A method for automatically settling and managing a preferential medical subsidy

Abstract

The present application relates to the technical field of data settlement, and particularly relates to an automatic settlement management method for preferential treatment and medical assistance. The method comprises the following steps: S100, obtaining basic data of preferential treatment objects, and performing first identification on the basic data of the preferential treatment objects, wherein the basic data comprises an identity category; S200, transmitting the basic data of the preferential treatment objects that have been identified in the preferential treatment and medical assistance settlement platform to a provincial medical insurance information platform through a secure network, and performing corresponding second identification on information belonging to the preferential treatment objects in full-amount insurance data of the provincial medical insurance information platform; S300, performing medical treatment marking when the provincial medical insurance information settles medical treatment of the preferential treatment objects, and obtaining medical insurance settlement data of the preferential treatment objects after the preferential treatment objects complete medical insurance settlement and marking in a medical institution; and S400, the preferential treatment and medical assistance settlement platform regularly obtains the medical insurance settlement data of the preferential treatment objects that have been marked from the provincial medical insurance information platform.

Description

Technical Field

[0001] This invention relates to the field of data settlement technology, and specifically to an automatic settlement management method for preferential medical subsidies for veterans. Background Technology

[0002] The current reimbursement model, primarily based on offline manual operations, has given rise to a series of complex technical and management issues. Firstly, at the data processing level, the lack of automated data exchange channels between systems prevents accurate correlation and real-time verification of medical insurance settlement data and the identity information of beneficiaries, relying entirely on manual document review and data entry. This not only leads to low business processing efficiency but also risks inconsistent subsidy standards and calculation accuracy due to the difficulty in achieving uniformity in manual judgment, while also imposing a heavy administrative burden on grassroots staff. Secondly, regarding data security and privacy protection, the repeated transmission and copying of numerous medical and identity documents containing sensitive personal information in physical space significantly increases the risk of personal information leakage and misuse, contradicting current legal requirements and social consensus on strengthening personal information protection. Thirdly, from a system collaboration perspective, medical insurance policies, drug catalogs, and settlement standards vary across provinces. Any technical solution aiming to achieve cross-platform automated settlement must fully consider the challenges of deep integration and compatibility with specific provincial medical security information platforms, such as Guizhou Province's. Insufficient compatibility can easily lead to data interaction errors, chaotic settlement logic, and even business interruptions. Finally, from a user experience perspective, the entire application process is lengthy, involves numerous steps and cumbersome materials, and lacks ease of use. It fails to reflect the service-oriented philosophy and falls short of the convenient and efficient service goals pursued by digital government.

[0003] Therefore, the existing technology system has revealed multiple defects in supporting the settlement of preferential medical subsidies for veterans, such as cumbersome processes, low efficiency, data fragmentation, poor controllability of security risks, and poor user experience. There is an urgent need for an innovative solution that can fundamentally break down information barriers, realize intelligent data flow, and ensure the safe and automatic operation of the business. Summary of the Invention

[0004] The technical problem solved by this invention is to provide an automatic settlement and management method for preferential medical subsidies, which can fundamentally break down information barriers, realize intelligent data flow, and ensure the safe and automatic operation of business.

[0005] The basic solution provided by this invention is an automatic settlement and management method for preferential medical subsidies for veterans, comprising the following steps:

[0006] S100. The preferential treatment medical subsidy settlement platform obtains the basic data of the preferential treatment recipients from the national preferential treatment information management system, and performs a first identification on the basic data of the preferential treatment recipients in the preferential treatment medical subsidy settlement platform. The basic data includes identity category.

[0007] S200. The basic data of the veterans who have been identified in the preferential medical subsidy settlement platform are transmitted to the provincial medical security information platform through a secure network, and the information belonging to the veterans is identified in the full amount of insurance data of the provincial medical security information platform.

[0008] S300 and provincial medical security information are used to mark medical treatment when veterans seek medical treatment and settle their medical expenses. After veterans complete medical insurance settlement and mark the medical treatment at medical institutions, their medical insurance settlement data is obtained.

[0009] S400, the preferential medical subsidy settlement platform for veterans regularly obtains medical insurance settlement data of the marked veterans from the provincial medical security information platform;

[0010] S500: Calculate the amount of preferential medical subsidies based on the subsidy policy indicators pre-set by the preferential medical subsidy settlement platform and associated with the identity category of the preferential treatment recipients;

[0011] S600: The generated amount of preferential medical subsidies is reviewed and confirmed, and a disbursement report is generated based on the confirmation result. According to the disbursement report, the preferential medical subsidies are disbursed to the corresponding preferential treatment recipient's account.

[0012] The principle and advantages of this invention are as follows: First, the beneficiaries are identified and confirmed within the preferential treatment platform. Then, this confirmed list is transmitted to the medical insurance platform via a secure government network. Within the vast pool of insured individuals on the medical insurance platform, these beneficiaries are accurately identified a second time. Thus, when these specially marked beneficiaries seek medical treatment and settle their medical insurance claims at a hospital, the system can automatically identify them and filter their settlement data separately. The preferential treatment platform can then periodically obtain this exclusive settlement data, automatically calculate the amount of subsidy to be paid according to pre-set subsidy policies, and finally, after manual review and confirmation, directly deposit the subsidy into the individual's account.

[0013] Compared to existing technologies, this system fundamentally changes the outdated model where veterans and their families had to submit paper documents in person. Through automatic data transfer and processing between systems, the entire subsidy settlement process is automated. This not only significantly saves veterans and their families time and energy, allowing them to truly feel the care and convenience provided by the state, but also significantly reduces the workload of grassroots staff in reviewing documents and manually entering data, improving the efficiency of government departments. Furthermore, because the data flows electronically throughout the closed e-government network, the risk of personal information leakage is greatly reduced compared to paper document delivery.

[0014] Furthermore, S550 is included between S500 and S600, and S550 includes the following steps:

[0015] S551. Match the medical insurance settlement data with a preset abnormal behavior rule base, the abnormal behavior rule base including single treatment cost threshold rules, visit frequency rules within a preset time period, and drug and diagnosis matching degree rules;

[0016] S552. Input the medical insurance settlement data into the pre-trained abnormal behavior recognition model and output a risk score;

[0017] S553. Generate early warning information based on the risk score output in S552.

[0018] After the system automatically calculates the subsidy but before submitting it to staff for review, an automated review step is inserted. This step first uses a pre-set set of rigid rules, such as limiting the cost of a single medical visit to a certain amount and prohibiting frequent visits in a short period, to quickly screen the settlement data. Simultaneously, it introduces a more intelligent machine learning model to perform deeper analysis of the settlement data and provide a risk score. The system then automatically generates different levels of warning prompts based on this score, adding a reliable pre-emptive defense to manual review. During staff review, the system automatically identifies potentially problematic applications, such as those with abnormally high costs or suspicious behavioral patterns, and assigns them a risk level. This allows staff to prioritize high-risk applications and focus more on complex and questionable cases, significantly improving the accuracy and efficiency of the review process. This effectively prevents and identifies the risk of improper medical practices or misappropriation of subsidy funds, ensuring the safe and fair use of national subsidy funds.

[0019] Furthermore, S552 includes the following steps:

[0020] S5521. Aggregate medical insurance settlement data according to the preferential treatment recipients to generate a personal medical behavior sequence containing time-series information. The medical insurance settlement data includes consultation time, medical institution information, disease diagnosis code, drug or treatment item code and its quantity and unit price, single diagnosis cost, and total cost.

[0021] S5522. Based on the medical treatment behavior sequence, extract a feature vector for analysis, the feature vector including cost features, frequency features and behavior matching degree features;

[0022] The cost characteristics are calculated by determining the quantile of the cost of a single visit under its corresponding disease diagnosis code, the monthly cumulative cost, and its month-on-month growth rate.

[0023] The frequency characteristics are calculated as the number of visits within a preset sliding time window and the number of different medical institutions involved;

[0024] The behavioral matching feature is based on a pre-established medical knowledge graph to determine whether the compliance matching degree between the medicine or treatment item at the start of the current medical visit and the disease diagnosis code is abnormal, and to count the cumulative number of abnormal matching degrees in the individual's historical sequence.

[0025] S5523. Input the feature vector into the pre-trained isolated forest and logistic regression fusion model for analysis to obtain the risk score.

[0026] This approach strings together the records of multiple medical visits by a single beneficiary in chronological order, forming a dynamic sequence of medical behavior, rather than simply looking at isolated records. Then, it extracts three key features from this sequence: first, cost characteristics, examining the level of each visit's expense compared to others with the same illness, and the trend of monthly total costs; second, frequency characteristics, identifying abnormally high hospital visits within a specific timeframe and whether the individual has visited multiple hospitals; and third, behavioral matching characteristics, using a standard medical knowledge base to determine whether the prescribed medication matches the diagnosis—for example, prescribing heart medication for a cold would be a mismatch. Finally, these features are input into a fusion model combining two algorithms for comprehensive analysis. The advantage of this method is that its judgment is no longer based on a single, rigid rule, but rather on a comprehensive and continuous profile of an individual's medical behavior, making risk assessment more comprehensive and accurate. It can detect obvious anomalies such as "spending exorbitant prices on a single medical visit" as well as hidden risks that require multiple interactions to identify, such as "seeing the same doctor for the same illness at different hospitals and frequently getting prescriptions." This greatly enhances the intelligence level of identifying fraud and abuse.

[0027] Furthermore, S5522 specifically includes the following steps:

[0028] S5522a, The formula for calculating the percentile position of a single medical visit cost is:

[0029]

[0030] in Indicates the position of the quantile. This indicates the ascending ranking of the single medical expense for diagnosis j for beneficiary i among all historical medical expenses for that diagnosis. This represents the total number of historical medical visits for diagnosis j;

[0031] The formula for calculating the month-on-month growth rate of cumulative expenses is:

[0032]

[0033] in This represents the month-on-month growth rate. This represents the cumulative medical expenses for the current month. This represents the cumulative medical expenses for the previous month.

[0034] S5522b, the formula for calculating the frequency index of visits within the sliding time window is:

[0035]

[0036] in This is a frequency index for medical visits. For the number of days in the sliding window, Let t be the weight of day t. Let t be the number of visits on day t. The number of different medical institutions involved in the window. The penalty coefficient for the number of institutions;

[0037] S5522c, The formula for calculating the matching degree of diagnostic and drug compliance is:

[0038]

[0039] in To determine the match rate for this medical visit, The number of medicines available for outpatient visits. For the kth drug, This is the set of drugs adapted to the diagnostic coding standard obtained based on a medical knowledge graph. This is an indicator function; its value is 1 when the drug belongs to this set, and 0 otherwise.

[0040] The system precisely calculates three abstract features—cost, frequency, and matching degree—using specific mathematical formulas. For example, the cost feature is no longer a vague "high" or "low," but rather uses "quantiles" to scientifically determine the specific position of the current cost within the historical costs of similar diseases. The frequency feature is not simply "frequent," but uses a "frequency index" that comprehensively considers time decay and penalties for cross-institutional visits to quantify its degree of abnormality. The matching degree is objectively measured by calculating the proportion of compliant drugs in the total drug supply. The advantage of these formulas is that they transform risk assessment from subjective experience-based judgment to objective data-driven analysis, making the model's input more scientific and consistent. Through quantile comparison, the system can adapt to price differences across different diseases and fairly determine whether costs are abnormal. Through the frequency index, it can more accurately capture intensive medical visits and "multiple prescriptions" within a short period. By calculating the matching degree, it can effectively identify unreasonable medical behaviors such as "over-treatment of minor illnesses" or "indiscriminate prescriptions" based on medical common sense, laying a solid and reliable data foundation for subsequent model analysis.

[0041] Furthermore, the 5523 includes the following steps:

[0042] S5523a. Global sparse outliers in the feature space are detected by the isolated forest model. Behaviors that deviate significantly from the behavior patterns of other preferential treatment recipients in the cost features and frequency features above a preset threshold are identified, and an anomaly isolation score is output.

[0043] S5523b: Identify the feature vector using a logistic regression model trained based on labeled historical fraud and normal settlement data, and output a fraud probability value;

[0044] S5523c: Combining the anomaly isolation score and the fraud probability value, a final comprehensive risk score is generated through a weighted fusion algorithm.

[0045] Furthermore, in S5523c, the risk score is calculated using the following formula:

[0046]

[0047] in Risk scoring, with a value range of: A higher value indicates a greater risk. This represents the anomaly isolation score output by the Isolation Forest model. A higher value indicates that the sample is more anomalous in the global feature space. This is a mapping function that maps the anomaly isolation score to a range of values. The interval; The fraud probability value output by the logistic regression model, with a value range of [value missing]. , These are the weighting coefficients. The penalty coefficient for high-risk behavior. For indicator functions, when Exceeding the preset threshold The value is 1 if it is true, and 0 otherwise. Attached Figure Description

[0048] Figure 1 This is a schematic diagram of an embodiment of the present invention, which is a method for automatically settling and managing preferential medical subsidies for veterans. Detailed Implementation

[0049] The following detailed description illustrates the specific implementation method:

[0050] The basic implementation examples are as follows: Figure 1 As shown:

[0051] A method for automatic settlement and management of preferential medical subsidies for veterans includes the following steps:

[0052] S100. The preferential treatment medical subsidy settlement platform obtains basic data of preferential treatment recipients from the National Preferential Treatment Information Management System, and performs a first identification on the basic data of preferential treatment recipients within the platform. The basic data includes identity category. In the implementation of the method described in this invention, the basic data of preferential treatment recipients obtained by the preferential treatment medical subsidy settlement platform from the National Preferential Treatment Information Management System, including their identity category, etc., is subject to the requirements of relevant laws and policies on personal information protection, and has been explicitly authorized and consented to by the preferential treatment recipients in advance. It is limited to the legitimate purpose of providing them with legally mandated automatic settlement services for medical subsidies. The platform takes strict security protection measures to process all data in accordance with the law.

[0053] S200. The basic data of the veterans who have been identified in the preferential medical subsidy settlement platform are transmitted to the provincial medical security information platform through a secure network, and the information belonging to the veterans is identified in the full amount of insurance data of the provincial medical security information platform.

[0054] S300 and provincial medical security information are used to mark medical treatment when veterans seek medical treatment and settle their medical expenses. After veterans complete medical insurance settlement and mark the medical treatment at medical institutions, their medical insurance settlement data is obtained.

[0055] S400, the preferential medical subsidy settlement platform for veterans regularly obtains medical insurance settlement data of the marked veterans from the provincial medical security information platform;

[0056] S500: Calculate the amount of preferential medical subsidies based on the subsidy policy indicators pre-set by the preferential medical subsidy settlement platform and associated with the identity category of the preferential treatment recipients;

[0057] S600: The generated amount of preferential medical subsidies is reviewed and confirmed, and a disbursement report is generated based on the confirmation result. According to the disbursement report, the preferential medical subsidies are disbursed to the corresponding preferential treatment recipient's account.

[0058] First, the beneficiaries are identified and verified within the preferential treatment platform. This confirmed list is then transmitted securely to the medical insurance platform via a government network, where a second layer of identification is precisely assigned to these beneficiaries among the vast number of insured individuals. This allows the system to automatically identify these specially marked beneficiaries when they seek medical treatment and settle their medical insurance claims at hospitals, and to isolate their settlement data. The preferential treatment platform can then periodically retrieve this exclusive settlement data, automatically calculate the appropriate subsidy amount based on pre-set subsidy policies, and finally, after manual review and confirmation, directly deposit the subsidy into the individual's account.

[0059] Compared to existing technologies, this system fundamentally changes the outdated model where veterans and their families had to submit paper documents in person. Through automatic data transfer and processing between systems, the entire subsidy settlement process is automated. This not only significantly saves veterans and their families time and energy, allowing them to truly feel the care and convenience provided by the state, but also significantly reduces the workload of grassroots staff in reviewing documents and manually entering data, improving the efficiency of government departments. Furthermore, because the data flows electronically throughout the closed e-government network, the risk of personal information leakage is greatly reduced compared to paper document delivery.

[0060] Between S500 and S600, there is also S550, which includes the following steps:

[0061] S551. Match the medical insurance settlement data with a preset abnormal behavior rule base, the abnormal behavior rule base including single treatment cost threshold rules, visit frequency rules within a preset time period, and drug and diagnosis matching degree rules;

[0062] S552. Input the medical insurance settlement data into the pre-trained abnormal behavior recognition model and output a risk score;

[0063] S553. Generate early warning information based on the risk score output in S552.

[0064] After the system automatically calculates the subsidy but before submitting it to staff for review, an automated review step is inserted. This step first uses a pre-set set of rigid rules, such as limiting the cost of a single medical visit to a certain amount and prohibiting frequent visits in a short period, to quickly screen the settlement data. Simultaneously, it introduces a more intelligent machine learning model to perform deeper analysis of the settlement data and provide a risk score. The system then automatically generates different levels of warning prompts based on this score, adding a reliable pre-emptive defense to manual review. During staff review, the system automatically identifies potentially problematic applications, such as those with abnormally high costs or suspicious behavioral patterns, and assigns them a risk level. This allows staff to prioritize high-risk applications and focus more on complex and questionable cases, significantly improving the accuracy and efficiency of the review process. This effectively prevents and identifies the risk of improper medical practices or misappropriation of subsidy funds, ensuring the safe and fair use of national subsidy funds.

[0065] S552 includes the following steps:

[0066] S5521. Aggregate medical insurance settlement data according to the preferential treatment recipients to generate a personal medical behavior sequence containing time-series information. The medical insurance settlement data includes consultation time, medical institution information, disease diagnosis code, drug or treatment item code and its quantity and unit price, single diagnosis cost, and total cost.

[0067] S5522. Based on the medical treatment behavior sequence, extract a feature vector for analysis, the feature vector including cost features, frequency features and behavior matching degree features;

[0068] The cost characteristics are calculated by determining the quantile of the cost of a single visit under its corresponding disease diagnosis code, the monthly cumulative cost, and its month-on-month growth rate.

[0069] The frequency characteristics are calculated as the number of visits within a preset sliding time window and the number of different medical institutions involved;

[0070] The behavioral matching feature is based on a pre-established medical knowledge graph to determine whether the compliance matching degree between the medicine or treatment item at the start of the current medical visit and the disease diagnosis code is abnormal, and to count the cumulative number of abnormal matching degrees in the individual's historical sequence.

[0071] S5523. Input the feature vector into the pre-trained isolated forest and logistic regression fusion model for analysis to obtain the risk score.

[0072] This approach strings together the records of multiple medical visits by a single beneficiary in chronological order, forming a dynamic sequence of medical behavior, rather than simply looking at isolated records. Then, it extracts three key features from this sequence: first, cost characteristics, examining the level of each visit's expense compared to others with the same illness, and the trend of monthly total costs; second, frequency characteristics, identifying abnormally high hospital visits within a specific timeframe and whether the individual has visited multiple hospitals; and third, behavioral matching characteristics, using a standard medical knowledge base to determine whether the prescribed medication matches the diagnosis—for example, prescribing heart medication for a cold would be a mismatch. Finally, these features are input into a fusion model combining two algorithms for comprehensive analysis. The advantage of this method is that its judgment is no longer based on a single, rigid rule, but rather on a comprehensive and continuous profile of an individual's medical behavior, making risk assessment more comprehensive and accurate. It can detect obvious anomalies such as "spending exorbitant prices on a single medical visit" as well as hidden risks that require multiple interactions to identify, such as "seeing the same doctor for the same illness at different hospitals and frequently getting prescriptions." This greatly enhances the intelligence level of identifying fraud and abuse.

[0073] S5522 specifically includes the following steps:

[0074] S5522a, The formula for calculating the percentile position of a single medical visit cost is:

[0075]

[0076] in Indicates the position of the quantile. This indicates the ascending ranking of the single medical expense for diagnosis j for beneficiary i among all historical medical expenses for that diagnosis. This represents the total number of historical medical visits for diagnosis j;

[0077] The formula for calculating the month-on-month growth rate of cumulative expenses is:

[0078]

[0079] in This represents the month-on-month growth rate. This represents the cumulative medical expenses for the current month. This represents the cumulative medical expenses for the previous month.

[0080] S5522b, the formula for calculating the frequency index of visits within the sliding time window is:

[0081]

[0082] in This is a frequency index for medical visits. For the number of days in the sliding window, Let t be the weight of day t. Let t be the number of visits on day t. The number of different medical institutions involved in the window. The penalty coefficient for the number of institutions;

[0083] S5522c, The formula for calculating the matching degree of diagnostic and drug compliance is:

[0084]

[0085] in To determine the match rate for this medical visit, The number of medicines available for outpatient visits. For the kth drug, This is the set of drugs adapted to the diagnostic coding standard obtained based on a medical knowledge graph. This is an indicator function; its value is 1 if the drug belongs to this set, and 0 otherwise.

[0086] The system precisely calculates three abstract features—cost, frequency, and matching degree—using specific mathematical formulas. For example, the cost feature is no longer a vague "high" or "low," but rather uses "quantiles" to scientifically determine the specific position of the current cost within the historical costs of similar diseases. The frequency feature is not simply "frequent," but uses a "frequency index" that comprehensively considers time decay and penalties for cross-institutional visits to quantify its degree of abnormality. The matching degree is objectively measured by calculating the proportion of compliant drugs in the total drug supply. The advantage of these formulas is that they transform risk assessment from subjective experience-based judgment to objective data-driven analysis, making the model's input more scientific and consistent. Through quantile comparison, the system can adapt to price differences across different diseases and fairly determine whether costs are abnormal. Through the frequency index, it can more accurately capture intensive medical visits and "multiple prescriptions" within a short period. By calculating the matching degree, it can effectively identify unreasonable medical behaviors such as "over-treatment of minor illnesses" or "indiscriminate prescriptions" based on medical common sense, laying a solid and reliable data foundation for subsequent model analysis.

[0087] The 5523 includes the following steps:

[0088] S5523a. Global sparse outliers in the feature space are detected by the isolated forest model. Behaviors that deviate significantly from the behavior patterns of other preferential treatment recipients in the cost features and frequency features above a preset threshold are identified, and an anomaly isolation score is output.

[0089] S5523b: Identify the feature vector using a logistic regression model trained based on labeled historical fraud and normal settlement data, and output a fraud probability value;

[0090] S5523c: Combining the anomaly isolation score and the fraud probability value, a final comprehensive risk score is generated through a weighted fusion algorithm.

[0091] In S5523c, the risk score is calculated using the following formula:

[0092]

[0093] in Risk scoring, with a value range of: A higher value indicates a greater risk. This represents the anomaly isolation score output by the Isolation Forest model. A higher value indicates that the sample is more anomalous in the global feature space. This is a mapping function that maps the anomaly isolation score to a range of values. The interval; The fraud probability value output by the logistic regression model, with a value range of [value missing]. , These are the weighting coefficients. The penalty coefficient for high-risk behavior. For indicator functions, when Exceeding the preset threshold The value is 1 if it is true, and 0 otherwise.

[0094] The above are merely embodiments of the present invention. Commonly known structures and characteristics are not described in detail here. Those skilled in the art are aware of all common technical knowledge in the field prior to the application date or priority date, are aware of all existing technologies in that field, and have the ability to apply conventional experimental methods prior to that date. Those skilled in the art can, under the guidance of this application, improve and implement this solution in combination with their own capabilities. Some typical known structures or methods should not be obstacles for those skilled in the art to implement this application. It should be noted that those skilled in the art can make several modifications and alterations without departing from the structure of the present invention, and these should also be considered within the scope of protection of the present invention. These modifications will not affect the effectiveness of the implementation of the present invention or the practicality of the patent. The scope of protection claimed in this application should be determined by the content of its claims, and the specific embodiments described in the specification can be used to interpret the content of the claims.

Claims

1. A method for automatic settlement and management of preferential medical subsidies for veterans, characterized in that: Includes the following steps: S100. The preferential treatment medical subsidy settlement platform obtains the basic data of the preferential treatment recipients from the national preferential treatment information management system, and performs a first identification on the basic data of the preferential treatment recipients in the preferential treatment medical subsidy settlement platform. The basic data includes identity category. S200. The basic data of the veterans who have been identified in the preferential medical subsidy settlement platform are transmitted to the provincial medical security information platform through a secure network, and the information belonging to the veterans is identified in the full amount of insurance data of the provincial medical security information platform. S300 and provincial medical security information are used to mark medical treatment when veterans seek medical treatment and settle their medical expenses. After veterans complete medical insurance settlement and mark the medical treatment at medical institutions, their medical insurance settlement data is obtained. S400, the preferential medical subsidy settlement platform for veterans regularly obtains medical insurance settlement data of the marked veterans from the provincial medical security information platform; S500: Calculate the amount of preferential medical subsidies based on the subsidy policy indicators pre-set by the preferential medical subsidy settlement platform and associated with the identity category of the preferential treatment recipients; S600: The generated amount of preferential medical subsidies is reviewed and confirmed, and a disbursement report is generated based on the confirmation result. According to the disbursement report, the preferential medical subsidies are disbursed to the corresponding preferential treatment recipient's account. Between S500 and S600, there is also S550, which includes the following steps: S551. Match the medical insurance settlement data with a preset abnormal behavior rule base, the abnormal behavior rule base including single treatment cost threshold rules, visit frequency rules within a preset time period, and drug and diagnosis matching degree rules; S552. Input the medical insurance settlement data into the pre-trained abnormal behavior recognition model and output a risk score; S553. Generate early warning information based on the risk score output in S552; S552 includes the following steps: S5521. Aggregate medical insurance settlement data according to the preferential treatment recipients to generate a personal medical behavior sequence containing time-series information. The medical insurance settlement data includes consultation time, medical institution information, disease diagnosis code, drug or treatment item code and its quantity and unit price, single diagnosis cost, and total cost. S5522. Based on the medical treatment behavior sequence, extract a feature vector for analysis, the feature vector including cost features, frequency features and behavior matching degree features; The cost characteristics are calculated by determining the quantile of the cost of a single visit under its corresponding disease diagnosis code, the monthly cumulative cost, and its month-on-month growth rate. The frequency characteristics are calculated as the number of visits within a preset sliding time window and the number of different medical institutions involved; The behavioral matching feature is based on a pre-established medical knowledge graph to determine whether the compliance matching degree between the prescription for the current medical visit and the disease diagnosis code is abnormal, and to count the cumulative number of abnormal matching degrees in the individual's historical sequence. S5523. Input the feature vector into the pre-trained isolated forest and logistic regression fusion model for analysis to obtain a risk score; The 5523 includes the following steps: S5523a. Global sparse outliers in the feature space are detected by the isolated forest model. Behaviors that deviate significantly from the behavior patterns of other preferential treatment recipients in the cost features and frequency features above a preset threshold are identified, and an anomaly isolation score is output. S5523b: Identify the feature vector using a logistic regression model trained based on labeled historical fraud and normal settlement data, and output a fraud probability value; S5523c. Combining the anomaly isolation score and the fraud probability value, a final comprehensive risk score is generated through a weighted fusion algorithm. In S5523c, the risk score is calculated using the following formula: ； in Risk scoring, with a value range of: A higher value indicates a greater risk. This represents the anomaly isolation score output by the Isolation Forest model. A higher value indicates that the sample is more anomalous in the global feature space. This is a mapping function that maps the anomaly isolation score to a range of values. The interval; The fraud probability value output by the logistic regression model, with a value range of [value missing]. , These are the weighting coefficients. The penalty coefficient for high-risk behavior. For indicator functions, when Exceeding the preset threshold The value is 1 if it is true, and 0 otherwise.

2. The method for automatic settlement and management of preferential medical subsidies for veterans according to claim 1, characterized in that: S5522 specifically includes the following steps: S5522a, The formula for calculating the percentile position of a single medical visit cost is: ； in Indicates the position of the quantile. This indicates the ascending ranking of the single medical expense for diagnosis j for beneficiary i among all historical medical expenses for that diagnosis. This represents the total number of historical medical visits for diagnosis j; The formula for calculating the month-on-month growth rate of cumulative expenses is: ； in This represents the month-on-month growth rate. This represents the cumulative medical expenses for the current month. This represents the cumulative medical expenses for the previous month. S5522b, the formula for calculating the frequency index of visits within the sliding time window is: ； in This is a frequency index for medical visits. For the number of days in the sliding window, Let t be the weight of day t. Let t be the number of visits on day t. The number of different medical institutions involved in the window. The penalty coefficient for the number of institutions; S5522c, The formula for calculating the matching degree of diagnostic and drug compliance is: ； in To determine the match rate for this medical visit, The number of medicines available for outpatient visits. For the kth drug, This is the set of drugs adapted to the diagnostic coding standard obtained based on a medical knowledge graph. This is an indicator function; its value is 1 when the drug belongs to this set, and 0 otherwise.

Images