Information operation dimension intelligent management and control platform system and method, electronic equipment and storage medium

Abstract

The invention discloses an information operation dimension intelligent management and control platform system and method, electronic equipment and a storage medium, and belongs to the field of intelligent operation and maintenance of an information communication technology. The system comprises an anomaly detection module, a multi-modal feature fusion module and the like, multi-source data is processed through an unstructured data preprocessing unit and a time sequence data alignment unit, model parameters are dynamically adjusted by utilizing a self-adaptive online learning strategy, and dynamic resource scheduling is realized by combining a reinforcement learning algorithm. The method comprises the steps of data receiving preprocessing, feature fusion alignment, model training analysis and the like. The electronic equipment adopts a heterogeneous computing architecture, and the storage medium stores corresponding program codes. According to the method, the problems of weak generalization ability, insufficient unstructured data processing and the like of a traditional platform model are solved, the anomaly detection precision and the response speed are improved, a closed-loop operation and maintenance system is formed, the operation and maintenance decision intelligent level is improved, and the generalization recognition ability and the environmental adaptability of the model to an unknown abnormal mode are remarkably enhanced.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of information communication technology application, and in particular to an information operation and maintenance intelligent management and control platform system and method, an electronic device and a storage medium. BACKGROUND

[0002] In traditional information communication technology operation and maintenance, manual monitoring and manual operation dominate, which leads to low operation and maintenance efficiency, long response time, and difficulty in effectively managing large-scale complex systems. In recent years, with the development of big data, artificial intelligence and other technologies, intelligent operation and maintenance has gradually become an important means to improve operation and maintenance efficiency and reduce operation cost.

[0003] Common intelligent operation and maintenance platforms assist operation and maintenance decision-making through pre-set rule bases and basic machine learning algorithms. These platforms can achieve a certain degree of automated early warning and fault diagnosis, but this approach has weak model generalization ability and insufficient adaptability, especially when dealing with unstructured data processing and unknown abnormal events, its diagnosis accuracy and response speed often cannot meet the requirements of high-availability systems, and cannot meet the work requirements of information communication technology application. Therefore, an information operation and maintenance intelligent management and control platform system and method, an electronic device and a storage medium are proposed. SUMMARY

[0004] The present application provides the following technical solutions: an information operation and maintenance intelligent management and control platform system, comprising: an anomaly detection module, a multi-modal feature fusion module, an adaptive online learning strategy module, an operation and maintenance management center module and a resource scheduling module, the anomaly detection module running on a server of a data center and receiving data from each operation and maintenance node through a dedicated interface, the anomaly detection module being configured to identify normal behavior patterns of the system by unsupervised learning and training using large-scale historical operation and maintenance data, and to analyze received data in real time; the multi-modal feature fusion module being configured to combine multiple data sources such as system logs, network traffic and hardware status, the adaptive online learning strategy module being executed by an SDN controller, the SDN controller being connected to the server and each operation and maintenance node, the adaptive online learning strategy module being configured to dynamically adjust parameters of the anomaly detection model according to real-time operation and maintenance data; the operation and maintenance management center module being configured to start a fault diagnosis process based on alarm information, and to update parameters of the anomaly detection model by calling the adaptive online learning strategy, the resource scheduling module having a reinforcement learning algorithm inside, the resource scheduling module being configured to dynamically adjust resource allocation of each node using the reinforcement learning algorithm based on the current system state and historical operation and maintenance data; The multimodal feature fusion module internally integrates an unstructured data preprocessing unit and a time series data alignment unit, the adaptive online learning strategy module internally provides a concept drift detection submodule and a model incremental update submodule, and the resource scheduling module internally integrates a multidimensional state space construction unit and a hierarchical action decision unit.

[0005] The application provides an information operation and maintenance intelligent management and control method based on the information operation and maintenance intelligent management and control platform system, and comprises the following steps: S1 data receiving and preprocessing: Firstly, the abnormality detection module receives data from each operation and maintenance node, and pre-processes the received data; S2 feature fusion and alignment: Different types of data sources of system logs, network traffic and hardware states are firstly subjected to unstructured data preprocessing unit to extract key features by using natural language processing technology, and then subjected to time series data alignment unit to combine a sliding window and interpolation fitting in a manner of time stamp as a reference to complete the spatio-temporal alignment of multi-source data; S3 abnormality detection model training and real-time analysis: The abnormality detection module is used to train the model based on large-scale historical operation and maintenance data by using unsupervised learning, identify the normal behavior mode of the system, and analyze the real-time operation and maintenance data, and when an abnormality is detected, an alarm information is generated; S4 model parameter adaptive online adjustment: The adaptive online learning strategy module is used to obtain real-time operation and maintenance data through the SDN controller, and when it is detected that the data distribution is significantly deviated, the model incremental update submodule is automatically triggered, and the abnormality detection model parameters are updated by using the online learning algorithm based on the newly added real-time data and gradually iterated; S5 fault diagnosis and resource dynamic scheduling: The fault diagnosis process is started through the operation and maintenance management center module based on the alarm information generated by the abnormality detection module, and the adaptive online learning strategy is used to update the abnormality detection model parameters, and the resource scheduling module is used to combine the real-time state parameters of CPU load, memory usage and network bandwidth with the historical operation and maintenance data behavior mode through the multidimensional state space construction unit to construct a multidimensional state vector comprising the current running state, resource occupation and historical trend; The application provides an electronic device of an information operation and maintenance intelligent management and control platform, adopts the information operation and maintenance intelligent management and control method, and comprises the following: A processor for executing a computer program, a memory for storing program instructions of the information operation and maintenance intelligent management and control platform system, the program instructions comprising program codes of an anomaly detection module, a multi-modal feature fusion module, an adaptive online learning strategy module, an operation and maintenance management center module, and a resource scheduling module; A communication interface for data interaction with each operation and maintenance node, receiving multi-source operation and maintenance data of system logs, network traffic, and hardware states, and sending alarm information and resource scheduling instructions; The application provides a storage medium of an information operation and maintenance intelligent management and control platform, an electronic device based on the information operation and maintenance intelligent management and control platform, and the like. A read card internally storing a computer program, the computer program realizing an information operation and maintenance intelligent management and control method when executed by a processor, the computer program comprising program codes realizing functions of an anomaly detection module, a multi-modal feature fusion module, an adaptive online learning strategy module, an operation and maintenance management center module, and a resource scheduling module.

[0006] Preferably, the unstructured data preprocessing unit extracts key features through natural language processing technology, converts unstructured log content into structured feature vectors, and simultaneously performs protocol analysis and field extraction on unstructured fields in network traffic data, so that the multi-modal feature fusion module can uniformly process different types of data sources.

[0007] Preferably, the time series data alignment unit adopts a combination of a sliding window and interpolation fitting, and performs spatio-temporal alignment processing on multi-source data based on timestamps, the concept drift detection submodule is used for dynamically evaluating the change trend of the system operation mode by continuously monitoring the difference between the distribution characteristics of real-time operation and maintenance data and historical training data, and automatically triggering the model incremental update submodule when a significant deviation in data distribution is detected.

[0008] Preferably, the model incremental updating submodule is based on newly added real-time operation and maintenance data, and uses an online learning algorithm to iteratively update the anomaly detection model, while retaining historical training knowledge and incorporating new data features. The multi-dimensional state space construction unit combines real-time system state parameters such as CPU load, memory usage, and network bandwidth with behavior patterns in historical operation and maintenance data to construct a multi-dimensional state vector containing current operating state, resource occupation, and historical trends. The model incremental updating submodule iteratively updates model parameters based on online learning. The multi-dimensional state space construction unit combines real-time and historical data to improve resource scheduling rationality.

[0009] Preferably, in step S1, the anomaly detection module performs a multi-level verification mechanism on the received raw data, including integrity verification, format compliance checking, and preliminary screening of abnormal values. For unstructured data, the preprocessing unit identifies key operation nodes and state descriptions in the log through semantic analysis techniques, and combines regular expression matching to achieve standardized extraction of sensitive fields. The anomaly detection module performs multi-level data verification, and the preprocessing unit combines semantic analysis and regular matching to improve data preprocessing quality and efficiency.

[0010] Preferably, the processor is a heterogeneous computing architecture that includes CPU and GPU co-processing units. The CPU in the CPU and GPU co-processing units is used to perform logical control and general computing tasks, while the GPU is used to accelerate natural language processing operations in unstructured data preprocessing and matrix operations in multi-modal feature fusion. The memory uses a solid-state storage and cache combined architecture. The solid-state storage is used to persistently store large-scale historical operation and maintenance data and trained anomaly detection models. The cache is used to temporarily store real-time received operation and maintenance data and intermediate parameters during model incremental updating. Through CPU and GPU co-processing in the heterogeneous computing architecture and the solid-state storage and cache combined architecture, data processing speed and model training efficiency are improved.

[0011] Preferably, the program code in the computer program that implements the multi-modal feature fusion module function includes an unstructured data preprocessing unit program and a time series data alignment unit program. The unstructured data preprocessing unit program is used to perform semantic analysis on system logs and extract key features through natural language processing techniques and convert them into structured feature vectors. At the same time, it performs protocol analysis on unstructured fields in network traffic data, so that different types of data sources can be uniformly processed. The time series data alignment unit program uses a combination of sliding window and interpolation fitting, and aligns multi-source time series data in space and time based on timestamps. Through semantic analysis, protocol analysis, and time series alignment by the multi-modal feature fusion module program, standardized processing and space-time alignment of multi-source data are achieved.

[0012] In summary, compared with the prior art, the information operation and maintenance intelligent management and control platform system, method, electronic equipment and storage medium provided by the present application have the following beneficial effects: 1、The multi-modal feature fusion module of the present application converts system logs, network traffic and other unstructured data into structured feature vectors through an unstructured data preprocessing unit, and eliminates the time deviation of multi-source data using a time series data alignment unit, solving the problem of insufficient unstructured data processing capability of traditional platforms, providing comprehensive and accurate feature input for anomaly detection, and through the anomaly detection module and in combination with the unsupervised learning framework and the adaptive online learning strategy, the concept drift detection submodule is used to monitor the data distribution changes in real time, triggering the model incremental update submodule to iteratively update the parameters based on real-time data online, avoiding the decline in detection accuracy caused by data distribution drift of the traditional preset rule base, and enhancing the generalization recognition ability and environmental adaptability of the model to unknown abnormal patterns; 2、The present application uses a resource scheduling module and a reinforcement learning algorithm to integrate real-time system state and historical data through a multi-dimensional state space construction unit, and realizes hierarchical optimization of resource allocation through a hierarchical action decision unit, so that when an anomaly occurs, the operation and maintenance management center dynamically adjusts the affected node resources through the resource scheduling module, shortens the response time of unknown abnormal events, improves the system fault tolerance and recovery efficiency, and guarantees the service continuity of the high-availability system, so that the above-mentioned modules form a closed-loop system of detection, fusion, optimization and scheduling, and realize data interaction and policy linkage through the SDN controller, avoiding the problems of model solidification and decision lag in traditional solutions, and improving the intelligent level of operation and maintenance decision as a whole. BRIEF DESCRIPTION OF DRAWINGS

[0013] Figure 1 is the system architecture diagram of the present application.

[0014] Figure 2 is the method flowchart of the present application. DETAILED DESCRIPTION

[0015] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0016] Please refer to Figure 1 The present application provides a technical solution, an information operation and maintenance intelligent management and control platform system, comprising: The anomaly detection module, the multi-modal feature fusion module, the adaptive online learning strategy module, the operation and maintenance management center module and the resource scheduling module, the anomaly detection module runs on the server of the data center, and receives data from each operation and maintenance node through a special interface, the anomaly detection module is used for training by unsupervised learning mode and using large-scale historical operation and maintenance data to identify the normal behavior mode of the system, and analyzing the received data in real time; The multi-modal feature fusion module is used for combining various data sources of system logs, network traffic and hardware states, the adaptive online learning strategy module is executed through an SDN controller, the SDN controller is connected with the server and each operation and maintenance node, and the adaptive online learning strategy module is used for dynamically adjusting parameters of the anomaly detection model according to real-time operation and maintenance data; The operation and maintenance management center module is used for starting a fault diagnosis process based on alarm information, and calling the adaptive online learning strategy to update parameters of the anomaly detection model, and the resource scheduling module is internally provided with a reinforcement learning algorithm, and the resource scheduling module is used for dynamically adjusting resource allocation of each node by using the reinforcement learning algorithm according to the current system state and historical operation and maintenance data; The multi-modal feature fusion module is internally integrated with an unstructured data preprocessing unit and a time series data alignment unit, the adaptive online learning strategy module is internally provided with a concept drift detection submodule and a model incremental update submodule, the resource scheduling module is internally integrated with a multi-dimensional state space construction unit and a hierarchical action decision unit, the unstructured data preprocessing unit extracts key features by using a natural language processing technology, converts unstructured log content into a structured feature vector, and performs protocol analysis and field extraction on unstructured fields in network traffic data, so that the multi-modal feature fusion module can uniformly process different types of data sources; The time series data alignment unit adopts a combination of a sliding window and interpolation fitting, and performs spatio-temporal alignment processing on multi-source data based on timestamps, the concept drift detection submodule is used for dynamically evaluating the change trend of the system operation mode by continuously monitoring the difference between the distribution characteristics of real-time operation and maintenance data and historical training data, automatically triggering the model incremental update submodule when a significant deviation of data distribution is detected, the model incremental update submodule gradually iteratively updates the anomaly detection model by using an online learning algorithm based on the newly added real-time operation and maintenance data, and integrates new data features while retaining historical training knowledge, and the multi-dimensional state space construction unit is used for combining real-time system state parameters such as CPU load, memory usage and network bandwidth with behavior patterns in historical operation and maintenance data to construct a multi-dimensional state vector containing the current operation state, resource occupation and historical trend.

[0017] Please refer to Figure 2The application provides an information operation and maintenance intelligent management and control method based on the information operation and maintenance intelligent management and control platform system, and comprises the following steps: S1 data receiving and preprocessing: Firstly, the abnormality detection module receives data from each operation and maintenance node, and pre-processes the received data. The abnormality detection module performs a multi-level checking mechanism on the received raw data, including integrity checking, format compliance checking and abnormal value preliminary screening. For unstructured data, the preprocessing unit identifies key operation nodes and state descriptions in the log through semantic analysis technology, and realizes standardized extraction of sensitive fields by combining regular expression matching. The specific implementation process of the above method is as follows: Firstly, the abnormality detection module establishes stable connection with each operation and maintenance node through a special interface. These nodes may be distributed in different positions such as data centers, network devices and server clusters, and continuously generate multiple data such as system logs, performance indicators and network traffic. The module adopts a multi-threaded or asynchronous communication mechanism to ensure the real-time and high efficiency of data receiving, and avoid data loss caused by network delay or node load; The received raw data first enters the multi-level checking mechanism. The first layer is integrity checking, and the module checks whether the data packet contains necessary fields and information, such as whether the log entry contains timestamp, event type, source node and other key attributes. If the detected data is incomplete, the module will trigger a retransmission request or record a missing log to ensure the coherence of subsequent analysis. The second layer is format compliance checking, and the module verifies whether the data format conforms to the standard according to the preset regular expression or pattern matching rule. For example, system logs need to follow a specific text format, and network traffic data need to comply with the TCP / IP protocol specification. For data with abnormal format, the module will correct the format or mark it as an abnormal sample. The third layer is abnormal value preliminary screening, and the module uses statistical methods or rule engines to identify obvious abnormal values in the data, such as CPU usage exceeding 100%, illegal memory address, etc. Such abnormal values may be caused by sensor failure or data collection error, and the module will isolate and record them to avoid interference with subsequent analysis; For non-structured data that passes the checking, such as system logs in free text format and binary network traffic, the module calls the non-structured data preprocessing unit for in-depth analysis. The unit first performs semantic analysis on the log text, and uses natural language processing technology to identify key operation nodes and state descriptions. For example, through named entity recognition, user ID, operation type, resource name and other entities are extracted, and through sentiment analysis, the severity of the event is judged. At the same time, the unit combines regular expression matching to standardize the extraction of sensitive fields in the log, such as converting different formats of timestamps to standard time format, and parsing IP addresses to standardized strings; For unstructured fields in network traffic data, such as HTTP request headers, TCP payload content, etc., the preprocessing unit performs protocol parsing and field extraction. The unit calls the corresponding protocol parser according to the packet type to extract key field information. For example, extract URL path, request method, user agent, etc. from HTTP request, and extract application layer protocol features from TCP payload. Through such parsing, the unit converts unstructured network traffic into structured feature vectors, facilitating subsequent multi-modal feature fusion; S2 feature fusion and alignment: Different types of data sources such as system logs, network traffic and hardware status are first processed by the unstructured data preprocessing unit using natural language processing techniques to extract key features, converting unstructured logs into structured feature vectors. Then, the time series data alignment unit uses a combination of sliding window and interpolation fitting to complete the spatio-temporal alignment of multi-source data based on timestamps. The specific implementation process of the above method is as follows: Feature fusion stage: The multi-modal feature fusion module first receives standardized data from the data preprocessing stage, which covers multiple dimensions such as system logs, network traffic, hardware status, etc. For unstructured data types, the module calls the unstructured data preprocessing unit for deep analysis. For system log text, the unit uses natural language processing techniques to extract key operation nodes through steps such as word segmentation, part-of-speech tagging, entity recognition, etc. For example, from the log "User admin performs database backup operation at 14:23", the unit will identify "user entity", "operation type", "resource object", "timestamp" and other core elements, and convert them into structured feature vectors. At the same time, the unit combines regular expression matching rules to standardize the extraction of fields such as status code, error level in the log, ensuring the comparability of different formats of logs. For network traffic data, the preprocessing unit performs protocol parsing and field extraction operations. The unit automatically selects parsing strategies according to packet types, such as extracting URL path, request method, response status code, etc. for HTTP protocol, and parsing query domain name and response IP address for DNS protocol. Through such parsing, the unit converts raw binary traffic into structured data containing protocol features, transmission behavior, and content attributes. For hardware status data such as CPU utilization, memory occupancy, etc., the unit directly extracts numerical features and marks abnormal states based on threshold rules.

[0018] Spatio-temporal alignment stage: After feature extraction, the multi-modal feature fusion module enters the time series data alignment process. This process is dominated by the time series data alignment unit, whose core task is to solve the problem of temporal and spatial inconsistency caused by different collection frequencies and transmission delays. The unit first aligns the feature vectors of each data source based on timestamps. For data with missing timestamps or large timestamp deviations, the unit uses a sliding window mechanism to search for matching features within a pre-set time window. For example, if a log record has a timestamp of 14:23:05 and the corresponding network traffic data is missing, the unit will search for the closest traffic snapshot within a ±5 second window as a replacement; To address the problem of inconsistent data collection frequencies, the unit uses interpolation fitting techniques for fine-grained alignment. For high-frequency data (such as CPU utilization collected every second), the unit uses linear interpolation or spline interpolation methods to generate intermediate values between the time points of low-frequency data (such as log entries recorded every minute), thereby constructing a continuous time series. For periodic data (such as network traffic periodic fluctuations), the unit extracts time-domain features by combining Fourier transform or wavelet analysis, and ensures the synchronization of periodic patterns in multi-source data through phase alignment; After feature fusion and spatio-temporal alignment, the original multi-source heterogeneous data is transformed into a structured feature matrix containing spatio-temporal consistency. This matrix not only retains the key information of each data source, but also reveals the potential patterns of system operation through feature intersection and spatio-temporal correlation. For example, by fusing operation records in logs and transmission features in network traffic, abnormal login behavior accompanied by illegal data transmission patterns can be identified; by correlating hardware status indicators and error events in logs, specific operation nodes causing performance degradation can be located; S3 Abnormality Detection Model Training and Real-time Analysis: Using the anomaly detection module, the model is trained based on large-scale historical operation and maintenance data through unsupervised learning, to identify normal behavior patterns of the system and analyze real-time operation and maintenance data. When an anomaly is detected, an alarm message is generated. The specific implementation process of the above method is as follows: Model Training Phase: The anomaly detection module first loads large-scale historical operation and maintenance data from the storage. These data cover multiple dimensions such as system logs, network traffic, and hardware status, and have been transformed into structured feature matrices through data preprocessing and feature fusion processes. The module uses an unsupervised learning framework to automatically mine potential patterns in the data without manual annotation. Specifically, the module identifies the core cluster of data distribution through clustering algorithms, and defines the frequently occurring feature combination as the normal behavior pattern of the system. For example, by analyzing the operation sequence and network traffic pattern in historical logs, the module can establish a "normal operation behavior atlas" to mark frequent user login, service invocation, and other operations as regular behavior; Meanwhile, the module constructs a dynamic baseline using time series analysis techniques. By decomposing the time series of historical data, the module extracts periodic trends and seasonal fluctuations. For example, it encodes the regularity of CPU utilization peak during daily business peak hours, network traffic low during weekends, and other rules into time series rules. On this basis, the module sets an abnormal threshold value by combining statistical methods, such as calculating the normal range boundary based on the mean-standard deviation model, or determining the extreme value critical point using the quantile method. Finally, the module encapsulates the normal behavior pattern and dynamic baseline as a detection model, and stores it persistently in a solid-state storage unit for real-time analysis stage calls; Real-time analysis stage: When new data flows in, the anomaly detection module first receives the pre-processed structured feature vector through the communication interface. These data have been processed through multi-level verification, unstructured analysis, and time series alignment processes to ensure data quality and spatio-temporal consistency. The module inputs real-time features into the trained detection model, and calculates the deviation of current data from normal behavior patterns through similarity matching, distance measurement, or probability estimation methods. For example, using a density-based anomaly detection algorithm, it evaluates the neighborhood density of the real-time feature vector in the feature space, and if the density is lower than the pre-set threshold, it is determined as a potential anomaly; To reduce false positives, the module introduces a multi-dimensional cross-validation mechanism. First, it checks whether the real-time data conforms to the periodicity rule by combining the time series baseline, such as whether the current CPU utilization exceeds the historical fluctuation range. Second, it verifies the logical consistency between multiple features through correlation analysis, such as whether the network traffic surge is accompanied by corresponding business operation logs. If the real-time data shows abnormal characteristics in multiple dimensions, the module will generate an alarm information and send an event notification to the operation and management center module through the communication interface; Dynamic optimization mechanism: The anomaly detection module and the adaptive online learning strategy module form a collaborative closed loop. The SDN controller continuously streams real-time operation and maintenance data into the concept drift detection submodule, which compares the statistical characteristics of the current data distribution and historical training data using the two-sample test method. For example, it monitors the change trend of network traffic entropy, log event type frequency, and other indicators. If it detects a significant shift in data distribution (such as a sustained decrease in traffic entropy, indicating a single protocol type), the submodule will automatically trigger the model incremental update process; The model incremental updating submodule adopts an online learning algorithm to input the newly added real-time data in a small batch manner, and adjusts the model parameters through a gradient descent or Bayesian updating mechanism. During the updating process, the module retains historical training knowledge through regularization techniques to avoid model drift caused by new data noise. For example, when updating the cluster centers, a dynamic weight strategy is adopted to make the influence of historical data on the model decay over time, while the influence of new data gradually increases. Finally, the updated model parameters are synchronized to the anomaly detection module to ensure that the detection logic is synchronized with the system operation mode; S4 model parameter adaptive online adjustment: Relying on the adaptive online learning strategy module, real-time operation and maintenance data are obtained through the SDN controller. When it is detected that the data distribution deviates significantly, the model incremental updating submodule is automatically triggered, and the anomaly detection model parameters are iteratively updated based on the newly added real-time data and the online learning algorithm; S5 fault diagnosis and dynamic resource scheduling: The fault diagnosis process is started through the operation and maintenance management center module based on the alarm information generated by the anomaly detection module, and the adaptive online learning strategy is called to update the anomaly detection model parameters. The resource scheduling module combines the real-time state parameters of CPU load, memory usage and network bandwidth with historical operation and maintenance data behavior patterns to construct a multi-dimensional state space, and constructs a multi-dimensional state vector containing the current running state, resource occupation and historical trend; The present application provides an electronic device for information operation and maintenance intelligent management and control platform, which adopts the above-mentioned information operation and maintenance intelligent management and control method, comprising: A processor, a memory and a communication interface, the processor is used for executing a computer program, the memory is used for storing program instructions of the information operation and maintenance intelligent management and control platform system, the program instructions include program codes of the anomaly detection module, the multi-modal feature fusion module, the adaptive online learning strategy module, the operation and maintenance management center module and the resource scheduling module; The communication interface is used for data interaction with each operation and maintenance node, receives multi-source operation and maintenance data of system logs, network traffic and hardware state, and sends alarm information and resource scheduling instructions; The processor is a heterogeneous computing architecture, which includes CPU and GPU cooperative processing units. The CPU in the CPU and GPU cooperative processing units is used for executing logical control and general computing tasks, while the GPU is used for accelerating natural language processing operations in unstructured data preprocessing and matrix operations in multi-modal feature fusion. The memory adopts a solid state storage and cache combined architecture. The solid state storage is used for persistent storage of large-scale historical operation and maintenance data and trained anomaly detection models, and the cache is used for temporary storage of real-time received operation and maintenance data and intermediate parameters in the model incremental updating process; The application provides a storage medium of an information operation and maintenance intelligent management and control platform, an electronic device based on the information operation and maintenance intelligent management and control platform, and the like. The read card internally stores a computer program, and the computer program realizes the information operation and maintenance intelligent management and control method when executed by the processor. The computer program includes program codes for realizing the functions of the abnormality detection module, the multi-modal feature fusion module, the adaptive online learning strategy module, the operation and maintenance management center module and the resource scheduling module. The program codes for realizing the function of the multi-modal feature fusion module in the computer program include an unstructured data preprocessing unit program and a time series data alignment unit program. The unstructured data preprocessing unit program is used for performing semantic analysis on system logs, extracting key features through natural language processing technology and converting the key features into a structured feature vector, and performing protocol analysis on unstructured fields in network traffic data, so that different types of data sources can be uniformly processed. The time series data alignment unit program combines a sliding window and an interpolation fitting mode, and performs space-time alignment on multi-source time series data based on timestamps.

[0019] The multi-modal feature fusion module in the scheme converts system logs, network traffic and other unstructured data into structured feature vectors through an unstructured data preprocessing unit, and eliminates the time deviation of multi-source data through a time series data alignment unit, solves the problem of insufficient processing capacity of traditional platforms for unstructured data, provides comprehensive and accurate feature input for abnormality detection, and simultaneously realizes real-time monitoring of data distribution changes through a concept drift detection submodule, triggers a model incremental update submodule to iteratively update parameters based on real-time data, avoids the decline in detection accuracy caused by data distribution drift of a traditional preset rule library, and significantly enhances the generalization recognition ability and environmental adaptability of the model for unknown abnormal patterns.

[0020] The scheme integrates real-time system states and historical data through a multi-dimensional state space construction unit by using a resource scheduling module and a reinforcement learning algorithm, realizes hierarchical optimization of resource allocation through a hierarchical action decision unit, so that when an abnormality occurs, the operation and maintenance management center dynamically adjusts the resources of affected nodes through the resource scheduling module, shortens the response time of unknown abnormal events, improves the system fault tolerance and recovery efficiency, guarantees the service continuity of a high-availability system, forms a closed-loop system of detection, fusion, optimization and scheduling through the above modules, realizes data interaction and policy linkage through an SDN controller, avoids the problems of model solidification and decision lag in traditional schemes, and improves the intelligent level of operation and maintenance decisions as a whole.

[0021] It is to be understood that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting; it is not intended to exclude myriad other embodiments of the present application that other inventors can develop based on the same general inventive concepts embodied by the described embodiments. That is, although the present application is described in terms of particular embodiments and illustrative figures, it should be apparent that the scope of the present application is not limited to these specific embodiments.

[0022] While the embodiments of the application have been shown and described herein, it will be understood by those skilled in the art that many changes, modifications, substitutions and alterations to these embodiments can be made without departing from the principles and spirits of the application, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An information operation and maintenance intelligent management and control platform system, characterized in that, Comprise: Abnormality detection module, multi-modal feature fusion module, adaptive online learning strategy module, operation and maintenance management center module and resource scheduling module, the abnormality detection module runs on the server of data center, and receives data from each operation and maintenance node through special interface, the abnormality detection module is used to identify the normal behavior mode of system through unsupervised learning mode, and real-time analysis is carried out on the received data by using large-scale historical operation and maintenance data; Multi-modal feature fusion module, for combining system log, network traffic and multiple data sources of hardware state, the adaptive online learning strategy module is executed through SDN controller, the SDN controller is connected with server and each operation and maintenance node, the adaptive online learning strategy module is used for dynamically adjusting the parameters of abnormality detection model according to real-time operation and maintenance data; Operation and maintenance management center module, for starting fault diagnosis process based on alarm information, and calling adaptive online learning strategy to update parameters of abnormality detection model, the resource scheduling module is internally provided with reinforcement learning algorithm, the resource scheduling module is used for dynamically adjusting resource allocation of each node according to current system state and historical operation and maintenance data by using reinforcement learning algorithm; Multi-modal feature fusion module, internally integrated with unstructured data preprocessing unit and time series data alignment unit, the adaptive online learning strategy module is internally provided with concept drift detection submodule and model incremental update submodule, the resource scheduling module is internally integrated with multi-dimensional state space construction unit and hierarchical action decision unit.

2. The information operation and maintenance intelligent management and control platform system according to claim 1, characterized in that: The unstructured data preprocessing unit extracts key features through natural language processing technology, converts unstructured log content into structured feature vector, and performs protocol analysis and field extraction on unstructured fields in network traffic data, so that multi-modal feature fusion module can uniformly process different types of data sources.

3. The information operation and maintenance intelligent management and control platform system of claim 1, wherein: The time series data alignment unit adopts the combination of sliding window and interpolation fitting, and performs space-time alignment processing on multi-source data based on timestamp, the concept drift detection submodule is used for dynamically evaluating the change trend of system operation mode by continuously monitoring the difference between real-time operation and maintenance data and historical training data, and automatically triggering model incremental update submodule when detecting that data distribution deviates significantly.

4. The information operation and maintenance intelligent management and control platform system of claim 1, wherein: The model incremental update submodule updates the abnormality detection model step by step based on the newly added real-time operation and maintenance data by using online learning algorithm, and integrates new data features while retaining historical training knowledge, the multi-dimensional state space construction unit is used for combining real-time system state parameters such as CPU load, memory usage and network bandwidth with behavior patterns in historical operation and maintenance data, to construct multi-dimensional state vector containing current running state, resource occupation and historical trend.

5. The information operation and maintenance digital management and control method based on the information operation and maintenance digital management and control platform system of any one of claims 1-4, characterized in that, Comprise the following steps: S1 data receiving and preprocessing: Firstly, the data received from each operation and maintenance node is received through the abnormality detection module, and the received data is pretreated; S2 feature fusion and alignment: Different types of data sources of system logs, network traffic and hardware status are firstly processed by the unstructured data preprocessing unit to extract key features by using natural language processing technology, and then the time series data alignment unit is used to complete the spatio-temporal alignment of multi-source data by combining sliding window and interpolation fitting based on timestamp. S3 Abnormality detection model training and real-time analysis: The abnormality detection module is used to identify normal behavior patterns of the system through unsupervised learning based on large-scale historical operation and maintenance data, and analyze real-time operation and maintenance data. When an anomaly is detected, an alarm message is generated. S4 Model parameter adaptive online adjustment: Based on the adaptive online learning strategy module, real-time operation and maintenance data are obtained through the SDN controller. When the data distribution is significantly deviated, the model incremental update submodule is automatically triggered, and the abnormality detection model parameters are updated gradually based on the new real-time data and online learning algorithm. S5 Fault diagnosis and resource dynamic scheduling: The fault diagnosis process is started by the operation and maintenance management center module based on the alarm information generated by the abnormality detection module, and the abnormality detection model parameters are updated by the adaptive online learning strategy. The resource scheduling module combines the real-time state parameters of CPU load, memory usage and network bandwidth with historical operation and maintenance data behavior patterns to construct a multi-dimensional state vector containing current operating state, resource occupation and historical trends through the multi-dimensional state space construction unit.

6. The information operation and maintenance digital management and control method according to claim 5, characterized in that: In step S1, the abnormality detection module performs a multi-level verification mechanism on the received raw data, including integrity verification, format compliance check and abnormal value preliminary screening. For unstructured data, the preprocessing unit identifies key operation nodes and state descriptions in the log through semantic analysis technology, and realizes standardized extraction of sensitive fields by combining regular expression matching.

7. The electronic device of the information operation and maintenance digital management and control platform, adopts the information operation and maintenance digital management and control method of any one of claims 5-6, characterized in that, Comprises: A processor, a memory and a communication interface, the processor is used to execute computer programs, the memory is used to store the program instructions of the information operation and maintenance intelligent management and control platform system, the program instructions include the program codes of the abnormality detection module, the multi-modal feature fusion module, the adaptive online learning strategy module, the operation and maintenance management center module and the resource scheduling module; The communication interface is used for data interaction with each operation and maintenance node, receives multi-source operation and maintenance data of system logs, network traffic and hardware status, and sends alarm information and resource scheduling instructions.

8. The electronic device of claim 7, wherein: The processor is a heterogeneous computing architecture, which includes CPU and GPU co-processing units. The CPU in the CPU and GPU co-processing units is used to perform logical control and general computing tasks, while the GPU is used to accelerate natural language processing operations in unstructured data preprocessing and matrix operations in multi-modal feature fusion. The memory adopts a solid state storage and cache combined architecture. The solid state storage is used to persistently store large-scale historical operation and maintenance data and trained abnormality detection models, and the cache is used to temporarily store real-time received operation and maintenance data and intermediate parameters in the model incremental update process.

9. A storage medium of the information operation and maintenance digital management and control platform, based on the electronic device of the information operation and maintenance digital management and control platform of any one of claims 7-8, characterized in that, Comprises: The read card internally stores a computer program, which, when executed by a processor, implements an information operation and maintenance intelligent management and control method, and the computer program includes program codes for implementing the functions of an anomaly detection module, a multi-modal feature fusion module, an adaptive online learning strategy module, an operation and maintenance management center module, and a resource scheduling module. 10.The storage medium of claim 9, wherein: The program code for implementing the function of the multi-modal feature fusion module in the computer program includes an unstructured data preprocessing unit program and a time series data alignment unit program. The unstructured data preprocessing unit program is used to perform semantic analysis on system logs, extract key features through natural language processing technology, and convert them into structured feature vectors. Meanwhile, the unstructured fields in network traffic data are protocol-analyzed, so that different types of data sources are uniformly processed. The time series data alignment unit program combines sliding window and interpolation fitting, and aligns multi-source time series data in space and time based on timestamps.

Images