Risk warning method of intelligent seal management system

Abstract

This application provides a risk warning method for an intelligent seal management system, which involves the field of information technology, including: using natural language processing methods to perform semantic analysis on the seal use reason text in the seal use approval form, extract keywords, and determine whether the seal use reason belongs to the common category of the corresponding seal-related business. If it deviates from the common category and the seal belongs to a high-heat or related core business, a warning is triggered and the risk level is increased; based on the seal use frequency warning, further analyze whether the seal use behavior that triggers the warning occurs outside of normal working hours. If high-frequency abnormal seal use occurs during non-working hours, the warning level is increased to trigger a higher priority risk prompt. Effectively prevent the risk of seal abuse and improve the security and standardization of seal management.

Description

Technical Field

[0001] The present invention relates to the field of information technology, and in particular to a risk early warning method for an intelligent seal management system. Background Art

[0002] In the intelligent seal management system, how to conduct risk warning based on the gradient change of seal usage frequency and seal heat is a complex technical problem. The gradient change of seal usage frequency reflects the dynamic trend of seal usage intensity, while seal heat represents the importance of the seal and the historical cumulative usage intensity. The two are intertwined and need to be analyzed in detail in actual business scenarios. The system needs to track the changes in the usage frequency of each seal in real time and calculate its gradient value. The size of the gradient value represents the severity of the change. However, the meaning of the gradient change of the usage frequency of seals with different heat is not the same. High-heat seals represent highly sensitive businesses, and even a small frequency gradient change may mean risk; while the frequency fluctuation of low-heat seals is relatively less sensitive. Therefore, the system needs to set different warning thresholds for seals with different heat. However, relying solely on frequency gradient values ​​and heat thresholds is not accurate enough. Seal usage behavior must also be considered in a specific time dimension. For example, the frequent use of low-heat seals during non-working hours is abnormal, even if the frequency gradient changes are not large. The use of high-heat seals at critical time points may involve sensitive business and require high attention. In addition, the use of seals must be comprehensively judged in combination with multiple attributes such as the identity of the user and the reason for using the seal. The combination of different attributes corresponds to different risk patterns. The system needs to automatically discover these patterns from massive historical data and apply them to real-time warnings. In short, the intelligent seal management system needs to build an effective risk warning model from multiple angles such as frequency gradient changes, seal heat, time dimension, and seal usage attributes. This requires a deep understanding of the business and data mining capabilities. At the same time, the system must also have flexible warning threshold configuration and real-time data processing capabilities, and ultimately achieve risk warnings that are neither false positives nor missed positives. Summary of the invention

[0003] In order to solve the above technical problems, the present invention provides a risk warning method for an intelligent seal management system, which mainly includes:

[0004] The historical usage frequency data of each seal is obtained, and a prediction model for the usage frequency of the seal is established through a time series analysis algorithm. The usage frequency change gradient of each seal is monitored in real time. The usage frequency change gradient includes the usage growth rate, growth amplitude and mutation degree. When the usage frequency change gradient exceeds the preset change threshold, an early warning is triggered, and an early warning signal of the seal usage frequency dimension is generated; according to the historical usage data of each seal, a clustering algorithm is used to divide the seals into three categories: high heat, medium heat and low heat. Combined with the absolute frequency of seal use and relative ranking, differentiated usage frequency early warning thresholds are set for different categories of seals, and early warning signals of the heat category dimension are generated based on differentiated usage frequency early warning thresholds; organizational personnel system data is obtained to determine the list of authorized users of each seal, and the early warning information is associated with the identity of the user. If it is found that unauthorized personnel use the seal or the number of seal users increases abnormally, a high-risk early warning is triggered, forming an early warning signal of the authorized personnel dimension; through natural language processing methods, semantic analysis is performed on the usage reason text in the seal approval form, keywords are extracted, and the use is judged. Whether the reason for the seal belongs to the common scope of the corresponding seal-related business. If it deviates from the common scope and the seal belongs to high-heat or related core business, an early warning will be triggered and the risk level will be increased. When generating the early warning signal of the seal usage frequency dimension, analyze whether the seal usage behavior that triggers the early warning occurs outside normal working hours. If high-frequency abnormal seal usage occurs during non-working hours, the early warning level will be increased to trigger a higher priority risk prompt. The early warning signals of multiple dimensions such as the frequency of use of each seal, heat category, authorized personnel, reason for use, and working hours are combined, and a weighted calculation method is used to obtain a comprehensive risk score for each seal. The seal risk level is sorted according to the score to form a risk early warning map, highlighting abnormal seals whose frequency of use exceeds the preset frequency threshold. For abnormal seals whose frequency of use exceeds the preset frequency threshold, a risk early warning task is automatically created and sent to the person responsible for seal management for verification and confirmation. The person responsible is required to analyze the reasons for the abnormal changes in frequency and heat, provide supporting information on personnel, reasons, and time, and feedback the verification results to the system for continuous optimization of early warning strategies.

[0005] The technical solution provided by the embodiment of the present invention may have the following beneficial effects:

[0006] The present invention discloses a risk warning method for an intelligent seal management system. The method establishes a seal usage frequency prediction model by analyzing the historical seal usage data, and monitors the changes in the usage frequency of each seal in real time. When the actual frequency exceeds the preset threshold of the prediction value, a warning is triggered. The present invention uses a clustering algorithm to classify seals into three categories of high, medium and low heat, and sets differentiated warning thresholds for different categories. At the same time, a comprehensive analysis is performed in combination with multi-dimensional information such as seal usage time, reasons and personnel. The natural language processing technology is used to perform semantic analysis on the reasons for seal usage to determine whether it deviates from the common business scope. The list of authorized seal users is obtained to identify the seal usage behavior of unauthorized personnel. The present invention performs weighted calculation on multi-dimensional warning signals to obtain a comprehensive seal risk score and form a risk warning map. Automatically create a verification task for high-risk seals and feedback the results. The method realizes all-round intelligent monitoring of seal usage, effectively prevents the risk of seal abuse, and improves the security and standardization of seal management. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] Figure 1 The present invention is a flowchart of the risk warning method of the intelligent seal management system. DETAILED DESCRIPTION

[0008] In order to enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below in conjunction with the drawings in the embodiments of this specification. Obviously, the described embodiments are only part of the embodiments of this specification, not all of the embodiments. Based on the embodiments in this specification, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of this specification.

[0009] like Figure 1 The risk warning method of the smart seal management system in this embodiment may specifically include:

[0010] S101. Obtain the historical frequency data of each seal, establish a prediction model for the seal usage frequency through a time series analysis algorithm, and monitor the frequency change gradient of each seal in real time. The frequency change gradient includes the growth rate, growth amplitude and mutation degree of the seal. When the frequency change gradient exceeds a preset change threshold, a warning is triggered to generate a warning signal for the seal usage frequency dimension.

[0011] The time intervals of adjacent records are obtained according to the seal timestamp data, and the long short-term memory neural network is used to train the time interval data to establish a prediction model for the seal usage frequency. The seal usage frequency reference value is obtained based on the seal usage frequency prediction model; the actual usage frequency curve is statistically analyzed for the seal usage frequency reference value, and the growth rate value, growth amplitude value and mutation degree value are obtained by calculating the usage frequency curve through exponential weighted moving average; the warning threshold is set according to the seal usage frequency reference value, and the warning threshold includes a speed threshold, an amplitude threshold and a mutation threshold. If the growth rate value exceeds the speed threshold, it is determined to be a high-frequency warning, if the growth amplitude value exceeds the amplitude threshold, it is determined to be a high-volume warning, and if the mutation degree value exceeds the mutation threshold, it is determined to be an abnormal warning; the seal warning trigger rule is used to process the high-frequency warning, the high-volume warning and the abnormal warning to obtain a warning signal of the seal usage frequency dimension, and the warning signal of the seal usage frequency dimension includes a warning time period, a warning type and a warning level.

[0012] Exemplarily, a time series data set is generated based on the seal usage timestamp data recorded at the minute level in the time dimension, the interval between adjacent records is calculated for the timestamp data to obtain the time series characteristic value, the seal usage frequency prediction model is established by training the number of seal usage in a unit time period through a long short-term memory neural network, and the seal usage frequency benchmark value is obtained based on the seal usage frequency prediction model. For each seal, the actual number of seal usage curve is statistically analyzed according to a 10-minute time window, the number of seal usage growth rate value and the growth amplitude value are calculated using an exponential weighted moving average, and the mutation point position is obtained from the actual number of seal usage curve to obtain the mutation degree value. The warning threshold is set according to the seal usage benchmark value in different time intervals, and the warning threshold includes a speed threshold, an amplitude threshold, and a mutation threshold. When the growth rate value exceeds the speed threshold of 0.8, it is determined as a high-frequency warning, when the growth amplitude value exceeds the amplitude threshold of 0.6, it is determined as a high-volume warning, and when the mutation degree value exceeds the mutation threshold of 0.5, it is determined as an abnormal warning. The seal frequency warning result is generated through the seal warning trigger rule, and the warning time period, warning type, and warning level are extracted from the warning result. The seal warning record is stored and the disposal rule is assigned according to the warning level. The time series data set generation stage records the specific time point of each seal use. For example, a seal operation is recorded at 9:15, 9:18, and 9:20, and the intervals between adjacent time points are calculated to be 3 minutes and 2 minutes respectively, so as to obtain the time interval characteristics of seal use. These time interval characteristic values ​​can reflect the density of seal use in different time periods. The long short-term memory neural network predicts the number of seal uses in the future by learning the seal usage rules in historical data. For example, by learning the rule that a seal is used an average of 15 times between 9 and 11 am every day, the seal usage baseline value in the same period of the next day is predicted. In actual application scenarios, 10 minutes is used as a time window to count the number of times the seal is used. For example, 8 seal operations are recorded between 9:00 and 9:10, and 12 seal operations are recorded between 9:10 and 9:20. By calculation, it can be concluded that the number of seals increases by 4 times every 10 minutes. The exponential weighted moving average method is used to calculate the growth rate. When the number of seals used during 9:20-9:30 suddenly increases to 25 times, the growth rate in this period reaches 108%, and a mutation point is formed in the time series. The setting of the warning threshold range is based on the historical usage data of the seal. For example, if the average number of seals used per hour on a normal working day is 30 times, and the standard deviation is 5 times, the growth rate threshold can be set to 0.8, that is, when the number of seals used per unit time increases by more than 24 times, a high-frequency warning is triggered. The growth rate threshold is set to 0.6, that is, when the increase in the number of seals used per unit time exceeds 60%, a high-volume warning is triggered. The mutation threshold is set to 0.5, which means that when the number of times the seal is used at a certain time point differs by more than 50% from the number of times the seal is used at the previous and next time points, an abnormal warning is triggered.Different disposal rules are set for different types of early warning signals. For example, the abnormal seal usage frequency warning is recorded as the time period of 9:20-9:30, the warning type is high-volume warning, and the warning level is level 2. The disposal rules include reviewing all seal usage records during this period, questioning and verifying the seal user, and suspending the subsequent use of the seal. Different disposal measures are taken for different warning levels. For example, the first-level warning requires immediate suspension of seal use and on-site inspection, the second-level warning requires manual review of seal usage records, and the third-level warning only requires record filing. The safety of seal use is ensured through a standardized early warning disposal process.

[0013] S102. Based on the historical usage data of each seal, a clustering algorithm is used to divide the seals into three categories: high heat, medium heat and low heat. Combined with the absolute frequency of seal use and relative ranking, differentiated seal usage frequency warning thresholds are set for different categories of seals. Based on the differentiated seal usage frequency warning thresholds, warning signals of the heat category dimension are generated.

[0014] The total number of seal usage and the average daily number of seal usage are statistically calculated based on the historical seal usage data, and the statistical values ​​are normalized by the maximum and minimum values ​​to obtain a feature matrix; the seal usage frequency index and ranking index are extracted through the feature matrix, and the K-means clustering algorithm is used to group the feature vectors to obtain a high-heat group, a medium-heat group and a low-heat group; differentiated seal usage frequency warning thresholds are set for the seals in the high-heat group, the medium-heat group and the low-heat group, and the seal usage frequency warning threshold is obtained by multiplying the average monthly number of seal usage of each group by a preset coefficient.

[0015] Exemplarily, the total number of seal usage per month, the average number of seal usage per day, the monthly usage frequency, and the relative usage ranking are counted based on the historical seal usage data, and the statistical values ​​are normalized to the maximum and minimum values. The seal usage feature matrix is ​​constructed through the standardized values, and the seal usage frequency index, ranking index, and trend index are extracted from the feature matrix to obtain the feature vector. K-means clustering is used to classify the feature vectors. Before clustering, the optimal number of clusters is determined to be 3 through the silhouette coefficient. A random initial cluster center point is set for the feature vector, and the feature vector is grouped and iterated by calculating the Euclidean distance until the cluster center point position converges to obtain the seal heat grouping result, and the seals are marked from high to low according to the average value of the seal usage frequency of each group as a high heat group, a medium heat group, and a low heat group. Differentiated frequency warning thresholds are set for the seals in the high heat group, medium heat group, and low heat group. The high heat group threshold is set to 0.7 times the average monthly number of seals used in this group, the medium heat group threshold is set to 0.8 times the average monthly number of seals used in this group, and the low heat group threshold is set to 0.9 times the average monthly number of seals used in this group. The trend of each seal's real-time use data is extracted through time series decomposition. The number of times the seal is used, the heat group it belongs to, and the proportion of the same group are obtained from the real-time seal use data. When the number of times the seal is used exceeds the threshold of the corresponding heat group, a frequency warning signal is generated, and the seal number, warning time, and the multiple of the threshold exceeded in the warning signal are recorded. The warning level is determined based on the recent change rate of the seal's use frequency, and the warning signals are classified and stored according to the heat group. The construction of seal usage features involves multiple dimensions of data. For example, a seal is used 120 times a month, 4 times a day on average, and 0.8 times a month, ranking fifth among all seals. After normalization, the values ​​of these raw data are unified to between 0 and 1. The normalized value of the monthly usage is 0.85, the normalized value of the daily usage is 0.78, the normalized value of the usage frequency is 0.82, and the normalized value of the ranking is 0.92. The normalized data is more suitable for subsequent clustering. K-means clustering calculates the silhouette coefficient under different cluster numbers when determining the number of clusters. When the number of clusters is 3, the silhouette coefficient reaches the maximum value of 0.68, indicating that the clustering effect is best at this time. The initial cluster center point is determined by randomly selecting three sample points. For example, three samples with printed feature vector values ​​of [0.9, 0.85, 0.88, 0.9], [0.6, 0.58, 0.62, 0.65], and [0.3, 0.28, 0.32, 0.35] are selected as the initial center points. After multiple rounds of iterations, three stable cluster center points are finally obtained.Differentiated threshold settings take into account the frequency of seal usage. Taking the seal grouping of a certain company as an example, the high-heat group includes important seals such as the general manager's seal and the financial seal, with an average monthly usage of 150 times, and the threshold is set to 105 times. The medium-heat group includes departmental business seals, with an average monthly usage of 80 times, and the threshold is set to 64 times. The low-heat group includes low-frequency seals such as archive seals, with an average monthly usage of 30 times, and the threshold is set to 27 times. The generation of seal warning signals is based on real-time monitoring data. For example, a seal in a high-heat group is used 140 times during the monitoring period, which is 1.33 times higher than the threshold of 105 times. At the same time, the frequency of use of the seal has been rising rapidly recently, with a month-on-month growth rate of 50%. Comprehensive analysis generates a secondary warning signal. The seal number recorded in the early warning signal is SEAL001, and the warning time is 10:00 on March 15, 2024. The current high-heat group where the seal belongs accounts for 15% of the total number of corporate seals. The early warning signal is stored in the high-heat group early warning record set for subsequent statistical analysis. Different early warning response mechanisms are adopted for seals in different heat groups. The high-heat group early warning signal has the highest priority and triggers the real-time disposal process.

[0016] S103. Obtain data from the organizational personnel system, determine the list of authorized users of each seal, and associate the warning information with the identity of the seal user. If it is found that unauthorized personnel use the seal or the number of seal users increases abnormally, a high-risk warning will be triggered, forming a warning signal in the authorized personnel dimension.

[0017] Obtain the basic information record of the seal user from the organizational personnel database, the basic information record of the seal user including the seal user identification number and the seal authorization period; use the random forest algorithm to train a seal usage behavior feature model based on the basic information record of the seal user, and the seal usage behavior feature model is used to calculate the seal usage behavior similarity mark value; count the high-frequency warning seal records for the seal usage behavior similarity mark value, and the high-frequency warning seal records include the seal number and the actual number of seal users; associate the high-frequency warning seal records with unauthorized seal usage events to obtain a warning signal in the authorized personnel dimension, and the warning signal in the authorized personnel dimension includes the seal identification and the number of abnormal behavior persons. If the proportion of the number of abnormal behavior persons reaches a preset threshold, a warning of the corresponding level is triggered.

[0018] Exemplarily, the basic information of each seal user is obtained based on organizational personnel data, the list of seal authorizers and the validity period of seal usage rights are extracted from the seal authorization record, and whether the authorization record is within the validity period at the current time point is calculated. The number of times each seal user uses each seal, the average daily frequency of use, and the most recent time of use are counted to generate the authorized person's seal usage behavior characteristics. Random forest is used to train the authorized person's seal usage behavior characteristics, and the seal usage time and seal user information are read from the seal usage record. When it is found that the seal user is not in the authorized person list or the seal usage rights have expired, the unauthorized seal usage event is recorded. The seal usage behavior similarity is calculated based on the authorized person's seal usage feature comparison, and the records with a deviation exceeding 0.8 are marked as abnormal behavior. High-frequency warning seals are extracted from the seal heat warning results, and the actual number of seal users and the number of authorized persons in the past 7 days are counted for the seal, and the month-on-month growth rate of the number of seal users is calculated. When the month-on-month growth rate exceeds 50%, an abnormal warning of the number of seal users is triggered, and the seal number, heat category, and warning time are obtained from the abnormal warning record. According to the abnormal warning record of the number of seal users, the unauthorized seal use event is associated with the abnormal behavior mark to generate an early warning signal of the authorized personnel dimension. The early warning signal of the authorized personnel dimension records the seal identification, the number of authorized persons, the number of actual seal users, the number of unauthorized persons, and the number of abnormal behavior persons. When the abnormal degree of seal users reaches 80%, it is set as a first-level warning, when it reaches 50%, it is a second-level warning, and when it reaches 20%, it is a third-level warning. The authorization management of seal users is based on the basic information and authorization records of personnel. Taking a company's financial seal as an example, Zhang, as the financial manager, obtains the authorization to use the seal for half a year. The authorization period is from January 1, 2024 to June 30, 2024. The system records that he used the seal a total of 320 times during the authorization period, with an average daily frequency of 2.1 times, and the most recent seal use time is March 15, 2024. These data constitute the behavioral characteristics of the seal user. Random forests learn from the historical characteristics of seal usage and establish a standard for determining seal usage. For example, when Li is detected using a special seal for finance, it is found that he is not in the list of authorized persons, and is directly determined as unauthorized seal usage. Although Wang is in the list of authorized persons, his authorization period expires on February 28, 2024, and the system also marks him as unauthorized seal usage. Zhao's seal usage behavior is significantly different from the historical pattern. The frequency of seal usage in a single day suddenly increases to 12 times, and the behavior similarity is only 0.2, which is lower than the threshold standard and is marked as abnormal behavior. The correlation analysis between seal heat and the number of seal users shows that a special seal for finance has triggered a high-frequency warning in the past 7 days. According to statistics, the number of authorized persons for the seal is 10, while the actual number of seal users reaches 18, with a month-on-month growth rate of 80%, exceeding the warning threshold of 50%. The system automatically generates an abnormal warning for the number of seal users, recording the seal number SEAL001, the heat category as the high heat group, and the warning time as March 15, 2024.The warning signal from the authorizer dimension combines multiple abnormal indicators. For the above-mentioned financial seal, the record shows that there are 10 authorized persons, 18 actual seal users, 8 unauthorized seal users, and 6 abnormal behavior users. The abnormal degree of seal users is calculated to be 70%, triggering the second-level warning. The warning signal records in detail the identity information and seal usage time of each unauthorized seal user, as well as the specific reasons for each abnormal behavior mark, including overclocking, exceeding authority, overdue and other dimensions. For the first-level warning seal, it is immediately locked and the relevant person in charge is notified. The second-level warning requires an additional approval link, and the third-level warning is for filing and monitoring.

[0019] S104. Through natural language processing methods, semantic analysis is performed on the seal usage reason text in the seal usage approval form, keywords are extracted, and it is determined whether the seal usage reason belongs to the common category of the corresponding seal-related business. If it deviates from the common category and the seal belongs to a high-heat or related core business, an early warning is triggered and the risk level is increased.

[0020] The Jieba word segmenter is used to segment the seal history approval form, and nouns, verbs, and adjectives are extracted from the word segmentation results to construct phrases to obtain a seal business thesaurus; the top 20% of the phrases in frequency are filtered according to the seal business thesaurus to obtain a seal core business keyword set, and the seal core business keyword set is used to calculate the business matching degree; a bidirectional long short-term memory neural network is used to extract semantic features of the newly added seal use reason text, and the cosine similarity is calculated through the seal core business keyword set to obtain a business matching degree score, and the business matching degree score is used to determine whether the newly added seal use reason belongs to the common category of the corresponding seal-related business; if it deviates from the common category, the seal use warning level is determined according to the business matching degree score and the key monitoring seal list, and the key monitoring seal list includes high-heat seals and core business seals. If the seal belongs to both a high-heat seal and a core business seal, it is determined as a first-level risk, if the seal belongs to only one of them, it is determined as a second-level risk, and other situations are determined as third-level risks.

[0021] Exemplarily, the text of the seal usage reason is obtained according to the seal history approval form, and the text is segmented by Jieba word segmenter. Nouns, verbs, and adjectives in the text are extracted to construct phrases. A business theme word list is generated for the seal history reasons through text theme extraction, and the phrases with the top 20% frequency of occurrence are selected from the business theme word list as the seal core business keyword set. A bidirectional long short-term memory neural network is used to extract semantic features of the newly added seal usage reason text, and a word vector matrix is ​​generated based on word2vec. The cosine similarity between each phrase in the seal usage reason text and the seal core business keywords is calculated, and the similarity value is weighted averaged to obtain the text business matching score. When the business matching score is higher than 0.8, it is determined to be a common business category. The business line to which each seal belongs is read from the seal attribute record, and the seals are grouped and counted according to the business line. The business lines in the top 30% of the statistical values ​​are marked as core businesses. Based on the seal heat classification results obtained in the previous order, high-heat seals and core business seals are selected to form a key monitoring seal list. According to the business matching score and the key monitoring seal list, the warning signal of the seal use reason is generated, and the seal number, business line, and heat classification mark are added to the warning signal. When the seal is both a high-heat seal and a core business seal, it is set to the first-level risk, when it is only a high-heat seal or a core business seal, it is set to the second-level risk, and in other cases, it is set to the third-level risk. The processing of the seal use reason text first requires word segmentation and part-of-speech tagging. For example, for the reason text such as "applying for the seal to handle the bank loan contract", the jieba word segmenter splits it into "application / seal / handling / bank / loan / contract", and identifies "bank, loan, contract" as nouns and "application, handling" as verbs. These phrases constitute the basic semantic unit. Through the analysis of the historical reasons for the use of financial special seals, the business subject words with high frequency such as "loan, account opening, transfer, settlement" are extracted, among which the word frequency of "loan, account opening" accounts for more than 20%, which is determined to be the core business keyword of the seal. The bidirectional long short-term memory neural network can understand the text semantics more accurately by bidirectionally scanning the context of the sentence. Taking "applying for a bank mortgage loan agreement" as an example, each word is converted into a 300-dimensional word vector, and the semantic similarity between the text and the core business keywords of the financial seal is calculated. The cosine similarity between "mortgage loan" and the core keyword "loan" is 0.92, and the similarity between "bank" and the core keyword is 0.85. The business matching score of the overall text is 0.88, which exceeds the threshold of 0.8 and is determined to belong to the common business category. In the business attribute analysis of seals, seals are divided according to business lines. Financial seals and contract seals belong to the financial business line, and personnel seals belong to the human resources line. Statistics on the number of seals in each business line show that the financial business line accounts for 28%, ranking in the top 30%, and is marked as a core business. At the same time, the financial seal is a high-heat seal in the heat classification, so it is included in the list of key monitoring seals.The generation of the warning signal for the use of seals takes into account multiple dimensions. For example, the text of the reason for a seal application is "application for seal advertising contract", and the matching score between this text and the core business keywords of the financial seal is only 0.35, which is obviously deviated from the common business scope. Since the financial seal has the dual attributes of a high-heat seal and a core business seal, the system automatically sets the warning as a first-level risk, and records the warning signal including the seal number SEAL001, financial business line, high-heat classification, business matching degree 0.35 and other information. Different control measures are taken for different risk levels, and the first-level risk requires the approval link of the responsible leader.

[0022] S105. When generating a warning signal for the seal usage frequency dimension, analyze whether the seal usage behavior that triggers the warning occurs outside normal working hours. If high-frequency abnormal seal usage occurs during non-working hours, the warning level will be increased, triggering a higher priority risk warning.

[0023] According to the historical seal usage time records, the time marks of standard working time periods and non-standard working time periods are obtained; the moving average method is used to divide the historical seal usage time records into time windows, and the statistical value of the number of seal usage in each time period and the seal usage benchmark ratio value are obtained through the time window; the warning time point is extracted from the seal warning signal, and the current time period type is obtained according to the warning time point, and the seal usage ratio value is obtained by calculating the number of times the seal is used in non-standard time periods within the past twenty-four hours; if the seal usage ratio value exceeds twice the seal usage benchmark ratio value, a time period abnormality record is generated, and the warning level of the original warning signal is upgraded according to the time period abnormality record to obtain a warning upgrade signal.

[0024] Exemplarily, based on the historical seal usage time records, the time marks of working days and holidays are obtained, and the normal working time intervals are divided according to industry standards. The working hours from 8 am to 12 pm and from 2 pm to 6 pm on working days are recorded as standard working hours, and the rest of the time is recorded as non-standard working hours. The seal usage time points of the working hours and non-working hours are marked. The moving average is used to divide the historical seal usage data into time windows, and the number of seal usage in each hour window is calculated. The peak and trough periods of seal usage are identified from the window data, and the historical seal usage ratio benchmark values ​​of the standard working hours and non-standard working hours are summarized according to the time period type. The warning time point and seal number are extracted from the seal frequency warning signal, and the current time period type is determined based on the association of the time point. At the same time, the proportion of the seal usage times in the non-standard working hours in the past 24 hours is calculated. When the proportion value exceeds the historical seal usage ratio benchmark value by 2 times, a time period abnormality record is generated. An early warning upgrade signal is generated based on the abnormal records of the time period and the original early warning level. When the seal is in a high-frequency use state during non-standard working hours, the early warning level is increased by one based on the original early warning signal. The seal number, number of uses, original early warning level, upgraded early warning level, and trigger time point of the abnormal period are recorded, and the early warning upgrade records are classified and stored according to the seal number. The division of working time periods follows the actual operating rules of the enterprise. Taking the historical use records of a company's financial special seal as an example, the system divides working days into two standard working periods from 8 am to 12 pm and from 2 pm to 6 pm, and other time points are marked as non-standard working periods. Statistics of the seal usage records in March 2024 show that 952 seal operations were completed during standard working hours and 48 seal operations were completed during non-standard working hours, which clearly reflects the time distribution characteristics of seal use. The setting of the mobile time window is based on the usage pattern of the seal. For the financial special seal, one hour is used as the window unit for statistics. The calculation shows that the average number of seals used per hour between 9:00 and 11:00 is 8 times, and the average number of seals used per hour between 14:00 and 16:00 is 6 times, which constitutes two peak periods of seal usage. However, the average number of seals used per hour between 12:00 and 14:00 is only 1, forming an obvious low period of seal usage. Through long-term data accumulation, it is known that the benchmark value of the seal usage ratio during non-standard working hours is stable at around 5%. In actual monitoring, the system detected that the seal triggered a frequency warning signal on March 15, 2024, and the warning time point was 20:30, which belongs to a non-standard working period. Further statistics show that the seal has generated a total of 48 seal usage operations in the past 24 hours, of which 22 occurred during non-standard working hours. The proportion of seal usage during non-standard periods reached 45.8%, which is significantly more than twice the historical benchmark value of 5%. The system automatically generates abnormal records for the time period.During the early warning upgrade process, the original frequency warning level of the financial special seal was level 2. Due to the high frequency of seal use during non-standard working hours, the warning level was automatically upgraded to level 1. The system recorded detailed warning upgrade information: seal number SEAL001, total number of seal uses on the day 48 times, 22 times during non-standard working hours, original warning level 2, upgraded warning level 1, and trigger time March 15, 2024 20:30. The upgraded level 1 warning triggers stricter control measures, including real-time notification of seal administrators, restriction of subsequent seal use permissions, and detailed recording of seal use during non-standard working hours. This multi-dimensional early warning mechanism effectively prevents the risk of abnormal use of seals.

[0025] S106. Based on the early warning signals from multiple dimensions, including the frequency of use of each seal, heat category, authorized personnel, reasons for use, and working hours, a weighted calculation method is used to obtain a comprehensive risk score for each seal. The seal risk levels are sorted according to the scores to form a risk early warning map, highlighting abnormal seals whose frequency of use exceeds the preset frequency threshold.

[0026] According to the warning signals of multiple dimensions such as the frequency of use of each seal, heat category, authorized personnel, reason for use, and working hours, the quantitative value of the warning level is extracted, and the seal warning score is obtained by multiplying the quantitative value of the warning level with the warning trigger frequency; the analytic hierarchy process is used to set the monitoring weights of the frequency of use, heat category, authorized personnel, reason type and time period for the seal warning score to obtain the comprehensive score of seal risk; according to the comprehensive score of seal risk, the risk level threshold interval is trained through convolutional neural network comparative learning to generate a seal risk ranking list with seal number and risk level; a time window analysis is performed on the risk values ​​of each seal in the seal risk ranking list, and when the slope of the trend of the change in the frequency of use exceeds the preset slope threshold or the increment of the number of warnings exceeds the preset increment threshold, it is marked as an abnormal seal; a monitoring matrix is ​​constructed according to the seal risk score, and the risk level is mapped to red, orange and yellow according to the abnormal seal mark to obtain a risk warning map.

[0027] Exemplarily, the quantitative value of the warning level is extracted according to the warning signals of each dimension in the seal warning record, and the various warning scores are calculated according to the product of the warning trigger frequency and the warning level. The warning type weight is set by the hierarchical analysis method, and the seal frequency score weight is 0.4, the heat category score weight is 0.15, the authorized personnel score weight is 0.15, the cause type score weight is 0.2, and the time period monitoring score weight is 0.1. The warning scores are weighted and summarized to obtain the seal risk comprehensive score. The seal risk comprehensive score is compared and learned by a convolutional neural network, and the seal risk level threshold interval is trained based on historical warning data. The risk level judgment rule is optimized by back propagation. The seal risk ranking list is generated in descending order of the risk comprehensive score, and the seal number, risk score, and risk level are recorded according to the list. The current risk value of each seal is read from the seal risk ranking list, and the change trend of the seal frequency in the last 24 hours and the last 7 days is compared with the change curve of the number of warning triggers. When the slope of the change trend exceeds 0.6 or the number of warnings increases by more than 10 times, the seal is marked as an abnormal seal. According to the seal risk score, a monitoring matrix is ​​constructed. The horizontal axis is set as the monitoring time interval, and the vertical axis is set as the seal risk score interval. The risk level value is filled in the matrix. The risk level is color mapped according to the abnormal seal mark. The risk level 1 is marked in red, the level 2 is marked in orange, and the level 3 is marked in yellow to generate a risk warning map. The quantitative processing of the warning signal is based on the weighted calculation of the warning level of different dimensions. Taking a certain enterprise's financial special seal as an example, the seal triggers 5 frequency warnings, 3 authorization warnings, 2 cause warnings, and 1 time period warning within 24 hours. The frequency warning is a first-level warning with 3 points, the authorization warning is a second-level warning with 2 points, the cause warning is a second-level warning with 2 points, and the time period warning is a third-level warning with 1 point. The weights set by the hierarchical analysis method reflect the importance of various warnings. Finally, the comprehensive risk score of the seal is (5×3×0.4+3×2×0.3+2×2×0.2+1×1×0.1)=8.3 points. The convolutional neural network has established a risk level determination standard by learning from historical warning data. Based on a large number of historical cases, it was found that a comprehensive risk score of 8 points or more corresponds to a first-level risk, 6-8 points correspond to a second-level risk, and 4-6 points correspond to a third-level risk. The financial seal scored 8.3 points and was determined to be a first-level risk, ranking second in the risk ranking of all seals, showing a high risk situation. The risk situation judgment tracks risk changes from a time dimension. The frequency of use of the financial seal increased from 2 times per hour to 6 times per hour in the last 24 hours, and the slope of the change trend reached 0.8, exceeding the threshold of 0.6. At the same time, the number of warning triggers increased from 3 times the previous day to 11 times, an increase of 8 times, close to the threshold standard of 10 times. Combining these two indicators, the seal is marked as an abnormal seal and needs to be paid special attention.The risk warning map intuitively displays the risk level through the depth of color. The horizontal axis shows the monitoring time from March 15 to March 21, and the vertical axis shows the risk score range of 0-10 points. The risk score of the financial seal reached 8.3 points on March 19, which is a level 1 risk and is marked in red on the heat map. Through the color change, it can be clearly observed that the risk level of the seal rose from yellow (level 3 risk) to orange (level 2 risk) on March 17, and then to red (level 1 risk) on March 19, forming a clear risk escalation trajectory. The color mapping rules of the heat map are simple and intuitive. The red area represents the seal with the highest risk and the most attention, which makes it easy for managers to quickly identify high-risk seals.

[0028] S107. For abnormal seals whose usage frequency changes exceed the preset frequency threshold, a risk warning task is automatically created and sent to the person responsible for seal management for verification and confirmation. The person responsible is required to analyze the reasons for the abnormal changes in frequency and temperature, provide supporting information on personnel, reasons, and time, and feed back the verification results to the system for continuous optimization of the warning strategy.

[0029] The seal frequency value and heat level are obtained according to the seal monitoring data, and a warning state is obtained when the frequency value exceeds the warning threshold and the heat level is at the highest level; the frequency value is segmentedly calculated by Fourier transform according to the warning state, and the maximum value point and the fluctuation range are obtained from the segmented calculation results; the verification feedback text is semantically classified by the naive Bayes classifier for the maximum value point and the fluctuation range, and the feedback element coverage rate is calculated from the semantic classification results; if the feedback element coverage rate is lower than the coverage rate threshold, the warning threshold and element rules are extracted from the historical verification records, the warning threshold is floatingly adjusted, and the element rules are increased or decreased and updated.

[0030] Exemplarily, the seal frequency abnormality and heat abnormality status are judged according to the seal risk monitoring data. When the frequency exceeds the set threshold of 0.8 and the heat level is at the highest level, an early warning verification task is generated, and the information of the person responsible for the seal is obtained through the organizational structure data, and the early warning task number, seal number, person responsible number, verification deadline, early warning trigger condition, and verification element list are recorded. The seal frequency change value is segmented and calculated according to the hourly level by Fourier transform, and the maximum value point and fluctuation range in each time period are obtained. The seal application record in each time period is compared, and the seal user information, seal use reason content, and seal use time data in the application record are extracted to form an early warning task verification item. The verification feedback text is semantically classified by the naive Bayes classifier, and the personnel description, reason explanation, and time description in the text are identified. The feedback element coverage rate is calculated, and when the element coverage rate is lower than 0.6, an element supplement reminder is automatically generated, and the verification feedback result is recorded in the early warning task sheet. The frequency warning threshold, heat warning threshold, and verification element rules are extracted from the historical verification task records. The threshold values ​​are dynamically adjusted based on the verification feedback content. The frequency warning threshold fluctuates up and down by 0.1, and the verification element rules increase or decrease the number of entries. The adjustment results are stored according to the seal number and the warning rule library is updated. The generation of the warning task is based on the judgment of the abnormal state of the seal. Taking a certain company's financial seal as an example, the seal has been used 6 times per hour recently, exceeding the frequency warning threshold of 0.8, and the heat level has been raised to the highest level, triggering the generation conditions of the warning verification task. The system automatically creates the warning task order TC20240319001, records the seal number SEAL001, the responsible person number USER005, and the verification deadline of March 20, 2024 at 17:00, and lists the two warning trigger conditions of frequency abnormality and heat upgrade. Fourier transform was used to analyze the seal frequency data periodically, and it was identified that the seal had frequency peaks in two time periods, from 9:00 to 11:00 and from 14:00 to 16:00 on March 19. The maximum value in the time period from 9:00 to 11:00 was 8 times per hour, and the maximum value in the time period from 14:00 to 16:00 was 7 times per hour. The seal application corresponding to these two time periods showed that in the morning, Zhang and Li from the Finance Department applied for the seal, and the reasons were bank loan contracts and fund settlement statements. In the afternoon, Wang applied for the seal, and the reasons were supplier payment vouchers. These detailed information constitute the key points of the early warning task. The naive Bayes classifier processed the verification feedback submitted by the responsible person, and extracted 20% of the personnel-related description, 30% of the reasons, and 10% of the time information from the text. The calculated element coverage rate was 0.6, which met the minimum requirement standard. The feedback content showed that the increase in frequency was due to the centralized processing of fund settlement at the end of the month, and the increase in popularity was due to the addition of large-scale loan business. The reasons for the use of the seal belonged to the normal business scope, and although the time was concentrated, it was within the working hours.Based on the verification feedback results, the system optimizes the early warning rules for the seal. Taking into account the characteristics of concentrated business at the end of the month, the frequency warning threshold is raised from 0.8 to 0.9, and the "whether it is the end-of-month settlement period" judgment item is added to the verification factors. The new early warning rules are more in line with business reality and avoid false alarms caused by routine business peaks at the end of the month. This dynamic adjustment mechanism enables the early warning rules to be continuously improved and more accurately identify real abnormal situations. The verification results and rule adjustment records are stored in the early warning rule library for reference in subsequent early warning judgments.

[0031] Based on the above embodiments of the present invention, relevant personnel can make various changes and modifications without departing from the technical concept of the present invention through the above description. The technical scope of the present invention is not limited to the content in the specification, and its technical scope must be determined according to the scope of the claims.

Claims

1. The risk warning method of the intelligent seal management system is characterized by: The method comprises: The historical usage frequency data of each seal is obtained, and a prediction model for the usage frequency of the seal is established through a time series analysis algorithm. The usage frequency change gradient of each seal is monitored in real time. The usage frequency change gradient includes the usage growth rate, growth amplitude and mutation degree. When the usage frequency change gradient exceeds the preset change threshold, an early warning is triggered, and an early warning signal of the seal usage frequency dimension is generated; according to the historical usage data of each seal, a clustering algorithm is used to divide the seals into three categories: high heat, medium heat and low heat. Combined with the absolute frequency of seal use and relative ranking, differentiated usage frequency early warning thresholds are set for different categories of seals, and early warning signals of the heat category dimension are generated based on differentiated usage frequency early warning thresholds; organizational personnel system data is obtained to determine the list of authorized users of each seal, and the early warning information is associated with the identity of the user. If it is found that unauthorized personnel use the seal or the number of seal users increases abnormally, a high-risk early warning is triggered, forming an early warning signal of the authorized personnel dimension; through natural language processing methods, semantic analysis is performed on the usage reason text in the seal approval form, keywords are extracted, and the use is judged. Whether the reason for the seal belongs to the common scope of the corresponding seal-related business. If it deviates from the common scope and the seal belongs to a high-heat or related core business, an early warning will be triggered and the risk level will be increased. When generating the early warning signal of the seal usage frequency dimension, analyze whether the seal usage behavior that triggers the early warning occurs outside normal working hours. If high-frequency abnormal seal usage occurs during non-working hours, the early warning level will be increased to trigger a higher priority risk prompt. The early warning signals of multiple dimensions such as the frequency of use of each seal, the heat category, the authorized personnel, the reason for the use of the seal, and the working hours are combined, and the weighted calculation method is used to obtain the comprehensive risk score of each seal. The seal risk level is sorted according to the score to form a risk early warning map, highlighting the abnormal seals whose frequency of use exceeds the preset frequency threshold. For abnormal seals whose frequency of use exceeds the preset frequency threshold, a risk early warning task is automatically created and sent to the person responsible for seal management for verification and confirmation. The person responsible is required to analyze the reasons for the abnormal changes in frequency and heat, provide supporting information on personnel, reasons, and time, and feedback the verification results to the system for continuous optimization of early warning strategies.

2. The method according to claim 1, characterized in that The historical frequency data of each seal is obtained, and a prediction model of the seal frequency is established through a time series analysis algorithm. The change gradient of the seal frequency of each seal is monitored in real time. The change gradient of the seal frequency includes the growth rate, growth amplitude and mutation degree of the seal. When the change gradient of the seal frequency exceeds a preset change threshold, an early warning is triggered, and an early warning signal of the seal frequency dimension is generated, including: Acquire the time intervals of adjacent records according to the seal timestamp data, use the long short-term memory neural network to train the time interval data to establish a prediction model for the seal usage frequency, and obtain the seal usage frequency reference value based on the prediction model for the seal usage frequency; According to the reference value of the frequency of use of the seal, a curve of the actual number of times of use of the seal is calculated, and the growth rate value, the growth amplitude value and the mutation degree value are obtained by calculating the curve of the number of times of use of the seal by exponential weighted moving average; According to the seal frequency reference value, a warning threshold is set, and the warning threshold includes a speed threshold, an amplitude threshold, and a mutation threshold. If the growth speed value exceeds the speed threshold, it is determined as a high-frequency warning. If the growth amplitude value exceeds the amplitude threshold, it is determined as a high-volume warning. If the mutation degree value exceeds the mutation threshold, it is determined as an abnormal warning. The high-frequency warning, the high-volume warning and the abnormal warning are processed by using seal warning trigger rules to obtain a warning signal in the seal usage frequency dimension. The warning signal in the seal usage frequency dimension includes a warning time period, a warning type and a warning level.

3. The method according to claim 1, characterized in that The method uses a clustering algorithm to classify seals into three categories: high heat, medium heat, and low heat according to the historical seal usage data of each seal. In combination with the absolute frequency of seal usage and relative ranking, differentiated seal usage frequency warning thresholds are set for seals of different categories. Based on the differentiated seal usage frequency warning thresholds, warning signals of heat category dimensions are generated, including: According to the historical seal usage data, the total number of seal usage and the average number of seal usage per day are counted, and the statistical values ​​are normalized to the maximum and minimum values ​​to obtain the feature matrix; The seal usage frequency index and ranking index are extracted through the feature matrix, and the feature vectors are grouped using the K-means clustering algorithm to obtain a high-heat group, a medium-heat group, and a low-heat group; Differentiated seal usage frequency warning thresholds are set for the seals in the high-heat group, medium-heat group and low-heat group. The seal usage frequency warning thresholds are obtained by multiplying the average monthly seal usage times of each group of seals by a preset coefficient.

4. The method according to claim 1, characterized in that The data of the organizational personnel system is obtained to determine the list of authorized users of each seal, and the warning information is associated with the identity of the user. If it is found that unauthorized personnel use the seal or the number of people using the seal increases abnormally, a high-risk warning is triggered, forming a warning signal of the authorized personnel dimension, including: Obtaining a seal user's basic information record from an organizational personnel database, wherein the seal user's basic information record includes a seal user identification number and a seal authorization validity period; A random forest algorithm is used to train a seal-using behavior feature model based on the seal-using person's basic information record, and the seal-using behavior feature model is used to calculate a seal-using behavior similarity mark value; According to the seal-using behavior similarity mark value, high-frequency warning seal records are counted, and the high-frequency warning seal records include the seal number and the number of actual seal users; According to the high-frequency warning seal record and the unauthorized seal use event, a warning signal of the authorized personnel dimension is obtained. The warning signal of the authorized personnel dimension includes the seal identification and the number of abnormal behavior persons. If the proportion of the number of abnormal behavior persons reaches a preset threshold, the corresponding level of warning is triggered.

5. The method according to claim 1, characterized in that: The natural language processing method is used to perform semantic analysis on the seal use reason text in the seal use approval form, extract keywords, and determine whether the seal use reason belongs to the common category of the corresponding seal-related business. If it deviates from the common category and the seal belongs to a high-heat or related core business, an early warning is triggered and the risk level is increased, including: Jieba word segmenter is used to segment the seal history approval form, and nouns, verbs and adjectives are extracted from the word segmentation results to construct phrases, and the seal business subject word list is obtained; According to the seal business subject word list, the top 20% of the word frequency phrases are selected to obtain a seal core business keyword set, and the seal core business keyword set is used to calculate the business matching degree; A bidirectional long short-term memory neural network is used to extract semantic features from the newly added seal use reason text, and the cosine similarity is calculated through the seal core business keyword set to obtain a business matching score, and the business matching score is used to determine whether the newly added seal use reason belongs to the common category of the corresponding seal-related business; If it deviates from the commonly used scope, the seal use warning level will be determined based on the business matching score and the key monitored seal list. The key monitored seal list includes high-heat seals and core business seals. If the seal is a high-heat seal and a core business seal, it will be judged as a first-level risk. If the seal only belongs to one of them, it will be judged as a second-level risk. Other situations will be judged as third-level risks.

6. The method according to claim 1, characterized in that When generating the warning signal of the seal usage frequency dimension, it is analyzed whether the seal usage behavior that triggers the warning occurs outside normal working hours. If high-frequency abnormal seal usage occurs during non-working hours, the warning level is increased to trigger a higher priority risk warning, including: Obtain time stamps for standard working hours and non-standard working hours based on the historical seal usage time records; The historical seal usage time record is divided into time windows by using a moving average method, and the seal usage count statistics and seal usage benchmark ratio values ​​for each time period are obtained through the time windows; Extract the warning time point from the seal warning signal, obtain the current time period type according to the warning time point, and obtain the seal usage ratio by calculating the number of times the seal is used in non-standard time periods within the past 24 hours; If the seal usage ratio exceeds twice the seal usage base ratio, a time period abnormality record is generated, and the warning level of the original warning signal is upgraded according to the time period abnormality record to obtain a warning upgrade signal.

7. The method according to claim 1, characterized in that The method uses a weighted calculation method to obtain a comprehensive risk score for each seal by combining the seal usage frequency, heat category, authorized personnel, seal usage reasons, and working hours of each seal. The seal risk level is sorted according to the score to form a risk warning map, highlighting the abnormal seals whose usage frequency changes exceed the preset frequency threshold, including: Extract the warning level quantitative value based on the warning signals of multiple dimensions such as the frequency of use of each seal, heat category, authorized personnel, reason for use, and working hours, and obtain the seal warning score by multiplying the warning level quantitative value by the warning trigger frequency; The analytic hierarchy process is used to set the seal use frequency, heat category, authorized personnel and reason type and time period monitoring weight for the seal early warning score to obtain the seal risk comprehensive score; According to the comprehensive seal risk score, a seal risk ranking list with seal numbers and risk levels is generated by training the risk level threshold interval through convolutional neural network comparative learning; Performing a time window analysis on the risk values ​​of each seal in the seal risk ranking list, when the slope of the seal usage frequency change trend exceeds a preset slope threshold or the increment of the number of warnings exceeds a preset increment threshold, marking it as an abnormal seal; A monitoring matrix is ​​constructed based on the seal risk score, and the risk levels are mapped to red, orange and yellow according to abnormal seal marks to obtain a risk warning map.

8. The method according to claim 7, characterized in that The monitoring matrix is ​​constructed according to the seal risk score, and the risk level is mapped to red, orange and yellow according to the abnormal seal mark to obtain a risk warning map, including: A monitoring matrix is ​​constructed based on the seal risk score, with the horizontal axis set to the monitoring time interval and the vertical axis set to the seal risk score interval. The risk level values ​​are filled in the matrix, and the risk levels are color mapped according to the abnormal seal marks. The first risk level is marked red, the second risk level is marked orange, and the third risk level is marked yellow to generate a risk warning map.

9. The method according to claim 1, characterized in that: For abnormal seals whose usage frequency changes exceed the preset frequency threshold, a risk warning task is automatically created and sent to the seal management responsible person for verification and confirmation. The responsible person is required to analyze the reasons for the abnormal changes in frequency and temperature, provide supporting information on personnel, reasons, and time, and feed back the verification results to the system for continuous optimization of the warning strategy, including: Acquire the seal frequency value and heat level according to the seal monitoring data, and obtain a warning state when the frequency value exceeds the warning threshold and the heat level is at the highest level; Using the warning state, Fourier transform is used to perform segmented calculation on the frequency value, and the maximum value point and fluctuation range are obtained from the segmented calculation results; Using a naive Bayes classifier to semantically classify the verification feedback text according to the maximum value point and the fluctuation range, and calculating the feedback element coverage rate from the semantic classification results; If the feedback factor coverage is lower than the coverage threshold, the warning threshold and factor rules are extracted from the historical verification records, the warning threshold is adjusted in a floating manner, and the factor rules are updated by additions and subtractions.

10. The method according to claim 9, characterized in that The step of performing segmented calculation on the frequency value by using Fourier transform through the warning state, and obtaining the maximum value point and the fluctuation range from the segmented calculation results, includes: Through the warning state, Fourier transform is used to calculate the seal frequency change value in segments according to the hourly level, and the maximum value point and fluctuation range in each time period are obtained. The seal application record in each time period is compared, and the seal user information, seal use reason content, and seal use time data in the application record are extracted to form a warning task verification item.

Images