An intelligent monitoring system and method applied to text recognition

Abstract

The application discloses an intelligent monitoring system and method applied to text recognition, relates to the technical field of medicine warehousing monitoring, and comprises the following steps: obtaining original text data of an invoice two-dimensional code of a medicine, and constructing business multi-source data of business context information packages of the original text data; constructing an abnormal mode vector of a business probe data set, analyzing a silent failure condition of a monitoring center platform in a current time window, and obtaining silent failure data; according to the business multi-source data and the silent failure data, analyzing a comprehensive health condition of text recognition service of the monitoring center platform in the current time window, and obtaining target health data; and according to the target health data, performing abnormal alarm on the text recognition service of the monitoring center platform in the current time window, so that management personnel can make corresponding adjustment in time, the quality of text recognition is fundamentally ensured, and the safety of normal business processes is protected.

Description

Technical Field

[0001] This invention relates to the field of pharmaceutical inventory monitoring technology, specifically an intelligent monitoring system and method applied to text recognition. Background Technology

[0002] Text recognition services have been widely used in various fields such as drug traceability, invoice verification, and automated document processing. For example, in drug warehousing systems, text recognition services are used to scan the QR codes on invoices using barcode scanners, thereby automatically extracting key drug information such as invoice numbers and drug types. Furthermore, these services are integrated with third-party tax invoice platforms to verify the authenticity of invoices and conduct three-party comparison verification. As the demand for drug warehousing continues to increase in reality, monitoring the stability and availability of text recognition services during the process has become a core element in ensuring the stable operation of drug warehousing operations in the pharmaceutical field.

[0003] Currently, the health status monitoring of text recognition services and other distributed services mainly adopts a range-based health monitoring system based on a heartbeat mechanism. Specifically, the monitored service instances need to periodically send heartbeat signals to the monitoring center at preset time intervals. The monitoring center determines whether the service is in normal operation based on the arrival of the heartbeat information. However, the heartbeat mechanism has a technical flaw that is easily overlooked: a normal heartbeat signal only proves that the monitored process is alive, but cannot prove that the service's business functions are available. Therefore, in practice, there may still be situations where the service is alive but the business functions are abnormal, which will lead to abnormal text recognition results and thus cause fatal damage to the normal operation of the business process. Summary of the Invention

[0004] The purpose of this invention is to provide an intelligent monitoring system and method for text recognition, so as to solve the problems raised in the prior art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: an intelligent monitoring method applied to text recognition, the method comprising: Step S1: Obtain the raw text data of the drug invoice QR code and construct the business context information package of the raw text data as multi-source business data; Step S2: Analyze the health status of the heartbeat signal of the monitoring center platform within the current time window, analyze the comprehensive changes of the business probes of the monitoring center platform within the current time window, construct the abnormal pattern vector of the business probe dataset, analyze the silent failure status of the monitoring center platform within the current time window, and obtain silent failure data. Step S3: Based on multi-source business data and silent failure data, analyze the overall health status of the text recognition service of the monitoring center platform within the current time window to obtain target health data; Step S4: Based on the target health data, issue an anomaly alert for the text recognition service of the monitoring center platform within the current time window.

[0006] Furthermore, step S2 includes: Obtain the minimum time difference p between the business probe dataset and the heartbeat signal dataset from multi-source business data. t Set the time-series threshold p', when p t When p ≥ p', the time window W is not processed; otherwise, the time window W is modified to obtain the modified time window W'. Get the last timestamp t of the heartbeat signal dataset last Get the timestamp t of the current moment. now Calculate the resting heart rate duration Δt = t now -t last Obtain the modified tolerance dynamic coefficient α´; Set the heart rate health variable H, the single probe variable q, and the comprehensive probe variable Q. Set the value of the single probe variable for each business probe result vector in the business probe dataset. When Δt≤α´·T, set H=1, otherwise set H=0. When there is a business probe result vector in the business probe dataset with a single probe variable of 0, set the comprehensive probe variable Q to 0, otherwise set the comprehensive probe variable Q to 1. When the comprehensive probe variable Q=1, the business probe dataset is not processed. When the comprehensive probe variable Q=0, an abnormal mode vector Q´ is constructed, the silent failure variable S of the monitoring center platform in the current time window is set, and the heartbeat health variable H and comprehensive probe variable Q of the monitoring center platform in the current time window are obtained. When H=1 and Q=0, S is set to 1, otherwise S is set to 0. When S=0, it is determined that there is no silent failure in the monitoring center platform within the current time window. When S=1, the severity D of the silent failure of the monitoring center platform within the current time window is calculated, the timeliness feature value m is set, and the severity D of the silent failure within the first m correction time windows W´ within the current time window is obtained. m The silent failure trend slope k of the monitoring center platform within the current time window is calculated. The silent failure trend slope k, silent failure variable S, silent failure severity D and abnormal mode vector Q´ are aggregated to obtain the silent failure data of the monitoring center platform within the current time window. The above steps enable proactive identification and precise location of the traditional monitoring blind spot of normal heartbeat but abnormal business operations. Furthermore, by forcibly binding the heartbeat signal and business probe to the same invoice recognition transaction for analysis, the mismatch between heartbeat signal and business probe caused by pure time window grouping is completely avoided, significantly reducing the false alarm rate of silent failure. At the same time, the acquisition of silent failure data also provides the monitoring center platform with refined quantitative basis for subsequent graded anomaly warnings of text recognition services.

[0007] Furthermore, step S1 includes: The raw text data of the drug invoice QR code is obtained from the scanner device, and the fields are extracted and parsed from the raw text data to obtain the business context information package; According to a preset fixed period, the monitoring center platform sends heartbeat monitoring requests to the monitoring center platform interface. After receiving the heartbeat monitoring requests, the monitoring center platform updates the stored service heartbeat data table and generates a heartbeat signal vector. The invoice number is obtained from the business context information package, a simulated text request probe is constructed, the simulated text request probe is sent to the monitoring center platform, the core quantitative indicators returned by the monitoring center platform based on the simulated text request probe are obtained, and the business probe result vector of the business context information package is generated. Get the timestamp t of the business context information packet in the current time window, get the preset time window W, and based on the timestamp t, extract all the business probe result vectors and heartbeat signal vectors within the time window W and aggregate them to obtain the vector dataset of the business context information packet in the current time window. Then, aggregate the business probe result vectors and heartbeat signal vectors in the vector dataset according to the primary key ID of the business context information packet to obtain the relevant vector dataset. The business probe result vector and heartbeat signal vector are separately partitioned and aggregated from the relevant vector dataset to obtain the business probe dataset and the heartbeat signal dataset, respectively. The minimum time difference p between the business probe dataset and the heartbeat signal dataset is calculated for each dataset. t ; The minimum time difference p between the business probe dataset and the heartbeat signal dataset t The relevant vector dataset and business probe dataset are aggregated to generate multi-source business data for the business context information package.

[0008] Furthermore, step S3 includes: Acquire the silent failure data and business context information packet of the monitoring center platform within the current time window, calculate the heartbeat stability value G of the monitoring center platform within the current time window, calculate the probe call success rate U and probe identification quality value Y of the monitoring center platform within the current time window, and calculate the response performance value Z of the monitoring center platform within the current time window. Obtain preset weight coefficients λ G , λ U , λ Y and λ Z for the heartbeat stability value G, probe call success rate U, probe identification quality value Y and response performance value Z of the monitoring center platform, and dynamically correct the weight coefficients λ G , λ U , λ Y and λ Z according to the silent failure data of the monitoring center platform within the current time window; Calculate the static failure penalty term P S of the monitoring center platform within the current time window, calculate the comprehensive health score A of the text recognition service of the monitoring center platform within the current time window, calculate the change value △A of the comprehensive health score of the text recognition service of the monitoring center platform within the current time window, and aggregate the comprehensive health score A and the change value △A to obtain the target health data of the text recognition service of the monitoring center platform within the current time window.

[0009] Further, step S4 includes: Obtain the target health data of the text recognition service of the monitoring center platform within the current time window, and acquire the comprehensive health score A and change value △A of the text recognition service of the monitoring center platform within the current time window from the target health data; Set a comprehensive health score threshold A´ and a health score change rate threshold △A´, wherein A´>0 and △A´>0; When A<A´, send a serious abnormality alarm for the text recognition service of the monitoring center platform within the current time window, and stop the text recognition service of the monitoring center platform within the current time window; When A≥A´ and △A>△A´, send an abnormality alarm for the text recognition service of the monitoring center platform within the current time window, and restart the text recognition service of the monitoring center platform within the current time window; When A≥A´ and △A≤△A´, send a detection reminder to the manager of the text recognition service through the monitoring center platform to remind the manager to perform service function detection on the text recognition service of the monitoring center platform.

[0010] To better implement the above method, an intelligent monitoring system applied to text recognition is also provided, the system includes a service data construction module, a silent failure analysis module, a service health analysis module and an abnormality alarm module; the service data construction module is configured to acquire original text data of drug invoice QR codes, and construct multi-source service data of service context information packages for the original text data; The silent failure analysis module is used to analyze the silent failure status of the monitoring center platform within the current time window and obtain silent failure data. The service health analysis module is used to analyze the overall health status of the text recognition service on the monitoring center platform within the current time window to obtain target health data; The anomaly alarm module is used to issue anomaly alarms for the text recognition service of the monitoring center platform within the current time window.

[0011] Furthermore, the business data construction module includes a business information package generation unit and a business data construction unit; The business information package generation unit is used to extract and parse fields from raw text data to generate a business context information package. The business data construction unit is used to construct multi-source business data based on the business context information package of the original text data.

[0012] Furthermore, the silent failure analysis module includes a service probe change analysis unit and a silent failure analysis unit; The business probe change analysis unit is used to analyze the health status of the heartbeat signal of the monitoring center platform within the current time window, analyze the comprehensive change status of the business probes of the monitoring center platform within the current time window, and construct the abnormal pattern vector of the business probe dataset. The silent failure analysis unit is used to analyze the silent failure status of the monitoring center platform within the current time window based on the abnormal pattern vector of the business probe dataset, and obtain silent failure data.

[0013] Furthermore, the service health analysis module includes a weight dynamic adjustment unit and a service health analysis unit; The weighted dynamic correction unit is used to dynamically correct the preset weight coefficients of the center jump stability value, probe call success rate, probe identification quality value and response performance value of the monitoring center platform based on the silent failure data. The service health analysis unit is used to calculate the static failure penalty items of the monitoring center platform within the current time window, analyze the overall health status of the text recognition service of the monitoring center platform within the current time window, and obtain target health data.

[0014] Furthermore, the anomaly alarm module includes anomaly alarm units; The anomaly alarm unit is used to identify anomalies in the text recognition service of the monitoring center platform within the current time window based on the target health data, and to issue anomaly alarms for the text recognition service.

[0015] Compared with existing technologies, the beneficial effects of this invention are as follows: This invention creatively performs logical intersection operations between heartbeat signals and business probe results to proactively identify the reverse pattern of normal heartbeat but abnormal business probe during text recognition service. Compared with traditional solutions, this method can identify the shortcomings of traditional heartbeat monitoring, which cannot cover the monitoring blind spots, filling the technical gap in silent failure detection of text recognition services. Furthermore, it deepens the monitoring of text recognition services from communication interruption monitoring to semantic accuracy monitoring, thereby discovering hidden faults unique to text recognition scenarios. The specific health status scoring mechanism for text recognition services can accurately reflect silent failures. Subsequent predictive anomaly alarms allow managers to make timely adjustments, fundamentally ensuring the quality of text recognition and protecting the security of normal business processes. Attached Figure Description

[0016] Figure 1 This is a flowchart of an intelligent monitoring method applied to text recognition according to the present invention; Figure 2 This is a flowchart of a module of an intelligent monitoring system for text recognition according to the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Example: Figures 1-2 As shown, the present invention provides a technical solution, an intelligent monitoring method applied to text recognition, the method comprising: Step S1: Obtain the raw text data of the drug invoice QR code and construct the business context information package of the raw text data as multi-source business data; Step S1 includes: The raw text data of the drug invoice QR code is obtained from the scanner device, and the fields are extracted and parsed from the raw text data to obtain the business context information package; For example, the original text data is as follows: The structured string obtained after decoding the invoice QR code includes fields such as invoice type identifier, invoice code, and invoice number. Then, fields are extracted and parsed from the original text data to obtain the QR code parsing record. The specific extraction and parsing process is as follows: According to the preset data format specifications of the invoice QR code, the invoice type, invoice code, invoice number, invoice date, total price and tax, hospital ID and delivery company ID are extracted from the original text data. Full-width characters are automatically converted to half-width characters and collected in a preset order to obtain the business context information package. According to a preset fixed period, the monitoring center platform sends heartbeat monitoring requests to the monitoring center platform interface. After receiving the heartbeat monitoring requests, the monitoring center platform updates the stored service heartbeat data table and generates a heartbeat signal vector. For example, the service heartbeat data table is a core runtime status table deployed in the monitoring center platform, specifically used to store the heartbeat signals periodically reported by each text recognition service instance. For example, the heartbeat signal vector is generated as follows: The monitoring center platform obtains the business context information packet from the heartbeat monitoring request, imports the ID field from the business context information packet into the fields in the heartbeat monitoring request, and aggregates them in a preset order to generate a heartbeat signal vector. The invoice number is obtained from the business context information package, a simulated text request probe is constructed, the simulated text request probe is sent to the monitoring center platform, the core quantitative indicators returned by the monitoring center platform based on the simulated text request probe are obtained, and the business probe result vector of the business context information package is generated. For example, the invoice number is obtained from the business context information packet, and a simulated text request probe is constructed. The specific construction process is as follows: The monitoring center platform obtains the invoice number from the business context information packet, retrieves the identification invoice structure corresponding to the business context information packet from the standard library in the monitoring center platform, and constructs a simulated text request probe. For example, the process of obtaining the core quantitative indicators returned by the monitoring center platform based on the simulated text request probe, and generating a business probe result vector of business context information packet, is as follows: The core metrics for obtaining simulated text request probes returned by the monitoring center platform are: response time, response status code, and recognition similarity. The identification similarity is calculated by calling the existing Levenshtein Distance algorithm to calculate the similarity between the simulated text request probe returned text and the preset standard text in the standard library, and recorded as the identification similarity. The ID field, timestamp, and various core quantitative indicators in the business context information package are sorted in a preset order to generate a business probe result vector of the business context information package. Get the timestamp t of the business context information packet in the current time window, get the preset time window W, and based on the timestamp t, extract all the business probe result vectors and heartbeat signal vectors within the time window W and aggregate them to obtain the vector dataset of the business context information packet in the current time window. Then, aggregate the business probe result vectors and heartbeat signal vectors in the vector dataset according to the primary key ID of the business context information packet to obtain the relevant vector dataset. For example, the business probe result vectors and heartbeat signal vectors in the vector dataset are aggregated based on the primary key ID of the business context information packet to obtain the relevant vector dataset. The specific acquisition process is as follows: Obtain the primary key IDs of the business probe result vectors and heartbeat signal vectors in the vector dataset, where the primary key ID is a unique tracking ID for the actual business transaction; If the primary key ID of the business probe result vector or heartbeat signal vector in the vector dataset is the same as the primary key ID of the simulated text request probe, they will be aggregated into the relevant vector dataset. For example, the time window W is 120 seconds in a specific embodiment; The business probe result vector and heartbeat signal vector are separately partitioned and aggregated from the relevant vector dataset to obtain the business probe dataset and the heartbeat signal dataset, respectively. The minimum time difference p between the business probe dataset and the heartbeat signal dataset is calculated for each dataset. t ; The minimum time difference p between the business probe dataset and the heartbeat signal dataset t The relevant vector dataset and business probe dataset are aggregated to generate multi-source business data for the business context information package; For example, the minimum time difference p between the business probe dataset and the heartbeat signal dataset t The specific calculation process is as follows: Obtain the reporting timestamps of each business probe result vector from the business probe dataset, obtain the reporting timestamps of each heartbeat signal vector from the heartbeat signal dataset, and find the minimum difference Δt between the reporting timestamps of each business probe result vector and each heartbeat signal vector. min Calculate the minimum time difference p between the business probe dataset and the heartbeat signal dataset. t : , Wherein, τ is a preset decay constant, and in a specific embodiment, τ is 10s.

[0019] Step S2: Analyze the health status of the heartbeat signal of the monitoring center platform within the current time window, analyze the comprehensive changes of the business probes of the monitoring center platform within the current time window, construct the abnormal pattern vector of the business probe dataset, analyze the silent failure status of the monitoring center platform within the current time window, and obtain silent failure data. Step S2 includes: Obtain the minimum time difference p between the business probe dataset and the heartbeat signal dataset from multi-source business data. t Set the time-series threshold p', when p t When p ≥ p', the time window W is not processed; otherwise, the time window W is modified to obtain the modified time window W'. For example, the specific calculation process for correcting the time window W' is as follows: ,in, As a preset minimum constant, in specific embodiments 10 -6 ; Get the last timestamp t of the heartbeat signal dataset last Get the timestamp t of the current moment. now Calculate the resting heart rate duration Δt = t now -t last Obtain the modified tolerance dynamic coefficient α´; For example, the end timestamp t of the heartbeat signal dataset last The specific acquisition process is as follows: Obtain the timestamp of the last heartbeat signal reported corresponding to the heartbeat signal vector from the heartbeat signal dataset, and record it as the end timestamp of the heartbeat signal dataset; For example, the specific calculation process for obtaining the modified tolerance dynamic coefficient α´ is as follows: Obtain the preset timeout tolerance coefficient α, and calculate the corrected tolerance dynamic coefficient α´=α×(W´ / W); Set the heart rate health variable H, the single probe variable q, and the comprehensive probe variable Q. Set the value of the single probe variable for each business probe result vector in the business probe dataset. When Δt≤α´·T, set H=1, otherwise set H=0. When there is a business probe result vector in the business probe dataset with a single probe variable of 0, set the comprehensive probe variable Q to 0, otherwise set the comprehensive probe variable Q to 1. For example, the specific process for setting the values ​​of single-probe variables for each business probe result vector in the business probe dataset is as follows: The core metrics are obtained from the business probe result vector. Specifically, the core metrics are: response time, response status code, and recognition similarity. The status code setting, response time threshold, and similarity identification threshold of the business probe result vector are set. In a specific embodiment, the response status code is 200, the response time threshold is 2000 milliseconds, and the similarity identification threshold is 0.85. When the response status code of the business probe result vector is the same as the status code setting value, the response time is less than the response time threshold, and the recognition similarity is greater than the recognition similarity threshold, the single probe variable of the business probe result vector is set to 1; otherwise, the single probe variable of the business probe result vector is set to 0. When the comprehensive probe variable Q=1, the business probe dataset is not processed. When the comprehensive probe variable Q=0, an abnormal mode vector Q´ is constructed, the silent failure variable S of the monitoring center platform in the current time window is set, and the heartbeat health variable H and comprehensive probe variable Q of the monitoring center platform in the current time window are obtained. When H=1 and Q=0, S is set to 1, otherwise S is set to 0. For example, when the comprehensive probe variable Q=0, the abnormal pattern vector Q´ of the business probe dataset is constructed. The specific construction process is as follows: Set the status code exception variable F of the business probe dataset. code Response abnormal variable F RT And identifying similar outlier variables F sim ; If any business probe result vector in the business probe dataset has a different status code setting value, then set F. code If it is 1, then set F. code =0; If the response time of any business probe result vector in the business probe dataset is greater than or equal to the response time threshold, then F is set. RT If it is 1, then set F. RT =0; If the similarity of any business probe result vector in the business probe dataset is less than or equal to the similarity threshold, then F is set. sim If it is 1, then set F. sim =0; The status code exception variable F of the business probe dataset code Response abnormal variable F RT And identifying similar outlier variables F sim The data is aggregated to obtain the anomaly pattern vector Q´ of the business probe dataset; When S=0, it is determined that there is no silent failure in the monitoring center platform within the current time window. When S=1, the severity D of the silent failure of the monitoring center platform within the current time window is calculated, the timeliness feature value m is set, and the severity D of the silent failure within the first m correction time windows W´ within the current time window is obtained. m The silent failure trend slope k of the monitoring center platform within the current time window is calculated. The silent failure trend slope k, silent failure variable S, silent failure severity D and abnormal mode vector Q´ are aggregated to obtain the silent failure data of the monitoring center platform within the current time window. For example, the specific calculation process for the silent failure severity D of the monitoring center platform is as follows: Obtain the anomaly pattern vector Q' of the business probe dataset, and extract the status code anomaly variable F of the business probe dataset from the anomaly pattern vector Q'. code Response abnormal variable F RT And identifying similar outlier variables F sim ; The severity of the silent failure of the monitoring center platform within the current time window is calculated as D=w. code ×F code +w RT ×F RT +w sim ×F sim w code w RT and w sim These are the preset status code weighting coefficient, response time weighting coefficient, and similarity weighting coefficient, respectively, where w code w RT and w sim All are greater than 0 and their sum is 1; For example, the timeliness feature value m is set to further improve the timeliness of the warning; in the specific embodiment, the timeliness feature value m is 2. For example, the slope k of the silent failure trend of the monitoring center platform within the current time window is calculated using the following formula: k = (DD) m ) / (m×W´).

[0020] Step S3: Based on multi-source business data and silent failure data, analyze the overall health status of the text recognition service of the monitoring center platform within the current time window to obtain target health data; Step S3 includes: Acquire the silent failure data and business context information packet of the monitoring center platform within the current time window, calculate the heartbeat stability value G of the monitoring center platform within the current time window, calculate the probe call success rate U and probe identification quality value Y of the monitoring center platform within the current time window, and calculate the response performance value Z of the monitoring center platform within the current time window. For example, the heart rate stability value G of the monitoring center platform during text recognition is calculated as follows: Obtain the mean μ of the time interval between the reporting timestamps of adjacent heartbeat signal vectors in the heartbeat signal dataset from the multi-source business data. t and standard deviation σ t The heart rate stability value of the monitoring center platform during the text recognition process is calculated as G = exp(-(σ t / μ t )); For example, the success rate U of probe calls on the monitoring center platform within the current time window is calculated as follows: Obtain the values ​​of single probe variables for each business probe result vector within the current time window from the multi-source business data. Also obtain the total number N of business probe result vectors with a single probe variable value of 1 from the multi-source business data. Finally, obtain the total number N of all business probe result vectors from the multi-source business data. sum The success rate of probe calls within the current time window is calculated as U = N / N. sum ; For example, to calculate the probe identification quality value Y of the monitoring center platform within the current time window, the specific calculation process is as follows: calculate the average identification similarity of the result vectors of each business probe, and record it as the probe identification quality value Y of the monitoring center platform within the current time window; For example, to calculate the response performance value Z of the monitoring center platform within the current time window, the specific calculation process is as follows: The average response time T of each business probe result vector is obtained from the business probe dataset of multi-source business data. RT Obtain the average response time μ of the business probe result vectors of the monitoring center platform within each historical time window. RT and standard deviation σ RT Calculate the response performance value Z of the monitoring center platform within the current time window: Z = exp(-(T) RT -μ RT ) / σ RT ); The monitoring center platform's center jump stability value G, probe call success rate U, probe identification quality value Y, and response performance value Z are obtained using preset weighting coefficients λ. G , λ U , λY and λ Z Based on the silent failure data from the monitoring center platform within the current time window, the weighting coefficient λ is adjusted. G , λ U , λ Y and λ Z Perform dynamic correction; For example, based on the silent failure data of the monitoring center platform within the current time window, the repetition coefficient λ is adjusted. G , λ U , λ Y and λ Z Dynamic correction is performed, and the specific process is as follows: Obtain the preset weight coefficient λ G , λ U , λ Y and λ Z Obtain the response anomaly variable F from the anomaly pattern vector Q´ in the silent failure data of the monitoring center platform within the current time window. RT And identifying similar outlier variables F sim Obtain the silent failure variable S from the silent failure data and set the dynamic adjustment coefficient γ, where 1>γ>0; When S=1 and F sim When =1, then for λ Y Adjust upwards for λ. G and λ Z The adjustment will be made downward, specifically: λ Y Multiply by (1+γ) to upregulate, respectively, λ G and λ Z Multiply by γ to downregulate; When S=1 and F RT When =1, then for λ Z Adjust upwards for λ. G The adjustment will be made downward, specifically: λ Z Up-regulate by multiplying by (1+γ), and then adjust λ. G Multiply by γ to downregulate; When S=0, the weighting coefficient λ is not applied. G , λ U , λ Y and λ Z Make adjustments; Obtain the dynamically adjusted weight coefficients λ G , λ U , λ Y and λ Z After normalization, we obtain λ. (△,G) , λ (△,U) , λ (△,Y) and λ (△,Z) , where λ (△,G) , λ(△,U) , λ (△,Y) and λ (△,Z) The normalization process is the same for all of them, λ (△,G) =λ (△,G) / (λ (△,G) +λ (△,Y) +λ (△,U) +λ (△,Z) ); The static failure penalty item P of the calculation monitoring center platform within the current time window. S Calculate the comprehensive health score A of the text recognition service of the monitoring center platform within the current time window, calculate the change value △A of the comprehensive health score of the text recognition service of the monitoring center platform within the current time window, and combine the comprehensive health score A and the change value △A to obtain the target health data of the text recognition service of the monitoring center platform within the current time window. For example, the static failure penalty item P of the monitoring center platform within the current time window. S The specific calculation process is as follows: Obtain the severity D of the silent failure from the silent failure data. When S=0, P S =0, when S=1, P S =D×(1-λ Y )×η, where η is a preset penalty intensity coefficient, and in a specific embodiment, η is taken as 0.5; For example, the comprehensive health score A and change value △A of the text recognition service within the current time window are calculated using the following formula: , Obtain the previous time window of the monitoring center platform in the current time window, calculate the difference between the text recognition service in the current time window and the text recognition service in the previous time window, and obtain the change value △A.

[0021] Step S4: Based on the target health data, issue an anomaly alert for the text recognition service of the monitoring center platform within the current time window; Step S4 includes: Obtain the target health data of the text recognition service of the monitoring center platform within the current time window, and obtain the comprehensive health score A and change value △A of the text recognition service of the monitoring center platform within the current time window from the target health data; Set a comprehensive health score threshold A´ and a health score change rate threshold △A´, where A´>0 and △A´>0; For example, in a specific embodiment, △A´ takes the value of 0.05, and A´ takes the value of 0.65; When A < A', a serious abnormal alarm is issued for the text recognition service of the monitoring center platform within the current time window, and the text recognition service of the monitoring center platform in the current time window is stopped; When A ≥ A' and △A > △A', an abnormal alarm is issued for the text recognition service of the monitoring center platform within the current time window, and the text recognition service of the monitoring center platform in the current time window is restarted; When A ≥ A' and △A ≤ △A', a detection reminder is sent to the manager of the text recognition service through the monitoring center platform to remind the manager to perform service function detection on the text recognition service of the monitoring center platform.

[0022] In order to better implement the above method, an intelligent monitoring system applied to text recognition is also provided. The system includes a service data construction module, a silent failure analysis module, a service health analysis module and an abnormal alarm module; The service data construction module is configured to obtain original text data of the two-dimensional code on drug invoices, and construct multi-source business data of a business context information package for the original text data; The silent failure analysis module is configured to analyze the silent failure condition of the monitoring center platform within the current time window, and obtain silent failure data; The service health analysis module is configured to analyze the comprehensive health condition of the text recognition service of the monitoring center platform within the current time window, and obtain target health data; The abnormal alarm module is configured to issue an abnormal alarm for the text recognition service of the monitoring center platform within the current time window.

[0023] Wherein, the service data construction module includes a service information package generation unit and a service data construction unit; The service information package generation unit is configured to extract and parse fields from the original text data, and generate the business context information package; The service data construction unit is configured to construct multi-source business data of the business context information package of the original text data according to the business context information package of the original text data.

[0024] Wherein, the silent failure analysis module includes a service probe change analysis unit and a silent failure analysis unit; The service probe change analysis unit is configured to analyze the health condition of the heartbeat signal of the monitoring center platform within the current time window, analyze the comprehensive change condition of the service probe of the monitoring center platform within the current time window, and construct an abnormal mode vector of the service probe data set; The silent failure analysis unit is configured to analyze the silent failure condition of the monitoring center platform within the current time window according to the abnormal mode vector of the service probe data set, and obtain the silent failure data.

[0025] The service health analysis module includes a weight dynamic adjustment unit and a service health analysis unit. The weighted dynamic correction unit is used to dynamically correct the preset weight coefficients of the center jump stability value, probe call success rate, probe identification quality value and response performance value of the monitoring center platform based on the silent failure data. The service health analysis unit is used to calculate the static failure penalty items of the monitoring center platform within the current time window, analyze the overall health status of the text recognition service of the monitoring center platform within the current time window, and obtain target health data.

[0026] The abnormal alarm module includes an abnormal alarm unit; The anomaly alarm unit is used to identify anomalies in the text recognition service of the monitoring center platform within the current time window based on the target health data, and to issue anomaly alarms for the text recognition service.

[0027] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. An intelligent monitoring method applied to text recognition, characterized in that, The method includes: Step S1: Obtain the raw text data of the drug invoice QR code and construct the business context information package of the raw text data as multi-source business data; Step S2: Analyze the health status of the heartbeat signal of the monitoring center platform within the current time window, analyze the comprehensive changes of the business probes of the monitoring center platform within the current time window, construct the abnormal pattern vector of the business probe dataset, analyze the silent failure status of the monitoring center platform within the current time window, and obtain silent failure data. Step S3: Based on multi-source business data and silent failure data, analyze the overall health status of the text recognition service of the monitoring center platform within the current time window to obtain target health data; Step S4: Based on the target health data, issue an anomaly alert for the text recognition service of the monitoring center platform within the current time window.

2. The intelligent monitoring method for text recognition according to claim 1, characterized in that, Step S2 includes: Obtain the minimum time difference p between the business probe dataset and the heartbeat signal dataset from multi-source business data. t Set the time-series threshold p', when p t When p ≥ p', the time window W is not processed; otherwise, the time window W is modified to obtain the modified time window W'. Get the last timestamp t of the heartbeat signal dataset last Get the timestamp t of the current moment. now Calculate the resting heart rate duration Δt = t now -t last Obtain the modified tolerance dynamic coefficient α´; Set the heart rate health variable H, the single probe variable q, and the comprehensive probe variable Q. Set the value of the single probe variable for each business probe result vector in the business probe dataset. When Δt≤α´·T, set H=1, otherwise set H=0. When there is a business probe result vector in the business probe dataset with a single probe variable of 0, set the comprehensive probe variable Q to 0, otherwise set the comprehensive probe variable Q to 1. When the comprehensive probe variable Q=1, the business probe dataset is not processed. When the comprehensive probe variable Q=0, an abnormal mode vector Q´ is constructed, the silent failure variable S of the monitoring center platform in the current time window is set, and the heartbeat health variable H and comprehensive probe variable Q of the monitoring center platform in the current time window are obtained. When H=1 and Q=0, S is set to 1, otherwise S is set to 0. When S=0, it is determined that there is no silent failure in the monitoring center platform within the current time window. When S=1, the severity D of the silent failure of the monitoring center platform within the current time window is calculated, the timeliness feature value m is set, and the severity D of the silent failure within the first m correction time windows W´ within the current time window is obtained. m The silent failure trend slope k of the monitoring center platform within the current time window is calculated. The silent failure trend slope k, silent failure variable S, silent failure severity D and abnormal mode vector Q´ are aggregated to obtain the silent failure data of the monitoring center platform within the current time window.

3. The intelligent monitoring method for text recognition according to claim 1, characterized in that, Step S1 includes: The raw text data of the drug invoice QR code is obtained from the scanner device, and the fields are extracted and parsed from the raw text data to obtain the business context information package; According to a preset fixed period, the monitoring center platform sends heartbeat monitoring requests to the monitoring center platform interface. After receiving the heartbeat monitoring requests, the monitoring center platform updates the stored service heartbeat data table and generates a heartbeat signal vector. The invoice number is obtained from the business context information package, a simulated text request probe is constructed, the simulated text request probe is sent to the monitoring center platform, the core quantitative indicators returned by the monitoring center platform based on the simulated text request probe are obtained, and the business probe result vector of the business context information package is generated. Get the timestamp t of the business context information packet in the current time window, get the preset time window W, and based on the timestamp t, extract all the business probe result vectors and heartbeat signal vectors within the time window W and aggregate them to obtain the vector dataset of the business context information packet in the current time window. Then, aggregate the business probe result vectors and heartbeat signal vectors in the vector dataset according to the primary key ID of the business context information packet to obtain the relevant vector dataset. The business probe result vector and heartbeat signal vector are separately partitioned and aggregated from the relevant vector dataset to obtain the business probe dataset and the heartbeat signal dataset, respectively. The minimum time difference p between the business probe dataset and the heartbeat signal dataset is calculated for each dataset. t ; The minimum time difference p between the business probe dataset and the heartbeat signal dataset t The relevant vector dataset and business probe dataset are aggregated to generate multi-source business data for the business context information package.

4. The intelligent monitoring method for text recognition according to claim 1, characterized in that, Step S3 includes: Acquiring multi-source service data including silent failure data and service context information package of the monitoring center platform within the current time window, calculating the heartbeat stability value G of the monitoring center platform within the current time window, calculating the probe call success rate U and the probe identification quality value Y of the monitoring center platform within the current time window, and calculating the response performance value Z of the monitoring center platform within the current time window; The monitoring center platform's center jump stability value G, probe call success rate U, probe identification quality value Y, and response performance value Z are obtained using preset weighting coefficients λ. G , λ U , λ Y and λ Z Based on the silent failure data from the monitoring center platform within the current time window, the weighting coefficient λ is adjusted. G , λ U , λ Y and λ Z Perform dynamic correction; The static failure penalty item P of the calculation monitoring center platform within the current time window. S The comprehensive health score A of the text recognition service of the monitoring center platform within the current time window is calculated, and the change value △A of the comprehensive health score of the text recognition service of the monitoring center platform within the current time window is calculated. The comprehensive health score A and the change value △A are aggregated to obtain the target health data of the text recognition service of the monitoring center platform within the current time window.

5. The intelligent monitoring method for text recognition according to claim 1, characterized in that, Said step S4 comprises: Acquiring target health data of the text recognition service of the monitoring center platform within the current time window, and obtaining the comprehensive health score A and the change value △A of the text recognition service of the monitoring center platform within the current time window from the target health data; Setting a comprehensive health score threshold A´ and a health score change rate threshold △A´, wherein A´>0 and △A´>0; When A<A´, issuing a serious abnormality alarm for the text recognition service of the monitoring center platform within the current time window, and stopping the text recognition service of the monitoring center platform within the current time window; When A≥A´ and △A>△A´, issuing an abnormality alarm for the text recognition service of the monitoring center platform within the current time window, and restarting the text recognition service of the monitoring center platform within the current time window; When A≥A´ and △A≤△A´, sending a detection reminder to the manager of the text recognition service through the monitoring center platform, to remind the manager to perform service function detection on the text recognition service of the monitoring center platform.

6. An intelligent monitoring system for text recognition, used to execute the intelligent monitoring method for text recognition as described in any one of claims 1-5, characterized in that, Said system comprises a service data construction module, a silent failure analysis module, a service health analysis module and an abnormality alarm module; Said service data construction module is configured to acquire original text data of the two-dimensional code on drug invoices, and construct multi-source service data of the service context information package for the original text data; Said silent failure analysis module is configured to analyze the silent failure condition of the monitoring center platform within the current time window, and obtain silent failure data; Said service health analysis module is configured to analyze the comprehensive health condition of the text recognition service of the monitoring center platform within the current time window, and obtain target health data; Said abnormality alarm module is configured to issue an abnormality alarm for the text recognition service of the monitoring center platform within the current time window.

7. The intelligent monitoring system for text recognition according to claim 6, characterized in that, Said service data construction module comprises a service information package generation unit and a service data construction unit; Said service information package generation unit is configured to perform field extraction and parsing from the original text data, and generate the service context information package; Said service data construction unit is configured to construct, according to the service context information package of the original text data, the multi-source service data of the service context information package for the original text data.

8. The intelligent monitoring system for text recognition according to claim 6, characterized in that, Said silent failure analysis module comprises a service probe change analysis unit and a silent failure analysis unit; Said service probe change analysis unit is configured to analyze the health condition of heartbeat signals of the monitoring center platform within the current time window, analyze the comprehensive change condition of service probes of the monitoring center platform within the current time window, and construct an abnormal mode vector of a service probe data set; The silent failure analysis unit is used to analyze the silent failure status of the monitoring center platform within the current time window based on the abnormal pattern vector of the business probe dataset, and obtain silent failure data.

9. The intelligent monitoring system for text recognition according to claim 6, characterized in that, The service health analysis module includes a weight dynamic correction unit and a service health analysis unit; The weight dynamic correction unit is used to dynamically correct the preset weight coefficients of the center jump stability value, probe call success rate, probe identification quality value and response performance value of the monitoring center platform based on the silent failure data. The service health analysis unit is used to calculate the static failure penalty items of the monitoring center platform within the current time window, analyze the overall health status of the text recognition service of the monitoring center platform within the current time window, and obtain target health data.

10. The intelligent monitoring system for text recognition according to claim 6, characterized in that, The anomaly alarm module includes anomaly alarm units; The anomaly alarm unit is used to identify anomalies in the text recognition service of the monitoring center platform within the current time window based on the target health data, and to issue an anomaly alarm for the text recognition service.

Images