Data information operation and maintenance management method of chain trust mechanism based on AI

Abstract

The invention relates to the technical field of data information operation and maintenance management, and discloses a data information operation and maintenance management method based on an AI chain trust mechanism, and the method comprises the steps: obtaining multi-source heterogeneous data through a distributed collection terminal composed of an edge calculation node, an Internet of Things sensor and a third-party interface module; and after format conversion, a data trust value is calculated by using a multi-dimensional trust evaluation model containing a source reliable item, a content consistent item and the like, then the data trust value and a historical trust record in a block chain distributed account book are subjected to association verification, and a trusted data identifier is generated through a consensus algorithm. And constructing an on-chain storage index based on the data type, the timestamp and the source node, encapsulating the data by adopting a hierarchical structure of a metadata layer, a content layer and a verification layer, and outputting a storage address. A monitoring layer of the dynamic operation and maintenance control model identifies abnormity in real time, a decision-making layer generates an operation and maintenance strategy in combination with credible data identification and historical records, and an execution layer triggers repairing and updating operation through an intelligent contract, so that on-chain credible operation and maintenance management of data information is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of data information operation and maintenance management, and specifically to a data information operation and maintenance management method based on an AI-based on-chain trust mechanism. Background Art

[0002] In the current field of data information operation and maintenance management, with the development of technologies such as the Internet of Things and the Industrial Internet, the collection and management of multi-source heterogeneous data faces many challenges. On the one hand, traditional data collection methods cannot efficiently integrate structured, unstructured, and semi-structured data generated by multiple source terminals such as edge computing nodes, IoT sensors, and third-party interfaces. Data with different protocols and formats are prone to compatibility issues during transmission and conversion, resulting in compromised data integrity. On the other hand, there is a lack of a systematic multi-dimensional model for the credibility assessment of multi-source data. Existing solutions often only evaluate from a single dimension (such as data source or content consistency), without fully considering dynamic factors such as the historical contribution of source nodes and data update frequency, making it difficult to accurately reflect the true credibility of the data.

[0003] In terms of data storage and management, traditional hierarchical storage structures fail to effectively integrate the characteristics of blockchain technology. They lack an on-chain storage index system categorized by data type, timestamp, and source node. This results in inefficient data retrieval and difficulty verifying the association of historical trust records. Furthermore, blockchain consensus algorithms face challenges in verifying data trust, including cumbersome node identity verification processes and a lack of trust transfer and decay mechanisms. This prevents rapid on-chain confirmation of data integrity and credibility.

[0004] Existing data operations management models lack dynamic response mechanisms. The monitoring layer relies on manually pre-set rules to identify anomalies in real-time data streams, making it difficult to adapt to the ever-changing nature of data. The decision-making layer lacks the ability to generate policies based on historical operations records and trusted data identifiers, resulting in insufficiently targeted and effective operations strategies. The execution layer lacks automated remediation mechanisms driven by smart contracts, resulting in poor execution efficiency and traceability of operations instructions. These issues make existing solutions unable to meet the trusted operations requirements of data information in complex business scenarios, necessitating a systematic solution that integrates AI and blockchain technologies. Summary of the Invention

[0005] The purpose of the present invention is to provide a data information operation and maintenance management method based on an AI-based on-chain trust mechanism to solve the problems raised in the above background technology.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a data information operation and maintenance management method based on an AI-based on-chain trust mechanism, the method comprising:

[0007] Acquire multi-source heterogeneous data information through a distributed data acquisition terminal, the distributed data acquisition terminal including an edge computing node, an IoT sensor, and a third-party interface module; perform credibility analysis on the multi-source heterogeneous data information based on a trust assessment model to obtain a data trust value; associate the data trust value with historical trust records on the chain, which are stored in a blockchain distributed ledger; perform node confirmation on the association verification result based on a consensus algorithm to generate a trusted data identifier;

[0008] Building an on-chain storage index based on the trusted data identifier, wherein the on-chain storage index is classified by data type, timestamp, and source node, and uses a hierarchical storage structure to encapsulate data information into blocks, wherein the hierarchical storage structure includes a metadata layer, a content layer, and a verification layer; outputting a data storage address based on the on-chain storage index;

[0009] A dynamic operation and maintenance control model is established according to the data storage address, and the dynamic operation and maintenance control model includes a monitoring layer, a decision layer and an execution layer, wherein the monitoring layer identifies abnormal conditions based on real-time data streams, the decision layer generates operation and maintenance strategies based on trusted data identifiers and historical operation and maintenance records, and the execution layer triggers data repair or update operations based on smart contracts; operation and maintenance instructions are output through the dynamic operation and maintenance control model to realize on-chain trust mechanism operation and maintenance management of data information.

[0010] Preferably, performing credibility analysis on the multi-source heterogeneous data information based on the trust evaluation model to obtain the data trust value includes:

[0011] Acquire multi-source heterogeneous data information, including structured table data, unstructured text data, and semi-structured log data; construct a feature extraction space based on the data type, and construct an evaluation dimension based on the historical contribution of the data source node, data update frequency, and data completeness;

[0012] A multi-dimensional trust evaluation function is constructed based on the feature extraction space and the evaluation dimension. The multi-dimensional trust evaluation function includes a source reliability item, a content consistency item, a timeliness matching item, and an association verification item. The source reliability item is calculated by the historical trust value of the data source node, the content consistency item is calculated by the matching degree between the data content and the standard template, the timeliness matching item is calculated by the deviation value between the data generation time and the business demand time, and the association verification item is calculated by the logical consistency of the data and the related data on the chain;

[0013] Constructing a trust assessment model, the trust assessment model includes an initial parameter set, a feature mapping layer, and a result output layer. The initial parameter set includes initial weight values ​​for each assessment dimension. The feature mapping layer converts raw data features into assessment feature vectors through linear transformation. The result output layer calculates the data trust value based on a weighted summation method.

[0014] The initial parameter set is updated using an iterative optimization method, the weight value of each evaluation dimension is adjusted through a gradient descent algorithm, and the data trust value is calculated based on the updated parameter set. The data trust value includes source credibility, content credibility, timeliness credibility, and association credibility.

[0015] Preferably, the data trust value is associated with the historical trust record on the chain for verification, the historical trust record on the chain is stored in the blockchain distributed ledger, and the association verification result is node-confirmed based on the consensus algorithm to generate a trusted data identifier, including:

[0016] Construct a query interface for historical trust records on the chain. The query interface uses the data source node, data type, and time range as query conditions to obtain historical trust records associated with the current data information;

[0017] Constructing a trust association matrix based on the historical trust records, wherein the trust association matrix includes trust transfer coefficients between different source nodes, trust influence coefficients between different data types, and trust decay coefficients within different time ranges;

[0018] A consensus algorithm is used to perform node confirmation on the association verification results. The consensus algorithm includes a node identity verification step, a data hash comparison step, and a trust value weighted voting step. The node identity verification step verifies the legitimacy of participating nodes through digital signatures. The data hash comparison step verifies data integrity by calculating the match between the hash value of the current data and the hash value stored on the chain. The trust value weighted voting step performs a weighted calculation of the voting results based on the historical contribution of each node.

[0019] A verification result is output based on the consensus algorithm, and the verification result includes fully credible, partially credible, and untrustworthy. A corresponding trusted data identifier is generated according to the verification result, and the trusted data identifier includes a verification level, a verification node set, and a verification timestamp.

[0020] Preferably, the monitoring layer performs abnormal state identification based on real-time data streams including:

[0021] Segmenting the real-time data stream using sliding window technology to divide the real-time data stream into continuous data segments within a time window, where the length of the time window is positively correlated with the data update frequency;

[0022] Building an abnormal feature library based on data types, the abnormal feature library includes range-crossing features of numerical data, format error features of text data, and logical conflict features of log data;

[0023] Perform feature extraction on each data segment, including the maximum, minimum, and average values ​​of numerical data, the frequency of keyword occurrences in text data, and the continuity of operation sequences in log data.

[0024] Matching the extracted features with the abnormal feature library and calculating the feature matching degree, wherein the feature matching degree is calculated by the absolute value of the difference between the feature value and the abnormal feature threshold;

[0025] Constructing abnormal state judgment rules, which include single feature matching judgment, multi-feature joint judgment, and continuous time window judgment, wherein the single feature matching judgment is used to identify single feature abnormalities, the multi-feature joint judgment is used to identify simultaneous abnormalities of multiple features, and the continuous time window judgment is used to identify continuous abnormalities;

[0026] An abnormal state identifier is output based on the abnormal state determination rule, where the abnormal state identifier includes an abnormal type, an abnormal location, and an abnormal time.

[0027] Preferably, the decision-making layer generates an operation and maintenance strategy based on trusted data identification and historical operation and maintenance records, including:

[0028] Construct an operation and maintenance strategy knowledge base, which includes basic repair strategies corresponding to different exception types, update priorities corresponding to different data types, and processing permissions corresponding to different source nodes;

[0029] Obtaining trusted data identification and historical operation and maintenance records. The trusted data identification includes the verification level, verification node set, and verification timestamp. The historical operation and maintenance records include the processing results of the same exception type, the update frequency of the same data type, and the processing feedback of the same source node;

[0030] A strategy matching model is constructed based on the operation and maintenance strategy knowledge base, trusted data identifiers, and historical operation and maintenance records. The strategy matching model calculates the matching score of each basic repair strategy using a weighted matching method. The weight coefficient of the weighted matching method is positively correlated with the severity of the anomaly type, the importance of the data type, and the credibility of the source node.

[0031] Parameters of the basic repair strategy with the highest matching score are adjusted, including the time delay of the repair operation, the execution node of the repair operation, and the verification method of the repair operation. Based on the parameter adjustment, a customized operation and maintenance strategy is generated, which includes the repair target, repair steps, and repair verification requirements.

[0032] Preferably, the execution layer triggers data repair or update operations based on smart contracts, outputs operation and maintenance instructions through the dynamic operation and maintenance control model, and implements the on-chain trust mechanism operation and maintenance management of data information, including:

[0033] Establishing trigger conditions for smart contracts, including the type of abnormal status identification, the level of trusted data identification, and the verification results of customized operation and maintenance strategies;

[0034] Construct the execution logic of the smart contract, which includes data repair logic and data update logic. The data repair logic corrects abnormal data by calling a preset repair function, and the data update logic obtains the latest data and overwrites the old data by calling the data acquisition interface.

[0035] The execution process of the smart contract is recorded on the chain, including the execution time, execution node, execution result and execution log;

[0036] Constructing an output format for operation and maintenance instructions, wherein the output format includes instruction type, instruction content, execution node, and execution time;

[0037] Based on the execution results of the smart contract, operation and maintenance instructions are generated, and the operation and maintenance instructions are sent to the target execution node through the blockchain broadcast mechanism, realizing the on-chain trust mechanism operation and maintenance management of data information.

[0038] Preferably, obtaining multi-source heterogeneous data information through a distributed data acquisition terminal includes:

[0039] The edge computing node obtains the production equipment operation data through the industrial bus protocol, which includes the Modbus protocol, PROFIBUS protocol and CAN bus protocol;

[0040] IoT sensors acquire environmental monitoring data through wireless transmission protocols, including LoRa, ZigBee, and NB-IoT.

[0041] The third-party interface module obtains external system data through the API interface, which includes RESTful API, SOAP API and GraphQL API;

[0042] The data information obtained by each terminal is format converted, and data information of different protocols and formats is converted into a unified JSON format to generate multi-source heterogeneous data information.

[0043] Preferably, building a trust assessment model includes:

[0044] A uniformly distributed initialization parameter set is used, wherein the parameter set includes initial weight values ​​of a source reliability item, a content consistency item, a timeliness matching item, and an association verification item;

[0045] The original data features are converted into evaluation feature vectors through a feature mapping layer, and the feature mapping layer compresses the original data features by a linear transformation matrix;

[0046] Calculate the data trust value based on the evaluation feature vector and the initial parameter set, and calculate the data trust value by summing the product of the feature value of each evaluation dimension and the corresponding weight value;

[0047] Calculate the error between the data trust value and the actual trustworthy label, where the actual trustworthy label is determined through manual verification or historical verification results;

[0048] The weight values ​​in the parameter set are updated using a gradient descent algorithm, wherein the gradient descent algorithm adjusts the weight values ​​along the negative gradient direction by calculating the partial derivative of the error value with respect to the weight value;

[0049] Repeat the feature mapping, trust value calculation and parameter update steps until the error value is less than the preset threshold, and obtain the trained trust assessment model.

[0050] Preferably, using a consensus algorithm to perform node confirmation on the association verification result includes:

[0051] Participating nodes authenticate their identities through digital certificates issued by trusted CAs, which contain the node's public key, node identity information, and certificate validity period.

[0052] Calculate the hash value of the current data information, which is obtained by hashing the data content using the SHA-256 algorithm;

[0053] Obtain the historical hash value stored on the chain, which has the same source node, data type, and time range as the current data information;

[0054] Compare the current hash value with the historical hash value. If the match is greater than the preset threshold, the voting phase begins. Otherwise, the data is marked as incomplete.

[0055] Participating nodes vote on the data trust value, and the voting value is weighted according to the node's historical contribution, which is calculated by the number of times the node has participated in consensus and the accuracy of the verification results;

[0056] The weighted voting results are counted. If the number of votes in favor exceeds the preset ratio, a trusted data identifier is generated; otherwise, it is marked as untrustworthy data.

[0057] Preferably, constructing abnormal state determination rules includes:

[0058] Single feature matching judgment: If the matching degree of a certain feature is greater than the single feature threshold, it is judged as a single feature anomaly;

[0059] Multi-feature joint judgment: If the matching degree of two or more features is greater than the multi-feature threshold at the same time, it is judged as a multi-feature joint anomaly;

[0060] Continuous time window judgment: If the matching degree of the same feature in N consecutive time windows is greater than the continuous threshold, it is judged as a continuous anomaly;

[0061] Different priorities are set for different exception types. The priority is determined by the degree of impact of the exception on the business. Single-feature exceptions are low priority, multi-feature combined exceptions are medium priority, and continuous exceptions are high priority.

[0062] An abnormality status identifier is generated based on the abnormality type and priority, wherein the abnormality status identifier includes an abnormality type code, a priority level, and an abnormality detail description.

[0063] Compared with the prior art, the present invention has the following beneficial effects:

[0064] This invention uses distributed data acquisition terminals to achieve standardized collection and format conversion of multi-source heterogeneous data. Edge computing nodes, IoT sensors, and third-party interface modules acquire data through industrial bus protocols, wireless transmission protocols, and API interfaces, respectively, and convert the data into JSON format. This ensures the comprehensiveness and compatibility of data collection and lays the foundation for subsequent processing. Based on a multi-dimensional trust assessment model, an evaluation function is constructed from dimensions such as source reliability, content consistency, timeliness matching, and association verification. The gradient descent algorithm is used to iteratively optimize weight parameters to achieve precise quantitative analysis of data credibility, effectively improving the accuracy of data trust value calculation.

[0065] The trust value of data is verified by correlating it with historical trust records in the blockchain's distributed ledger. By constructing a trust association matrix and consensus algorithm (including node identity verification, hash comparison, and weighted voting steps), on-chain consensus confirmation of data credibility is achieved. This generates a trusted data identifier that includes the verification level, node set, and time stamp, ensuring the decentralization and immutability of data trust verification. A hierarchical storage structure (metadata layer, content layer, and verification layer) based on on-chain storage indexes encapsulates blocks based on data type, timestamp, and source node, significantly improving the standardization of data storage and retrieval efficiency. Data integrity is also ensured through hash verification.

[0066] The monitoring layer of the dynamic operation and maintenance control model uses sliding window technology and anomaly feature library to realize segmented processing of real-time data streams and multi-dimensional anomaly identification, and combines single-feature, multi-feature and continuous time window judgment rules to accurately locate the type and location of anomalies; the decision-making layer builds a strategy matching model based on the operation and maintenance strategy knowledge base, trusted data identification and historical operation and maintenance records, and generates customized operation and maintenance strategies through weighted matching to improve the pertinence and scientific nature of operation and maintenance decisions; the execution layer triggers data repair or update operations through smart contracts to realize the automatic execution and on-chain recording of operation and maintenance instructions, greatly improving the efficiency and traceability of operation and maintenance management. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Figure 1 This is a working principle diagram of the data information operation and maintenance management method based on the AI-based on-chain trust mechanism described in the present invention;

[0068] Figure 2 Flowchart for credibility analysis of multi-source heterogeneous data;

[0069] Figure 3 A flowchart generated for on-chain association verification and trusted data identification;

[0070] Figure 4 Flowchart generated for the operation and maintenance strategy. DETAILED DESCRIPTION

[0071] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0072] See also Figures 1-4 The present invention relates to a data information operation and maintenance management method based on an AI-based on-chain trust mechanism, and the specific implementation steps are as follows:

[0073] Acquire multi-source heterogeneous data information through a distributed data acquisition terminal, the distributed data acquisition terminal including an edge computing node, an IoT sensor, and a third-party interface module; perform credibility analysis on the multi-source heterogeneous data information based on a trust assessment model to obtain a data trust value; associate the data trust value with historical trust records on the chain, which are stored in a blockchain distributed ledger; perform node confirmation on the association verification result based on a consensus algorithm to generate a trusted data identifier;

[0074] Building an on-chain storage index based on the trusted data identifier, wherein the on-chain storage index is classified by data type, timestamp, and source node, and uses a hierarchical storage structure to encapsulate data information into blocks, wherein the hierarchical storage structure includes a metadata layer, a content layer, and a verification layer; outputting a data storage address based on the on-chain storage index;

[0075] A dynamic operation and maintenance control model is established according to the data storage address, and the dynamic operation and maintenance control model includes a monitoring layer, a decision layer and an execution layer, wherein the monitoring layer identifies abnormal conditions based on real-time data streams, the decision layer generates operation and maintenance strategies based on trusted data identifiers and historical operation and maintenance records, and the execution layer triggers data repair or update operations based on smart contracts; operation and maintenance instructions are output through the dynamic operation and maintenance control model to realize on-chain trust mechanism operation and maintenance management of data information.

[0076] Example 1:

[0077] When analyzing the credibility of multi-source, heterogeneous data to determine its trust value, it's necessary to obtain this information. This data is rich and diverse, encompassing structured tabular data, unstructured text data, and semi-structured log data. Different types of data have distinct characteristics and structures. For example, structured tabular data is typically organized in rows and columns, with well-defined fields and formats. Unstructured text data, such as documents and articles, lacks a fixed structure. Semi-structured log data falls somewhere in between, with a definite format but less rigid uniformity.

[0078] After obtaining this data, a feature extraction space must be constructed based on the data type. Because different types of data contain different information and features, designing a specialized feature extraction method for each data type can more accurately extract the key features of the data. At the same time, it is also necessary to construct evaluation dimensions based on the historical contribution of the data source node, the frequency of data updates, and the completeness of the data. The historical contribution of the data source node reflects the performance and reliability of the node in the past data provision process; the frequency of data updates reflects the timeliness and freshness of the data; and the data completeness indicates the comprehensiveness and completeness of the data content. These three dimensions provide important basis for evaluating data credibility from different aspects.

[0079] On the basis of constructing the feature extraction space and evaluation dimensions, a multi-dimensional trust evaluation function is further constructed. This function includes source reliability items, content consistency items, timeliness matching items, and association verification items. Among them, the calculation of the source reliability item depends on the historical trust value of the data source node. By analyzing and processing the historical trust value, the reliability of the current data source can be inferred; the content consistency item is determined by calculating the matching degree between the data content and the standard template. The standard template is formulated according to the normal specifications and requirements of this type of data. The higher the matching degree, the more the data content meets the requirements; the timeliness matching item is calculated by the deviation value between the data generation time and the business demand time. The smaller the deviation value, the higher the time matching degree of the data, and the more it can meet the time requirements of the business; the association verification item is calculated by examining the logical consistency of the data with the related data on the chain. Logically consistent data is more credible.

[0080] Next, a trust assessment model needs to be constructed. This model consists of an initial parameter set, a feature mapping layer, and a result output layer. The initial parameter set contains initial weights for each assessment dimension, which preliminarily determine the importance of each dimension in the credibility calculation. The feature mapping layer converts raw data features into assessment feature vectors through linear transformation, thus converting complex raw data features into a form more suitable for evaluation calculations. The result output layer calculates the data trust value based on a weighted summation method. The final preliminary data trust value is obtained by multiplying the feature values ​​of each assessment dimension by the corresponding weights and then summing them.

[0081] When building the trust assessment model, a uniform distribution is used to initialize the parameter set. Initial weights are set for the source reliability, content consistency, timeliness match, and association verification criteria. The linear transformation matrix in the feature mapping layer compresses the raw data features and converts them into evaluation feature vectors. This step reduces the data dimension, removes redundant information, and improves computational efficiency. Based on the resulting evaluation feature vectors and the initial parameter set, the data trust value is calculated by multiplying the eigenvalues ​​of each evaluation dimension by their corresponding weights.

[0082] To improve model accuracy, it's necessary to calculate the error between the data's trust value and the actual trusted label. The actual trusted label can be determined through manual verification or by referring to historical verification results. It represents the true credibility of the data. By calculating the error, we can understand the gap between the current model's predictions and the actual situation. A gradient descent algorithm is used to update the weights in the parameter set. This algorithm calculates the partial derivative of the error with respect to the weight value to determine the direction and magnitude of weight adjustment. The weights are adjusted along the negative gradient, gradually reducing the error.

[0083] The process of feature mapping, trust value calculation, and parameter update is repeated. Each iteration adjusts the weights based on the new error value, bringing the model's predictions closer to the actual trusted labels. This process continues until the error value falls below a preset threshold, indicating that the model's accuracy has met the required standards. This completes the training of the trust assessment model.

[0084] Utilizing the trained trust assessment model and an updated parameter set, the acquired multi-source heterogeneous data is processed to calculate a data trust value. This trust value encompasses source credibility, content credibility, timeliness credibility, and relevance credibility, comprehensively reflecting the data's credibility from multiple dimensions. Analysis of these credibility indicators provides an important reference for subsequent data processing and application, ensuring the high reliability and trustworthiness of the data used.

[0085] Example 2:

[0086] When correlating the data trust value with the historical trust record on the chain and generating a trusted data identifier, it is necessary to build a query interface for the historical trust record on the chain. This query interface uses the data source node, data type, and time range as query conditions, where the data source node can clearly identify the provider of the data, the data type is used to limit the category of the query data, and the time range can filter out historical records within a specific time period. Through the combined query of these three dimensions, the historical trust records associated with the current data information can be accurately obtained. For example, when processing the operating data of a production equipment, the historical trust records of this type of data within the time period can be queried by specifying the edge computing node corresponding to the equipment as the source node, the data type as equipment operating data, and the time range as the past week.

[0087] After obtaining historical trust records, a trust association matrix is ​​constructed based on these records. This matrix includes the trust transfer coefficients between nodes from different sources, the trust influence coefficients between different data types, and the trust decay coefficients within different timeframes. The trust transfer coefficients between nodes from different sources reflect the ability of trust relationships to be transferred between nodes. For example, if nodes A and B have long-term, stable interactions and a high level of trust, the transfer coefficient between them is likely to be large. The trust influence coefficients between different data types reflect the degree to which the credibility of one type of data affects other types of data. For example, changes in the credibility of production data may have a correlated impact on the credibility of quality inspection data. The trust decay coefficients within different timeframes account for the impact of time on trust values. Records older than the current time are likely to have a greater trust decay coefficient, and their reference value to current data is relatively reduced.

[0088] A consensus algorithm is used to verify the results of the association verification process. The specific steps are as follows: First, participating nodes must authenticate themselves using digital certificates. Digital certificates are issued by trusted CAs and contain the node's public key, node identity, and certificate validity period. Verification involves checking whether the certificate is valid, whether the identity information matches the node's actual state, and the validity of the public key. This ensures the legitimacy of participating nodes and prevents unauthorized nodes from participating in the consensus process.

[0089] Calculate the hash value of the current data. The hash value is calculated by hashing the data content using the SHA-256 algorithm. This algorithm is highly secure and unique, and can convert data of any length into a fixed-length hash value. Any slight change in the data content will result in a significant difference in the hash value. At the same time, obtain the historical hash value stored on the chain. This historical hash value has the same source node, data type, and time range as the current data information, ensuring that the historical records of the same type of data are being compared.

[0090] Compare the current hash value with the historical hash value. If the match is greater than the preset threshold, it indicates that the current data and the historical data are highly consistent in content, the data integrity is good, and the voting phase can be entered. If the match does not reach the preset threshold, the data is marked as incomplete and further verification is required to determine if there are any problems with the data source or transmission process.

[0091] During the voting phase, participating nodes vote on the data's trust value, with voting weights based on the node's historical contribution. Historical contribution is calculated by the number of times a node has participated in consensus and the accuracy of its verification results. Nodes with high consensus participation and high accuracy receive a relatively higher voting weight, reflecting trust and recognition for active and reliable nodes. For example, if node X participated in consensus 100 times with a verification accuracy of 95%, and node Y participated in consensus 50 times with a verification accuracy of 80%, node X's historical contribution is higher than node Y's, and its voting weight is also greater.

[0092] When counting the weighted voting results, if the number of votes in favor exceeds the preset ratio, it indicates that the majority of trusted nodes recognize the trust value of the current data, and a trusted data identifier is generated at this time; if the number of votes in favor does not reach the preset ratio, it is marked as untrustworthy data and requires further review or processing.

[0093] Verification results are output based on the consensus algorithm and are categorized as fully trusted, partially trusted, and untrusted. Fully trusted means the data has been verified in terms of source, content, and timeliness, resulting in a high degree of credibility. Partially trusted means the data may have issues in certain aspects, but overall still has some reference value. Untrusted means the data has failed verification and cannot be trusted. Based on the verification results, a corresponding trusted data identifier is generated, which includes information such as the verification level, the set of validating nodes, and the verification timestamp. The verification level clarifies the data's credibility, the set of validating nodes records the nodes that participated in the verification and approved the data, and the verification timestamp marks the specific time when the verification was completed. Together, these pieces of information constitute the data's trustworthiness, providing a reliable basis for subsequent data storage, use, and operation. Through this rigorous series of steps, the associated verification of the data trust value and the generation of the trusted data identifier are achieved, ensuring the credibility and reliability of the on-chain data and laying a solid foundation for the subsequent steps of the entire data information operation and maintenance management method.

[0094] Example 3:

[0095] When the monitoring layer identifies anomalies based on real-time data streams, it uses sliding window technology to segment the real-time data streams. Real-time data streams are continuous and dynamically changing, and sliding window technology can divide them into continuous data segments within a time window. The length of the time window setting is positively correlated with the frequency of data updates. That is, the higher the data update frequency, the shorter the time window length to ensure timely capture of data changes. The lower the data update frequency, the longer the time window length to avoid the increased processing burden caused by an excessive number of data segments due to a short window. For example, for data updated once a second, a 10-second time window can be set to divide the data stream into continuous data segments every 10 seconds. For data updated every minute, a 5-minute time window might be set.

[0096] Build an anomaly feature library based on data type. Different types of data have different anomaly manifestations, so corresponding anomaly features need to be built for numerical data, text data, and log data. For numerical data, the anomaly feature library includes range-out-of-bounds features, meaning that the data value exceeds the normal value range of that type of data. For example, temperature data should be between 0-100°C, but if it reaches 150°C, it is considered a range out-of-bounds. For text data, it includes format error features, such as the requirement that text must be in a specific JSON format. Missing fields or confusing formats are considered an anomaly. For log data, it includes logical conflict features. For example, the operation sequence should follow the order of logging in first and then operating. If there is a record of operating without logging in, it is a logical conflict. The construction of these anomaly features is based on the analysis and summary of normal business scenarios for various types of data, ensuring that the anomaly feature library can fully cover possible anomalies.

[0097] Perform feature extraction on each data segment. For numerical data segments, extract the maximum, minimum, and average values. These statistics can reflect the overall distribution and fluctuation of the data. For example, by calculating the maximum, minimum, and average values ​​of the device voltage data within a certain time period, it can be determined whether the voltage fluctuates within the normal range. For text data segments, extract the frequency of keyword occurrence. The occurrence of keywords can reflect the rationality of the text content. For example, in the text of a fault report, if the frequency of keywords such as "abnormal" and "fault" increases abnormally, it may indicate the presence of an abnormal situation. For log data segments, extract the continuity of the operation sequence. The continuity of the operation sequence is an important basis for determining whether the log data is normal. If the operation sequence is interrupted or chaotic, it may mean that there is a problem with the system operation.

[0098] The extracted features are matched against the anomaly feature library and the feature matching degree is calculated. The feature matching degree is calculated by the absolute difference between the feature value and the anomaly threshold. The smaller the absolute difference, the higher the matching degree between the feature and the anomaly feature, and the greater the likelihood of an anomaly in the data. For example, if the feature value of a numerical data item is 120, the anomaly threshold is 100, and the absolute difference is 20, then this can be converted into a matching degree using a certain conversion method. If another feature value is 130 and the absolute difference is 30, then the matching degree is higher and the data is more likely to be an anomaly.

[0099] Establishing abnormal state determination rules is a key step. These rules include single-feature matching determination, multi-feature joint determination, and continuous time window determination. Single-feature matching determination is used to identify anomalies in a single feature. If the matching degree of a feature exceeds the single-feature threshold, it is considered a single-feature anomaly. For example, if the matching degree of a format error feature of a piece of text exceeds the set threshold, the text can be determined to have a format anomaly. Multi-feature joint determination is used to identify situations where multiple features are abnormal at the same time. If the matching degrees of two or more features are simultaneously greater than the multi-feature threshold, it is considered a multi-feature joint anomaly. For example, if a data segment simultaneously has a value range out-of-bounds and a log logic conflict, the severity of the multi-feature joint anomaly is higher than that of a single-feature anomaly. Continuous time window determination is used to identify persistent anomalies. If the matching degree of the same feature in N consecutive time windows is greater than the continuous threshold, it is considered a persistent anomaly. The value of N can be set according to specific business needs and data characteristics. For example, setting N to 3 means that if a feature has abnormal matches in three consecutive time windows, it is considered a persistent anomaly.

[0100] Different priorities are set for different types of anomalies, and the priority is determined based on the degree of impact of the anomaly on the business. Single-feature anomalies have a relatively small impact on the business and are set to low priority; multi-feature joint anomalies may affect multiple business links and are set to medium priority; continuous anomalies indicate that the abnormal situation persists and may have a serious impact on the business and are set to high priority. For example, a single-feature out-of-bounds error in a certain numerical data may only affect the accuracy of a single data point and is set to low priority; a multi-feature joint anomaly with both a numerical out-of-bounds error and a log logic conflict may affect some functions of the system and is set to medium priority; if the numerical out-of-bounds error continues to occur in multiple consecutive time windows, it may cause system failure and is set to high priority.

[0101] An exception status identifier is generated based on the exception type and priority. The exception status identifier includes an exception type code, priority level, and a detailed description of the exception. The exception type code concisely identifies the type of exception, facilitating system recording and processing. The priority level clearly defines the urgency of the exception, providing a basis for subsequent O&M decisions. The detailed description of the exception details the specific circumstances of the exception, including its location, time, and characteristics. For example, an exception type code of "NT001" indicates a value out-of-bounds, with a high priority level. The detailed description of the exception would read, "During the period from 10:00 AM to 10:10 AM on June 24, 2025, the voltage data of device A exceeded the normal upper limit of 100V for three consecutive time windows, reaching a maximum of 120V, representing a continuous voltage out-of-bounds anomaly." Through this series of steps, the monitoring layer can accurately identify abnormal conditions in real-time data streams, providing reliable exception information for subsequent decision-making and execution of dynamic O&M control models. This ensures timely detection and resolution of data anomalies, ensuring the normal operation and reliability of data information.

[0102] Example 4:

[0103] When the decision-making layer generates operation and maintenance strategies based on trusted data identification and historical operation and maintenance records, it is necessary to build an operation and maintenance strategy knowledge base. The construction of this knowledge base is based on the analysis and summary of various data anomalies, including basic repair strategies corresponding to different anomaly types, update priorities corresponding to different data types, and processing permissions corresponding to different source nodes. For example, for out-of-range anomalies of numerical data, the basic repair strategy may be to make corrections based on the mean of historical normal data; for format error anomalies of text data, the basic repair strategy may be to call a preset format verification tool for repair. Among different data types, the update priority of real-time operation data of production equipment is usually higher than that of historical statistical report data; among different source nodes, high-trustworthiness edge computing nodes that have been verified multiple times have relatively high data processing permissions.

[0104] Obtain trusted data identification and historical operation and maintenance records. The trusted data identification includes information such as the verification level, the verification node set, and the verification timestamp. For example, in the trusted data identification of a piece of environmental monitoring data, the verification level is fully trusted, the verification node set includes 3 IoT sensor nodes and 2 edge computing nodes, and the verification timestamp is 09:30:15 on June 24, 2025. The historical operation and maintenance records include the processing results of the same exception type, the update frequency of the same data type, and the processing feedback of the same source node. For example, for the same type of temperature data out-of-bounds exception, the historical operation and maintenance records may record the correction strategy adopted during the last processing, the data recovery status after processing, and the feedback from the source node after processing.

[0105] A strategy matching model is constructed based on an operation and maintenance strategy knowledge base, trusted data identifiers, and historical operation and maintenance records. This model calculates the matching score for each basic repair strategy using a weighted matching method. The weight coefficient of the weighted matching method is positively correlated with the severity of the anomaly type, the importance of the data type, and the credibility of the source node. Taking the equipment operation data of a factory as an example, if a multi-feature joint anomaly in the motor speed data occurs (including both numerical value out-of-bounds and log logic conflicts), the severity of this anomaly type is high, and the corresponding weight coefficient is large. Equipment operation data is critical production data and is highly important, so the weight coefficient is also increased accordingly. If this data comes from an edge computing node with a high historical contribution, its credibility is high, and the weight coefficient is also increased. When calculating the matching score, the degree of match between each basic repair strategy and the current anomaly is multiplied by the corresponding weight coefficient to obtain the matching score for each strategy.

[0106] Parameters are adjusted for the basic repair strategy with the highest matching score. These adjustments include the repair operation's time delay, the execution node for the repair operation, and the verification method for the repair operation. For example, if the basic repair strategy with the highest matching score is to perform a mean correction on the motor speed data, the repair operation's time delay might be set to 5 minutes, depending on the actual situation, to avoid impacting equipment operation during peak production periods. Edge computing nodes with high processing authority are selected as execution nodes to ensure the accuracy and efficiency of the repair operation. Verification is set to dual verification, with initial verification performed by the execution node followed by a second, high-trustworthiness node for reconciliation.

[0107] Generate a customized operation and maintenance strategy based on parameter adjustment. The customized operation and maintenance strategy includes repair goals, repair steps, and repair verification requirements. Continuing with the above-mentioned motor speed data anomaly as an example, the repair goal is set to restore the motor speed to the normal range (1500-1800 rpm); the specific repair steps are as follows: Step 1, the designated edge computing node obtains the normal speed data of the motor for nearly 1 hour and calculates the average; Step 2, the average is used to correct the abnormal data; Step 3, the corrected data is sent to the verification node. The repair verification requirements are: The verification node must complete the verification within 10 minutes after receiving the corrected data. The verification content includes whether the data is within the normal range and whether it is consistent with the logic of other related data (such as current data). If the verification passes, the repair is recorded as successful. If it fails, the repair steps are re-executed.

[0108] Throughout the process of generating an operation and maintenance strategy, it is necessary to constantly reference the verification level and verification node information in the trusted data identifier. If the verification level of the trusted data identifier is partially trusted, it may be necessary to add additional verification steps to the operation and maintenance strategy or improve the accuracy of the repair operation. The trustworthiness of the nodes in the verification node set will also affect the selection of execution nodes, and high-trustworthiness nodes in the verification node set will be preferentially selected as execution nodes for repair operations. At the same time, the processing results and feedback in historical operation and maintenance records also serve as an important reference. If the historical records show that a certain type of anomaly frequently recurs after using a certain repair strategy, the strategy will be adjusted or preventive measures will be added when the current strategy is formulated.

[0109] For example, when processing a batch of text-based quality inspection report data formatting errors, the corresponding basic repair strategy in the operation and maintenance strategy knowledge base invokes a format conversion tool for repair. Reviewing the trusted data identifier revealed that the data's verification level was partially trusted, and one of the verification nodes reported data with potential missing fields. Historical operation and maintenance records revealed that similar anomalies had previously resulted in incomplete repairs after using this basic repair strategy, with the feedback indicating the need for manual review. Therefore, when constructing the strategy matching model, in addition to calculating the matching score for the basic repair strategy, the necessity for manual review in historical feedback was also factored in, adjusting the matching score weighting. When adjusting the parameters of the basic repair strategy with the highest matching score, the repair operation delay was set to allow for a preliminary manual check for missing fields. The execution node was selected to have both format conversion capabilities and manual review permissions. A manual review step was added to the repair verification requirements to ensure that the repaired quality inspection report data was formatted correctly and had complete fields.

[0110] In this way, the decision-making layer can fully combine the information of trusted data identification and historical operation and maintenance records, and generate customized operation and maintenance strategies for specific abnormal situations based on the operation and maintenance strategy knowledge base through strategy matching and parameter adjustment, providing clear and feasible guidance for the operations of the subsequent execution layer, ensuring the effectiveness and pertinence of data information operation and maintenance management, and achieving accurate processing and efficient repair of data anomalies.

[0111] Example 5:

[0112] When triggering data repair or update operations based on smart contracts at the execution layer and outputting operation and maintenance instructions through a dynamic operation and maintenance control model to implement on-chain trust mechanism operation and maintenance management of data information, a complete closed loop must be formed from the establishment of trigger conditions, construction of execution logic, on-chain recording, instruction format design, and instruction generation and sending. Taking the abnormal handling of equipment operation data in a smart manufacturing factory as an example, the specific implementation method is as follows:

[0113] Establish trigger conditions for smart contracts, which must include the type of abnormal status identification, the level of trusted data identification, and the verification results of customized operation and maintenance strategies. For example, when the monitoring layer identifies that the real-time pressure data of a stamping equipment has a continuous time window abnormality (such as the pressure value continuously exceeds the normal upper limit of 100MPa in three consecutive 10-minute windows), and the trusted data identification verification level of this data is "partially trusted" (indicating that there are historical trust fluctuations in the data source node), and the customized operation and maintenance strategy generated by the decision-making layer does not fully meet the repair requirements after preliminary verification, the trigger conditions of the smart contract are activated. At this time, the abnormal status identification The abnormal type code (such as "PT002" indicates that the pressure exceeds the limit), the priority level (high priority), and the abnormality details (equipment number, time range, etc.) are passed into the smart contract as trigger parameters.

[0114] Construct the execution logic of the smart contract, including data repair logic and data update logic. The data repair logic targets the abnormal data itself. For example, if the pressure data of the stamping equipment mentioned above is abnormal, the smart contract will call the preset repair function. This function is based on the pressure data distribution of the equipment during normal operation (such as the mean of 120MPa and the standard deviation of 5MPa in the past week). It performs regression correction on the abnormal data and adjusts the pressure value that exceeds the threshold to a reasonable range. The data update logic is used to obtain the latest data. If the repair logic cannot effectively correct the data (such as a sensor hardware failure causing continuous erroneous collection), the smart contract will call the data collection interface of the IoT sensor to re-acquire the pressure data of the device and overwrite the old data stored on the chain. The execution logic also includes an error handling mechanism. For example, when the interface call fails, it will automatically switch to the backup sensor node for data collection.

[0115] During smart contract execution, every step of the operation must be recorded on-chain. This record includes the execution time, execution node, execution result, and execution log. For example, if a smart contract is triggered at 14:30:22 on June 24, 2025, the edge computing node numbered "ECN-007" will perform a data repair operation, with the result "Pressure data repair successful, current value 98MPa." The execution log will detail the repair function call parameters, data repair process, and intermediate calculation results. These records are stored using blockchain's distributed ledger technology, ensuring immutability and traceability, facilitating subsequent audits and troubleshooting.

[0116] Constructing the output format of operational instructions is key to ensuring accurate communication. The format includes the instruction type, instruction content, execution node, and execution time. For example, for data repair, the instruction type is marked "DATA_REPAIR," and the instruction content includes the storage address of the abnormal data, repair parameters (such as historical mean and correction threshold), verification requirements, etc. The execution node is designated as "ECN-007," and the execution time is accurate to the second (2025-06-24 14:30:22). If multi-node collaboration is involved, the instruction content will clearly specify the operation sequence and interaction method of each node, such as the edge computing node performing the repair first, followed by the blockchain consensus node performing the hash check.

[0117] After generating an O&M instruction based on the smart contract's execution results, the instruction is sent to the target execution node via the blockchain broadcast mechanism. For example, once the smart contract completes stress data repair and generates an execution log, the system automatically encapsulates the O&M instruction and broadcasts it to the "ECN-007" node via the P2P network. Upon receiving the instruction, the node first verifies the instruction's digital signature (signed by the system administrator's private key) to ensure it has not been tampered with, and then executes the specific operation according to the instruction's content. In a data update scenario, if the smart contract determines that sensor data needs to be updated, an O&M instruction is sent to the corresponding IoT sensor node (such as "WSN-015"). The instruction includes parameters such as the data collection frequency (e.g., increasing from once per minute to five times per minute), the transmission protocol (NB-IoT), and the timeout retry mechanism (three retries, each with a 10-second interval).

[0118] In actual applications, the combination of trigger conditions can be even more complex. For example, in an environmental monitoring system, if a temperature and humidity sensor data exhibits a single-character anomaly (the temperature value displays -50°C, outside the normal range of -20°C to 60°C), the trusted data identifier's verification level is "untrusted" (the sensor node has recently experienced frequent data anomalies due to communication failures), and the customized O&M strategy requires a node restart before data repair, the smart contract will sequentially trigger the node restart and data repair instructions. The execution logic will first call the hardware control interface to restart the sensor, then wait three minutes before executing the data repair to prevent repair failures due to node operation anomalies. On-chain records will detail the restart operation time, the execution node (gateway node "GN-003"), the restart log (e.g., "Device restart successful, firmware version V2.1.0"), and the subsequent data repair process.

[0119] The generation of operational instructions also needs to consider the blockchain's consensus mechanism. For example, when an executing node completes a data repair, the operational instruction will include the node's signature and the data hash value. After being broadcast to the blockchain network, other consensus nodes will verify the repair results (such as recalculating the data hash value and comparing it with the historical hash value stored on the chain). Only if more than two-thirds of the consensus nodes confirm that the repair operation is correct will the repair operation be written into the blockchain and a new block generated. If the verification fails (for example, the repaired data still contains logical conflicts), the smart contract will automatically trigger a secondary repair process and adjust the repair parameters (such as increasing the time window of historical data).

[0120] Through this implementation approach, the execution layer can leverage the automated triggering and execution capabilities of smart contracts, combined with the distributed consensus and immutable nature of blockchain, to translate operational strategies into specific instructions, ensuring accurate execution and full traceability. From repairing pressure data on stamping equipment to restarting nodes for environmental monitoring sensors, every operation at the execution layer is closely centered around data credibility. Through a dynamic operational control model, full lifecycle management of on-chain data is achieved, ultimately establishing a trust mechanism for data information operations and ensuring the system can efficiently and reliably complete repairs and updates in the face of various data anomalies.

[0121] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "includes," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0122] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A data information operation and maintenance management method based on an AI-based on-chain trust mechanism, characterized in that: include: Acquire multi-source heterogeneous data information through a distributed data acquisition terminal, which includes edge computing nodes, IoT sensors, and third-party interface modules; Conducting credibility analysis on the multi-source heterogeneous data information based on a trust evaluation model to obtain a data trust value; performing correlation verification on the data trust value and the historical trust record on the chain, wherein the historical trust record on the chain is stored in the blockchain distributed ledger; performing node confirmation on the correlation verification result based on a consensus algorithm to generate a trusted data identifier; Building an on-chain storage index based on the trusted data identifier, wherein the on-chain storage index is classified by data type, timestamp, and source node, and uses a hierarchical storage structure to encapsulate data information into blocks, wherein the hierarchical storage structure includes a metadata layer, a content layer, and a verification layer; outputting a data storage address based on the on-chain storage index; A dynamic operation and maintenance control model is established according to the data storage address, and the dynamic operation and maintenance control model includes a monitoring layer, a decision layer and an execution layer, wherein the monitoring layer identifies abnormal conditions based on real-time data streams, the decision layer generates operation and maintenance strategies based on trusted data identifiers and historical operation and maintenance records, and the execution layer triggers data repair or update operations based on smart contracts; operation and maintenance instructions are output through the dynamic operation and maintenance control model to realize on-chain trust mechanism operation and maintenance management of data information.

2. The data information operation and maintenance management method based on the AI-based on-chain trust mechanism according to claim 1 is characterized in that: The credibility of the multi-source heterogeneous data information is analyzed based on the trust evaluation model to obtain the data trust value including: Acquire multi-source heterogeneous data information, including structured table data, unstructured text data, and semi-structured log data; construct a feature extraction space based on the data type, and construct an evaluation dimension based on the historical contribution of the data source node, data update frequency, and data completeness; A multi-dimensional trust evaluation function is constructed based on the feature extraction space and the evaluation dimension. The multi-dimensional trust evaluation function includes a source reliability item, a content consistency item, a timeliness matching item, and an association verification item. The source reliability item is calculated by the historical trust value of the data source node, the content consistency item is calculated by the matching degree between the data content and the standard template, the timeliness matching item is calculated by the deviation value between the data generation time and the business demand time, and the association verification item is calculated by the logical consistency of the data and the related data on the chain; Constructing a trust assessment model, the trust assessment model includes an initial parameter set, a feature mapping layer, and a result output layer. The initial parameter set includes initial weight values ​​for each assessment dimension. The feature mapping layer converts raw data features into assessment feature vectors through linear transformation. The result output layer calculates the data trust value based on a weighted summation method. The initial parameter set is updated using an iterative optimization method, the weight value of each evaluation dimension is adjusted through a gradient descent algorithm, and the data trust value is calculated based on the updated parameter set. The data trust value includes source credibility, content credibility, timeliness credibility, and association credibility.

3. The data information operation and maintenance management method based on the AI-based on-chain trust mechanism according to claim 1 is characterized in that: The data trust value is associated with the historical trust record on the chain for verification. The historical trust record on the chain is stored in the blockchain distributed ledger. The association verification result is confirmed by nodes based on the consensus algorithm to generate a trusted data identifier, including: Construct a query interface for historical trust records on the chain. The query interface uses the data source node, data type, and time range as query conditions to obtain historical trust records associated with the current data information; Constructing a trust association matrix based on the historical trust records, wherein the trust association matrix includes trust transfer coefficients between different source nodes, trust influence coefficients between different data types, and trust decay coefficients within different time ranges; A consensus algorithm is used to perform node confirmation on the association verification results. The consensus algorithm includes a node identity verification step, a data hash comparison step, and a trust value weighted voting step. The node identity verification step verifies the legitimacy of participating nodes through digital signatures. The data hash comparison step verifies data integrity by calculating the match between the hash value of the current data and the hash value stored on the chain. The trust value weighted voting step performs a weighted calculation of the voting results based on the historical contribution of each node. A verification result is output based on the consensus algorithm, and the verification result includes fully credible, partially credible, and untrustworthy. A corresponding trusted data identifier is generated according to the verification result, and the trusted data identifier includes a verification level, a verification node set, and a verification timestamp.

4. The data information operation and maintenance management method based on the AI-based on-chain trust mechanism according to claim 1 is characterized in that: The monitoring layer identifies abnormal conditions based on real-time data streams, including: Segmenting the real-time data stream using sliding window technology to divide the real-time data stream into continuous data segments within a time window, where the length of the time window is positively correlated with the data update frequency; Building an abnormal feature library based on data types, the abnormal feature library includes range-crossing features of numerical data, format error features of text data, and logical conflict features of log data; Perform feature extraction on each data segment, including the maximum, minimum, and average values ​​of numerical data, the frequency of keyword occurrences in text data, and the continuity of operation sequences in log data. Matching the extracted features with the abnormal feature library and calculating the feature matching degree, wherein the feature matching degree is calculated by the absolute value of the difference between the feature value and the abnormal feature threshold; Constructing abnormal state judgment rules, which include single feature matching judgment, multi-feature joint judgment, and continuous time window judgment, wherein the single feature matching judgment is used to identify single feature abnormalities, the multi-feature joint judgment is used to identify simultaneous abnormalities of multiple features, and the continuous time window judgment is used to identify continuous abnormalities; An abnormal state identifier is output based on the abnormal state determination rule, where the abnormal state identifier includes an abnormal type, an abnormal location, and an abnormal time.

5. The data information operation and maintenance management method based on the AI-based on-chain trust mechanism according to claim 1 is characterized in that: The decision-making layer generates operation and maintenance strategies based on trusted data identification and historical operation and maintenance records, including: Construct an operation and maintenance strategy knowledge base, which includes basic repair strategies corresponding to different exception types, update priorities corresponding to different data types, and processing permissions corresponding to different source nodes; Obtaining trusted data identification and historical operation and maintenance records. The trusted data identification includes the verification level, verification node set, and verification timestamp. The historical operation and maintenance records include the processing results of the same exception type, the update frequency of the same data type, and the processing feedback of the same source node; A strategy matching model is constructed based on the operation and maintenance strategy knowledge base, trusted data identifiers, and historical operation and maintenance records. The strategy matching model calculates the matching score of each basic repair strategy using a weighted matching method. The weight coefficient of the weighted matching method is positively correlated with the severity of the anomaly type, the importance of the data type, and the credibility of the source node. Parameters of the basic repair strategy with the highest matching score are adjusted, including the time delay of the repair operation, the execution node of the repair operation, and the verification method of the repair operation. Based on the parameter adjustment, a customized operation and maintenance strategy is generated, which includes the repair target, repair steps, and repair verification requirements.

6. The data information operation and maintenance management method based on the AI-based on-chain trust mechanism according to claim 1 is characterized in that: The execution layer triggers data repair or update operations based on smart contracts, outputs operation and maintenance instructions through the dynamic operation and maintenance control model, and implements the on-chain trust mechanism operation and maintenance management of data information, including: Establishing trigger conditions for smart contracts, including the type of abnormal status identification, the level of trusted data identification, and the verification results of customized operation and maintenance strategies; Construct the execution logic of the smart contract, which includes data repair logic and data update logic. The data repair logic corrects abnormal data by calling a preset repair function, and the data update logic obtains the latest data and overwrites the old data by calling the data acquisition interface. The execution process of the smart contract is recorded on the chain, including the execution time, execution node, execution result and execution log; Constructing an output format for operation and maintenance instructions, wherein the output format includes instruction type, instruction content, execution node, and execution time; Based on the execution results of the smart contract, operation and maintenance instructions are generated, and the operation and maintenance instructions are sent to the target execution node through the blockchain broadcast mechanism, realizing the on-chain trust mechanism operation and maintenance management of data information.

7. The data information operation and maintenance management method based on the AI-based on-chain trust mechanism according to claim 1 is characterized in that: Acquiring multi-source heterogeneous data information through distributed data acquisition terminals includes: The edge computing node obtains the production equipment operation data through the industrial bus protocol, which includes the Modbus protocol, PROFIBUS protocol and CAN bus protocol; IoT sensors acquire environmental monitoring data through wireless transmission protocols, including LoRa, ZigBee, and NB-IoT. The third-party interface module obtains external system data through the API interface, which includes RESTful API, SOAP API and GraphQL API; The data information obtained by each terminal is format converted, and data information of different protocols and formats is converted into a unified JSON format to generate multi-source heterogeneous data information.

8. The data information operation and maintenance management method based on the AI-based on-chain trust mechanism according to claim 2 is characterized in that: Building a trust assessment model includes: A uniformly distributed initialization parameter set is used, wherein the parameter set includes initial weight values ​​of a source reliability item, a content consistency item, a timeliness matching item, and an association verification item; The original data features are converted into evaluation feature vectors through a feature mapping layer, and the feature mapping layer compresses the original data features by a linear transformation matrix; Calculate the data trust value based on the evaluation feature vector and the initial parameter set, and calculate the data trust value by summing the product of the feature value of each evaluation dimension and the corresponding weight value; Calculate the error between the data trust value and the actual trustworthy label, where the actual trustworthy label is determined through manual verification or historical verification results; The weight values ​​in the parameter set are updated using a gradient descent algorithm, wherein the gradient descent algorithm adjusts the weight values ​​along the negative gradient direction by calculating the partial derivative of the error value with respect to the weight value; Repeat the feature mapping, trust value calculation and parameter update steps until the error value is less than the preset threshold, and obtain the trained trust assessment model.

9. The data information operation and maintenance management method based on the AI-based on-chain trust mechanism according to claim 3 is characterized in that: The consensus algorithm used to confirm the associated verification results includes: Participating nodes authenticate their identities through digital certificates issued by trusted CAs, which contain the node's public key, node identity information, and certificate validity period. Calculate the hash value of the current data information, which is obtained by hashing the data content using the SHA-256 algorithm; Obtain the historical hash value stored on the chain, which has the same source node, data type, and time range as the current data information; Compare the current hash value with the historical hash value. If the match is greater than the preset threshold, the voting phase begins. Otherwise, the data is marked as incomplete. Participating nodes vote on the data trust value, and the voting value is weighted according to the node's historical contribution, which is calculated by the number of times the node has participated in consensus and the accuracy of the verification results; The weighted voting results are counted. If the number of votes in favor exceeds the preset ratio, a trusted data identifier is generated; otherwise, it is marked as untrustworthy data.

10. The data information operation and maintenance management method based on the AI-based on-chain trust mechanism according to claim 4 is characterized in that: Building abnormal status judgment rules includes: Single feature matching judgment: If the matching degree of a certain feature is greater than the single feature threshold, it is judged as a single feature anomaly; Multi-feature joint judgment: If the matching degree of two or more features is greater than the multi-feature threshold at the same time, it is judged as a multi-feature joint anomaly; Continuous time window judgment: If the matching degree of the same feature in N consecutive time windows is greater than the continuous threshold, it is judged as a continuous anomaly; Different priorities are set for different exception types. The priority is determined by the degree of impact of the exception on the business. Single-feature exceptions are low priority, multi-feature combined exceptions are medium priority, and continuous exceptions are high priority. An abnormality status identifier is generated based on the abnormality type and priority, wherein the abnormality status identifier includes an abnormality type code, a priority level, and an abnormality detail description.

Images