CN-122019302-A - Log data intelligent operation and maintenance method, operation and maintenance system, electronic equipment and storage medium

CN122019302ACN 122019302 ACN122019302 ACN 122019302ACN-122019302-A

Abstract

The invention discloses an intelligent operation and maintenance method, an operation and maintenance system, electronic equipment and a storage medium for log data, which belong to the technical field of electric digital data processing; the near-end strategy optimization rewarding function is a weighted sum of the recovery gradient of monitoring indexes in a preset time window for executing the repairing scheme, the execution cost of the repairing scheme and the confidence of the repairing scheme in the simulation environment. The input features are fused with text vectors, time differences and position codes, so that the prediction model can sense the occurrence frequency and interval of faults, the near-end strategy optimization can guide the prediction model to generate an effective maintenance scheme, the effectiveness of the maintenance scheme is improved, the faults and early warning are predicted in time based on the analysis of log data by the prediction model, and a corresponding repair scheme is generated.

Inventors

ZHENG GUOLIN
LIU YIMING
LUO CHENG
WANG AOYU

Assignees

杭州谐云科技有限公司

Dates

Publication Date: 20260512
Application Date: 20260126

Claims (10)

1. The intelligent operation and maintenance method for the log data is characterized by comprising the following steps of: Predicting the log data through a prediction model based on deep learning and near-end strategy optimization to obtain a fault classification and repair scheme; Wherein the near-end policy-optimized reward function is a weighted sum of the following metrics: monitoring the recovery gradient of the index, the execution cost of the repair scheme and the confidence of the repair scheme in a preset time window for executing the repair scheme in the simulation environment.
2. The intelligent operation and maintenance method of log data according to claim 1, wherein the reward function R is expressed as: R = a* M recovery – b*C action +c *S confidence ; Wherein a, b and C are weight data respectively, M recovery is a recovery gradient of monitoring indexes in a preset time window for executing the repair scheme in a simulation environment, C action is the execution cost of the repair scheme, and S confidence is the confidence level of the repair scheme; and guiding the prediction model to generate a maintenance scheme in a mode of maximizing rewards.
3. The method for intelligently operating and maintaining log data according to claim 1, wherein the training method of the prediction model comprises the following steps: Collecting historical log data; preprocessing the history log data to obtain a training set; encoding the historical log data of the training set to obtain a text vector; obtaining time difference and position code of history log data; Fusing the text vector, the time difference and the position code to obtain input characteristics; The method is based on deep learning and near-end strategy optimization, and a large language model is trained and/or fine-tuned through a training set and input features of the training set to obtain a prediction model.
4. The intelligent operation and maintenance method for log data according to claim 3, wherein the method for generating the input features comprises: mapping the time difference to a high-dimensional space to obtain a time embedded vector; and after the characteristic connection operation is carried out on the text vector and the time embedded vector, the text vector and the time embedded vector are fused with the position code, and the input characteristic is obtained.
5. The intelligent operation and maintenance method for log data according to claim 4, wherein the method for obtaining the position code comprises the following steps: coding source nodes of log data based on a graph neural network to obtain position codes; The input feature E final is represented as: E final = Concat(E text , E time ) +E position ; Wherein Concat () is represented as a feature join operation, + is represented as a feature fusion operation, E text is a text vector, E position is a position code, and E time is a time embedded vector.
6. A log data intelligent operation and maintenance method according to claim 3, wherein the input feature E final is expressed as: E final = E text + E position + E time ; Where E text is a text vector, E position is a position code, E time is a time embedded vector, + is denoted as a feature fusion operation.
7. The intelligent operation and maintenance method of log data according to claim 3, wherein the preprocessing comprises data cleaning, denoising, alignment, normalization, time correction and data alignment, and log data parsing and field extraction; the method also comprises the following steps of: and generating an alarm message according to the fault classification and repair scheme.
8. An operation and maintenance system, which is characterized in that the operation and maintenance system is used for realizing the intelligent operation and maintenance method of log data according to any one of claims 1-7, comprising a prediction module, The prediction module is used for predicting the log data through a prediction model based on deep learning and near-end strategy optimization to obtain fault classification and a repair scheme.
9. A storage medium, characterized in that the storage medium holds code, which when processed, performs the log data intelligent operation and maintenance method according to any one of claims 1-7.
10. An electronic device comprising a processor and a storage medium as claimed in claim 9.

Description

Log data intelligent operation and maintenance method, operation and maintenance system, electronic equipment and storage medium Technical Field The invention relates to the technical field of electric digital data processing, in particular to an intelligent operation and maintenance method, an operation and maintenance system, electronic equipment and a storage medium for log data. Background Log data is an important component of modern network security policies that provides a rich source of information about events occurring in an organized digital environment. Conventional operation and maintenance methods generally rely on manual analysis of logs, and fault localization and processing is performed through experience of operation and maintenance personnel, but with the rapid increase of log data volume, manual analysis becomes inefficient and prone to errors. The existing log data operation and maintenance mainly realizes screening and alarm pushing of logs through regular expressions, but most technologies still stay at the level of identifying known fault modes, lack intelligent learning and prediction capabilities for novel faults, and cannot cope with complex and changeable system behaviors and fault scenes. The statistical analysis can be simply carried out on the log data, and potential information in the log data cannot be deeply mined, so that bottlenecks or potential risks in the system cannot be found in time. This way of operation has not been satisfactory in the face of large-scale, complex IT environments. The presently disclosed technology attempts to use models such as BERT or T5 for text classification of log data to identify faults, but lacks the dynamics and result guidance of fault handling. Specifically, during the derivation process, the model cannot learn the validity of the generated maintenance scheme, and the coding of the time sequence is also lacked. The patent with publication number CN120872672A discloses a fault root cause positioning method based on a large model, and the text vector and the time sequence feature are fused based on a cross attention mechanism to obtain the initial fusion feature corresponding to the target log data. The time sequence characteristic is the time sequence frequency of the target log data, and the calculation mode of the time sequence frequency is not recorded. Therefore, the semantic understanding capability of the large model is combined with the time state feedback, and the log data intelligent operation and maintenance method with result guidance is an important development direction. Disclosure of Invention Aiming at the technical problems in the prior art, the invention provides an intelligent operation and maintenance method for log data, an operation and maintenance system, electronic equipment and a storage medium, and has good result guidance. The invention discloses an intelligent operation and maintenance method for log data, which comprises the following steps of predicting the log data through a prediction model based on deep learning and near-end strategy optimization to obtain a fault classification and repair scheme, wherein a near-end strategy optimization rewarding function is a weighted sum of indexes including recovery gradient of monitoring indexes in a preset time window for executing the repair scheme in a simulation environment, execution cost of the repair scheme and confidence level of the repair scheme. Preferably, the reward function R is expressed as: R = a* Mrecovery– b*Caction+c *Sconfidence; Wherein a, b and C are weight data respectively, M recovery is a recovery gradient of monitoring indexes in a preset time window for executing the repair scheme in a simulation environment, C action is the execution cost of the repair scheme, and S confidence is the confidence level of the repair scheme; and guiding the prediction model to generate a maintenance scheme in a mode of maximizing rewards. Preferably, the training method of the prediction model includes: Collecting historical log data; preprocessing the history log data to obtain a training set; encoding the historical log data of the training set to obtain a text vector; obtaining time difference and position code of history log data; Fusing the text vector, the time difference and the position code to obtain input characteristics; The method is based on deep learning and near-end strategy optimization, and a large language model is trained and/or fine-tuned through a training set and input features of the training set to obtain a prediction model. Preferably, the method for generating the input features includes: mapping the time difference to a high-dimensional space to obtain a time embedded vector; and after the characteristic connection operation is carried out on the text vector and the time embedded vector, the text vector and the time embedded vector are fused with the position code, and the input characteristic is obtained. Preferably, the method for o