An intelligent operation and maintenance method, system and device for a heterogeneous system based on an LSTM

Abstract

The application relates to the technical field of disaster intelligent operation and maintenance, and discloses a heterogeneous system intelligent operation and maintenance method, system and device based on an LSTM. The method collects heterogeneous system data of a plurality of data sources; the plurality of heterogeneous system data is classified into first data and second data; a first anomaly detection model is used to perform anomaly prediction on the first data at a center node to generate a first prediction result; a second anomaly detection model is used to perform anomaly prediction on the second data at an edge node to generate a second prediction result; the first prediction result and the second prediction result are combined to determine abnormal data; the abnormal data is synchronously processed based on preset network requirements; and the abnormal data is backed up based on the preset network requirements and the importance of the abnormal data. The application realizes fault prediction and intelligent disaster recovery, and can improve the accuracy and real-time performance of data processing and prediction.

Description

Technical Field

[0001] The present invention relates to the field of disaster recovery intelligent operation and maintenance technology, and in particular to an LSTM-based heterogeneous system intelligent operation and maintenance method, system and device. Background Art

[0002] Distributed heterogeneous systems are complex systems widely used for large-scale data management and processing. These systems primarily consist of multiple subsystems of varying types and functions, distributed across different physical locations within a province. These subsystems, connected through networks and working collaboratively, collectively implement data collection, processing, storage, and transmission. The heterogeneity of distributed heterogeneous systems is reflected in the diversity of their hardware and software, including diverse computing devices (such as servers, edge devices, and sensors), as well as a variety of operating systems and application software. Such system designs fully leverage the computing resources and storage capabilities of diverse devices, enabling efficient data processing and optimized resource allocation. Provincial-level distributed heterogeneous systems present both significant advantages and challenges in practical applications. Their advantage lies in their ability to effectively share computing and storage loads through a distributed architecture, avoiding single points of failure and improving system robustness and scalability. Furthermore, heterogeneous systems can flexibly adapt to diverse data and application requirements, providing efficient resource utilization.

[0003] However, distributed heterogeneous systems currently face challenges such as complex data synchronization and difficult disaster recovery management. Due to differences in data formats and communication protocols between subsystems, achieving data consistency and real-time synchronization presents a major technical challenge. Furthermore, how to quickly restore data and services to ensure system continuity and reliability in the event of a system failure or disaster is another pressing issue for provincial distributed heterogeneous systems. These challenges provide a realistic background and technical basis for the present invention. Therefore, research on an LSTM-based intelligent operation and maintenance system for data synchronization and disaster recovery in provincial distributed heterogeneous systems is of great practical significance. Summary of the Invention

[0004] The present invention provides a LSTM-based intelligent operation and maintenance method, system, and device for heterogeneous systems, which realizes fault prediction and intelligent disaster recovery, and can improve the accuracy and real-time performance of data processing and prediction.

[0005] In order to solve the above technical problems, the present invention provides a heterogeneous system intelligent operation and maintenance method based on LSTM, comprising:

[0006] Collect heterogeneous system data from multiple data sources;

[0007] Classifying the plurality of heterogeneous system data into first data and second data;

[0008] Performing anomaly prediction on the first data using a first anomaly detection model at a central node to generate a first prediction result;

[0009] Performing anomaly prediction on the second data using a second anomaly detection model at the edge node to generate a second prediction result;

[0010] Determining abnormal data by combining the first prediction result and the second prediction result;

[0011] Based on preset network requirements, synchronously process the abnormal data;

[0012] Based on preset network requirements and the importance of the abnormal data, the abnormal data is backed up.

[0013] Furthermore, the collection of heterogeneous system data from multiple data sources is specifically as follows:

[0014] Identifying multiple data sources of heterogeneous systems and establishing connections between the data sources;

[0015] Collecting a number of raw data from each of the data sources;

[0016] Performing format standardization processing on the original data;

[0017] Preprocess the data after format unification;

[0018] Use encrypted transmission protocol to encrypt pre-processed data;

[0019] Determine the encrypted data as heterogeneous system data

[0020] Furthermore, the classifying of the plurality of heterogeneous system data into first data and second data is specifically:

[0021] Performing data size analysis and computational complexity analysis on the data of the plurality of heterogeneous systems;

[0022] Determining heterogeneous system data whose data size does not exceed a preset first data range and whose calculation complexity does not exceed a preset first calculation range as first data;

[0023] The heterogeneous system data whose data size exceeds a preset first data range or whose calculation complexity exceeds a preset first calculation range is determined as the second data.

[0024] Furthermore, the central node uses a first anomaly detection model to perform anomaly prediction on the first data to generate a first prediction result, specifically:

[0025] performing normalization processing on the first data to obtain normalized data;

[0026] Performing feature extraction on the normalized data to obtain first feature data;

[0027] A first anomaly detection model is used to perform anomaly prediction on the first feature data to obtain a first prediction result.

[0028] Furthermore, the first anomaly detection model is specifically:

[0029] Use the LSTM model to learn historical heterogeneous system data and form a time series prediction model;

[0030] Introducing an Attention mechanism into the time series prediction model, training the time series prediction model, and adjusting the attention of the time series prediction model;

[0031] The trained time series prediction model is determined as the first anomaly detection model.

[0032] Furthermore, the edge node uses a second anomaly detection model to perform anomaly prediction on the second data to generate a second prediction result, specifically:

[0033] performing data preprocessing on the second data to obtain preprocessed data;

[0034] Performing feature extraction on the preprocessed data to obtain second feature data;

[0035] A second anomaly detection model is used to perform anomaly prediction on the second feature data to obtain a second prediction result; wherein the second anomaly detection model is an LSTM model.

[0036] Furthermore, the abnormal data is synchronously processed based on the preset network requirements, specifically:

[0037] Detect the real-time network status of heterogeneous systems;

[0038] Determine a data synchronization strategy based on preset network requirements and the real-time network status; wherein the data synchronization strategy includes a timestamp synchronization strategy, an event-driven synchronization strategy, and a scheduled task synchronization strategy;

[0039] Based on the data synchronization strategy, a distributed computing framework is used to synchronize the abnormal data to each system node of the heterogeneous system.

[0040] Furthermore, the abnormal data is backed up based on the preset network requirements and the importance of the abnormal data, specifically:

[0041] Performing importance analysis on the abnormal data to generate importance analysis results;

[0042] Based on the importance analysis result, screening important abnormal data from the abnormal data;

[0043] Determine a data backup strategy based on preset network requirements; wherein the data backup strategy includes full backup and incremental backup;

[0044] Based on the data backup strategy, the abnormal data is backed up to a backup heterogeneous system.

[0045] The present invention provides an LSTM-based intelligent operation and maintenance method for heterogeneous systems, which collects heterogeneous system data from multiple data sources; classifies the multiple heterogeneous system data into first data and second data; uses a first anomaly detection model at a central node to predict anomalies on the first data and generate a first prediction result; uses a second anomaly detection model at an edge node to predict anomalies on the second data and generate a second prediction result; determines the abnormal data by combining the first prediction result and the second prediction result; synchronously processes the abnormal data based on preset network requirements; and backs up the abnormal data based on the preset network requirements and the importance of the abnormal data. The present invention realizes fault prediction and intelligent disaster recovery; combines edge computing technology to delegate some data processing tasks to edge nodes, thereby reducing the load on the central node, reducing network bandwidth occupancy, and improving data transmission efficiency; and improves the accuracy and real-time performance of data processing and prediction.

[0046] The present invention also provides an LSTM-based heterogeneous system intelligent operation and maintenance system for executing the LSTM-based heterogeneous system intelligent operation and maintenance method as described above, comprising: an acquisition module, a classification module, a first prediction module, a second prediction module, a data determination module, a synchronization module, and a backup module;

[0047] The data acquisition module is used to collect heterogeneous system data from various data sources in the heterogeneous system;

[0048] The data processing module is used to perform in-depth analysis and processing on the heterogeneous system data;

[0049] The data synchronization module is used to perform data synchronization processing on the heterogeneous system data;

[0050] The disaster recovery management module is used to perform data backup processing on the heterogeneous system data;

[0051] The edge computing module is used to perform edge computing processing on the heterogeneous system data using edge nodes.

[0052] The present invention provides an LSTM-based intelligent operation and maintenance system for heterogeneous systems, including a data acquisition module, a data processing module, a data synchronization module, a disaster recovery management module, and an edge computing module. By introducing an LSTM network with an Attention mechanism, the accuracy and real-time performance of data processing and prediction are significantly improved. A distributed training and inference framework is adopted to fully utilize computing resources within the province and improve the training and inference efficiency of the model. In combination with edge computing technology, some data processing tasks are delegated to edge nodes, thereby reducing the load on central nodes. The present invention not only achieves fault prediction and intelligent disaster recovery, but also ensures the transparent migration of data and computing services through efficient data management technology, thereby improving the resilience and post-disaster recovery capabilities of heterogeneous systems.

[0053] The present invention also provides a heterogeneous system intelligent operation and maintenance device based on LSTM, comprising: an extraction module, a sending module, a receiving module and a replacement module;

[0054] The acquisition module is used to collect heterogeneous system data from several data sources;

[0055] The classification module is used to classify the plurality of heterogeneous system data into first data and second data;

[0056] The first prediction module is used to perform anomaly prediction on the first data using a first anomaly detection model at the central node to generate a first prediction result;

[0057] The second prediction module is used to perform anomaly prediction on the second data using a second anomaly detection model at the edge node to generate a second prediction result;

[0058] The data determination module is used to determine abnormal data by combining the first prediction result and the second prediction result;

[0059] The synchronization module is used to synchronize the abnormal data based on preset network requirements;

[0060] The backup module is used to perform backup processing on the abnormal data based on preset network requirements and the importance of the abnormal data.

[0061] The present invention provides an LSTM-based intelligent operation and maintenance device for heterogeneous systems. The device collects heterogeneous system data from multiple data sources based on the organic combination of modules; classifies the multiple heterogeneous system data into first data and second data; uses a first anomaly detection model at a central node to predict anomalies on the first data and generate a first prediction result; uses a second anomaly detection model at an edge node to predict anomalies on the second data and generate a second prediction result; determines the abnormal data by combining the first prediction result and the second prediction result; synchronously processes the abnormal data based on preset network requirements; and backs up the abnormal data based on the preset network requirements and the importance of the abnormal data. The present invention realizes fault prediction and intelligent disaster recovery; combines edge computing technology to delegate some data processing tasks to edge nodes, thereby reducing the load on the central node, reducing network bandwidth occupancy, and improving data transmission efficiency; and improves the accuracy and real-time performance of data processing and prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 A flowchart of an embodiment of the LSTM-based intelligent operation and maintenance method for heterogeneous systems provided by the present invention;

[0063] Figure 2 A schematic diagram of the structure of an embodiment of the LSTM-based intelligent operation and maintenance system for heterogeneous systems provided by the present invention;

[0064] Figure 3 A schematic structural diagram of an embodiment of a data acquisition module provided by the present invention;

[0065] Figure 4 A schematic structural diagram of an embodiment of a data processing module provided by the present invention;

[0066] Figure 5 A schematic structural diagram of an embodiment of a data synchronization module provided by the present invention;

[0067] Figure 6 A schematic structural diagram of an embodiment of a disaster recovery management module provided by the present invention;

[0068] Figure 7 A comparison chart of mean square error of prediction accuracy provided by the present invention;

[0069] Figure 8 A model loss mean absolute error convergence diagram provided by the present invention;

[0070] Figure 9 A load distribution comparison diagram provided by the present invention;

[0071] Figure 10 A processing time comparison chart provided by the present invention;

[0072] Figure 11 A fault response time comparison diagram provided by the present invention;

[0073] Figure 12 A data recovery time comparison chart provided by the present invention;

[0074] Figure 13 A system availability comparison chart provided by the present invention;

[0075] Figure 14 A schematic diagram of the structure of an embodiment of the LSTM-based intelligent operation and maintenance device for heterogeneous systems provided by the present invention. DETAILED DESCRIPTION

[0076] 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. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0077] The flowcharts shown in the accompanying drawings are for illustrative purposes only and do not necessarily include all contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may be decomposed, combined, or partially merged, so the actual execution order may vary depending on the actual situation.

[0078] The following embodiments of the present invention are described in detail with reference to the accompanying drawings. In the absence of conflict, the following embodiments and features in the embodiments may be combined with each other.

[0079] Example 1

[0080] See also Figure 1 , is a flow chart of an embodiment of the LSTM-based intelligent operation and maintenance method for heterogeneous systems provided by the present invention. The method includes steps 101 to 107, each of which is specifically as follows:

[0081] Step 101: Collect heterogeneous system data from multiple data sources.

[0082] Furthermore, in the first embodiment of the present invention, heterogeneous system data from several data sources are collected, specifically:

[0083] Identifying multiple data sources of heterogeneous systems and establishing connections between the data sources;

[0084] Collecting a number of raw data from each of the data sources;

[0085] Performing format standardization processing on the original data;

[0086] Preprocess the data after format unification;

[0087] Use encrypted transmission protocol to encrypt pre-processed data;

[0088] The encrypted data is determined to be heterogeneous system data.

[0089] In the first embodiment of the present invention, there are data sources of different types of devices in the heterogeneous system, including sensors, servers, edge computing nodes, databases, etc. When collecting data from the heterogeneous system, first identify the various data sources in the heterogeneous system, establish a stable connection, and collect raw data from different data sources. Based on the different data formats of different data sources, the acquired raw data is converted into a unified internal format for subsequent processing. The collected data may contain noise, incomplete or redundant information, so the raw data needs to be preprocessed. The preprocessing operations include noise filtering, data imputation, and redundancy elimination. Since the processed data needs to be efficiently and reliably transmitted to the data processing module, the data is encrypted to ensure that the data is not tampered with or leaked during transmission, and to ensure the integrity and security of the data during transmission.

[0090] The present invention provides stable and reliable data support for the complex and changeable heterogeneous system environment by collecting data corresponding to multiple data sources in the heterogeneous system, thereby ensuring the efficient operation of the heterogeneous system.

[0091] Step 102: Classify the plurality of heterogeneous system data into first data and second data.

[0092] Furthermore, in the first embodiment of the present invention, the plurality of heterogeneous system data are classified into first data and second data, specifically:

[0093] Performing data size analysis and computational complexity analysis on the data of the plurality of heterogeneous systems;

[0094] Determining heterogeneous system data whose data size does not exceed a preset first data range and whose calculation complexity does not exceed a preset first calculation range as first data;

[0095] The heterogeneous system data whose data size exceeds a preset first data range or whose calculation complexity exceeds a preset first calculation range is determined as the second data.

[0096] In the first embodiment of the present invention, after obtaining the processed heterogeneous system data, in order to effectively improve the efficiency of data processing and the overall performance of the system, the heterogeneous system data is analyzed, and larger data or data that requires complex calculations are placed on the edge nodes. The edge nodes are used to perform preliminary data processing and analysis, thereby reducing the load on the central node and achieving optimal configuration of system resources.

[0097] Step 103: Perform anomaly prediction on the first data using a first anomaly detection model at the central node to generate a first prediction result.

[0098] Furthermore, in the first embodiment of the present invention, the first anomaly detection model is used at the central node to perform anomaly prediction on the first data to generate a first prediction result, specifically:

[0099] performing normalization processing on the first data to obtain normalized data;

[0100] Performing feature extraction on the normalized data to obtain first feature data;

[0101] A first anomaly detection model is used to perform anomaly prediction on the first feature data to obtain a first prediction result.

[0102] In a first embodiment of the present invention, after classifying heterogeneous system data into first data and second data, a central node is used to perform in-depth analysis and processing on the first data to generate valuable information and prediction results. To ensure that data from different data sources is processed at the same scale, the data must first be normalized. To improve the processing efficiency and effectiveness of the subsequent first anomaly detection model, feature extraction is performed on the data to form first feature data. Feature extraction methods include time series features, frequency domain features, and statistical features. Based on the first anomaly detection model, anomaly prediction is performed on the first feature data to obtain a first prediction result. The processed data and the first prediction result are stored in a database.

[0103] Furthermore, in the first embodiment of the present invention, the first anomaly detection model is specifically:

[0104] Use the LSTM model to learn historical heterogeneous system data and form a time series prediction model;

[0105] Introducing an Attention mechanism into the time series prediction model, training the time series prediction model, and adjusting the attention of the time series prediction model;

[0106] The trained time series prediction model is determined as the first anomaly detection model. In the first embodiment of the present invention, the LSTM model is trained using historical heterogeneous system data to generate the first anomaly detection model. First, the LSTM model establishes a time series prediction model by learning from the historical heterogeneous system data. The input of the LSTM model is feature data from the historical heterogeneous system data, and the output is anomaly prediction results. The LSTM model is trained using a backpropagation algorithm to optimize the loss function. To further improve the accuracy of anomaly prediction, an attention mechanism is introduced. The attention mechanism dynamically adjusts the model's focus on different parts of the input data by calculating attention weights. Using the trained LSTM model and the attention mechanism, newly input feature data is predicted to generate a prediction result. Simultaneously, the heterogeneous system detects the actual anomaly corresponding to the prediction result in real time, sets a first threshold, and determines an anomaly when the deviation between the prediction result and the actual anomaly exceeds the first threshold.

[0107] Through processes such as data normalization, feature extraction, LSTM model training, Attention mechanism fusion, data prediction and anomaly detection, the present invention realizes in-depth processing and analysis of heterogeneous system data, improves the accuracy of data processing and prediction, and provides reliable data support for data synchronization and disaster recovery management of heterogeneous systems.

[0108] Step 104: Perform anomaly prediction on the second data using a second anomaly detection model at the edge node to generate a second prediction result.

[0109] Furthermore, in the first embodiment of the present invention, the edge node uses a second anomaly detection model to perform anomaly prediction on the second data to generate a second prediction result, specifically:

[0110] performing data preprocessing on the second data to obtain preprocessed data;

[0111] Performing feature extraction on the preprocessed data to obtain second feature data;

[0112] A second anomaly detection model is used to perform anomaly prediction on the second feature data to obtain a second prediction result; wherein the second anomaly detection model is an LSTM model.

[0113] In a first embodiment of the present invention, the edge node is used to process second data that is large in size or requires complex calculations. First, the edge node preprocesses the second data collected locally, including data cleaning, format conversion, and preliminary analysis. The edge node extracts features from the second data after data preprocessing and converts the second data into second feature data for subsequent analysis and processing. The edge node can train and predict local models, such as using an LSTM model for time series prediction to form a second anomaly detection model. After the model training is completed, the second feature data is detected for anomalies using the second anomaly detection model to generate a second prediction result. The edge node aggregates the processed and analyzed data to form simplified data and transmit it to the central node.

[0114] The present invention utilizes edge nodes to process data with large size or requiring complex calculations, can effectively process and analyze local data, and send simplified data to the central node, reducing the burden on the central node.

[0115] Step 105: Determine abnormal data by combining the first prediction result and the second prediction result.

[0116] In the first embodiment of the present invention, after processing the second data, the edge node transmits the data and the corresponding second prediction result to the central node, so that the central node combines the first prediction result and the second prediction result to obtain the abnormal data in the heterogeneous system data, which facilitates subsequent data synchronization and data backup based on the abnormal data.

[0117] Step 106: Synchronize the abnormal data based on preset network requirements.

[0118] Furthermore, in the first embodiment of the present invention, the abnormal data is synchronously processed based on preset network requirements, specifically:

[0119] Detect the real-time network status of heterogeneous systems;

[0120] Determine a data synchronization strategy based on preset network requirements and the real-time network status; wherein the data synchronization strategy includes a timestamp synchronization strategy, an event-driven synchronization strategy, and a scheduled task synchronization strategy;

[0121] Based on the data synchronization strategy, a distributed computing framework is used to synchronize the abnormal data to each system node of the heterogeneous system.

[0122] In the first embodiment of the present invention, after obtaining the abnormal data, data anomaly processing is performed based on the abnormal data. First, according to the actual needs of the heterogeneous system and the real-time network conditions, a suitable data synchronization strategy is selected. The data synchronization strategy includes timestamp synchronization, event-driven synchronization and scheduled task synchronization. After determining the data synchronization strategy, data synchronization processing is performed based on timestamp synchronization, event-driven synchronization and scheduled task synchronization. Specifically, in the timestamp synchronization strategy, each piece of data is timestamped to ensure data consistency of each node at the same time point. In the event-driven synchronization strategy, it is ensured that the synchronization operation is triggered immediately when the data changes. In the scheduled task synchronization strategy, the system periodically synchronizes incremental data to other nodes. After data synchronization, the synchronized data is stored in the database.

[0123] Optionally, during the data synchronization process, data verification and error correction are required to ensure data consistency and integrity.

[0124] Optionally, in order to improve the efficiency of data synchronization, a distributed computing framework is used to assign data synchronization tasks to various nodes, process them in parallel, reduce synchronization delays, and make full use of computing resources within the province.

[0125] By performing data synchronization processing, the present invention can provide stable and reliable data synchronization support in a complex and changeable heterogeneous system environment, thereby ensuring efficient operation of the system.

[0126] Step 107: Back up the abnormal data based on preset network requirements and the importance of the abnormal data.

[0127] Furthermore, in the first embodiment of the present invention, based on the preset network requirements and the importance of the abnormal data, the abnormal data is backed up, specifically:

[0128] Performing importance analysis on the abnormal data to generate importance analysis results;

[0129] Based on the importance analysis result, screening important abnormal data from the abnormal data;

[0130] Determine a data backup strategy based on preset network requirements; wherein the data backup strategy includes full backup and incremental backup;

[0131] Based on the data backup strategy, the abnormal data is backed up to a backup heterogeneous system.

[0132] In the first embodiment of the present invention, by building a highly available backup and recovery strategy, combined with time series analysis technology and LSTM models, the disaster recovery management capabilities of the system can be improved. After obtaining abnormal data, a reasonable data backup strategy is determined based on the business needs of the heterogeneous system and the data importance of the abnormal data. Among them, the backup strategy includes full backup and incremental backup. Full backup is a regular complete backup of all data, while incremental backup is a backup of changed data. The backup data is stored in a secure backup center and effectively managed. To ensure the security and availability of the data, the backup data needs to be encrypted, compressed and version controlled. When a fault is detected, the present invention can respond quickly, start the fault processing and data recovery mechanism, and ensure the rapid recovery of data and services. Among them, the recovery strategy includes full recovery and incremental recovery. When a fault is detected, the present invention can automatically switch to the backup heterogeneous system containing backup data to ensure business continuity.

[0133] In summary, the first embodiment of the present invention provides a method for intelligent operation and maintenance of heterogeneous systems based on LSTM, which collects heterogeneous system data from several data sources; classifies several heterogeneous system data into first data and second data; uses a first anomaly detection model at a central node to predict anomalies of the first data and generate a first prediction result; uses a second anomaly detection model at an edge node to predict anomalies of the second data and generate a second prediction result; determines abnormal data by combining the first prediction result and the second prediction result; synchronizes abnormal data based on preset network requirements; and backs up abnormal data based on preset network requirements and the importance of abnormal data. The present invention realizes fault prediction and intelligent disaster recovery; combines edge computing technology to delegate part of the data processing tasks to the edge nodes, thereby reducing the load of the central node, reducing the occupancy of network bandwidth, and improving the efficiency of data transmission; and improves the accuracy and real-time performance of data processing and prediction.

[0134] Example 2

[0135] See also Figure 2 , is a structural diagram of an embodiment of an LSTM-based heterogeneous system intelligent operation and maintenance system provided by the present invention, which includes a data acquisition module, a data processing module, a data synchronization module, a disaster recovery management module, and an edge computing module;

[0136] The data acquisition module is used to collect heterogeneous system data from various data sources in the heterogeneous system;

[0137] The data processing module is used to perform in-depth analysis and processing on the heterogeneous system data;

[0138] The data synchronization module is used to perform data synchronization processing on the heterogeneous system data;

[0139] The disaster recovery management module is used to perform data backup processing on the heterogeneous system data;

[0140] The edge computing module is used to perform edge computing processing on the heterogeneous system data using edge nodes.

[0141] In the second embodiment of the present invention, see Figure 3 , is a structural diagram of an embodiment of the data acquisition module provided by the present invention. The data acquisition module needs to process data sources from different types of devices and systems, including sensors, servers, edge computing nodes, databases, etc. The data format and communication protocol of each data source may be different, and the data acquisition module must have strong compatibility and adaptability. To achieve this, the data acquisition module adopts a variety of data acquisition technologies and communication protocols, such as HTTP, MQTT, CoAP, SNMP, etc., and designs a universal data interface to ensure that data can be efficiently and securely transmitted to the data processing module. Among them, the LSTM-based heterogeneous system intelligent operation and maintenance system also includes a data storage module for storing data collected by the data acquisition module.

[0142] The key steps in the data collection process are as follows:

[0143] 1. Data source identification and connection: The data acquisition module first identifies various data sources in the heterogeneous system and establishes a stable connection. Set each data source as S i , whose data transmission protocol is P i The data acquisition module connects to each data source through the protocol adapter and establishes a connection C i :C i =connect(S i ,P i ).

[0144] 2. Data format conversion: Since the data formats of different data sources are different, the data acquisition module needs to convert these data into a unified internal format for subsequent processing. Set the original data as D raw , the target format is D fmt , the conversion function is T, then: D fmt =T(D raw ).

[0145] 3. Data preprocessing: The collected data may contain noise, incomplete or redundant information. The data acquisition module cleans and filters the data through preprocessing steps. The preprocessing steps include noise filtering, data imputation and redundancy elimination. Set the processed data as D clean , the preprocessing function is F, then: Dclean =F(D fmt ).

[0146] 4. Data transmission and storage: In order to ensure that the processed data can be efficiently and reliably transmitted to the data processing module and to ensure the integrity and security of the data during the transmission process, data transmission adopts an encrypted transmission protocol (such as TLS / SSL) to ensure that the data is not tampered with or leaked during the transmission process. The data transmission function is T trans , the data after transmission is D trans , then: D trans =T trans (D clean ).

[0147] 5. Real-time monitoring and exception handling: The data acquisition module needs to monitor the data acquisition process in real time and promptly detect and handle abnormal situations such as connection interruption and data loss. Therefore, by setting the monitoring function M and the exception handling mechanism H, the continuity and reliability of data acquisition are ensured. The relationship between monitoring and exception handling is:

[0148] M(D trans )=status

[0149] if status≠normal,thenH(status)

[0150] Through the above-mentioned data source identification and connection, data format conversion, data preprocessing, data transmission and storage, and real-time monitoring and exception handling, the data acquisition module can provide stable and reliable data support in a complex and changeable heterogeneous system environment to ensure the efficient operation of the system.

[0151] In the second embodiment of the present invention, see Figure 4 , is a structural diagram of an embodiment of the data processing module provided by the present invention. Data processing is the core link of the provincial distributed heterogeneous system data synchronization and disaster recovery intelligent operation and maintenance system. By acquiring the collected and pre-processed data in the data storage module and performing in-depth analysis and processing, valuable information and prediction results are generated. After obtaining the prediction results using the result prediction module, the prediction results are transmitted to the data analysis module for further analysis. The data processing module of the present invention uses an improved LSTM network combined with the Attention mechanism to improve the accuracy of data processing and prediction. The specific steps in the data processing process are:

[0152] 1. Data normalization: To ensure that data from different data sources are processed at the same scale, the data must first be normalized. Set the input data to D trans , the normalized data is D norm , the normalization function is N, then: Where μ and σ are the mean and standard deviation of the data, respectively.

[0153] 2. Feature extraction: In order to improve the processing efficiency and effect of the model, it is necessary to extract features from the data. The feature extraction function is F, and the extracted feature data is F feat , then: F feat =F(D norm ). Among them, feature extraction methods include time series features, frequency domain features and statistical features.

[0154] 3. LSTM model training: The LSTM model builds a time series prediction model by learning from historical data. The input of LSTM is feature data F feat , the output is the prediction result The LSTM model training process is:

[0155] The LSTM model is trained using the back-propagation algorithm to optimize the loss function L, which is defined as:

[0156]

[0157] Where N is the number of samples, Y i is the actual value, is the predicted value.

[0158] 4. Attention mechanism integration: To further improve prediction accuracy, the Attention mechanism is introduced. The Attention mechanism dynamically adjusts the model's focus on different parts of the input data by calculating attention weights.

[0159] 5. LSTM model anomaly detection training: Train the LSTM model based on the Attention mechanism and generate anomaly detection results based on historical data At the same time, the system detects anomalies in the prediction results in real time and sets an anomaly detection function D. When the deviation between the anomaly detection result and the actual value exceeds the threshold ∈, it is judged as an anomaly: if then anomaly detected.

[0160] 6. Result prediction: Based on the trained LSTM model, the newly input data is predicted and the prediction results are generated. The processed data and prediction results are stored in the database and fed back to other modules of the system for subsequent operations. For example, the prediction results are fed back to the data synchronization module and the disaster recovery management module for corresponding processing and adjustment. The storage function is S and the stored data is D store ,but:

[0161] The data processing module realizes in-depth processing and analysis of collected data, improves the accuracy of data processing and prediction, and provides reliable data support for the system's data synchronization module and disaster recovery management.

[0162] In the second embodiment of the present invention, see Figure 5 , is a schematic diagram of the structure of an embodiment of the data synchronization module provided by the present invention. When abnormal data is detected, the data stored in the central node and the data stored in the edge node are synchronized using the data synchronization module. The data synchronization module uses an improved LSTM network and a distributed computing framework to achieve efficient and accurate data synchronization. The specific steps of the data synchronization process are:

[0163] 1. Data synchronization strategy selection: According to the actual needs of the system and the network conditions, select the appropriate data synchronization strategy. Data synchronization strategies include timestamp synchronization, event-driven synchronization, and scheduled task synchronization. Set the timestamp synchronization strategy to S t , the event-driven synchronization strategy is S e , the scheduled task synchronization strategy is S d , then the final synchronization strategy is: S=(S t ,S e ,S d ).

[0164] 2. Timestamp synchronization: In the timestamp synchronization strategy, each piece of data is timestamped to ensure data consistency on each node at the same time point.

[0165] 3. Event-driven synchronization: In the event-driven synchronization strategy, the synchronization operation is triggered immediately when the data changes.

[0166] 4. Scheduled task synchronization: In the scheduled task synchronization strategy, the system regularly synchronizes incremental data to other nodes.

[0167] 5. Data verification and error correction: During the data synchronization process, data verification and error correction are required to ensure data consistency and integrity.

[0168] 6. Application of a distributed computing framework: To improve data synchronization efficiency, this paper uses a distributed computing framework to fully utilize the computing resources within the province. The distributed computing framework distributes data synchronization tasks to various nodes, allowing for parallel processing and reducing synchronization delays.

[0169] 7. Storage and feedback of data synchronization results: The synchronized data needs to be stored in the database, and the synchronization results need to be fed back to other modules of the system.

[0170] The data synchronization module ensures data consistency and integrity through timestamp synchronization, event-driven synchronization, and scheduled task synchronization strategies, combined with data verification and error correction. The data synchronization module can provide stable and reliable data synchronization support in complex and changing heterogeneous system environments, ensuring the efficient operation of the system.

[0171] In the second embodiment of the present invention, see Figure 6 , is a schematic diagram of the structure of an embodiment of the disaster recovery management module provided by the present invention. The disaster recovery management module enhances the system's disaster recovery management capabilities by building a highly available backup and recovery strategy, combining time series analysis technology and the LSTM model. After obtaining an abnormality prediction result from the data analysis module, the disaster recovery response module is activated based on the prediction result, and the response data backup module is also activated. This allows the disaster recovery management module to activate a backup heterogeneous system in the event of an abnormal failure and transmit recovery data to the data recovery module to achieve data recovery. The specific technical implementation steps of the disaster recovery management module are as follows:

[0172] 1. Data backup strategy: Develop a reasonable data backup strategy based on the system's business needs and data importance. The backup strategy includes full backup and incremental backup. Full backup regularly backs up all data, while incremental backup backs up only the changed data. Set the full backup cycle to T f , the incremental backup cycle is T i , the full backup function is B f , the incremental backup function is B i , the backed-up data is:

[0173] 2. Storage and management of backup data: The backup data needs to be stored in a safe backup center and managed effectively. The backup data storage function is S b , the backup data after storage is D store , then: D store =S b (D backup ).

[0174] Among them, backup management includes encryption, compression and version control of backup data to ensure data security and availability.

[0175] 3. Fault prediction and detection: Through time series analysis technology and LSTM model, the system's potential fault points are predicted, and timely detection and warning are given. Set the historical data as H and the fault prediction model as P f , the predicted fault point is F p , then: F p =P f (H).

[0176] The fault detection function is Df , the detected fault point is F d , then: F d =D f (D current ).

[0177] Among them, D current The current system data.

[0178] 4. Fast fault response: When a fault is detected, the disaster recovery management module needs to respond quickly and start the fault handling and data recovery mechanism. The fault handling function is R f , the data after fault processing is D recover , then: D recover =R f (F d ).

[0179] 5. Execution of data recovery strategy: After a disaster or failure occurs, execute the data recovery strategy to ensure the rapid recovery of data and services. The recovery strategy includes full recovery and incremental recovery. The full recovery function is R full , the incremental recovery function is R inc , the restored data is:

[0180] 6. System disaster recovery design: System disaster recovery design includes active / standby switching and multiple redundant configurations. When the primary system fails, it automatically switches to the backup system to ensure business continuity. Set the active / standby switching function to S switch , the system state after switching is S new , then: S new =S switch (S current ).

[0181] Multiple redundant configurations improve system reliability and availability by deploying redundant systems in multiple physical locations.

[0182] 7. Disaster recovery drills and evaluation: Conduct disaster recovery drills regularly to evaluate the effectiveness of the disaster recovery strategy and the system's recovery capabilities. The disaster recovery drill function is T d , the exercise evaluation function is E d , the drill and evaluation results are:

[0183] R drill =T d (S current )

[0184] R eval =E d (R drill )

[0185] The disaster recovery management module achieves efficient and reliable disaster recovery management by formulating reasonable data backup strategies, combining fault prediction and detection, rapid fault response, data recovery strategies and system disaster recovery design.

[0186] In the second embodiment of the present invention, in a distributed heterogeneous system, the introduction of edge computing effectively improves the efficiency of data processing and the overall performance of the system. By delegating part of the data processing tasks to the edge nodes, the load on the central node can be reduced and the optimal allocation of system resources can be achieved. Data processing at the edge nodes is the core task of edge computing. By performing preliminary data processing and analysis at the edge nodes, the computing burden of the central node can be greatly reduced.

[0187] The process steps of the edge computing module are as follows:

[0188] 1. Data preprocessing: The edge node first preprocesses the locally collected data, including data cleaning, format conversion and preliminary analysis. Set the original data as D raw , the preprocessing function is P, then the preprocessed data is: D prep =P(D raw ).

[0189] 2. Feature extraction: The edge node performs feature extraction and converts the original data into feature vectors for subsequent analysis and processing. The feature extraction function is F, and the extracted feature vector is F feat , then: F feat =F(D prep ).

[0190] 3. Local model training and prediction: Edge nodes can perform local model training and prediction, such as using LSTM models for time series prediction. Set the LSTM model to M and the input feature vector to F. feat , the output prediction result is but:

[0191] 4. Preliminary data aggregation: The edge node aggregates the processed and analyzed data to form simplified data for transmission to the central node. The aggregation function is A, and the aggregated data is D agg ,but:

[0192] By performing preliminary data processing at the edge node, the computing and storage pressure of the central node can be significantly reduced. Specifically, the edge node transmits the preprocessed and aggregated data to the central node, reducing the amount of transmitted data; by reducing the amount of transmitted data, the network bandwidth occupancy is reduced, the efficiency of data transmission is improved, and data transmission is optimized. After the central node receives the aggregated data from the edge node, it performs further processing and analysis. The processing tasks of the central node are reduced, and it can focus on complex data analysis and global optimization. Through edge computing, some computing tasks are delegated to the edge node, achieving load balancing of the system. The edge node processes data nearby, improving the real-time response capability of the system. By introducing edge computing, the present invention realizes data preprocessing and local analysis of the edge node, significantly reducing the load on the central node. Through data transmission optimization, task allocation and load balancing, the overall performance and real-time response capability of the system are improved.

[0193] As an example of the second embodiment of the present invention, the technical effect of the present invention can be verified by designing a test. To improve the comprehensiveness and accuracy of the test, the test environment includes hardware configuration and software tools.

[0194] In terms of hardware, the test environment consists of multiple heterogeneous nodes, including high-performance servers, edge computing nodes, and sensor devices. Servers are primarily used for centralized data processing and storage, edge computing nodes for distributed data processing, and sensor devices for data collection. The specific configuration is as follows: 8 high-performance servers, each equipped with two Intel Xeon E5-2690 v4 processors, 128GB of memory, and 1TB of SSD storage; 20 edge computing nodes, each equipped with an Intel Core i7-8550U processor, 16GB of memory, and 256GB of SSD storage; and 50 sensor nodes, distributed across different physical locations, to collect real-time environmental data and device operating status.

[0195] In terms of operating systems, servers and edge computing nodes run the Ubuntu 20.04 LTS operating system, and sensor nodes run an embedded Linux system. The data acquisition software is written in Python and supports multiple protocols (such as HTTP, MQTT, and CoAP) to achieve real-time collection of sensor data. The data processing platform uses the TensorFlow framework to implement an improved LSTM model and Attention mechanism for data processing and prediction. The data synchronization service uses Apache Kafka and Apache Pulsar to achieve real-time synchronization and transmission of data, ensuring data consistency between nodes in the system. The disaster recovery management system uses Ansible and Terraform to write disaster recovery management scripts, automate backup and recovery operations, and ensure high availability of data and services.

[0196] For example, under the simulation parameters set in Table 1, comprehensive performance testing and functional verification are carried out. The test content includes the accuracy of data processing and prediction, the real-time and consistency of data synchronization, the response speed and recovery capability of disaster recovery management, and the optimization effect of edge computing on system load and efficiency. By setting and adjusting these test parameters, we can comprehensively evaluate the performance of this invention in practical applications and verify its effectiveness in improving data synchronization and disaster recovery intelligent operation and maintenance in provincial distributed heterogeneous systems.

[0197] Table 1: Parameter settings

[0198]

[0199] See also Figure 7 , is a comparison chart of the mean squared error of prediction accuracy provided by the present invention. By conducting a detailed comparison of the improved LSTM model (with the Attention mechanism), the traditional LSTM model, and the GRU model in terms of data processing and prediction accuracy, the changing trends of the mean squared error (MSE) of each model during the training process can be obtained. The MSE of the improved LSTM model (with the Attention mechanism) decreases significantly faster than that of the traditional LSTM model and the GRU model in the early stages of training. This shows that the Attention mechanism has a clear advantage in improving the model's attention to important data segments, enabling the model to learn key features in the data more quickly and reduce prediction errors. As the number of training rounds increases, the MSE of the three models all show a trend of gradually decreasing and stabilizing. Among them, the final MSE of the improved LSTM model is the lowest, followed by the traditional LSTM model, and the GRU model has the highest MSE. This further verifies the effectiveness of the Attention mechanism in enhancing the LSTM model's ability to process long sequence data, enabling it to achieve higher prediction accuracy under the same training conditions.

[0200] See also Figure 8, the present invention provides a model loss mean absolute error convergence diagram. By making a detailed comparison of the improved LSTM model (combined with the Attention mechanism), the traditional LSTM model and the GRU model in terms of data processing and prediction accuracy, the changing trend of the mean absolute error (MAE) of each model during the training process can be obtained. The improved LSTM model (combined with the Attention mechanism) always maintains a low MAE during the training process, and the convergence speed is significantly faster than the traditional LSTM model and the GRU model. This shows that the Attention mechanism can not only improve the prediction accuracy of the model, but also accelerate the convergence process of the model and improve the training efficiency. When the number of training rounds reaches 100, the MAE of the improved LSTM model is the lowest, followed by the traditional LSTM model, and the MAE of the GRU model is the highest. The improved LSTM model shows a relatively stable error downward trend throughout the training process, and eventually converges to a lower MAE value, showing its superior performance in processing complex time series data.

[0201] A detailed comparison was made on the changes in load distribution of central nodes and edge nodes before and after the introduction of edge computing, as well as the overall data processing time of the system with and without the introduction of edge computing under different load conditions. It can be concluded that the changes in load distribution of central nodes and edge nodes before and after the introduction of edge computing, as well as the comparison of the overall data processing time of the system with and without the introduction of edge computing under different load conditions, can be obtained.

[0202] See also Figure 9 , is a load distribution comparison chart provided by the present invention, which shows the changes in load distribution of central nodes and edge nodes before and after the introduction of edge computing. Before the introduction of edge computing, the load of the central node was very high, with a load percentage of about 90%, and gradually decreased with the increase of the node number, but the overall load was still heavy. In this case, the load of the central node is too high, which can easily lead to system bottlenecks and performance degradation. After the introduction of edge computing, the load of the central node is significantly reduced, and the load percentage drops to below 30%. At the same time, the load of the edge node remains at a relatively stable level of about 30%. This shows that edge computing effectively disperses part of the data processing tasks to the edge nodes, reduces the burden on the central node, thereby optimizing the load distribution of the system and improving the overall operating efficiency.

[0203] See also Figure 10, is a processing time comparison chart provided by the present invention, showing the overall data processing time comparison of the system under different load conditions, with and without the introduction of edge computing. It can be seen that as the load conditions (number of nodes) increase, the processing time of the centralized data processing method increases significantly, from the initial 100 seconds to about 90 seconds. In contrast, the processing time is significantly reduced after the introduction of edge computing, from the initial 70 seconds to about 50 seconds. In addition, compared with the traditional centralized data processing method, the distributed computing framework based on Hadoop and Spark also has certain advantages in processing time, but it is still higher than the processing time of the introduction of edge computing. Specifically, Hadoop's processing time is reduced from the initial 85 seconds to about 63 seconds, and Spark's processing time is reduced from the initial 80 seconds to about 58 seconds. This result shows that edge computing has significant advantages in processing efficiency. By distributing data processing tasks to edge nodes, the processing time of the central node can be greatly reduced, improving the overall processing efficiency of the system. Compared with traditional centralized data processing methods and other distributed computing frameworks (Hadoop and Spark), edge computing has shorter processing time and better performance.

[0204] See also Figure 11 , is a fault response time comparison chart provided by the present invention, which shows the comparison of fault response time between the traditional disaster recovery management method and the method of the present invention. After a fault occurs, the response time of the traditional disaster recovery management method is relatively long, and as time goes by, the fault response time gradually decreases, but still remains at a high level (approximately between 70 and 90 seconds). In contrast, the fault response time of the method of the present invention is significantly shorter, with a response time between 30 and 50 seconds, and the response time is more stable over time. This shows that the present invention has significant advantages in improving fault response speed, can detect and handle faults more quickly, reduce system downtime, and improve system reliability and availability.

[0205] See also Figure 12 , is a data recovery time comparison chart provided by the present invention, which shows the comparison of data recovery time between traditional disaster recovery management methods and the method of the present invention in different disaster scenarios. The results show that the data recovery time of the traditional disaster recovery management method is relatively long, the recovery time is approximately between 90 and 120 seconds, and it fluctuates greatly with changes in the disaster scenario. The data recovery time of the method of the present invention is significantly shorter, the recovery time is between 40 and 60 seconds, and it shows higher stability in different disaster scenarios. This shows that the present invention has significant advantages in reducing data recovery time, can restore system data more quickly, ensure data integrity and consistency, and improve the system's post-disaster recovery capabilities.

[0206] See also Figure 13 , is a system availability comparison chart provided by the present invention, which shows the availability comparison between the traditional disaster recovery management method and the method of the present invention when the system is running for a long time. It can be seen that the system availability of the traditional disaster recovery management method is relatively low, approximately between 90% and 95%, and the availability gradually increases with the passage of time, but the improvement is small. In contrast, the system availability of the method of the present invention is significantly higher, approximately between 95% and 99.8%, and it shows higher stability and consistency during long-term operation. This shows that the present invention has significant advantages in improving system reliability and availability, can better ensure the continuous operation of the system, reduce the risk of system downtime, and improve user experience and service quality.

[0207] In summary, the second embodiment of the present invention provides an intelligent operation and maintenance system for heterogeneous systems based on LSTM, including a data acquisition module, a data processing module, a data synchronization module, a disaster recovery management module and an edge computing module. By introducing the LSTM network with the Attention mechanism, the accuracy and real-time performance of data processing and prediction are significantly improved; a distributed training and reasoning framework is adopted to make full use of the computing resources within the province to improve the training and reasoning efficiency of the model; combined with edge computing technology, some data processing tasks are delegated to edge nodes, thereby reducing the load on the central node. The present invention not only realizes fault prediction and intelligent disaster recovery, but also ensures the transparent migration of data and computing services through efficient data management technology, thereby improving the resilience and post-disaster recovery capabilities of heterogeneous systems.

[0208] Example 3

[0209] See also Figure 14 , is a schematic structural diagram of an embodiment of an LSTM-based intelligent operation and maintenance device for heterogeneous systems provided by the present invention, the device comprising a collection module 201, a classification module 202, a first prediction module 203, a second prediction module 204, a data determination module 205, a synchronization module 206, and a backup module 207;

[0210] The acquisition module 201 is used to collect heterogeneous system data from several data sources;

[0211] The classification module 202 is used to classify the plurality of heterogeneous system data into first data and second data;

[0212] The first prediction module 203 is configured to perform anomaly prediction on the first data using a first anomaly detection model at a central node to generate a first prediction result;

[0213] The second prediction module 204 is configured to perform anomaly prediction on the second data using a second anomaly detection model at the edge node to generate a second prediction result;

[0214] The data determination module 205 is used to determine abnormal data by combining the first prediction result and the second prediction result;

[0215] The synchronization module 206 is used to synchronize the abnormal data based on the preset network requirements;

[0216] The backup module 207 is configured to perform backup processing on the abnormal data based on preset network requirements and the importance of the abnormal data.

[0217] Furthermore, in a third embodiment of the present invention, heterogeneous system data from several data sources is collected, specifically:

[0218] Identifying multiple data sources of heterogeneous systems and establishing connections between the data sources;

[0219] Collecting a number of raw data from each of the data sources;

[0220] Performing format standardization processing on the original data;

[0221] Preprocess the data after format unification;

[0222] Use encrypted transmission protocol to encrypt pre-processed data;

[0223] Determine the encrypted data as heterogeneous system data

[0224] Furthermore, in a third embodiment of the present invention, a plurality of heterogeneous system data are classified into first data and second data, specifically:

[0225] Performing data size analysis and computational complexity analysis on the data of the plurality of heterogeneous systems;

[0226] Determining heterogeneous system data whose data size does not exceed a preset first data range and whose calculation complexity does not exceed a preset first calculation range as first data;

[0227] The heterogeneous system data whose data size exceeds a preset first data range or whose calculation complexity exceeds a preset first calculation range is determined as the second data.

[0228] Furthermore, in a third embodiment of the present invention, a first anomaly detection model is used at a central node to perform anomaly prediction on the first data to generate a first prediction result, specifically:

[0229] performing normalization processing on the first data to obtain normalized data;

[0230] Performing feature extraction on the normalized data to obtain first feature data;

[0231] A first anomaly detection model is used to perform anomaly prediction on the first feature data to obtain a first prediction result.

[0232] Furthermore, in the third embodiment of the present invention, the first anomaly detection model is specifically:

[0233] Use the LSTM model to learn historical heterogeneous system data and form a time series prediction model;

[0234] Introducing an Attention mechanism into the time series prediction model, training the time series prediction model, and adjusting the attention of the time series prediction model;

[0235] The trained time series prediction model is determined as the first anomaly detection model.

[0236] Furthermore, in a third embodiment of the present invention, an anomaly prediction is performed on the second data using a second anomaly detection model at the edge node to generate a second prediction result, specifically:

[0237] performing data preprocessing on the second data to obtain preprocessed data;

[0238] Performing feature extraction on the preprocessed data to obtain second feature data;

[0239] A second anomaly detection model is used to perform anomaly prediction on the second feature data to obtain a second prediction result; wherein the second anomaly detection model is an LSTM model.

[0240] Furthermore, in a third embodiment of the present invention, the abnormal data is synchronously processed based on preset network requirements, specifically:

[0241] Detect the real-time network status of heterogeneous systems;

[0242] Determine a data synchronization strategy based on preset network requirements and the real-time network status; wherein the data synchronization strategy includes a timestamp synchronization strategy, an event-driven synchronization strategy, and a scheduled task synchronization strategy;

[0243] Based on the data synchronization strategy, a distributed computing framework is used to synchronize the abnormal data to each system node of the heterogeneous system.

[0244] Furthermore, in a third embodiment of the present invention, the abnormal data is backed up based on preset network requirements and the importance of the abnormal data, specifically:

[0245] Performing importance analysis on the abnormal data to generate importance analysis results;

[0246] Based on the importance analysis result, screening important abnormal data from the abnormal data;

[0247] Determine a data backup strategy based on preset network requirements; wherein the data backup strategy includes full backup and incremental backup;

[0248] Based on the data backup strategy, the abnormal data is backed up to a backup heterogeneous system.

[0249] In summary, the third embodiment of the present invention provides a heterogeneous system intelligent operation and maintenance device based on LSTM, which collects heterogeneous system data from several data sources based on the organic combination of modules; classifies several heterogeneous system data into first data and second data; uses a first anomaly detection model at the central node to predict anomalies of the first data and generate a first prediction result; uses a second anomaly detection model at the edge node to predict anomalies of the second data and generate a second prediction result; determines abnormal data by combining the first prediction result and the second prediction result; synchronizes the abnormal data based on preset network requirements; and backs up the abnormal data based on preset network requirements and the importance of the abnormal data. The present invention realizes fault prediction and intelligent disaster recovery; combines edge computing technology to delegate part of the data processing tasks to the edge nodes, thereby reducing the load of the central node, reducing the occupancy of the network bandwidth, and improving the efficiency of data transmission; and improves the accuracy and real-time performance of data processing and prediction.

[0250] The specific embodiments described above further illustrate the objectives, technical solutions, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. An intelligent operation and maintenance method for heterogeneous systems based on LSTM, characterized in that: include: Collect heterogeneous system data from multiple data sources; classifying the plurality of heterogeneous system data into first data and second data; Performing anomaly prediction on the first data using a first anomaly detection model at a central node to generate a first prediction result; Performing anomaly prediction on the second data using a second anomaly detection model at the edge node to generate a second prediction result; Determining abnormal data by combining the first prediction result and the second prediction result; Based on preset network requirements, synchronously process the abnormal data; Based on preset network requirements and the importance of the abnormal data, the abnormal data is backed up.

2. The LSTM-based intelligent operation and maintenance method for heterogeneous systems according to claim 1 is characterized in that: The collection of heterogeneous system data from multiple data sources is specifically as follows: Identifying multiple data sources of heterogeneous systems and establishing connections between the data sources; Collecting a number of raw data from each of the data sources; Performing format standardization processing on the original data; Preprocess the data after format unification; Use encrypted transmission protocol to encrypt pre-processed data; The encrypted data is determined to be heterogeneous system data.

3. The LSTM-based intelligent operation and maintenance method for heterogeneous systems according to claim 2 is characterized in that: The classifying of the plurality of heterogeneous system data into first data and second data is specifically: Performing data size analysis and computational complexity analysis on the data of the plurality of heterogeneous systems; Determining heterogeneous system data whose data size does not exceed a preset first data range and whose calculation complexity does not exceed a preset first calculation range as first data; The heterogeneous system data whose data size exceeds a preset first data range or whose calculation complexity exceeds a preset first calculation range is determined as the second data.

4. The LSTM-based intelligent operation and maintenance method for heterogeneous systems according to claim 3 is characterized in that: The first anomaly detection model is used at the central node to perform anomaly prediction on the first data to generate a first prediction result, specifically: performing normalization processing on the first data to obtain normalized data; Performing feature extraction on the normalized data to obtain first feature data; A first anomaly detection model is used to perform anomaly prediction on the first feature data to obtain a first prediction result.

5. The LSTM-based intelligent operation and maintenance method for heterogeneous systems according to claim 4 is characterized in that: The first anomaly detection model is specifically: Use the LSTM model to learn historical heterogeneous system data and form a time series prediction model; Introducing an Attention mechanism into the time series prediction model, training the time series prediction model, and adjusting the attention of the time series prediction model; The trained time series prediction model is determined as the first anomaly detection model.

6. The LSTM-based intelligent operation and maintenance method for heterogeneous systems according to claim 3 is characterized in that: The edge node uses the second anomaly detection model to perform anomaly prediction on the second data to generate a second prediction result, specifically: performing data preprocessing on the second data to obtain preprocessed data; Performing feature extraction on the preprocessed data to obtain second feature data; A second anomaly detection model is used to perform anomaly prediction on the second feature data to obtain a second prediction result; wherein the second anomaly detection model is an LSTM model.

7. The LSTM-based intelligent operation and maintenance method for heterogeneous systems according to claim 1, characterized in that: The synchronization processing of the abnormal data based on the preset network requirements is specifically as follows: Detect the real-time network status of heterogeneous systems; Determine a data synchronization strategy based on preset network requirements and the real-time network status; wherein the data synchronization strategy includes a timestamp synchronization strategy, an event-driven synchronization strategy, and a scheduled task synchronization strategy; Based on the data synchronization strategy, a distributed computing framework is used to synchronize the abnormal data to each system node of the heterogeneous system.

8. The LSTM-based intelligent operation and maintenance method for heterogeneous systems according to claim 1, characterized in that: The backup process of the abnormal data based on the preset network requirements and the importance of the abnormal data is specifically as follows: Performing importance analysis on the abnormal data to generate importance analysis results; Based on the importance analysis result, screening important abnormal data from the abnormal data; Determine a data backup strategy based on preset network requirements; wherein the data backup strategy includes full backup and incremental backup; Based on the data backup strategy, the abnormal data is backed up to a backup heterogeneous system.

9. An LSTM-based intelligent operation and maintenance system for heterogeneous systems, characterized by: Used to execute the LSTM-based intelligent operation and maintenance method for heterogeneous systems according to any one of claims 1 to 8, comprising: a data acquisition module, a data processing module, a data synchronization module, a disaster recovery management module and an edge computing module; The data acquisition module is used to collect heterogeneous system data from various data sources in the heterogeneous system; The data processing module is used to perform in-depth analysis and processing on the heterogeneous system data; The data synchronization module is used to perform data synchronization processing on the heterogeneous system data; The disaster recovery management module is used to perform data backup processing on the heterogeneous system data; The edge computing module is used to perform edge computing processing on the heterogeneous system data using edge nodes.

10. An intelligent operation and maintenance device for heterogeneous systems based on LSTM, characterized in that: include: An acquisition module, a classification module, a first prediction module, a second prediction module, a data determination module, a synchronization module, and a backup module; The acquisition module is used to collect heterogeneous system data from several data sources; The classification module is used to classify the plurality of heterogeneous system data into first data and second data; The first prediction module is used to perform anomaly prediction on the first data using a first anomaly detection model at the central node to generate a first prediction result; The second prediction module is used to perform anomaly prediction on the second data using a second anomaly detection model at the edge node to generate a second prediction result; The data determination module is used to determine abnormal data by combining the first prediction result and the second prediction result; The synchronization module is used to synchronize the abnormal data based on preset network requirements; The backup module is used to perform backup processing on the abnormal data based on preset network requirements and the importance of the abnormal data.

Images