Backup method for computer system and computer system

Abstract

The invention relates to a computer system data backup technology, and discloses a dynamic high-reliability verifiable backup method and system. Multi-dimensional features such as data entropy and access frequency are extracted and classified into key, important and common data through machine learning; the hierarchical backup strategy comprises key data real-time increment, important data dynamic partitioning and common data period full amount, and edge node loads are dispatched in combination with federal learning; pBFT consensus edge cluster distributed storage is adopted, a backup data hash uplink evidence is stored, and the consistency is regularly verified by a smart contract; parallel processing is carried out according to priorities during recovery, and the integrity is verified through on-chain hash. The problems of static strategies, single-point faults, integrity loss and efficiency are solved, the data reliability is improved, the recovery efficiency and compliance are improved, and the method adapts to high-security scenes such as finance and medical treatment.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer system data backup, in particular to a backup method for a computer system and a computer system. BACKGROUND

[0002] With the deepening of digital transformation, computer systems have become the core support in key fields such as finance, medical treatment, and government affairs. The amount of data stored by computer systems is growing exponentially, and the data correlation and real-time requirements are significantly improving. For example, financial transaction systems need to ensure real-time backup and traceability of each transaction data, and medical systems need to ensure long-term safe storage of patient medical record data. Such scenarios have put forward very high requirements for the reliability, dynamic adaptability, and integrity verification of backup technology. Current mainstream backup technology mainly focuses on data transmission and storage optimization, trying to balance backup efficiency and resource consumption by adjusting backup periods and compressing data volumes, but has not yet formed an integrated solution combining the characteristics of data itself and the advantages of distributed architecture, making it difficult to fully meet the multi-dimensional needs of complex scenarios.

[0003] The existing computer system backup method has obvious defects in actual application and cannot meet the high requirements of key fields: firstly, the static backup strategy has significant limitations. Traditional full backup and incremental backup often use fixed periods (such as daily full backup and hourly incremental backup), which cannot dynamically adjust backup priorities according to data value, resulting in high-value data (such as real-time transaction records) being at risk of loss due to backup delays, while low-value data (such as temporary cache files) is excessively backed up, occupying a large amount of storage resources; secondly, the centralized storage architecture has single-point failure risks. Existing solutions often rely on a single cloud server or local storage array, which is vulnerable to network attacks, hardware failures, or natural disasters, leading to the loss of backup data as a whole, and the recovery process relies on a single node, with weak fault tolerance; thirdly, the integrity verification mechanism is missing. The authenticity of existing backup data is often verified by local hash comparison, lacking cross-node and tamper-proof evidence, making it difficult to detect data tampering and meet the data tracing requirements of high-security scenarios; fourthly, the storage efficiency is low. Although block-level incremental backup can reduce the amount of data transmitted at a time, it does not optimize the blocking strategy based on the intrinsic characteristics of data, resulting in repeated storage of the same type of data blocks, high redundancy, and increased storage costs and management complexity. SUMMARY

[0004] Therefore, the present application aims to provide a dynamic, highly reliable, and verifiable backup method and system, which optimizes backup efficiency through intelligent strategies driven by data characteristics, realizes distributed storage by combining blockchain and edge computing, and solves the problems of static strategies, single-point risks, integrity verification, and storage efficiency in existing technologies.

[0005] The technical solution of the embodiment of the present application is as follows:

[0006] A backup method for a computer system, comprising the following steps:

[0007] S1. Multi-dimensional feature extraction is performed on target data in the computer system, and the features include at least data entropy, access frequency and business correlation degree;

[0008] S2. Based on the features, a machine learning model is used to classify the importance of the target data to obtain key data, important data and ordinary data;

[0009] S3. According to the classification results, a dynamic backup strategy is generated, wherein the key data is backed up in real time, the important data is dynamically divided and backed up based on the entropy value, and the ordinary data is periodically fully backed up;

[0010] S4. Backup data shards are stored in an edge node cluster, and the hash values of each shard are written into a blockchain to form an unalterable evidence record;

[0011] S5. Based on a smart contract of the blockchain, the consistency of the backup data and the evidence record is verified periodically, and if it is inconsistent, an abnormal processing mechanism is triggered.

[0012] Preferably, in step S1, the data entropy is calculated by the formula , wherein p(x i ) is the probability distribution of the data block x i , and the larger the value of H(X) is, the higher the data uncertainty is.

[0013] Preferably, in step S2, the machine learning model is a lightweight random forest model, which learns the mapping relationship between data features and importance through training samples, and the training samples contain feature labels and business priority labels of historical data.

[0014] Preferably, in step S3, the block size of the dynamic block backup based on the entropy value is determined by the formula block size = a·H(X)+β, wherein a and β are preset empirical parameters, and H(X) is the entropy value of the important data calculated in step S1.

[0015] Preferably, in step S4, the edge node cluster uses a PBFT consensus mechanism, and when any node with a survival rate of not less than 30% is alive, the complete backup data can be recovered through shard reorganization, and the shards stored by each node are dynamically allocated through a federated learning model to achieve load balancing.

[0016] Preferably, in step S5, the smart contract has a preset verification period and an abnormal processing rule, and the abnormal processing mechanism includes sending an alarm to an administrator, automatically starting a data repair process, and writing the processing result into the blockchain evidence.

[0017] Preferably, the data recovery step is further included: after receiving the recovery request, the key data is recovered according to the data importance level, and the important data and the ordinary data are recovered in parallel; after the recovery is completed, the hash value of the recovered data is calculated and compared with the blockchain storage record, and the recovery is confirmed to be successful if the verification is passed.

[0018] Preferably, a backup system for a computer system comprises:

[0019] The feature extraction module is configured to extract data entropy values, access frequencies and business correlation degrees from the target data.

[0020] The classification module is connected to the feature extraction module and classifies the data according to importance through a machine learning model.

[0021] The strategy generation module is connected to the classification module and generates corresponding backup strategies according to the classification results.

[0022] The distributed storage module is connected to the strategy generation module and includes an edge node cluster and a blockchain node, and is configured to store data shards and hash storage records.

[0023] The verification module is connected to the distributed storage module and realizes consistency verification of the backup data and the storage record through a smart contract.

[0024] Preferably, in the distributed storage module, the load scheduling of the edge node cluster is realized based on a federated learning model, the model aggregates local training parameters of each node to generate a global scheduling strategy, and the local data of the node is not disclosed.

[0025] Preferably, the recovery module is connected to the distributed storage module and the verification module, and is configured to recover key data according to the data importance level, and verify the integrity of the recovered data through hash comparison.

[0026] The embodiments of the present application have the following advantages due to the use of the above technical solutions:

[0027] I. Reliability is significantly improved: the distributed storage architecture avoids the single point failure risk of centralized storage, and combined with the multi-node redundancy design, the data recovery success rate is greatly improved, meeting the high requirements of key fields for data security.

[0028] II. Backup and recovery efficiency is optimized: through data importance layering and dynamic strategy, key data can be efficiently synchronized, and the recovery process is processed in parallel according to priority, effectively shortening the data recovery time and improving the overall business continuity.

[0029] III. Data integrity and compliance assurance: The blockchain storage mechanism provides tamper-proof timestamps and hash proofs for backup data, meeting the requirements of GDPR, etc. for data traceability and authenticity, and is suitable for high-security scenarios such as finance, healthcare, and government affairs.

[0030] IV. Storage resource utilization improvement: Dynamic block strategy based on data entropy and hash deduplication technology are combined to reduce redundant data storage, while federated learning load scheduling reduces edge node resource waste, balancing economy and environmental protection.

[0031] The above summary is intended to illustrate only and is not intended to be limiting in any way. Further aspects, embodiments and features of the present application will be apparent from the drawings and the following detailed description. BRIEF DESCRIPTION OF DRAWINGS

[0032] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or prior art description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and those skilled in the art can obtain other drawings according to these drawings without creative labor.

[0033] Fig. 1 The backup method flowchart of the present application is shown in the figure.

[0034] Fig. 2 The system module architecture diagram of the present application is shown in the figure.

[0035] Fig. 3 The dynamic block logic diagram of the present application is shown in the figure. DETAILED DESCRIPTION

[0036] In the following, only some exemplary embodiments are simply described. As those skilled in the art can recognize, the described embodiments can be modified in various different ways without departing from the spirit or scope of the present application. Therefore, the drawings and description are considered to be exemplary in nature rather than limiting.

[0037] It should be noted that the terms "first", "second", "symmetric", "array" and the like are only used for the purpose of distinguishing description and position description, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined with "first", "symmetric" and the like can explicitly or implicitly include one or more of the features; similarly, for some features that are not limited in number by the words "two", "three" and the like, it should be noted that the features also belong to explicitly or implicitly including one or more feature quantities.

[0038] In the present application, unless otherwise explicitly specified and limited, the terms "mounting", "connecting", "fixing" and the like should be understood in a broad sense; for example, it can be fixed connection, or detachable connection, or integral molding; it can be mechanical connection, it can be direct connection, it can be welding, it can be indirect connection through an intermediate medium, it can be internal communication of two elements or interaction relationship between two elements. For those skilled in the art, the specific meaning of the above terms in the present application can be understood according to the specification and drawings in combination with specific circumstances.

[0039] The embodiments of the present application will be described in detail below in combination with the drawings.

[0040] As Figs. 1-3 shown, the present application provides a backup method for a computer system, comprising the following steps: S1. Multi-dimensional feature extraction is performed on target data in the computer system, the features at least including data entropy value, access frequency and business correlation degree, the data entropy value in step S1 is calculated by the formula , wherein p(x i ) is the data block x iS1. Calculate the entropy value of the target data based on the feature, the greater the H(X) value represents the higher the data uncertainty; S2. Classify the importance of the target data based on the feature through a machine learning model to obtain key data, important data and ordinary data, the machine learning model in step S2 is a lightweight random forest model, and the mapping relationship between the data feature and the importance is learned through training samples, and the training samples include feature labels and business priority labels of historical data; S3. Generate a dynamic backup strategy according to the classification result, wherein the key data adopts real-time incremental backup, the important data adopts dynamic block backup based on the entropy value, and the ordinary data adopts periodic full backup, and the block size of the dynamic block backup based on the entropy value in step S3 is determined by the formula block size = a H(X) + b, wherein a and b are preset empirical parameters, and H(X) is the entropy value of the important data calculated in step S1; S4. Store the backup data shards to an edge node cluster, and write the hash values of each shard into a block chain to form an unalterable evidence record, the edge node cluster in step S4 adopts a PBFT consensus mechanism, and when any node with a survival rate of not less than 30% is alive, the complete backup data can be recovered through shard recombination, and the shards stored by each node are dynamically allocated through a federated learning model to realize load balancing; S5. Based on a block chain smart contract, the consistency of the backup data and the evidence record is verified periodically, and if the consistency is inconsistent, an abnormal processing mechanism is triggered, the smart contract in step S5 presets a verification period and an abnormal processing rule, the abnormal processing mechanism includes sending an alarm to an administrator, automatically starting a data repair process, and writing the processing result into the block chain evidence. Further comprising a data recovery step: after receiving a recovery request, the key data is recovered according to the data importance level, and the important data and the ordinary data are recovered in parallel; after the recovery is completed, the hash value of the recovered data is calculated and compared with the block chain evidence record, and if the verification is passed, the recovery is confirmed to be successful.

[0041] Break through the limitations of traditional backup "static rules, single point dependence": dynamic feature collection allows accurate identification of high-value data (such as financial transactions and medical records), intelligent classification replaces the lag of manual rules, dynamic block + deduplication reduces storage redundancy by 50%, PBFT consensus + block chain evidence reduces the risk of data loss to below 0.01%, and smart contract self-healing realizes a "problem automatic discovery-repair-audit" closed loop, which solves the industry pain points of resource mismatch, low reliability, high cost and audit deficiency from the root.

[0042] A backup system for a computer system, comprising:

[0043] The feature extraction module is used for extracting data entropy values, access frequencies and service correlation degree features from target data, integrates kernel probes (capture file system / database metadata), network traffic analyzers (identify service data transmission paths), and offline feature engines (process historical archived data), supports concurrent collection of multi-source data, and dynamically extends interfaces through API to adapt to feature definition rules of new service systems (such as government document circulation priority features);

[0044] The classification module is connected to the feature extraction module, and the importance of data is classified through a machine learning model. A lightweight random forest model is used to realize “offline batch training + online incremental learning”. In the offline stage, the basic model is trained based on historical labeled data, and in the online stage, new business rules (such as newly added medical diagnosis and treatment links priority) and data feature changes (such as temporary data of sudden public health events) are absorbed in real time. The classification result is pushed through the MQTT message queue with low delay;

[0045] The policy generation module is connected to the classification module, and generates corresponding backup strategies according to the classification results. A hierarchical strategy engine is built in, which maps the classification results and backup rules (such as RDMA high-speed synchronization of key data and dynamic block parameters of important data), and dynamically adjusts the backup task allocation combined with the output of the federated learning load scheduling model (real-time load of each edge node). At the same time, a visual interface configuration is supported, and administrators can customize backup period, redundancy and other parameters to adapt to business changes;

[0046] The distributed storage module is connected to the policy generation module and includes an edge node cluster and a blockchain node, which is used for storing data shards and hash records. The edge node cluster adopts a PBFT consensus mechanism, and data is split into encrypted shards according to the Shamir algorithm. Through the federated learning model, the node load characteristics are aggregated (homomorphic encryption protects node privacy), and global scheduling is realized. It is compatible with SSD (key data), HDD (ordinary data) and other heterogeneous storage media, which reduces the storage cost. The blockchain node receives shard hash (including timestamp and data type), and stores it through smart contract;

[0047] The verification module is connected to the distributed storage module, and realizes consistency verification of backup data and storage records through smart contract. The preset time verification (daily full amount) and trigger verification (before recovery / audit) rules are set, and when an exception occurs, redundant shard repair data is automatically called (key data is repaired first), and the verification / repaired result is written into the blockchain to form an audit closed loop;

[0048] Also includes a recovery module connected to the distributed storage module and the verification module, for restoring key data according to the importance level of the data, and verifying the integrity of the restored data through hash comparison, supporting priority parallel recovery (key data second-level hot backup, multi-node parallel transmission), breakpoint resume (recording recovery progress, subsequent transmission after interruption), version rollback (relying on the blockchain timeline to restore historical versions), automatic triggering of hash verification after recovery, and confirmation of successful recovery by the party;

[0049] Through the cooperation of the six modules of "feature-classification-strategy-storage-verification-recovery", a modular, scalable and self-healing backup system is constructed: multi-source feature collection covers complex scenarios such as finance, medical care and government affairs, dual-mode classification adapts to data dynamic changes, visualized strategy reduces the operation and maintenance threshold, distributed storage + heterogeneous storage balances safety and cost, smart contract self-healing realizes "anti-tampering + self-repair", and priority recovery ensures business continuity. From the architectural level, the problems of traditional solutions such as "strong scene limitation, high strategy rigidity, large resource waste and low recovery efficiency" are solved, and full-process controllable and full-cycle auditable backup capabilities are provided for key areas.

[0050] In this embodiment, the specific application in operation:

[0051] First, data feature collection and preprocessing are performed: through a kernel-level data probe, file system changes, database transactions and memory data fluctuations of a computer system are captured in real time, and a stream processing engine is combined to extract data entropy values (based on a sliding time window to calculate data block access probability, through a formula , the higher the entropy value, the stronger the data uncertainty), access frequency (introducing a time decay function W(t) = e -kt quantifying recent access heat, distinguishing between high-frequency business data and low-frequency archived data), business correlation (relying on a pre-defined business topology map to match the priority label of the data belonging to the business link), forming a standardized feature vector; synchronous filtering of noise data (such as invalid system logs, duplicate cache files) to ensure feature quality.

[0052] Then, intelligent classification and strategy generation are performed: the pre-processed feature vector is input into a lightweight random forest model, which adopts an "offline batch training + online incremental learning" mechanism:

[0053] Offline phase: based on the feature labels (entropy value, access frequency) and business priority labels (such as manual labeling of financial transactions and medical records) of historical data, a basic classification model is trained to cover typical business scenarios;

[0054] Online stage: Real-time absorption of new business rules (such as the addition of government document circulation priority) and data feature changes (such as temporary data blocks for sudden access), continuously optimize classification logic through incremental learning, dynamically output key data (millisecond-level backup required), important data (minute-level backup required), and ordinary data (hour-level backup required) classification results, and push them to the backup strategy module through the MQTT message queue with low delay.

[0055] Strategy module calls hierarchical backup rules:

[0056] Key data (such as real-time financial transaction records): Trigger RDMA high-speed direct channel to achieve microsecond-level real-time incremental backup of data changes, and synchronize log real-time writing to the blockchain pending chain queue;

[0057] Important data (such as medical record change logs): Perform dynamic chunking backup according to the formula block size = a * H(X) + b (a / b optimized through financial and medical scene simulation), the higher the entropy value (the more complex the data), the smaller the block (fine-grained backup); At the same time, combined with the CDCChunking content chunking algorithm, identify duplicate data blocks and upload only new shards;

[0058] Ordinary data (such as system temporary cache): Start periodic full backup, rely on SHA-256 hash deduplication technology, only store the index of unique data blocks, and reduce storage redundancy.

[0059] At the same time, the federated learning load scheduling model aggregates the storage utilization, network bandwidth, CPU load, and other features of each edge node (parameter aggregation uses homomorphic encryption to protect node data privacy), generates a global backup task allocation strategy, avoids single-point overload, and realizes cross-node load balancing.

[0060] Then complete distributed storage and blockchain evidence:

[0061] Data sharding storage: Backup data is split into N encrypted shards (N ≥ 5, meeting the PBFT consensus node number requirement) according to the Shamir secret sharing algorithm, and is dynamically allocated to a heterogeneous edge node cluster compatible with SSD (key data, priority low latency) and HDD (ordinary data, priority large capacity) according to data importance; Edge nodes ensure shard consistency through PBFT consensus mechanism, and any ≥ Node survival can restore complete data;

[0062] Hash chain storage: The hash value (including timestamp, data type identifier) of each shard is written into the blockchain consortium chain through a smart contract, forming an unalterable backup timeline; Edge nodes synchronize shard status through Gossip protocol to ensure global consistency.

[0063] Finally trigger verification, self-healing and recovery process:

[0064] Smart contract verification: preset daily full verification, business audit directional verification rules, and regular traversal of on-chain hash and edge node shard hash comparison. If inconsistencies are detected, automatic triggering of hierarchical exception handling:

[0065] Key data exception: immediately send SMS + email alerts, and preferentially call redundant shards to start repair processes (such as financial transaction data repair priority over ordinary logs);

[0066] Repair record evidence: write verification process and repair results into the blockchain in real time to form a closed-loop audit chain.

[0067] Priority recovery: after receiving the recovery request, the priority recovery engine schedules according to importance level:

[0068] Key data: through preloading hot standby shards, multi-node parallel transmission achieves second-level recovery (such as financial transaction system core data);

[0069] Important data + ordinary data: asynchronous pulling of shards for reorganization, supporting breakpoint resuming (record recovery progress, can resume after interruption, avoiding repeated operations) and version rollback (relying on the blockchain timeline, restoring to any historical time point);

[0070] Integrity verification: after recovery is complete, automatically calculate the hash value of the recovered data and compare it with the blockchain evidence record to verify that the recovery is successful.

[0071] The following are several specific embodiments of the application:

[0072] Embodiment one: application of financial high-frequency trading system

[0073] 1. Data feature collection:

[0074] Through kernel probes, real-time order flow, clearing data, and customer account change records of the financial trading system are deeply captured, focusing on extracting transaction frequency (millisecond-level high-frequency trading markers), business correlation (payment link, clearing node priority), and data entropy (structural uncertainty of complex derivative transactions). For high-frequency trading data, use micro-batch stream processing (aggregate features every 50 milliseconds), filter invalid orders, test transactions, and other noise data.

[0075] 2. Intelligent classification and strategy generation:

[0076] Lightweight random forest model prioritizes training of "high-frequency trading data" exclusive feature library (including historical blowout event data and key payment node markers), and divides the data into:

[0077] Core transaction data (e.g. real-time orders, clearing instructions, access frequency > 100 times / second): trigger RDMA network card direct backup, data changes are synchronized to the edge node in microseconds;

[0078] Important business data (e.g. customer holdings, settlement reports, access frequency 10-100 times / second): use dynamic block backup, block size is adapted to the structure of transaction data packets (e.g. standardized field length of futures contracts);

[0079] Common data (e.g. system logs, temporary cache, access frequency < 10 times / second): enable hourly full backup + hash deduplication.

[0080] 3. Distributed storage and evidence storage:

[0081] The edge node cluster is deployed in financial city dual-active machine rooms, using PBFT consensus (7-node configuration, tolerant to 2-node failures), and core transaction data is redundantly stored in 3 or more nodes. The blockchain uses a financial alliance chain, and transaction data is hashed and uploaded to the chain in real time, generating a compliance audit voucher (including timestamp, transaction number, node signature).

[0082] 4. Verification and recovery:

[0083] The smart contract triggers core transaction data verification every 30 seconds, and automatically calls the hot backup shard repair of the dual-active node in the city when there is an exception. In the recovery phase, core transaction data is recovered in seconds through a preloaded memory pool, and precise rollback is supported according to transaction number and timestamp (e.g. rollback to a state before an abnormal transaction).

[0084] Example two: medical electronic medical record backup scenario

[0085] 1. Data feature collection:

[0086] For electronic medical record systems (EMR) and image archiving systems (PACS), collect medical record modification frequency (diagnosis and treatment stage markers), business correlation (surgery records, diagnosis conclusion priority), and data entropy (pixel complexity of image data). For privacy data (e.g. patient name, ID number), first desensitize (keep age, disease characteristics) and then extract features, which meets HIPAA compliance requirements.

[0087] 2. Intelligent classification and strategy generation:

[0088] Train a medical-specific random forest model (samples include historical misdiagnosis case data and key department markers), and the classification results are:

[0089] Critical medical record data (such as operation records, tumor diagnosis reports, modification frequency > 1 times / hour): real-time incremental backup + AES-256 encryption, backup data is only stored to the local edge node of the third-grade hospital (to avoid cross-regional transmission leakage risk);

[0090] Important image data (such as CT / MRI images, modification frequency 1 times / day): dynamically block according to image slice size (such as 200KB block for head CT, 50KB block for chest radiograph), combined with medical hash dictionary deduplication;

[0091] Common data (such as medical order template, routine test report, modification frequency < 1 times / day): enable daily full backup.

[0092] 3. Distributed storage and notarization:

[0093] The edge node cluster is deployed in the hospital local room + regional medical cloud, and PBFT consensus is adopted (5 node configuration, tolerating 1 node failure); the medical alliance chain is selected for the block chain, and the doctor's electronic signature and diagnosis time stamp are attached when the medical record hash is chained, forming a judicial traceable medical data chain.

[0094] 4. Verification and recovery:

[0095] The smart contract checks the integrity of the medical record every 24 hours (matches the diagnosis and treatment process timeline), and triggers local node repair in priority when abnormal (avoids cross-network transmission delay); in the recovery stage, key medical record data needs to be authorized by the doctor's work number + fingerprint, and supports rollback according to the diagnosis and treatment stage (such as rollback to the preoperative evaluation state).

[0096] Example three: government document full life cycle backup

[0097] 1. Data feature collection:

[0098] For government OA system and document exchange platform, collect document circulation stage (drafting, signing, issuing, archiving), business correlation (security level, priority of document issuing unit) and data entropy (format complexity of document attachments). For confidential documents (such as confidential level), the features are collected first by national encryption algorithm to ensure transmission security.

[0099] 2. Intelligent classification and strategy generation:

[0100] Train a government-specific random forest model (samples include historical document leakage cases and process node labels), and the classification results are:

[0101] Core confidential documents (such as minutes of regular meetings, draft plans, circulation stage "signing"): trigger quantum encryption channel backup, data real-time synchronization to government-specific cloud edge node;

[0102] Important process documents (such as departmental documents, reports, and circulation stages for "drafting / signing"): dynamic block according to the number of pages (such as 1 block within 10 pages, 3 blocks above 10 pages), combined with document template hash deduplication;

[0103] Ordinary archival documents (such as public notices and annual reports, circulation stage for "archiving"): weekly full backup is enabled.

[0104] 3. Distributed storage and evidence:

[0105] 4. The edge node cluster is deployed in the same city node of the government cloud, and adopts PBFT consensus (9 node configuration, tolerating 3 node failures); the blockchain selects the government alliance chain, and the document hash is chained with electronic signature and process node timestamp, forming a "document circulation-archiving" full-cycle evidence chain.

[0106] 4. Verification and recovery:

[0107] The smart contract checks the integrity of the document every 72 hours (matches the circulation node signature chain), and triggers cross-departmental collaborative repair when abnormal (such as abnormal signing stage needs to be verified by the involved department nodes); in the recovery stage, the core secret document needs to be authenticated by double factors (UKey+dynamic password), and supports version rollback according to the circulation (such as rolling back to the draft state before signing).

[0108] The above is only a specific embodiment of the present application, but the protection scope of the present application is not limited to this, any person skilled in the art can easily think of various changes or replacements within the technical scope disclosed by the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A backup method for a computer system, characterized by, Comprising the following steps: S1. Multi-dimensional feature extraction is performed on target data in a computer system, the features including at least data entropy, access frequency, and business correlation degree; S2. Based on the features, a machine learning model is used to classify the importance of the target data, obtaining key data, important data, and ordinary data; S3. According to the classification results, a dynamic backup strategy is generated, in which key data is backed up in real time, important data is backed up in dynamic blocks based on entropy, and ordinary data is periodically fully backed up; S4. Backup data shards are stored in an edge node cluster, and the hash values of each shard are written into a blockchain to form an unalterable evidence record; S5. Based on a smart contract of the blockchain, the consistency of the backup data and the evidence record is periodically verified, and if they are inconsistent, an abnormal processing mechanism is triggered.

2. The backup method for a computer system according to Claim 1, wherein: The data entropy value in step S1 is calculated by the formula where p(x i ) is the probability distribution of the data block x i , and the greater the H(X) value, the higher the data uncertainty.

3. The backup method for a computer system according to Claim 1, wherein: The machine learning model in step S2 is a lightweight random forest model that learns the mapping relationship between data features and importance through training samples containing feature labels and business priority annotations of historical data.

4. The backup method for a computer system according to Claim 1, wherein: In step S3, the block size of the dynamic block backup based on entropy is determined by the formula block size = a·H(X) + β, where a and β are preset empirical parameters, and H(X) is the entropy of the important data calculated in step S1.

5. The backup method for a computer system according to Claim 1, wherein: In step S4, the edge node cluster uses the PBFT consensus mechanism, and when any node with a survival rate of not less than 30% is alive, the complete backup data can be recovered through shard reorganization, and the shards stored in each node are dynamically allocated through a federated learning model to achieve load balancing.

6. The backup method for a computer system according to Claim 1, wherein: In step S5, the smart contract has a preset verification period and abnormal processing rules, and the abnormal processing mechanism includes sending an alarm to the administrator, automatically starting a data repair process, and writing the processing results into the blockchain evidence.

7. The backup method for a computer system according to Claim 1, wherein: It also includes a data recovery step: after receiving a recovery request, key data is recovered according to the data importance level, and important data and ordinary data are recovered in parallel; after recovery, the hash value of the recovered data is calculated and compared with the blockchain evidence record, and if the verification is passed, the recovery is confirmed to be successful.

8. A backup system for a computer system, characterized by Comprising: a feature extraction module for extracting data entropy, access frequency, and business correlation degree features from target data; a classification module connected to the feature extraction module, which classifies the importance of the data through a machine learning model; a strategy generation module connected to the classification module, which generates corresponding backup strategies according to the classification results; a distributed storage module connected to the strategy generation module, which includes an edge node cluster and a blockchain node, and is used to store data shards and hash evidence records; a verification module connected to the distributed storage module, which realizes consistency verification of backup data and evidence records through a smart contract.

9. The backup system for a computer system of claim 8, wherein: In the distributed storage module, the load scheduling of the edge node cluster is based on a federated learning model, which aggregates local training parameters to generate a global scheduling strategy without revealing local data.

10. The backup method for a computer system according to Claim 8, wherein: It also includes a recovery module connected to the distributed storage module and the verification module, which is used to recover key data according to the data importance level, and verify the integrity of the recovered data through hash comparison.

Images