A method and system for intelligent early warning and intelligent attribution analysis of indicators

By employing an indicator fluctuation self-optimization algorithm and multi-dimensional decomposition intelligent attribution analysis, the existing indicator management platform addresses the issues of static setting of early warning thresholds and insufficient attribution analysis. This enables intelligent dynamic early warning and multi-dimensional analysis, improving the accuracy of early warnings and the timeliness of business decisions.

CN123117873APending Publication Date: 2026-11-17王文逸
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411321549.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-09-23
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing indicator management platforms lack dynamic flexibility and intelligence in setting early warning thresholds, resulting in frequent false alarms. They also lack effective multi-dimensional attribution analysis capabilities, making it difficult to adapt to changes in the business environment and provide timely and accurate analytical support.

Method used

An intelligent dynamic early warning threshold setting is adopted based on an indicator fluctuation self-optimization algorithm. Through multi-dimensional decomposition and intelligent attribution analysis mechanism, an intelligent early warning and intelligent attribution analysis system for indicators is constructed, including indicator threshold self-optimization and intelligent attribution analysis method for abnormal indicators. Using additive, multiplicative, divisional and implicit relationship attribution calculation models, multi-angle analysis of business data is realized.

Benefits of technology

It enables intelligent and dynamic adjustment of early warning thresholds, reduces false alarms, improves the accuracy and timeliness of early warnings, provides multi-dimensional business analysis capabilities, can quickly identify the causes and extent of anomalies, and improves the timeliness and accuracy of business decisions.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123117873A_ABST
    Figure CN123117873A_ABST
Patent Text Reader

Abstract

This invention discloses an intelligent early warning and intelligent attribution analysis method and system for indicators, relating to the field of indicator platform technology. The indicator threshold self-optimization method of this invention includes initiating a timed task execution, reading historical data of the observed indicators, calculating future thresholds for the observed indicators according to a time cycle, setting new thresholds for the observed indicators for the next day, monitoring the thresholds of the observed indicators, and issuing business alarms. The intelligent attribution analysis method for abnormal indicators includes reading the relationship definition of the observed indicators, calculating abnormal attributions, calculating explicit relationship attributions, calculating implicit relationship attributions, locating abnormal attribution factors, analyzing the impact of abnormal attribution factors, and issuing an abnormal attribution analysis report. This invention reduces the investment of human resources, not only identifying key influencing factors of business performance but also revealing the degree of influence of these factors on the observed indicators, exhibiting a high degree of automation and intelligence.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention belongs to the field of indicator platform technology, and in particular relates to an intelligent indicator early warning and intelligent attribution analysis method and system. Background Technology

[0002] Currently, indicator platforms are playing an increasingly crucial role as important platforms for enterprise digital transformation. Traditional indicator platforms have made many efforts and improvements in areas such as centralized management of enterprise indicators. To solve the problem of timely early warning of business indicators, traditional indicator management platforms typically: set one or more early warning thresholds based on historical data and business objectives; these thresholds can be fixed values ​​or comparisons with historical performance or data from the same period. Specific rules for triggering early warnings are defined, such as indicators falling below or exceeding a certain fixed value, or comparisons with data from the previous period or the same period. At the same time, considering the temporal fluctuations of business activities, early warning periods are configured to adapt to the business characteristics of different time periods. However, with the increasing digitalization of enterprises and the massive accumulation of business data, problems have also emerged regarding how to effectively set indicator early warnings and the inadequacy of indicator attribution analysis.

[0003] Meanwhile, in terms of attribution analysis of indicators, traditional indicator management platforms are designed with the effective management of the enterprise indicator system in mind. Therefore, they lack the business expert experience, analytical models, and multi-dimensional data correlation analysis required to establish an indicator attribution analysis system. Even when indicator management platforms do exist that involve attribution analysis, the analytical methods they apply are often very simplistic, focusing on the breakdown of indicator structures. In practical applications, this type of analysis cannot fully cover the diverse attribution analysis needs of various business scenarios.

[0004] With the development of indicator platforms, the early warning and attribution analysis in existing indicator management platforms mainly have the following two prominent problems: (1) They require a lot of manual intervention and lack a mechanism for setting indicator early warning thresholds that are dynamic, effective and intelligent: The early warning settings of current indicator platforms generally lack dynamic flexibility and intelligent threshold adaptive adjustment. In actual business scenarios, the business environment and market conditions are constantly changing, so the early warning thresholds are required to adapt to these changes. For example, new product launches, large-scale promotional activities or holidays will lead to expected growth in traffic and sales, and the thresholds need to be dynamically adjusted accordingly to avoid false alarms. At the same time, the existing static indicator threshold settings lack close alignment with business objectives. For example, for a business objective aimed at improving user engagement, the early warning threshold may need to be adjusted according to changes in user activities. In addition, if the business indicator thresholds lack intelligent and dynamic adjustment, for business departments, the system is out of touch with the actual changes in business and frequently issues false alarms, which may cause users to lose confidence in the early warnings. This phenomenon is called "early warning fatigue". In order to adapt to the needs of business development and change, some indicator platforms provide a mechanism for manual intervention in indicator settings. Although this reduces the frequency of indicator false alarms to a certain extent, it obviously poses risks of high operating costs and reliability. (2) Lack of effective intelligent attribution analysis: As a method for in-depth understanding of business performance and the root causes of problems, indicator attribution analysis is both an opportunity and a challenge for business personnel. Business personnel often face the following challenges when conducting indicator attribution analysis. First, business personnel need to have a certain data analysis foundation, including understanding of statistical principles, mastery of data analysis tools, and sensitivity to business data. Attribution analysis often involves complex data processing and analysis model construction, which may be an insurmountable obstacle for business personnel without relevant background. In addition, with the continuous accumulation of business big data and the development of artificial intelligence technology, business personnel also need to continuously learn new tools and technologies to keep their analytical capabilities up-to-date. In the face of massive business data, it can better cover multi-dimensional indicator attribution. Second, even if individual business personnel can complete basic indicator attribution analysis, how to transform the analysis results into easily understandable business insights and make effective decisions based on them is also a challenge. Attribution analysis results may involve the interaction of multiple dimensions and factors. Business personnel need to be able to interpret these complex relationships and integrate them with actual business scenarios. Existing indicator management platforms often emphasize the management functions of the business indicator system, while lacking the accumulation of professional knowledge, experience, and analytical models for indicator attribution analysis. Therefore, for the reasons mentioned above, the indicator attribution analysis conclusions drawn by different business personnel based on existing indicator management platforms may vary significantly, or even be biased and incomplete.Meanwhile, during business development, although business personnel hope to provide attribution support for business metrics through the existing platform's metric analysis, the timeliness of such support is often unsatisfactory. Therefore, this technical solution proposes a method and system for intelligent metric early warning and intelligent attribution analysis based on an intelligent dynamic early warning algorithm for metric fluctuations and a multi-dimensional decomposition intelligent attribution modeling and analysis mechanism, which can effectively solve the shortcomings of current metric management platforms. Summary of the Invention

[0005] This invention provides a method and system for intelligent indicator early warning and intelligent attribution analysis. The problem it addresses is: how to use an indicator fluctuation self-optimization algorithm to achieve intelligent dynamic early warning of indicators and a modeling and analysis mechanism for multi-dimensional decomposition and intelligent attribution of business data, thereby improving the accuracy, effectiveness, and timeliness of overall business indicator early warning and attribution analysis, and providing business personnel with a systematic and intelligent means for intelligent indicator early warning and attribution analysis. Returning to the core business application scenario of the indicator management platform, the problem it faces is: when the market changes, how to assist business personnel in independently and effectively issuing early warnings for the business indicators they care about, and further determine whether the indicator is normal or abnormal in the current business scenario; if abnormal, what are the core reasons for the abnormality, and the degree of impact of different abnormal factors on the business indicators, so that business personnel can quickly locate the problem and take accurate actions to achieve the predetermined goals of the indicators. This solution uses an indicator fluctuation self-optimization algorithm to achieve intelligent dynamic early warning of indicators and constructs a modeling and analysis mechanism for multi-dimensional decomposition and intelligent attribution of business data, thereby solving the above problems.

[0006] To solve the above-mentioned technical problems, the present invention is achieved through the following technical solution:

[0007] The present invention provides an intelligent early warning and intelligent attribution analysis method for indicators, including an indicator threshold self-optimization method and an abnormal indicator intelligent attribution analysis method.

[0008] The self-optimization method for the indicator threshold includes the following steps:

[0009] S1, Start: The starting point for setting indicator thresholds and alarm processes;

[0010] S2. Start scheduled tasks: Automatically start calculation and monitoring tasks, including threshold self-optimization calculation and threshold anomaly monitoring, according to the set time period;

[0011] S3. Read historical data of observed indicators: Extract historical data of indicators from the historical database. This data flows from the historical database to the threshold self-optimization calculation and threshold monitoring module, providing a basis for dynamically calculating indicator thresholds.

[0012] S4. Calculate the future threshold of the observation index according to the time period: Use the algorithm provided in the observation index threshold calculation algorithm library to periodically calculate the optimal value of the index threshold, and output the result of the threshold optimization calculation to the index new threshold setting module.

[0013] S5. Set new thresholds for observation indicators for the next day: Set thresholds for observation indicators based on the optimization calculation results;

[0014] S6. Monitoring of observation index thresholds: Real-time monitoring of index values, comparison with set thresholds, and real-time monitoring of business indicators, comparison with newly set thresholds. Abnormal results are sent to the business alarm module.

[0015] S7. Business Alarm: When the indicator value exceeds the threshold and an anomaly is triggered, the alarm mechanism is activated to notify relevant personnel, and the attribution analysis of the abnormal indicator can be enabled.

[0016] The intelligent attribution analysis method for abnormal indicators includes the following steps:

[0017] P1, Start: The starting point for anomaly attribution analysis;

[0018] P2. Read the observed indicator relationship definition: Obtain the relationship between indicators from the indicator relationship definition in the database, including explicit and implicit relationships; the observed indicator definition is obtained from the indicator relationship definition library and output to the attribution calculation module.

[0019] P3. Attribution Calculation of Indicator Anomalies: Attribution analysis is performed on indicator anomalies to identify the influencing factors and their degree of influence, including attribution calculation of explicit relationship between indicators and attribution calculation of implicit relationship between indicators.

[0020] Attribution calculation of explicit relationships between indicators: Attribution analysis is performed on the direct relationships between indicators; if the observed indicators have additive, multiplicative, and divisive attribution relationships, then the attribution factors for anomalies are located; the attribution calculation of explicit relationships between indicators also includes attribution calculation using the correlation coefficient method, which is attribution analysis of indicators using the correlation coefficient method;

[0021] Implicit Relationship Attribution Calculation: Attribution analysis is performed on the implicit or indirect relationships between indicators;

[0022] P4. Locating the attribution factors of anomalies: Through the above series of calculations, the specific factors that cause the abnormality of the indicators are determined, and the attribution factor location and impact analysis module is output.

[0023] P5. Analysis of the impact of anomaly attribution factors: Through the above series of calculations, the degree of influence of key anomaly attribution factors on the indicators is determined.

[0024] P6. Anomaly Attribution Analysis Report: The final attribution analysis report is generated for business analysts to refer to.

[0025] Furthermore, the threshold self-optimization calculation is to calculate the rate of change of the indicator for each day over the past n days compared to the previous day, specifically as follows:

[0026] (D n -D n-1 ) / D n-1 *100%;

[0027] Where n is the nth day, D n This refers to the indicator data for day n.

[0028] Furthermore, the optimal value of the indicator threshold is calculated automatically and dynamically based on changes in business indicators to find and set the optimal threshold for the business observation indicator.

[0029] Calculate the rate of change of the indicator for each day over the past n days compared to the previous day, and sum the positive rates of change of all indicator values ​​(deposit) to obtain R. p The sum of the negative rates of change of all deposit values ​​is R. n Therefore, we can initially deduce that the upper and lower limits of tomorrow's (or future) deposit warning threshold are calculated as follows: [Today's deposit index value * R] p, Today's deposit metric value *R n 】;

[0030] This threshold is usually not the optimal warning threshold. Furthermore, an algorithm is used to adjust [R]. p, R n Recursive variation optimization, when R p At what limit value will the maximum number of threshold alarms under normal operation over the past n days not exceed one? The algorithm uses an iterative optimization method to successively reduce R. p The value is determined according to the formula: Upper Threshold = deposit * R p After calculating the new threshold, if the number of alarms in the past n days is zero under this threshold condition, then R is further reduced. p The value is determined according to the formula: Upper Threshold = deposit * R p Calculate a new threshold until, under this threshold assumption, the number of valid alarms in the past n days is greater than or equal to 1; then, the upper limit of the threshold can be finally set to the currently set R. p value of deposit*R p Similarly, it can be based on R nThe recursive optimization is used to calculate the final lower limit value of the set threshold. Since the historical data is continuously changing over an n-day window, the above optimization algorithm k can be calculated once a day at a fixed time to obtain the upper and lower limits of the future threshold of the indicator deposit, thereby realizing the threshold setting under the self-optimization algorithm of indicator fluctuation.

[0031] Furthermore, the additive attribution calculation uses an additive model to perform attribution analysis on the indicators; the additive model is as follows:

[0032] Assuming the observed index is Y, the index Y that can be decomposed using addition is as follows:

[0033] Y = X1 + X2 + ... + X n ;

[0034] The change ratio of the observed index Y is:

[0035] ΔY%=(Y1-Y0) / Y0;

[0036] Therefore, measuring a constituent element X i The influence of the change ratio ΔY% is calculated as follows:

[0037] ΔX i =(X i1 -X i0 ) / Y0.

[0038] Furthermore, the multiplicative attribution calculation uses a multiplicative model to perform attribution analysis on the indicators, and the multiplicative model is as follows:

[0039] Assuming the observed index is Y, the index Y that can be decomposed using multiplication is as follows:

[0040] Y = X1 * X2 * ... * X n

[0041] The change ratio of the observed index Y is:

[0042] ΔY%=(Y1-Y0) / Y0

[0043] Therefore, measuring a constituent element X i The influence of the change ratio ΔY% is calculated as follows (converted to logarithmic form):

[0044] LogX i =Log(Y1-Y0)*lg(X) i1 / X i0 );

[0045] ΔX i =LogX i / Y0=Log(Y1-Y0)*lg(X i1 / Xi0 ) / Y0.

[0046] Furthermore, the division-type attribution calculation uses a division model to perform attribution analysis on the indicators, and the division model is as follows:

[0047] Assuming the observed index is Y, the index Y that can be decomposed using division is as follows:

[0048] Y = M / N;

[0049] M and N are two sub-indicators associated with the observed index Y. The influence of M on the observed index Y is positively correlated, that is, the greater the change in M, the greater the positive change in Y. The influence of N on the observed index Y is negatively correlated, that is, the greater the change in N, the greater the negative change in Y.

[0050] First calculate [Y0,Y1,...,Y] separately. n-1 ] and [M0,M1,...,M n-1 Correlation coefficient value C ym , and [Y0,Y1,...,Y n-1 ] and [N0,N1,...,N n-1 Correlation coefficient value C yn Then, for each dataset, add one more Y value. n M n N n The correlation coefficient value was recalculated using numerical values ​​to obtain C. ym-new and C yn-new The change in correlation coefficient values ​​was calculated using several sets of data, and C was used. ym and C ym-new To further calculate which factor, M or N, has the greatest impact on Y, and what the degree of impact is; factors with large changes in correlation coefficient values ​​mean that they have a significant impact on the observed index Y anomaly, and the degree of impact on the observed index Y anomaly can be calculated by the rate of change of correlation coefficient values.

[0051] Furthermore, when changes in certain indicators that have an implicit relationship with the observed indicator cause an anomaly in the observed indicator, a dataset is created that contains the observed indicator Y and the indicators with an implicit relationship. The dataset format is [Y]. i ,X 隐性指标1-i, X 隐性指标2-i,..., X 隐性指标n-i ], where i represents each row in the dataset; when the observed indicator Y becomes abnormal in the nth row, the changes in the correlation coefficients between Y and the latent indicators in the following two datasets can be calculated to locate which latent indicator has an impact on the observed indicator, and to determine the magnitude of the impact; the dataset format is as follows:

[0052] Dataset 1 used to calculate correlation coefficient values:

[0053] [Y0,X 隐性指标1-0, X 隐性指标2-0,..., X 隐性指标n-0

[0054] Y1,X 隐性指标1-1, X 隐性指标2-1,..., X 隐性指标n-1

[0055] Y2,X 隐性指标1-2, X 隐性指标2-2,..., X 隐性指标n-2

[0056] Y3,X 隐性指标1-3, X 隐性指标2-3,..., X 隐性指标n-3 ......

[0058] Y n-1 ,X 隐性指标1-n-1 ,X 隐性指标2-n-1 , ...,X 隐性指标n-n-1 ]

[0059] Dataset 2 used to calculate correlation coefficient values:

[0060] [Y0,X 隐性指标1-0, X 隐性指标2-0,..., X 隐性指标n-0

[0061] Y1,X 隐性指标1-1, X 隐性指标2-1,..., X 隐性指标n-1

[0062] Y2,X 隐性指标1-2, X 隐性指标2-2,..., X 隐性指标n-2

[0063] Y3,X 隐性指标1-3, X 隐性指标2-3,..., X 隐性指标n-3 ......

[0065] Y n ,X 隐性指标1-n, X 隐性指标2-n,..., X 隐性指标n-n ].

[0066] A system for intelligent early warning and intelligent attribution analysis of indicators, comprising:

[0067] Scheduled task execution module: Schedules and executes threshold calculations for observed metrics at regular intervals. The execution cycle of scheduled tasks is set to different cycles based on the business scenario and specific metric business characteristics.

[0068] Historical data reading module: When the scheduled task execution module initiates a scheduled task, the historical data reading module is triggered to execute. The historical data reading module reads the past historical data of the observed indicators that require threshold calculation, so that the threshold optimization calculation module can use it for numerical calculation. The amount of historical data read depends on the business scenario and business characteristics of the observed indicator.

[0069] Threshold optimization calculation module: Based on the historical data of the observed indicators, it calls the algorithm in the early warning threshold algorithm library to calculate the upper and lower limits of the threshold at a future time.

[0070] Threshold update module: Updates and saves the new threshold for the observed indicator;

[0071] Metric threshold monitoring and alarm module: When the business progresses to the next cycle, it compares the actual business metric value with the new threshold value set for the observed metric; if the actual value exceeds the threshold value of the observed metric, an abnormal alarm is issued.

[0072] API output module: Used to send anomaly alerts to relevant business systems so that they ultimately reach the business personnel who are aware of them.

[0073] Early warning threshold algorithm management module: Used to manage the algorithms in the early warning threshold algorithm library, and can add or optimize the algorithms according to the actual application business needs;

[0074] The Observational Indicator Anomaly Attribution Analysis Module: When an observed indicator becomes abnormal, or when business personnel have questions about the anomaly of a certain business indicator, the Observational Indicator Anomaly Attribution Analysis Module can be triggered by initiating the anomaly attribution analysis. This module uses a predefined indicator relationship database to perform a comprehensive analysis of the anomalies in the observed indicators. The algorithm for the comprehensive analysis of anomaly attribution comes from the anomaly attribution algorithm library module. The algorithms in the anomaly attribution algorithm library module include explicit correlation indicator analysis algorithms and implicit correlation indicator analysis algorithms, which are correlation coefficient analysis methods. Explicit correlation indicator analysis algorithms include additive, multiplicative, and division types. The Observational Indicator Anomaly Attribution Analysis Module calculates and analyzes the factors that may cause the anomalies in the indicators from different perspectives and the quantifiable degree of influence. Finally, the Anomaly Attribution Report Analysis Module outputs the key factors and influence degree that caused the anomalies in the observed indicators, and, combined with past business operation experience, provides corresponding suggestions for improving the business indicators.

[0075] Anomaly Attribution Algorithm Management Module: This module manages the algorithms in the anomaly attribution algorithm library and allows for the continuous addition or optimization of relevant algorithms based on actual application business needs.

[0076] The hardware of this system includes:

[0077] Unified API Gateway Server: Serves as the unified access point for the system to the outside world, handling external requests and responses;

[0078] Firewall load balancing: Provides system network security and balanced traffic distribution functions;

[0079] Scheduled task execution module: Responsible for executing tasks within the system on a scheduled basis;

[0080] Historical data reading module: Reads historical data from the indicator historical database;

[0081] Threshold optimization calculation module + threshold update module: responsible for intelligent optimization calculation of indicators, and dynamically updating the threshold of indicators based on the calculation results;

[0082] K8S server: Kubernetes server, used for container orchestration and management; and can scalably support the running of various business function modules;

[0083] Indicator threshold monitoring and alarm module: Monitors the real-time value of indicators and issues an alarm when an anomaly occurs;

[0084] Anomaly Attribution Algorithm Library: Contains algorithms for anomaly attribution of indicators, including explicit indicator analysis algorithms and implicit indicator analysis algorithms;

[0085] Early warning threshold algorithm library: contains self-optimizing algorithms for indicator early warning;

[0086] MySQL server: A relational database server used to store and manage historical metric data;

[0087] Observation index anomaly attribution calculation module: performs attribution calculation for index anomalies;

[0088] Anomaly Attribution Algorithm Management Module: This module manages anomaly attribution algorithms.

[0089] Early warning threshold algorithm management module: This module manages the early warning threshold algorithm.

[0090] Anomaly Attribution Report Analysis Module: The module for generating anomaly attribution analysis reports;

[0091] API Output Module: Provides API interfaces to output the data and results generated by the system of this invention.

[0092] The receipt interaction relationships and flow of this system include:

[0093] If there are external requests, they enter the system through a unified API gateway server; the firewall load balancer distributes internal and external requests to different K8S servers for processing; the scheduled task execution module triggers periodic tasks of the historical data reading module and the threshold optimization calculation module; the historical data reading module extracts data from the historical indicator database for use by the threshold optimization calculation module; the threshold optimization calculation module + threshold update module calculates and updates the threshold of the indicator, and the new threshold is read and used by the indicator threshold monitoring and alarm module; the indicator threshold monitoring and alarm module receives real-time data, compares it with the threshold of the observed indicator, and issues an alarm when there is an anomaly; the anomaly attribution algorithm library and the early warning threshold algorithm library provide algorithm support for the observed indicator anomaly early warning and attribution calculation module; the observed indicator anomaly attribution calculation module retrieves algorithms from the anomaly attribution algorithm library and performs attribution analysis, and passes the analysis results to the anomaly attribution report analysis module; the anomaly attribution algorithm management module and the early warning threshold algorithm management module are responsible for the updating and maintenance of the algorithms; the API output module outputs the data and results of threshold alarms and anomaly analysis to internal and external systems or users through the API interface.

[0094] The present invention has the following advantages over the prior art:

[0095] (1) The adaptive early warning threshold setting achieves intelligent dynamic adjustment of the early warning threshold by introducing an indicator fluctuation self-optimization algorithm. This algorithm can monitor subtle changes in the business environment and market conditions in a timely manner and automatically adjust the early warning parameters accordingly, ensuring that the sensitivity and accuracy of the early warning system keep pace with actual business needs. In traditional early warning systems, static threshold settings often cannot adapt to rapidly changing business scenarios, leading to frequent false alarms and early warning fatigue, which reduces the trust of business personnel in the system. The adaptive threshold optimization function of this invention can accurately capture the normal fluctuation range of business indicators through the algorithm's self-optimization of the early warning threshold, and automatically adjust the threshold when necessary, thereby greatly reducing the occurrence of false alarms.

[0096] (2) Adaptive threshold optimization can also provide customized early warning services based on different business objectives. For example, during new product launches or large-scale promotional events, the system can identify the normal growth trend of business indicators and adjust the early warning threshold accordingly to avoid false alarms caused by business growth. This intelligent early warning mechanism not only improves the accuracy of early warnings but also reduces the workload of business personnel, allowing them to focus more on business development and decision-making.

[0097] (3) This invention also reduces reliance on business personnel through automated early warning threshold processes, thereby reducing human resource investment. This cost-effectiveness improvement is not only reflected in direct financial savings, but also in improved overall operational efficiency and response speed; by reducing manual intervention, the system can respond more quickly to market changes, adjust early warning strategies in a timely manner, and ensure the timeliness and effectiveness of business decisions. This efficient early warning mechanism and cost control strategy enable enterprises to maintain competitiveness in a highly competitive market and achieve sustainable development;

[0098] (4) Multi-dimensional intelligent attribution analysis is another innovative function of this invention. It provides a comprehensive and in-depth business analysis method by comprehensively considering the multi-dimensional correlations of business indicator data. This mechanism can not only identify the key influencing factors of business performance, but also reveal the degree of influence of these factors on the observed indicators. Compared with traditional single structured attribution analysis, the multi-dimensional correlation intelligent attribution analysis of this invention can more accurately identify and explain business problems. It constructs a three-dimensional business attribution analysis model by comprehensively considering the explicit and implicit relationships between indicators, as well as the correlation data of multiple dimensions such as additive, multiplicative, and divisive relationships that are explicitly related to the indicators. At the same time, correlation analysis can also discover the influence of certain implicit indicators on the observed indicators. The above attribution analysis model can comprehensively reveal the internal logic of business performance from different perspectives, helping business personnel to understand the reasons for business anomalies more comprehensively.

[0099] (5) The multi-dimensional association intelligent attribution analysis of the present invention also has the characteristics of high automation and intelligence; through the pre-set attribution algorithm and model, it can automatically extract key information from massive business data and perform in-depth analysis and attribution; this automated analysis process not only improves the efficiency of analysis, but also reduces the interference of human operation, and ensures the objectivity and accuracy of the analysis results.

[0100] (6) Compared with traditional indicator management platforms, this invention not only addresses the shortcomings mentioned in the background section but also significantly improves the timeliness of business operations. It can quickly capture and respond to business changes, providing timely analytical support for business decisions. This rapid response capability ensures that decisions are synchronized with market trends during business development, thereby quickly adapting to market dynamics. With continuous learning and optimization of the model, this invention can also accumulate rich professional knowledge and experience, making anomaly warnings and attribution analysis results more accurate, reducing bias, and improving the depth and breadth of analysis. Furthermore, due to the improved accuracy and timeliness of the warning and attribution analysis involved in this invention, business users' trust in the system has been significantly enhanced. This increased trust not only strengthens users' reliance on intelligent warning and attribution analysis functions but also promotes the wider application of intelligent warning and attribution analysis in actual business operations.

[0101] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0102] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0103] Figure 1 This is a flowchart illustrating the steps of an intelligent early warning and intelligent attribution analysis method for indicators according to the present invention.

[0104] Figure 2 A schematic diagram illustrating the steps and principles of intelligent dynamic threshold optimization and early warning for indicators, intelligent observation of indicators, and attribution analysis of indicators.

[0105] Figure 3 The present invention provides a method for intelligent early warning and intelligent attribution analysis of indicators, based on a program module structure, with the following processing steps:

[0106] Figure 4 This is a hardware structure diagram of an intelligent early warning and intelligent attribution analysis system for indicators according to the present invention. Detailed Implementation

[0107] 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.

[0108] like Figure 2 As shown, the present invention addresses the shortcomings of existing indicator management platform systems from the following aspects:

[0109] 1) Employ an indicator fluctuation-based self-optimization algorithm to achieve intelligent dynamic early warning of indicators:

[0110] In the course of business operations, business data is constantly changing. A policy, a trending event, or a marketing activity can all affect the changes in business observation metrics. These changes may be significant or prominent compared to routine, daily business metric values. Furthermore, the impact of an event or activity may continue to affect the observed metric values ​​for a period of time. Therefore, for metric fluctuations in such scenarios, the algorithm of this invention needs to automatically and dynamically find and set the optimal threshold for the business observation metrics based on changes in the business metrics. Additionally, the impact of an event or activity may continue to affect the observed metric values ​​for a period of time.

[0111] Therefore, from a business analysis perspective, significant changes in observed indicator values ​​do not necessarily indicate an anomaly. Thus, using traditional static threshold setting methods may obviously lead to false alarms. Therefore, the self-optimizing algorithm of this invention solves this by setting a historical data sampling period for the observed indicator value, such as 30 days or 90 days. The length of the time period is based on whether the observation period includes multiple possible business events or activities, giving the observed indicator a certain periodicity. In the following description of the invention, assuming a 30-day historical data period, the indicator to be observed is defined as the daily deposit balance: deposit. Therefore, the question to be answered is: how to calculate and set tomorrow's deposit warning threshold based on the historical data of the past 30 days? The self-optimizing algorithm of this invention calculates the rate of change of the indicator for each day compared to the previous day over the past 30 days, and sums the positive rate of change values ​​of all indicator values ​​(deposit) to obtain R. p The sum of the negative rates of change of all deposit values ​​is R. n Therefore, the preliminary calculation for tomorrow's deposit warning threshold is: [Today's deposit value * R] p, Today's deposit value * R n However, this threshold is usually not the optimal warning threshold. Furthermore, an algorithm is used to adjust [R]. p, R n Recursive optimization, the optimization algorithm is: when R p At what value will the maximum number of threshold alarms under normal operation over the past 30 days not exceed one? The algorithm uses an iterative optimization method, successively reducing R... p The value is determined according to the formula: Upper Threshold = deposit * R p。 After calculating the new threshold, if the number of alarms in the past 30 days is zero under this threshold condition, then continue to decrease R. p The value is determined according to the formula: Upper Threshold = deposit * R pA new threshold is calculated until, under this threshold assumption, the number of valid alarms in the past 30 days is greater than or equal to 1. Therefore, the upper limit of the threshold can be finally set to the currently set R. p Value. Similarly, it can be determined based on R. n The algorithm iteratively optimizes and calculates the final lower limit value of the set threshold. Since the historical data has a continuously changing 30-day window, the above optimization algorithm is calculated once a day at a fixed time to determine the upper and lower limits of the threshold for the indicator deposit for the next day; thus realizing the threshold setting under the self-optimization algorithm for indicator fluctuation.

[0112] Taking the daily deposit balance indicator as an example. Assume the deposit value on day 7 is 50, and the deposit value on day 8 is 60. Therefore, the rate of change of the indicator on day 8 relative to day 7 is: (60-50) / 50*100% = 20%; the daily rate of change of the indicator can be expressed as: (D n -D n-1 ) / D n-1 *100%. Here, n represents the nth day. D n It is the daily deposit balance on day n.

[0113] The self-optimization algorithm for indicators needs to set two values ​​for future thresholds based on historical data: an upper limit value and a deposit*R value. p and a lower bound value deposit*R n For example: The threshold for the indicator is [60, 20], where the upper limit is 60 and the lower limit is 20. An alarm is triggered when the indicator exceeds the upper limit of 60. An alarm is triggered when the indicator exceeds the lower limit of 20.

[0114] Therefore, the index self-optimization algorithm finds the optimal R based on historical data. p and R n, Thus, [deposit*R] is found. p, deposit*R n ]; R involved in the lower limit of the indicator n R involved in the lower limit of the indicator p The algorithms are the same. The only difference is that one uses the positive rate of change of the indicator, while the other uses the negative rate of change. Note: The lower limit is not R. p, The lower limit value is obtained through R n It is calculated by summing the positive rate of change of all indicator values ​​(deposit) in the historical data to obtain R. p The sum of the negative rates of change of all deposit values ​​is R. n .

[0115] The above solution dynamically and intelligently solves the problem of setting observation indicator thresholds. It uses an N-day window as a rolling period to periodically and dynamically calculate new observation indicator thresholds. Simultaneously, it utilizes the rate of change of the indicator over the past N-day window to recursively calculate the optimal observation indicator threshold. This intelligently improves the accuracy of indicator early warnings and reduces the false alarm rate.

[0116] 2) Construction of a modeling and analysis mechanism for multi-dimensional decomposition and intelligent attribution of business data:

[0117] In the attribution analysis of an indicator, in addition to identifying the factors affecting the observed indicator, it is also necessary to quantitatively calculate the degree of influence of each factor on the observed indicator by analyzing the relationship between the indicator and each factor. Through the degree of influence, a quantitative attribution analysis can identify the key factors affecting the observed indicator and the subsequent business actions that need to be taken to correct the operational deviations of the observed indicator and bring it back onto the operational track of achieving its goals.

[0118] In practical applications, the root cause of an anomaly in an observed indicator may be a change in certain atomic indicators that have an explicit relationship with that indicator, leading to the anomaly. Another scenario is that a change in certain indicators that have an implicit relationship with that indicator may also cause the anomaly. For the first scenario, the solution of this invention is to analyze the relationship between the observed indicator and its associated atomic indicators. The relationship between an observed indicator and its associated atomic indicators can be additive, multiplicative, or divisive. An additive relationship means that the observed indicator is composed of several atomic indicators that are additive, such as the total national sales figure being the sum of sales figures for each province or region. A multiplicative relationship means that the observed indicator is calculated by multiplying several indicators involved in the process flow, such as online transaction volume equals traffic multiplied by conversion rate multiplied by average order value. A divisive relationship means that the observed indicator is a rate-value indicator derived from several related indicators, such as order conversion rate equals the number of completed orders divided by total customer traffic. Therefore, this invention will decompose an observed indicator into atomic indicators affecting its anomaly based on these three relationships, i.e., to discover the possible factors causing the anomaly of the observed indicator. After identifying potential influencing factors, the degree of influence of these factors on the occurrence of anomalies in the observed indicators can be further determined using the algorithm described below:

[0119] 1) Additive relationship

[0120] Assuming the observed index is Y, the index Y that can be decomposed using addition is as follows:

[0121] Y = X1 + X2 + ... + X n

[0122] X represents all the relevant sub-indicators that make up the observed indicator Y. For example, the total national sales volume is composed of the sum of sales volumes from each province or region. Here, the total national sales volume is Y, and the sales volumes from each province or region are the different X's.

[0123] The change ratio of the observed index Y is:

[0124] ΔY%=(Y1-Y0) / Y0

[0125] The observed indicator Y changes daily. Y1 represents the observed indicator Y value for a given day, and Y0 represents the observed indicator Y value for the previous day. For example, the national total sales indicator Y was 15 million on July 15th and 14 million on July 14th; therefore, Y1 = 15 million and Y0 = 14 million.

[0126] Therefore, measuring a constituent element X i The influence of the change ratio ΔY% is calculated as follows:

[0127] ΔX i =(X i1 -X i0 ) / Y0

[0128] The constituent elements or sub-indicators of each observed indicator are different. For example, the national total sales index is Y. This index can be calculated using provinces as the unit element. That is, the national total sales index Y is the sum of the sales indices of all provinces. The constituent element of the national total sales index Y is the sales indices of each province. Alternatively, the national total sales index Y can also be calculated using regions as the unit element. Therefore, the constituent elements of the national total sales index Y are the sales indices of each region, such as East China, South China, North China, and Northeast China. For example, the national total sales index is Y. Here, Xi represents the total sales of Shanghai. Therefore, Xi1 represents the total sales of Shanghai on a certain day, while Xi0 represents the total sales of Shanghai on the day before that day. This allows us to calculate the impact of changes in Shanghai's sales over two days on the overall national total sales index.

[0129] Due to the different application scenarios of business metrics, the constituent element X i It will also be diverse, so X here i "i" is a general concept, referring to all sub-indicators that make up the observed indicator Y. Here, i can be 1, 2, or n, depending on the business indicator.

[0130] 2) Multiplicative Relationship

[0131] Assuming the observed index is Y, the index Y that can be decomposed using multiplication is as follows:

[0132] Y = X1 * X2 * ... * X n

[0133] The change ratio of the observed index Y is:

[0134] ΔY%=(Y1-Y0) / Y0

[0135] Therefore, measuring a constituent element X i The influence of the change ratio ΔY% is calculated as follows (converted to logarithmic form):

[0136] LogX i =Log(Y1-Y0)*lg(X) i1 / X i0 )

[0137] ΔX i =LogX i / Y0=Log(Y1-Y0)*lg(X i1 / X i0 ) / Y0

[0138] 3) Division-type relationship

[0139] Assuming the observed index is Y, the index Y that can be decomposed using division is as follows:

[0140] Y = M / N

[0141] M and N are two sub-indicators associated with the observed indicator Y. For example, the deposit customer conversion rate is indicator Y, which can be equal to the number of actual deposit customers divided by the number of actual bank customers. Therefore, here M = the number of actual deposit customers, and N = the number of actual bank customers. Thus, M has a positive correlation with the observed indicator Y; that is, the greater the change in M, the greater the positive change in Y. N has a negative correlation with the observed indicator Y; that is, the greater the change in N, the greater the negative change in Y.

[0142] This invention first calculates [Y0, Y1, ..., Y] separately. n-1 ] and [M0,M1,...,M n-1 Using the above example as an illustration, the suffixes M and N represent the actual number of deposit customers and the actual number of bank customers per day, respectively. Let M0 represent the actual number of deposit customers on July 15th, and M1 represent the actual number of deposit customers on July 16th. N0 represents the actual number of bank customers on July 15th, and N1 represents the actual number of bank customers on July 16th.

[0143] Correlation coefficient value C ym , and [Y0,Y1,...,Y n-1 ] and [N0,N1,...,N n-1 Correlation coefficient value C yn;

[0144] Using the above example as an illustration, C ym What is the correlation between the deposit customer conversion rate (Y) and the actual number of deposit customers? (C) yn This represents the correlation between the deposit customer conversion rate (Y) and the actual number of bank customers. This value can be calculated using the correlation coefficient method. While not part of this invention, this method will be applied here to address the calculation of the influence of a division-type relationship. Then, a new digit for Y is added to each dataset. n M n N n The correlation coefficient value was recalculated using numerical values ​​to obtain C. ym-new and C yn-new The change in correlation coefficient values ​​calculated from several sets of data, such as C. ym and C ym-new To further calculate which factor, M or N, has the greatest impact on Y, and the degree of its impact. In this invention, factors with large changes in correlation coefficient values ​​indicate a significant impact on the observed index Y anomaly, and the degree of impact on the observed index Y anomaly is calculated using the rate of change of the correlation coefficient value.

[0145] The above algorithm analyzes the possibility that the root cause of an anomaly in an observed metric might be changes in certain atomic metrics that have an explicit relationship with that observed metric, leading to the anomaly. For another scenario, where changes in certain metrics that have an implicit relationship with the observed metric cause the anomaly, the solution of this invention is to create a dataset of the observed metric Y and the metrics with implicit relationships, in the format [Y...]. i ,X 隐性指标1-i, X 隐性指标2-i,..., X 隐性指标n-i ], where i represents each row in the dataset. When the observed indicator Y becomes abnormal in the nth row, the changes in the correlation coefficients between Y and the latent indicators in the following two datasets can be calculated to locate which latent indicator has an impact on the observed indicator and how significant that impact is.

[0146] Dataset 1 used to calculate correlation coefficient values:

[0147] [Y0,X 隐性指标1-0, X 隐性指标2-0,..., X 隐性指标n-0

[0148] Y1,X 隐性指标1-1, X 隐性指标2-1,..., X 隐性指标n-1

[0149] Y2,X 隐性指标1-2, X 隐性指标2-2,..., X 隐性指标n-2

[0150] Y3,X 隐性指标1-3, X 隐性指标2-3,..., X 隐性指标n-3 ......

[0152] Y n-1 ,X 隐性指标1-n-1 ,X 隐性指标2-n-1 , ...,X 隐性指标n-n-1 ]

[0153] Dataset 2 used to calculate correlation coefficient values:

[0154] [Y0,X 隐性指标1-0, X 隐性指标2-0,..., X 隐性指标n-0

[0155] Y1,X 隐性指标1-1, X 隐性指标2-1,..., X 隐性指标n-1

[0156] Y2,X 隐性指标1-2, X 隐性指标2-2,..., X 隐性指标n-2

[0157] Y3,X 隐性指标1-3, X 隐性指标2-3,..., X 隐性指标n-3 ......

[0159] Y n ,X 隐性指标1-n, X 隐性指标2-n,..., X 隐性指标n-n ]

[0160] The above expressions are abstract representations of specific data.

[0161] Y and X 隐性指标1, X 隐性指标2,..., X 隐性指标n There are n latent indicators that are potentially correlated; the question is how to calculate the degree of correlation between Y and each latent indicator. Therefore, it is necessary to construct a dataset using historical data, specifically the values ​​of Y and the n latent indicators on different days. Here, the first subscript represents the n latent indicators, and the second n represents the value of the latent indicator on the nth day. X 隐性指标n-n This represents the specific value of the nth implicit indicator on day n.

[0162] like Figure 1 As shown, this invention solves the business application problems existing in the current indicator management platform from two aspects. One is intelligent dynamic threshold optimization and early warning for indicators, and the other is intelligent attribution analysis of observed indicators. Therefore, the steps to implement intelligent dynamic threshold optimization and early warning for indicators are:

[0163] 1. Historical data retrieval: Extract historical data of the indicators from the historical database of observed indicators.

[0164] 2. Intelligent self-optimization calculation of observation indicators: Using historical data and self-optimization calculation of observation indicators, the threshold of the observation indicators for the next day is dynamically calculated.

[0165] 3. Setting the threshold for the observation indicator for the next day: Based on the self-optimization calculation results, set the threshold for the observation indicator for the next day.

[0166] 4. Metric Threshold Monitoring: Monitor the current value of metrics in real time and compare it with the set metric threshold.

[0167] 5. Normal / Abnormal Judgment: Based on the comparison results, determine whether the indicator status is normal or abnormal. If the current indicator is greater than or equal to the upper threshold, or if the current indicator is less than or equal to the lower threshold, an abnormal alarm is issued. Otherwise, the currently observed indicator is in a normal state.

[0168] 6. Threshold monitoring and alarm: When the indicator status is abnormal, an alarm mechanism is triggered to notify relevant business personnel to intervene or perform attribution analysis, etc.

[0169] The specific data interaction is as follows: Historical data is read from the historical database of observation indicators and then flows to the intelligent self-optimization calculation module of observation indicators. The result of the intelligent self-optimization calculation is output to the threshold setting module for the next day of observation indicators. The comparison between real-time monitoring data and dynamically set thresholds is performed in the indicator threshold monitoring. If the comparison result is normal, monitoring continues; if an anomaly occurs, an alarm is pushed to the threshold monitoring.

[0170] The steps of intelligent attribution analysis for observed indicators are:

[0171] 1. Attribution Analysis of Anomalies in Observed Indicators: When an indicator's state is abnormal, attribution analysis is initiated. Relationship definitions between indicators are retrieved from the observed indicator relationship definition library. Based on these defined relationships, explicit and implicit relationship analyses are performed on the abnormal indicators.

[0172] 2. Analysis of Explicit Relationships Among Observed Indicators: Based on the defined relationships between the indicators, such as the following relationships, attribution calculations are performed. Specifically, this is divided into three types: additive attribution calculation, multiplicative attribution calculation, and divisional attribution calculation.

[0173] 3. Analysis of Implicit Relationships Among Observational Indicators: Based on the implicit relationships between the indicators, the correlation coefficient method is used to perform attribution calculations on the indicators.

[0174] 4. Impact Factor Identification and Impact Calculation: Whether performing explicit or implicit relationship analysis on abnormal indicators, the most critical business factor or sub-indicator for the abnormal impact of that indicator can ultimately be determined.

[0175] The specific data flow is as follows: The definitions of explicit and implicit relationship indicators are obtained from the observation indicator relationship definition library and flow to the attribution analysis module. The calculation results of attribution analysis (additive, multiplicative, divisive, and correlation coefficient methods) flow to the attribution factor influence analysis module. The results of abnormal attribution factor influence analysis are used for further decision analysis and support.

[0176] Indicator threshold self-optimization function:

[0177] 1. Start: The starting point for setting indicator thresholds and alarm processes.

[0178] 2. Start scheduled tasks: Automatically start relevant calculation and monitoring tasks according to the set time period; such as threshold self-optimization calculation and threshold anomaly monitoring.

[0179] 3. Read historical data of observed indicators: Extract historical data of indicators from the historical database to provide a basis for dynamically calculating indicator thresholds.

[0180] 4. Optimize and calculate future thresholds for observed indicators according to time periods: Use the algorithms provided in the observed indicator threshold calculation algorithm library to periodically calculate the optimal values ​​of indicator thresholds.

[0181] 5. Set new thresholds for observation indicators for the next day: Set thresholds for observation indicators based on the optimization calculation results.

[0182] 6. Monitoring of observation index thresholds: Real-time monitoring of index values ​​and comparison with set thresholds.

[0183] 7. Business Alarms: When an indicator value exceeds the threshold and triggers an anomaly, an alarm mechanism is activated to notify relevant personnel, and attribution analysis of the abnormal indicator can be enabled.

[0184] Intelligent attribution analysis function for abnormal indicators:

[0185] 1. Start: The starting point for anomaly attribution analysis.

[0186] 2. Read the relationship definition of observation indicators: Obtain the relationship between indicators from the indicator relationship definition in the database, including explicit and implicit relationships.

[0187] 3. Attribution Calculation for Indicator Anomalies: Attribution analysis is performed on indicator anomalies to identify potential influencing factors and their degree of influence. This includes attribution calculations for explicit and implicit relationships among indicators.

[0188] 4. Attribution Calculation of Explicit Relationships Between Indicators: This involves attribution analysis of the direct relationships between indicators. If the observed indicators exhibit the following relationships, this calculation includes: additive attribution calculation, multiplicative attribution calculation, and divisive attribution calculation.

[0189] 5. Implicit Relationship Attribution Calculation: Attribution analysis is performed on the implicit or indirect relationships between indicators.

[0190] 6. Additive Attribution Calculation: Attribution analysis of indicators is performed using an additive model.

[0191] 7. Multiplicative Attribution Calculation: Attribution analysis of indicators is performed using a multiplicative model.

[0192] 8. Division-based attribution calculation: Attribution analysis of indicators is performed using a division model.

[0193] 9. Correlation coefficient method attribution calculation: Use the correlation coefficient method to perform attribution analysis on the indicators.

[0194] 10. Identify the attribution factors for anomalies: Through the above series of calculations, determine the specific factors that cause the abnormality of the indicators.

[0195] 11. Analysis of the impact of anomaly attribution factors: Through the above series of calculations, the degree of influence of key anomaly attribution factors on the indicators is determined.

[0196] 12. Anomaly Attribution Analysis Report: Finally, an attribution analysis report is generated for business analysts to refer to.

[0197] Data interaction status:

[0198] Historical indicator data: Flows from the historical indicator database to the attribution calculation and threshold monitoring modules; Threshold optimization calculation results: Output to the indicator new threshold setting module; Real-time monitoring data: Real-time observation of business indicators and comparison with newly set thresholds; Abnormal results flow to the business alarm module; Observation indicator relationship definitions: Obtained from the indicator relationship definition library and output to the attribution calculation module; Attribution calculation results: Output to the abnormal attribution factor location and impact analysis module, and output to the abnormal attribution analysis report; Abnormal attribution analysis report: Outputs a report for business analysts' reference.

[0199] like Figure 3As shown, the main function of the scheduled task execution module is to schedule and execute threshold calculations for observed indicators on a regular basis. The execution cycle of scheduled tasks can be set to different cycles based on the business scenario and specific indicator characteristics, such as once a day or once an hour. When the scheduled task execution module initiates a scheduled task, it first triggers the execution of the "Read Historical Indicator Data Module". This module reads the historical data of the observed indicator that requires threshold calculation, so that it can be used by the threshold optimization calculation module for numerical calculation. The amount of historical data read depends on the business scenario and characteristics of the observed indicator, such as reading business indicator data from the past 24 hours or from the past 30 days. Based on the historical data of the observed indicator, the "Threshold Optimization Calculation Module" calls the algorithm in the "Early Warning Threshold Algorithm Library" to calculate the upper and lower limits of the threshold at a future time. The "Threshold Update Module" then updates and saves the new threshold for the observed indicator. When the business progresses to the next cycle, the "Indicator Threshold Monitoring and Alarm Module" compares the new threshold of the observed indicator with the actual business indicator value. If the actual value exceeds the threshold of the observed indicator, an abnormal alarm is issued. Anomaly alerts can be sent to relevant business systems via the "API output module" so that business personnel are ultimately informed. In this invention, the "early warning threshold algorithm management module" manages the algorithms in the "early warning threshold algorithm library," and can add or optimize relevant algorithms according to actual application business needs.

[0200] exist Figure 3 In this system, when observed indicators become abnormal, or when business personnel have questions about the anomalies of a certain business indicator, the "Observed Indicator Anomaly Attribution Analysis Module" can be triggered by initiating the anomaly indicator attribution analysis. This module primarily utilizes a predefined "indicator relationship database" to perform a comprehensive analysis of the causes of anomalies in the observed indicators. The algorithm for this comprehensive anomaly attribution analysis comes from the "Anomaly Attribution Algorithm Library" module. In this invention, the main algorithms for explicit correlation indicator analysis are additive, multiplicative, and divisive. The algorithm for implicit correlation indicator analysis is the correlation coefficient analysis method. The "Observed Indicator Anomaly Attribution Analysis Module" analyzes the factors that may cause anomalies in the indicators from different perspectives and the quantifiable degree of their impact. Finally, the "Anomaly Attribution Report Analysis Module" outputs the key factors and their impact causing the anomalies in the observed indicators, and, combined with past business operation experience, provides corresponding suggestions for improving the business indicators. Figure 3 The "Anomaly Attribution Algorithm Management Module" manages the algorithms in the "Anomaly Attribution Algorithm Library," allowing for the continuous addition or optimization of relevant algorithms based on actual application business needs.

[0201] like Figure 4As shown: the threshold optimization calculation module, the early warning threshold algorithm library, the observation index anomaly attribution calculation module, and the anomaly attribution algorithm library constitute the core foundation of this invention. In this invention, Figure 4 Each module in the system corresponds to a server cluster in the computer hardware architecture. Except for the historical data metrics, which runs on a MySQL server cluster, the other modules run on a Kubernetes server cluster.

[0202] The threshold optimization calculation module, the early warning threshold algorithm library, the observation index anomaly attribution calculation module, and the anomaly attribution algorithm library are implemented based on Python and its components. The algorithm models in the early warning threshold algorithm library and the anomaly attribution algorithm library are implemented using Python and corresponding development components. In the threshold optimization calculation module and the observation index anomaly attribution calculation module, you only need to call the algorithm model API provided by the algorithm library.

[0203] The scheduled task execution module, historical data reading module, indicator threshold update module, and indicator threshold monitoring and alarm module are implemented based on Java and Spring Boot components. Through these four modules, the server cluster completes the early warning threshold management and update mechanism for abnormal observed indicators involved in this invention, ensuring that the application of business indicators conforms to actual business scenarios and improving the effectiveness, accuracy, and timeliness of indicator application.

[0204] The anomaly attribution algorithm management module, anomaly attribution report module, and API output module are also implemented based on Java and Spring Boot components. Through the first two modules, the server cluster can manage anomaly attribution algorithms related to observed metrics and generate anomaly attribution analysis reports. This provides business personnel with an efficient and intelligent way to identify and analyze anomalies in business operations, quickly locate the source of problems, and provide data support for business decisions. Furthermore, the generated attribution analysis reports not only include detailed descriptions and cause analysis of the anomalies but also provide corresponding solutions and preventative measures. These reports can be customized according to the needs of different business personnel to ensure the accuracy and usability of the information. The API output module is an important auxiliary component responsible for providing the server cluster's processed warning thresholds or attribution analysis data to external systems or business users in a standardized interface format.

[0205] The historical data database cluster forms the storage foundation for the data involved in this invention. Structured data within the application system is stored using a database cluster approach; this invention employs a MySQL cluster server as support. However, the database cluster used in this invention is not limited to MySQL.

[0206] If there are external requests, they enter the system through the unified API gateway server.

[0207] Firewall load balancing distributes internal and external requests to different Kubernetes servers for processing. A scheduled task execution module triggers periodic tasks from the historical data reading module and the threshold optimization calculation module. The historical data reading module extracts data from the historical indicator database for use by the threshold optimization calculation module. The threshold optimization calculation module + threshold update module calculates and updates the indicator thresholds; the new thresholds are then read and used by the indicator threshold monitoring and alarm module. The indicator threshold monitoring and alarm module receives real-time data, compares it with the observed indicator thresholds, and issues alarms when anomalies occur. Anomaly attribution algorithm libraries and early warning threshold algorithm libraries provide algorithmic support for the observed indicator anomaly early warning and attribution calculation modules. The observed indicator anomaly attribution calculation module retrieves algorithms from the anomaly attribution algorithm library, performs attribution analysis, and passes the analysis results to the anomaly attribution report analysis module. The anomaly attribution algorithm management module and early warning threshold algorithm management module are responsible for algorithm updates and maintenance. The API output module outputs the threshold alarm and anomaly analysis data and results to internal and external systems or users via API interfaces.

[0208] The scheduled task execution module triggers periodic tasks for reading historical indicator data and calculating thresholds. The indicator threshold monitoring and alarm module reads the latest observed indicator thresholds, compares them with the current actual indicator values, and triggers an alarm action to the API output module when an anomaly is detected. The anomaly attribution algorithm management module and the early warning threshold algorithm management module manage and update algorithms, and provide algorithm services to the threshold optimization calculation module, the threshold update module, and the observed indicator anomaly attribution calculation module. The API output module outputs alarm data and anomaly analysis results.

[0209] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A method for intelligent early warning and intelligent attribution analysis of indicators, characterized in that, This includes methods for automatically optimizing indicator thresholds and intelligent attribution analysis methods for abnormal indicators; The self-optimization method for the indicator threshold includes the following steps: S1, Start: The starting point for setting indicator thresholds and alarm processes; S2. Start scheduled tasks: Automatically start calculation and monitoring tasks, including threshold self-optimization calculation and threshold anomaly monitoring, according to the set time period; S3. Read historical data of observed indicators: Extract historical data of indicators from the historical database. This data flows from the historical database to the threshold self-optimization calculation and threshold monitoring module, providing a basis for dynamically calculating indicator thresholds. S4. Calculate the future threshold of the observation index according to the time period: Use the algorithm provided in the observation index threshold calculation algorithm library to periodically calculate the optimal value of the index threshold, and output the result of the threshold optimization calculation to the index new threshold setting module. S5. Set new thresholds for observation indicators for the next day: Set thresholds for observation indicators based on the optimization calculation results; S6. Monitoring of observation index thresholds: Real-time monitoring of index values, comparison with set thresholds, and real-time monitoring of business indicators, comparison with newly set thresholds. Abnormal results are sent to the business alarm module. S7. Business Alarm: When the indicator value exceeds the threshold and an anomaly is triggered, the alarm mechanism is activated to notify relevant personnel, and the attribution analysis of the abnormal indicator can be enabled. The intelligent attribution analysis method for abnormal indicators includes the following steps: P1, Start: The starting point for anomaly attribution analysis; P2. Read the relationship definition of observation indicators: Obtain the relationship between indicators from the indicator relationship definition in the database, including explicit and implicit relationships; The definitions of observation indicators are obtained from the indicator relationship definition library and output to the attribution calculation module; P3. Attribution Calculation of Indicator Anomalies: Attribution analysis is performed on indicator anomalies to identify the influencing factors and their degree of influence, including attribution calculation of explicit relationship between indicators and attribution calculation of implicit relationship between indicators. Attribution calculation of explicit relationships between indicators: Attribution analysis of the direct relationships between indicators; If the observed indicators have additive, multiplicative, and divisive attribution relationships, then the abnormal attribution factors are located; the explicit relationship attribution calculation of the indicators also includes attribution calculation using the correlation coefficient method, which is to perform attribution analysis on the indicators using the correlation coefficient method. Implicit Relationship Attribution Calculation: Attribution analysis is performed on the implicit or indirect relationships between indicators; P4. Locating the attribution factors of anomalies: Through the above series of calculations, the specific factors that cause the abnormality of the indicators are determined, and the attribution factor location and impact analysis module is output. P5. Analysis of the impact of anomaly attribution factors: Through the above series of calculations, the degree of influence of key anomaly attribution factors on the indicators is determined. P6. Anomaly Attribution Analysis Report: The final attribution analysis report is generated for business analysts to refer to.

2. The intelligent early warning and intelligent attribution analysis method for indicators according to claim 1, characterized in that, The threshold self-optimization calculation is to calculate the rate of change of the indicator for each day over the past n days compared to the previous day, specifically as follows: (D n -D n-1 ) / D n-1 *100%; where n is the nth day, D n is the indicator data for the nth day.

3. The intelligent early warning and intelligent attribution analysis method for indicators according to claim 1, characterized in that, The optimal value of the indicator threshold is calculated automatically and dynamically based on changes in business indicators to find and set the optimal threshold for the business observation indicator. Calculate the rate of change of each day compared to the previous day in the past n days, and sum all the positive rate of change values of the deposit index value as R p , and sum all the negative rate of change values of the deposit index value as R n ; thereby preliminarily derive the preliminary tomorrow (or future) deposit warning threshold upper and lower limits = 【today's deposit index value * R p, today's deposit index value * R n 】; This threshold is usually not the optimal warning threshold. Furthermore, an algorithm is used to adjust [R]. p, R n Recursive variation optimization, when R p At what limit value will the maximum number of threshold alarms under normal operation over the past n days not exceed one? The algorithm uses an iterative optimization method to successively reduce R. p The value is determined according to the formula: Upper Threshold = deposit * R p After calculating the new threshold, if the number of alarms in the past n days is zero under this threshold condition, then R is further reduced. p The value is determined according to the formula: Upper Threshold = deposit * R p Calculate a new threshold until, under this threshold assumption, the number of valid alarms in the past n days is greater than or equal to 1; then, the upper limit of the threshold can be finally set to the currently set R. p value of deposit*R p Similarly, it can be based on R n The recursive optimization is used to calculate the final lower limit value of the set threshold. Since the historical data is continuously changing over an n-day window, the above optimization algorithm k can be calculated once a day at a fixed time to obtain the upper and lower limits of the future threshold of the indicator deposit, thereby realizing the threshold setting under the self-optimization algorithm of indicator fluctuation.

4. The intelligent early warning and intelligent attribution analysis method for indicators according to claim 1, characterized in that, The additive attribution calculation uses an additive model to perform attribution analysis on the indicators; the additive model is as follows: Assuming the observed index is Y, the index Y that can be decomposed using addition is as follows: Y=X1+X2+....+X n ; The change ratio of the observed index Y is: ΔY%=(Y1-Y0) / Y0; Therefore, measuring a constituent element X i The influence of the change ratio ΔY% is calculated as follows: ΔX i =(X i1 -X i0 ) / Y0。 5. The intelligent early warning and intelligent attribution analysis method for indicators according to claim 1, characterized in that, The multiplicative attribution calculation uses a multiplicative model to perform attribution analysis on the indicators. The multiplicative model is as follows: Assuming the observed index is Y, the index Y that can be decomposed using multiplication is as follows: Y=X1*X2*....*X n The change ratio of the observed index Y is: ΔY%=(Y1-Y0) / Y0 Therefore, measuring a constituent element X i The influence of the change ratio ΔY% is calculated as follows (converted to logarithmic form): LogX i =Log(Y1-Y0)*lg(X i1 / X i0 ); ΔX i =LogX i / Y0=Log(Y1-Y0)*lg(X i1 / X i0 ) / Y0。 6. The intelligent early warning and intelligent attribution analysis method for indicators according to claim 1, characterized in that, The division-based attribution calculation uses a division model to perform attribution analysis on the indicators. The division model is as follows: Assuming the observed index is Y, the index Y that can be decomposed using division is as follows: Y = M / N; M and N are two sub-indicators associated with the observed index Y. The influence of M on the observed index Y is positively correlated, that is, the greater the change in M, the greater the positive change in Y. The influence of N on the observed index Y is negatively correlated, that is, the greater the change in N, the greater the negative change in Y. First calculate [Y0,Y1,...,Y] separately. n-1 ] and [M0,M1,...,M n-1 Correlation coefficient value C ym ,and [Y0,Y1,...,Y n-1 ] and [N0,N1,...,N n-1 Correlation coefficient value C yn Then, for each dataset, add one more Y value. n M n N n The correlation coefficient value was recalculated using numerical values ​​to obtain C. ym-new and C yn-new The change in correlation coefficient values ​​was calculated using several sets of data, and C was used. ym and C ym-new To further calculate which factor, M or N, has the greatest impact on Y, and what the degree of impact is; factors with large changes in correlation coefficient values ​​mean that they have a significant impact on the observed index Y anomaly, and the degree of impact on the observed index Y anomaly can be calculated by the rate of change of correlation coefficient values.

7. The intelligent early warning and intelligent attribution analysis method for indicators according to claim 1, characterized in that, When changes in certain indicators that have an implicit relationship with the observed indicator cause an anomaly in the observed indicator, a dataset is created that contains the observed indicator Y and the indicators with an implicit relationship. The dataset format is [Y]. i ,X 隐性指标1-i, X 隐性指标2-i,..., X 隐性指标n-i ], where i represents each row in the dataset; when the observed indicator Y becomes abnormal in the nth row, the changes in the correlation coefficients between Y and the latent indicators in the following two datasets can be calculated to locate which latent indicator has an impact on the observed indicator, and to determine the magnitude of the impact; the dataset format is as follows: Dataset 1 used to calculate correlation coefficient values: [Y0,X 隐性指标1-0 ,X 隐性指标2-0 ,...,X 隐性指标n-0 Y1,X 隐性指标1-1 ,X 隐性指标2-1 ,...,X 隐性指标n-1 Y2,X 隐性指标1-2 ,X 隐性指标2-2 ,...,X 隐性指标n-2 Y3,X 隐性指标1-3 ,X 隐性指标2-3 ,...,X 隐性指标n-3 ....... Y n-1 ,X 隐性指标1-n-1 ,X 隐性指标2-n-1 ,...,X 隐性指标n-n-1 ] Dataset 2 used to calculate correlation coefficient values: [Y0,X 隐性指标1-0 ,X 隐性指标2-0 ,...,X 隐性指标n-0 Y1,X 隐性指标1-1 ,X 隐性指标2-1 ,...,X 隐性指标n-1 Y2,X 隐性指标1-2 ,X 隐性指标2-2 ,...,X 隐性指标n-2 Y3,X 隐性指标1-3 ,X 隐性指标2-3 ,...,X 隐性指标n-3 ....... Y n ,X 隐性指标1-n ,X 隐性指标2-n ,...,X 隐性指标n-n ]。 8. An intelligent indicator early warning and intelligent attribution analysis system, used to implement the intelligent indicator early warning and intelligent attribution analysis method as described in any one of claims 1-7, characterized in that, The system includes: Scheduled task execution module: Schedules and executes threshold calculations for observed metrics at regular intervals. The execution cycle of scheduled tasks is set to different cycles based on the business scenario and specific metric business characteristics. Historical data reading module: When the scheduled task execution module initiates a scheduled task, the historical data reading module is triggered to execute. The historical data reading module reads the past historical data of the observed indicators that require threshold calculation, so that the threshold optimization calculation module can use it for numerical calculation. The amount of historical data read depends on the business scenario and business characteristics of the observed indicator. Threshold optimization calculation module: Based on the historical data of the observed indicators, it calls the algorithm in the early warning threshold algorithm library to calculate the upper and lower limits of the threshold at a future time. Threshold update module: Updates and saves the new threshold for the observed indicator; Metric threshold monitoring and alarm module: When the business progresses to the next cycle, it compares the actual business metric value with the new threshold value set for the observed metric; if the actual value exceeds the threshold value of the observed metric, an abnormal alarm is issued. API output module: Used to send abnormal alerts to relevant business systems so that they can ultimately reach business personnel and make them aware of the situation; Early warning threshold algorithm management module: Used to manage the algorithms in the early warning threshold algorithm library, and can add or optimize the algorithms according to the actual application business needs; The Observational Indicator Anomaly Attribution Analysis Module: When an observed indicator becomes abnormal, or when business personnel have questions about the anomaly of a certain business indicator, the Observational Indicator Anomaly Attribution Analysis Module can be triggered by initiating the anomaly attribution analysis. This module uses a predefined indicator relationship database to perform a comprehensive analysis of the anomalies in the observed indicators. The algorithm for the comprehensive analysis of anomaly attribution comes from the anomaly attribution algorithm library module. The algorithms in the anomaly attribution algorithm library module include explicit correlation indicator analysis algorithms and implicit correlation indicator analysis algorithms, which are correlation coefficient analysis methods. Explicit correlation indicator analysis algorithms include additive, multiplicative, and division types. The Observational Indicator Anomaly Attribution Analysis Module calculates and analyzes the factors that may cause the anomalies in the indicators from different perspectives and the quantifiable degree of influence. Finally, the Anomaly Attribution Report Analysis Module outputs the key factors and influence degree that caused the anomalies in the observed indicators, and, combined with past business operation experience, provides corresponding suggestions for improving the business indicators. Anomaly Attribution Algorithm Management Module: This module manages the algorithms in the anomaly attribution algorithm library and allows for the continuous addition or optimization of relevant algorithms based on actual application business needs.

9. The intelligent early warning and intelligent attribution analysis system for indicators according to claim 8, characterized in that, The hardware of this system includes: Unified API Gateway Server: Serves as the unified access point for the system to the outside world, handling external requests and responses; Firewall load balancing: Provides system network security and balanced traffic distribution functions; Scheduled task execution module: Responsible for executing tasks within the system on a scheduled basis; Historical data reading module: Reads historical data from the indicator historical database; Threshold optimization calculation module + threshold update module: responsible for intelligent optimization calculation of indicators, and dynamically updating the threshold of indicators based on the calculation results; K8S server: Kubernetes server, used for container orchestration and management; and can scalably support the running of various business function modules; Indicator threshold monitoring and alarm module: Monitors the real-time value of indicators and issues an alarm when an anomaly occurs; Anomaly Attribution Algorithm Library: Contains algorithms for anomaly attribution of indicators, including explicit indicator analysis algorithms and implicit indicator analysis algorithms; Early warning threshold algorithm library: contains self-optimizing algorithms for indicator early warning; MySQL server: A relational database server used to store and manage historical metric data; Observation index anomaly attribution calculation module: performs attribution calculation for index anomalies; Anomaly Attribution Algorithm Management Module: This module manages anomaly attribution algorithms. Early warning threshold algorithm management module: This module manages the early warning threshold algorithm. Anomaly Attribution Report Analysis Module: The module for generating anomaly attribution analysis reports; API Output Module: Provides API interfaces to output the data and results generated by the system of this invention.

10. The intelligent early warning and intelligent attribution analysis system for indicators according to claim 9, characterized in that, The receipt interaction relationships and flow of this system include: If there are external requests, they enter the system through a unified API gateway server; the firewall load balancer distributes internal and external requests to different K8S servers for processing; the scheduled task execution module triggers periodic tasks of the historical data reading module and the threshold optimization calculation module; the historical data reading module extracts data from the historical indicator database for use by the threshold optimization calculation module; the threshold optimization calculation module + threshold update module calculates and updates the threshold of the indicator, and the new threshold is read and used by the indicator threshold monitoring and alarm module; the indicator threshold monitoring and alarm module receives real-time data, compares it with the threshold of the observed indicator, and issues an alarm when there is an anomaly; the anomaly attribution algorithm library and the early warning threshold algorithm library provide algorithm support for the observed indicator anomaly early warning and attribution calculation module; the observed indicator anomaly attribution calculation module retrieves algorithms from the anomaly attribution algorithm library and performs attribution analysis, and passes the analysis results to the anomaly attribution report analysis module; the anomaly attribution algorithm management module and the early warning threshold algorithm management module are responsible for the updating and maintenance of the algorithms; the API output module outputs the data and results of threshold alarms and anomaly analysis to internal and external systems or users through the API interface.