Algorithm component remote upgrading method

Abstract

The invention belongs to the technical field of algorithm component upgrading, and relates to an algorithm component remote upgrading method, which is characterized in that a data acquisition module established based on an SCADA or EMS system monitors and collects performance data of each algorithm component in real time, and the system can adapt to the change of a power grid environment in real time by setting and dynamically adjusting a performance threshold value. The algorithm performance is ensured to be always in a reasonable range, if the performance of a certain algorithm component is lower than a threshold value, the system automatically triggers a data collection process, and an anomaly detection algorithm is applied to identify data deviation based on collected input data, output results and operation log information, so that a new data set is constructed through comprehensive analysis; the corresponding optimization strategy is generated through the automatic optimization algorithm based on the data set, the algorithm component upgrade package is created through the generated optimization strategy, and remote upgrade operation is performed, so that the problems of insufficient manual intervention, slow response, inaccurate optimization and the like in the prior art are solved, and the intelligent level and the adaptive capacity of the power grid monitoring system are remarkably improved.

Description

Technical Field

[0001] The present application belongs to the technical field of algorithm component upgrade, and more specifically, to a method for remotely upgrading an algorithm component. Background Art

[0002] In modern power systems, especially in the monitoring and management of large-scale power grids, the performance of algorithm components directly affects the stability and operation efficiency of the power grid. Monitoring systems in power systems (such as SCADA or EMS) usually rely on a large number of algorithm modules to perform tasks such as real-time data acquisition, fault diagnosis, load forecasting, and operation optimization. These algorithm modules play a vital role in the security, stability, and response speed of the system.

[0003] As the scale of power grids continues to expand and the complexity increases, traditional power grid monitoring systems rely on fixed algorithms to process tasks. However, the operating environment of the power grid is dynamically changing, including load fluctuations, equipment failures, external meteorological changes, etc. These factors may cause changes to the input data of the algorithm, resulting in the performance of the original algorithm being degraded or no longer adapting to the new power grid operating conditions. For example, the quality of sensor data in the power grid may be affected by noise, failure or damage, the power load forecasting algorithm may fail to respond to drastic load fluctuations in a timely manner, and even the fault diagnosis algorithm may not be able to effectively identify abnormal conditions in the power grid in some cases. These problems will not only affect the operating efficiency of the power grid, but may also cause unnecessary power outages, equipment damage and other safety hazards in the power grid. Summary of the invention

[0004] The present invention provides a remote upgrade method for an algorithm component, which is intended to solve the technical problem that the performance of the algorithm component is degraded or no longer adaptable to the operating conditions of the power grid due to dynamic changes in the power grid.

[0005] A remote upgrade method for an algorithm component comprises the following steps:

[0006] Step 1: Establish a data acquisition module based on the power grid monitoring system SCADA or EMS to obtain the performance data of each algorithm component;

[0007] Step 2: Set the performance threshold of each algorithm component based on historical data and business needs, and dynamically adjust the performance threshold based on changes in the power grid operating environment;

[0008] Step 3: If the performance of an algorithm component is lower than the performance threshold, the data collection process is automatically triggered to collect the input data, output results, operation logs and error logs of the relevant algorithm components;

[0009] Step 4: Based on the collected input data, output results, operation logs, and error logs, anomaly detection algorithms are used to detect data that deviates from expectations, and comprehensive analysis tools are used to conduct comprehensive analysis of the calculation components to extract relevant data;

[0010] Step 5: Based on the analysis results in step 4, an automatic optimization algorithm is used to generate an optimization strategy, including hyperparameter optimization of algorithm components and input parameter optimization;

[0011] Step 6: Generate an algorithm component upgrade package based on the optimization strategy generated by the automatic optimization algorithm, including optimization parameters, algorithm code, and configuration files;

[0012] Step 7: Execute the upgrade of relevant algorithm components based on the generated algorithm component upgrade package. After the upgrade is completed, detect the performance data of the upgraded algorithm components and determine whether the performance has been improved based on the performance threshold. If the performance is reduced, roll back to the previous version and issue an error report for technical personnel to adjust. If the performance is improved, issue an upgrade success report.

[0013] The present invention systematically solves the performance degradation problem that may occur in the power grid algorithm component in the dynamic power grid environment by constructing an automated data collection, performance monitoring, anomaly detection, optimization and upgrade process. First, based on the data acquisition module established by the SCADA or EMS system, the performance data of each algorithm component is monitored and collected in real time, which provides a basis for subsequent performance evaluation and optimization. By setting and dynamically adjusting the performance threshold, the system can adapt to the changes in the power grid environment in real time to ensure that the algorithm performance is always within a reasonable range. Secondly, if the performance of a certain algorithm component is lower than the threshold, the system automatically triggers the data collection process, and based on the collected input data, output results, operation log and other information, the anomaly detection algorithm is applied to identify data deviations, and then a new data set is constructed through comprehensive analysis. Based on the newly constructed data set, the automatic optimization algorithm generates corresponding optimization strategies, including adjustments to the algorithm hyperparameters and input parameters, so as to improve the algorithm performance. Then, the algorithm component upgrade package is created through the generated optimization strategy, and the remote upgrade operation is performed. Finally, the system detects the performance of the upgraded algorithm component, and determines whether it is improved through the performance threshold. If the performance is not improved, it rolls back to the previous version to ensure the safety of the power grid operation. If the performance is improved, an upgrade success report is output. Through full-process automation, dynamic adjustment and intelligent optimization, this solution not only solves the problems of insufficient traditional manual intervention, slow response, and inaccurate optimization, but also significantly improves the intelligence level and adaptive capabilities of the power grid monitoring system, thereby ensuring that the power grid algorithm always maintains efficient and reliable operation in a constantly changing environment.

[0014] As a possible implementation of this embodiment, step 2 includes the following steps:

[0015] Initial performance threshold setting: Use the mean and standard deviation of historical data to set an initial threshold based on the tolerance range:

[0016] T initial =μ+k·σ;

[0017] Where: T initial represents the initial performance threshold; μ represents the mean of historical data; σ represents the standard deviation of historical data; k represents the empirical coefficient, which is between 1.5 and 2, indicating tolerance for a certain degree of fluctuation;

[0018] Dynamic adjustment using the time window method: For each time window, the mean and standard deviation of the performance data within that time period are recalculated, and the threshold is reset:

[0019] T window =μ window +k·σ window ;

[0020] Where: μ window Represents the mean value of the performance data in the current event window; σ window represents the standard deviation of the performance data in the current time window; k represents the empirical coefficient.

[0021] As a possible implementation of this embodiment, environmental factor compensation is introduced into the dynamic adjustment of the performance threshold, as follows:

[0022] T adjusted =T window +α·E factor ;

[0023] Where: T adjusted represents the performance threshold obtained after the final adjustment; E factor Represents the value of external environmental factors; α represents the adjustment weight coefficient of environmental factors, which ranges from 0 to 1;

[0024] in:

[0025] E factor =w1·L fluctuation +w2·C factor +w3·N delay +w4·F rate ;

[0026] Where: w1, w2, w3 and w4 represent the weights of each environmental factor, the sum of which equals 1; F rate Indicates the equipment failure rate; N delay Indicates network delay; C factor Indicates the difference between the current temperature and the historical average temperature; L fluctuationIndicates load fluctuations over a period of time.

[0027] As a possible implementation of this embodiment, step 4 includes the following steps:

[0028] Anomaly detection: The isolation forest algorithm is used to calculate the anomaly score of the data point. If the calculated anomaly score exceeds the preset threshold, the corresponding data point is marked as abnormal;

[0029] Performance bottleneck analysis: The principal component analysis method is used to calculate the covariance matrix of the sample data. By calculating the eigenvalues ​​and eigenvectors of the covariance matrix, the eigenvectors corresponding to the largest first k eigenvalues ​​are selected as the principal components, and the contribution rate of each principal component is calculated:

[0030]

[0031] Where: i represents the eigenvalue of the i-th principal component; m represents the number of principal components;

[0032] Feature importance analysis: The random forest model is used to analyze the impact of input parameters on output results, and multiple decision trees are constructed to predict data and calculate the importance of each feature.

[0033] As a possible implementation of this embodiment, step 5 includes the following steps:

[0034] Data processing: Based on the results of anomaly detection in step 4, abnormal data is proposed, and the features extracted by performance bottleneck analysis and feature importance analysis are combined to form training set data;

[0035] Grid search: By traversing the given parameter space, find the best hyperparameter combination. Let the hyperparameters of the model be θ1, θ2, …, θ k , whose value ranges are And so on;

[0036] The best hyperparameters are selected by training and validating all possible combinations through grid search, where the training is as follows:

[0037] For each set of hyperparameter combinations, the model is trained using the training set data, and the model accuracy is calculated to evaluate the model performance;

[0038] Select the optimal combination: Select the hyperparameter combination that optimizes model performance.

[0039] The beneficial effects of the present invention include:

[0040] The present invention systematically solves the performance degradation problem that may occur in the power grid algorithm component in the dynamic power grid environment by constructing an automated data collection, performance monitoring, anomaly detection, optimization and upgrade process. First, based on the data acquisition module established by the SCADA or EMS system, the performance data of each algorithm component is monitored and collected in real time, which provides a basis for subsequent performance evaluation and optimization. By setting and dynamically adjusting the performance threshold, the system can adapt to the changes in the power grid environment in real time to ensure that the algorithm performance is always within a reasonable range. Secondly, if the performance of a certain algorithm component is lower than the threshold, the system automatically triggers the data collection process, and based on the collected input data, output results, operation log and other information, the anomaly detection algorithm is applied to identify data deviations, and then a new data set is constructed through comprehensive analysis. Based on the newly constructed data set, the automatic optimization algorithm generates corresponding optimization strategies, including adjustments to the algorithm hyperparameters and input parameters, so as to improve the algorithm performance. Then, the algorithm component upgrade package is created through the generated optimization strategy, and the remote upgrade operation is performed. Finally, the system detects the performance of the upgraded algorithm component, and determines whether it is improved through the performance threshold. If the performance is not improved, it rolls back to the previous version to ensure the safety of the power grid operation. If the performance is improved, an upgrade success report is output. Through full-process automation, dynamic adjustment and intelligent optimization, this solution not only solves the problems of insufficient traditional manual intervention, slow response, and inaccurate optimization, but also significantly improves the intelligence level and adaptive capabilities of the power grid monitoring system, thereby ensuring that the power grid algorithm always maintains efficient and reliable operation in a constantly changing environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0042] Figure 1 An overall step block diagram provided for an embodiment of the present invention. DETAILED DESCRIPTION

[0043] In order to make the technical problems, technical solutions and beneficial effects to be solved by the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0044] See also Figure 1 As shown, the best embodiment of the present invention is further described;

[0045] A remote upgrade method for an algorithm component comprises the following steps:

[0046] Step 1: Establish a data acquisition module based on the power grid monitoring system SCADA or EMS to obtain the performance data of each algorithm component;

[0047] Performance data include: accuracy indicators (e.g., prediction error, load error, etc.);

[0048] Response time (e.g., response delay of fault detection algorithms);

[0049] Algorithm stability (e.g., the algorithm’s runtime error rate);

[0050] Computational efficiency (e.g. computation time or number of iterations).

[0051] Step 2: Set the performance threshold of each algorithm component based on historical data and business needs, and dynamically adjust the performance threshold based on changes in the power grid operating environment;

[0052] The step 2 comprises the following steps:

[0053] Initial performance threshold setting: Use the mean and standard deviation of historical data to set an initial threshold based on the tolerance range:

[0054] T initial =μ+k·σ;

[0055] Where: T initial represents the initial performance threshold; μ represents the mean of historical data; σ represents the standard deviation of historical data; k represents the empirical coefficient, which is between 1.5 and 2, indicating tolerance for a certain degree of fluctuation;

[0056] Dynamic adjustment using the time window method: For each time window, the mean and standard deviation of the performance data within that time period are recalculated, and the threshold is reset:

[0057] T window =μ window +k·σ window ;

[0058] Where: μ window Represents the mean value of the performance data in the current event window; σ window represents the standard deviation of the performance data in the current time window; k represents the empirical coefficient.

[0059] As a possible implementation of this embodiment, environmental factor compensation is introduced into the dynamic adjustment of the performance threshold, as follows:

[0060] T adjusted =t window +α·E factor;

[0061] Where: T adjusted represents the performance threshold obtained after the final adjustment; E factor Represents the value of external environmental factors; α represents the adjustment weight coefficient of environmental factors, which ranges from 0 to 1;

[0062] in:

[0063] E factor =w1·L fluctuation +w2·C factor +w3·N delay +w4·F rate ;

[0064] Where: w1, w2, w3 and w4 represent the weights of each environmental factor, the sum of which equals 1; F rate Indicates the equipment failure rate; N delay Indicates network delay; C factor Indicates the difference between the current temperature and the historical average temperature; L fluctuation Indicates load fluctuations over a period of time.

[0065] Step 3: If the performance of an algorithm component is lower than the performance threshold, the data collection process is automatically triggered to collect the input data, output results, operation logs and error logs of the relevant algorithm components;

[0066] Step 4: Based on the collected input data, output results, operation logs, and error logs, anomaly detection algorithms are used to detect data that deviates from expectations, and comprehensive analysis tools are used to conduct comprehensive analysis of the calculation components to extract relevant data;

[0067] As a possible implementation of this embodiment, step 4 includes the following steps:

[0068] Anomaly detection: The isolation forest algorithm is used to calculate the anomaly score of the data point. If the calculated anomaly score exceeds the preset threshold, the corresponding data point is marked as abnormal;

[0069] Performance bottleneck analysis: The principal component analysis method is used to calculate the covariance matrix of the sample data. By calculating the eigenvalues ​​and eigenvectors of the covariance matrix, the eigenvectors corresponding to the largest first k eigenvalues ​​are selected as the principal components, and the contribution rate of each principal component is calculated:

[0070]

[0071] Where: i represents the eigenvalue of the i-th principal component; m represents the number of principal components;

[0072] Feature importance analysis: The random forest model is used to analyze the impact of input parameters on output results, and multiple decision trees are constructed to predict data and calculate the importance of each feature.

[0073] Step 5: Based on the analysis results in step 4, an automatic optimization algorithm is used to generate an optimization strategy, including hyperparameter optimization of algorithm components and input parameter optimization;

[0074] As a possible implementation of this embodiment, step 5 includes the following steps:

[0075] Data processing: Based on the results of anomaly detection in step 4, abnormal data is proposed, and the features extracted by performance bottleneck analysis and feature importance analysis are combined to form training set data;

[0076] Grid search: By traversing the given parameter space, find the best hyperparameter combination. Let the hyperparameters of the model be θ1, θ2, …, θ k , whose value ranges are And so on;

[0077] The best hyperparameters are selected by training and validating all possible combinations through grid search, where the training is as follows:

[0078] For each set of hyperparameter combinations, the model is trained using the training set data, and the model accuracy is calculated to evaluate the model performance;

[0079] Select the optimal combination: Select the hyperparameter combination that optimizes model performance.

[0080] When performing the grid search, unnecessary calculations are reduced by "adaptively reducing the grid". The specific steps are as follows:

[0081] Initial coarse grid search: First, a coarse grid search is performed. The hyperparameter intervals of the coarse grid are larger. Through the coarse grid search, we can roughly find a range of hyperparameters. For example, if we set the grid range to:

[0082] η∈{0.01,0.1,1.0},λ∈{0.1,1.0,10.0};

[0083] Through preliminary training and verification, the hyperparameter region with the best performance is selected.

[0084] Refined grid search: For the region where the hyperparameter combinations perform well, further refine the grid search. For example, if the coarse grid search shows that the learning rate η is most effective between [0.01, 0.1], it can be further refined to:

[0085] η∈{0.02,0.04,0.06,0.08};

[0086] Dynamically adjust the grid: Dynamically adjust the search range and step size of hyperparameters based on training and validation results. For example, during the iteration process, if it is found that the change of a certain hyperparameter has little effect on model performance, the step size of the hyperparameter can be reduced, or a more refined search can be performed within a certain range;

[0087] The dynamic grid adjustment includes dynamically adjusting the grid step size and dynamically adjusting the hyperparameter search range, as follows:

[0088] Dynamically adjust the grid step size:

[0089]

[0090] Where: represents the hyperparameter step size after the t+1th iteration; represents the hyperparameter step size after the tth iteration; Pref(t) represents the model performance after the tth iteration; α represents the learning rate control factor; β represents the influence coefficient of performance change;

[0091] The hyperparameter step size is dynamically adjusted by calculating the relative performance change rate. If the performance improvement is small in consecutive iterations, the step size will be reduced, reducing the search for the hyperparameter range and focusing on a finer hyperparameter range; conversely, if the performance improvement is significant, the step size will be maintained or increased to expand the search range and explore a wider range of hyperparameter space.

[0092] Dynamically adjust the hyperparameter search range:

[0093]

[0094] Where: represents the hyperparameter search range at the tth iteration; γ represents the sensitivity of the control range adjustment; δ represents the coefficient of the performance change on the search range adjustment. If the model performs well in the current hyperparameter region, the search range will be moderately expanded to further explore more hyperparameter combinations; if the performance improvement is small, the search range will be narrowed to focus on a smaller and more refined parameter range.

[0095] By utilizing the information about the change in model performance after each iteration, the hyperparameter step size and search range are dynamically adjusted. This dynamic adjustment based on performance feedback can more accurately focus on the most promising hyperparameter area, thereby reducing unnecessary calculations; by introducing hyperparameter control factors, the strategy in the search process can be flexibly adjusted, making the algorithm more adaptable to different tasks or data sets. For example, in the early stages of hyperparameter search, a larger step size and a wider search range can speed up initial screening; when performance is relatively stable, refining the step size and range can improve search accuracy.

[0096] Step 6: Generate an algorithm component upgrade package based on the optimization strategy generated by the automatic optimization algorithm, including optimization parameters, algorithm code, and configuration files;

[0097] The optimized algorithm code may need to be updated according to the adjusted hyperparameters and changes in input data. If it is a machine learning model, it may be necessary to modify the model's training process, loss function settings, optimization algorithm, etc. according to the hyperparameter optimization results.

[0098] Algorithm code update: Based on the results of automatic optimization, the automated script will modify the code of the algorithm component. For example, if the learning rate is adjusted from 0.01 to 0.001 during the optimization process, the part of the code related to the learning rate will be automatically modified.

[0099] Version control: Each time the algorithm code is updated, a new version is generated. A version control system (such as Git) is used to manage the version of the algorithm code to ensure that it can be traced back to the previous version.

[0100] The optimized input data processing strategy includes how to standardize the input data, perform feature selection, etc. For example, if the automatic optimization algorithm finds that some features have little impact on the model, feature selection code can be added to the generated upgrade package to remove unimportant features.

[0101] Data preprocessing update: For each input data set, the automatic optimization algorithm will generate an optimal data preprocessing process. This process may include feature selection, feature engineering, data standardization and other steps.

[0102] Adjustments to standardization / normalization methods: If the automatic optimization results indicate that certain features require different standardization or normalization, the upgrade package will include updated data preprocessing code.

[0103] The configuration file is one of the core parts of the algorithm component. It records the environment configuration, hyperparameters, input and output formats, etc. of the algorithm. Generate the corresponding configuration file based on the output of the automatic optimization algorithm in step 5.

[0104] Update the contents of the configuration file: The configuration file will contain new hyperparameter values, data processing methods, operating environment configuration, etc. For example, a hyperparameter (such as learning rate) will be updated to α opt , the way the input data is processed (such as whether to standardize it) will also be updated according to the optimization strategy.

[0105] Configuration file format: The configuration file can be in JSON, YAML, or XML format to facilitate automatic reading and updating.

[0106] The generated upgrade package is a complete set of files, including the following:

[0107] Optimized algorithm code: including all updated model training, loss function, and optimization algorithm code.

[0108] Optimized input data processing scripts: including updated feature selection, standardization / normalization, feature engineering and other codes.

[0109] Configuration file: including hyperparameters, data processing methods and other configurations.

[0110] Version information: Record the version of the current upgrade package to ensure that rollback operations can be performed.

[0111] The upgrade package can be packaged into a compressed file (such as ZIP format) for easy distribution and deployment.

[0112] Release and version management of upgrade packages:

[0113] Each time you generate an upgrade package, you need to generate a unique version number for tracking and management. The version number can be generated based on a timestamp or iteration number.

[0114] Step 7: Execute the upgrade of relevant algorithm components based on the generated algorithm component upgrade package. After the upgrade is completed, detect the performance data of the upgraded algorithm components and determine whether the performance has been improved based on the performance threshold. If the performance is reduced, roll back to the previous version and issue an error report for technical personnel to adjust. If the performance is improved, issue an upgrade success report.

[0115] The present invention systematically solves the performance degradation problem that may occur in the power grid algorithm component in the dynamic power grid environment by constructing an automated data collection, performance monitoring, anomaly detection, optimization and upgrade process. First, based on the data acquisition module established by the SCADA or EMS system, the performance data of each algorithm component is monitored and collected in real time, which provides a basis for subsequent performance evaluation and optimization. By setting and dynamically adjusting the performance threshold, the system can adapt to the changes in the power grid environment in real time to ensure that the algorithm performance is always within a reasonable range. Secondly, if the performance of a certain algorithm component is lower than the threshold, the system automatically triggers the data collection process, and based on the collected input data, output results, operation log and other information, the anomaly detection algorithm is applied to identify data deviations, and then a new data set is constructed through comprehensive analysis. Based on the newly constructed data set, the automatic optimization algorithm generates corresponding optimization strategies, including adjustments to the algorithm hyperparameters and input parameters, so as to improve the algorithm performance. Then, the algorithm component upgrade package is created through the generated optimization strategy, and the remote upgrade operation is performed. Finally, the system detects the performance of the upgraded algorithm component, and determines whether it is improved through the performance threshold. If the performance is not improved, it rolls back to the previous version to ensure the safety of the power grid operation. If the performance is improved, an upgrade success report is output. Through full-process automation, dynamic adjustment and intelligent optimization, this solution not only solves the problems of insufficient traditional manual intervention, slow response, and inaccurate optimization, but also significantly improves the intelligence level and adaptive capabilities of the power grid monitoring system, thereby ensuring that the power grid algorithm always maintains efficient and reliable operation in a constantly changing environment.

[0116] The above are only preferred embodiments of the present application and are not intended to limit the present application. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present application should be included in the protection scope of the present application.

Claims

1. A remote upgrade method for algorithm components, characterized in that: The following steps are involved: Step 1: Establish a data acquisition module based on the power grid monitoring system SCADA or EMS to obtain the performance data of each algorithm component; Step 2: Set the performance threshold of each algorithm component based on historical data and business needs, and dynamically adjust the performance threshold based on changes in the power grid operating environment; Step 3: If the performance of an algorithm component is lower than the performance threshold, the data collection process is automatically triggered to collect the input data, output results, operation logs and error logs of the relevant algorithm components; Step 4: Based on the collected input data, output results, operation logs, and error logs, anomaly detection algorithms are used to detect data that deviates from expectations, and comprehensive analysis tools are used to conduct comprehensive analysis of the calculation components to extract relevant data; Step 5: Based on the analysis results in step 4, an automatic optimization algorithm is used to generate an optimization strategy, including hyperparameter optimization of algorithm components and input parameter optimization; Step 6: Generate an algorithm component upgrade package based on the optimization strategy generated by the automatic optimization algorithm, including optimization parameters, algorithm code, and configuration files; Step 7: Execute the upgrade of relevant algorithm components based on the generated algorithm component upgrade package. After the upgrade is completed, detect the performance data of the upgraded algorithm components and determine whether the performance has been improved based on the performance threshold. If the performance is reduced, roll back to the previous version and issue an error report for technical personnel to adjust. If the performance is improved, issue an upgrade success report.

2. The algorithm component remote upgrade method according to claim 1, characterized in that: The step 2 comprises the following steps: Initial performance threshold setting: Use the mean and standard deviation of historical data to set an initial threshold based on the tolerance range: T initial =μ+k·μ; Where: T initial represents the initial performance threshold; μ represents the mean of historical data; σ represents the standard deviation of historical data; k represents the empirical coefficient, which is between 1.5 and 2, indicating tolerance for a certain degree of fluctuation; Dynamic adjustment using the time window method: For each time window, the mean and standard deviation of the performance data within that time period are recalculated, and the threshold is reset: T window =μ window +k·s window ; Where: μ window Represents the mean value of the performance data in the current event window; σ window represents the standard deviation of the performance data in the current time window; k represents the empirical coefficient.

3. The algorithm component remote upgrade method according to claim 2, characterized in that: Environmental factor compensation is introduced into the dynamic adjustment of the performance threshold, as follows: T adjusted =T window +a·E factor ; Where: T adjusted represents the performance threshold obtained after the final adjustment; E factor Represents the value of external environmental factors; α represents the adjustment weight coefficient of environmental factors, which ranges from 0 to 1; in: E factor =w1·L fluctuation +w2·C factor +w3·N delay +w4·F rate ; Where: w1, w2, w3 and w4 represent the weights of each environmental factor, the sum of which equals 1; F rate Indicates the equipment failure rate; N delay Indicates network delay; C factor Indicates the difference between the current temperature and the historical average temperature; L fluctuation Indicates load fluctuations over a period of time.

4. The method for remotely upgrading an algorithm component according to claim 1, characterized in that: The step 4 comprises the following steps: Anomaly detection: The isolation forest algorithm is used to calculate the anomaly score of the data point. If the calculated anomaly score exceeds the preset threshold, the corresponding data point is marked as abnormal; Performance bottleneck analysis: The principal component analysis method is used to calculate the covariance matrix of the sample data. By calculating the eigenvalues ​​and eigenvectors of the covariance matrix, the eigenvectors corresponding to the largest first k eigenvalues ​​are selected as the principal components, and the contribution rate of each principal component is calculated: Where: i represents the eigenvalue of the i-th principal component; m represents the number of principal components; Feature importance analysis: The random forest model is used to analyze the impact of input parameters on output results, and multiple decision trees are constructed to predict data and calculate the importance of each feature.

5. The algorithm component remote upgrade method according to claim 1, characterized in that: The step 5 comprises the following steps: Data processing: Based on the results of anomaly detection in step 4, abnormal data is proposed, and the features extracted by performance bottleneck analysis and feature importance analysis are combined to form training set data; Grid search: By traversing the given parameter space, find the best hyperparameter combination. Let the hyperparameters of the model be θ1, θ2, …, θ k , whose value ranges are And so on; The best hyperparameters are selected by training and validating all possible combinations through grid search, where the training is as follows: For each set of hyperparameter combinations, the model is trained using the training set data, and the model accuracy is calculated to evaluate the model performance; Select the optimal combination: Select the hyperparameter combination that optimizes model performance.

Images