Non-tampering operation and maintenance auditing system based on block chain
By using a blockchain-based immutable operation and maintenance audit system, the problems of unreliable data, incomplete coverage, poor compliance adaptation, delayed early warning, and insufficient consistency across scenarios in traditional operation and maintenance audit systems are solved, achieving efficient, reliable, and consistent operation and maintenance auditing under complex IT architectures.
Patent Information
- Application Number
- CN202512024654.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-30
- Publication Date
- 2026-05-01
AI Technical Summary
Traditional operation and maintenance audit systems suffer from problems such as unreliable data, incomplete coverage, poor compliance adaptation, delayed early warning, and insufficient consistency across scenarios, making them unable to meet the operation and maintenance needs of complex IT architectures.
The system adopts a blockchain-based immutable operation and maintenance audit system, which includes a multi-source audit data acquisition module, a blockchain evidence storage engine, a smart contract audit module, and a cross-scenario collaborative audit module. Combined with intelligent algorithms, it achieves trusted evidence storage of audit data throughout the entire chain, real-time risk perception, automatic compliance adaptation, and consistent traceability across scenarios.
It enhances the credibility and compliance of audit data, enables proactive risk identification and cross-scenario consistency, reduces operational interference and resource consumption, and adapts to the operational needs of complex IT architectures across multiple industries.
Smart Images

Figure CN121961756A_ABST
Abstract
Description
Blockchain-based immutable operation and maintenance audit system Technical Field
[0001] This invention relates to the field of computer information security technology, and in particular to an immutable operation and maintenance audit system based on blockchain. Background Technology
[0002] Operations and maintenance auditing is a core component of ensuring the security and compliance of information systems. Traditional operations and maintenance auditing solutions rely on centralized storage and manual verification, which have significant shortcomings under complex IT architectures and multi-scenario operations and maintenance needs: Insufficient audit data credibility: Under centralized storage, logs are easily tampered with or deleted, and cannot serve as effective evidence for compliance accountability; Limited audit coverage: Multiple systems and cross-environment (cloud-native / physical machine / container / blockchain) create "data silos," supporting only basic operations such as SSH and RDP, lacking audit support for new operations such as API calls and smart contract interactions; Low efficiency in compliance adaptation and traceability: Fixed rules are difficult to meet the personalized compliance requirements of different industries, and manual retrieval of massive logs is time-consuming and labor-intensive, easily overlooking key records; Passive and delayed risk warning: Primarily based on "post-event auditing," lacking real-time risk monitoring capabilities, unable to identify high-risk behaviors such as unauthorized operations and operations during abnormal periods in advance, leading to the expansion of risks; Poor consistency across scenarios: Under multi-cloud and multi-chain architectures, audit data synchronization is difficult, easily resulting in data inconsistencies, timestamp errors, and other problems, affecting the accuracy of audit results. Summary of the Invention
[0003] This invention provides a blockchain-based immutable operation and maintenance audit system, which addresses the pain points of traditional operation and maintenance audit systems, such as unreliable data, incomplete coverage, poor compliance adaptation, delayed early warning, and insufficient consistency across scenarios. Through the deep integration of blockchain technology and intelligent algorithms, it achieves trusted evidence storage of audit data throughout the entire chain, real-time risk perception, automatic compliance adaptation, and consistent traceability across scenarios. Under the premise of ensuring the security and compliance of information systems, it improves audit efficiency, reduces operation and maintenance interference and resource consumption, and meets the operation and maintenance audit needs of complex IT architectures in multiple industries such as finance and government.
[0004] To achieve the above objectives, the present invention adopts the following technical solution: a blockchain-based immutable operation and maintenance audit system, including a multi-source audit data acquisition module, a blockchain evidence storage engine, a smart contract audit module, a cross-scenario collaborative audit module, and a visual audit platform. Each module collaborates according to preset logic to form a closed-loop audit system. The multi-source audit data acquisition module is adaptable to cloud-native, physical machine, container, and blockchain multi-operation and maintenance environments. It collects traditional operation and maintenance data, API calls, smart contract interactions, and cross-chain operation and maintenance instruction data from these environments, and simultaneously collects adversarial feature data related to the operation data. It has a built-in dynamic risk perception model to analyze the collected operation data and adversarial feature data, and outputs risk levels and structured attribution reports. The blockchain evidence storage engine is based on a consortium blockchain architecture and uses national cryptographic algorithms to encrypt the collected operation data, adversarial feature data, risk levels, and structured attribution reports. The encrypted complete audit data is stored on the blockchain and a unique evidence storage number is generated. The efficiency of audit data retrieval is improved by using data sharding storage methods divided by day or operation and maintenance environment. The smart contract audit module... The contract audit module includes rule contracts, permission contracts, and early warning contracts. It integrates a multi-objective optimization decision model, using risk level, structured attribution reports, and historical audit data as inputs to dynamically adjust compliance verification thresholds and execute compliance verification through smart contracts. The cross-scenario collaborative audit module builds a cross-environment data synchronization channel based on the XCM protocol, integrating adaptive verification strategies. It uses audit data, evidence numbers, and risk levels stored in the blockchain evidence engine as inputs to optimize cross-environment data synchronization strategies and ensure consistency of audit data across scenarios. The visual audit platform uses evidence numbers generated by the blockchain evidence engine, compliance verification results from the smart contract audit module, and structured attribution reports as core data. It provides audit log query, risk alarm display, and compliance report generation functions, and connects to the regulatory database to complete data submission. The audit process is as follows: initial deployment, multi-source data collection and dynamic risk perception model analysis, encrypted audit data storage on the blockchain, dynamic adjustment of thresholds and compliance verification by the multi-objective optimization decision model, adaptive verification strategy optimization for cross-scenario synchronization, visual compliance audit, and permanent archiving of audit data.
[0005] In this specification, the multi-source audit data acquisition module adopts corresponding acquisition methods for different operation and maintenance environments and operation types: traditional operation and maintenance operations of physical machines and virtual machines are carried out by deploying a lightweight acquisition agent and transmitting data through an encrypted channel; API call operations of the cloud platform are carried out by periodically pulling logs through the open API of the cloud platform; operations in the container environment are carried out by mounting the container log directory and collecting logs through the Fluentd tool; smart contract interaction operations in the blockchain environment are collected by connecting to the blockchain node log interface; cross-chain operation and maintenance instructions are collected by listening to the XCM protocol cross-chain message channel. All collected data ensures that the adversarial feature data corresponding to the operation data is obtained synchronously.
[0006] In this specification, the dynamic risk perception model adopts a dual-channel analysis logic: Channel 1 calculates the deviation between the operation and maintenance and the preset behavioral baseline, which is constructed based on the operator's historical frequently used IP, operation time, high-frequency commands, and operation objects; Channel 2 calculates the adversarial similarity between the current operation and the known attack sequence through the DTW algorithm, which is derived from externally connected threat intelligence sources; the model is configured with five initial weights for permission compliance, abnormal behavior, command risk, environmental risk, and adversarial risk, and dynamically adjusts the weights of each dimension in combination with external threat intelligence, quantifies and calculates the total score and maps it to risk levels (L1-L4), and automatically generates a structured attribution report when the risk level is L3 or L4.
[0007] In this specification, the blockchain evidence storage engine adopts the Hyperledger Fabric consortium blockchain architecture, is configured with 4 consensus nodes and follows the Byzantine fault tolerance mechanism, and uses the PBFT consensus algorithm to generate blocks; the national cryptographic algorithms include the SM3 hash algorithm, the SM4 data encryption algorithm and the SM2 signature verification algorithm; data sharding storage is divided by day or operation and maintenance environment, with audit data from the past 3 months being stored as hot data first, and audit data from 3 months ago being archived as cold data. At the same time, it supports real-time cross-node backup of audit data and daily offline encrypted backup to ensure data security and recoverability.
[0008] In this specification, the multi-objective optimization decision-making model takes compliance, performance overhead, and operational interference as its three major optimization objectives. It takes the risk level, structured attribution report, and historical audit data output by the dynamic risk perception model as input, and learns the optimal adjustment strategy through reinforcement learning algorithm. For features with high contribution in the structured attribution report, the corresponding compliance verification threshold is tightened, and for features with low contribution, the corresponding compliance verification threshold is appropriately relaxed. At the same time, the threshold is dynamically adjusted in combination with the preset compliance coefficient K and the CPU and memory utilization of the operation and maintenance environment. The adjusted threshold takes effect through on-chain hot update.
[0009] In this specification, the adaptive verification strategy analyzes historical network performance indicators and system resource status data using an LSTM model to predict communication quality and resource sufficiency for the next 5 minutes; it uses the NSGA-II algorithm to construct a multi-objective optimization problem and outputs a Pareto optimal synchronization scheduling scheme; it sets a basic synchronization frequency based on the risk level, with higher risk levels resulting in higher synchronization frequencies, and dynamically adjusts the synchronization frequency and task priority based on the prediction results; it performs consistency verification by comparing the hash value, timestamp, and risk level of cross-environment audit data, and if all three are consistent, the process passes; otherwise, resynchronization is triggered.
[0010] In this specification, the smart contract auditing module supports hot contract updates and can quickly adapt to multiple industry compliance standards such as the Information Security Protection Standard 2.0, SOX, and GDPR. By configuring different values of the compliance coefficient K, it can adapt to the compliance requirements of different scenarios. The smart contract has reentrancy attack protection and permission isolation mechanisms. It triggers hierarchical alarms based on compliance verification results. High-risk non-compliant operations are pushed to the security manager and administrator, while ordinary non-compliant operations are pushed to the administrator. It supports automatic blocking of high-risk non-compliant operations and generates violation audit records for on-chain evidence storage.
[0011] In this specification, the structured attribution report adopts JSON format, which includes the total score corresponding to the risk level and the marginal contribution of each input feature to the total score, and clearly defines the core attribution factors such as high-risk instruction hit, deviation from behavioral baseline, adversarial similarity matching, and abnormal operation time period. After the report is stored on the blockchain evidence storage engine, it is associated with the corresponding evidence storage number and serves as the basis for compliance verification, risk tracing and regulatory review. It supports querying the original record of the report through the evidence storage number on the blockchain explorer.
[0012] In this manual, the dynamic risk perception model, multi-objective optimization decision model, and adaptive verification strategy are all configured with degradation strategies: When the dynamic risk perception model crashes, it maintains the most recent effective risk level for 30 minutes; if it does not recover within 30 minutes, it switches to the default risk level L2; when the multi-objective optimization decision model fails, it maintains operation using the initial compliance verification threshold; if it does not recover within 1 hour, it switches to the industry default threshold; when the prediction module of the adaptive verification strategy fails, it reverts to the instantaneous network status and system load perception strategy, maintaining the basic synchronization frequency. When all models are degraded, corresponding alarms are triggered to notify the administrator to troubleshoot the fault.
[0013] In this specification, the visual auditing platform connects to the regulatory database via HTTPS+JSON-RPC2.0 protocol, employs two-way certificate authentication and HMAC-SHA256 signature mechanism to ensure data transmission security, and the signature parameters include request timestamp, interface name, compliance coefficient K, and random number; it supports three core interfaces: audit log upload, compliance report submission, and high-risk event push, and can export compliance audit reports containing structured attribution reports and judgment criteria with one click, meeting the regulatory authorities' verification requirements for audit evidence.
[0014] In summary, this invention has at least the following beneficial effects: Significantly enhanced audit data credibility: Based on the blockchain's evidence storage characteristics, audit data is permanently immutable after being encrypted and uploaded to the chain, serving as a legal basis for compliance accountability and gaining regulatory approval; Enhanced risk perception and foresight: Intelligent algorithms enable accurate identification of advanced threats, proactively intercepting unauthorized operations, malicious tampering, and other risky behaviors, preventing core data leakage or increased losses; Optimized compliance traceability and adaptability: The audit process is highly interpretable, clearly defining risk assessment criteria, adapting to multi-industry compliance standards, and enabling compliance report generation and regulatory coordination without extensive manual intervention; Achieving a balance between compliance and efficiency: Dynamically adjusting compliance verification strategies reduces system resource consumption and minimizes interference with normal operations while meeting high compliance requirements; Enhanced cross-scenario collaborative auditing capabilities: Solving the "data silo" problem across multiple environments, ensuring consistency of audit data under different IT architectures, and adapting to complex operation scenarios; Outstanding system availability and self-evolution capabilities: Possessing comprehensive fault degradation strategies, supporting 24 / 7 uninterrupted auditing, and achieving self-optimization through bidirectional feedback between algorithms, continuously improving audit accuracy. Attached Figure Description
[0015] Figure 1 is a schematic diagram of the initialization and data acquisition stages involved in this invention.
[0016] Figure 2 is a schematic diagram of the risk perception and compliance determination stage involved in this invention.
[0017] Figure 3 is a schematic diagram of the synchronization and audit archiving stage involved in this invention. Detailed Implementation
[0018] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.
[0019] As shown in Figures 1, 2, and 3, this embodiment provides a blockchain-based immutable operation and maintenance audit system, including a multi-source audit data acquisition module, a blockchain evidence storage engine, a smart contract audit module, a cross-scenario collaborative audit module, and a visual audit platform. Each module collaborates according to preset logic to form a closed-loop audit system. The multi-source audit data acquisition module is adaptable to cloud-native, physical machine, container, and blockchain multi-operation and maintenance environments. It collects traditional operation and maintenance data, API calls, smart contract interactions, and cross-chain operation and maintenance command data from these environments, and simultaneously collects adversarial feature data related to the operation data. It has a built-in dynamic risk perception model to analyze the collected operation data and adversarial feature data, outputting risk levels and structured attribution reports. The blockchain evidence storage engine is based on a consortium blockchain architecture and uses national cryptographic algorithms to encrypt the collected operation data, adversarial feature data, risk levels, and structured attribution reports. The encrypted complete audit data is stored on the blockchain and a unique evidence number is generated. Data sharding storage by day or operation and maintenance environment improves audit data query efficiency. The smart contract audit module... The audit module includes rule contracts, permission contracts, and early warning contracts, integrating a multi-objective optimization decision model. It dynamically adjusts compliance verification thresholds based on risk level, structured attribution reports, and historical audit data, and executes compliance verification through smart contracts. The cross-scenario collaborative audit module builds a cross-environment data synchronization channel based on the XCM protocol, integrating adaptive verification strategies. It optimizes cross-environment data synchronization strategies based on audit data stored in the blockchain evidence engine, evidence numbers, and risk levels, ensuring consistency of audit data across scenarios. The visual audit platform uses evidence numbers generated by the blockchain evidence engine, compliance verification results from the smart contract audit module, and structured attribution reports as core data. It provides audit log query, risk alarm display, and compliance report generation functions, and connects to the regulatory database to complete data submission. The audit process consists of initial deployment, multi-source data collection and dynamic risk perception model analysis, encrypted on-chain audit data storage, dynamic adjustment of thresholds and compliance verification using the multi-objective optimization decision model, adaptive verification strategy optimization for cross-scenario synchronization, visual compliance audit, and permanent archiving of audit data.
[0020] In some embodiments, the multi-source audit data acquisition module adopts corresponding acquisition methods for different operation and maintenance environments and operation types: traditional operation and maintenance operations of physical machines and virtual machines are carried out by deploying a lightweight acquisition agent and transmitting data through an encrypted channel; API call operations of the cloud platform are carried out by periodically pulling logs through the open API of the cloud platform; operations in the container environment are carried out by mounting the container log directory and collecting logs through the Fluentd tool; smart contract interaction operations in the blockchain environment are collected by connecting to the blockchain node log interface; cross-chain operation and maintenance instructions are collected by listening to the XCM protocol cross-chain message channel. All collected data ensures that the adversarial feature data corresponding to the operation data is obtained synchronously.
[0021] In some embodiments, the dynamic risk perception model employs a dual-channel analysis logic: Channel 1 calculates the deviation between the operation and maintenance activities and a preset behavioral baseline, which is constructed based on the operator's historically frequently used IPs, operation times, high-frequency commands, and operation objects; Channel 2 calculates the adversarial similarity between the current operation and known attack sequences using the DTW algorithm, where the known attack sequences originate from externally connected threat intelligence sources; the model configures initial weights for five dimensions: permission compliance, abnormal behavior, command risk, environmental risk, and adversarial risk, dynamically adjusts the weights of each dimension in conjunction with external threat intelligence, quantifies and calculates the total score, and maps it to a risk level (L1-L4). When the risk level is L3 or L4, a structured attribution report is automatically generated.
[0022] In some embodiments, the blockchain evidence storage engine adopts a Hyperledger Fabric consortium blockchain architecture, configures 4 consensus nodes and follows the Byzantine fault tolerance mechanism, and uses the PBFT consensus algorithm to generate blocks; the national cryptographic algorithms include the SM3 hash algorithm, the SM4 data encryption algorithm and the SM2 signature verification algorithm; data sharding storage is divided by day or operation and maintenance environment, with audit data from the past 3 months being stored as hot data first, and audit data from 3 months ago being archived as cold data. At the same time, it supports real-time cross-node backup of audit data and daily offline encrypted backup to ensure data security and recoverability.
[0023] In some embodiments, the multi-objective optimization decision model takes compliance, performance overhead, and operational interference as three major optimization objectives. It inputs the risk level, structured attribution report, and historical audit data output by the dynamic risk perception model, and learns the optimal adjustment strategy through reinforcement learning algorithm. For features with high contribution in the structured attribution report, the corresponding compliance verification threshold is tightened, and for features with low contribution, the corresponding compliance verification threshold is appropriately relaxed. At the same time, the threshold is dynamically adjusted in combination with the preset compliance coefficient K and the CPU and memory utilization of the operation and maintenance environment. The adjusted threshold takes effect through on-chain hot update.
[0024] In some embodiments, the adaptive verification strategy analyzes historical network performance indicators and system resource status data using an LSTM model to predict communication quality and resource sufficiency for the next 5 minutes; it constructs a multi-objective optimization problem using the NSGA-II algorithm to output a Pareto optimal synchronization scheduling scheme; it sets a basic synchronization frequency based on the risk level, with higher risk levels resulting in higher synchronization frequencies, and dynamically adjusts the synchronization frequency and task priority based on the prediction results; it performs consistency verification by comparing the hash value, timestamp, and risk level of cross-environment audit data, and if all three are consistent, the system passes the test; otherwise, it triggers resynchronization.
[0025] In some embodiments, the smart contract auditing module supports hot contract updates and can quickly adapt to multiple industry compliance standards such as the Information Security Protection Standard 2.0, SOX, and GDPR. It can adapt to the compliance requirements of different scenarios by configuring different compliance coefficients K. The smart contract has reentrancy attack protection and permission isolation mechanisms. It triggers hierarchical alarms based on compliance verification results. High-risk non-compliant operations are pushed to the security manager and administrator, while ordinary non-compliant operations are pushed to the administrator. It supports automatic blocking of high-risk non-compliant operations and generates violation audit records for on-chain storage.
[0026] In some embodiments, the structured attribution report adopts JSON format, including the total score corresponding to the risk level and the marginal contribution of each input feature to the total score, and clearly identifies core attribution factors such as high-risk instruction hits, deviation from behavioral baselines, adversarial similarity matching, and abnormal operation periods. After the report is stored on the blockchain through the blockchain evidence storage engine, it is associated with the corresponding evidence storage number and serves as the basis for compliance verification, risk tracing, and regulatory review. It also supports querying the original report record on the blockchain explorer through the evidence storage number.
[0027] In some embodiments, the dynamic risk perception model, the multi-objective optimization decision model, and the adaptive verification strategy are all configured with a degradation strategy: when the dynamic risk perception model crashes as a whole, it maintains the most recent effective risk level for 30 minutes; if it does not recover within 30 minutes, it switches to the default risk level L2; when the multi-objective optimization decision model fails to execute, it maintains operation using the initial compliance verification threshold; if it does not recover within 1 hour, it switches to the industry default threshold; when the prediction module of the adaptive verification strategy fails, it reverts to the instantaneous network status and system load perception strategy, maintains the basic synchronization frequency, and triggers corresponding alarms to notify the administrator to troubleshoot the fault when all models are degraded.
[0028] In some embodiments, the visual audit platform connects to the regulatory database via HTTPS+JSON-RPC2.0 protocol, employs two-way certificate authentication and HMAC-SHA256 signature mechanism to ensure data transmission security, and the signature parameters include request timestamp, interface name, compliance coefficient K and random number; it supports three core interfaces: audit log upload, compliance report submission, and high-risk event push, and can export compliance audit reports containing structured attribution reports and judgment criteria with one click, meeting the regulatory authorities' verification requirements for audit evidence.
[0029] The technical concept of this invention is as follows: This invention constructs an immutable operation and maintenance audit system based on "blockchain evidence storage + smart contract driving + multi-source data collection + intelligent risk warning". The core system comprises five modules and three intelligent algorithms: Five core modules: Multi-source audit data collection module (cross-scenario data collection and preliminary risk perception), blockchain evidence storage engine (encrypted evidence storage and efficient query), smart contract audit module (compliance verification and access control), cross-scenario collaborative audit module (cross-environment data synchronization and consistency assurance), and a visual audit platform (log query, alarm display, and report generation); Three core algorithms: Dynamic risk perception model (A1 algorithm, quantifying five-dimensional risks and generating attribution reports), multi-objective optimization decision model (A2 algorithm, dynamically adjusting compliance verification thresholds), and adaptive verification strategy (A3 algorithm, optimizing cross-scenario data synchronization scheduling); Full-process logic: Operation and maintenance operation and adversarial characteristic data are acquired through multi-source collection, encrypted and stored on the blockchain, and then the intelligent algorithm completes risk perception and compliance verification, triggering tiered alarms or blocking, while simultaneously achieving cross-scenario data synchronization and compliance auditing, ultimately forming a closed-loop system of "collection-evidence storage-analysis-verification-synchronization-audit".
[0030] The system consists of five core modules, which work together to deeply integrate algorithmic logic and blockchain characteristics: The multi-source audit data acquisition module supports cross-scenario operation and maintenance data acquisition, covering traditional operations such as SSH / RDP, API calls, smart contract interactions, and cross-chain operation and maintenance instructions, adapting to multiple environments including cloud-native, physical machines, containers, and blockchain. It incorporates a dynamic risk perception model (A1 algorithm) based on threat intelligence, behavioral profiling, and adversarial simulation, monitoring operational behavior in real time, calculating adversarial similarity, and outputting L1-L4 risk levels and structured attribution reports, providing decision-making basis for audit warnings and compliance verification. It also supports multiple acquisition methods such as Syslog, API, and Agent, simultaneously collecting adversarial data such as known attack pattern libraries and operation sequence characteristics to ensure comprehensive data acquisition.
[0031] Blockchain Evidence Storage Engine: Based on a consortium blockchain architecture (using Hyperledger Fabric), it generates immutable block records from audit data (operator, operation object, operation instruction, timestamp, five-dimensional risk data, adversarial similarity, attribution report, etc.), and uses the national cryptographic SM3 hash algorithm for data encryption to ensure data integrity and security. It supports standardized format processing of audit data before it is uploaded to the blockchain, adapting to compatibility with data from different sources; it has a built-in data sharding storage mechanism, sharding by day / operational environment, improving the storage and query efficiency of massive amounts of audit data (single log query response time ≤ 1 second).
[0032] The smart contract audit module includes rule contracts (compliance rule configuration), permission contracts (tiered authorization management), and alert contracts (risk alarm triggering). It integrates a reinforcement learning-based multi-objective optimization decision-making model (A2 algorithm), and dynamically adjusts compliance verification thresholds and alert strategies based on the attribution report output by the A1 algorithm, balancing the three objectives of "compliance," "performance overhead," and "operational efficiency." It supports hot contract updates, enabling rapid adaptation to different industry compliance requirements (such as Cybersecurity Classified Protection 2.0, SOX, GDPR), and also features security mechanisms such as reentrancy attack protection and permission isolation.
[0033] Cross-scenario collaborative auditing module: Based on the Polkadot XCM protocol, a cross-environment audit data synchronization channel is built to achieve real-time interoperability of audit data under multi-cloud, multi-chain, and hybrid architectures. It integrates an adaptive verification strategy based on "prediction-Pareto" (A3 algorithm), using an LSTM model to predict network quality and resource status for the next 5 minutes, and combining the NSGA-II algorithm to output a Pareto-optimal scheduling scheme, dynamically adjusting the data synchronization frequency and verification priority to ensure cross-scenario audit data consistency; it supports rapid integration with mainstream cloud platforms (Alibaba Cloud, AWS) and blockchain networks (Ethereum, Polygon), adapting to complex IT architectures.
[0034] Visualized Audit Platform: Provides a web-based interface supporting audit log querying (multi-dimensional filtering), risk event alerts, compliance report generation, and permission configuration. Built-in data visualization charts intuitively display operational trends, risk distribution, compliance rates, and algorithm scheduling effects. Supports integration with regulatory databases (such as CFCA in the financial industry and e-government platforms in the government sector), and allows for one-click export of compliance audit reports with attribution evidence. Administrators can also manually switch A3 algorithm scheduling strategies (consistency priority / bandwidth saving / efficiency priority) to improve audit convenience and compliance.
[0035] 1. Full-process initialization and deployment of operation and maintenance audit: Deploy consortium blockchain nodes + configure collection agent / interface + set permission matrix + connect to regulatory database (including compliance coefficient K configuration) + connect to external threat intelligence sources + algorithm parameter initialization (A1 five-dimensional initial weights, A2 reward function parameters, A3 LSTM model parameters / NSGA-II configuration) Data collection: Multi-source collection module collects operation and maintenance data in real time + adversarial feature data (operation sequence, attack mode matching information) + format standardization processing + SM3 hash encryption blockchain evidence storage: Encrypted audit data (including risk level, attribution report) is stored on the blockchain + generate a unique evidence hash + return evidence number Risk perception: A1 algorithm dual-channel analysis (behavioral baseline deviation + adversarial similarity) + dynamic weight adjustment + output risk level (L1-L4) + generate structured attribution report Compliance verification: A2 algorithm combined with attribution The system utilizes a combination of reporting, dynamic calculation of compliance verification thresholds, and smart contract execution compliance rule verification trigger conditions. It combines risk level and compliance verification results: a. Compliant and low-risk (L1 / L2): Continuously collect and store evidence; b. Compliant but high-risk (L3 / L4): Trigger risk warnings (including attribution summaries) + record warning logs; c. Non-compliant: Trigger tiered alarms (including attribution basis) + block illegal operations (optional) + generate violation audit records (on-chain evidence storage); Cross-scenario synchronization: A3 algorithm predicts network / resource status + outputs Pareto optimal scheduling scheme + dynamically adjusts synchronization frequency and priority + cross-environment audit data synchronization + consistency verification; Compliance audit: The visualization platform supports log query / risk tracing (including attribution report viewing) / compliance report generation + connects to the regulatory database to upload reports (including judgment basis); Log archiving: Audit data is permanently archived on the blockchain + offline backup (optional).
[0036] 2. Key Process Details (1) Initialization Deployment Parameter Configuration: Users configure the data collection scope (operation type, operation and maintenance environment), permission matrix (auditor, operation and maintenance personnel, administrator), compliance rule template (adapting to industry standards), and regulatory database connection parameters (interface address, authentication key, compliance coefficient K) through the visualization platform; at the same time, they set the initialization parameters of the three major algorithms: A1 five-dimensional initial weights (permission compliance 25%, behavior abnormality 30%, instruction risk 20%, environmental risk 10%, adversarial risk 15%), A2 target weights (core business α=0.6 / β=0.2 / γ=0.2, non-core α=0.4 / β=0.3 / γ=0.3), A3 LSTM training window (2 hours), NSGA-II configuration (population 100, crossover probability 0.9, mutation probability 0.1, iteration 200). Among them, the value of the compliance coefficient K is strongly correlated with the compliance risk. When K=2, it meets the requirements of the financial industry's Level 3 of the Information Security Protection Standard 2.0. K=1.5 is suitable for ordinary government affairs scenarios. In this example, K=2 is selected according to business needs.
[0037] Consortium blockchain deployment: Deploy 4 consensus nodes (following the N=4 / f=1 Byzantine fault tolerance mechanism) to ensure normal system operation when a single node fails; adopt the PBFT consensus algorithm to improve block generation efficiency (block generation time ≤ 3 seconds); configure data sharding strategy to shard and store data according to the operation and maintenance environment / time dimension.
[0038] Data collection module deployment: Deploy the data collection agent in the target operation and maintenance environment (physical machine, container, cloud server), or configure the API interface to connect to the cloud platform / blockchain node, and test the data collection success rate (≥99.9%); connect to external threat intelligence sources (such as the National Information Security Vulnerability Database, industry security announcements) to obtain the abnormal operation and maintenance behavior feature database and 47 attack mode TTPs (minimum privilege bypass, log erasure, etc.).
[0039] Algorithm degradation strategy configuration: Preset degradation scheme after algorithm execution failure, such as the default risk level of A1 algorithm when it crashes is L2, and its weight is reduced to 5% when the adversarial module fails; the initial verification threshold is used when A2 algorithm fails, and the industry default value is switched if it does not recover within 1 hour; the basic synchronization frequency is used when A3 algorithm fails, and it falls back to instantaneous state awareness when the prediction module fails.
[0040] (2) Data collection: The multi-source collection module collects operation and maintenance data in real time through Agent / API / Syslog and other methods: traditional operations (SSH login, command execution, file transfer), API calls (cloud platform interface operations, blockchain contract calls), cross-chain operation and maintenance instructions (cross-chain node configuration modification), etc.; and collects adversarial feature data simultaneously, including operation sequence vectors, matching information with the attack pattern library, DTW similarity scores, etc.
[0041] The collected data is standardized by unifying the data format (operator ID, object ID, operation instruction, operation timestamp, IP address, operation result, adversarial similarity, risk level, and attribution report identifier) and using the national cryptographic SM3 hash algorithm to generate hash values for the operation data to ensure that the data is not tampered with during transmission.
[0042] The standardized raw data, adversarial feature data, and hash values are synchronized to the blockchain evidence storage engine and simultaneously pushed to the A1 algorithm for risk analysis.
[0043] (3) Blockchain Evidence Storage After receiving the collected data, the blockchain evidence storage engine verifies the data integrity (compares hash values) and writes the data into the consortium blockchain block through a smart contract. The block contains a block header (previous block hash, timestamp, Merkle root) and a block body (audit data record, five-dimensional risk data, adversarial similarity, JSON format attribution report) to ensure that the data is immutable.
[0044] A unique evidence number (composed of block height + data index) is generated and returned to the visualization platform and collection module for subsequent audit log query and traceability.
[0045] A data sharding storage mechanism is adopted, and data is sharded by day / by operation and maintenance environment to improve the query efficiency of massive audit data (the response time for a single log query is ≤1 second).
[0046] (4) Risk perception and compliance verification: The A1 algorithm receives collected data and external threat intelligence, and performs dual-channel analysis in real time: Channel 1 calculates the deviation δ between the operation and the normal behavior baseline, and Channel 2 calculates the maximum similarity σ with the attack sequence through the DTW algorithm; Combined with threat intelligence, the five-dimensional risk weights are dynamically adjusted to quantify the five-dimensional risk scores (authority compliance, abnormal behavior, instruction risk, environmental risk, adversarial risk), and output the risk level (L1-L4); If it is an L3 / L4 level risk, a structured attribution report is immediately generated and stored on the blockchain.
[0047] The A2 algorithm dynamically calculates compliance verification thresholds based on the risk level, attribution report (high contribution feature list), compliance requirements, and historical audit data output by A1. It learns the optimal adjustment strategy through the Q-Learning algorithm, tightens the threshold for features with a contribution of ≥30% in the attribution report (×0.95), and appropriately relaxes the threshold for features with a contribution of ≤5% (×1.05). After balancing the three objectives, it outputs the optimal verification rules and warning thresholds.
[0048] The smart contract performs compliance verification on the operation and maintenance data according to the verification rules output by the A2 algorithm: comparing the operator's permissions, the legality of the operation instructions, the compliance of the operation scope, the adversarial similarity threshold, etc., and generates a compliance verification result (compliant / non-compliant).
[0049] (5) Triggering conditions are handled in a regular and low-risk manner (L1 / L2): Continue to execute the data collection and evidence storage process without triggering any alarms.
[0050] Compliant but high-risk (L3 / L4): When a contract triggers a risk alert, it is pushed to the operations and maintenance administrator via a visualization platform, SMS, or email (including risk level, attribution summary, and adversarial similarity information), and the alert is recorded in the alert log without blocking operations and maintenance.
[0051] Non-compliance: The warning contract triggers a tiered alert (high-risk non-compliance is pushed to the security manager and administrator, and ordinary non-compliance is pushed to the administrator). The alert information includes evidence of the violation, risk level, compliance clause basis, and link to the attribution report. You can choose whether to block the violation operation based on the configuration (such as automatically blocking when modifying core configurations across permissions). Generate a violation audit record (including a complete attribution report), store it on the blockchain, and synchronize it to the regulatory docking module.
[0052] (6) Cross-scenario collaborative auditing The cross-scenario collaborative auditing module builds cross-environment audit data synchronization messages based on the XCM protocol, including core fields such as evidence number, operation data hash, risk level, compliance result, adversarial similarity, and timestamp, to ensure the conciseness and completeness of the synchronized data.
[0053] The A3 algorithm performs look-ahead scheduling: it analyzes 120 past data points (30-second granularity) using an LSTM model to predict communication quality scores for the next 5 minutes. With abundant resources The synchronization task is modeled as a four-objective optimization problem (maximizing consistency, minimizing latency, minimizing peak bandwidth, and minimizing resource consumption), and 8-12 Pareto optimal solutions are output through the NSGA-II algorithm. The system defaults to selecting the solution closest to the ideal point, and supports administrators to manually switch personalized strategies (consistency priority / bandwidth saving / efficiency priority).
[0054] Synchronization frequency is dynamically adjusted based on risk level and prediction results: Level L4 base frequency 5 minutes / time (3 minutes / time under favorable conditions), Level L3 10 minutes / time (8 minutes / time under favorable conditions), Level L2 30 minutes / time, Level L1 2 hours / time; if prediction ≤4 or ≤0.3, L1 / L2 frequency ×0.8, L3 / L4 maintain basic frequency and reserve resources; if a business peak is predicted, complete the synchronization of non-core high-risk data one window in advance.
[0055] Consistency verification logic: The hash value, timestamp, and risk level of cross-environment audit data are compared. If all three are consistent, the data is deemed to have passed. If any one of them is inconsistent, data resynchronization is triggered (based on the main storage node data, up to 3 times, with an interval of 60 seconds). For predicted network fluctuation periods, the core data verification coverage is increased to 100%, and the non-core data coverage is increased to 80% to ensure synchronization consistency.
[0056] (7) Compliance audit and log archiving: Auditors can use the visualization platform to query audit logs by multi-dimensional filtering conditions (operator, time range, risk level, compliance status, adversarial similarity) and support traceability of evidence number (jump to the blockchain explorer to view the original block record and attribution report).
[0057] The system automatically calculates indicators such as compliance rate, risk event occurrence rate, and advanced threat detection rate, and generates a compliance audit report (supporting export in PDF / Excel format). The report includes the audit scope, compliance clause matching results, risk event details (including attribution basis), and rectification suggestions.
[0058] It supports integration with regulatory databases, automatically uploading compliance audit reports, key audit logs, and attribution reports to meet industry regulatory requirements for "explanation of judgment criteria"; audit data is permanently archived on the blockchain and supports offline backup (encrypted storage on a private server) to ensure data traceability.
[0059] 3. Key Technical Details: Multi-Source Data Acquisition and Adaptation Solution: Acquisition Method Adaptation: Data Format Standardization: The unified audit data format is JSON structure, with the following core fields: {"operationId": "OP202407010001", / / Unique identifier for the operation "operator": "admin_001", / / Operator ID "operatorIp": "192.168.1.100", / / Operation IP address "operationObject": "server_01", / / Operation object (server / contract / cloud resource) "operationType": "SSH command execution", / / Operation type "operationCmd": "rm -rf / test", / / Operation command (sensitive information anonymized) "timestamp": "2024-07-01T10:30:00Z", / / Operation timestamp (UTC) "hashValue": "sm3 (raw data)", / / Data hash value "riskLevel": "L4", / / Risk level (A1 algorithm output) "adversarialSimilarity": 85, / / Adversarial similarity (σ value) "attributionReportId": "AR202407010001", / / Unique identifier for attribution report "complianceStatus": "Non-compliant" / / Compliance status (A2 algorithm + smart contract verification result)}.
[0060] Core configuration of consortium blockchain evidence storage: ; Regulatory database interface standard (taking CFCA in the financial industry as an example): Secure connection is achieved by using HTTPS+JSON-RPC2.0 protocol to meet the compliance requirements of the financial industry: (1) Basic interface configuration protocol: HTTPS (TLS1.3), port 443, timeout time 30 seconds.
[0061] Authentication: Two-way certificate authentication (the system deploys a client certificate issued by CFCA, and the supervisor deploys a server certificate). The request header carries an HMAC-SHA256 signature (signature parameters: request timestamp, interface name, compliance coefficient K, random number, and the key is assigned by CFCA).
[0062] Compliance coefficient K: When K≥2, the Level 3 compliance requirements of the Information Security Protection Scheme 2.0 are met. In this example, K=2 is used to ensure compliance.
[0063] (2) Core interface definition ;3. Three major algorithm innovations and interaction logic Based on the dynamic risk perception model (A1 algorithm) of threat intelligence and behavior analysis (1) Algorithm objective: Real-time quantification of five-dimensional risks of operation and maintenance operation: "permission compliance - abnormal behavior - dangerous command - vulnerable environment - adversarial", output L1-L4 level, and automatically generate structured risk attribution report to provide traceable basis for compliance verification, early warning triggering and administrator decision-making.
[0064] (2) Input parameters (3) Dynamic weight adjustment mechanism initial weight configuration: W = (permission compliance 25%, abnormal behavior 30%, instruction risk 20%, environmental risk 10%, adversarial risk 15%). The system maintains a "threat sensitivity coefficient" αi for each risk dimension (initial value 1.0, range 0.5-1.5). When connecting with external threat intelligence sources, if the average number of certain risk events per day is ≥3 in the past 7 days or a new adversarial attack pattern is detected, the corresponding dimension αi +0.1 (upper limit 1.5) will be increased, and the other dimensions αj will be scaled proportionally to ensure Σ(Wi×αi) = 1.
[0065] Weight calculation formula: Dynamic weight Wi′ of a certain dimension = initial weight Wi of the dimension × threat sensitivity coefficient αi of the dimension / Σ (initial weight Wj of each dimension × threat sensitivity coefficient αj of each dimension) (j=1→5).
[0066] (4) Dual-channel analysis logic ① Behavioral baseline channel (original logic optimization): Establish a behavioral baseline for each operator (common IP, operation time, high-frequency instructions, operation object), and calculate the behavioral deviation δ=Σ(single-type behavioral deviation value × behavioral weight). Behavioral weight configuration: IP matching degree 30%, time matching degree 30%, instruction matching degree 25%, object matching degree 15%; single-type behavioral deviation value = 1 - the proportion of actual behavior conforming to the baseline (range 0-1). When δ≥0.7 (serious deviation), trigger the risk level jump.
[0067] ② Adversarial Channel: The DTW algorithm is used to calculate the maximum similarity σ (range 0-100) between the current operation and the attack sequence generated by reinforcement learning. The higher the σ value, the more similar the operation is to the high-risk attack sequence. A new "Adversarial Risk Sub-item" score Sa = σ × W5′ (W5′ is the dynamic weight of adversarial risk); when Sa ≥ 12 points, it directly jumps to level L3 or above to ensure that no high-risk adversarial operations are missed.
[0068] (5) Risk Quantification Calculation (Total Score 100 points) Permission Compliance Risk Score (W1′): Unauthorized Operation 100 points, Operation with Expired Permission 80 points, Operation with Unregistered Permission 60 points, Compliance Operation 0 points; Final Score = Score of this Dimension × W1′.
[0069] Behavioral abnormality risk score (W2′): score = δ × 100 × W2′; final score = score of this dimension.
[0070] Command risk score (W3′): High-risk commands (such as rm -rf, sudosu) 100 points, sensitive commands (such as database modification) 80 points, ordinary commands 0 points; final score = score of this dimension × W3′.
[0071] Environmental risk score (W4′): 60 points for high-risk environment operation, 50 points for environment with unpatched high-risk vulnerabilities, and 0 points for low-risk environment compliance operation; final score = score of this dimension × W4′.
[0072] Adversarial risk score (W5′): Score = Sa = σ × W5′; Final score = score for this dimension.
[0073] Total score = W1′ score + W2′ score + W3′ score + W4′ score + W5′ score, corresponding to risk level (0-20 points: L1, 21-40 points: L2, 41-60 points: L3, 61-100 points: L4).
[0074] (6) Risk source tracing and attribution explanation: When the A1 algorithm outputs L3 / L4 high risk level, SHAPKernel Explainer is immediately called to quantify the marginal contribution of each input feature to the final score, generate a structured JSON attribution report and store it on the blockchain.
[0075] Example report: "This case is classified as a Level 4 risk (out of 75 points). The core reasons are: the command 'sudo su' hit the high-risk command library (contribution +35%); the source IP of the operation has no overlap with the historical baseline (contribution +30%); the similarity with the 'log erasure' type adversarial sequence DTW is 85% (contribution +18%); the operation time falls within the abnormal window in the early morning (contribution +20%); all high-risk vulnerabilities in the operation and maintenance environment have been patched (contribution -8%)."
[0076] Feedback value: The attribution report serves as input data for the A2 and A3 algorithms, providing a basis for subsequent threshold adjustments and scheduling optimizations, thus achieving a "interpretation-optimization" closed loop.
[0077] (7) The overall algorithm of the downgrade strategy crashes: the most recent effective risk level is used to maintain the risk level for 30 minutes. If it is not restored within 30 minutes, it will switch to the default level L2 and trigger an alarm.
[0078] Individual failure of adversarial module: The dynamic weight W5′ of adversarial risk dimension will be temporarily reduced to 5%, while the weights of the other four dimensions will be scaled proportionally according to the initial ratio to continue to ensure the core risk detection capability. At the same time, a special alarm will be triggered to notify the administrator to investigate.
[0079] (1) Algorithm Objective: To balance the three objectives of "compliance C, performance overhead P, and operation and maintenance interference I", and to dynamically adjust the compliance verification threshold by combining the risk attribution report of the A1 algorithm, so as to achieve accurate compliance and optimal resource efficiency.
[0080] (2) Input parameters: (3) Multi-objective optimization function definition: Optimization function F=α×C-β×P-γ×I, where: C: compliance index (0-100 points), C=compliance rate×100, compliance rate=compliance operation number / total operation number.
[0081] P: Performance overhead index (0-100 points), P = (number of verifications × resource consumption per verification) / resource threshold × 100.
[0082] I: Operation and maintenance interference index (0-100 points), I = number of falsely judged violations / total number of operations × 100, false judgment refers to normal operations being judged as non-compliant.
[0083] α, β, and γ are the target weights: α=0.6, β=0.2, and γ=0.2 for core business scenarios; and α=0.4, β=0.3, and γ=0.3 for non-core business scenarios.
[0084] (4) Reinforcement learning feedback mechanism agent: A2 algorithm itself.
[0085] State space: includes risk level (L1-L4), feature bits that contribute ≥30% to the attribution report, business type, environmental performance indicators, and compliance coefficient K.
[0086] Action space: Selectable range of compliance verification thresholds (including high-risk instruction verification threshold 0-100 points, behavior deviation verification threshold 0.3-0.8 points, and adversarial similarity verification threshold 50-90 points).
[0087] Reward function: R=F×λ (λ=1.0). R is positive when the compliance rate increases, resource consumption decreases, and the false judgment rate decreases, and negative when these factors increase.
[0088] Learning process: The Q-value table is updated using the Q-Learning algorithm after training with the verification-feedback data of the past 3 months. After the Q-table is trained offline, it is updated on-chain to take effect, and the verification threshold corresponding to high contribution features is optimized first.
[0089] (5) Verification threshold adjustment factor and risk level factor: Level L4 verification threshold × 0.8 (strict), Level L3 × 0.9, Level L2 × 1.0, Level L1 × 1.1 (lenient).
[0090] Compliance factor: K=2 × 0.9 (strict), K=1.5 × 1.0, K=1 × 1.1 (lenient).
[0091] Performance overhead factor: ×1.1 when CPU utilization is ≥80% (lenient), ×0.9 when CPU utilization is ≤60% (strict).
[0092] Risk attribution factor: If the contribution of a certain feature in the attribution report is ≥30%, the corresponding verification threshold is ×0.95 (strictly targeted); if the contribution of a certain feature is ≤5%, the corresponding verification threshold is ×1.05 (appropriately relaxed).
[0093] (6) Output results: optimal five-dimensional compliance verification threshold (including adversarial risk verification threshold), graded early warning triggering strategy (such as whether to push SMS alarm for L3 level risk), attribution summary, synchronized to smart contract audit module.
[0094] (7) Degradation strategy: When the algorithm fails to execute, it maintains operation by using the initial compliance verification threshold and triggers an alarm; if it does not recover within 1 hour, it switches to the industry default threshold (such as high-risk instruction verification threshold = 80 points, adversarial similarity verification threshold = 70 points); if the adversarial module of the A1 algorithm fails, the A2 algorithm will fix the adversarial similarity threshold to 70 points.
[0095] Based on the predictive-Pareto adaptive verification strategy (A3 algorithm) (1) Algorithm objective: Utilize the network and system load prediction capability + multi-objective Pareto front optimization to achieve a 5-minute look-ahead scheduling and globally optimal synchronization scheme, while ensuring the consistency of cross-environment audit data and minimizing synchronization delay, bandwidth consumption and CPU / memory usage.
[0096] (2) Input parameters (3) The time series prediction model adopts a univariate LSTM model with a training window of 2 hours and a prediction step size of 5 minutes. It predicts the network status and system load respectively: Communication quality score in the next 5 minutes: (t+5) = LSTM_Q(past 120 data points), the calculation logic is 80 - latency / 100 × 20 - packet loss rate × 2 - bandwidth utilization rate × 0.5, the range is 0-10 minutes.
[0097] Resource sufficiency in the next 5 minutes: (t+5) = LSTM_R(past 120 data points), the calculation logic is (1-CPU utilization) × 0.4 + (1-memory utilization) × 0.3 + (1-I / O utilization) × 0.3, the range is 0-1.
[0098] Prediction validity determination: When the prediction error MAE ≤ 5%, the prediction result is uploaded to drive scheduling; otherwise, the prediction data is discarded and an alarm is triggered.
[0099] (4) Multi-objective Pareto optimization objective function (four objectives): Maximize consistency contribution F1=wi×synchronization success rate (wi is the task risk weight, L4=1, L3=0.8, L2=0.5, L1=0.2); Minimize synchronization delay F2=average synchronization time; Minimize bandwidth peak F3=peak bandwidth usage during synchronization; Minimize resource usage F4=average CPU + memory usage during synchronization.
[0100] Algorithm and execution logic: The NSGA-II multi-objective evolutionary algorithm is adopted, with the following parameters: population size 100, crossover probability 0.9, mutation probability 0.1, and number of iterations 200. The output is a Pareto optimal solution set containing 8-12 schemes.
[0101] The system defaults to selecting the scheme that is "closest to the ideal point" (minimum Euclidean distance); administrators can manually switch personalized strategies on the web interface, including "consistency priority", "bandwidth saving priority" and "efficiency priority".
[0102] (5) Dynamic synchronization frequency and priority basic synchronization frequency (based on risk level) Forward scheduling strategy (based on prediction results) prediction ≤4 or ≤0.3 (imminent congestion / resource shortage): L1 / L2 level synchronization frequency × 0.8, L3 / L4 level maintains base frequency and reserves resources in advance; prediction ≥8 and ≥0.8 (high-quality communication / abundant resources): L3 / L4 level synchronization frequency × 0.6 (if L4 level is compressed to 3 minutes / time), L1 / L2 level frequency × 1.2 (if L1 level is compressed to 1.6 hours / time); predict peak business periods: complete the synchronization of non-core high-risk data one synchronization window in advance, and only retain the synchronization tasks of core business-related audit data during peak periods.
[0103] Task priority calculation: Priority formula: P = 0.6 × P0 + 0.4 × [(1- / 10)×0.5+(1- [)×0.5], where P0 is the basic priority (L4=10, L3=8, L2=5, L1=3).
[0104] (6) The consistency verification logic adopts a three-way comparison mechanism of "hash value + timestamp + risk level". If the three are consistent, the data is considered to have passed; if they are inconsistent, resynchronization is triggered (up to 3 times, with an interval of 60 seconds). If the retry fails, an alarm is triggered and the data is marked as abnormal. For the predicted network fluctuation period, the core data verification coverage is increased to 100%, and the non-core data coverage is increased to 80%, reducing the risk of synchronization failure.
[0105] (7) Degradation strategy prediction module failure: fall back to the "instantaneous network status + system load" perception strategy and maintain the basic synchronization frequency; NSGA-II algorithm timeout: directly use the basic synchronization frequency to execute the synchronization task, trigger an alarm to notify the administrator, and ensure that cross-scenario data synchronization is not interrupted.
[0106] The three algorithms interact logically: if the risk level output by A1 plus the adversarial similarity σ is ≥80 points, or if the contribution of a certain feature in the attribution report is ≥30%, the A2 algorithm is immediately triggered to recalculate the verification threshold; at the same time, the verification threshold corresponding to the high contribution feature is multiplied by an additional 0.95 to enhance the accuracy of verification.
[0107] L4 / L3 level synchronous tasks carry adversarial similarity σ values. In the Pareto objective F1, the A3 algorithm assigns this value to the task's weight × 1.2, prioritizing consistency of high-risk adversarial data. The risk level of A1 is similar to that of A3. , The forecast results together drive the forward adjustment of the synchronization frequency.
[0108] If a certain type of operation is deemed non-compliant by A2 ≥ 3 times within 24 hours, the threat sensitivity coefficient αi of the corresponding risk dimension of A1 will be increased by 0.1 (maximum 1.5), and the dynamic weight will be adjusted accordingly; at the same time, the sequence of operations will be written into A1's "adversarial attack library" to support online incremental learning.
[0109] When the cross-scenario synchronization failure rate is ≥10%, A2 relaxes the compliance verification threshold for non-core businesses by 5%; the resource shortage prediction warning pushed by A3 triggers A2 to relax the verification threshold for non-core businesses in advance to reduce system resource consumption.
[0110] If the cross-scenario data anomaly rate of a certain operation and maintenance environment is ≥5%, the threat sensitivity coefficient α4+0.1 of the risk dimension of the A1 environment will be increased to improve the risk monitoring sensitivity of the environment.
[0111] All attribution reports for L3 / L4 level risks are broadcast via on-chain events. The A2 and A3 algorithms read the high-contribution features in the reports and use them to dynamically adjust the verification threshold and scheduling weight, forming a self-evolving closed loop of "interpretation-optimization".
[0112] The advantages of this invention are: Audit data is tamper-proof, significantly improving credibility: Based on a blockchain evidence storage engine, audit data (including five-dimensional risk data, adversarial similarity, and attribution reports) is encrypted and uploaded to the blockchain using SM3 hashing. The block records are permanently immutable and can serve as legal evidence for compliance and accountability, achieving 100% data credibility. After application by a financial institution, it successfully traced three operational misoperation incidents, and the log credibility was recognized by regulatory authorities.
[0113] Proactive risk awareness significantly improves the detection rate of advanced threats: The A1 algorithm, through dual-channel analysis and adversarial simulation, accurately identifies advanced threats such as slow APT attacks and malicious "compliance" operations by internal personnel, achieving a detection rate of 96%. Risk warning response time has been reduced from hours to seconds, and the incidence of risk events has decreased by more than 80%. After being applied to a government platform, it proactively intercepted 5 unauthorized operations and 2 malicious log erasure attempts, preventing the leakage of core data.
[0114] Highly interpretable, compliance traceability efficiency doubled: SHAP attribution reports clearly define the basis for risk assessment, reducing the average administrator processing time to 9 minutes; reports are automatically uploaded to the blockchain for evidence storage and connected to the regulatory system, meeting the "Explanation of Judgment Basis" requirements, resulting in a 100% first-time pass rate for regulatory audits. After implementation, a bank reduced compliance audit time from 15 days to 1 day, improving audit efficiency by 93.3%.
[0115] Multi-objective optimization decision-making, perfect balance between compliance and efficiency: The A2 algorithm, combined with attribution reports, dynamically adjusts thresholds to balance compliance, performance overhead and operational efficiency. The compliance rate is increased to 99.8%, system resource consumption is reduced by 30%, and the normal operation and maintenance misjudgment rate is reduced to below 0.1%, solving the pain point of traditional solutions where "strict compliance leads to low efficiency".
[0116] Highly efficient cross-scenario collaboration and outstanding consistency assurance capabilities: The A3 algorithm achieves forward scheduling through prediction and Pareto optimization. In network jitter scenarios, cross-scenario data consistency is achieved one synchronization window ahead, bandwidth peak usage decreases by 18%, and CPU peak usage decreases by 12%. The cross-scenario data consistency rate is improved to 99.8%, solving the problems of "data silos" and short-sighted strategies.
[0117] Strong compliance adaptability, meeting the needs of multiple industries: Supports multiple industry compliance standards such as Cybersecurity Classified Protection 2.0, SOX, GDPR, etc., with explicit configuration of the compliance coefficient K. Smart contracts can quickly adapt to personalized compliance rules, and compliance reports are automatically generated and connected with regulators, adapting to the needs of different industries such as finance, government affairs, and healthcare.
[0118] Combining self-evolution capabilities with high availability: The three algorithms achieve self-evolution through a two-way feedback closed loop, and the misjudgment rate during normal operation and maintenance is reduced by 0.3 percentage points within 30 days of going live; the consortium blockchain nodes adopt N=4 / f=1 Byzantine fault-tolerant configuration, and all three algorithms are equipped with comprehensive degradation strategies, with overall system availability ≥99.99% and support for 24 / 7 uninterrupted auditing.
[0119] Taking the operation and maintenance audit requirements of a large commercial bank (multi-cloud architecture: Alibaba Cloud + AWS, hybrid environment: physical machine + container + blockchain payment node) as an example, the bank needs to cover operation and maintenance scenarios such as SSH operation, cloud API call, and blockchain contract interaction, and meet the Level 3 compliance requirements of the Information Security Protection 2.0. The implementation process is as follows: (1) Initialization deployment stage (2-3 weeks) Consortium chain deployment: Deploy 4 Hyperledger Fabric consensus nodes (N=4 / f=1), adopt the PBFT consensus algorithm (block time 3 seconds), configure SM3 / SM4 / SM2 national cryptographic algorithms; divide data into shards (sharded according to cloud platform / physical machine / blockchain environment), test block storage and query efficiency (single log query ≤ 1 second); ensure the storage and query performance of core information such as attribution reports and adversarial data.
[0120] Data collection module deployment: The data collection agent was deployed on 100 physical machines / virtual machines, configured to connect to Alibaba Cloud ActionTrail and AWS CloudTrail APIs, deployed Fluentd to collect container logs, and connected to the blockchain node log interface; the test data collection success rate was 99.95%, ensuring the complete collection and encrypted transmission of data such as operation sequences and adversarial characteristics in compliance with regulations; and connected to external threat intelligence sources to obtain a database of 47 attack pattern TTPs.
[0121] Parameter and permission configuration: Compliance coefficient K=2 (meeting Level 3 of the Information Security Protection Standard 2.0), initial compliance verification thresholds (high-risk instructions = 80 points, behavior deviation = 0.6, adversarial similarity = 70 points). A1 algorithm initial weights (permission compliance 25%, behavior abnormality 30%, instruction risk 20%, environmental risk 10%, adversarial risk 15%), A2 target weights (α=0.6, β=0.2, γ=0.2), A3 LSTM model training window = 2 hours, NSGA-II configuration (population 100, crossover 0.9, mutation 0.1, iteration 200). Permission matrix: administrators (3 people), auditors (5 people), operations and maintenance personnel (20 people), with permission isolation and operation logging; configure visualization platform policy switching permissions (only administrators can operate). Regulatory integration: complete two-way authentication with the CFCA regulatory database, test the success rate of interface calls such as auditLogUpload and complianceReportSubmit to 100%, and ensure the compliant transmission of parameters such as attribution report ID and adversarial similarity.
[0122] (2) Data collection and evidence storage during daily operation: The multi-source collection module collects operational data and adversarial features such as SSH commands, cloud API calls, and blockchain contract interactions in real time. After standardization processing, the data is encrypted and uploaded to the blockchain using SM3 to generate an evidence storage number. The average daily number of audit records uploaded to the blockchain is 100,000+, and the query response time is ≤0.8 seconds.
[0123] Algorithm-linked execution: The A1 algorithm analyzes operation data every 10 seconds, calculates δ and σ values through dual channels, dynamically adjusts weights, and outputs the risk level; if it is an L3 / L4 level risk, it generates an attribution report and uploads it to the blockchain immediately; the A2 algorithm calculates the compliance verification threshold every hour and optimizes the high contribution feature threshold in conjunction with the attribution report; the A3 algorithm predicts the network / resource status every 5 minutes and outputs the Pareto optimal scheduling scheme, and when the communication quality is excellent, the L4 level data synchronization frequency is increased to 3 minutes / time.
[0124] Risk warning case: At 2:00 AM, an operations engineer executed the high-risk command "sudosu" from an unknown IP address. The A1 algorithm detected a deviation of δ=0.85 and a similarity of σ=88% with adversarial sequences of the "privilege bypass" type, with a total score of 89.75 points (Level 4), triggering a level jump; an attribution report was generated, clearly identifying the four major contribution characteristics; the A2 algorithm immediately adjusted the verification threshold (high-risk command = 70 points, adversarial similarity = 65 points), and the smart contract triggered a tiered alarm (pushed to the security manager and administrator). The operation was not blocked, but complete audit evidence (including the attribution report) was recorded.
[0125] Compliance audit: Auditors can filter logs by “operator + time range + risk level” through the visualization platform, and support the tracing of evidence numbers to the blockchain explorer to view the original records and attribution reports; the system automatically generates a compliance audit report every month, which can be exported and uploaded to the CFCA regulatory database with one click, with a compliance rate of 99.8%.
[0126] (3) Handling of violations during the emergency response phase: An operator modified the core database configuration without authorization. The A1 algorithm determined the permission compliance risk to be 100 points and the adversarial similarity σ=75%, with a total score of 92 points (Level 4), and generated an attribution report. The A2 algorithm verified that it was non-compliant, and the smart contract triggered an alarm and automatically blocked the operation. The violation audit record (including operation evidence, attribution report, and compliance clause basis) was generated and stored on the blockchain. The administrator traced the entire operation process through the platform and completed the rectification within 2 hours.
[0127] Cross-scenario synchronization anomaly: Increased network latency between Alibaba Cloud and AWS, predicted by A3 algorithm. =3.5、 =0.25 (imminent congestion and resource shortage), automatically reduce the synchronization frequency of L1 / L2 level non-core audit data (from 30 minutes / time to 24 minutes / time), switch L4 level risk data to the backup dedicated line channel and reserve resources, and maintain a synchronization success rate of 99.9%; automatically restore the original frequency after the network recovers.
[0128] Algorithm downgrade operation: Due to a failure of the external threat intelligence source interface, the A1 algorithm's adversarial module was unable to obtain the attack sequence library, and the adversarial risk weight was automatically reduced to 5%, while the weights of the other four dimensions were scaled up proportionally to maintain the core risk detection capability; during this period, the administrator fixed the interface failure, the algorithm returned to normal, and the audit process was not affected.
[0129] (4) Implementation effect data (3 months later) The implementation results show that this system effectively solves the pain points of traditional operation and maintenance audit data such as unreliability, delayed early warning, and poor compliance adaptation. Through the linkage of blockchain evidence storage and intelligent algorithms, it realizes the automation, intelligence, and credibility of the entire audit process, meets the compliance and security needs of key industries such as finance and government, and has the value for large-scale promotion.
Claims
1. A blockchain-based immutable operation and maintenance audit system, characterized in that: The system comprises a multi-source audit data acquisition module, a blockchain evidence storage engine, a smart contract audit module, a cross-scenario collaborative audit module, and a visual audit platform. These modules work together according to pre-defined logic to form a closed-loop audit system. The multi-source audit data acquisition module is compatible with cloud-native, physical machine, container, and blockchain multi-operation and maintenance environments. It collects traditional operation and maintenance data, API calls, smart contract interactions, and cross-chain operation and maintenance command data from these environments. Simultaneously, it collects adversarial characteristic data related to the operation data. A built-in dynamic risk perception model analyzes the collected operation data and adversarial characteristic data, outputting risk levels and structured attributions. The report states that the blockchain evidence storage engine is based on a consortium blockchain architecture and uses national cryptographic algorithms to encrypt the collected operational data, adversarial feature data, risk levels, and structured attribution reports. The encrypted, complete audit data is then stored on the blockchain and a unique evidence storage number is generated. Data sharding storage, either by day or by the operational environment, improves the efficiency of audit data retrieval. The smart contract audit module includes rule contracts, permission contracts, and early warning contracts, integrating a multi-objective optimization decision-making model. Using risk levels, structured attribution reports, and historical audit data as inputs, it dynamically adjusts compliance verification thresholds and executes compliance verification through smart contracts. The cross-scenario collaborative audit module constructs a cross-environment data synchronization channel based on the XCM protocol, integrates adaptive verification strategies, and optimizes cross-environment data synchronization strategies using audit data, evidence numbers, and risk levels stored in the blockchain evidence engine as input to ensure consistency of audit data across scenarios. The visual audit platform uses evidence numbers generated by the blockchain evidence engine, compliance verification results from the smart contract audit module, and structured attribution reports as core data, providing audit log query, risk alarm display, and compliance report generation functions, and connects to the regulatory database to complete data submission. The audit process is as follows: initial deployment, multi-source data collection and dynamic risk perception model analysis, encrypted on-chain evidence storage of audit data, dynamic adjustment of thresholds and compliance verification by a multi-objective optimization decision model, optimization of cross-scenario synchronization using adaptive verification strategies, visual compliance audit, and permanent archiving of audit data.
2. The blockchain-based immutable operation and maintenance audit system according to claim 1, characterized in that, The multi-source audit data collection module adopts corresponding collection methods for different operation and maintenance environments and operation types: traditional operation and maintenance operations of physical machines and virtual machines are carried out by deploying a lightweight collection agent and transmitting data through an encrypted channel; API call operations of the cloud platform are carried out by calling the open API of the cloud platform to pull logs periodically; operations in the container environment are carried out by mounting the container log directory and collecting logs through the Fluentd tool; smart contract interaction operations in the blockchain environment are collected by connecting to the blockchain node log interface. Cross-chain operation and maintenance instructions are collected by listening to the XCM protocol cross-chain message channel, and all collected data ensures that adversarial feature data corresponding to the operation data is obtained synchronously.
3. The blockchain-based immutable operation and maintenance audit system according to claim 1, characterized in that, The dynamic risk perception model employs a dual-channel analysis logic: Channel 1 calculates the deviation between the operation and maintenance and the preset behavioral baseline, which is constructed based on the operator's historically frequently used IPs, operation times, high-frequency commands, and operation objects; Channel 2 calculates the adversarial similarity between the current operation and known attack sequences using the DTW algorithm, where the known attack sequences originate from externally connected threat intelligence sources; the model configures initial weights for five dimensions: permission compliance, abnormal behavior, command risk, environmental risk, and adversarial risk, and dynamically adjusts the weights of each dimension in conjunction with external threat intelligence, quantifies and calculates the total score, and maps it to risk levels L1-L4. When the risk level is L3 or L4, a structured attribution report is automatically generated.
4. The blockchain-based immutable operation and maintenance audit system according to claim 1, characterized in that, The blockchain evidence storage engine adopts the Hyperledger Fabric consortium blockchain architecture, configured with 4 consensus nodes and following the Byzantine fault tolerance mechanism, and uses the PBFT consensus algorithm to generate blocks; the national cryptographic algorithms include the SM3 hash algorithm, the SM4 data encryption algorithm, and the SM2 signature verification algorithm; data sharding storage is divided by day or operation and maintenance environment, with audit data from the past 3 months being stored as hot data first, and audit data from 3 months ago being archived as cold data. At the same time, it supports real-time cross-node backup of audit data and daily offline encrypted backup to ensure data security and recoverability.
5. The blockchain-based immutable operation and maintenance audit system according to claim 1, characterized in that, The multi-objective optimization decision-making model takes compliance, performance overhead, and operational interference as its three major optimization objectives. It takes the risk level, structured attribution report, and historical audit data output by the dynamic risk perception model as input, and learns the optimal adjustment strategy through reinforcement learning algorithm. For features that contribute significantly to the structured attribution report, the corresponding compliance verification thresholds are tightened; for features that contribute less significantly, the corresponding compliance verification thresholds are appropriately relaxed. At the same time, the thresholds are dynamically adjusted in conjunction with the preset compliance coefficient K and the CPU and memory usage of the operation and maintenance environment. The adjusted thresholds take effect through on-chain hot updates.
6. The blockchain-based immutable operation and maintenance audit system according to claim 1, characterized in that, The adaptive verification strategy analyzes historical network performance indicators and system resource status data using an LSTM model to predict communication quality and resource sufficiency for the next 5 minutes. It employs the NSGA-II algorithm to construct a multi-objective optimization problem and outputs a Pareto optimal synchronization scheduling scheme. A basic synchronization frequency is set according to the risk level, with higher risk levels resulting in higher synchronization frequencies. The synchronization frequency and task priority are dynamically adjusted based on the prediction results. Consistency verification is achieved by comparing the hash value, timestamp, and risk level of cross-environment audit data. If all three are consistent, the process passes; otherwise, resynchronization is triggered.
7. The blockchain-based immutable operation and maintenance audit system according to claim 1, characterized in that, The smart contract auditing module supports hot contract updates, adapts to compliance standards across multiple industries, and can adapt to compliance requirements in different scenarios by configuring different compliance coefficients K. The smart contract has reentrancy attack protection and permission isolation mechanisms. It triggers tiered alarms based on compliance verification results, pushes high-risk non-compliant operations to the security manager and administrator, and pushes ordinary non-compliant operations to the administrator. It supports automatic blocking of high-risk non-compliant operations and generates violation audit records for on-chain storage.
8. The blockchain-based immutable operation and maintenance audit system according to claim 1, characterized in that, The structured attribution report uses JSON format and includes the total score corresponding to the risk level and the marginal contribution of each input feature to the total score. It clarifies the core attribution factors, including high-risk instruction hits, deviation from behavioral baseline, adversarial similarity matching, and abnormal operation time periods. After the report is stored on the blockchain through the blockchain evidence storage engine, it is associated with the corresponding evidence storage number and serves as the basis for compliance verification, risk tracing, and regulatory review. The original record of the report can be queried on the blockchain explorer by the evidence storage number.
9. The blockchain-based immutable operation and maintenance audit system according to claim 1, characterized in that, The dynamic risk perception model, multi-objective optimization decision-making model and adaptive verification strategy are all configured with a degradation strategy: when the dynamic risk perception model crashes as a whole, it maintains the most recent effective risk level for 30 minutes. If it does not recover within 30 minutes, it switches to the default risk level L2. When the multi-objective optimization decision model fails to execute, it will continue to run using the initial compliance verification threshold. If it fails to recover within 1 hour, it will switch to the industry default threshold. When the prediction module of the adaptive verification strategy fails, it reverts to the instantaneous network state and system load awareness strategy, maintains the basic synchronization frequency, and triggers corresponding alarms to notify the administrator to troubleshoot the fault when all models are degraded.
10. The blockchain-based immutable operation and maintenance audit system according to claim 1, characterized in that, The visual auditing platform connects to the regulatory database via HTTPS+JSON-RPC2.0 protocol, and uses two-way certificate authentication and HMAC-SHA256 signature mechanism to ensure data transmission security. The signature parameters include request timestamp, interface name, compliance coefficient K and random number. It supports three core interfaces: audit log upload, compliance report submission, and high-risk event push. It can export a compliance audit report containing a structured attribution report and judgment basis with one click, meeting the regulatory authorities' verification requirements for audit basis.
Citation Information
Cited By
Statistical data compliance evidence storage system and method based on distributed storage and blockchain
CN122529917A
