Business-aware database operation and maintenance AI diagnosis method and device

Abstract

This invention provides a business-aware AI-based diagnostic method and apparatus for database operation and maintenance. The method includes: collecting candidate database performance index data of a target database; correcting the candidate database performance index data based on the current version information of the target database to obtain database performance index data; determining the impact coefficient and business volume impact value of each database performance index on business data by combining business data and database performance index data; generating business data context by combining business data, a preset baseline business volume, impact coefficient, and business volume impact value; processing the business data by combining clustering algorithms, preset business semantics, and business types to determine the periodic characteristics of the business data; extracting features from the business data context, periodic features, normalized database performance index data, and business data; and using a pre-built target model to perform diagnostic reasoning on the extracted features to obtain a diagnostic result.

Description

Technical Field

[0001] The embodiments of the present invention relate to the field of computer technology, and in particular to a business-aware AI diagnostic method and apparatus for database operation and maintenance. Background Technology

[0002] As the infrastructure of a company's core business systems, the stable operation of databases is crucial. With the expansion of business scale and the increase in system complexity, database operation and maintenance present the following challenges: (a) Disadvantages of rule-based diagnostic systems 1. Incomplete rule coverage: The rule base cannot cover all fault scenarios, and new fault types require manual addition of rules; 2. High maintenance costs: The rule base needs continuous updates and maintenance, and as the system complexity increases, the maintenance costs grow exponentially. 3. Inability to handle complex scenarios: For complex faults involving multiple coupled factors, rule-based systems struggle to make accurate judgments; 4. Lack of learning ability: Unable to automatically learn new failure modes from historical data; (II) Disadvantages of Machine Learning-Based Diagnostic Systems 1. Poor interpretability: The model output only classifies faults and cannot explain the basis for the judgment; 2. Lack of root cause analysis: The contribution of each indicator cannot be quantified, making it difficult for DBAs to understand the AI's judgment process; 3. Limited feature engineering: Primarily uses basic performance metrics and lacks in-depth feature mining; 4. Inability to adapt to business changes: The model is fixed after training and cannot adapt to fluctuations in business volume and system changes; (III) Disadvantages of Deep Learning-Based Diagnostic Systems 1. High demand for training data: A large amount of labeled data is required, which is difficult for small and medium-sized enterprises to meet; 2. Difficulty in explanation: Deep learning models are black boxes, making it difficult to provide understandable explanations; 3. Poor real-time performance: The model inference takes a long time, making it difficult to meet the needs of real-time diagnosis; 4. Poor generalization ability: Difficult to migrate between different database types; (iv) Disadvantages of Diagnostic Systems Based on Explainable AI 1. General methods are not optimized: General methods such as SHAP and LIME are not optimized for database operation and maintenance scenarios; 2. Unsuitable output format: The output format is not suitable for operation and maintenance scenarios, making it difficult for DBAs to use directly. 3. Lack of business context: Business context information such as business volume and version is not considered; 4. Lack of periodic analysis: The periodic characteristics of business data are not fully utilized; (v) Disadvantages of periodic detection methods based on time series analysis 1. Limited by single methods: Using only a single method such as FFT or autocorrelation cannot fully capture periodic features; 2. Lack of a comprehensive evaluation mechanism: It is impossible to provide a comprehensive score and confidence assessment of the cycle strength; 3. Unable to automatically detect unknown periods: A period range needs to be preset, and unknown periods cannot be detected; 4. Unable to map to business semantics: The detected mathematical cycles cannot be automatically mapped to business semantics (such as "daily cycle" or "weekly cycle"). (vi) Disadvantages of AI-based diagnostic systems 1. Lack of real-time capability: LLM inference speed is slow, making it difficult to meet the needs of real-time diagnosis; 2. Lack of precision: The diagnostic results output by LLM may be inaccurate and difficult to quantify; 3. Lack of interpretability: LLM is a black box model, making it difficult to provide interpretable diagnostic evidence; 4. Lack of business context: Business context information such as business volume and version is not considered; 5. High cost: LLM inference is costly and difficult to deploy on a large scale in production environments; (vii) Disadvantages of diagnostic systems based on experience knowledge bases 1. Incomplete case coverage: The historical case library cannot cover all failure scenarios, and there are no cases to refer to for new failure types; 2. Inaccurate similarity calculation: Simple similarity calculation cannot accurately match similar cases; 3. Lagging experience updates: The experience knowledge base is not updated in a timely manner and cannot reflect the latest failure modes and optimization suggestions; 4. Lack of intelligent reasoning: It relies solely on case matching and lacks intelligent reasoning capabilities; 5. Lack of interpretability: The case matching results lack interpretability, making it difficult for DBAs to understand the matching criteria; (viii) Common shortcomings of existing technologies 1. Insufficient modeling of the impact of business volume: The model fails to distinguish between "increased indicators driven by business" and "increased indicators caused by system problems," resulting in a high false alarm rate when business volume fluctuates. 2. The impact of version updates was not modeled: the baseline drift caused by version updates was not considered, and the false positive rate increased significantly after the version update; 3. Poor cross-database adaptability: It mainly targets general metrics and lacks in-depth analysis of metrics specific to specific databases (DB2, Oracle, MySQL, PostgreSQL); 4. Insufficient number of features: The existing system feature engineering only uses basic performance indicators and lacks multi-dimensional features such as business features, periodic features, and experience features, which cannot fully characterize the database status. Summary of the Invention

[0003] To address the aforementioned technical problems, embodiments of the present invention provide a business-aware AI-based diagnostic method for database operation and maintenance, comprising: The candidate database performance metrics data of the target database are collected from multiple data sources. The candidate database performance metrics data are related to the operating system, database characteristics, and business. Based on the current version information of the target database, the performance index data of the candidate database is corrected to obtain the database performance index data; Monitor the business data of the target database, and combine the business data and database performance index data to determine the impact coefficient and business volume impact value of each database performance index on the business data; A business data context is generated by combining the aforementioned business data, preset baseline business volume, impact coefficient, and business volume impact value; The business data is processed by combining clustering algorithms, preset business semantics, and business types to determine the periodic characteristics of the business data; Feature extraction is performed on the business data context, periodic features, and normalized database performance index data and business data. The normalization is related to business volume perception and target database version perception. The extracted features are used to perform diagnostic reasoning using a pre-built target model to obtain diagnostic results.

[0004] In one embodiment, the method further includes: Based on the diagnostic results and the extracted features, the SHAP value of each feature is determined; The importance and contribution of the features are ranked based on the SHAP values. A feature list is generated based on the ranking results and the SHAP values ​​of each of the aforementioned features.

[0005] In one embodiment, the method further includes: When the diagnostic results characterize the target database failure, the root cause analysis of the failure is performed by combining the SHAP value and the preset SHAP value grading standard to determine the root cause of the failure, and the evidence chain matching the reasoning path and reasoning process is recorded.

[0006] In one embodiment, the method further includes: Determine the multi-dimensional credibility assessment results of the diagnostic results, and determine the comprehensive confidence level based on the credibility assessment results. The multi-dimensional credibility assessment results include the target model confidence level, SHAP confidence level, historical confidence level determined based on historical similar data, and business confidence level corresponding to business uncertainty. The business uncertainty is determined based on business volume fluctuations, version updates, and system configuration changes. Based on the credibility assessment results and the preset credibility rating criteria, the credibility level and corresponding operational recommendations regarding the diagnostic results are determined.

[0007] In one embodiment, the method further includes: An enhanced diagnostic report is generated by processing the diagnostic results, feature list, evidence chain, business data context, and operation suggestions through a large language model. The evidence chain includes the entire reasoning process for determining the root cause of the fault by combining the SHAP value and a preset SHAP value grading standard to perform root cause analysis of the fault. The content of the enhanced diagnostic report and the DBA feedback data in the target database are combined and stored in a pre-built knowledge base to form a case study; An index and tags are constructed for the cases stored in the knowledge base.

[0008] In one embodiment, the step of correcting the candidate database performance indicator data based on the current version information of the target database to obtain database performance indicator data includes: Calculate the version impact factor based on the database performance metrics before and after the version update; The correction period for the current correction is calculated by combining the version impact factor and the preset basic correction period. The correction progress is calculated based on the correction period and the time interval between the current time and the update time of the current version; The revised database performance metrics are calculated based on the metric values ​​of the database performance metrics to be revised, the version impact factor, and the revision progress.

[0009] In one embodiment, monitoring the business data of the target database and determining the impact coefficient and business volume impact value of each database performance indicator on the business data in conjunction with the business data and database performance indicator data includes: The rate of change in business volume is determined based on the data volume of the aforementioned business data and the baseline business volume. The business volume change rate and the statistical database performance index change rate are learned based on the linear regression method, and the current influence coefficient is updated based on the learning results. The initial value of the influence coefficient is a preset value. The impact value of the business volume is determined based on the rate of change of the business volume and the impact coefficients of the updated corresponding different database performance indicators. The method further includes: Obtain information about the false alarm rate for alarms; The false alarm rate and the rate of change in traffic volume are learned using the gradient descent method, and the current adjustment coefficient is adjusted based on the learning results. The initial value of the adjustment coefficient is a preset value. The preset static threshold is updated based on the business volume change rate and the adjusted adjustment coefficient to obtain the dynamic threshold. The static threshold is the range of index values ​​for the database performance index. The process of generating a business data context by combining the business data, a preset baseline business volume, an impact coefficient, and a business volume impact value includes: The business data context is generated by combining the aforementioned business data, baseline business volume, impact coefficient, business volume impact value, business volume change rate, and dynamic threshold.

[0010] In one embodiment, the step of processing the business data by combining clustering algorithms, preset business semantics, and business types to determine the periodic characteristics of the business data includes: The business data is preprocessed; Spectral analysis and peak detection are performed on the preprocessed business data, and the spectral analysis results and peak detection results are learned based on unsupervised learning methods to obtain the first candidate period list; The first candidate period in the candidate period list is clustered based on a clustering algorithm to obtain the second candidate period list; The second candidate period in the second candidate period list is verified by the autocorrelation algorithm to obtain the period whose autocorrelation coefficient meets the preset requirements; The business semantics of the period are determined based on a preset semantic mapping rule containing weights, and the business semantics describe the temporal characteristics of the period. The periodic characteristics of the period are determined based on the periodic attribute features, business semantics, weights, and periodic confidence, and the periodic confidence is calculated based on the peak detection results and autocorrelation coefficients.

[0011] In one embodiment, the feature extraction of the business data context, periodic features, and normalized database performance index data and business data includes: The business data context, periodic features, and normalized database performance index data and business data are respectively extracted using basic features and specific features. The basic features are common features of different target databases, and the specific features are unique features of different target databases. A baseline model is constructed based on database performance metrics data from the first time period of the historical period, and a recent model is constructed based on database performance metrics data from the second time period of the historical period. The baseline model is used to determine the long-term baseline of database performance metrics, and the recent model is used to capture recent business change trends. The first time period is longer than the second time period. The rate of change in business volume is determined based on the data volume of the aforementioned business data and the baseline business volume. The weight parameters of the baseline model and the recent model are determined based on the rate of change in business volume. The target baseline is determined based on the baseline model, recent model, and weight parameters. Based on the aforementioned basic features, specific features, target baseline, and business volume impact value, determine the normalized features corresponding to the aforementioned basic features and specific features; The method further includes: In the event that the target baseline has drifted, the drifted target baseline shall be corrected based on the version impact factor and correction period. The normalized features are updated based on the corrected target baseline.

[0012] Another embodiment of the present invention also provides a business-aware database operation and maintenance AI diagnostic device, comprising: The acquisition module is used to collect candidate database performance index data of the target database from multiple data sources. The candidate database performance index data is related to the operating system, database characteristics, and business. The correction module is used to correct the candidate database performance index data based on the current version information of the target database to obtain database performance index data. The first determining module is used to monitor the business data of the target database and, in combination with the business data and database performance index data, determine the impact coefficient and business volume impact value of each database performance index on the business data. The generation module is used to generate a business data context by combining the business data, the preset baseline business volume, the impact coefficient, and the business volume impact value. The second determining module is used to process the business data by combining clustering algorithms, preset business semantics and business types, and determine the periodic characteristics of the business data; The extraction module is used to extract features from the business data context, periodic features, and normalized database performance index data and business data. The normalization is related to business volume perception and target database version perception. The reasoning module is used to perform diagnostic reasoning on the extracted features using a pre-built target model to obtain diagnostic results.

[0013] Other features and advantages of this application will be set forth in the following description. The objectives and other advantages of this application can be realized and obtained through the structures particularly pointed out in the written description and drawings.

[0014] The technical solution of this application will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0015] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0016] Figure 1 This is a flowchart illustrating the business-aware AI-based database operation and maintenance diagnostic method in this embodiment of the invention.

[0017] Figure 2 This is a flowchart of the model reasoning process in an embodiment of the present invention.

[0018] Figure 3 This is a flowchart illustrating the SHAP value calculation process in an embodiment of the present invention.

[0019] Figure 4 This is a flowchart of the credibility quantification process in an embodiment of the present invention.

[0020] Figure 5 This is a flowchart illustrating the generation process of the enhanced diagnostic report in an embodiment of the present invention.

[0021] Figure 6 This is a flowchart illustrating the generation and application of the experience knowledge base in an embodiment of the present invention.

[0022] Figure 7 This is a structural block diagram of the business-aware database operation and maintenance AI diagnostic device in an embodiment of the present invention. Detailed Implementation

[0023] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings, but these are not intended to limit the scope of the invention.

[0024] It should be understood that various modifications can be made to the embodiments disclosed herein. Therefore, the following description should not be considered as limiting, but merely as an example of embodiments. Other modifications within the scope of this disclosure will be apparent to those skilled in the art.

[0025] The accompanying drawings, which are included in and form part of this specification, illustrate embodiments of the present disclosure and, together with the general description of the disclosure given above and the detailed description of the embodiments given below, serve to explain the principles of the disclosure.

[0026] These and other features of the invention will become apparent from the following description of preferred forms of embodiments given as non-limiting examples, with reference to the accompanying drawings.

[0027] It should also be understood that although the invention has been described with reference to some specific examples, those skilled in the art can certainly implement many other equivalent forms of the invention, which have the features described in the claims and are therefore all within the scope of protection defined herein.

[0028] The above and other aspects, features and advantages of this disclosure will become more apparent when taken in conjunction with the accompanying drawings and in view of the following detailed description.

[0029] Specific embodiments of the present disclosure are described thereafter with reference to the accompanying drawings; however, it should be understood that the disclosed embodiments are merely examples of the present disclosure and can be implemented in various ways. Well-known and / or repeated functions and structures are not described in detail to avoid unnecessary or redundant details that could obscure the present disclosure. Therefore, the specific structural and functional details disclosed herein are not intended to be limiting, but merely to serve as the basis and representative basis for the claims to teach those skilled in the art to use the present disclosure in a variety of substantially any suitable detailed structures.

[0030] This specification may use the phrases “in one embodiment,” “in another embodiment,” “in yet another embodiment,” or “in still another embodiment,” all of which may refer to one or more of the same or different embodiments according to this disclosure.

[0031] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0032] To address the problems in existing solutions, such as numerous database performance metrics, difficulty in fault location, high system maintenance costs, and wide-ranging impact of business failures, such as... Figure 1 As shown, this embodiment of the invention provides a business-aware AI-based database operation and maintenance diagnostic method, including: S1: Collect candidate database performance index data of the target database from multiple data sources. The candidate database performance index data is related to the operating system, database characteristics, and business. S2: Based on the current version information of the target database, the performance index data of the candidate database is corrected to obtain the database performance index data; S3: Monitor the business data of the target database, and combine the business data and database performance index data to determine the impact coefficient and business volume impact value of each database performance index on the business data; S4: Generate business data context by combining the business data, preset baseline business volume, impact coefficient, and business volume impact value; S5: Process the business data by combining clustering algorithms, preset business semantics, and business types to determine the periodic characteristics of the business data; S6: Extract features from the business data context, periodic features, and normalized database performance index data and business data. The normalization is related to business volume perception and target database version perception. S7: Use the pre-built target model to perform diagnostic reasoning on the extracted features to obtain diagnostic results.

[0033] Based on the above, it can be seen that the method in this embodiment transforms AI diagnosis from a "black box" to a "white box," and from "frequent false alarms" to "accurate and reliable." Firstly, the method in this embodiment, through business volume-aware normalization and version-aware correction, enables AI to distinguish between "indicator increases caused by normal business fluctuations" and "genuine system failures," completely resolving the previous problem of soaring false alarm rates after business peaks or version upgrades, making diagnostic results more accurate and reliable. In short, the method in this embodiment transforms database operation and maintenance from "passive" to "proactive and controllable," and from "relying on personal experience" to "AI-interpretable assisted decision-making," achieving effective maintenance of the database through AI models that accurately match actual business conditions.

[0034] In one embodiment, using DB2 database as an example, the data acquisition phase refers to collecting database performance indicator data from multiple data sources to provide a data foundation for subsequent feature extraction and model inference. This acquisition includes: Basic metric collection, such as collecting fundamental performance metrics of the operating system and database, provides foundational data for subsequent analysis. The collected data includes, for example, CPU utilization, disk I / O wait times, network latency, number of active transactions, number of slow queries, etc.

[0035] Database-specific metrics collection, such as performance metrics specific to DB2 databases, provides data support for dedicated DB2 database diagnostics. Collected metrics include, for example, DB2 CPU utilization, DB2 sort heap utilization, DB2 packet cache hit rate, and DB2 lock wait time.

[0036] Business metric collection, such as collecting business-related performance metrics, provides data support for adapting to business scenarios. The collected content includes, for example, business transaction volume metrics (including policy inquiry volume, claims inquiry volume, insurance purchase volume, claims upload volume, etc.), business cycle metrics (including whether it is a claims season, monthly business level, hourly business level, whether it is a policy change period, etc.), and business trend metrics (including business transaction volume trends, etc.).

[0037] Furthermore, the step of correcting the candidate database performance indicator data based on the current version information of the target database to obtain database performance indicator data includes: S201: Calculate the version impact factor based on the database performance metrics before and after the version update; S202: Calculate the correction period to adapt to the current correction by combining the version impact factor and the preset basic correction period; S203: Calculate the correction progress based on the correction period and the time interval between the current time and the update time of the current version; S204: Calculate the corrected database performance index value based on the index value of the database performance index to be corrected, the version impact factor, and the correction progress.

[0038] For example, identifying the impact of database version updates on performance metrics, correcting metric data, and reducing false alarm rates. The version-aware correction process in this embodiment includes: 1. Get version information: Retrieve database version information, including the current version, previous version, and update time; 2. Calculate the version impact factor: Version impact factor = (Indicator (value) after version update - Indicator (value) before version update) / Indicator (value) before version update; 3. Calculate the correction progress: Correction progress = Time from the current version update to the current time / Correction period (between 0 and 1); 4. Performance metric correction: Corrected metric = Original metric / (1 + Version impact factor × Correction progress); 5. Baseline Adjustment: Adjusted Baseline = Original Baseline × (1 - Correction Progress × Version Impact Factor) 6. Output the corrected indicator: The output includes data structures such as timestamp, indicator name, original indicator value, and corrected indicator value.

[0039] The version impact factor is calculated by subtracting the original version's metric from the updated metric and then dividing by the original metric. The version impact factors for various database performance metrics are shown in the table below:

[0040] The revision period mechanism, that is, the mechanism related to the revision period of indicator data: Revision period length: Revision period = Base revision period x Version impact factor 2 The basic fix period is usually 7 days. Major version upgrades have longer fix periods, such as 14-21 days, minor version upgrades have a fix period of 7-14 days, and patch updates have shorter fix periods, such as 1-7 days.

[0041] Baseline adjustment during the correction period: Adjusted baseline = Original baseline x (1 - (Correction progress x Version impact factor). Where correction progress = Time since version update ÷ Correction period.

[0042] For the baseline, including both the pre-adjustment and post-adjustment baselines, baseline drift detection is also involved: Drift detection value = Current database performance metric value - |Expected metric value| / Expected metric value. If the drift detection value is greater than a preset drift threshold, it is determined to be baseline drift, triggering a baseline update.

[0043] Baseline update strategy: Updated baseline = (1 - Update coefficient) x Old baseline + Update coefficient x Recent baseline average. The update coefficient is larger during the correction period and smaller after the correction period.

[0044] The correction formula for version-aware performance metrics is: Corrected metric = Original metric ÷ (1 + (Version impact factor x Correction progress)). Next, the corrected metric is normalized: Normalized value = (Corrected metric - Baseline) / Standard deviation of the corrected metric.

[0045] After the above adjustments, the database characteristic indicators are obtained and output based on the following data structure: timestamp, indicator name, original indicator value, version information (current version, previous version, update time), version impact factor, correction period, correction progress, corrected indicator value, adjusted baseline, normalized value, whether it is within the correction period, and explanation.

[0046] Next, real-time monitoring of changes in business volume is performed to establish a model of the relationship between business volume and performance metrics, providing business context for subsequent feature extraction and model inference. In this embodiment, monitoring the business data of the target database and determining the impact coefficient and business volume impact value of each database performance metric on the business data in conjunction with the business data and database performance metric data includes: S301: Determine the rate of change of business volume based on the data volume of the business data and the baseline business volume; S302: Based on the linear regression method, learn the rate of change of the business volume and the rate of change of the statistical database performance index values, and update the current influence coefficient in combination with the learning results. The initial value of the influence coefficient is a preset value. S303: Determine the impact value of the business volume based on the business volume change rate and the impact coefficients of the updated corresponding different database performance indicators.

[0047] For example, monitoring business data may include, but is not limited to: Policy inquiry volume: The number of policy inquiries per unit of time; Claims inquiry volume: The number of claims inquiries per unit of time; Insurance coverage: the number of times insurance is purchased per unit of time; Claim upload volume: The number of claims uploaded per unit of time; Business transaction volume: The total number of business transactions per unit of time.

[0048] Next, scene identification can be performed based on monitored business data. For example, unsupervised learning (K-Means clustering) can be used to automatically discover business scenes, avoiding scene distortion caused by hard coding. Specifically, feature engineering should be applied to the business data to obtain at least the following features: Time characteristics: hour, day of the week, month, whether it is the beginning of the month, whether it is the end of the month, whether it is a holiday; Business volume characteristics: policy inquiry volume, claims inquiry volume, number of new policies, number of claims uploaded, total business transaction volume; Business volume change characteristics: business volume change rate, business volume trend (rising / falling / stable), business volume fluctuation variance; Historical comparison features: comparison with the same period yesterday, comparison with the same period last week, and comparison with the same period last month.

[0049] The obtained features are standardized, for example, using Z-Score standardization, to eliminate the influence of dimensions: Z = (X - μ) / σ Where X is the original feature value, μ is the historical mean (past 30 days), and σ is the historical standard deviation (past 30 days).

[0050] The standardized features are then clustered. K-Means clustering (K=8, corresponding to 8 typical business scenarios) can be selected as the algorithm. Euclidean distance can be used as the distance metric, and the K-Means++ algorithm can be used for initializing cluster centers. The clustering process includes using the K-Means algorithm to assign the current business state to the nearest cluster center based on the standardized features. Then, based on the characteristics of the cluster centers, the business scenario is automatically named (e.g., "Morning Peak," "Monthly Peak," etc.), and the current business scenario type, confidence level, and similar historical scenarios are output. Scenario naming rules can include: naming based on time features: if the hourly features of the cluster centers are concentrated between 9-11 AM, it is named "Morning Peak"; naming based on business volume features: if the business volume features of the cluster centers are more than 1.5 times the average, it is named "Peak"; naming based on month features: if the month feature of the cluster centers is December, it is named "Annual Peak"; naming based on claims features: if the claims volume features of the cluster centers are more than 1.5 times the average, it is named "Claims Peak."

[0051] Furthermore, the identified scenarios are dynamically changing and have an online learning mechanism. This includes updating cluster centers daily with new data; merging two cluster centers into one scenario if the distance is less than a threshold; splitting a cluster into two scenarios if the variance within a cluster exceeds a threshold; and marking a scenario as expired and deleting it if it hasn't been triggered for an extended period. During scenario identification, the automatically discovered scenario names (e.g., "morning rush hour," "monthly peak") are used to determine the confidence level of scenario identification based on the distance from the current business status to the cluster center; the smaller the distance, the higher the confidence level. Simultaneously, based on historical data, five or more similar scenarios and their timestamps are identified, and relevant data for the identified scenario is output accordingly.

[0052] After determining the scenario, the system executes another process to determine the impact value of business volume. First, it determines the relationship between database performance metrics and baseline values, business volume impact values, and system problem impact values. In this embodiment, the performance metric equals the baseline value plus the business volume impact value plus the system problem impact value. The baseline value represents the normal range of the performance metric, the business volume impact value is the performance metric change caused by changes in business volume, and the system problem impact value is the performance metric change caused by system problems. Business volume impact value = Business volume change rate x Impact coefficient. The business volume change rate = (Current business volume - Baseline business volume) ÷ Baseline business volume, and the impact coefficient is the sensitivity of different performance metrics to business volume. The impact coefficient is determined through online learning. The learning objective is to dynamically adjust the impact coefficient based on historical data to more accurately reflect the actual impact of business volume on performance metrics. The learning method can be, but is not limited to, linear regression, calculating the impact coefficient based on historical data. The learning formula is: Impact coefficient = Σ[(business volume change rate × performance index change rate)] / Σ[(business volume change rate)²].

[0053] The specific learning steps include collecting historical business volume data and performance indicator data, calculating the business volume change rate and performance indicator change rate, calculating the impact coefficient using linear regression, smoothly updating the calculated impact coefficient with the current impact coefficient, and verifying the accuracy of the impact coefficient using a validation set. The update formula includes: New impact coefficient = Learning rate × Calculated impact coefficient + (1 - Learning rate) × Current impact coefficient, where the learning rate is 0.01, but not unique. Afterwards, a validation set can be used for verification, such as using business data from the most recent 7 days as the validation set. The mean squared error (MSE) is used to evaluate the accuracy of the impact coefficient. If the MSE is greater than a threshold, the impact coefficient is recalculated.

[0054] Furthermore, the method also includes: S304: Obtain information about the false alarm rate; S305: The gradient descent method is used to learn the false alarm rate and the rate of change in traffic volume, and the current adjustment coefficient is adjusted based on the learning results. The initial value of the adjustment coefficient is a preset value. S306: Update the preset static threshold based on the business volume change rate and the adjusted adjustment coefficient to obtain the dynamic threshold, wherein the static threshold is the index value range of the database performance index. The process of generating a business data context by combining the business data, a preset baseline business volume, an impact coefficient, and a business volume impact value includes: S401: Generate business data context by combining the business data, baseline business volume, impact coefficient, business volume impact value, business volume change rate, and dynamic threshold.

[0055] For example, in this embodiment, the dynamic threshold = static threshold x (1 + traffic volume change rate x adjustment coefficient). Here, the static threshold is the normal range threshold for the performance indicator, the traffic volume change rate = (current traffic volume - baseline traffic volume) / baseline traffic volume, and the adjustment coefficient represents the sensitivity of different performance indicators to threshold adjustments. The adjustment coefficient is specifically determined using an online learning method. For example, the learning objective is to dynamically adjust the adjustment coefficient based on the false positive rate / false negative rate to more accurately reflect the impact of traffic volume on the threshold. The learning method includes using gradient descent to dynamically adjust the adjustment coefficient based on the false positive rate / false negative rate. The learning formulas involved include: Change in adjustment coefficient = Learning rate × (Target false positive rate - Actual false positive rate) × Change in traffic volume; New adjustment coefficient = Current adjustment coefficient + Change in adjustment coefficient

[0056] The learning process includes collecting historical alarm data, business volume data, and performance indicator data. The actual false alarm rate (number of false alarms / total number of alarms) is calculated. Based on the difference between the false alarm rate and the target false alarm rate, the change in the adjustment coefficient is calculated. This change is added to the current adjustment coefficient to complete the coefficient update. Afterwards, business data from the most recent 7 days can be used as a validation set to evaluate the accuracy of the adjustment coefficient using the false alarm rate and the missed alarm rate. If the false alarm rate is greater than 10% or the missed alarm rate is greater than 5%, the adjustment coefficient is recalculated.

[0057] After completing the above calculations, a business context can be generated based on the data obtained from the calculations. Its data structure includes: timestamp, business volume information (policy inquiry volume, claims inquiry volume, insurance volume, claims upload volume, total business transaction volume), business scenario type, business volume change rate, dynamic threshold (dynamic threshold of each performance indicator), and business impact coefficient (impact coefficient of each performance indicator).

[0058] In one embodiment, the system uses machine learning to automatically discover the periodic characteristics of business operations, avoiding periodic distortion caused by hard coding and providing periodic information for business scenario adaptation. The process of combining clustering algorithms, preset business semantics, and business types to process the business data and determine its periodic characteristics includes: S501: Preprocess the business data; S502: Perform spectral analysis and peak detection on the preprocessed business data, and learn the spectral analysis results and peak detection results based on unsupervised learning methods to obtain the first candidate period list; S503: Cluster the first candidate period in the candidate period list based on the clustering algorithm to obtain the second candidate period list; S504: The second candidate period in the second candidate period list is verified by the autocorrelation algorithm to obtain the period whose autocorrelation coefficient meets the preset requirements; S505: Determine the business semantics of the period based on a preset semantic mapping rule containing weights, wherein the business semantics describes the temporal characteristics of the period; S506: Determine the periodic characteristics of the period based on the periodic attribute features, business semantics, weights, and periodic confidence, wherein the periodic confidence is calculated based on peak detection results and autocorrelation coefficients.

[0059] For example, Z-Score can be used to standardize operational data to eliminate the influence of dimensions. Then, linear interpolation is used to fill missing values, followed by IQR detection and handling of outliers. Spectral analysis and power spectral density calculation are performed on the preprocessed operational data, including using Fast Fourier Transform (FFT) to transform the time series from the time domain to the frequency domain, automatically identifying periods, and automatically detecting significant peaks in the power spectrum. The minimum peak height is 10% of the maximum power spectrum value, and the minimum peak spacing is 5% of the sampling frequency. After obtaining the periods, DBSCAN clustering is used to merge similar periods to avoid duplicate detections. Clustering parameters include neighborhood radius (0.1 × minimum detection period) and minimum sample size, which can be 1.

[0060] Next, periodic verification is performed. The autocorrelation coefficient of the detection period is calculated and must be greater than 0.5. The detection period is then mapped to a preset business semantic. The confidence score is calculated using the following formula: Confidence level = Normalized peak intensity × 0.6 + Autocorrelation coefficient × 0.4.

[0061] The semantic mapping rules can be found in the table below:

[0062] The business weight equals the base weight multiplied by the business type coefficient. Different business types have different coefficients; for example, the monthly coefficient for auto insurance is 1.5, the quarterly coefficient is 1.8, and the annual coefficient is 2.0.

[0063] After determining the above parameters, the output periodic features include timestamp, data source, sampling interval, list of detected periods (each period includes period length, unit, business semantics, intensity, weight, and confidence), and recommendation suggestions.

[0064] Furthermore, the feature extraction from the business data context, periodic features, and normalized database performance metrics and business data includes: S601: Extract basic features and specific features from the business data context, periodic features, and normalized database performance index data and business data, respectively. The basic features are common features of different target databases, and the specific features are unique features of different target databases. S602: Construct a baseline model based on database performance index data for the first time period of the historical period, and construct a recent model based on database performance index data for the second time period of the historical period. The baseline model is used to determine the long-term baseline of database performance indexes, and the recent model is used to capture recent business change trends. The first time period is longer than the second time period. S603: Determine the rate of change of business volume based on the data volume of the business data and the baseline business volume; S604: Determine the weight parameters of the baseline model and the recent model based on the business volume change rate; S605: Determine the target baseline based on the baseline model, recent model, and weight parameters; S606: Determine the normalized features corresponding to the basic features and specific features based on the basic features, specific features, target baseline, and traffic volume impact value; Furthermore, the method further includes: S8: In the event that the target baseline has drifted, the drifted target baseline shall be corrected based on the version impact factor and correction period; S9: Update the normalized features based on the corrected target baseline.

[0065] The purpose of the above feature extraction process is to extract feature-engineered features from database performance index data, providing standardized and structured input for subsequent model inference. In this embodiment, a two-layer feature architecture of "universality + specificity" is used for feature extraction, ensuring both generality and deep adaptation. The universality layer is used to extract general features, applicable to all database types and operational scenarios; the specificity layer is used to extract specific features, deeply optimized for database types such as DB2, Oracle, MySQL, and PostgreSQL. The extracted features may include, but are not limited to, core diagnostic features and auxiliary explanatory features. Core diagnostic features include: Basic characteristics: CPU utilization, disk I / O wait, network latency, number of active transactions, number of lock waits, number of slow queries, and other basic performance metrics; Time characteristics: hour, day of the week, whether it is the beginning of the month, whether it is a holiday, time since the last anomaly, etc. Historical features: historical statistical features such as CPU trends, average CPU usage, I / O trends, and changes in the number of transactions; Correlation characteristics: correlation characteristics such as CPU and disk I / O correlation, CPU and transaction correlation, disk and transaction correlation; Anomaly characteristics: Anomaly detection characteristics such as whether the CPU is abnormal, whether disk I / O is abnormal, whether transactions are abnormal, and whether the network is abnormal; Periodic characteristics: current periodic type, periodic strength, periodic confidence, periodic phase, multi-period encoding, periodic weight, and other periodic features.

[0066] Auxiliary explanatory features include: Business characteristics include: policy inquiry volume, claims inquiry volume, number of new policies, number of claims uploaded, whether it is the claims season, monthly business level, hourly business level, whether it is a policy change period, business transaction volume trends, and other business dimension characteristics. Business scenario characteristics: business scenario type, business scenario confidence level, number of similar historical scenarios, business scenario coding, business scenario time characteristics, and other business scenario characteristics; Business impact characteristics: Business scenario type, business impact score, priority of recommended measures, and other business impact characteristics; Empirical features: number of similar cases, highest similarity, recommended fault type, recommended root cause, empirical confidence, etc. Other auxiliary features: diagnostic confidence, fault classification, severity, etc.

[0067] The extracted features are normalized to distinguish between "increased indicators driven by business" and "increased indicators caused by system problems," thus avoiding false alarms caused by fluctuations in business volume.

[0068] Next, models are built, including a baseline model and a recent model. The baseline model uses database performance metric data from the past 3-6 months to establish a long-term baseline for performance metrics. The recent model uses performance metric data from the most recent 7-14 days to capture recent business trends.

[0069] In this embodiment, the normalization formula for traffic volume perception is: Normalized value = (Original value - Baseline value - Business volume impact value) / Dynamic standard deviation.

[0070] Wherein, the baseline model weight = 1 - the rate of change in business volume²; the recent model weight = the rate of change in business volume². When business volume is stable, the baseline model weight is close to 1, and the baseline model is mainly used; when business volume fluctuates greatly, the recent model weight increases, and the recent model is used more often.

[0071] Since database performance metrics are also affected by the database version, it is necessary to adjust the metrics in conjunction with the database version, that is, to perform version-aware correction on the extracted features.

[0072] Version impact factor = (Indicators after version update - Indicators before version update) / Indicators before version update; Revision period = Base revision period × Version impact factor²; Based on the above correction mechanism, the extracted features are corrected as follows: Corrected feature = Original feature / (1 + Version impact factor × Correction progress); Fix progress = Time until version update / Fix period.

[0073] After completing the feature data correction, the final features are output, including timestamps, core diagnostic features, auxiliary explanatory features, normalized features, corrected features, and feature source descriptions.

[0074] Furthermore, as mentioned earlier, the obtained baseline needs to be subjected to drift detection. If drift is detected, baseline correction is required. Details will not be elaborated further.

[0075] After processing the feature data, it needs to be input into a trained AI model to infer the feature-engineered features and obtain fault diagnosis results for the target database. The model can be, but is not limited to, XGBoost (eXtreme Gradient Boosting). The inference process can be combined with... Figure 2 As shown. The model's output includes: Fault types: CPU problems, disk I / O problems, network problems, lock wait problems, slow query problems, transaction problems; Severity levels: Informational, Warning, Emergency; Confidence: The confidence level of the model in the diagnostic results (a floating-point number between 0 and 1).

[0076] After completing model diagnosis, the system will perform an interpretability calculation step. This step calculates the SHAP value of the feature-engineered features to obtain their contribution and explain the decision-making basis of the AI ​​model. The SHAP value calculation process can be found in [link to documentation]. Figure 3 As shown. For example, correspondingly, the method further includes: S9: Based on the diagnostic results and extracted features, determine the SHAP value of each feature; S10: Rank the features by importance and contribution based on the SHAP values; S11: Generate a feature list based on the ranking results and the SHAP values ​​of each of the aforementioned features.

[0077] During operation, SHAP values ​​can be mapped to business terms using a pre-built business semantic mapping rule table, making it easier for DBAs in the database to understand. Through the calculation process described above, the system outputs a feature importance ranking: a list of features sorted from highest to lowest contribution; a detailed analysis of SHAP values: the SHAP value, contribution, and impact description for each feature; and a list of high-contribution features: a list of features with a contribution greater than 0.05.

[0078] Through SHAP value analysis, root cause classification, reasoning path recording, and evidence chain construction, the system fully demonstrates to DBAs in the database every step of the decision-making process "from input features to output conclusions". DBAs are no longer faced with an incomprehensible fault label, but can see the contribution of each indicator to the fault and the complete reasoning process, thus fully trusting the AI's judgment.

[0079] In one embodiment, the method further includes: S12: When the diagnostic results characterize the target database fault, the root cause analysis of the fault is performed in combination with the SHAP value and the preset SHAP value grading standard to determine the root cause of the fault, and the evidence chain matching the reasoning path and reasoning process is recorded.

[0080] The purpose of the above process steps is to determine the root cause of the failure based on the feature contribution and to record the reasoning path. The root cause classification criteria in this embodiment include: The main root cause is that the SHAP value contributes more than 0.3, which has a direct and major impact on the fault. Secondary root cause: The contribution of SHAP value is between 0.15 and 0.3, which has a certain impact on the fault. Minor root cause: SHAP value contribution is less than 0.05, with minimal impact on the fault.

[0081] The inference path recording format (JSON structured data) may include, but is not limited to, the following: diagnosis timestamp, database name, model version, confidence level, root cause analysis (primary root cause, secondary root cause, minor root cause), and inference path (list of inference steps, detailed information for each step, and timeline).

[0082] The evidence chain is an ordered sequence of feature values, SHAP values, impact descriptions, and timestamps. This evidence chain is used to record the complete reasoning process from input to output, supporting path backtracking and verification.

[0083] In one embodiment, the method further includes: S13: Determine the multi-dimensional credibility assessment results of the diagnostic results, and determine the comprehensive confidence level based on the credibility assessment results. The multi-dimensional credibility assessment results include the target model confidence level, SHAP confidence level, historical confidence level determined based on historical similar data, and business confidence level corresponding to business uncertainty. The business uncertainty is determined based on business volume fluctuations, version updates, and system configuration changes. S14: Based on the credibility assessment results and the preset credibility rating criteria, determine the credibility level and the corresponding operational recommendations regarding the diagnostic results.

[0084] For example, the purpose of the above method is to conduct a multi-dimensional credibility assessment of AI diagnostic results, provide comprehensive confidence level and graded rating, and help DBAs in the target database judge the credibility of the diagnostic results and decide whether to adopt the recommendations.

[0085] In this embodiment, the trusted metric architecture, such as... Figure 4 As shown, it includes: 1. Diagnostic Results: Input the diagnostic results from the model; 2. Multi-dimensional confidence assessment: Model confidence: The confidence level of the model in the prediction results (0-1); SHAP confidence: The percentage of contribution of the main features to the prediction, equal to the sum of the SHAP values ​​of the first k features divided by the sum of the SHAP values ​​of all features, ranging from 0 to 1; Historical confidence level: Number of correct cases among similar cases / Total number of similar cases; Business confidence level: 1 - Business uncertainty; The business uncertainty is equal to the square root of the sum of the squares of the variance of business volume fluctuation, the impact of version updates, and the impact of configuration changes; 3. Weighted fusion: Overall confidence level = Model confidence level × 30% + SHAP confidence level × 25% + Historical confidence level × 25% + Business confidence level × 20%; 4. Overall Confidence: Outputs an overall confidence value between 0 and 1; the overall confidence value is equal to the sum of the confidence values ​​of each dimension multiplied by their corresponding weights. The recommended weights are: model confidence 30%, SHAP confidence 25%, historical confidence 25%, and business confidence 20%.

[0086] 5. Grading and rating: Based on the overall confidence level (A: 0.9-1.0, B: 0.7-0.9, C: 0.5-0.7, D: 0.3-0.5, E: 0-0.3); different confidence levels have different recommended actions.

[0087] 6. Visualization: The dashboard displays the overall confidence level, and the radar chart displays the confidence level for each dimension. For example, the dashboard displays: Overall Confidence Level: Circular dashboard (0-100%); Confidence Levels for Each Dimension: Bar chart; Rating: Letter rating (AE); Radar Chart Display: X-axis: Model confidence level, SHAP confidence level, historical confidence level, business confidence level; Y-axis: Confidence value (0-1). Graph: Quadrilateral, the larger the area, the higher the confidence level.

[0088] The multi-dimensional confidence assessment criteria include: Dimension 1: Model confidence: The confidence level of the model in the prediction results, ranging from 0 to 1.

[0089] Dimension 2: SHAP Confidence: The proportion of the contribution of the main features to the prediction, which is equal to the sum of the SHAP values ​​of the first k features divided by the sum of the SHAP values ​​of all features, and ranges from 0 to 1.

[0090] Dimension 3: Historical Confidence: The historical accuracy of similar cases, which is equal to the number of correct cases divided by the total number of similar cases, ranging from 0 to 1.

[0091] Dimension 4: Business Confidence: The impact of the business environment on the diagnostic results, equal to 1 minus business uncertainty (such as business volume fluctuations, version updates, etc.), ranging from 0 to 1.

[0092] After completing the above process, a data structure containing timestamps, diagnostic IDs, confidence levels for each dimension, overall confidence level, levels, recommendations, and visualization data (dashboard data, radar chart data) can be output.

[0093] In one embodiment, the method further includes: S15: Generate an enhanced diagnostic report by processing the diagnostic results, feature list, evidence chain, business data context, and operation suggestions through a large language model. The evidence chain includes the entire reasoning process for determining the root cause of the fault by combining the SHAP value and the preset SHAP value grading standard to perform root cause analysis of the fault. S16: Combine the content of the enhanced diagnostic report and the DBA feedback data in the target database and store it in the pre-built knowledge base to form a case. S17: Build an index and tags for the cases stored in the knowledge base.

[0094] This involves generating an AI diagnostic report that includes the root cause of the fault, feature contribution, reasoning path, and recommended measures. Then, an AI Large Model (LLM) is used to enhance the readability, understandability, and interpretability of the diagnostic report, providing reports tailored to different roles. Next, an experience knowledge base is established, storing historical fault diagnosis results, DBA feedback, and optimization suggestions, allowing for the retrieval of similar cases to assist in fault diagnosis.

[0095] Specifically, the generated report can be in Markdown and JSON formats. The report content structure can include: Fault Overview: Fault time, fault database, fault type, severity, and confidence level; Root cause analysis: primary root cause, secondary root cause, minor root cause; Feature contribution: Ranking of feature importance, detailed analysis of SHAP values, and list of high-contribution features; Reasoning path: a list of reasoning steps, detailed information for each step, and a timeline; Recommended measures: immediate action measures, short-term optimization measures, and long-term planning measures.

[0096] AI large model enhancement process, such as Figure 5 As shown, it includes: Diagnostic results: Input fault overview, root cause analysis, feature contribution, inference path, business context (including business volume, business scenario, dynamic threshold, etc.); LLM Diagnostic Report Generation: Generate natural language diagnostic reports using LLM; describe fault conditions, explain root causes and impacts, provide actionable recommendations in natural language, and keep report length within a specified word limit. Output Example: Includes fault time, fault database, fault type, severity, fault description, root cause analysis, impact range, and recommended actions.

[0097] LLM Root Cause Inference: Analyzes causal relationships between features and infers the propagation path of root causes; for example, input data includes: feature importance ranking, detailed SHAP value analysis, and inference path; LLM prompt templates: analyze causal relationships between features, infer the propagation path of root causes, and provide interpretability explanations for root causes. Output example: includes root cause propagation path, root cause analysis, and interpretability explanations.

[0098] LLM Interpretability Statement: Explains the contribution of each feature to the failure and the relationships between features; for example, input data: feature importance ranking, detailed SHAP value analysis, root cause analysis; LLM prompt template: explains the contribution of each feature to the failure, explains the relationships between features, and provides understandable explanations. Output example: includes feature contribution analysis, feature relationships, and understandable explanations.

[0099] LLM suggestion generation: Generates immediate action measures, near-term optimization measures, long-term planning measures, and DB2-specific optimization measures; for example, input data: root cause analysis, business context, fault type; LLM suggestion word templates: provide immediate action measures (urgent level faults), provide near-term optimization measures (warning level faults), provide long-term planning measures (information level faults), and provide DB2-specific optimization measures. Output example: Includes immediate action measures, near-term optimization measures, long-term planning measures, and DB2-specific optimization measures.

[0100] Output Enhancement Report: The output includes data structures such as natural language diagnostic reports, root cause inferences, interpretability descriptions, and recommended actions. For example, the enhanced report data structure can include: timestamps, natural language diagnostic reports, root cause inferences, interpretability descriptions, recommended actions, target roles, and report versions for different roles.

[0101] The formation process of an experience knowledge base, such as Figure 6 As shown, it includes: Experience accumulation: Historical fault diagnosis results, DBA feedback, and optimization suggestions are accumulated in the knowledge base; Case storage: Establish a case library, index library, and tag library to support multi-dimensional retrieval; Case similarity calculation: Case similarity = Fault type × 25% + Database type × 20% + Business type × 15% + Root cause × 25% + Feature × 15%; Experience citation: Recommending diagnostic results and suggested measures for similar cases based on similarity; Output similar cases: Output the top 5 similar cases (the specific values ​​are not fixed) in descending order of similarity.

[0102] The experience accumulation phase includes the following: Content accumulated: Fault Case: Fault Time, Fault Database, Fault Type, Severity, Diagnostic Results; Diagnostic results: root cause analysis, feature contribution, and reasoning path; DBA Feedback: Feedback and suggestions for improvement from the DBA regarding the diagnostic results; Optimization suggestions: immediate action measures, short-term optimization measures, and long-term planning measures; Implementation results: Implementation results and effectiveness evaluation of the recommended measures.

[0103] Sedimentation time: After each fault diagnosis is completed; After the DBA reports the diagnostic results; After the recommended measures are implemented; The data structure includes: case ID, timestamp, database information, fault information, diagnostic results, DBA feedback, suggested measures, and execution results.

[0104] The case storage phase involves the following: Storage structure: Case library: Stores all historical failure cases; Index: Create a case index to support fast retrieval; Tag library: Tag cases and support multi-dimensional search; Index fields: Fault type (e.g., CPU problem, disk I / O problem, network problem, etc.); Database type (e.g., DB2, Oracle, MySQL, PostgreSQL); Business type (e.g., auto insurance, life insurance, property insurance). Severity level (e.g., emergency, warning, information); Time frame (e.g., by month, by quarter, by year); Tagging system: includes fault type tags, database type tags, business type tags, severity tags, and root cause tags.

[0105] Case similarity calculation: Case similarity is equal to the sum of the similarity scores of each dimension multiplied by their corresponding weights. The weights are as follows: fault type (25%), database type (20%), business type (15%), root cause (25%), and feature (15%).

[0106] In this embodiment, the similarity calculation method specifically includes: Fault type similarity: 1 if the current fault type is equal to the historical fault type, 0 otherwise; Database type similarity: 1 if the current database type is equal to the historical database type, otherwise 0.0; Business type similarity: 1 if the current business type is equal to the historical business type, otherwise 0; Root cause similarity: the number of common root causes divided by the total number of root causes; Feature similarity: Calculated using cosine similarity; Feature vector composition: The feature vector equals [CPU utilization, disk I / O wait, network latency, number of active transactions, number of lock waits, number of slow queries].

[0107] For example, the similarity calculation example is the calculation of the similarity between the current fault and historical cases, and the final case similarity is 0.82.

[0108] The experience knowledge base in this embodiment also supports experience referencing, and its referencing strategy includes: High similarity cases (similarity greater than 0.8): Directly recommend diagnostic results and suggested measures; Medium similarity cases (between 0.6 and 0.8): Recommended diagnosis and suggested measures; Low similarity cases (similarity less than or equal to 0.6): For reference only, direct application is not recommended; The output format for cited cases includes a list of similar cases (each case includes case ID, similarity score, timestamp, failure type, root cause, recommended actions, and execution results) and recommendations.

[0109] Specifically, when outputting similar cases, the following should be included: Sort by similarity in descending order, and output at most the Top 5 (the specific number is not unique) similar cases. The output information includes basic case information, similarity, diagnosis results, suggested measures, and implementation results.

[0110] Based on the above, the diagnostic methods for accumulating and referencing experience knowledge bases include accumulating historical fault diagnosis results, DBA feedback, and optimization suggestions into the knowledge base - case storage, including: establishing a case library, index library, and tag library - case similarity calculation: multi-dimensional similarity calculation - experience referencing: recommending diagnostic results and suggested measures for similar cases based on similarity.

[0111] Based on the above solutions, this embodiment, through business cycle identification, business scenario adaptation, and LLM-enhanced report generation, can automatically identify business cycles (daily / weekly / monthly / annual) and generate targeted diagnostic reports and optimization suggestions based on the current business scenario (such as peak claims periods or monthly settlements). The reports adopt different language styles for different roles (DBA, operations and maintenance, and business personnel), which greatly improves the operability and adoption rate of the suggestions.

[0112] Meanwhile, through the accumulation and citation of experience knowledge base and the credibility quantification rating (AE level 5), the system can reuse historical diagnostic experience and tell the DBA "how reliable the current diagnostic results are", enabling the DBA to quickly determine whether to adopt them directly or need manual review. Ultimately, this has achieved an overall effect of reducing the average fault location time by more than 50%, the suggestion adoption rate by more than 95%, and the fault recovery time by 30-50%.

[0113] In another embodiment, the target database also includes a business scenario adaptation function, specifically, identifying business periodic characteristics, assessing business impact, and generating recommended measures for specific business scenarios.

[0114] Business cycle characteristic identification includes monthly business level identification, where different months have different business levels and corresponding weights. Monthly business level weight = Average business volume for that month / Average business volume for the preceding 6 months. Next, hourly business level identification is performed, with different time periods corresponding to different business levels and weights. Hourly business level weight = Average business volume for that hour / Average business volume for the preceding 9 hours. Finally, peak business period identification is conducted, with different types of peak periods corresponding to different weights, which can be preset weight values.

[0115] Next, based on the identified information, a business impact assessment is performed. This algorithm includes: Business Impact Score = Severity × Fault Type Weight × Peak Business Hour Weight × Monthly Business Level Weight × Hourly Business Level Weight × Policy Change Weight × Database Type Weight

[0116] The recommended measures will vary depending on the different assessment results, for example: Immediate action (emergency level fault): Check DB2 diagnostic logs, review slow query logs, check lock wait status, and suspend non-critical business operations; Recent optimization measures (warning-level faults): Adjusting DB2 sort heap size, optimizing DB2 package cache, running DB2 statistics updates, optimizing index structure, and cleaning up fragmented data; Long-term planning measures (information-level failures): Establish a capacity planning mechanism, optimize business processes, improve monitoring coverage, and conduct regular performance tests; DB2-specific optimization measures: Adjust SORTHEAP, optimize PKGCache, adjust BUFFERPOOL, run RUNSTATS, and run REORG INDEXES.

[0117] Furthermore, the target database in this embodiment also supports displaying the inference path in the form of a visual chart, supporting interactive viewing and path backtracking. The types of visual charts can be, but are not limited to: Flowchart: Shows the causal relationship of the reasoning path, the complete process from input to output; Timeline: Displays the chronological order of reasoning steps, visually showing the time point and duration of each reasoning step; Node graph: Shows the relationships between features. Nodes represent features, and edges represent causal relationships or correlations between features. Heatmap: Shows the importance and impact of features; the intensity of the color indicates the importance or impact of the feature.

[0118] In addition, the chart also supports interactive features with users, such as: Node Expand / Collapse: Clicking a node expands or collapses its child nodes; Path backtracking: Clicking on a node will display its predecessor nodes and the inference path; Feature information viewing: Clicking on a node allows you to view detailed information about the feature (feature value, SHAP value, impact description, etc.); Timeline display: View the time point and duration of each inference step; Zoom function: Supports zooming in and out to view inference paths at different levels; Export function: Supports exporting charts to PNG, SVG, PDF and other formats.

[0119] In addition, in practical applications, some of the above methods can be replaced by the following alternatives: Alternative Solution 1: Alternative to the Feature Engineering Layer Original solution: Hard-coding of universal features + loading of specific feature parameters; Alternative solution: 1. All hard-coded: All database-specific features are hard-coded into the system. Advantages: Simple to implement, no configuration files required; Disadvantages: Large code size, difficult to maintain, and poor scalability.

[0120] 2. Load all parameters: Universal features are also loaded via parameter loading. Advantages: Unified loading method, easy to manage; Disadvantages: Slightly lower performance and complex configuration files.

[0121] 3. Hybrid approach: Loading universal feature parameters + hard encoding specific features; Advantages: High flexibility; Disadvantage: Complex logic.

[0122] Alternative Solution 2: An alternative to periodic feature recognition methods Original solution: Multi-method fusion (FFT + autocorrelation + STL + spectral entropy).

[0123] Alternative solution: 1. Single method: using only FFT or only autocorrelation; –Advantages: Simple to implement, fast calculation; –Disadvantages: Low accuracy and poor robustness.

[0124] 2. Deep Learning: Using deep learning models such as LSTM and Transformer; –Advantages: Automatic learning cycle characteristics; – Disadvantages: Requires a large amount of training data and has poor interpretability.

[0125] 3. Statistical methods: Use statistical tests (such as the KPSS test); –Advantages: Rigorous theory; –Disadvantage: It can only detect the existence of a period, but cannot identify the length of the period.

[0126] Alternative Solution 3: Alternatives to the Business Volume Normalization Method Original solution: dual-model architecture (baseline + recent) + dynamic weights.

[0127] Alternative solution: 1. Single model: Using only the baseline model or only the recent model; –Advantages: Simple to implement; – Disadvantages: Poor stability or adaptability.

[0128] 2. Fixed weights: Baseline and recent weights are fixed; –Advantages: Simple to implement; – Disadvantage: It cannot adapt to changes in data quality.

[0129] 3. Online learning: The model is updated in real time using online learning algorithms; –Advantages: High real-time performance; – Disadvantages: Complex to implement and requires more computing resources.

[0130] Alternative Solution 4: Alternatives to Version Correction Methods Original plan: Version impact factor + correction period mechanism + linear transition; Alternative solution: 1. Instant Switch: Switch to the new model immediately after the version update; –Advantages: Simple to implement; –Disadvantage: High false alarm rate.

[0131] 2. Exponential Smoothing: Uses exponential smoothing instead of a non-linear transition; –Advantages: Smoother; –Disadvantage: Parameter tuning is complex.

[0132] 3. No correction period: No correction period is set; the model is updated directly. –Advantages: Simple to implement; – Disadvantages: High false positive rate after version update.

[0133] Alternative Option 5: Alternatives for the AI ​​Model Layer Original solution: XGBoost + hierarchical modeling + model fusion; Alternative solution: 1. Single model: using only a general model or only a specific model; –Advantages: Simple to implement; –Disadvantage: Lower accuracy.

[0134] 2. Deep learning: using LSTM, Transformer, etc.; –Advantages: Automatic feature learning; – Disadvantage: Requires a large amount of training data.

[0135] 3. Ensemble learning: Using ensemble methods such as Bagging and Boosting; –Advantages: High accuracy; –Disadvantage: High computational complexity.

[0136] Alternative Option Six: Alternatives to the Interpretability Layer Original solution: SHAP + reasoning path + chain of evidence; Alternative solution: 1. LIME: Use LIME instead of SHAP; –Advantages: Fast computation; – Disadvantage: Poor stability.

[0137] 2. Feature Importance: Only feature importance is used; –Advantages: Simple to implement; – Disadvantage: Weak explanatory power.

[0138] 3. Decision Tree: Using decision trees as an interpretable model; –Advantages: Naturally explainable; –Disadvantage: Lower accuracy.

[0139] like Figure 7 As shown, another embodiment of the present invention also provides a business-aware database operation and maintenance AI diagnostic device, including: The acquisition module is used to collect candidate database performance index data of the target database from multiple data sources. The candidate database performance index data is related to the operating system, database characteristics, and business. The correction module is used to correct the candidate database performance index data based on the current version information of the target database to obtain database performance index data. The first determining module is used to monitor the business data of the target database and, in combination with the business data and database performance index data, determine the impact coefficient and business volume impact value of each database performance index on the business data. The first generation module is used to generate a business data context by combining the business data, the preset baseline business volume, the impact coefficient, and the business volume impact value. The second determining module is used to process the business data by combining clustering algorithms, preset business semantics and business types, and determine the periodic characteristics of the business data; The extraction module is used to extract features from the business data context, periodic features, and normalized database performance index data and business data. The normalization is related to business volume perception and target database version perception. The reasoning module is used to perform diagnostic reasoning on the extracted features using a pre-built target model to obtain diagnostic results.

[0140] In one embodiment, the device further includes: The third determining module is used to determine the SHAP value of each feature based on the diagnostic results and the extracted features; The ranking module is used to rank the features based on their importance and contribution by combining the SHAP values; The second generation module is used to generate a feature list based on the ranking results and the SHAP values ​​of each feature.

[0141] In one embodiment, the device further includes: The analysis module is used to perform root cause analysis of the fault by combining the SHAP value and the preset SHAP value classification standard when the diagnostic results characterize the fault of the target database, to determine the root cause of the fault, and to record the evidence chain that matches the reasoning path and reasoning process.

[0142] In one embodiment, the device further includes: The fourth determination module is used to determine the multi-dimensional credibility assessment results of the diagnostic results, and to determine the comprehensive confidence level based on the credibility assessment results. The multi-dimensional credibility assessment results include the target model confidence level, SHAP confidence level, historical confidence level determined based on historical similar data, and business confidence level corresponding to business uncertainty. The business uncertainty is determined based on business volume fluctuations, version updates, and system configuration changes. The fifth determination module is used to determine the credibility level and corresponding operational suggestions regarding the diagnostic results based on the credibility assessment results and preset credibility rating standards.

[0143] In one embodiment, the device further includes: The processing module is used to process the diagnostic results, feature list, evidence chain, business data context, and operation suggestions through a large language model to generate an enhanced diagnostic report. The evidence chain includes the entire reasoning process of combining the SHAP value and the preset SHAP value grading standard to perform root cause analysis of the fault and determine the root cause of the fault. The sedimentation module is used to combine the content of the enhanced diagnostic report and the DBA feedback data in the target database and sediment it into a pre-built knowledge base to form a case. The construction module is used to build indexes and tags for the cases stored in the knowledge base.

[0144] In one embodiment, the step of correcting the candidate database performance indicator data based on the current version information of the target database to obtain database performance indicator data includes: Calculate the version impact factor based on the database performance metrics before and after the version update; The correction period for the current correction is calculated by combining the version impact factor and the preset basic correction period. The correction progress is calculated based on the correction period and the time interval between the current time and the update time of the current version; The revised database performance metrics are calculated based on the metric values ​​of the database performance metrics to be revised, the version impact factor, and the revision progress.

[0145] In one embodiment, monitoring the business data of the target database and determining the impact coefficient and business volume impact value of each database performance indicator on the business data in conjunction with the business data and database performance indicator data includes: The rate of change in business volume is determined based on the data volume of the aforementioned business data and the baseline business volume. The business volume change rate and the statistical database performance index change rate are learned based on the linear regression method, and the current influence coefficient is updated based on the learning results. The initial value of the influence coefficient is a preset value. The impact value of the business volume is determined based on the rate of change of the business volume and the impact coefficients of the updated corresponding different database performance indicators. The device further includes: The module is used to obtain the false alarm rate of alarms; The adjustment module is used to learn the false alarm rate and the rate of change of business volume using the gradient descent method, and adjust the current adjustment coefficient based on the learning result. The initial value of the adjustment coefficient is a preset value. The first update module is used to update the preset static threshold based on the business volume change rate and the adjusted adjustment coefficient to obtain the dynamic threshold, wherein the static threshold is the index value range of the database performance index. The process of generating a business data context by combining the business data, a preset baseline business volume, an impact coefficient, and a business volume impact value includes: The business data context is generated by combining the aforementioned business data, baseline business volume, impact coefficient, business volume impact value, business volume change rate, and dynamic threshold.

[0146] In one embodiment, the step of processing the business data by combining clustering algorithms, preset business semantics, and business types to determine the periodic characteristics of the business data includes: The business data is preprocessed; Spectral analysis and peak detection are performed on the preprocessed business data, and the spectral analysis results and peak detection results are learned based on unsupervised learning methods to obtain the first candidate period list; The first candidate period in the candidate period list is clustered based on a clustering algorithm to obtain the second candidate period list; The second candidate period in the second candidate period list is verified by the autocorrelation algorithm to obtain the period whose autocorrelation coefficient meets the preset requirements; The business semantics of the period are determined based on a preset semantic mapping rule containing weights, and the business semantics describe the temporal characteristics of the period. The periodic characteristics of the period are determined based on the periodic attribute features, business semantics, weights, and periodic confidence, and the periodic confidence is calculated based on the peak detection results and autocorrelation coefficients.

[0147] In one embodiment, the feature extraction of the business data context, periodic features, and normalized database performance index data and business data includes: The business data context, periodic features, and normalized database performance index data and business data are respectively extracted using basic features and specific features. The basic features are common features of different target databases, and the specific features are unique features of different target databases. A baseline model is constructed based on database performance metrics data from the first time period of the historical period, and a recent model is constructed based on database performance metrics data from the second time period of the historical period. The baseline model is used to determine the long-term baseline of database performance metrics, and the recent model is used to capture recent business change trends. The first time period is longer than the second time period. The rate of change in business volume is determined based on the data volume of the aforementioned business data and the baseline business volume. The weight parameters of the baseline model and the recent model are determined based on the rate of change in business volume. The target baseline is determined based on the baseline model, recent model, and weight parameters. Based on the aforementioned basic features, specific features, target baseline, and business volume impact value, determine the normalized features corresponding to the aforementioned basic features and specific features; The device further includes: The correction module is used to correct the drifted target baseline based on the version impact factor and correction period when the target baseline drifts. The second update module is used to update the normalized features based on the corrected target baseline.

[0148] Another embodiment of the present invention also provides an electronic device, comprising: One or more processors; Memory, configured to store one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the business-aware database operation and maintenance AI diagnostic method as described in any one of the above descriptions.

[0149] Furthermore, one embodiment of the present invention also provides a storage medium storing a computer program thereon, which, when executed by a processor, implements the business-aware database operation and maintenance AI diagnostic method described above. It should be understood that the various solutions in this embodiment have the corresponding technical effects in the above-described method embodiments, and will not be repeated here.

[0150] Furthermore, embodiments of the present invention also provide a computer program product, which is tangibly stored on a computer-readable medium and includes computer-readable instructions, which, when executed, cause at least one processor to perform a business-aware database operation and maintenance AI diagnostic method as described in the embodiments above.

[0151] It should be noted that the computer storage medium of the present invention can be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, system, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, a random access storage medium (RAM), a read-only storage medium (ROM), an erasable programmable read-only storage medium (EPROM or flash memory), an optical fiber, a portable compact disk read-only storage medium (CD-ROM), an optical storage medium, a magnetic storage medium, or any suitable combination thereof. In the present invention, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, system, or device. In the present invention, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media can also be any computer-readable medium other than computer-readable storage media, which can send, propagate, or transmit a program configured for use by or in connection with an instruction execution system, system, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wireless, antenna, optical fiber, RF, etc., or any suitable combination thereof.

[0152] Furthermore, those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage and optical storage) containing computer-usable program code.

[0153] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A system that specifies functions in one or more boxes.

[0154] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including an instruction set implemented in a process. Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0155] Those skilled in the art should understand that the discussion of any of the above embodiments is merely exemplary and is not intended to imply that the scope of protection of this application is limited to these examples; within the framework of this application, the technical features of the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other variations of different aspects of one or more embodiments of this application as described above, which are not provided in detail for the sake of brevity.

Claims

1. A business-aware AI-based diagnostic method for database operation and maintenance, characterized in that, include: The candidate database performance metrics data of the target database are collected from multiple data sources. The candidate database performance metrics data are related to the operating system, database characteristics, and business. Based on the current version information of the target database, the performance index data of the candidate database is corrected to obtain the database performance index data; Monitor the business data of the target database, and combine the business data and database performance index data to determine the impact coefficient and business volume impact value of each database performance index on the business data; A business data context is generated by combining the aforementioned business data, preset baseline business volume, impact coefficient, and business volume impact value; The business data is processed by combining clustering algorithms, preset business semantics, and business types to determine the periodic characteristics of the business data; Feature extraction is performed on the business data context, periodic features, and normalized database performance index data and business data. The normalization is related to business volume perception and target database version perception. The extracted features are used to perform diagnostic reasoning using a pre-built target model to obtain diagnostic results.

2. The business-aware database operation and maintenance AI diagnostic method according to claim 1, characterized in that, The method further includes: Based on the diagnostic results and the extracted features, the SHAP value of each feature is determined; The importance and contribution of the features are ranked based on the SHAP values. A feature list is generated based on the ranking results and the SHAP values ​​of each of the aforementioned features.

3. The business-aware database operation and maintenance AI diagnostic method according to claim 2, characterized in that, The method further includes: When the diagnostic results characterize the target database failure, the root cause analysis of the failure is performed by combining the SHAP value and the preset SHAP value grading standard to determine the root cause of the failure and record the evidence chain that matches the reasoning path and reasoning process.

4. The business-aware database operation and maintenance AI diagnostic method according to claim 2, characterized in that, The method further includes: Determine the multi-dimensional credibility assessment results of the diagnostic results, and determine the comprehensive confidence level based on the credibility assessment results. The multi-dimensional credibility assessment results include the target model confidence level, SHAP confidence level, historical confidence level determined based on historical similar data, and business confidence level corresponding to business uncertainty. The business uncertainty is determined based on business volume fluctuations, version updates, and system configuration changes. Based on the credibility assessment results and the preset credibility rating criteria, the credibility level and corresponding operational recommendations regarding the diagnostic results are determined.

5. The business-aware database operation and maintenance AI diagnostic method according to claim 4, characterized in that, The method further includes: An enhanced diagnostic report is generated by processing the diagnostic results, feature list, evidence chain, business data context, and operation suggestions through a large language model. The evidence chain includes the entire reasoning process for determining the root cause of the fault by combining the SHAP value and the preset SHAP value grading standard to perform root cause analysis of the fault. The content of the enhanced diagnostic report and the DBA feedback data in the target database are combined and stored in a pre-built knowledge base to form a case study; An index and tags are constructed for the cases stored in the knowledge base.

6. The business-aware database operation and maintenance AI diagnostic method according to claim 1, characterized in that, The process of correcting the candidate database performance indicator data based on the current version information of the target database to obtain database performance indicator data includes: Calculate the version impact factor based on the database performance metrics before and after the version update; The correction period for the current correction is calculated by combining the version impact factor and the preset basic correction period. The correction progress is calculated based on the correction period and the time interval between the current time and the update time of the current version; The revised database performance metrics are calculated based on the metric values ​​of the database performance metrics to be revised, the version impact factor, and the revision progress.

7. The business-aware database operation and maintenance AI diagnostic method according to claim 1, characterized in that, The process involves monitoring the business data of the target database and, in conjunction with the business data and database performance metrics, determining the impact coefficient and business volume impact value of each database performance metric on the business data, including: The rate of change in business volume is determined based on the data volume of the aforementioned business data and the baseline business volume. The business volume change rate and the statistical database performance index change rate are learned based on the linear regression method, and the current influence coefficient is updated based on the learning results. The initial value of the influence coefficient is a preset value. The impact value of the business volume is determined based on the rate of change of the business volume and the impact coefficients of the updated corresponding different database performance indicators. The method further includes: Obtain information about the false alarm rate for alarms; The false alarm rate and the rate of change in traffic volume are learned using the gradient descent method, and the current adjustment coefficient is adjusted based on the learning results. The initial value of the adjustment coefficient is a preset value. The preset static threshold is updated based on the business volume change rate and the adjusted adjustment coefficient to obtain the dynamic threshold. The static threshold is the range of index values ​​for the database performance index. The process of generating a business data context by combining the business data, a preset baseline business volume, an impact coefficient, and a business volume impact value includes: The business data context is generated by combining the aforementioned business data, baseline business volume, impact coefficient, business volume impact value, business volume change rate, and dynamic threshold.

8. The business-aware database operation and maintenance AI diagnostic method according to claim 1, characterized in that, The process of combining clustering algorithms, preset business semantics, and business types to process the business data and determine the periodic characteristics of the business data includes: The business data is preprocessed; Spectral analysis and peak detection are performed on the preprocessed business data, and the spectral analysis results and peak detection results are learned based on unsupervised learning methods to obtain the first candidate period list; The first candidate period in the candidate period list is clustered based on a clustering algorithm to obtain the second candidate period list. The second candidate period in the second candidate period list is verified by the autocorrelation algorithm to obtain the period whose autocorrelation coefficient meets the preset requirements; The business semantics of the period are determined based on a preset semantic mapping rule containing weights, and the business semantics describe the temporal characteristics of the period. The periodic characteristics of the period are determined based on the periodic attribute features, business semantics, weights, and periodic confidence, and the periodic confidence is calculated based on the peak detection results and autocorrelation coefficients.

9. The business-aware database operation and maintenance AI diagnostic method according to claim 6, characterized in that, The feature extraction of the business data context, periodic features, and normalized database performance metrics and business data includes: The business data context, periodic features, and normalized database performance index data and business data are respectively extracted using basic features and specific features. The basic features are common features of different target databases, and the specific features are unique features of different target databases. A baseline model is constructed based on database performance metrics data from the first time period of the historical period, and a recent model is constructed based on database performance metrics data from the second time period of the historical period. The baseline model is used to determine the long-term baseline of database performance metrics, and the recent model is used to capture recent business change trends. The first time period is longer than the second time period. The rate of change in business volume is determined based on the data volume of the aforementioned business data and the baseline business volume. The weight parameters of the baseline model and the recent model are determined based on the rate of change in business volume. The target baseline is determined based on the baseline model, recent model, and weight parameters. Based on the aforementioned basic features, specific features, target baseline, and business volume impact value, determine the normalized features corresponding to the aforementioned basic features and specific features; The method further includes: In the event that the target baseline has drifted, the drifted target baseline shall be corrected based on the version impact factor and correction period. The normalized features are updated based on the corrected target baseline.

10. A business-aware AI-based diagnostic device for database operation and maintenance, characterized in that, include: The acquisition module is used to collect candidate database performance index data of the target database from multiple data sources. The candidate database performance index data is related to the operating system, database characteristics, and business. The correction module is used to correct the candidate database performance index data based on the current version information of the target database to obtain database performance index data. The first determining module is used to monitor the business data of the target database and, in combination with the business data and database performance index data, determine the impact coefficient and business volume impact value of each database performance index on the business data. The first generation module is used to generate a business data context by combining the business data, the preset baseline business volume, the impact coefficient, and the business volume impact value. The second determining module is used to process the business data by combining clustering algorithms, preset business semantics and business types, and determine the periodic characteristics of the business data; The extraction module is used to extract features from the business data context, periodic features, and normalized database performance index data and business data. The normalization is related to business volume perception and target database version perception. The reasoning module is used to perform diagnostic reasoning on the extracted features using a pre-built target model to obtain diagnostic results.

Images