield correction factor to obtain the posterior expected yield per acre; The posterior standard deviation is obtained by multiplying the prior standard deviation by the absolute value of the difference between the yield correction factor and 1, plus 1. A probability distribution of the yield in the form of a normal distribution is constructed using the posterior expected yield per acre and the posterior standard deviation. 3.The rural credit investigation and credit system based on multi-source heterogeneous data fusion according to claim 1, characterized in that, The steps performed by the price prediction module include: Obtain the historical price time series of high-quality fruit and ordinary fruit of the target agricultural product; The probability distribution of high-quality fruit prices and the probability distribution of ordinary fruit prices were predicted using the mixed-frequency GARCH-MIDAS model, respectively. Obtain the historical management level score of the target farmer, and determine the weight of high-quality fruit based on the historical management level score; The price probability distribution is obtained by multiplying the price probability distribution of high-quality fruit by the weight of high-quality fruit and the price probability distribution of ordinary fruit by one minus the weight of high-quality fruit.

4. The rural, credit and credit investigation system based on multi-source heterogeneous data fusion according to claim 1, characterized in that, The steps performed by the income distribution generation module include: Obtain the correlation coefficient between the historical yield and historical price of the target crop; The Monte Carlo simulation method is used to extract a preset number of sample pairs from the probability distribution of output and the probability distribution of price. Each time a sample is extracted, a negative correlation between output and price is introduced based on the correlation coefficient. Multiply the yield sample extracted each time by the planting area, and then multiply by the price sample to obtain the total income for a single simulation. Arrange the simulated total revenue of the preset number of times in ascending order, take the 10th percentile as the first predetermined percentile revenue, take the 50th percentile as the second predetermined percentile revenue, and take the 90th percentile as the third predetermined percentile revenue. 5.The rural credit investigation and credit system based on multi-source heterogeneous data fusion according to claim 1, characterized in that, The steps performed by the credit granting decision module include: When the price cycle pressure coefficient is greater than or equal to the first preset threshold, a full credit authorization instruction is output, and the credit authorization ratio corresponding to the full credit authorization instruction is 100%. When the price cycle pressure coefficient is less than the second preset threshold, a restricted credit instruction is output. The credit ratio corresponding to the restricted credit instruction is 50% and is accompanied by a risk hedging indicator. When the price cycle pressure coefficient is between the second preset threshold and the first preset threshold, the first predetermined income is further compared with the total loan principal and interest: If the income of the first predetermined number is greater than or equal to the total principal and interest of the loan, then a full credit line instruction is output with a flexible repayment indicator. If the income of the first predetermined number is less than the total principal and interest of the loan, a credit compression instruction is output. The credit compression instruction corresponds to a credit ratio of 80% and is accompanied by a flexible repayment indicator. 6.The rural credit investigation and credit system based on multi-source heterogeneous data fusion according to claim 1, characterized in that, The steps performed by the repayment plan generation module include: When the full credit instruction is received without the flexible repayment indicator, the repayment mode in the generated repayment plan parameters is the normal equal principal and interest repayment mode. When the full credit instruction or the compressed credit instruction is received and accompanied by a flexible repayment identifier, the repayment mode identifier in the generated repayment plan parameters is the income ratio repayment mode, and the repayment ratio is set to a first preset ratio or a second preset ratio, wherein the first preset ratio is greater than the second preset ratio. When the restricted credit instruction is received along with a risk hedging identifier, the repayment mode identifier in the generated repayment plan parameters is a price threshold triggered deferred mode, and the price trigger threshold is set to the third predetermined number of the historical market price time series. 7.The rural credit investigation and credit system based on multi-source heterogeneous data fusion according to claim 1, characterized in that, The steps performed by the post-loan monitoring module include: When the repayment mode is identified as the income ratio repayment mode, the cumulative agricultural product sales revenue of the target farmer in the previous repayment cycle is obtained at the end of each repayment cycle. Multiply the cumulative agricultural product sales revenue by the repayment ratio to obtain the amount due for the current period. If the cumulative sales revenue of agricultural products is zero, then the amount due for the current period is zero, no default is recorded, and only deferred interest is recorded; The amount due for the current period will be automatically deducted from the settlement account linked to the target farmer until the loan principal and interest are fully repaid.

8. The rural credit reporting system based on multi-source heterogeneous data fusion according to claim 1, characterized in that, The steps performed by the post-loan monitoring module also include: When the repayment mode is identified as a price threshold-triggered deferred mode, the real-time market price data is obtained daily. Determine whether the real-time market price data has been continuously below the price trigger threshold for a predetermined duration; If so, the principal deferral operation will be automatically triggered: the current principal due will be moved to the end of the next harvest season for repayment, and only interest will be charged during the deferral period. If the target farmer is found to hold valid price insurance, the interest during the deferral period will be deducted from the insurance claim first.

9. The rural credit reporting system based on multi-source heterogeneous data fusion according to claim 1, characterized in that, The steps performed by the post-loan monitoring module also include: After the loan is disbursed, the actual sales revenue of agricultural products of the target farmers will be continuously monitored; When the actual sales revenue of agricultural products is higher than the third predetermined income level, the excess portion is calculated, and a predetermined percentage of the excess portion is allocated to the price stabilization account of the target farmer. The price stabilization account is managed by a bank and cannot be freely withdrawn by the target farmer. When the principal deferral operation is triggered, funds are automatically released from the price stabilization account to repay part of the principal and interest; Once the loan principal and interest have been fully repaid, the remaining balance in the price stabilization account will be returned to the target farmer.

10. The rural credit reporting system based on multi-source heterogeneous data fusion according to claim 1, characterized in that, The steps performed by the post-loan monitoring module also include: In the post-loan stage, the real-time updated value of the price cycle pressure coefficient is recalculated based on the real-time agricultural product sales revenue and the real-time market price data; When the real-time updated value of the price cycle pressure coefficient decreases by more than a preset change threshold within a preset monitoring window, a first warning signal is generated and pushed to the bank's customer manager terminal. When the real-time market price data is continuously lower than the price trigger threshold for a predetermined period of time, a second early warning signal is generated and pushed to the target farmer's mobile terminal.

Images