Rural financial risk real-time monitoring and early warning system

Abstract

The invention discloses a rural financial risk real-time monitoring and early warning system, which relates to the technical field of risk early warning and comprises an evaluation score statistical unit, a prediction model analysis unit, an initial individual screening unit, a hyper-parameter optimization unit, a hyper-parameter determination unit and a risk identification unit. The risk identification unit is used to acquire the price and the yield of the specified crop in the prediction period to evaluate the repayment capability of the farmer, the farmer is informed of the improvement of the risk level in time when the repayment capability is insufficient in multiple days, early warning is directly given to the superior when the repayment capability is seriously insufficient, the actual condition is verified manually, and the repayment efficiency is improved. Management personnel can conveniently take measures in advance for intervention.

Description

Technical Field

[0001] The present invention relates to the technical field of risk early warning, and in particular to a rural financial risk real-time monitoring and early warning system. Background Art

[0002] A real-time monitoring and early warning system for rural financial risks monitors rural financial risks in real time, integrates agricultural product evaluation, price and output data, accurately assesses farmers' repayment ability, issues real-time early warnings for risks, helps managers intervene in a timely manner, and ensures rural financial stability. The invention patent with application number 202311463818.2 discloses "a rural financial service system based on basic credit data, including: an agricultural credit platform and an external access front-end module, the external access front-end module is connected to the rural land information system and the loan information system; the agricultural credit platform includes: a management module and a back-end module, the management module is connected to the external access front-end module through the back-end module; the management module includes: a system public management module, a loan parameter management module, a loan management module, a report management module, a monitoring management module and a log management module, the system public management module is used for querying and maintaining public information; the loan parameter management module is used to manage the elements of loan products; the loan management module is used for loan evaluation model maintenance, land mortgage management, land mortgage release management, and loan repayment management. The present invention can provide efficient and convenient rural financial services."

[0003] The above-mentioned existing technologies solve the problem that some archived credit information data cannot be updated in a timely manner due to the lack of information and data on farmers and rural credit. However, when the system is running, it is impossible to quantify the impact of online evaluations on agricultural product prices through evaluation scores, and because the credibility differences of various platforms are not taken into account, the reference value of the evaluation data is low, which may mislead price judgments. At the same time, the model hyperparameters are not optimized, the accuracy of the price prediction model is poor, and there is a lack of risk identification mechanism. It is impossible to combine prices and output to evaluate farmers' repayment ability, and it is difficult to detect repayment risks in advance. Farmers cannot be reminded in time, and management personnel cannot intervene in advance, which may lead to farmer defaults, economic losses and other problems. Risk response is delayed and passive. Summary of the Invention

[0004] The purpose of the present invention is to provide a rural financial risk real-time monitoring and early warning system to solve the problems raised in the above background technology.

[0005] To achieve the above-mentioned object, the present invention provides the following technical solutions: a rural financial risk real-time monitoring and early warning system, comprising an evaluation score statistics unit, a prediction model analysis unit, an initial individual screening unit, a hyperparameter optimization unit, a hyperparameter determination unit, and a risk identification unit;

[0006] An initial individual screening unit, which counts all hyperparameters to be optimized for the XGBoost model and the LSTM model, determines the constraint range of each hyperparameter to be optimized, sets the number of individuals, constructs an individual sequence based on the constraint range and the number of individuals, determines the accuracy of different individuals, screens individuals according to the accuracy and deletes them, calculates the corresponding derived individuals using the remaining individuals, and reconstructs a new individual sequence. After counting the accuracies corresponding to the individuals of different generation methods in the three ways, the final initial individuals are selected according to the accuracy.

[0007] A hyperparameter optimization unit, wherein the hyperparameter optimization unit calculates the updated individuals before and after adding the mutation factor using the hyperparameters of the initial individuals after statistically analyzing the phase factors and transformation coefficients, calculates the accuracy of the original individuals and the updated individuals, and replaces the hyperparameters of the original individuals with the hyperparameters within the individuals with higher accuracy;

[0008] A risk identification unit uses an XGBoost model and an LSTM model to analyze historical price information to obtain a daily forecast price for a specified crop within a forecast period, calculates the farmer's daily repayment ability based on the forecast price, and determines the risk level using a preset repayment amount and repayment ability.

[0009] Preferably, the evaluation score statistics unit includes an evaluation data acquisition module, a word segmentation feature extraction module, a label information determination module, a weight setting module, a comprehensive score calculation module and an actual score output module. The evaluation data acquisition module uses a crawler tool to obtain the evaluation data of the specified crop, wherein the evaluation data includes text content, number of visits, time and data source platform, and adds corresponding label information to each review data, wherein the label information is divided into two types, namely positive evaluation and negative evaluation. The word segmentation feature extraction module uses the evaluation data to construct multiple samples, extracts the word segmentation features of each sample, and then transmits the word segmentation features and label information of the sample to the naive Bayes classification model to determine the model parameters and accuracy. The label information determination module obtains all evaluation data of the specified crop on the same day, extracts the word segmentation features of each evaluation data, and transmits it to the naive Bayes classification model for analysis to determine the label information of the evaluation data. The weight setting module counts all official data source platforms and unofficial data source platforms, sets a corresponding weight coefficient for each source platform, stores it in the database, reads the source platform of each evaluation data, and queries the database according to the source platform. If the query is successful, the weight coefficient corresponding to the current platform is output. If the query fails, the current source platform is determined to be an unknown platform, and the default weight coefficient is used as the weight coefficient of the unknown platform. The comprehensive score calculation module determines the source platform. Number of visits to all reviews with positive reviews tags and the number of visits to reviews with negative review tags ,in Representation Platform The number of messages with positive evaluation labels, Representation Platform The number of negative evaluation labels, according to and Calculate the platform's comprehensive score ,in The actual score output module calculates the comprehensive scores of all source platforms And the corresponding weight coefficient Afterwards, according to and Calculate the actual evaluation score of the specified crop ,in , Indicates parameters, Indicates the total number of source platforms.

[0010] Preferably, the prediction model analysis unit includes a historical price acquisition module and a model generation module. After acquiring historical price information on the national agricultural product market price information service platform, the historical price acquisition module reads the actual historical evaluation score of the specified crop, performs data dimension reduction on the attribute representations contained in the historical price information and the actual evaluation score, and then performs standardization processing on them. The processed attribute representations are used to construct multiple samples, and the samples are classified according to the model generation module. The ratio is stored in the training set and the test set respectively. The model generation module transmits the sample data in the training set to the XGBoost model and the LSTM model for analysis. After the model training is completed, the accuracy of the combined XGBoost model and the LSTM model is determined by the sample data in the test set. The historical price information includes the daily price, output, temperature, extreme weather impact index, per capita disposable income of residents, transportation cost, food price index, import amount, export amount and e-commerce logistics index of target agricultural products.

[0011] Preferably, the initial individual screening unit includes a label adding module and an individual generating module. After the label adding module counts all the hyperparameters to be optimized in the XGBoost model and the LSTM model, it sets an update cycle and divides the hyperparameters to be optimized into two categories, namely high-frequency hyperparameters and low-frequency hyperparameters, and labels them, wherein the high-frequency hyperparameters are labeled as 1 and the low-frequency hyperparameters are labeled as 0. The individual generating module determines the constraint range of each hyperparameter to be optimized, and sets the number of individuals and the maximum number of cycles, and constructs an individual sequence according to the constraint range and the number of individuals. ,in represents the number of individuals in the initial sequence, , Indicates the Individuals, Indicates the Intra-individual hyperparameters, Indicates the Intra-individual hyperparameters, represents the total number of hyperparameters, represents the number of individual numbers in the initial sequence, , Indicates the The lower limit of the constraint range corresponding to the hyperparameter, Indicates the The upper limit of the constraint range corresponding to the hyperparameters, among which the hyperparameters to be optimized for the XGBoost model include the learning rate, sample sampling ratio, maximum tree depth, feature sampling ratio and the number of trees; the hyperparameters to be optimized for the LSTM model include the number of neurons, learning rate, dropout rate and number of stacking layers; the joint hyperparameters of the XGBoost model and LSTM model include the LSTM output weight ratio, among which the high-frequency hyperparameters include the learning rate, sample sampling ratio of the XGBoost model, the learning rate, dropout rate and LSTM output weight ratio of the LSTM model; the low-frequency hyperparameters include the maximum tree depth, feature sampling ratio, the number of trees, the number of neurons and the number of stacking layers of the XGBoost model.

[0012] Preferably, the initial individual screening unit also includes a sorting and screening module and an initial individual determination module. The sorting and screening module inputs the corresponding hyperparameters of each individual into the XGBoost model and the LSTM model in sequence, uses the samples in the training set and the test set to determine the accuracy corresponding to different individuals, sorts the individuals according to the accuracy, and calculates the accuracy mean of the current individual sequence, deletes the individuals with accuracy lower than the mean, and the initial individual determination module obtains the current individual sequence. ,in Indicates the Individuals, Indicates the individual sequence number in the current sequence, Indicates the number of individuals in the current sequence, , Indicates the Intra-individual hyperparameters, Indicates the Intra-individual Hyperparameters, setting the spatial coefficient ,according to and Calculate the derived individual ,in , , Indicates the The first derivative individual hyperparameters, Represents a random number, repeating the operation until all individuals in the current sequence have generated derivative individuals, and reconstructing multiple different individuals according to the constraint range and number of individuals ,in , Represents the reconstructed Intra-individual hyperparameters, ,statistics 、 and After the corresponding accuracy, sort by accuracy, and put the first individuals as the initial individuals and delete the other individuals.

[0013] Preferably, the hyperparameter optimization unit includes an update cycle judgment module, a coefficient calculation module, a coefficient combination module and a preliminary update module. The update cycle judgment module divides the initial individuals into two groups, stores them in the first set and the second set respectively, records the current cycle number, and judges whether the current cycle number is a multiple of the update cycle. If it is not a multiple of the update cycle, all initial individuals are counted. The hyperparameters marked as 1 in ,in Indicates the ordinal number. If it is a multiple of the update period, all initial individuals are counted. The coefficient calculation module calculates the phase factor and the transformation coefficient according to the current number of cycles and the maximum number of cycles. If the phase factor is less than the first threshold, the coefficient calculation module calculates the phase factor and the transformation coefficient according to the hyperparameter and constraint range Calculate the updated hyperparameters when no mutation factor is added , repeat until all hyperparameters have been calculated, where , Indicates the first The value of the hyperparameter, Indicates A random number selected from , Indicates the The lower limit of the constraint range corresponding to the hyperparameter, Indicates the The upper limit of the constraint range corresponding to the hyperparameter is determined by combining the coefficient with the module to determine the weight coefficient corresponding to the current number of cycles. After adding the mutation factor, the parameter optimization algorithm is used to combine the mutation factor, weight coefficient and Combine to get the updated hyperparameters when adding mutation factors , repeat the operation until all hyperparameters have been calculated, the initial update module counts the accuracy of the original individual, the updated individual without adding the mutation factor and the updated individual with adding the mutation factor, and replaces the hyperparameters of the original individual with the hyperparameters within the individual with high accuracy.

[0014] Preferably, the hyperparameter determination unit includes a mean analysis module, an optimal individual screening module, a two-round updating module, a parameter fusion module and an individual output module. If the stage factor of the mean analysis module is greater than or equal to the first threshold and the transformation coefficient is greater than the second threshold, the similarity between each individual and the other individuals is calculated, and the individual with a similarity greater than a preset value is used as a candidate individual of the current individual. The accuracy corresponding to all candidate individuals is counted, and the candidate individuals with an accuracy lower than the current individual are deleted. The remaining candidate individuals and the current individual are used to calculate the mean of each hyperparameter, and an updated individual is constructed by the mean. The optimal individual screening module selects the individual with the highest current accuracy and uses it as the optimal individual. The corresponding individual of each individual is fine-tuned by the parameter fine-tuning algorithm. The hyperparameters, hyperparameters of the best individual and the transformation coefficient are calculated to obtain an updated individual. The two-round update module compares the accuracy of the original individual and the two different updated individuals, and replaces the hyperparameters of the original individual with the hyperparameters within the individual with high accuracy. If the stage factor is greater than or equal to the first threshold and the transformation coefficient is less than or equal to the second threshold, the parameter fusion module determines the individual with the highest accuracy in the first set and the second set, and uses the parameter fusion algorithm to calculate the individual with the highest accuracy in the first set and the second set to obtain new hyperparameters for each individual. The operation is repeated until the end of the cycle. The individual output module outputs the individual with the highest accuracy, and its corresponding hyperparameters are input into the XGBoost model and the LSTM model as the optimal hyperparameters.

[0015] Preferably, the risk identification unit includes a daily price generation module, a capacity calculation module and an early warning execution module. The daily price generation module obtains the maximum daily yield and the minimum daily yield of the specified crops planted by the current farmer within the forecast period, and transmits the historical price information and the actual evaluation score of the specified crops to the XGBoost model and the LSTM model for analysis to obtain the daily forecast price of the specified crops within the forecast period. The capacity calculation module uses the maximum daily yield, the minimum daily yield and the daily forecast price to calculate the daily repayment capacity of the current farmer for planting the specified crops. The early warning execution module divides the risk level into high risk, medium risk and low risk, reads the daily repayment amount preset by the current farmer, and if the daily repayment amount within the forecast period is less than or equal to the daily repayment capacity, it is judged that the current risk level is a low risk level and no early warning operation is performed. If the number of times the daily repayment amount exceeds the repayment capacity within the forecast period is greater than one and less than four times, it is judged that the current risk level is a medium risk level and an early warning prompt message is issued to the farmer. If the number of times the daily repayment amount exceeds the repayment capacity within the forecast period is greater than four times, the manual review mechanism is activated.

[0016] Compared with the prior art, the present invention has the following beneficial effects:

[0017] After obtaining the evaluation data of the specified crops through the evaluation score statistics unit, the present invention calculates the actual evaluation score of the specified crops according to the number of positive and negative evaluations. This design can fully reflect the impact of online evaluations on agricultural product prices, and the evaluation credibility of different source platforms is also different. A corresponding weight coefficient is set for each platform, so that the final actual evaluation score can be more reasonable and reliable. The initial individual screening unit, the hyperparameter optimization unit and the hyperparameter determination unit are used to optimize the various hyperparameters in the XGBoost model and the LSTM model to ensure that the obtained hyperparameters are the optimal hyperparameters. During the analysis process, individuals are continuously updated and screened to avoid individuals with low accuracy from occupying space for a long time, thereby further improving the operating speed. The risk identification unit is used to obtain the price and yield of the specified crops within the forecast period to evaluate the repayment ability of farmers. When the repayment ability is insufficient for many days, the farmers are promptly informed of the increased risk level, and when the repayment ability is seriously insufficient, a warning will be directly issued to the superiors. The actual situation is manually verified, which facilitates management personnel to take measures to intervene in advance. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 A schematic diagram of the overall system flow is provided for an embodiment of the present invention. DETAILED DESCRIPTION

[0019] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0020] Example:

[0021] See also Figure 1 ,The present invention provides a technical solution: a rural financial risk real-time monitoring and early warning system, including an evaluation score statistics unit, a prediction model analysis unit, an initial individual screening unit, a hyperparameter optimization unit, a hyperparameter determination unit and a risk identification unit;

[0022] The initial individual screening unit counts all the hyperparameters to be optimized for the XGBoost model and the LSTM model, determines the constraint range of each hyperparameter to be optimized, and sets the number of individuals. It constructs an individual sequence based on the constraint range and the number of individuals, determines the accuracy of different individuals, and screens individuals for deletion based on the accuracy. It uses the remaining individuals to calculate the corresponding derived individuals and reconstructs a new individual sequence. After counting the corresponding accuracies of individuals generated in different ways in the three ways, it selects the final initial individuals based on the accuracy.

[0023] The hyperparameter optimization unit calculates the phase factors and transformation coefficients, uses the hyperparameters of the initial individuals to calculate the updated individuals with and without the mutation factor, calculates the accuracy of the original individuals and the updated individuals, and replaces the hyperparameters of the original individuals with the hyperparameters of the individuals with higher accuracy.

[0024] Risk identification unit: The risk identification unit uses the XGBoost model and LSTM model to analyze historical price information to obtain the daily predicted price of specified crops within the prediction period. Based on the predicted price, it calculates the farmer's daily repayment ability and determines the risk level using the preset repayment amount and repayment ability.

[0025] The evaluation score statistics unit includes an evaluation data acquisition module, a word segmentation feature extraction module, a label information determination module, a weight setting module, a comprehensive score calculation module and an actual score output module. The evaluation data acquisition module uses a crawler tool to obtain the evaluation data of specified crops, where the evaluation data includes text content, visits, time and data source platform, and adds corresponding label information to each review data. The label information is divided into two types, namely positive evaluation and negative evaluation. The word segmentation feature extraction module uses the evaluation data to construct multiple samples. After extracting the word segmentation features of each sample, the word segmentation features and label information of the sample are transmitted to the naive Bayes classification model to determine the model parameters and accuracy. The information determination module obtains all evaluation data of the specified crop on the same day, extracts the word segmentation features of each evaluation data, and transmits it to the naive Bayes classification model for analysis to determine the label information of the evaluation data. The weight setting module counts all official data source platforms and unofficial data source platforms, sets the corresponding weight coefficient for each source platform, stores it in the database, reads the source platform of each evaluation data, and queries the database according to the source platform. If the query is successful, the weight coefficient corresponding to the current platform is output. If the query fails, the current source platform is determined to be an unknown platform, and the default weight coefficient is used as the weight coefficient of the unknown platform. The comprehensive score calculation module determines the source platform Number of visits to all reviews with positive reviews tags and the number of visits to reviews with negative review tags ,in Representation Platform The number of messages with positive evaluation labels, Representation Platform The number of negative evaluation labels, according to and Calculate the platform's comprehensive score ,in The actual score output module calculates the comprehensive score of all source platforms And the corresponding weight coefficient Afterwards, according to and Calculate the actual evaluation score of the specified crop ,in , Indicates parameters, Indicates the total number of source platforms;

[0026] The prediction model analysis unit includes a historical price acquisition module and a model generation module. The historical price acquisition module obtains historical price information from the national agricultural product market price information service platform, reads the historical evaluation scores of the specified crops, performs data dimension reduction on the attribute representations contained in the historical price information and the evaluation scores, and then standardizes them. The processed attribute representations are used to construct multiple samples and are divided into two groups according to the model generation module. The ratio of the two models is stored in the training set and the test set respectively. The model generation module transmits the sample data in the training set to the XGBoost model and the LSTM model for analysis. After the model training is completed, the accuracy of the combined XGBoost model and the LSTM model is determined by the sample data in the test set. The historical price information includes the daily price, output, temperature, extreme weather impact index, per capita disposable income of residents, transportation cost, food price index, import amount, export amount and e-commerce logistics index of the target agricultural products;

[0027] The initial individual screening unit includes a label adding module and an individual generation module. After the label adding module counts all the hyperparameters to be optimized in the XGBoost model and the LSTM model, it sets the update cycle and divides the hyperparameters to be optimized into two categories, namely high-frequency hyperparameters and low-frequency hyperparameters, and labels them. The high-frequency hyperparameters are labeled as 1 and the low-frequency hyperparameters are labeled as 0. The individual generation module determines the constraint range of each hyperparameter to be optimized, and sets the number of individuals and the maximum number of cycles. The individual sequence is constructed according to the constraint range and the number of individuals. ,in represents the number of individuals in the initial sequence, , Indicates the Individuals, Indicates the Intra-individual hyperparameters, Indicates the Intra-individual hyperparameters, represents the total number of hyperparameters, represents the number of individual numbers in the initial sequence, , Indicates the The lower limit of the constraint range corresponding to the hyperparameter, Indicates the The upper limit of the constraint range corresponding to the hyperparameters, among which the hyperparameters to be optimized for the XGBoost model include the learning rate, sample sampling ratio, maximum tree depth, feature sampling ratio and number of trees; the hyperparameters to be optimized for the LSTM model include the number of neurons, learning rate, dropout rate and number of stacking layers; the joint hyperparameters of the XGBoost model and LSTM model include the LSTM output weight ratio, among which the high-frequency hyperparameters include the learning rate, sample sampling ratio of the XGBoost model, the learning rate, dropout rate and LSTM output weight ratio of the LSTM model; the low-frequency hyperparameters include the maximum tree depth, feature sampling ratio, number of trees, number of neurons and number of stacking layers of the XGBoost model;

[0028] The initial individual screening unit also includes a sorting and screening module and an initial individual determination module. The sorting and screening module inputs the corresponding hyperparameters of each individual into the XGBoost model and the LSTM model in sequence, uses the samples in the training set and the test set to determine the accuracy of different individuals, sorts the individuals according to the accuracy, and calculates the accuracy mean of the current individual sequence. Individuals with accuracy lower than the mean are deleted. The initial individual determination module obtains the current individual sequence. ,in Indicates the Individuals, Indicates the individual sequence number in the current sequence, Indicates the number of individuals in the current sequence, , Indicates the Intra-individual hyperparameters, Indicates the Intra-individual Hyperparameters, setting the spatial coefficient ,according to and Calculate the derived individual ,in , , Indicates the The first derivative individual hyperparameters, Represents a random number, repeating the operation until all individuals in the current sequence have generated derivative individuals, and reconstructing multiple different individuals according to the constraint range and number of individuals ,in , Represents the reconstructed Intra-individual hyperparameters, ,statistics 、 and After the corresponding accuracy, sort by accuracy, and put the first individuals as the initial individuals and delete other individuals;

[0029] The hyperparameter optimization unit includes an update cycle judgment module, a coefficient calculation module, a coefficient combination module and an initial round update module. The update cycle judgment module divides the initial individuals into two groups, stores them in the first set and the second set respectively, records the current cycle number, and judges whether the current cycle number is a multiple of the update cycle. If it is not a multiple of the update cycle, all initial individuals are counted. The hyperparameters marked as 1 in ,in Indicates the ordinal number. If it is a multiple of the update period, all initial individuals are counted. The coefficient calculation module calculates the phase factor and transformation coefficient according to the current number of cycles and the maximum number of cycles. If the phase factor is less than the first threshold, the coefficient calculation module calculates the phase factor and transformation coefficient according to the hyperparameters. and constraint range Calculate the updated hyperparameters when no mutation factor is added , repeat until all hyperparameters have been calculated, where , Indicates the first The value of the hyperparameter, Indicates A random number selected from , Indicates the The lower limit of the constraint range corresponding to the hyperparameter, Indicates the The upper limit of the constraint range corresponding to the hyperparameter is determined by combining the coefficient module to determine the weight coefficient corresponding to the current number of cycles. After adding the mutation factor, the parameter optimization algorithm is used to combine the mutation factor, weight coefficient and Combine to get the updated hyperparameters when adding mutation factors Repeat the operation until all hyperparameters have been calculated. The initial update module counts the accuracy of the original individual, the updated individual without adding the mutation factor, and the updated individual with adding the mutation factor. The hyperparameters of the original individual are replaced with the hyperparameters of the individual with high accuracy. The parameter optimization algorithm is as follows:

[0030]

[0031] in, represents the updated hyperparameters when adding mutation factors, represents the value of the randomly selected intra-individual hyperparameter, represents a random number, Indicates the current number of cycles. Indicates the maximum number of cycles, represents the original hyperparameters, represents the weight coefficient;

[0032] The hyperparameter determination unit includes a mean analysis module, an optimal individual screening module, a second-round updating module, a parameter fusion module and an individual output module. If the stage factor of the mean analysis module is greater than or equal to the first threshold and the transformation coefficient is greater than the second threshold, the similarity between each individual and the other individuals is calculated, and the individuals with similarity greater than the preset value are taken as candidates for the current individual. The accuracy corresponding to all candidate individuals is counted, and the candidate individuals with accuracy lower than the current individual are deleted. The remaining candidate individuals and the current individual are used to calculate the mean of each hyperparameter, and the updated individual is constructed by the mean. The optimal individual screening module selects the individual with the highest current accuracy and takes it as the optimal individual. The hyperparameters and the optimal individual corresponding to each individual are adjusted by the parameter fine-tuning algorithm. The hyperparameters and transformation coefficients of the individual are calculated to obtain the updated individual. The second-round update module compares the accuracy of the original individual and the two different updated individuals, and replaces the hyperparameters of the original individual with the hyperparameters of the individual with higher accuracy. If the stage factor is greater than or equal to the first threshold and the transformation coefficient is less than or equal to the second threshold, the parameter fusion module determines the individual with the highest accuracy in the first set and the second set. The parameter fusion algorithm is used to calculate the individuals with the highest accuracy in the first set and the second set to obtain new hyperparameters for each individual. The operation is repeated until the end of the loop. The individual output module outputs the individual with the highest accuracy, and its corresponding hyperparameters are input as the optimal hyperparameters into the XGBoost model and the LSTM model. The parameter fine-tuning algorithm is as follows:

[0033]

[0034] in, represents the update of hyperparameters, represents the hyperparameters of the best individual, represents a random number, represents the transformation coefficient, represents the original hyperparameters;

[0035] The parameter fusion algorithm is specifically as follows:

[0036]

[0037] in, represents the updated hyperparameters of individuals in the specified set, represents the best individual hyperparameters within the specified set, represents the original hyperparameters, represents a random number, represents the adaptation coefficient, represents the stage factor;

[0038] The risk identification unit includes a daily price generation module, a capacity calculation module, and an early warning execution module. The daily price generation module obtains the maximum and minimum daily yields of the specified crops planted by the current farmer during the forecast period, and transmits the historical price information and actual evaluation scores of the specified crops to the XGBoost model and LSTM model for analysis to obtain the daily forecast price of the specified crops during the forecast period. The capacity calculation module uses the maximum and minimum daily yields and daily forecast prices to calculate the daily repayment capacity of the current farmer for planting the specified crops. The early warning execution module divides the risk level into high risk, medium risk, and low risk, and reads the daily repayment amount preset by the current farmer. If the daily repayment amount during the forecast period is less than or equal to the daily repayment capacity, the current risk level is judged to be low risk and no early warning operation is executed. If the daily repayment amount exceeds the repayment capacity more than once and less than four times during the forecast period, the current risk level is judged to be medium risk and an early warning prompt message is issued to the farmer. If the daily repayment amount exceeds the repayment capacity more than four times during the forecast period, the manual review mechanism is activated.

[0039] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0040] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A real-time monitoring and early warning system for rural financial risks, comprising an evaluation score statistics unit, a prediction model analysis unit, an initial individual screening unit, a hyperparameter optimization unit, a hyperparameter determination unit, and a risk identification unit, characterized in that: An initial individual screening unit, which counts all hyperparameters to be optimized for the XGBoost model and the LSTM model, determines the constraint range of each hyperparameter to be optimized, sets the number of individuals, constructs an individual sequence based on the constraint range and the number of individuals, determines the accuracy of different individuals, screens individuals according to the accuracy and deletes them, calculates the corresponding derived individuals using the remaining individuals, and reconstructs a new individual sequence. After counting the accuracies corresponding to the individuals of different generation methods in the three ways, the final initial individuals are selected according to the accuracy. A hyperparameter optimization unit, wherein the hyperparameter optimization unit calculates the updated individuals before and after adding the mutation factor using the hyperparameters of the initial individuals after statistically analyzing the phase factors and transformation coefficients, calculates the accuracy of the original individuals and the updated individuals, and replaces the hyperparameters of the original individuals with the hyperparameters within the individuals with higher accuracy; A risk identification unit uses an XGBoost model and an LSTM model to analyze historical price information to obtain a daily forecast price for a specified crop within a forecast period, calculates the farmer's daily repayment ability based on the forecast price, and determines the risk level using a preset repayment amount and repayment ability.

2. A rural financial risk real-time monitoring and early warning system according to claim 1, characterized in that: The evaluation score statistics unit includes an evaluation data acquisition module, a word segmentation feature extraction module, a label information determination module, a weight setting module, a comprehensive score calculation module and an actual score output module. The evaluation data acquisition module uses a crawler tool to obtain the evaluation data of the specified crop, wherein the evaluation data includes text content, number of visits, time and data source platform, and adds corresponding label information to each review data, wherein the label information is divided into two types, namely positive evaluation and negative evaluation. The word segmentation feature extraction module uses the evaluation data to construct multiple samples, extracts the word segmentation features of each sample, and transmits the word segmentation features and label information of the sample to the naive Bayes classification model to determine the model parameters and accuracy. The label information determination module obtains all evaluation data of the specified crop on the same day, extracts the word segmentation features of each evaluation data, and transmits it to the naive Bayes classification model for analysis to determine the label information of the evaluation data. The weight setting module counts all official data source platforms and unofficial data source platforms, sets a corresponding weight coefficient for each source platform, stores it in the database, reads the source platform of each evaluation data, and queries the database according to the source platform. If the query is successful, the weight coefficient corresponding to the current platform is output. If the query fails, the current source platform is determined to be an unknown platform, and the default weight coefficient is used as the weight coefficient of the unknown platform. The comprehensive score calculation module determines the source platform Number of visits to all reviews with positive reviews tags and the number of visits to reviews with negative review tags ,in Representation Platform The number of messages with positive evaluation labels, Representation Platform The number of negative evaluation labels, according to and Calculate the platform's comprehensive score ,in The actual score output module calculates the comprehensive scores of all source platforms And the corresponding weight coefficient Afterwards, according to and Calculate the actual evaluation score of the specified crop ,in , Indicates parameters, Indicates the total number of source platforms.

3. A rural financial risk real-time monitoring and early warning system according to claim 1, characterized in that: The prediction model analysis unit includes a historical price acquisition module and a model generation module. After obtaining historical price information on the national agricultural product market price information service platform, the historical price acquisition module reads the actual historical evaluation score of the specified crop, performs data dimension reduction on the attribute representations contained in the historical price information and the actual evaluation score, and standardizes them. The processed attribute representations are used to construct multiple samples and are used to generate the samples. The model generation module transmits the sample data in the training set to the XGBoost model and the LSTM model for analysis. After the model training is completed, the accuracy of the combined XGBoost model and the LSTM model is determined by the sample data in the test set.

4. A rural financial risk real-time monitoring and early warning system according to claim 1, characterized in that: The initial individual screening unit includes a label adding module and an individual generation module. After the label adding module counts all the hyperparameters to be optimized in the XGBoost model and the LSTM model, it sets an update cycle and divides the hyperparameters to be optimized into two categories, namely high-frequency hyperparameters and low-frequency hyperparameters, and labels them, where the high-frequency hyperparameters are labeled as 1 and the low-frequency hyperparameters are labeled as 0. The individual generation module determines the constraint range of each hyperparameter to be optimized, and sets the number of individuals and the maximum number of cycles. The individual sequence is constructed according to the constraint range and the number of individuals. ,in represents the number of individuals in the initial sequence, , Indicates the Individuals, Indicates the Intra-individual hyperparameters, represents the total number of hyperparameters, Indicates the individual sequence number in the initial sequence.

5. A rural financial risk real-time monitoring and early warning system according to claim 4, characterized in that: The initial individual screening unit also includes a sorting and screening module and an initial individual determination module. The sorting and screening module inputs the corresponding hyperparameters of each individual into the XGBoost model and the LSTM model in sequence, uses the samples in the training set and the test set to determine the accuracy corresponding to different individuals, sorts the individuals according to the accuracy, and calculates the accuracy mean of the current individual sequence, deletes the individuals with accuracy lower than the mean, and obtains the accuracy of the current individual sequence. ,in Indicates the Individuals, Indicates the individual sequence number in the current sequence, Indicates the number of individuals in the current sequence, , set the space coefficient ,according to and Calculate the derived individual ,in , , Represents a random number, repeating the operation until all individuals in the current sequence have generated derivative individuals, and reconstructing multiple different individuals according to the constraint range and number of individuals ,in ,statistics 、 and After the corresponding accuracy, sort by accuracy, and put the first individuals as the initial individuals and delete the other individuals.

6. A rural financial risk real-time monitoring and early warning system according to claim 1, characterized in that: The hyperparameter optimization unit includes an update cycle judgment module, a coefficient calculation module, a coefficient combination module and a first round update module. The update cycle judgment module divides the initial individuals into two groups, stores them in the first set and the second set respectively, records the current cycle number, and judges whether the current cycle number is a multiple of the update cycle. If it is not a multiple of the update cycle, all initial individuals are counted. The hyperparameters marked as 1 in , if it is a multiple of the update period, all initial individuals are counted The coefficient calculation module calculates the phase factor and the transformation coefficient according to the current number of cycles and the maximum number of cycles. If the phase factor is less than the first threshold, the coefficient calculation module calculates the phase factor and the transformation coefficient according to the hyperparameter and constraint range Calculate the updated hyperparameters when no mutation factor is added , repeat until all hyperparameters have been calculated, where , Indicates the first The value of the hyperparameter, Indicates A random number selected from The coefficient is combined with the module to determine the weight coefficient corresponding to the current number of cycles. After adding the variation factor, the variation factor, weight coefficient and Combine to get the updated hyperparameters when adding mutation factors , repeat the operation until all hyperparameters have been calculated, the initial update module counts the accuracy of the original individual, the updated individual without adding the mutation factor and the updated individual with adding the mutation factor, and replaces the hyperparameters of the original individual with the hyperparameters within the individual with high accuracy.

7. A rural financial risk real-time monitoring and early warning system according to claim 1, characterized in that: The hyperparameter determination unit includes a mean analysis module, a best individual screening module, a two-round updating module, a parameter fusion module and an individual output module. If the stage factor is greater than or equal to the first threshold and the transformation coefficient is greater than the second threshold, the mean analysis module calculates the similarity between each individual and the other individuals, and takes the individual with a similarity greater than a preset value as the candidate individual of the current individual. The accuracy corresponding to all candidate individuals is counted, and after deleting the candidate individuals with an accuracy lower than the current individual, the remaining candidate individuals and the current individual are used to calculate the mean of each hyperparameter, and an updated individual is constructed by the mean. The best individual screening module selects the individual with the highest current accuracy and takes it as the best individual. The hyperparameters corresponding to each individual are adjusted by the parameter fine-tuning algorithm. The number of individuals, the hyperparameters of the best individual and the transformation coefficient are calculated to obtain an updated individual. The two-round update module compares the accuracy of the original individual and the two different updated individuals, and replaces the hyperparameters of the original individual with the hyperparameters of the individual with higher accuracy. If the stage factor is greater than or equal to the first threshold and the transformation coefficient is less than or equal to the second threshold, the parameter fusion module determines the individual with the highest accuracy in the first set and the second set, and uses the parameter fusion algorithm to calculate the individual with the highest accuracy in the first set and the second set to obtain new hyperparameters for each individual. The operation is repeated until the end of the cycle. The individual output module outputs the individual with the highest accuracy, and its corresponding hyperparameters are input into the XGBoost model and the LSTM model as the optimal hyperparameters.

8. The rural financial risk real-time monitoring and early warning system according to claim 1 is characterized by: The risk identification unit includes a daily price generation module, a capacity calculation module, and an early warning execution module. The daily price generation module obtains the maximum daily yield and minimum daily yield of the specified crops planted by the current farmer within the forecast period, and transmits the historical price information and actual evaluation score of the specified crops to the XGBoost model and the LSTM model for analysis to obtain the daily forecast price of the specified crops within the forecast period. The capacity calculation module uses the maximum daily yield, the minimum daily yield, and the daily forecast price to calculate the daily repayment capacity of the current farmer for planting the specified crops. The early warning execution module divides the risk level into high risk, medium risk, and low risk, reads the daily repayment amount preset by the current farmer, and if the daily repayment amount within the forecast period is less than or equal to the daily repayment capacity, the current risk level is judged to be low risk and no early warning operation is executed. If the daily repayment amount exceeds the repayment capacity more than once and less than four times within the forecast period, the current risk level is judged to be medium risk and an early warning prompt message is issued to the farmer. If the daily repayment amount exceeds the repayment capacity more than four times within the forecast period, a manual review mechanism is activated.

Images