Kylin system performance optimization method based on machine learning

Abstract

The invention discloses a machine learning-based Kylin system performance optimization method, which comprises the steps of data processing, scene recognition, bottleneck detection and performance optimization, and is characterized in that performance data are acquired through Poptimization and Promeheus, are stored in a MySQL database after verification, and key features are screened by using feature engineering; multiple machine learning models are adopted for scene recognition, an optimal model is screened after evaluation, and a scene recognition model is generated through fusion and used for dynamically matching a scene; constructing a range isolation forest model to detect abnormity, and positioning a bottleneck root cause in combination with a technology; and generating an optimal configuration parameter combination based on a Bayesian optimization algorithm and the proxy model, and dynamically adjusting and verifying the effect. According to the method, efficient processing, accurate identification, rapid positioning and automatic optimization of system performance are realized, the system performance is improved, and the method is suitable for a high-performance scene.

Description

Technical Field

[0001] The present invention belongs to the technical field of operating system optimization, and specifically relates to a Kirin system performance optimization method based on machine learning. Background Art

[0002] With the rapid development and continuous expansion of the domestic operating system field, more and more enterprises and individual users are choosing domestic operating systems as tools for their daily business activities; Kylin system, as an outstanding representative of domestic operating systems, has begun to cover key areas such as servers, cloud computing, and high-performance computing. Despite this, the system often encounters internal process problems such as memory leaks, abnormal disk I / O usage, and a sharp increase in CPU usage, which lead to overall performance degradation. In the face of these problems, how to quickly and accurately identify abnormal processes and take effective measures to enhance system performance has become a crucial task. This not only involves the stability and reliability of the system, but also directly affects the user experience and work efficiency. Therefore, how to use intelligent technology to improve system performance and ensure that the Kylin operating system can maintain efficient and stable operation even in complex and changing environments has become the focus of industry attention.

[0003] Currently, performance optimization primarily relies on the following technical approaches: manually analyzing system logs and adjusting parameters; automatically executing optimization measures based on preset thresholds; and allocating resources through rules and conditions. However, these methods have limitations: the first approach relies too heavily on experience, resulting in slow processing speeds, making it difficult to adapt to rapidly changing environments and prone to overlooking potential performance bottlenecks; the second approach often uses static threshold settings, lacking the necessary dynamic adaptability, resulting in delayed resource adjustments; and the third approach lacks sufficient flexibility, making it difficult to adapt to complex scenarios and requiring high maintenance costs.

[0004] Therefore, the present invention provides a Kirin system performance optimization method based on machine learning to solve the above technical problems. Summary of the Invention

[0005] In response to the above problems, the purpose of the present invention is to provide a Kirin system performance optimization method based on machine learning, aiming to achieve efficient processing of Kirin system performance data, accurate identification of scenarios, rapid location of bottlenecks and automated performance optimization, thereby improving the system's response speed, stability and resource utilization.

[0006] The present invention provides a method for optimizing the performance of a Kylin system based on machine learning, comprising:

[0007] S1, Data Processing: Collect Kylin system performance data in real time, verify the integrity of the collected performance data via Socket communication, store the verified performance data in a MySQL database, and filter key features from the stored performance data through feature engineering;

[0008] S2, Scene Recognition: A comprehensive analysis of the performance data after screening out key features is performed using multiple machine learning models. The performance of each machine learning model is determined through model evaluation. The machine learning models with good performance are stacked and fused to generate a scene recognition model for dynamic matching of system scenarios.

[0009] S3, Bottleneck Detection: We build a range isolation forest model and use it to test Kylin system performance to identify performance anomalies. When Kylin system performance anomalies are detected, we combine causal analysis and FP-growth technology to conduct in-depth analysis of the anomaly to locate the root cause of the Kylin system performance bottleneck.

[0010] S4, performance optimization: Generate the optimal configuration parameter combination for the Kylin system based on the Bayesian optimization algorithm, dynamically adjust the operating parameters of the Kylin system based on the performance data collected in real time and in combination with the optimal configuration parameter combination; and perform performance verification on the adjusted operating parameters of the Kylin system to determine the optimization effect.

[0011] Preferably, the performance data includes the CPU usage, memory occupancy, disk I / O rate and network traffic data of the Kylin system.

[0012] Preferably, the data processing in S1 specifically includes the following steps:

[0013] Data Collection: Kylin system performance data is collected in real time through Psutil and Prometheus. A communication connection is established by binding a specified IP address and port through Socket, and the binary data sent by the client is received and parsed. The integrity of the performance data is verified based on the length field in the data message, and the verified binary data is deserialized into a structured performance data object.

[0014] Data storage: Classify the performance data objects according to data type and acquisition timestamp, store the classified performance data objects in a thread-safe queue, and perform batch write operations through the transaction management mechanism of the MySQL database to form raw performance data storage to ensure consistency and integrity of the performance data storage process;

[0015] Data preprocessing: After extracting the raw performance data from the MySQL database, invalid values ​​and abnormal data points are removed according to preset rules. Numerical features of different dimensions are normalized using a standardization method. Performance trend features of time series data are extracted based on a sliding time window algorithm. Categorical features are converted to binary vectors using one-hot encoding.

[0016] Feature Engineering: Use regular expressions to parse complex data fields, calculate the correlation between features using the Spearman correlation coefficient matrix, eliminate redundant features above the correlation threshold, and calculate the importance score of each feature to the system performance indicator based on the feature importance scoring mechanism of the random forest model, and screen out key features with importance scores higher than the preset threshold.

[0017] Preferably, the scene recognition in S2 specifically includes the following steps:

[0018] Model training: Use at least five machine learning models, including support vector machine (SVM), CatBoost, XGBoost, LightGBM, logistic regression, neural network, random forest (RF), and decision tree, to train the selected key features. Build a multi-classification model for typical Kylin system operating scenarios. Each model outputs a probability distribution vector representing the Kylin system in different scenarios.

[0019] Model evaluation: The classification performance of each model was quantified using the ROC curve and AUC value. The stability and generalization ability of each model on different data subsets were evaluated using the cross-validation method. Based on the evaluation results, the three models with the top three AUC values ​​and the smallest cross-validation variance were selected for stacking and fusion. The predicted probability vectors of the three selected models were used as input features, and the gated recurrent unit (GRU) was used as the meta-model to generate the final scene recognition model.

[0020] Dynamic matching: The performance data collected in real time and processed in step S1 is input into the trained scene recognition model to obtain the probability distribution of the current scene, and the difference in the probability distribution of two adjacent predictions is calculated. When the difference value exceeds the preset threshold, the dynamic adjustment process of the Kirin system configuration parameters in step S4 is triggered.

[0021] Preferably, the bottleneck detection in S3 specifically includes the following steps:

[0022] Anomaly Detection: The Kylin system's operating status is modeled using the Extreme Isolation Forest model. When identifying performance indicators that deviate from the normal range, the system automatically adjusts detection sensitivity to suit different system scenarios.

[0023] Root cause analysis: Use causal analysis methods to determine the causal relationship between abnormal indicators and system performance, and use the FP-growth algorithm to mine the association rules between abnormal indicators to locate the root cause of performance bottlenecks.

[0024] Preferably, the performance optimization in S4 specifically includes the following steps:

[0025] Configuration file scanning and parameter extraction: traverse the specified directory to obtain all configuration files, and call the corresponding parser based on the file format characteristics to extract configuration parameters;

[0026] Parameter screening: Build a random forest evaluation model and use it to evaluate the impact of configuration parameters on Kylin system performance, screening out key configuration parameters with significant impact;

[0027] Optimization strategy generation: Using the Bayesian optimization framework, combined with the CatBoost gradient boosting tree as a proxy model, and PI (Probability of Improvement) as the objective function, we reduce the search complexity in the high-dimensional parameter space through sequential experimental design and generate the optimal configuration parameter combination;

[0028] Dynamic optimization and verification: During Kylin system operation, key configuration parameters are dynamically adjusted based on real-time performance data. The optimization effect is verified by comparing the changes in system response time and resource utilization indicators before and after, and the optimization strategy library is continuously updated.

[0029] Preferably, the anomaly detection threshold of the extreme difference isolation forest model is dynamically adjusted through the quantile of historical data to adapt to the detection requirements of different system scenarios.

[0030] Preferably, the transaction mechanism of the MySQL database ensures consistency and integrity of data storage through ACID properties, which include atomicity, consistency, isolation and persistence.

[0031] Compared with related technologies, the machine learning-based Kylin system performance optimization method provided by the present invention has the following beneficial effects:

[0032] 1. Efficient and accurate data processing: The integration of Psutil and Prometheus tools enables real-time collection and monitoring of Kylin system performance indicators such as CPU usage and memory utilization. Compared with traditional manual or non-real-time collection methods, this greatly improves the timeliness of data acquisition. Data integrity verification is performed through Socket communication, and the transaction mechanism of the MySQL database ACID characteristics ensures the consistency and integrity of data storage, avoiding data loss or errors. Feature engineering methods such as regular expression parsing, Spearman correlation coefficient analysis, and random forest feature importance scoring are used to effectively eliminate redundant features and screen key features, significantly improving the efficiency and accuracy of data processing and laying a solid foundation for subsequent performance analysis.

[0033] 2. Accurate and real-time scene recognition: This system uses at least five machine learning models, including support vector machines and CatBoost, to fuse and stack them. Compared to a single model, it can more comprehensively mine data features and improve the accuracy of scene recognition. Through ROC curves, AUC value quantitative evaluation, and cross-validation, the best-performing model is selected for fusion, which enhances the robustness of the model, significantly improves the timeliness of scene recognition, and better adapts to dynamic changes during system operation.

[0034] 3. Rapid and accurate bottleneck location: This approach builds an extreme isolation forest model and dynamically adjusts the anomaly detection threshold using historical data quantiles. This model can accurately detect system performance bottlenecks. Compared with traditional fixed-threshold detection methods, it is more adaptable to different system scenarios. Combining causal analysis methods with FP-growth mining technology, it not only identifies performance anomalies but also deeply analyzes the causal relationships and potential correlations between abnormal indicators, enabling rapid location of system bottlenecks and anomaly root cause analysis, improving problem-solving efficiency.

[0035] 4. Automated performance optimization: Using the Bayesian optimization method combined with the CatBoost proxy model, with PI as the objective function, it efficiently searches for the optimal configuration parameter combination in the high-dimensional parameter space. Compared with blind adjustment or empirical parameter setting, it greatly improves the targetedness and effectiveness of optimization. It dynamically adjusts key configuration parameters based on real-time performance data, verifies the optimization effect by comparing system response time and resource utilization indicators, and continuously updates the optimization strategy library to achieve automated optimization and continuous improvement of system performance, significantly improving the system's operating efficiency, stability, and resource utilization. It is suitable for various high-performance scenarios such as cloud computing platforms, big data processing systems, and distributed systems, and provides strong support for Kylin system performance assurance under complex business loads. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 This is a flow chart of a method for optimizing Kylin system performance based on machine learning. DETAILED DESCRIPTION

[0037] The present invention provides a Kirin system performance optimization method based on machine learning, which aims to solve the problems that traditional optimization methods lack sufficient flexibility, are difficult to adapt to complex scenarios, and have high maintenance costs.

[0038] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0039] Please see the attached Figure 1 As shown, the present invention provides a method for optimizing the performance of a Kylin system based on machine learning, comprising:

[0040] S1, Data Processing: Collect Kylin system performance data in real time, verify the integrity of the collected performance data via Socket communication, store the verified performance data in a MySQL database, and filter key features from the stored performance data through feature engineering;

[0041] S2, Scene Recognition: A comprehensive analysis of the performance data after screening out key features is performed using multiple machine learning models. The performance of each machine learning model is determined through model evaluation. The machine learning models with good performance are stacked and fused to generate a scene recognition model for dynamic matching of system scenarios.

[0042] S3, Bottleneck Detection: We build a range isolation forest model and use it to test Kylin system performance to identify performance anomalies. When Kylin system performance anomalies are detected, we combine causal analysis and FP-growth technology to conduct in-depth analysis of the anomaly to locate the root cause of the Kylin system performance bottleneck.

[0043] S4, performance optimization: Generate the optimal configuration parameter combination for the Kylin system based on the Bayesian optimization algorithm, dynamically adjust the operating parameters of the Kylin system based on the performance data collected in real time and in combination with the optimal configuration parameter combination; and perform performance verification on the adjusted operating parameters of the Kylin system to determine the optimization effect.

[0044] In this embodiment, the performance data includes the CPU usage, memory occupancy, disk I / O rate and network traffic data of the Kylin system, but of course, it is not limited to this.

[0045] Specifically, the data processing in S1 includes the following steps: data collection: collect the performance data of the Kylin system in real time through Psutil and Prometheus, establish a communication connection through Socket binding to the specified IP address and port, receive and parse the binary format data sent by the client; verify the integrity of the performance data based on the length field in the data message, and deserialize the verified binary data into a structured performance data object; data storage: classify the performance data objects according to data type and collection timestamp, store the classified performance data objects in a thread-safe queue, and perform batch write operations through the transaction management mechanism of the MySQL database to form the original performance data storage to ensure the performance data storage process. Consistency and integrity; Data preprocessing: After extracting the raw performance data from the MySQL database, invalid values ​​and abnormal data points of the raw performance data are eliminated according to preset rules, and numerical features of different dimensions are normalized using a standardized method. The performance trend characteristics of time series data are extracted based on a sliding time window algorithm, and categorical features are subjected to one-hot encoding conversion to convert discrete categories into binary vector representations; Feature engineering: Regular expressions are used to parse data fields in complex formats, and the correlation between features is calculated using the Spearman correlation coefficient matrix. Redundant features above the correlation threshold are eliminated, and the feature importance scoring mechanism based on the random forest model is used to calculate the importance score of each feature to the system performance indicator, and screen out key features with importance scores higher than the preset threshold.

[0046] Furthermore, the scene recognition in S2 specifically includes the following steps: model training: using at least five machine learning models including support vector machine (SVM), CatBoost, XGBoost, LightGBM, logistic regression, neural network, random forest (RF) and decision tree to train the screened key features, and construct a multi-classification model for the typical operating scenarios of the Kylin system, and each model outputs a probability distribution vector that characterizes the Kylin system in different scenarios; model evaluation: quantifying the classification performance of each model through ROC curve and AUC value, and using cross-validation method to evaluate the stability and generalization ability of each model on different data subsets; based on the evaluation results, select the top three models with the smallest cross-validation variance in AUC value for stacking and fusion, use the predicted probability vectors of the three selected models as input features, and use the gated recurrent unit (GRU) as the meta-model to generate the final scene recognition model. Dynamic matching: The performance data collected in real time and processed in step S1 is input into the trained scene recognition model to obtain the probability distribution of the current scene, and the difference in the probability distribution of two adjacent predictions is calculated. When the difference value exceeds the preset threshold, the dynamic adjustment process of the Kirin system configuration parameters in step S4 is triggered.

[0047] It should be noted that support vector machine (SVM): suitable for high-dimensional data classification, has strong generalization ability, and can effectively handle nonlinear classification problems; CatBoost, XGBoost, LightGBM: gradient-boosting based integrated learning models, can efficiently process structured data, suitable for processing large-scale data sets, with high prediction accuracy and computational efficiency; logistic regression: suitable for linear classification problems, with high computational efficiency, suitable for processing low-dimensional data; neural network: can capture complex nonlinear relationships, suitable for processing high-dimensional and unstructured data, with strong feature extraction capabilities; random forest (RF) and decision tree: tree-based models, with good interpretability and anti-overfitting capabilities, suitable for processing medium-sized data sets.

[0048] In this embodiment, the bottleneck detection in S3 specifically includes the following steps: anomaly detection: modeling the Kylin system operating status through the extreme difference isolation forest model, and automatically adjusting the detection sensitivity to adapt to different system scenarios in the process of identifying performance indicators that deviate from the normal range; root cause analysis: using the causal analysis method to determine the causal relationship between abnormal indicators and system performance, and using the FP-growth algorithm to mine the association rules between abnormal indicators to locate the root cause of the performance bottleneck.

[0049] Furthermore, the performance optimization in S4 specifically includes the following steps: configuration file scanning and parameter extraction: traverse the specified directory to obtain all configuration files, and call the corresponding parser based on the file format features to extract configuration parameters; parameter screening: build a random forest evaluation model, and use the random forest model to evaluate the impact of configuration parameters on the performance of the Kylin system, and screen out key configuration parameters with significant impact; optimization strategy generation: adopt the Bayesian optimization framework, combine the CatBoost gradient boosting tree as the agent model, and use PI (Probability of Improvement) as the objective function. Through sequential experimental design, the search complexity of the high-dimensional parameter space is reduced to generate the optimal configuration parameter combination; dynamic optimization and verification: during the operation of the Kylin system, key configuration parameters are dynamically adjusted according to real-time performance data, and the optimization effect is verified by comparing the changes before and after the system response time and resource utilization indicators, and the optimization strategy library is continuously updated.

[0050] In this embodiment, the anomaly detection threshold of the extreme difference isolation forest model is dynamically adjusted through the quantile of historical data to adapt to the detection requirements of different system scenarios; the transaction mechanism of the MySQL database ensures the consistency and integrity of data storage through the ACID characteristics, and the ACID characteristics include atomicity, consistency, isolation and persistence.

[0051] Compared with related technologies, the machine learning-based Kylin system performance optimization method provided by the present invention has the following beneficial effects:

[0052] 1. Efficient and accurate data processing: The integration of Psutil and Prometheus tools enables real-time collection and monitoring of Kylin system performance indicators such as CPU usage and memory utilization. Compared with traditional manual or non-real-time collection methods, this greatly improves the timeliness of data acquisition. Data integrity verification is performed through Socket communication, and the transaction mechanism of the MySQL database ACID characteristics ensures the consistency and integrity of data storage, avoiding data loss or errors. Feature engineering methods such as regular expression parsing, Spearman correlation coefficient analysis, and random forest feature importance scoring are used to effectively eliminate redundant features and screen key features, significantly improving the efficiency and accuracy of data processing and laying a solid foundation for subsequent performance analysis.

[0053] 2. Accurate and real-time scene recognition: This system uses at least five machine learning models, including support vector machines and CatBoost, to fuse and stack them. Compared to a single model, it can more comprehensively mine data features and improve the accuracy of scene recognition. Through ROC curves, AUC value quantitative evaluation, and cross-validation, the best-performing model is selected for fusion, which enhances the robustness of the model, significantly improves the timeliness of scene recognition, and better adapts to dynamic changes during system operation.

[0054] 3. Rapid and accurate bottleneck location: This approach builds an extreme isolation forest model and dynamically adjusts the anomaly detection threshold using historical data quantiles. This model can accurately detect system performance bottlenecks. Compared with traditional fixed-threshold detection methods, it is more adaptable to different system scenarios. Combining causal analysis methods with FP-growth mining technology, it not only identifies performance anomalies but also deeply analyzes the causal relationships and potential correlations between abnormal indicators, enabling rapid location of system bottlenecks and anomaly root cause analysis, improving problem-solving efficiency.

[0055] 4. Automated performance optimization: Using the Bayesian optimization method combined with the CatBoost proxy model, with PI as the objective function, it efficiently searches for the optimal configuration parameter combination in the high-dimensional parameter space. Compared with blind adjustment or empirical parameter setting, it greatly improves the targetedness and effectiveness of optimization. It dynamically adjusts key configuration parameters based on real-time performance data, verifies the optimization effect by comparing system response time and resource utilization indicators, and continuously updates the optimization strategy library to achieve automated optimization and continuous improvement of system performance, significantly improving the system's operating efficiency, stability, and resource utilization. It is suitable for various high-performance scenarios such as cloud computing platforms, big data processing systems, and distributed systems, and provides strong support for Kylin system performance assurance under complex business loads.

[0056] The embodiments described above are to be understood as illustrative rather than limiting the scope of the present invention, which is to be determined by the claims. It will be apparent to those skilled in the art that non-essential improvements and adjustments to the present invention, without departing from the spirit and scope of the present invention, still fall within the scope of the present invention.

Claims

1. A method for optimizing Kylin system performance based on machine learning, characterized in that: The optimization method comprises: S1, Data Processing: Collect Kylin system performance data in real time, verify the integrity of the collected performance data via Socket communication, store the verified performance data in a MySQL database, and filter key features from the stored performance data through feature engineering; S2, Scene Recognition: A comprehensive analysis of the performance data after screening out key features is performed using multiple machine learning models. The performance of each machine learning model is determined through model evaluation. The machine learning models with good performance are stacked and fused to generate a scene recognition model for dynamic matching of system scenarios. S3, Bottleneck Detection: We build a range isolation forest model and use it to test Kylin system performance to identify performance anomalies. When Kylin system performance anomalies are detected, we combine causal analysis and FP-growth technology to conduct in-depth analysis of the anomaly to locate the root cause of the Kylin system performance bottleneck. S4, performance optimization: Generate the optimal configuration parameter combination for the Kylin system based on the Bayesian optimization algorithm, dynamically adjust the operating parameters of the Kylin system based on the performance data collected in real time and in combination with the optimal configuration parameter combination; and perform performance verification on the adjusted operating parameters of the Kylin system to determine the optimization effect.

2. The method for optimizing Kylin system performance based on machine learning according to claim 1, wherein: The performance data includes the CPU usage, memory usage, disk I / O rate and network traffic data of the Kylin system.

3. The method for optimizing Kylin system performance based on machine learning according to claim 2, wherein: The data processing in S1 specifically includes the following steps: Data Collection: Kylin system performance data is collected in real time through Psutil and Prometheus. A communication connection is established by binding a specified IP address and port through Socket, and the binary data sent by the client is received and parsed. The integrity of the performance data is verified based on the length field in the data message, and the verified binary data is deserialized into a structured performance data object. Data storage: Classify the performance data objects according to data type and acquisition timestamp, store the classified performance data objects in a thread-safe queue, and perform batch write operations through the transaction management mechanism of the MySQL database to form raw performance data storage to ensure consistency and integrity of the performance data storage process; Data preprocessing: After extracting the raw performance data from the MySQL database, invalid values ​​and abnormal data points are removed according to preset rules. Numerical features of different dimensions are normalized using a standardization method. Performance trend features of time series data are extracted based on a sliding time window algorithm. Categorical features are converted to binary vectors using one-hot encoding. Feature Engineering: Use regular expressions to parse complex data fields, calculate the correlation between features using the Spearman correlation coefficient matrix, eliminate redundant features above the correlation threshold, and calculate the importance score of each feature to the system performance indicator based on the feature importance scoring mechanism of the random forest model, and screen out key features with importance scores higher than the preset threshold.

4. The method for optimizing Kylin system performance based on machine learning according to claim 2, wherein: The scene recognition in S2 specifically includes the following steps: Model training: Use at least five machine learning models, including support vector machine (SVM), CatBoost, XGBoost, LightGBM, logistic regression, neural network, random forest (RF), and decision tree, to train the selected key features. Build a multi-classification model for typical Kylin system operating scenarios. Each model outputs a probability distribution vector representing the Kylin system in different scenarios. Model evaluation: The classification performance of each model was quantified using ROC curves and AUC values. Cross-validation was used to evaluate the stability and generalization ability of each model on different data subsets. Based on the evaluation results, the three models with the top three AUC values ​​and the smallest cross-validation variance were selected for stacking and fusion. The predicted probability vectors of the three selected models were used as input features, and the gated recurrent unit (GRU) was used as the meta-model to generate the final scene recognition model. Dynamic matching: The performance data collected in real time and processed in step S1 is input into the trained scene recognition model to obtain the probability distribution of the current scene, and the difference in the probability distribution of two adjacent predictions is calculated. When the difference value exceeds the preset threshold, the dynamic adjustment process of the Kirin system configuration parameters in step S4 is triggered.

5. The method for optimizing Kylin system performance based on machine learning according to claim 2, wherein: The bottleneck detection in S3 specifically includes the following steps: Anomaly Detection: The Kylin system's operating status is modeled using the Extreme Isolation Forest model. When identifying performance indicators that deviate from the normal range, the system automatically adjusts detection sensitivity to suit different system scenarios. Root cause analysis: Use causal analysis methods to determine the causal relationship between abnormal indicators and system performance, and use the FP-growth algorithm to mine the association rules between abnormal indicators to locate the root cause of performance bottlenecks.

6. The method for optimizing Kylin system performance based on machine learning according to claim 2, wherein: The performance optimization in S4 specifically includes the following steps: Configuration file scanning and parameter extraction: traverse the specified directory to obtain all configuration files, and call the corresponding parser based on the file format characteristics to extract configuration parameters; Parameter screening: Build a random forest evaluation model and use it to evaluate the impact of configuration parameters on Kylin system performance, screening out key configuration parameters with significant impact; Optimization strategy generation: Using the Bayesian optimization framework, combined with the CatBoost gradient boosting tree as a proxy model, and PI (Probability of Improvement) as the objective function, we reduce the search complexity in the high-dimensional parameter space through sequential experimental design and generate the optimal configuration parameter combination; Dynamic optimization and verification: During Kylin system operation, key configuration parameters are dynamically adjusted based on real-time performance data. The optimization effect is verified by comparing the changes in system response time and resource utilization indicators before and after, and the optimization strategy library is continuously updated.

7. The method for optimizing Kylin system performance based on machine learning according to claim 1, wherein: The anomaly detection threshold of the range isolation forest model is dynamically adjusted through the quantile of historical data to adapt to the detection requirements of different system scenarios.

8. The method for optimizing Kylin system performance based on machine learning according to claim 1, wherein: The transaction mechanism of the MySQL database ensures the consistency and integrity of data storage through ACID properties, which include atomicity, consistency, isolation, and persistence.

Images