Agricultural financing guarantee-oriented risk control model based on intelligent insurance multi-dimensional dynamic scoring

Abstract

The invention discloses an agricultural financing guarantee-oriented risk control model based on intelligent protection multi-dimensional dynamic scoring, and the model comprises a multi-dimensional adaptive prediction model and a multi-dimensional adaptive weight data selection module, and the main purpose of the multi-dimensional adaptive weight data selection module is to carry out risk assessment according to different agricultural production scenes and risk assessment demands. Distributing a reasonable weight for each data source; and dynamically adjusting the weight. The invention relates to the technical field of data processing and analysis, and has the beneficial effects that by integrating multi-source data including traditional financial data, social media data, network consumption data, water and electricity payment records and the like, data sources are widened, excessive dependence on single historical data is reduced, and the data processing efficiency is improved. The scoring system can more comprehensively reflect the real credit status and risk level of the customer, and adapts to the change of the market environment, the economic situation and the customer behavior mode.

Description

Technical Field

[0001] The present invention relates to the field of data processing and analysis, and in particular to a risk control model for agricultural financing guarantees based on intelligent insurance multi-dimensional dynamic scoring. Background Art

[0002] Agricultural financing guarantees currently face the following difficulties and technical bottlenecks:

[0003] Imperfect valuation technology: There is a lack of scientific, accurate and unified valuation standards and methods for the valuation of agricultural assets such as land management rights, crops, livestock, agricultural production facilities, etc. Different valuation agencies may use different valuation models and parameters, resulting in large differences in valuation results, which affects the value determination of collateral and the determination of financing quotas.

[0004] Inaccurate risk assessment technology: Existing risk assessment technology mainly relies on traditional financial indicators and historical data, and does not take into account the particularity and uncertainty of agricultural production. The lack of risk assessment models and tools for agricultural projects makes it difficult to accurately predict the impact of natural disasters, market fluctuations and other factors on agricultural operations and loan repayments, which increases the difficulty of risk control for guarantee institutions.

[0005] Insufficient data management and information technology application: Agricultural financing guarantees involve a large amount of information data, including basic information of agricultural operators, production and operation data, credit records, etc. However, the current data management level of guarantee institutions is low, and the ability to collect, organize, analyze and share information is insufficient. There is a lack of effective information management systems and big data technology applications, which affects the scientificity and efficiency of business decision-making.

[0006] Therefore, it is necessary to combine the actual business of agricultural financing guarantee, and use the Zhibao multi-dimensional dynamic scoring algorithm according to the characteristics of the target population and business reports. This can automatically select features based on the characteristics of agricultural financing, and normalize them to obtain a risk control model with high accuracy specifically for agricultural financing guarantee. Summary of the invention

[0007] The purpose of the present invention is to solve the above problems and to design a risk control model for agricultural financing guarantees based on intelligent insurance multi-dimensional dynamic scoring.

[0008] To achieve the above purpose, the technical solution of the present invention is to provide a risk control model for agricultural financing guarantee based on intelligent insurance multi-dimensional dynamic scoring, including a multi-dimensional adaptive prediction model and a multi-dimensional adaptive weight data selection module. The main purpose of the multi-dimensional adaptive weight data selection module is to assign reasonable weights to each data source according to different agricultural production scenarios and risk assessment requirements;

[0009] First, for each data source, a data quality assessment is performed;

[0010] Next, the correlation between each data source and agricultural production risk is analyzed; through statistical analysis and machine learning algorithms, the influence of different data sources on the prediction target is determined;

[0011] Then, consider the timeliness factor; newer data usually better reflects the current agricultural production situation and is therefore given a higher weight;

[0012] Finally, the weights are adjusted dynamically according to different agricultural production scenarios and risk assessment needs;

[0013] The multi-dimensional adaptive forecasting model combines a variety of forecasting methods, including time series models, regression models, and machine learning models;

[0014] Time series models include seasonal decomposition models and ARIMA models, which are mainly used to analyze data with time series characteristics;

[0015] Regression models include Bayesian probability model, Rogers regression and reverse neural network, which are used to establish the relationship between input variables and output variables;

[0016] Machine learning models include decision trees and random forests.

[0017] Furthermore, in the multi-dimensional adaptive weight data selection module, it is assumed that we have n data sources, namely D1, D2, ..., Dn; for each data source Di, we define its weight as wi, and satisfy ∑wi = 1;

[0018] First, calculate the reliability index Ri of the data source Di. The value of Ri can be determined by data quality assessment and historical data accuracy verification. For example, the accuracy and recall rate of the data in the data source Di can be calculated, and then these indicators can be combined into a reliability index Ri.

[0019] Next, the correlation index Ci between the data source Di and the prediction target Y is calculated; the value of Ci can be determined using the correlation coefficient and regression analysis method; for example, the Pearson correlation coefficient between the data in the data source Di and the prediction target Y can be calculated as the correlation index Ci;

[0020] Then, considering the timeliness factor, define the timeliness index Ti of the data source Di; the value of Ti can be determined according to the timeliness and update frequency of the data; for example, the timeliness index of recent data can be set to a higher value, and as time goes by, the timeliness index gradually decreases;

[0021] Finally, considering reliability, relevance and timeliness, the weight wi of the data source Di is calculated; the following formula can be used:

[0022] wi = α * Ri + β * Ci + γ * Ti

[0023] Among them, α, β, and γ are the weight coefficients of reliability, relevance, and timeliness, respectively, which can be adjusted according to specific application scenarios and requirements.

[0024] Furthermore, in the multi-dimensional adaptive prediction model, random forest is taken as an example. Random forest is an integrated model composed of multiple decision trees. Assume that we have a training data set X = {x1, x2,..., xm}, where xi is a d-dimensional input vector and the corresponding output is yi;

[0025] First, a portion of sample data is randomly selected as a training set to build multiple decision trees; for each decision tree, a portion of features is randomly selected from the training set for splitting until a certain stopping condition is reached;

[0026] Then, for a new input vector x, each decision tree will give a prediction value y^i; the final prediction result is the average of the prediction values ​​of all decision trees:

[0027] y^ = (1 / N) * ∑y^i

[0028] Where N is the number of decision trees, y^i is the predicted value of the i-th decision tree;

[0029] For the reverse neural network, suppose we have an input layer, a hidden layer, and an output layer; the input layer has d neurons, corresponding to the d features of the input vector x; the hidden layer has h neurons; the output layer has one neuron, corresponding to the predicted target y;

[0030] First, define the weight matrix W1 from the input layer to the hidden layer, with a size of h × d; the weight matrix W2 from the hidden layer to the output layer, with a size of 1 × h; the activation function of the hidden layer is f, and the activation function of the output layer is g;

[0031] For an input vector x, the output of the hidden layer is:

[0032] z = f (W1 * x)

[0033] Where * represents matrix multiplication;

[0034] The predicted value of the output layer is:

[0035] y^ = g (W2 * z)

[0036] Through training data, the back propagation algorithm is used to continuously adjust the weight matrices W1 and W2 to minimize the error between the predicted value y^ and the actual value y. The commonly used error function is the mean square error (MSE).

[0037] Furthermore, the seasonal decomposition model decomposes agricultural-related data according to time series and extracts the trend, seasonality and residual parts therein; by analyzing seasonal factors, it is possible to better understand the impact of different seasons on agricultural production and predict the risks that may be faced in a specific season.

[0038] Furthermore, the ARIMA model uses an autoregressive moving average model to analyze and predict data with time series characteristics; by analyzing the autocorrelation and moving average characteristics of historical data, future agricultural production indicators are predicted.

[0039] Furthermore, the Bayesian probability model estimates the posterior probability based on the Bayesian theory by combining the prior probability and the sample data;

[0040] The Rogers regression is applicable to binary classification problems and can be used to predict whether a certain event in agricultural production will occur;

[0041] The reverse neural network can learn the complex nonlinear relationship between input variables and output variables;

[0042] Decision tree, the decision tree model can classify and predict agricultural production risks based on different characteristic variables;

[0043] Random Forest,Random Forest is an integrated model composed of multiple decision trees.

[0044] Furthermore, it also includes: GIS meteorological data capture, GIS pest and disease data, NLP agricultural news, NLP public opinion analysis, and Baixing data.

[0045] Furthermore, in GIS meteorological data capture, the weekly average temperature data in a specific area is collected, the weekly average precipitation data in a specific area is obtained, and the meteorological data over a period of time in the past, including temperature, precipitation, and light, is stored and managed. Geographic Information System (GIS) technology is used to integrate geospatial data and meteorological data.

[0046] Furthermore, GIS pest and disease data, combined with GIS technology, collects and displays the occurrence of pests and diseases in a specific area, and stores and manages pest and disease occurrence data over a period of time, including pest and disease type, occurrence time, location, and degree of damage.

[0047] Furthermore, NLP agricultural news uses natural language processing (NLP) technology to analyze and extract key information from agriculture-related news;

[0048] NLP public opinion analysis: Use NLP technology to analyze agriculture-related public opinions on social media and forum channels.

[0049] Baihang Data evaluates the risk of guarantors in agricultural loan scenarios. By analyzing the guarantor's credit record, financial status, and career stability factors, the reliability and guarantee ability of the guarantor are judged.

[0050] The risk control model for agricultural financing guarantees based on intelligent insurance multi-dimensional dynamic scoring, which is produced using the technical solution of the present invention, broadens the data sources and reduces excessive reliance on single historical data by integrating multi-source data, including traditional financial data, social media data, online consumption data, water and electricity payment records, etc., so that the scoring system can more comprehensively reflect the customer's true credit status and risk level and adapt to changes in the market environment, economic situation and customer behavior patterns.

[0051] Improve the comprehensiveness of information. Utilize multi-dimensional data and adaptive algorithms to capture potential risk factors such as personal qualities and special family situations that are difficult to quantify, make up for the deficiencies in information coverage of existing scorecard models, and provide a more complete perspective for credit assessment, thereby making more accurate decisions.

[0052] Enhance timeliness. Introduce real-time data collection mechanisms and dynamic adjustment algorithms to ensure that the scoring system can promptly reflect recent sudden and short-term behavioral changes of customers, overcome the timeliness lag of existing behavioral scoring cards, and provide more timely support for loan management and risk warning.

[0053] Improve data quality adaptability. Design data cleaning and preprocessing modules to effectively handle errors, missing values, and outliers in the data, reduce the impact of data quality on scoring accuracy, and enable the scoring system to maintain high reliability and stability under different data quality conditions.

[0054] Improve the comprehensiveness of risk warning. Through multi-dimensional adaptive scoring, not only overdue customers can be evaluated, but also customers who are not overdue but have potential risks can be effectively identified, expanding the coverage of risk assessment and providing earlier warning signals for collection management and risk control.

[0055] Break through the limitations of existing model assumptions. Use advanced machine learning algorithms and adaptive model structures to get rid of the reliance of traditional models such as logistic regression on linear relationships and specific assumptions, better handle nonlinear relationships and complex data patterns, and improve the accuracy and adaptability of scoring.

[0056] Solve the problems of overfitting and instability. Use optimized algorithms and model verification mechanisms to prevent overfitting of models such as decision trees, improve the stability of models under different data sets and data sampling, and ensure the reliability and consistency of scoring results.

[0057] Improve model interpretability. When building a multi-dimensional adaptive scoring system, focus on the interpretability design of the model so that users can understand the scoring decision process and the basis of the prediction results, providing strong support for the verification, supervision and application of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 It is a block diagram of the risk control model for agricultural financing guarantee based on intelligent insurance multi-dimensional dynamic scoring described in the present invention. DETAILED DESCRIPTION

[0059] The present invention will be described in detail below in conjunction with the accompanying drawings. Figure 1 As shown in the figure, the risk control model for agricultural financing guarantee based on Zhibao multi-dimensional dynamic scoring, the main purpose of the multi-dimensional adaptive weight data selection module is to assign reasonable weights to each data source according to different agricultural production scenarios and risk assessment requirements, so as to improve the accuracy and reliability of the prediction. This module will comprehensively consider multiple factors, including the reliability, timeliness, relevance and importance of the data source in different scenarios.

[0060] First, for each data source, a data quality assessment is performed. This includes checking the completeness, accuracy, and consistency of the data. For example, for meteorological data, the accuracy of the sensor and the frequency of data collection are checked; for pest and disease data, the source of the data and the reliability of the detection method are verified.

[0061] Next, the correlation between each data source and agricultural production risk is analyzed. Through statistical analysis and machine learning algorithms, the influence of different data sources on the prediction target (such as crop yield, probability of pests and diseases, etc.) is determined. For example, the correlation coefficient between meteorological data (such as temperature and precipitation) and crop yield can be calculated to evaluate the importance of meteorological factors on yield.

[0062] Then, consider the timeliness factor. Newer data usually better reflect the current agricultural production situation, so it will be given a higher weight. For example, recent meteorological data and pest monitoring data will be given a larger weight, while historical data will be gradually reduced in weight according to its time.

[0063] Finally, the weights are adjusted dynamically according to different agricultural production scenarios and risk assessment needs. For example, the demand for different data sources may be different at different stages of crop growth. During the sowing period, soil data and weather forecasts may be more important; while during the high incidence of pests and diseases, pest and disease data and prevention and control information in agricultural news may be given higher weights.

[0064] Mathematical explanation

[0065] Assume that we have n data sources, D1, D2,..., Dn. For each data source Di, we define its weight as wi, and satisfy ∑wi = 1.

[0066] First, calculate the reliability index Ri of the data source Di. The value of Ri can be determined by data quality assessment and historical data accuracy verification. For example, you can calculate the accuracy, recall rate and other indicators of the data in the data source Di, and then combine these indicators into a reliability index Ri.

[0067] Next, the correlation index Ci between the data source Di and the predicted target Y is calculated. The value of Ci can be determined using methods such as correlation coefficient and regression analysis. For example, the Pearson correlation coefficient between the data in the data source Di and the predicted target Y can be calculated as the correlation index Ci.

[0068] Then, considering the timeliness factor, define the timeliness index Ti of the data source Di. The value of Ti can be determined according to the timeliness and update frequency of the data. For example, the timeliness index of recent data can be set to a higher value, and the timeliness index gradually decreases as time goes by.

[0069] Finally, considering reliability, relevance and timeliness, the weight wi of data source Di is calculated. The following formula can be used:

[0070] wi = α * Ri + β * Ci + γ * Ti

[0071] Among them, α, β, and γ are the weight coefficients of reliability, relevance, and timeliness, respectively, which can be adjusted according to specific application scenarios and requirements. By adjusting these coefficients, the importance of different factors in weight calculation can be highlighted.

[0072] Multidimensional adaptive prediction model

[0073] How it works

[0074] The multidimensional adaptive forecasting model combines multiple forecasting methods, including time series models, regression models, and machine learning models, to fully utilize the advantages of different models and improve the accuracy and adaptability of forecasting.

[0075] Time series models, such as seasonal decomposition models and ARIMA models, are mainly used to analyze data with time series characteristics. These models can capture trends, seasonality, and random fluctuations in data to predict future values. For example, seasonal decomposition models can decompose crop yield data into trend, seasonality, and residual parts, and predict yield changes in different seasons by analyzing seasonal factors.

[0076] Regression models, such as Bayesian probability models, Rogers regression, and inverse neural networks, are used to establish relationships between input variables and output variables. These models can consider multiple influencing factors and make more comprehensive predictions of agricultural production risks. For example, the Bayesian probability model can combine prior probabilities and sample data to estimate the probability of crop pests and diseases; Rogers regression can be used to predict whether crops are affected; and inverse neural networks can learn complex nonlinear relationships and predict crop yields.

[0077] Machine learning models, such as decision trees and random forests, have powerful classification and regression capabilities. These models can automatically select important feature variables, process high-dimensional data, and improve the accuracy of predictions. For example, decision trees can classify agricultural production risks based on different feature variables, and random forests can reduce overfitting risks and improve prediction stability by integrating multiple decision trees.

[0078] In practical applications, the multi-dimensional adaptive prediction model will select the appropriate model combination for prediction based on different data sources and prediction targets. At the same time, by continuously learning and updating model parameters, it can adapt to data changes and new situations, thereby improving the adaptability and accuracy of the prediction.

[0079] Mathematical explanation

[0080] Take random forest as an example, which is an integrated model composed of multiple decision trees. Suppose we have a training data set X = {x1, x2,..., xm}, where xi is a d-dimensional input vector and the corresponding output is yi.

[0081] First, a portion of sample data is randomly selected as a training set to build multiple decision trees. For each decision tree, a portion of features is randomly selected from the training set for splitting until a certain stopping condition is reached (such as the depth of the tree, the minimum number of samples in a leaf node, etc.).

[0082] Then, for a new input vector x, each decision tree will give a predicted value y^i. The final prediction result is the average of the predictions of all decision trees:

[0083] y^ = (1 / N) * ∑y^i

[0084] Where N is the number of decision trees and y^i is the predicted value of the ith decision tree.

[0085] For the reverse neural network, suppose we have an input layer, a hidden layer, and an output layer. The input layer has d neurons, corresponding to the d features of the input vector x; the hidden layer has h neurons; and the output layer has one neuron, corresponding to the predicted target y.

[0086] First, define the weight matrix W1 from the input layer to the hidden layer, with a size of h × d; the weight matrix W2 from the hidden layer to the output layer, with a size of 1 × h. The activation function of the hidden layer is f, and the activation function of the output layer is g.

[0087] For an input vector x, the output of the hidden layer is:

[0088] z = f (W1 * x)

[0089] Here, * indicates matrix multiplication.

[0090] The predicted value of the output layer is:

[0091] y^ = g (W2 * z)

[0092] Through the training data, the weight matrix W1 and W2 are continuously adjusted using the back propagation algorithm to minimize the error between the predicted value y^ and the actual value y. Commonly used error functions include mean square error (MSE) and so on.

[0093] Through the above methods, the multidimensional adaptive prediction model can use different mathematical models and algorithms to accurately predict agricultural production risks. At the same time, combined with the multidimensional adaptive weight data selection module, it can better utilize the information from different data sources and improve the reliability and adaptability of the prediction.

[0094] It should be noted that:

[0095] In time series models:

[0096] Seasonal Decomposition Model:

[0097] Function: Decompose agricultural data (such as crop yields, prices, etc.) according to time series and extract trends, seasonality and residuals. By analyzing seasonal factors, we can better understand the impact of different seasons on agricultural production and predict the risks that may be faced in a specific season. For example, for some crops with obvious seasonality, the model can help determine the yield fluctuation range and possible risk factors in different seasons.

[0098] ARIMA Model:

[0099] Function: Use the autoregressive moving average model to analyze and predict data with time series characteristics. By analyzing the autocorrelation and moving average characteristics of historical data, predict future agricultural production indicators such as crop yields, price trends, etc. The model can capture trends, seasonality, and random fluctuations in the data, providing a basis for agricultural production decisions.

[0100] Regression Model:

[0101] Bayesian probability model:

[0102] Function: Based on Bayesian theory, the model combines prior probability and sample data to estimate the posterior probability. In the agricultural field, the model can be used to predict crop yields, the probability of pests and diseases, etc. By continuously updating the prior probability and introducing new data, the model can gradually improve the accuracy of the prediction, while taking into account uncertainty factors and providing more comprehensive information for risk assessment.

[0103] Rogers returns:

[0104] Function: Applicable to binary classification problems, and can be used to predict whether a certain event in agricultural production will occur, such as whether crops are affected by disasters, whether pests and diseases break out, etc. By analyzing the relationship between multiple influencing factors and the occurrence of events, a regression model is established to predict the probability of an event. This model can help farmers and agricultural managers take preventive measures in advance to reduce risks.

[0105] Inverse Neural Network:

[0106] Function: Inverse neural network can learn complex nonlinear relationships between input variables and output variables. In agriculture, multiple factors such as weather, soil, pests and diseases can be used as input, and crop yield and quality can be used as output to predict the results of agricultural production by training the inverse neural network model. The model has strong learning and generalization capabilities, can process high-dimensional data, and provide more accurate predictions for agricultural production risk assessment.

[0107] Machine Learning Models:

[0108] Decision Tree:

[0109] Function: The decision tree model can classify and predict agricultural production risks based on different characteristic variables. By analyzing agricultural data, a decision tree is constructed to determine whether crops are facing high risks. For example, a decision tree model can be established based on characteristics such as soil type, meteorological conditions, and the occurrence of pests and diseases, providing an intuitive basis for risk judgment for agricultural production decisions.

[0110] Random Forest:

[0111] Function: Random forest is an integrated model composed of multiple decision trees, which has higher accuracy and stability. By integrating a large number of decision trees, the overfitting risk of a single decision tree can be reduced and the ability to predict agricultural production risks can be improved. Random forest can process high-dimensional data and automatically select important feature variables to provide more reliable results for agricultural production risk assessment.

[0112] Multi-dimensional adaptive weight data selection

[0113] Function: This module is responsible for automatically selecting appropriate data sources and assigning corresponding weights according to different agricultural production scenarios and risk assessment requirements. By analyzing the reliability, timeliness and relevance of each data source, it provides the most valuable data combination for different forecasting models. For example, when predicting crop yields, the weights of data sources such as meteorological data, pest and disease data, and agricultural news may be adjusted according to factors such as the current season and geographical location to improve the accuracy of the forecast.

[0114] GIS meteorological data capture

[0115] Weekly average temperature:

[0116] Function: Collect weekly average temperature data in a specific area. Temperature has an important impact on the growth and development of crops. By monitoring the weekly average temperature, we can understand the temperature changes in the crop growth environment. For example, too high or too low temperature may affect the germination, growth rate and yield of crops. This data can be used as one of the important factors for predicting agricultural production risks.

[0117] Average weekly precipitation:

[0118] Function: Get the average weekly precipitation data in a specific area. Precipitation is another key factor in agricultural production, and the right amount of precipitation is essential for the growth of crops. By monitoring the weekly average precipitation, it is possible to determine whether there are risks such as drought or flooding. For example, too much precipitation may cause waterlogging in farmland and affect the respiration of crop roots, while too little precipitation may cause water shortage in crops and affect yield.

[0119] Historical data:

[0120] Function: Store and manage meteorological data over a period of time, including temperature, precipitation, sunlight, etc. Historical data can be used to analyze the changing trends and seasonal patterns of meteorological factors, and provide a reference for predicting future meteorological conditions. For example, by analyzing meteorological data from the same period in history, it is possible to predict possible meteorological risks in the current season, helping farmers and agricultural managers to prepare countermeasures in advance.

[0121] GIS Information:

[0122] Function: Utilize Geographic Information System (GIS) technology to integrate geospatial data and meteorological data. Through GIS information, the distribution of meteorological conditions in different regions can be intuitively displayed, providing spatial analysis support for agricultural production planning and risk assessment. For example, GIS information can be used to determine the types of crops suitable for planting in different regions, as well as to assess the meteorological disaster risks faced by specific regions.

[0123] GIS pest and disease data

[0124] GIS pest and disease information:

[0125] Function: Combined with GIS technology, collect and display the occurrence of pests and diseases in a specific area. Through GIS pest and disease information, you can intuitively understand the distribution of pests and diseases in different regions, and provide spatial analysis support for pest and disease monitoring and prevention. For example, you can determine the key prevention and control areas based on pest and disease information and formulate targeted prevention and control measures.

[0126] Historical data on pests and diseases:

[0127] Function: Store and manage pest and disease occurrence data over a period of time, including pest and disease type, occurrence time, location, degree of damage, etc. Historical data can be used to analyze the occurrence patterns and trends of pests and diseases, and provide a reference for predicting future pest and disease risks. For example, by analyzing pest and disease data from the same period in history, it is possible to predict the types and severity of pests and diseases that may occur in the current season, helping farmers and agricultural managers to prepare for prevention and control in advance.

[0128] NLP Agriculture News

[0129] Function: Use natural language processing (NLP) technology to analyze and extract key information from agricultural news. By analyzing agricultural news, you can understand current agricultural policies, market trends, technological innovations, and other information, and provide reference for agricultural production decisions. For example, policy changes may affect the price and sales channels of agricultural products, changes in market demand may affect the variety and scale of crop planting, and technological innovation may improve agricultural production efficiency and reduce risks.

[0130] NLP public opinion analysis

[0131] Function: Analyze agriculture-related public opinions on social media, forums and other channels through NLP technology. Understand the public's focus, emotional tendencies, opinions and suggestions on agricultural issues. For example, if it is found that the public pays more attention to the quality and safety of a certain agricultural product, it may mean that the agricultural product faces certain market risks and needs to strengthen quality supervision and publicity guidance. Public opinion analysis can also help agricultural managers understand farmers' needs and problems in a timely manner and provide reference for policy formulation and service provision.

[0132] Hundreds of rows of data

[0133] Guarantor risk assessment:

[0134] Function: In agricultural loan scenarios, the guarantor's risk is assessed. By analyzing the guarantor's credit record, financial status, career stability and other factors, the reliability and guarantee ability of the guarantor can be judged. For example, if the guarantor has a bad credit record or unstable financial status, the risk of the loan may increase. This module can provide decision support for agricultural financial institutions and reduce loan risks.

[0135] The above technical solutions only reflect the preferred technical solutions of the technical solutions of the present invention. Some changes that may be made to certain parts thereof by technicians in this technical field all reflect the principles of the present invention and fall within the protection scope of the present invention.

Claims

1. The risk control model for agricultural financing guarantee based on Zhibao multi-dimensional dynamic scoring is characterized by: It includes a multi-dimensional adaptive prediction model and a multi-dimensional adaptive weight data selection module. The main purpose of the multi-dimensional adaptive weight data selection module is to assign reasonable weights to various data sources according to different agricultural production scenarios and risk assessment requirements; First, for each data source, a data quality assessment is performed; Next, the correlation between various data sources and agricultural production risks is analyzed; through statistical analysis and machine learning algorithms, the degree of influence of different data sources on the prediction target is determined; Then, consider the timeliness factor; newer data usually better reflects the current agricultural production situation and is therefore given a higher weight; Finally, the weights are adjusted dynamically according to different agricultural production scenarios and risk assessment needs; The multi-dimensional adaptive forecasting model combines a variety of forecasting methods, including time series models, regression models, and machine learning models; Time series models include seasonal decomposition models and ARIMA models, which are mainly used to analyze data with time series characteristics; Regression models include Bayesian probability model, Rogers regression and reverse neural network, which are used to establish the relationship between input variables and output variables; Machine learning models include decision trees and random forests.

2. The risk control model for agricultural financing guarantee based on intelligent insurance multi-dimensional dynamic scoring according to claim 1 is characterized in that: In the multi-dimensional adaptive weight data selection module, it is assumed that we have n data sources, namely D1, D2, ..., Dn; for each data source Di, we define its weight as wi, and satisfy ∑wi=1; First, calculate the reliability index Ri of the data source Di. The value of Ri can be determined by data quality assessment and historical data accuracy verification. For example, the accuracy and recall rate of the data in the data source Di can be calculated, and then these indicators can be combined into a reliability index Ri. Next, the correlation index Ci between the data source Di and the prediction target Y is calculated; the value of Ci can be determined using a correlation coefficient or a regression analysis method; for example, the Pearson correlation coefficient between the data in the data source Di and the prediction target Y can be calculated as the correlation index Ci; Then, considering the timeliness factor, define the timeliness index Ti of the data source Di; the value of Ti can be determined according to the timeliness and update frequency of the data; for example, the timeliness index of recent data can be set to a higher value, and as time goes by, the timeliness index gradually decreases; Finally, considering reliability, relevance and timeliness, the weight wi of the data source Di is calculated; the following formula can be used: wi=α*Ri+β*Ci+γ*Ti; Among them, α, β, and γ are the weight coefficients of reliability, relevance, and timeliness, respectively, which can be adjusted according to specific application scenarios and requirements.

3. The risk control model for agricultural financing guarantee based on intelligent insurance multi-dimensional dynamic scoring according to claim 1 is characterized in that: In the multi-dimensional adaptive prediction model, random forest is taken as an example. Random forest is an integrated model composed of multiple decision trees. Assume that we have a training data set X={x1, x2,..., xm}, where xi is a d-dimensional input vector and the corresponding output is yi; First, a portion of sample data is randomly selected as a training set to build multiple decision trees; for each decision tree, a portion of features is randomly selected from the training set for splitting until a certain stopping condition is reached; Then, for a new input vector x, each decision tree will give a prediction value y^i; the final prediction result is the average of the prediction values ​​of all decision trees: y^= (1 / N) *∑y^i; Where N is the number of decision trees, y^i is the predicted value of the i-th decision tree; For the reverse neural network, suppose we have an input layer, a hidden layer, and an output layer; the input layer has d neurons, corresponding to the d features of the input vector x; the hidden layer has h neurons; the output layer has one neuron, corresponding to the predicted target y; First, define the weight matrix W1 from the input layer to the hidden layer, with a size of h×d; the weight matrix W2 from the hidden layer to the output layer, with a size of 1×h; the activation function of the hidden layer is f, and the activation function of the output layer is g; For an input vector x, the output of the hidden layer is: z=f(W1*x); Where * represents matrix multiplication; The predicted value of the output layer is: y^=g(W2*z); Through training data, the back propagation algorithm is used to continuously adjust the weight matrices W1 and W2 to minimize the error between the predicted value y^ and the actual value y. The commonly used error function is the mean square error (MSE).

4. The risk control model for agricultural financing guarantee based on intelligent insurance multi-dimensional dynamic scoring according to claim 1 is characterized in that: The seasonal decomposition model decomposes agricultural-related data according to time series and extracts the trend, seasonality and residual parts therein; by analyzing seasonal factors, we can better understand the impact of different seasons on agricultural production and predict the risks that may be faced in a specific season.

5. The risk control model for agricultural financing guarantee based on intelligent insurance multi-dimensional dynamic scoring according to claim 1 is characterized in that: The ARIMA model uses an autoregressive moving average model to analyze and predict data with time series characteristics; by analyzing the autocorrelation and moving average characteristics of historical data, future agricultural production indicators are predicted.

6. The risk control model for agricultural financing guarantee based on intelligent insurance multi-dimensional dynamic scoring according to claim 1 is characterized in that: The Bayesian probability model, based on Bayesian theory, combines prior probability and sample data to estimate posterior probability; The Rogers regression is applicable to binary classification problems and can be used to predict whether a certain event in agricultural production will occur; The reverse neural network can learn the complex nonlinear relationship between input variables and output variables; Decision tree, the decision tree model can classify and predict agricultural production risks based on different characteristic variables; Random Forest,Random Forest is an integrated model composed of multiple decision trees.

7. The risk control model for agricultural financing guarantee based on intelligent insurance multi-dimensional dynamic scoring according to claim 1 is characterized in that: Also includes: GIS meteorological data capture, GIS pest and disease data, NLP agricultural news, NLP public opinion analysis, and hundreds of lines of data.

8. The risk control model for agricultural financing guarantee based on intelligent insurance multi-dimensional dynamic scoring according to claim 1 is characterized in that: In GIS meteorological data capture, the average temperature data and average precipitation data of a specific area are collected every week. The meteorological data of the past period of time, including temperature, precipitation, and light, are stored and managed. Geographic Information System (GIS) technology is used to integrate geographic spatial data and meteorological data.

9. The risk control model for agricultural financing guarantee based on intelligent insurance multi-dimensional dynamic scoring according to claim 1 is characterized in that: In GIS pest and disease data, GIS technology is combined to collect and display the occurrence of pests and diseases in a specific area, and to store and manage pest and disease occurrence data over a period of time, including pest and disease type, occurrence time, location, and degree of damage.

10. The risk control model for agricultural financing guarantee based on intelligent insurance multi-dimensional dynamic scoring according to claim 1 is characterized in that: NLP agricultural news, using natural language processing (NLP) technology to analyze and extract key information from agriculture-related news; NLP public opinion analysis: using NLP technology to analyze agriculture-related public opinions on social media and forum channels; Baihang Data evaluates the risk of guarantors in agricultural loan scenarios. It determines the reliability and guarantee ability of guarantors by analyzing their credit records, financial status, and career stability.

Images