Resource prediction and optimization method and device based on large model, equipment and medium
Through the large model, the limitations of traditional resource prediction are solved, adaptive resource management is realized, operation and maintenance costs are reduced and prediction accuracy is improved.
Patent Information
- Application Number
- CN202510665279.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-05-22
- Publication Date
- 2025-08-22
AI Technical Summary
Traditional resource prediction relies on historical data and fixed strategies, and is difficult to deal with burst traffic, resulting in waste of costs and response delays, and does not consider the impact of external factors.
The large model is used to automatically process resource prediction and optimization, and by analyzing resource prediction requests, collecting multi-modal data, generating resource prediction solutions, and converting them into configuration parameters for local parameter changes, and monitoring system indicators in real time to confirm the optimization effect.
Reduce intervention of operation and maintenance personnel, reduce operation and maintenance costs, improve response speed, generate adaptive and flexible optimization strategies, and improve prediction and optimization accuracy.
Smart Images
Figure CN120523601A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of computer technology, and in particular to a resource prediction and optimization method, device, equipment and medium based on a large model. Background Art
[0002] With the rapid development of the internet, resource forecasting has become increasingly important. Traditional resource forecasting relies primarily on statistical analysis of historical data and empirical rules, and resource allocation is often based on fixed policies or manual intervention. This often has practical limitations, such as in scenarios like e-commerce promotions and flash sales. Traditional models struggle to accurately predict peak traffic and cope with sudden bursts. They also tend to overlook external factors, such as the impact of holidays, marketing events, and other external variables on resource demand. Limited by fixed policies and manual intervention, redundant resources are reserved to cope with peak traffic, resulting in costly problems or manual capacity expansion after outages. Response delays can lead to service degradation and other issues. Summary of the Invention
[0003] In view of this, the purpose of the present invention is to provide a resource prediction and optimization method, device, equipment, and medium based on a large model, which can automatically handle resource prediction and optimization problems using a large model, reduce the intervention of operation and maintenance personnel, and lower operation and maintenance costs. The specific solution is as follows:
[0004] In a first aspect, the present application discloses a resource prediction and optimization method based on a large model, which is applied to a software system, comprising:
[0005] If a resource prediction request is received, the resource prediction request is parsed to determine a target resource prediction type;
[0006] Collecting multimodal data local to the software system and inputting the multimodal data into a target resource prediction model, so that the target resource prediction model performs resource analysis based on the target resource prediction type and the multimodal data, and generates a corresponding resource prediction solution based on the multimodal data;
[0007] generating a resource optimization strategy based on the resource prediction scheme, and converting the resource optimization strategy into configuration parameters in a preset target format, so as to perform local parameter changes based on the configuration parameters;
[0008] Monitor the real-time system indicators of the software system after the parameters are changed, and confirm whether resource optimization is completed or resource alarm is issued based on the real-time system indicators.
[0009] Optionally, if a resource prediction request is received, before parsing the resource prediction request to determine the target resource prediction type, the method further includes:
[0010] Perform model fusion on the preset text understanding model, the preset time series prediction model, and the preset multimodal model to obtain a resource prediction model to be trained;
[0011] Generate a training data set based on the collected local event logs, local configuration files, local operation and maintenance manuals, and historical fault handling records, and train the resource prediction model to be trained using the training data set to obtain a resource prediction model to be determined;
[0012] A policy generation prompt template corresponding to the software system is generated, and the policy generation prompt template and the to-be-determined resource prediction model are fused to obtain a target resource prediction model.
[0013] Optionally, if a resource prediction request is received, parsing the resource prediction request to determine a target resource prediction type includes:
[0014] If a resource prediction request is received, parsing the resource prediction request to determine whether the target resource prediction type of the resource prediction request is a short-term resource prediction request or a long-term resource prediction request;
[0015] A resource prediction task corresponding to the resource prediction request is generated according to the target resource prediction type.
[0016] Optionally, the collecting of the local multimodal data of the software system includes:
[0017] Collect local time series indicator data, log text data, configuration data, and operation and maintenance manuals of the software system;
[0018] Eliminate data from the time series indicator data based on a preset time series indicator range, and complete missing values in the time series indicator data to obtain target time series indicator data;
[0019] Parsing the log text data to obtain target log text data corresponding to a preset log requirement data format in the log text data;
[0020] The target time series indicator data, the target log text data, the configuration data and the operation and maintenance manual are used as target multimodal data.
[0021] Optionally, inputting the multimodal data into a target resource prediction model so that the target resource prediction model performs resource analysis based on the target resource prediction type and the multimodal data, and generates a corresponding resource prediction solution based on the multimodal data, includes:
[0022] Inputting the multimodal data into a target resource prediction model so that the target resource prediction model parses the multimodal data based on the target resource prediction type to determine the current system indicator health of the software system through the multimodal data;
[0023] A system indicator change trend corresponding to the system indicator is generated based on the system indicator health.
[0024] Optionally, generating a resource optimization strategy based on the resource prediction solution, and converting the resource optimization strategy into configuration parameters in a preset target format, so as to perform local parameter changes based on the configuration parameters, includes:
[0025] Generate a horizontal expansion and contraction adjustment strategy and a vertical expansion and contraction adjustment strategy based on the resource prediction scheme, and generate a resource dynamic scheduling strategy and a fault recovery strategy according to the resource prediction scheme;
[0026] Determine the parameter change plans corresponding to the horizontal scaling adjustment strategy, the vertical scaling adjustment strategy, the resource dynamic scheduling strategy, and the fault recovery strategy, respectively, and generate configuration parameters in a preset target format based on the parameter change plan to perform local parameter changes based on the configuration parameters.
[0027] Optionally, monitoring the real-time system indicators of the software system after the parameter change, and confirming whether resource optimization is completed or issuing a resource alarm based on the real-time system indicators, includes:
[0028] Monitoring the real-time system indicators of the software system after the parameters are changed, and determining whether the real-time system indicators meet the preset indicator operation threshold conditions;
[0029] If the real-time system indicator meets the preset indicator operation threshold condition, then end this resource optimization;
[0030] If the real-time system indicator does not meet the preset indicator operation threshold condition, the configuration is rolled back and a resource alarm operation is performed.
[0031] In a second aspect, the present application discloses a resource prediction and optimization device based on a large model, which is applied to a software system, comprising:
[0032] a prediction type determination module, configured to, upon receiving a resource prediction request, parse the resource prediction request to determine a target resource prediction type;
[0033] a prediction scheme generation module, configured to collect multimodal data local to the software system and input the multimodal data into a target resource prediction model, so that the target resource prediction model performs resource analysis based on the target resource prediction type and the multimodal data, and generates a corresponding resource prediction scheme based on the multimodal data;
[0034] a parameter change module, configured to generate a resource optimization strategy based on the resource prediction scheme, and convert the resource optimization strategy into configuration parameters in a preset target format, so as to perform local parameter changes based on the configuration parameters;
[0035] The optimization judgment module is used to monitor the real-time system indicators of the software system after the parameters are changed, and to confirm whether resource optimization is completed or to issue a resource alarm based on the real-time system indicators.
[0036] In a third aspect, the present application discloses an electronic device, comprising:
[0037] Memory, used to store computer programs;
[0038] The processor is used to execute the computer program to implement the resource prediction and optimization method based on the large model as described above.
[0039] In a fourth aspect, the present application discloses a computer-readable storage medium for storing a computer program, wherein the computer program, when executed by a processor, implements the aforementioned large model-based resource prediction and optimization method.
[0040] In the present application, if a resource prediction request is received, the resource prediction request is parsed to determine the target resource prediction type; the local multimodal data of the software system is collected, and the multimodal data is input into the target resource prediction model, so that the target resource prediction model performs resource analysis based on the target resource prediction type and the multimodal data, and generates a corresponding resource prediction plan based on the multimodal data; a resource optimization strategy is generated based on the resource prediction plan, and the resource optimization strategy is converted into configuration parameters in a preset target format, so as to perform local parameter changes based on the configuration parameters; the real-time system indicators of the software system after the parameter change are monitored, and based on the real-time system indicators, it is confirmed whether the resource optimization is completed or a resource alarm is issued.
[0041] As can be seen, the method of the present application can parse received resource prediction requests to determine the target resource prediction type required by the request. Then, multimodal data from the system's local environment can be collected so that the target resource prediction model can perform resource analysis based on the target resource prediction type and multimodal data, thereby generating a resource prediction solution for the software system. Based on the resource prediction solution, a resource optimization strategy can then be generated to optimize the local parameters of the software system. Subsequently, the resource optimization strategy needs to be converted into system-readable configuration parameters, and the parameters are configured to perform local parameter changes. Finally, system indicators after the parameter changes need to be monitored in real time to determine whether resource optimization has been completed. This allows, on the one hand, a large model to automatically maintain and handle most resource prediction and optimization issues, reducing the intervention of operation and maintenance personnel, lowering operation and maintenance costs, and improving the speed of operation and maintenance response. On the other hand, compared to traditional fixed-rule-based methods that require manual adjustment, the method of the present application can adaptively generate strategies, support natural language modification, and make strategies more flexible and convenient. Furthermore, compared to judging resource status based on a single indicator, the method of the present application supports multimodal fusion, thereby improving prediction and optimization accuracy. BRIEF DESCRIPTION OF THE DRAWINGS
[0042] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.
[0043] Figure 1 This is a flow chart of a resource prediction and optimization method based on a large model disclosed in this application;
[0044] Figure 2 This is a flow chart of a specific resource prediction and optimization method based on a large model disclosed in this application;
[0045] Figure 3 This is a schematic diagram of the structure of a resource prediction and optimization device based on a large model disclosed in this application;
[0046] Figure 4 This is a structural diagram of an electronic device disclosed in this application. DETAILED DESCRIPTION
[0047] 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.
[0048] In existing technologies, traditional resource forecasting mainly relies on statistical analysis of historical data and empirical rules, and resource allocation is usually based on fixed strategies or manual intervention, which often has limitations in actual operations.
[0049] In order to overcome the above technical problems, the present application discloses a resource prediction and optimization method, device, equipment and medium based on a large model, which can automatically handle resource prediction and optimization problems using a large model, reduce the intervention of operation and maintenance personnel, and reduce operation and maintenance costs.
[0050] See also Figure 1 As shown, an embodiment of the present invention discloses a resource prediction and optimization method based on a large model, which is applied to a software system, including:
[0051] Step S11: If a resource prediction request is received, the resource prediction request is parsed to determine a target resource prediction type.
[0052] In this embodiment, after receiving a resource prediction request, the request must be parsed to determine the resource prediction type. The specific process is as follows: If a resource prediction request is received, the request is parsed to determine whether the target resource prediction type is a short-term or long-term resource prediction request; and a resource prediction task corresponding to the resource prediction request is generated based on the target resource prediction type. Specifically, after receiving a resource prediction request, the request must be parsed to determine whether it contains an explicit field indicating whether the request is a long-term or short-term request. It should be noted that, in this application, a short-term request is a request for predicting resource conditions within the next hour, and a long-term request is a request for predicting resource conditions within the next 24 hours. If the request contains the explicit field, the request can be directly determined to be a long-term or short-term request based on the explicit field. Furthermore, if the request does not contain the explicit field, the time parameter contained in the request must be determined, and the request type, whether long-term or short-term, can be determined based on the time parameter. Finally, after the request type is determined, a corresponding long-term or short-term resource prediction task can be generated based on the request type. In this way, corresponding resource forecasting tasks can be generated according to demand, thereby ensuring the accuracy of resource forecasting.
[0053] Step S12: Collect the local multimodal data of the software system and input the multimodal data into the target resource prediction model, so that the target resource prediction model performs resource analysis based on the target resource prediction type and the multimodal data, and generates a corresponding resource prediction plan based on the multimodal data.
[0054] In this embodiment, in order to predict resources, it is first necessary to collect multimodal data of the system. Specifically, it is necessary to collect the local time series indicator data, log text data, configuration data and operation and maintenance manual of the software system. The time series indicator data corresponds to the task type. If the task type is a short-term resource prediction task, the time series indicator data of the past hour is collected. If the task type is a long-term resource prediction task, the time series indicator data of the past 24 hours is collected. The time series indicator data mainly includes the processor, memory, network, storage utilization rate of the Pod / Node, the number of Deployment copies, and the HPA (Horizontal Pod Autoscaler, automatic expansion and contraction mechanism) status; the data content of the log text data mainly includes the container standard output / error log, K8s events and application custom logs; the configuration data mainly includes the Deployment / Service YAML configuration and the service dependency diagram.
[0055] Furthermore, the collected sequence indicator data needs to be preprocessed. Specifically, data must be filtered based on a preset time series indicator range to remove data outside the preset range. For example, values outside the abnormal range (e.g., CPU (Central Processing Unit) utilization exceeding 100%) must be removed. Missing values in the time series indicator data must also be completed to obtain the target time series indicator data. Furthermore, the log text data must be parsed to obtain target log text data that corresponds to the preset log data format. For example, structured parsing of the log text can be performed to obtain key information such as timestamps, log levels, service names, and exception occurrence points. Critical security information must also be desensitized. Ultimately, the processed target time series indicator data, target log text data, along with the collected configuration data and operation and maintenance manuals, are used as the target multimodal data.
[0056] After obtaining the target multimodal data, the multimodal data needs to be input into the target resource prediction model so that the target resource prediction model can parse the multimodal data based on the target resource prediction type to determine the current system indicator health of the software system through the multimodal data. It should be noted that the target resource prediction model can analyze whether the current system indicators are healthy based on the multimodal data. For example, if a certain indicator is too high, it indicates that the current system indicator is unhealthy. The current system indicator health can be generated by combining various indicators, and the system indicator change trend corresponding to the system indicator can be generated based on the system indicator health, such as the number of Pod replicas, CPU / memory requirements, etc. in the next hour. In this way, by automatically maintaining and processing resource prediction and optimization issues through large models, the intervention of operation and maintenance personnel can be reduced, the cost of operation and maintenance personnel can be reduced, and the speed of operation and maintenance response can be improved.
[0057] Step S13: Generate a resource optimization strategy based on the resource prediction plan, and convert the resource optimization strategy into configuration parameters in a preset target format, so as to perform local parameter changes based on the configuration parameters.
[0058] In this embodiment, resource optimization strategies need to be generated based on the resource forecasting scheme. For example, horizontal and vertical scaling strategies are generated based on the number of Pod replicas, CPU / memory requirements, and other factors for the next hour. Horizontal scaling strategies adjust the number of Pod replicas to accommodate traffic fluctuations, while vertical scaling strategies optimize Pod resource requests (requests / limits) to avoid resource waste. Dynamic resource scheduling strategies and fault recovery strategies can also be generated based on the resource forecasting scheme. Dynamic resource scheduling strategies dynamically adjust resource allocation based on service importance (such as core payment services), while fault recovery strategies automatically restart abnormal Pods to trigger root cause analysis.
[0059] Furthermore, parameter change plans corresponding to horizontal scaling, vertical scaling, dynamic resource scheduling, and fault recovery policies are required. Based on these parameter change plans, configuration parameters in a preset target format are generated to enable local parameter changes based on these configuration parameters. Specifically, policies need to be dynamically converted into Kubernetes operational instructions and executed securely and efficiently, achieving a seamless transition from policy generation to implementation. For example, a model-generated policy can be converted into YAML-formatted configuration parameters based on a preset template. The conversion is as follows: Policy description: "Recommend scaling the payment service to 5 replicas, based on a CPU utilization of 70%." YAML output: "Generate HPA configuration, set minReplicas: 3, maxReplicas: 10, target CPU utilization of 70%." An automated execution mechanism then calls the Kubernetes Application Programming Interface (API) to execute these parameter changes. This allows the model-generated policy to be converted into system-readable configuration parameters, which can then be adjusted to optimize resources.
[0060] Step S14: monitor the real-time system indicators of the software system after the parameters are changed, and confirm whether resource optimization is completed or resource alarm is issued based on the real-time system indicators.
[0061] In this embodiment, after completing resource optimization, it is necessary to monitor the real-time system indicators after the parameter changes to determine whether the resource optimization is complete. Specifically, it is necessary to monitor the real-time system indicators after the software system parameters are changed and determine whether the real-time system indicators meet the preset indicator operation threshold conditions. For example, the strategy is judged to be in line with expectations based on multiple indicators such as resource utilization, service response time, and Pod status, such as whether the processor utilization is within the preset normal range. If the real-time system indicators meet the preset indicator operation threshold conditions, the resource optimization is terminated, indicating that the resource optimization is successful and the system is running stably. If the real-time system indicators do not meet the preset indicator operation threshold conditions, the configuration is rolled back and a resource alarm operation is performed. For example, if the processor utilization is greater than 95%, which is significantly higher than the normal value, it indicates that there may be problems with the optimization and the configuration needs to be rolled back. Then, an alarm is issued to notify the operation and maintenance personnel to make adjustments. In this way, it can be ensured that resource optimization will not have a negative impact on the operation of the system, and if the optimization fails, the operation and maintenance personnel are notified immediately to perform maintenance to avoid damage to the system, thereby ensuring the stability of resource optimization.
[0062] In this embodiment, received resource prediction requests can be parsed to determine the target resource prediction type required by the request. Local multimodal data can then be collected from the system, allowing the target resource prediction model to perform resource analysis based on the target resource prediction type and multimodal data, thereby generating a resource prediction solution for the software system. Based on the resource prediction solution, a resource optimization strategy can then be generated to optimize the local parameters of the software system. The resource optimization strategy then needs to be converted into system-readable configuration parameters, and the parameters can be configured to perform local parameter changes. Finally, system metrics after the parameter changes need to be monitored in real time to determine whether resource optimization has been completed. This allows, on the one hand, a large model to automatically maintain and handle most resource prediction and optimization issues, reducing the intervention of operations and maintenance personnel, lowering operation and maintenance costs, and improving the speed of operation and maintenance response. On the other hand, compared to traditional fixed-rule-based methods that require manual adjustment, the method of this application can adaptively generate strategies, support natural language modification, and provide greater flexibility and convenience. Furthermore, compared to determining resource status based on a single metric, the method of this application supports multimodal fusion, thereby improving prediction and optimization accuracy.
[0063] Based on the above embodiments, it can be seen that the method of this application requires the generation of an optimization strategy through a target resource prediction model. Therefore, this embodiment provides a detailed description of how to generate a target resource prediction model. Figure 2 As shown, an embodiment of the present invention discloses a resource prediction and optimization method based on a large model, which is applied to a software system, including:
[0064] Step S21: performing model fusion on the preset text understanding model, the preset time series prediction model, and the preset multimodal model to obtain a resource prediction model to be trained.
[0065] In this embodiment, it is first necessary to perform model fusion, and there are three models to be fused, namely the text understanding model, the time series prediction model and the multimodal model. It should be noted that the text understanding model has a strong semantic understanding ability, but requires additional adaptation to time series data. The time series prediction model has a strong processing ability for time series tasks, but lacks the ability to interact with text. The multimodal model has knowledge cross-modal reasoning, but consumes a lot of computing resources. It should be further explained that the role of the text understanding model is to parse log text and operation and maintenance documents, and then generate natural language strategies through the analysis of the above two types of data. The time series prediction model can predict the trends of indicators such as processors and memory, and the multimodal model can integrate multimodal analysis of indicators, logs, and configurations. It can be seen that by fusing the above three models, it is possible to achieve multi-dimensional prediction of software system resources and generate relevant optimization strategies.
[0066] Step S22: Generate a training data set based on the collected local event logs, local configuration files, local operation and maintenance manuals, and historical fault handling records, and train the resource prediction model to be trained using the training data set to obtain the resource prediction model to be determined.
[0067] In this embodiment, after the model fusion is completed and the resource prediction model to be trained is obtained, the model needs to be trained. Therefore, training data needs to be collected first. It should be noted that the training data mainly consists of local event logs, local configuration files, local operation and maintenance manuals, and historical fault handling records. Among them, local event logs are collected in a real-time streaming manner, and after collection, they are saved to the Elasticsearch database and indexed by log level. Hot data is retained for 7 days, and cold data is archived to object storage; local configuration files are collected by a method of triggering collection when configuration changes are made, and stored in the Neo4j database to store current service dependencies and historical version snapshots; local operation and maintenance manuals are collected by regular batch updates; historical fault handling records need to be searched in the database.
[0068] After data collection is complete, a dataset is generated based on the collected data to train the resource prediction model to be trained and fine-tune the model accordingly. Model fine-tuning can be performed using a multi-task joint training approach for large-scale model fine-tuning. Task types primarily include resource demand prediction tasks, such as regression tasks that predict processor utilization; policy generation tasks, such as text generation tasks that output HPA configurations; and root cause analysis tasks, such as classification tasks that determine the cause of anomalies. Through model training and fine-tuning, the resource prediction model to be determined is obtained. By collecting specific data and fine-tuning the model through relevant tasks, the model's capabilities in resource prediction and optimization can be effectively improved, thereby enhancing the accuracy of the model's resource prediction and optimization capabilities.
[0069] Step S23: Generate a policy generation prompt template corresponding to the software system, and fuse the policy generation prompt template and the to-be-determined resource prediction model to obtain a target resource prediction model.
[0070] In this embodiment, it is necessary to generate a policy generation prompt template and integrate the template into the model to improve the model's ability to generate policies. Specifically, K8s API documents and Best Practices can be converted into prompt templates (PromptTemplates) to guide the model to generate compliance policies. The prompt template is as follows: "You are a K8s operation and maintenance expert. Please generate scaling policies based on the following indicators and logs. Requirements: Minimum number of replicas ≥ 2, prioritize core services, and comply with SLA (Service Level Agreement) agreements. Input data: {"metrics": [CPU 80%, 85%, 90%], "logs": "Pod evicted due to OOM"}", and then a K8s operation and maintenance knowledge graph (such as service dependencies, fault handling processes) can be constructed as the context for model reasoning. In this way, integrating the template into the model can effectively improve the readability of the model generation policy and thus improve the prediction accuracy.
[0071] In this embodiment, a preset text understanding model, a preset time series prediction model, and a preset multimodal model are first fused to obtain a resource prediction model to be trained. A training dataset is then generated based on collected local event logs, local configuration files, local operation and maintenance manuals, and historical fault handling records. The resource prediction model to be trained is then trained using the training dataset to obtain a resource prediction model to be determined. Finally, a policy generation prompt template corresponding to the software system is generated, and the policy generation prompt template and the resource prediction model to be determined are fused to obtain a target resource prediction model. In this way, on the one hand, by fusing the above three models, multi-dimensional predictions of software system resources can be achieved and relevant optimization strategies can be generated. On the other hand, by collecting specific data and fine-tuning the model through relevant tasks, the model's capabilities in resource prediction and optimization can be effectively improved, thereby improving the model's accuracy in resource prediction and optimization.
[0072] See also Figure 3 As shown, an embodiment of the present invention discloses a resource prediction and optimization device based on a large model, which is applied to a software system, including:
[0073] The prediction type determination module 11 is configured to parse a resource prediction request upon receipt to determine a target resource prediction type;
[0074] A prediction solution generation module 12 is configured to collect multimodal data local to the software system and input the multimodal data into a target resource prediction model so that the target resource prediction model performs resource analysis based on the target resource prediction type and the multimodal data, and generates a corresponding resource prediction solution based on the multimodal data;
[0075] A parameter changing module 13 is configured to generate a resource optimization strategy based on the resource prediction scheme, and convert the resource optimization strategy into configuration parameters in a preset target format, so as to perform local parameter changes based on the configuration parameters;
[0076] The optimization judgment module 14 is used to monitor the real-time system indicators of the software system after the parameters are changed, and to confirm whether resource optimization is completed or to issue a resource alarm based on the real-time system indicators.
[0077] In this embodiment, received resource prediction requests can be parsed to determine the target resource prediction type required by the request. Local multimodal data can then be collected from the system, allowing the target resource prediction model to perform resource analysis based on the target resource prediction type and multimodal data, thereby generating a resource prediction solution for the software system. Based on the resource prediction solution, a resource optimization strategy can then be generated to optimize the local parameters of the software system. The resource optimization strategy then needs to be converted into system-readable configuration parameters, and the parameters can be configured to perform local parameter changes. Finally, system indicators after the parameter changes need to be monitored in real time to determine whether resource optimization has been completed. This allows, on the one hand, a large model to automatically maintain and handle most resource prediction and optimization issues, reducing the intervention of operations and maintenance personnel, lowering operation and maintenance costs, and improving the speed of operation and maintenance response. On the other hand, compared to traditional fixed-rule-based methods that require manual adjustment, the method of this application can adaptively generate strategies, support natural language modification, and make strategies more flexible and convenient. Furthermore, compared to judging resource status based on a single indicator, the method of this application supports multimodal fusion, thereby improving prediction and optimization accuracy.
[0078] In some embodiments, the large model-based resource prediction and optimization device may further include:
[0079] A model fusion unit is used to fuse a preset text understanding model, a preset time series prediction model, and a preset multimodal model to obtain a resource prediction model to be trained;
[0080] A model training unit is used to generate a training data set based on the collected local event logs, local configuration files, local operation and maintenance manuals, and historical fault handling records, and train the resource prediction model to be trained using the training data set to obtain a resource prediction model to be determined;
[0081] The model determination unit is used to generate a policy generation prompt template corresponding to the software system, and fuse the policy generation prompt template and the resource prediction model to be determined to obtain a target resource prediction model.
[0082] In some embodiments, the prediction type determination module 11 may specifically include:
[0083] a request type determining unit, configured to, upon receiving a resource prediction request, parse the resource prediction request to determine whether the target resource prediction type of the resource prediction request is a short-term resource prediction request or a long-term resource prediction request;
[0084] A task generating unit is configured to generate a resource prediction task corresponding to the resource prediction request according to the target resource prediction type.
[0085] In some embodiments, the prediction solution generation module 12 may specifically include:
[0086] A data collection unit, used to collect local time series indicator data, log text data, configuration data and operation and maintenance manuals of the software system;
[0087] A data preprocessing unit, configured to perform data elimination on the time series indicator data based on a preset time series indicator range, and to fill in missing values in the time series indicator data to obtain target time series indicator data;
[0088] A data parsing unit, configured to parse the log text data to obtain target log text data corresponding to a preset log requirement data format in the log text data;
[0089] A data integration unit is used to use the target time series indicator data, the target log text data, the configuration data and the operation and maintenance manual as target multimodal data.
[0090] In some embodiments, the prediction solution generation module 12 may specifically include:
[0091] a health determination unit, configured to input the multimodal data into a target resource prediction model, so that the target resource prediction model parses the multimodal data based on the target resource prediction type, and determines the current system indicator health of the software system through the multimodal data;
[0092] A change trend generating unit is used to generate a system indicator change trend corresponding to the system indicator based on the system indicator health.
[0093] In some embodiments, the parameter changing module 13 may specifically include:
[0094] a policy generating unit, configured to generate a horizontal scaling adjustment policy and a vertical scaling adjustment policy based on the resource prediction scheme, and to generate a resource dynamic scheduling policy and a fault recovery policy according to the resource prediction scheme;
[0095] A parameter change unit is used to determine the parameter change plans corresponding to the horizontal scaling adjustment strategy, the vertical scaling adjustment strategy, the resource dynamic scheduling strategy and the fault recovery strategy, and generate configuration parameters in a preset target format based on the parameter change plan to perform local parameter changes based on the configuration parameters.
[0096] In some embodiments, the optimization judgment module 14 may specifically include:
[0097] An indicator monitoring unit, configured to monitor the real-time system indicators of the software system after parameter changes are made, and determine whether the real-time system indicators meet preset indicator operation threshold conditions;
[0098] The optimization ending unit is configured to end the resource optimization if the real-time system indicator meets the preset indicator operation threshold condition;
[0099] The configuration rollback unit is used to roll back the configuration and perform a resource alarm operation if the real-time system indicator does not meet the preset indicator operation threshold condition.
[0100] Furthermore, the embodiment of the present application also discloses an electronic device, Figure 4 This is a structural diagram of an electronic device 20 according to an exemplary embodiment. The content in the diagram should not be considered as any limitation to the scope of application of the present application.
[0101] Figure 4 This is a schematic diagram of the structure of an electronic device 20 provided in an embodiment of the present application. The electronic device 20 may specifically include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input / output interface 25, and a communication bus 26. The memory 22 is used to store a computer program, which is loaded and executed by the processor 21 to implement the relevant steps of the large model-based resource prediction and optimization method disclosed in any of the aforementioned embodiments. Furthermore, the electronic device 20 in this embodiment may specifically be an electronic computer.
[0102] In this embodiment, the power supply 23 is used to provide operating voltage for each hardware device on the electronic device 20; the communication interface 24 can create a data transmission channel between the electronic device 20 and the external device. The communication protocol it follows is any communication protocol that can be applied to the technical solution of this application and is not specifically limited here; the input and output interface 25 is used to obtain external input data or output data to the outside world. Its specific interface type can be selected according to specific application needs and is not specifically limited here.
[0103] In addition, the memory 22, as a carrier for resource storage, can be a read-only memory, random access memory, disk or CD, etc. The resources stored thereon can include an operating system 221, a computer program 222, etc., and the storage method can be temporary storage or permanent storage.
[0104] The operating system 221 is used to manage and control the hardware devices on the electronic device 20 and the computer program 222, and can be Windows Server, Netware, Unix, Linux, etc. In addition to including computer programs that can be used to implement the large model-based resource prediction and optimization method performed by the electronic device 20 disclosed in any of the aforementioned embodiments, the computer program 222 can further include computer programs that can be used to perform other specific tasks.
[0105] Furthermore, this application discloses a computer-readable storage medium for storing a computer program; wherein, when executed by a processor, the computer program implements the aforementioned large-scale model-based resource prediction and optimization method. The specific steps of this method can be found in the corresponding content disclosed in the aforementioned embodiments and will not be repeated here.
[0106] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from the other embodiments. Reference can be made to the descriptions of the identical or similar parts between the various embodiments. For the devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple, and the relevant parts can be referred to the descriptions of the methods.
[0107] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the above description has generally described the components and steps of each example according to their functions. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.
[0108] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein may be implemented directly using hardware, a software module executed by a processor, or a combination of the two. The software module may be placed in random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.
[0109] Finally, it should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of additional identical elements in the process, method, article, or device comprising the element.
[0110] The above is a detailed introduction to the technical solution provided by the present application. Specific examples are used herein to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method of the present application and its core idea. At the same time, for those skilled in the art, according to the ideas of the present application, there may be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present application.
Claims
1. A resource prediction and optimization method based on a large model, characterized in that: Applied to software systems, including: If a resource prediction request is received, the resource prediction request is parsed to determine a target resource prediction type; Collecting multimodal data local to the software system and inputting the multimodal data into a target resource prediction model, so that the target resource prediction model performs resource analysis based on the target resource prediction type and the multimodal data, and generates a corresponding resource prediction solution based on the multimodal data; generating a resource optimization strategy based on the resource prediction scheme, and converting the resource optimization strategy into configuration parameters in a preset target format, so as to perform local parameter changes based on the configuration parameters; Monitor the real-time system indicators of the software system after the parameters are changed, and confirm whether resource optimization is completed or resource alarm is issued based on the real-time system indicators.
2. The resource prediction and optimization method based on a large model according to claim 1, characterized in that: If a resource prediction request is received, the method further includes parsing the resource prediction request to determine the target resource prediction type: Perform model fusion on the preset text understanding model, the preset time series prediction model, and the preset multimodal model to obtain a resource prediction model to be trained; Generate a training data set based on the collected local event logs, local configuration files, local operation and maintenance manuals, and historical fault handling records, and train the resource prediction model to be trained using the training data set to obtain a resource prediction model to be determined; A policy generation prompt template corresponding to the software system is generated, and the policy generation prompt template and the to-be-determined resource prediction model are fused to obtain a target resource prediction model.
3. The resource prediction and optimization method based on a large model according to claim 1, characterized in that: If a resource prediction request is received, parsing the resource prediction request to determine a target resource prediction type includes: If a resource prediction request is received, parsing the resource prediction request to determine whether the target resource prediction type of the resource prediction request is a short-term resource prediction request or a long-term resource prediction request; A resource prediction task corresponding to the resource prediction request is generated according to the target resource prediction type.
4. The resource prediction and optimization method based on a large model according to claim 1, characterized in that: The collecting of the local multimodal data of the software system includes: Collect local time series indicator data, log text data, configuration data, and operation and maintenance manuals of the software system; Eliminate data from the time series indicator data based on a preset time series indicator range, and complete missing values in the time series indicator data to obtain target time series indicator data; Parsing the log text data to obtain target log text data corresponding to a preset log requirement data format in the log text data; The target time series indicator data, the target log text data, the configuration data and the operation and maintenance manual are used as target multimodal data.
5. The resource prediction and optimization method based on a large model according to claim 1 is characterized in that: Inputting the multimodal data into a target resource prediction model so that the target resource prediction model performs resource analysis based on the target resource prediction type and the multimodal data, and generates a corresponding resource prediction solution based on the multimodal data, includes: Inputting the multimodal data into a target resource prediction model so that the target resource prediction model parses the multimodal data based on the target resource prediction type to determine the current system indicator health of the software system through the multimodal data; A system indicator change trend corresponding to the system indicator is generated based on the system indicator health.
6. The resource prediction and optimization method based on a large model according to claim 1 is characterized in that: Generating a resource optimization strategy based on the resource prediction solution, and converting the resource optimization strategy into configuration parameters in a preset target format to perform local parameter changes based on the configuration parameters, includes: Generate a horizontal expansion and contraction adjustment strategy and a vertical expansion and contraction adjustment strategy based on the resource prediction scheme, and generate a resource dynamic scheduling strategy and a fault recovery strategy according to the resource prediction scheme; Determine the parameter change plans corresponding to the horizontal scaling adjustment strategy, the vertical scaling adjustment strategy, the resource dynamic scheduling strategy, and the fault recovery strategy, respectively, and generate configuration parameters in a preset target format based on the parameter change plan to perform local parameter changes based on the configuration parameters.
7. The resource prediction and optimization method based on a large model according to any one of claims 1 to 6, characterized in that: The monitoring of the real-time system indicators of the software system after the parameter change, and confirming whether resource optimization is completed or resource alarm is issued based on the real-time system indicators, includes: Monitoring the real-time system indicators of the software system after the parameters are changed, and determining whether the real-time system indicators meet the preset indicator operation threshold conditions; If the real-time system indicator meets the preset indicator operation threshold condition, then end this resource optimization; If the real-time system indicator does not meet the preset indicator operation threshold condition, the configuration is rolled back and a resource alarm operation is performed.
8. A resource prediction and optimization device based on a large model, characterized in that: Applied to software systems, including: a prediction type determination module, configured to, upon receiving a resource prediction request, parse the resource prediction request to determine a target resource prediction type; a prediction scheme generation module, configured to collect multimodal data local to the software system and input the multimodal data into a target resource prediction model, so that the target resource prediction model performs resource analysis based on the target resource prediction type and the multimodal data, and generates a corresponding resource prediction scheme based on the multimodal data; a parameter change module, configured to generate a resource optimization strategy based on the resource prediction scheme, and convert the resource optimization strategy into configuration parameters in a preset target format, so as to perform local parameter changes based on the configuration parameters; The optimization judgment module is used to monitor the real-time system indicators of the software system after the parameters are changed, and to confirm whether resource optimization is completed or to issue a resource alarm based on the real-time system indicators.
9. An electronic device, characterized in that: include: Memory, used to store computer programs; A processor, configured to execute the computer program to implement the large model-based resource prediction and optimization method according to any one of claims 1 to 7.
10. A computer-readable storage medium, characterized in that Used to store a computer program, wherein when the computer program is executed by a processor, it implements the resource prediction and optimization method based on a large model as described in any one of claims 1 to 7.