A power production monitoring data governance and high-value data asset generation method and system based on a storage-computation integrated time sequence library

Abstract

This invention discloses a method and system for power production monitoring data governance and high-value data asset generation based on an in-memory computing time-series database, relating to the field of power data processing technology. The method includes: outputting a time-series dataset in an in-memory computing underlying support environment based on a time-series database; calling a pre-defined lightweight digital twin of power equipment to output a standardized time-series dataset; matching corresponding artificial intelligence algorithms to generate a standardized high-value feature set; constructing an intelligent annotation and multi-dimensional quality rating system to obtain a high-value dataset; classifying and structurally encapsulating the high-value dataset, and storing the encapsulation results in a pre-defined data asset database; and optimizing the data asset database by combining a pre-deployed power industrial control alliance chain and a power business knowledge graph. This invention solves the problems of high computing power loss and low processing efficiency caused by massive data cross-layer migration in traditional architectures, meeting the real-time governance needs of high-concurrency scenarios with tens of millions of monitoring points.

Description

Technical Field

[0001] This invention relates to the field of power data processing technology, and more specifically, to a method and system for power production monitoring data governance and high-value data asset generation based on an integrated storage and computing time-series database. Background Technology

[0002] With the accelerated construction of new power systems and the large-scale grid connection of new energy sources such as wind power and photovoltaics, power production monitoring scenarios are characterized by an exponential increase in the number of monitoring points, a continuous increase in data sampling frequency, and a petabyte-scale explosion in data volume. This has generated massive amounts of equipment operation monitoring data with high concurrency and strong time-series characteristics. Distributed time-series databases, with their high-throughput writes, efficient time-series queries, and distributed scalability, have become the core infrastructure for carrying power production monitoring time-series data. At the same time, the demand for data assetization in the power industry's digital transformation is becoming increasingly urgent. There is a pressing need to transform massive amounts of low-value-density raw production data into reusable, highly reliable, and standardized high-value data assets, providing high-quality data support for power digitalization operations such as load forecasting, anomaly detection, fault diagnosis, equipment health management, and energy efficiency analysis and optimization.

[0003] However, current technical solutions for power production monitoring time-series data governance and asset generation generally suffer from the following three major shortcomings, making it difficult to meet the development needs of new power systems:

[0004] First, the architecture generally adopts a post-processing governance design that separates storage and computation, requiring data governance to repeatedly migrate data between the storage and computation layers. Faced with business scenarios involving high-concurrency writes to tens of millions of measurement points and petabyte-level massive data, this architecture suffers from core bottlenecks such as low governance efficiency, high processing latency, and significant consumption of computing resources, making it unable to support the real-time governance needs of large-scale power production data.

[0005] Second, at the business level, there is a lack of deep integration with the scenario characteristics and business rules of power production monitoring, with most solutions adopting generalized data governance approaches. These solutions cannot adapt to the unique data characteristics and business needs of the power industry, lack differentiated governance strategies for the operating conditions of power equipment, and generate data assets with low value density, making it difficult to directly support AI model training and business application implementation.

[0006] Third, the security aspects do not strictly adhere to the core protection principles of security zoning, dedicated networks, horizontal isolation, and vertical authentication in power industrial control systems, resulting in a disconnect between data security management and governance processes. Existing solutions are insufficient to meet the industrial-grade security compliance requirements of large-scale power production control areas and cannot guarantee the authenticity, integrity, and trustworthiness of data assets.

[0007] To address the aforementioned issues, the industry has conducted relevant technological research. For example, Chinese patent CN112559502B discloses a wind farm data governance system based on a time-series database platform. This system achieves the collection, diagnosis, and storage of wind farm data through communication modules, data diagnosis modules, and data governance modules, which improves data access speed to some extent. However, this solution only targets the single business scenario of wind farms and does not cover the entire scenario of new power systems such as thermal power and centralized monitoring of new energy, thus limiting its scenario adaptability. At the same time, this solution only focuses on the basic data collection and diagnosis stage and does not build an AI-driven end-to-end data governance and high-value data asset generation system, failing to transform raw data into standardized and reusable data assets. Furthermore, this solution adopts a traditional time-series database architecture and does not achieve integrated storage and computing optimization. When facing high-concurrency scenarios with tens of millions of measurement points, data governance requires cross-layer data migration, resulting in significant computing power consumption and high processing latency. Chinese patent CN115238823A discloses a method for standardized governance of time-series data in the energy industry, relying on a data acquisition platform, a data storage management platform, and a data analysis and governance platform to achieve standardized governance of time-series data in the energy industry. However, this solution only targets slowly changing time-series data and cannot adapt to the scenarios of high-frequency sampling and drastically fluctuating fast-changing time-series data in power production monitoring, thus limiting the scope of data governance. Furthermore, the solution adopts a three-layer architecture separating acquisition, storage, and analysis / governance, failing to extend governance capabilities to the storage layer. When dealing with petabyte-scale historical data, it suffers from low batch governance efficiency and excessive resource consumption. In addition, the solution lacks a standardized power data asset encapsulation and management system, and lacks feature engineering and intelligent annotation capabilities specific to power business, making it unable to generate high-value datasets that meet the requirements of AI model training. Chinese patent CN116975138B discloses a method for governing safety production monitoring data, achieving governance of safety production monitoring data through steps such as real-time data acquisition, invalid data removal, and visualization. However, this solution only completes the basic cleaning and validity screening stages of data governance, failing to construct a full-process asset generation system encompassing feature extraction, intelligent labeling, quality rating, and asset encapsulation, thus unable to transform raw data into data assets. Furthermore, the solution employs a post-processing model, lacking deep integration with time-series databases, resulting in latency and resource consumption issues that cannot meet the real-time requirements of large-scale power production control areas during high-concurrency real-time data governance. Additionally, the solution lacks security and compliance design specific to the power industry, failing to integrate data encryption / decryption, access control, and governance processes, making it difficult to meet the security protection requirements of power industrial control systems. Chinese patent CN120216569B discloses an artificial intelligence-based energy and power data asset evaluation method that achieves real-time evaluation of the full lifecycle value of energy data assets by constructing an energy knowledge graph and a multi-dimensional value assessment model.However, this solution focuses on the value assessment of data assets rather than the generation process of data assets. It fails to build a full-chain AI governance system from raw time-series data to high-value datasets, and thus cannot solve the core problems of inconsistent data quality and low value density in power production monitoring. At the same time, the solution only uses the time-series database as a storage medium and does not utilize an integrated storage and computing architecture to achieve the decentralization of governance capabilities. The separation of data processing and storage leads to low processing efficiency and high resource consumption when dealing with massive amounts of time-series data. In addition, the solution is not designed in accordance with the security zoning requirements of the power production control area and cannot adapt to the security and compliance deployment requirements of power industrial control scenarios.

[0008] In summary, how to achieve efficient AI governance of massive time-series data and automated generation of high-value data assets while ensuring high-concurrency read / write, high-reliability storage, and end-to-end security of power time-series data remains a pressing technical challenge in the field of power production monitoring data processing.

[0009] No effective solutions have yet been proposed to address the problems in the relevant technologies. Summary of the Invention

[0010] To address the problems in related technologies, this invention proposes a method and system for power production monitoring data governance and high-value data asset generation based on an integrated storage and computing time-series database, in order to overcome the aforementioned technical problems existing in the current related technologies.

[0011] Therefore, the specific technical solution adopted by the present invention is as follows:

[0012] According to one aspect of the present invention, a method for power production monitoring data governance and high-value data asset generation based on an in-store computing time-series database is provided, the method comprising:

[0013] S1. In the underlying support environment of building a storage-computing integrated system based on a time-series database, the raw time-series data of multi-source power production monitoring is accessed, and storage management and security control are performed synchronously during the access process to output the time-series dataset.

[0014] S2. Based on the time-series dataset, call the preset lightweight digital twin of power equipment to generate a dynamic operating condition baseline, and combine the dual-track detection mechanism to perform anomaly detection and management on the time-series dataset, and output a standardized time-series dataset.

[0015] S3. Match the artificial intelligence algorithm corresponding to the standardized time series dataset, perform feature mining based on the matching results, and use a two-dimensional screening mechanism to screen and optimize the feature mining results to generate a standardized high-value feature set.

[0016] S4. Based on the preset power industry business specifications, construct an intelligent annotation and multi-dimensional quality rating system, and use the intelligent annotation and multi-dimensional quality rating system to annotate and filter the standardized high-value feature set to obtain a high-value dataset.

[0017] S5. Classify and structure high-value datasets, and store the encapsulation results in a pre-defined data asset library. Combine the pre-deployed power industrial control alliance chain and power business knowledge graph to optimize the data asset library, so as to realize the governance of power production monitoring data and the generation of high-value data assets.

[0018] According to another aspect of the present invention, a system for power production monitoring data governance and high-value data asset generation based on an in-store computing time-series database is also provided, the system comprising:

[0019] The data support and acquisition module is used to access the raw time-series data of multi-source power production monitoring in the construction of an in-store computing underlying support environment based on a time-series database, and to simultaneously perform storage management and security control during the access process, and output time-series datasets.

[0020] The anomaly detection and management module is used to generate a dynamic operating condition baseline by calling a preset lightweight digital twin of power equipment based on the time series dataset, and to perform anomaly detection and management on the time series dataset in combination with a dual-track detection mechanism, and output a standardized time series dataset.

[0021] The data filtering and optimization module is used to match the artificial intelligence algorithms corresponding to the standardized time series datasets, perform feature mining based on the matching results, and use a two-dimensional filtering mechanism to filter and optimize the feature mining results to generate a standardized high-value feature set.

[0022] The data asset generation module is used to construct an intelligent labeling and multi-dimensional quality rating system based on the preset power industry business specifications, and to use the intelligent labeling and multi-dimensional quality rating system to label and filter the standardized high-value feature set to obtain a high-value dataset.

[0023] The data asset management module is used to classify and structure high-value datasets and store the encapsulation results in a pre-set data asset library. Combined with a pre-deployed power industrial control alliance chain and power business knowledge graph, the data asset library is optimized to achieve power production monitoring data governance and high-value data asset generation.

[0024] The beneficial effects of this invention are as follows:

[0025] 1. This invention optimizes the underlying architecture of power time-series data governance. It combines the methods of on-site computing at storage nodes of the storage-computing integrated time-series library, the sinking of AI governance capabilities, and the collaborative optimization of read / write and governance. This reduces the end-to-end processing latency of data governance, improves the efficiency of storage-computing collaboration, and solves the problems of large computing power loss and low processing efficiency caused by the cross-layer migration of massive data in traditional architectures. It meets the real-time governance needs of high-concurrency scenarios with tens of millions of measurement points.

[0026] 2. This invention integrates the business rules of the power industry with the entire process of AI data governance. By optimizing the entire chain solution of intelligent preprocessing, power-specific feature engineering, scenario-based intelligent labeling and quality rating, it realizes the automated transformation of raw time-series data into high-value data assets, improves the overall quality pass rate of the dataset, and significantly enhances the value density and business scenario adaptability of power time-series data.

[0027] 3. This invention improves the full lifecycle security management and control system for power time-series data, combining data encryption and decryption, access control, abnormal data identification and processing, data governance, and asset generation into a single process. It accurately adapts to the compliance requirements of security partition protection for power industrial control systems, achieving full-link security protection for data assets and improving the credibility, security, and availability of power production monitoring data and data assets.

[0028] 4. This invention constructs a standardized generation and full lifecycle management mechanism for high-value datasets that conforms to power industry standards. Through the integrated design of classification system, encapsulation specifications, version control, and archiving and reuse, it realizes the traceability and reusability of power data assets, and can directly support the implementation of digital and intelligent business such as power load forecasting, fault diagnosis, and equipment health management. Attached Figure Description

[0029] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0030] Figure 1 This is a flowchart according to an embodiment of the present invention;

[0031] Figure 2 This is a principle block diagram according to an embodiment of the present invention;

[0032] Figure 3 This is a flowchart of a standardized time-series dataset according to an embodiment of the present invention;

[0033] Figure 4 This is an architecture diagram according to an embodiment of the present invention;

[0034] Figure 5 This is a flowchart illustrating a specific application of an embodiment of the present invention.

[0035] In the picture:

[0036] 1. Data support and acquisition module; 2. Anomaly detection and management module; 3. Data filtering and optimization module; 4. Data asset generation module; 5. Data asset management module. Detailed Implementation

[0037] To further illustrate the various embodiments, the present invention provides accompanying drawings, which are part of the disclosure of the present invention. These drawings are mainly used to illustrate the embodiments and can be used in conjunction with the relevant descriptions in the specification to explain the operating principles of the embodiments. With reference to these drawings, those skilled in the art should be able to understand other possible implementation methods and the advantages of the present invention.

[0038] According to embodiments of the present invention, a method and system for power production monitoring data governance and high-value data asset generation based on an in-store computing time-series database are provided.

[0039] The present invention will now be further described in conjunction with the accompanying drawings and specific embodiments, such as... Figure 1 As shown, according to an embodiment of the present invention, a method for power production monitoring data governance and high-value data asset generation based on an in-store computing time-series database includes:

[0040] S1. In the underlying support environment of building a storage-computing integrated system based on a time-series database, the raw time-series data of multi-source power production monitoring is accessed, and storage management and security control are performed synchronously during the access process to output the time-series dataset.

[0041] It should be further explained that, based on a self-developed distributed time-series database, an integrated storage and computing underlying support environment has been constructed to complete the access, high-concurrency storage, and basic security management of time-series data across all scenarios of power production monitoring. This provides a stable, efficient, and secure foundation for subsequent governance and asset generation, specifically including:

[0042] 1. Data Access and High-Concurrency Writing: Adapts to multi-source production monitoring systems such as power SCADA systems, wind turbine monitoring systems, thermal power unit DCS systems, and new energy centralized monitoring platforms, enabling high-concurrency access of time-series data sampled at millisecond / second levels. It supports writing tens of millions of measurement point data points per second to a single cluster. Through WAL logs and multi-replica mechanisms, it ensures data persistence and anomaly recovery capabilities, ensuring zero data loss.

[0043] 2. Data storage and query optimization: Adopting a distributed data layout and hierarchical storage mechanism, it supports stable storage of petabyte-level massive data. For typical power scenarios such as latest value query, historical curve query, and cross-section query, the access path is optimized to ensure that the query response latency is ≤1 second and the system CPU utilization is ≤25%, providing efficient data read and write support for subsequent data governance.

[0044] 3. End-to-end basic security control: Data transmission is encrypted based on the SSL protocol. Through static storage encryption, key lifecycle management and fine-grained access control, end-to-end security protection is achieved for data transmission, storage and access. At the same time, an abnormal data identification module is built in to preliminarily identify and mark typical abnormal data in power scenarios such as out-of-bounds, mutations and missing data, providing a data foundation for subsequent AI governance.

[0045] The process follows a linear progression of data access, storage management, and security control, as follows: Data access and high-concurrency writing inputs consist of raw time-series data from multi-source production monitoring systems such as power SCADA, DCS, and new energy centralized control. Outputs are inbound time-series data that has undergone log persistence, multi-replica redundancy backup, and format verification. Data that has not been persisted cannot enter the storage management stage, preventing data loss and inconsistency. Data storage and query optimization inputs are inbound time-series data, and outputs a stable, accessible time-series dataset that has undergone distributed sharding storage, cold / hot tiering, and query path optimization, along with supporting node load balancing and data proximity storage scheduling strategies. This provides the underlying data layout basis for subsequent storage-computing collaborative scheduling. End-to-end basic security control inputs are stable, accessible time-series datasets, and outputs a secure, compliant time-series dataset that has undergone transmission encryption, storage encryption, fine-grained access control, and initial screening and marking of abnormal data, along with end-to-end security protection rules. This provides the core processing objects and underlying security constraints for the entire process, serving as a prerequisite for compliance in all subsequent data processing stages.

[0046] Step S1 provides underlying data and architecture support for the entire process. The time-series dataset consists of raw time-series power production monitoring data that has undergone format verification, initial anomaly screening, and persistent storage. The system's underlying architecture features a distributed data sharding storage layout and node scheduling capability adapted to an integrated storage-computing architecture, and a full-link basic security protection foundation that meets power security zoning requirements. Based on this, AI preprocessing tasks need to be precisely distributed to the corresponding storage nodes for local execution through the storage-computing collaborative scheduling module, based on the storage location of the distributed data shards. The core processing object is the initially screened time-series dataset. At the same time, the basic security protection capability provides strict underlying security constraints for the entire process of data processing, ensuring that all subsequent data governance and asset generation processes strictly adhere to the set access permissions, encryption rules, and security zoning control requirements.

[0047] S2. Based on the time-series dataset, call the preset lightweight digital twin of power equipment to generate a dynamic operating condition baseline, and combine the dual-track detection mechanism to perform anomaly detection and management on the time-series dataset, and output a standardized time-series dataset.

[0048] It should be further explained that by decentralizing AI preprocessing capabilities to the time-series database storage nodes, data can be processed locally without large-scale migration across nodes. This enables automated cleaning and standardization of raw time-series data, specifically including:

[0049] 1. Intelligent filling of missing values: For scenarios with missing data such as data breakpoints and null values, the algorithm adaptively selects algorithms such as forward filling, mean filling, linear filling, and adaptive interpolation filling in combination with power business rules to accurately fill missing values ​​and ensure data continuity.

[0050] 2. Intelligent Anomaly Detection and Management: Based on AI algorithms such as 3-Sigma, Z-Score, and Isolation Forest, combined with power equipment operation threshold rules, it accurately identifies initially marked abnormal data, transmission noise, and mis-collected data. According to business configuration, it automatically filters, marks, or corrects the data, reducing the interference of abnormal data on subsequent asset generation.

[0051] 3. Data format standardization and normalization: Automatically identify format differences of multi-source data, unify the specifications of timestamps, units, ranges, etc., and complete data scale normalization based on algorithms such as Min-Max and Z-Score. Output high-quality preprocessed datasets that conform to power industry standards. The entire processing is executed locally on the storage node, with an end-to-end processing latency of ≤1 second.

[0052] Among them, such as Figure 3 As shown, the system follows a progressive closed-loop logic of data continuity assurance, anomaly data governance, and standardization and normalization, specifically as follows: The input for intelligent missing value imputation is a time-series dataset that has undergone initial screening and labeling. The output is a continuous time-series dataset after missing value identification, legal missing value labeling, and accurate imputation, along with a full-link log of missing value processing. Data without continuous repair has breakpoints, which can lead to misjudgments in anomaly detection and prevent accurate anomaly governance. The input for intelligent anomaly detection and governance is a continuous time-series dataset. The output is a time-series dataset free from anomaly interference after accurate anomaly identification, classification, filtering / correction / labeling, along with a full-link log of anomaly governance. Furthermore, anomaly identification results can be used to optimize the missing value imputation strategy, such as performing secondary imputation for new missing values ​​generated after anomaly removal. The input for data format standardization and normalization is a time-series dataset free from anomaly interference. The output is a standardized preprocessed dataset conforming to the power industry standard format and eliminating dimensional differences. Simultaneously, format verification results can be used to optimize processing rules, such as optimizing the range adaptation rules for anomaly detection caused by inconsistent units.

[0053] Step S2 provides a standardized, high-quality data foundation for the entire process. The standardized time-series dataset is high-quality data that has undergone missing value imputation, outlier management, and standardization normalization, and strictly follows the unified data format specifications adapted to the power industry standards. The output synchronously includes the full-link log of the data governance process and anomaly handling records. The anomaly handling records serve as the core basis for quality rating and can be used to achieve reverse optimization of the process. In addition, subsequent feature engineering processing must be carried out based on this standardized time-series dataset to effectively avoid the risk that the original data cannot achieve effective feature extraction and quantitative comparison due to differences in units, missing anomalies, etc.

[0054] In this optional embodiment, based on the time-series dataset, a preset lightweight digital twin of power equipment is invoked to generate a dynamic operating condition baseline. An anomaly detection and mitigation mechanism is then used to perform anomaly detection on the time-series dataset, outputting a standardized time-series dataset including:

[0055] S21. Based on the sharding information of the time series dataset, the preset lightweight digital twin of the power equipment is called locally on the storage node of the time series database, and the power equipment information and operating characteristics are combined to match the current operating conditions of the power equipment in real time and generate a dynamic operating condition baseline.

[0056] S22. Based on the dynamic operating condition baseline and the preset power equipment business rules, distinguish between legal missing data and abnormal missing data in the time series dataset, and adaptively select a differentiated filling strategy according to the current operating condition type to generate a continuous time series dataset.

[0057] S23. Based on the continuous time series dataset and the dynamic operating condition baseline, construct a dual-track detection mechanism based on baseline rule verification and anomaly detection. Use the dual-track detection mechanism to perform anomaly detection and governance, and output a time series dataset without anomaly interference.

[0058] In this optional embodiment, a dual-track detection mechanism based on baseline rule verification and anomaly detection is constructed according to the continuous time-series dataset and the dynamic operating condition baseline. This dual-track detection mechanism is used to perform anomaly detection and mitigation, outputting a time-series dataset free of anomaly interference, including:

[0059] S231. Based on the initial screening labeling results and fragmentation information in the continuous time series dataset, perform initial anomaly classification and detection on the time series data, and output the dataset to be detected.

[0060] S232. Based on the dynamic operating condition baseline and combined with the preset power business rule base, perform rule-level anomaly verification on the dataset to be tested, and generate the full dataset based on the verification results. Rule-level anomaly verification includes boundary out-of-bounds anomaly verification, mutation anomaly verification, and logic anomaly verification.

[0061] S233. A cascaded detection mode combining statistical algorithms and machine learning algorithms is adopted. On the local storage node, a secondary anomaly classification and detection is performed on the full dataset, and an anomaly data classification set is output.

[0062] S234. Combining the abnormal data classification set with the preset power business rule base, adaptively match the corresponding governance strategy;

[0063] S235. On the local storage node, perform abnormal data governance on the abnormal data classification set according to the matching governance strategy, and output a time series dataset without abnormal interference.

[0064] It should be further noted that the intelligent outlier detection and governance of this invention, based on a dual-track detection mechanism of baseline rule verification and outlier detection, is deeply embedded in the storage nodes of the in-memory computing time-series library for local execution, without requiring cross-node data migration throughout the process. The specific implementation steps are as follows:

[0065] 1. Data Input and Pre-processing Verification: On the local storage node, the continuous time-series dataset after missing value filling is received. The abnormal data initial screening and marking results output from step S1, the device information, measurement point attributes, business scenarios corresponding to the current data shard, and the current device real-time operating condition twin baseline output by the adaptive pre-processing decision engine are obtained synchronously as the core input of this step.

[0066] 2. Initial Classification and Priority Ranking of Outliers: Based on the initial screening and labeling results and the business attributes of the measurement points, the data is initially classified and prioritized for detection. Core measurement points, such as turbine speed, generator power, and wind turbine pitch angle, which directly affect equipment safety and production control, are set to the highest priority and undergo full dual-track detection. Auxiliary measurement points are set to secondary priority and undergo a detection mode that prioritizes rule verification and supplements it with AI algorithm sampling. Legitimate outlier data that has been marked as being under maintenance or shutdown conditions is directly marked and isolated, and is not included in subsequent detection stages to avoid invalid calculations.

[0067] 3. Twin Baseline Rule Verification (First Track Detection): Based on the twin baseline of the current equipment's real-time operating conditions, combined with the power business rule library, rule-level anomaly verification is performed. The specific logic is as follows: Boundary Exceedance Anomaly Verification: Compare the current data with the twin baseline's rated upper and lower limits, equipment nameplate thresholds, and the safe operating range stipulated in power industry regulations to identify out-of-boundary data exceeding reasonable ranges; Sudden Change Anomaly Verification: Based on the twin baseline's normal fluctuation rate of operating conditions, calculate the slope of change of adjacent data points to identify sudden change data exceeding the normal fluctuation rate, such as a sudden change anomaly where the turbine vibration value instantly jumps above the baseline threshold during normal unit operation; Logical Anomaly Verification: Based on power business coupling rules, verify the logical consistency of associated measurement points, such as logical contradictions where the generator's active power is 0 but the stator current displays the rated value, identifying logical anomalies caused by mis-collection or transmission errors; After the rule verification is completed, a rule anomaly marker dataset is output and simultaneously enters the AI ​​algorithm for secondary detection.

[0068] 4. AI Anomaly Detection (Second-Track Detection): For the full dataset that has undergone rule validation, a cascaded detection mode combining statistical and machine learning algorithms is used. AI anomaly detection is executed locally on the storage node. The specific algorithm application logic is as follows: First-level statistical algorithm for rapid screening: Using 3-Sigma and Z-Score statistical algorithms, rapid anomaly screening is performed on time-series data to identify outliers that significantly deviate from the normal distribution and output first-level anomaly labeling results. This step has low computational load and can achieve rapid processing of high-concurrency data; Second-level machine learning algorithm for precise identification: For suspected anomaly data marked by the first-level screening and the highest priority core measurement point data, anomaly identification is performed using... The system employs Isolation Forest and KNN algorithms for precise secondary detection. Isolation Forest identifies local anomalies in high-dimensional data, adapting to the identification of weak anomaly signals in the early stages of equipment failure. KNN identifies sequential anomalies in time-series data, adapting to the identification of gradual anomalies where equipment operating trends deviate from normal conditions. The system also integrates the detection results of statistical and machine learning algorithms, combining them with anomaly labeling results from rule validation to output the final anomaly data classification results. Anomalies are categorized into four levels: fatal anomalies (affecting safe equipment operation), severe anomalies (affecting data validity), general anomalies (minor noise), and minor anomalies (negligible fluctuations).

[0069] 5. Adaptive Matching of Business Rule-Driven Anomaly Management Strategies: Based on the classification of anomaly data levels, measurement point business attributes, and current equipment operating conditions, combined with the power business rule base, adaptive matching of corresponding management strategies is performed, as follows: For fatal anomalies, such as equipment exceeding limits or fault-related fatal anomalies, a marking and retention strategy with alarm triggering is implemented. No filtering or correction is performed, and the original anomaly data is fully preserved. Equipment alarms are triggered synchronously, providing core feature data for subsequent fault diagnosis and equipment health management. For severe anomalies, such as those caused by mis-collection, transmission errors, or logical contradictions, a filtering and removal strategy with missing data marking is implemented. Invalid data is removed, and the data is simultaneously marked as anomaly missing, feeding back to the missing value filling module for secondary precise filling. For general anomalies, such as slight over-range or small-amplitude changes, a smoothing correction strategy is implemented. Based on the twin baseline of the current operating conditions, a sliding window smoothing algorithm is used for correction, preserving the temporal continuity of the data. For minor anomalies, such as those caused by random noise, a marking and retention strategy is implemented without processing, avoiding the loss of effective features due to over-management.

[0070] 6. Governance Execution and Result Output: Locally on the storage node, abnormal data governance is performed according to the matching governance strategy, and the time-series dataset without abnormal interference after governance is output. Simultaneously, a full-link log of anomaly detection and governance is generated, which includes information such as anomaly location, anomaly type, level, governance strategy, and execution result. This log will serve as the core basis for data quality rating.

[0071] 7. Self-optimizing closed loop strategy: Based on the output data quality rating results and business adaptation verification feedback, for scenarios of anomaly misjudgment, missed judgment, and poor governance effect, the system automatically optimizes the twin baseline, rule verification threshold, AI algorithm model parameters and governance strategy matching rules to form a self-optimizing closed loop of anomaly detection and governance.

[0072] S24. Perform data format standardization and normalization processing on the time series dataset without abnormal interference to generate a standardized time series dataset.

[0073] It should be further explained that, in the end-to-end AI intelligent preprocessing of power time-series data based on in-memory computing collaboration, this invention introduces an adaptive preprocessing decision engine driven by digital twins of power equipment operating conditions. This engine is deeply integrated into the storage nodes of the in-memory computing integrated time-series database and works in conjunction with the in-memory computing collaborative scheduling module to achieve intelligent processing throughout the entire process, including dynamic perception of operating conditions, twin baseline matching, adaptive generation of preprocessing strategies, and iterative optimization of governance effects. This technical approach overcomes the core bottlenecks of existing technologies, such as the use of fixed algorithm models and static rules, inability to adapt to the complex dynamic operating conditions of power equipment, high susceptibility to misjudgments and omissions, and the separation of preprocessing and storage layers. The specific implementation method is as follows:

[0074] 1. Engine core architecture and storage-compute collaboration mechanism: The engine is built into each storage node of the time series library and is deeply integrated with the distributed storage management module and the storage-compute collaboration scheduling module. Each storage node has a lightweight digital twin of the device to which the corresponding shard data belongs. It can realize local working condition perception and preprocessing decision without migrating data across nodes. It is fully adapted to the storage-compute integrated architecture where the data does not move and the computation moves, eliminating the performance loss of cross-node data transmission from the root.

[0075] 2. Core implementation process, deeply integrated into all stages of preprocessing:

[0076] (1) Pre-processing: Construction and real-time perception of dynamic twin baselines of equipment operating conditions. In the data access stage of step S1, lightweight digital twins are constructed for various power equipment such as thermal power, wind power, and new energy. Based on equipment nameplate parameters, historical operating data, maintenance records, and business rules, a full-condition operating baseline library of equipment is generated, covering the normal operating parameter range, fluctuation pattern, and time series characteristics of equipment in all scenarios such as no-load, start-up, rated load, peak shaving, maintenance, and fault transients. This provides a dynamic benchmark for preprocessing, rather than a traditional fixed threshold. Before the preprocessing is executed in step S2, the engine calls the twin of the corresponding equipment locally on the storage node based on the sharding information of the current time series data, matches the current operating conditions of the equipment in real time, and generates a dynamic operating condition baseline as the core decision basis for preprocessing.

[0077] (2) Integration of intelligent missing value filling: The twin-driven adaptive missing value filling strategy engine distinguishes between legal and abnormal missing values ​​based on real-time perceived equipment operating conditions. For legal missing values ​​under operating conditions such as equipment maintenance, shutdown, and planned shutdown, the engine marks and isolates them based on the twin baseline and does not perform filling. For abnormal missing values ​​caused by data transmission interruption or acquisition failure, the engine adaptively selects the optimal filling algorithm based on the twin operating baseline of the current operating condition. Under steady-state operating conditions (stable operation under rated load), forward filling / mean filling is used to ensure processing efficiency. Under dynamic operating conditions (start-stop, peak shaving, load fluctuation), adaptive interpolation filling / predictive filling based on the twin model is used to accurately fit the changing trend of equipment parameters and avoid filling errors caused by traditional fixed algorithms. Under fault transient operating conditions, the original missing value mark is retained and filling is not performed to avoid destroying fault characteristics and to retain complete data for subsequent fault diagnosis.

[0078] (3) Integration of intelligent outlier detection and governance: The outlier detection and governance engine, which integrates twin baselines and AI algorithms, is based on real-time working conditions and constructs a dual-track detection mechanism that includes twin baseline rule verification and AI algorithm outlier detection. This solves the problems of high false positive rate and missed outlier detection in existing technologies.

[0079] (4) Integrating data format standardization and normalization: The working condition adaptive normalization strategy engine is based on the real-time perceived equipment working condition and adaptively adjusts the normalization algorithm and parameter range: For steady-state working conditions, fixed-range Min-Max normalization is adopted; for dynamic peak shaving and start-up / shutdown working conditions, dynamic range normalization is performed based on the twin baseline parameter range of the current working condition to eliminate the dimensional influence caused by working condition fluctuations and ensure that the normalized data can accurately reflect the real operating status of the equipment, rather than the feature distortion caused by traditional fixed-range normalization.

[0080] 3. Self-learning and self-iterative closed-loop mechanism: Based on the quality rating results of step S4 and the business adaptation verification feedback of step S5, the engine automatically optimizes the device twin baseline and preprocessing strategy. For working scenarios with low filling accuracy and high anomaly misjudgment rate, the twin baseline library is automatically updated, and the algorithm selection logic and parameter configuration are adjusted to achieve self-learning and self-iteration of the preprocessing strategy without manual intervention, which greatly improves the adaptability and accuracy of preprocessing.

[0081] S3. Match the artificial intelligence algorithm corresponding to the standardized time series dataset, perform feature mining based on the matching results, and use a two-dimensional screening mechanism to screen and optimize the feature mining results to generate a standardized high-value feature set.

[0082] It should be further explained that, for standardized time-series datasets, high-value time-series features are automatically extracted using AI algorithms to construct a power business-specific feature library, thereby increasing the business value density of the data. Specifically, this includes:

[0083] 1. Automatic extraction of basic time series features: Through sliding window calculation, AI automatically extracts basic time series features such as maximum and minimum values, mean, variance, slope, and periodicity of data to capture the fluctuation patterns and trend changes of equipment operation data.

[0084] 2. Power Business-Specific Feature Mining: Using algorithms such as KMeans clustering and PCA dimensionality reduction, combined with business rules for scenarios such as thermal power, wind power, and new energy centralized control, we can mine specific business features such as equipment health, load fluctuation, fault warning, and energy efficiency.

[0085] 3. Feature selection and optimization: Based on Pearson correlation coefficient and Spearman correlation coefficient, the extracted features are weighted and redundant features are removed. High-value features that are highly relevant to business objectives are retained, and a standardized high-value feature set is generated, laying the foundation for subsequent intelligent labeling and asset generation.

[0086] The process follows a hierarchical progression: basic feature extraction, business feature mining, and feature selection and optimization. Specifically: The input for automatic basic time-series feature extraction is a standardized preprocessed dataset, and the output is a basic feature set containing fundamental time-series features such as maximum / minimum values, mean, variance, slope, and periodicity. Basic time-series features form the foundation for business-specific feature mining; without basic feature extraction, higher-order business feature correlation mining cannot be performed. The input for power business-specific feature mining is the basic feature set, and the output is a set of specific business features that integrate power business rules and are strongly correlated with business objectives, such as equipment health, load fluctuations, fault warnings, and energy efficiency. Simultaneously, the business feature mining results can be used to optimize the basic feature extraction strategy, such as adjusting the sliding window size and feature calculation cycle for specific business scenarios. The input for feature selection and optimization is the full business feature set, and the output is a standardized high-value feature set after weight evaluation, redundancy removal, and dimensionality reduction optimization. This output is directly used as input for intelligent labeling and quality rating. Furthermore, the feature relevance evaluation results can be used to optimize the business feature mining strategy and the basic feature extraction strategy, eliminating feature extraction logic irrelevant to business objectives and improving feature extraction efficiency.

[0087] Step S3 is the core step in achieving a significant increase in data value density throughout the entire process. The standardized high-value feature set is a set of standardized high-value features extracted from basic time-series features, mined from business-specific features, and optimized through screening. It is accompanied by a report on the correlation weight assessment between the features and the goals of the power business, as well as an iterative update record of the power industry-specific feature library. This standardized high-value feature set serves as the core input for intelligent annotation and quality rating. It is not only aimed at the original data itself, but more importantly, at its business-specific annotation and value assessment, because the business relevance of the features directly determines the classification of data value levels. At the same time, the subsequent annotation results and business adaptation verification feedback can drive the continuous optimization of the feature library and the precise adjustment of the feature screening strategy.

[0088] In this optional embodiment, an artificial intelligence algorithm is matched to the standardized time-series dataset, feature mining is performed based on the matching results, and a two-dimensional screening mechanism is used to filter and optimize the feature mining results to generate a standardized high-value feature set, including:

[0089] S31. Use the sliding window technique to extract time-series features from the standardized time-series dataset, generate a time-series feature set, and match the corresponding power feature mining template according to the power equipment type and business target scenario.

[0090] S32. Based on the preset power business rules, perform operating condition identification and data segmentation on the standardized time series dataset to generate different operating condition segments, and construct the target feature dimension by combining the power feature mining template.

[0091] S33. Based on the target feature dimension, match the corresponding artificial intelligence algorithm, perform feature mining on the time series feature set based on the matching result, and generate full business features;

[0092] S34. Combining the preset power business rule base, historical fault case base and power industry standards and specifications, perform business validity verification on all business features, and merge the verified all business features with the time series feature set to generate a full feature set.

[0093] S35. A dual-dimensional screening mechanism combining mathematical correlation screening and power business weight assignment is used to screen and optimize the full feature set and generate a standardized high-value feature set.

[0094] In this optional embodiment, a two-dimensional screening mechanism combining mathematical relevance screening and power business weight assignment is used to screen and optimize the entire feature set, generating a standardized high-value feature set including:

[0095] S351. Based on the full feature set, calculate the Pearson correlation coefficient between the full features and the preset business target labels, and perform linear correlation screening in combination with the preset threshold.

[0096] S352. Based on the linear correlation screening results, calculate the Spearman correlation coefficient and maximum information coefficient between the full set of features and the preset business target labels, and perform nonlinear correlation screening based on the calculation results to obtain the mathematical correlation screening results.

[0097] S353. Based on the business objective scenario, construct a power business weight evaluation system, assign weights to all features from preset dimensions, prioritize all features based on the weight assignment results, and output a priority list of all features.

[0098] S354. Combining the mathematical correlation screening results with the full feature priority list, set a business weight threshold, and use the weight threshold to perform feature screening on the full feature set, outputting a standardized high-value feature set.

[0099] It should be further explained that the power business-specific feature mining of this invention deeply integrates power business rules and AI algorithms for all power production scenarios, including centralized monitoring of thermal power, wind power, and new energy, and is executed locally and distributed on the storage nodes of the in-store computing time-series library. The specific implementation steps are as follows:

[0100] 1. Data Input and Scenario Matching: On the local storage node, the basic time-series feature set is received, and the corresponding standardized time-series dataset, equipment type, business scenario, measurement point list, as well as the power business rule library, historical fault case library, and typical operating condition library are obtained synchronously as the core input for this step. Based on the equipment type and business scenario, the corresponding power-specific feature mining template is matched, such as the thermal power turbine vibration feature template, the wind power turbine health feature template, and the new energy load prediction feature template.

[0101] 2. Business Scenario Operating Condition Classification and Data Segmentation: Based on power business rules, operating condition identification and data segmentation are performed on the time-series dataset, dividing the data into different operating condition segments, such as start-up and shutdown, no-load, rated load, peak shaving, and maintenance operating conditions for thermal power; and grid connection, shutdown, rated wind speed, low wind speed, and power limitation operating conditions for wind power. For different operating condition segments, business feature mining is carried out separately to avoid feature distortion caused by cross-operating condition data mixing.

[0102] 3. Targeted Construction of Feature Dimensions Based on Business Rules: For each segmented operating condition, core feature dimensions are constructed based on power business rules, combined with matching feature mining templates, to clarify the core mining objectives for each business scenario: For equipment health management and fault diagnosis scenarios, targeted construction is carried out for equipment degradation trend features, fault precursor features, key component vibration features, temperature field distribution features, and oil quality features; For load forecasting scenarios, targeted construction is carried out for load fluctuation features, periodic characteristics, meteorological correlation features, grid dispatch command correlation features, and new energy output characteristics; For energy efficiency analysis and optimization scenarios, targeted construction is carried out for equipment energy consumption features, plant power consumption rate features, unit efficiency features, coal consumption / wind farm curtailment rate features, and economic operating range features.

[0103] 4. In-depth mining of business characteristics driven by AI algorithms:

[0104] For the targeted feature dimensions, corresponding AI algorithms are used in conjunction with the basic time-series feature set to conduct deep feature mining. The algorithm application logic and business rule integration mechanism are as follows:

[0105] (1) Mining of operating condition feature intervals based on KMeans clustering algorithm: For the basic feature set of each operating condition segment, the KMeans clustering algorithm is used to automatically cluster and generate the feature intervals of normal operation of equipment, abnormal warning feature intervals, and fault feature intervals under the operating condition. The clustering results are verified and corrected in combination with power business rules, and clusters that do not conform to business logic are eliminated. The standard operating feature intervals of each operating condition of the equipment are output, providing quantitative feature basis for equipment health status assessment. For example, for the peak shaving operating condition of thermal power units, the normal operating ranges of boiler main steam temperature and pressure under different peak shaving depths are generated by KMeans clustering, and the deterioration features that deviate from the normal range are identified.

[0106] (2) Equipment coupling feature mining based on association rule mining algorithm: Apriori and FP-Growth association rule mining algorithms are used to mine coupling association rules between measurement point parameters for multi-measurement point association data of the same equipment and the same system. Combined with the physical mechanism and business rules of power system equipment operation, strong association rules that conform to the equipment operation logic are selected to generate equipment multi-parameter coupling association features. For example, for the boiler system of thermal power unit, strong association rules between coal feed, air supply, furnace temperature and main steam pressure are mined to generate boiler combustion efficiency coupling features; for wind turbine, strong association rules between wind speed, blade pitch angle, generator speed and active power are mined to generate wind turbine wind energy utilization efficiency coupling features.

[0107] (3) High-dimensional sensitive feature mining based on PCA dimensionality reduction algorithm: For the high-dimensional basic feature set of fault diagnosis and anomaly detection scenarios, the PCA principal component analysis algorithm is used to reduce the dimensionality of the high-dimensional features, extract the core principal components that can reflect the changes in the operating status of the equipment, and combine the historical fault case library to screen the sensitive principal components that are strongly correlated with equipment faults and anomalies, generate fault precursor sensitive features, and capture the weak change signals in the early stage of equipment faults; for example, for the high-dimensional time series features of turbine bearing vibration, the PCA dimensionality reduction is used to extract the sensitive features that can reflect the early wear of the bearing, so as to realize the early warning of faults.

[0108] (4) Mining of periodic and trend features based on time series decomposition algorithm: The STL time series decomposition algorithm is used to decompose the time series data into trend items, periodic items and residual items. Combined with power business rules, the corresponding features are mined respectively: trend items, mining the deterioration trend features and load change trend features of long-term equipment operation, which are suitable for equipment health management and medium and long-term load forecasting scenarios; periodic items, mining the daily, weekly, monthly and annual cycle features of equipment operation, which are suitable for short-term load forecasting and economic operation scheduling scenarios; residual items, mining the abnormal fluctuation features in the data, which are suitable for anomaly detection and fault diagnosis scenarios.

[0109] (5) Business rule verification and feature validity verification: For the full range of business features mined by AI algorithm, combined with the power business rule library, historical fault case library and industry standards and specifications, business validity verification is carried out: invalid features that do not conform to the physical mechanism of equipment operation and business logic are removed; the relevance of features to business objectives is verified, for example, in the fault diagnosis scenario, whether the features can effectively distinguish between normal operating conditions and fault operating conditions; for features that pass the verification, business semantic tags are added to clarify the business meaning, applicable scenarios and calculation logic of the features, and form a power industry-specific business feature library.

[0110] (6) Feature set output and iterative optimization: Output the verified power business-specific feature set, merge it with the basic time series feature set to form a full feature set, and input it into the feature screening and optimization stage; at the same time, based on the quality rating results and business adaptation verification feedback, automatically optimize the feature mining algorithm parameters, feature templates and business rules, and continuously iterate and update the power-specific feature library.

[0111] S4. Based on the preset power industry business specifications, construct an intelligent annotation and multi-dimensional quality rating system, and use the intelligent annotation and multi-dimensional quality rating system to annotate and filter the standardized high-value feature set to obtain a high-value dataset.

[0112] It should be further explained that, based on the business standards of the power industry, an intelligent labeling and multi-dimensional quality rating system has been constructed to complete the business-oriented labeling and value stratification of data, specifically including:

[0113] 1. Business-scenario-based intelligent labeling: For fault data, operating condition data, alarm data, etc., AI combines historical fault cases and power business rules to achieve automatic classification and labeling, supports manual review and correction, and core fault samples achieve 100% expert review to ensure labeling accuracy ≥97%.

[0114] 2. Multi-dimensional data quality rating: Data quality is automatically scored and rated from five dimensions: data integrity, accuracy, timeliness, relevance, and security. Data is divided into high, medium, and low value levels, and high-value datasets are selected. The core acceptance indicators are data integrity ≥98% and data content accuracy ≥99%.

[0115] 3. Quality closed-loop optimization: For low- to medium-value data, feedback is sent to the preprocessing stage for secondary optimization, forming a closed-loop mechanism for data quality governance.

[0116] The closed-loop logic follows business-oriented annotation, quality rating, and closed-loop optimization, as follows: Business-scenario-based intelligent annotation takes a standardized high-value feature set and its corresponding time-series dataset as input, and outputs an automatically classified and annotated dataset that has been manually reviewed and corrected, along with an annotation accuracy assessment report. Data without business-oriented annotation cannot be quantitatively evaluated for business value, and targeted quality rating cannot be conducted. Multi-dimensional data quality rating takes an annotated dataset as input, and outputs quality scores for five dimensions: completeness, accuracy, timeliness, relevance, and security, as well as high, medium, and low-value data tier classification results and a high-value dataset. Quality deduction results directly serve as the decision-making basis for closed-loop optimization. Quality closed-loop optimization takes medium- and low-value data and its quality problem analysis report as input, and outputs secondary optimization processing instructions and optimization strategy suggestions. This output is directly fed back to the preprocessing step S2, driving the preprocessing stage to perform secondary governance of quality problems. Simultaneously, it can reverse-optimize annotation rules and quality rating models, forming a closed loop for data quality governance.

[0117] The feature selection and optimization method of this invention adopts a two-dimensional selection mechanism of mathematical statistical correlation selection and power business weight assignment, which breaks through the bottleneck of the existing technology of pure mathematical selection and ignoring business value. The specific implementation steps are as follows:

[0118] (1) Feature set input and preprocessing verification: Receive the full feature set, which includes basic time series features and power business-specific features. Simultaneously obtain the corresponding business target scenarios (such as load forecasting, fault diagnosis, equipment health management, etc.), power business rule base, and feature business semantic tags as input for this step. Perform preprocessing verification on the feature set, and remove invalid features with missing values ​​or variance of 0, that is, features whose feature values ​​have not changed in the full data and cannot provide effective information, to complete the initial dimensionality reduction.

[0119] (2) Linear correlation screening: Based on the Pearson correlation coefficient, linear correlation screening is carried out. The specific logic is as follows: calculate the Pearson correlation coefficient between each feature and the business target label, evaluate the linear correlation between the feature and the business target, and remove weakly linearly correlated features whose absolute value of the correlation coefficient is lower than the preset threshold (default threshold 0.1); calculate the Pearson correlation coefficient between features, and for highly redundant feature pairs whose absolute value of the correlation coefficient is higher than the preset threshold (default threshold 0.9), filter them in combination with business rules: retain features that are more relevant to the business target and have clearer business semantics, remove redundant features, and avoid feature collinearity problems.

[0120] (3) Nonlinear correlation screening: Based on Spearman correlation coefficient and maximum information coefficient (MIC), nonlinear correlation screening is carried out to make up for the deficiency that Pearson correlation coefficient can only identify linear correlation. The specific logic is as follows: calculate the Spearman correlation coefficient and maximum information coefficient between each feature and the business target label, evaluate the nonlinear correlation between the feature and the business target, and remove features that have no significant nonlinear correlation with the business target; for the feature set after linear screening, further identify nonlinear redundant features, combine with business rules, retain features that are more interpretable to the business target, and remove redundant features.

[0121] (4) Weighting and Priority Ranking of Power Business: Breaking through the limitations of pure mathematical screening in existing technologies, power business rules are introduced to assign weights to features. The specific logic is as follows: Based on the business target scenario, a power business weight evaluation system is constructed, and features are weighted from the following dimensions: Business coreness, whether the feature corresponds to the core operating parameters of power equipment and whether it directly affects the achievement of business objectives. Features corresponding to core measurement points have the highest weights; Scenario adaptability, the degree of adaptability of the feature to the current business scenario. For example, in the fault diagnosis scenario, the feature of fault precursors has the highest weight; Industry compliance, whether the feature conforms to the power industry standards and specifications and whether it is a core evaluation indicator recognized by the industry; Interpretability, whether the business semantics of the feature are clear and whether they conform to the physical mechanism of equipment operation. The stronger the interpretability, the higher the weight; Based on the above dimensions, a comprehensive weight is assigned to each feature. The weight range is 0-1. The higher the weight, the higher the business value of the feature; Based on the comprehensive weight, the features are prioritized to provide a business-level decision-making basis for subsequent feature screening.

[0122] (5) Redundant feature removal and optimal feature subset selection: Combining the mathematical correlation selection results and business weight assignment results, the final feature selection is carried out. The specific logic is as follows: Set a business weight threshold, with a default threshold of 0.3. Remove low business value features with weights lower than the threshold. Even if the mathematical correlation of the feature meets the standard, if the business value is insufficient, it will still be removed to ensure that the selected features have clear business significance. For the remaining features, sort them from high to low according to business weight. Combined with the feature dimension requirements of the business scenario, use the forward selection algorithm to gradually add features and verify the effect of the feature set on the business model until the model effect no longer significantly improves. Stop adding features and generate a preliminary optimal feature subset. For the preliminary optimal feature subset, use cross-validation to verify the stability and generalization ability of the feature set, remove unstable features, and finally generate a standardized high-value feature set.

[0123] (6) Feature set output and iterative optimization: Output the final standardized high-value feature set and input it into the intelligent annotation and quality rating stage; at the same time, based on the business scenario adaptation verification results, for the problem that the feature set does not perform well in the business model, reverse optimize the relevance screening threshold, business weight evaluation system and feature screening strategy to continuously improve the business adaptability of the feature set.

[0124] Step S4 involves the graded screening and business-oriented anchoring of data value throughout the entire process. The high-value dataset is a dataset that has undergone intelligent annotation based on business scenarios. It includes the results of high, medium, and low-value data hierarchy classification after multi-dimensional quality rating, a subset of high-value data that meets the core acceptance threshold, and suggestions for secondary optimization of medium and low-value data. The annotation results, quality rating report, and business scenario classification of the high-value dataset directly determine the classification, encapsulation content, and recommended application scenarios of the subsequent datasets. At the same time, the output medium and low-value data will be directly fed back to step S2 for secondary optimization, thus forming an internal closed loop in the data governance process.

[0125] S5. Classify and structure high-value datasets, and store the encapsulation results in a pre-defined data asset library. Combine the pre-deployed power industrial control alliance chain and power business knowledge graph to optimize the data asset library, so as to realize the governance of power production monitoring data and the generation of high-value data assets.

[0126] It should be further explained that, based on high-value datasets, the standardized packaging, archiving, and full lifecycle management of data assets are completed, ultimately transforming raw data into reusable data assets. This specifically includes:

[0127] 1. Power Scenarios Dedicated Dataset Classification: A high-value dataset classification system is constructed according to four categories: centralized monitoring of new energy, operation and management of thermal power, health of wind power equipment, and general power digitalization, to meet the reuse needs of different business scenarios.

[0128] 2. Standardized Asset Encapsulation of Datasets: In accordance with the power industry's data set construction standards, high-value data sets are structurally encapsulated, and dataset description documents are generated simultaneously, including information such as data source, feature dimension list, annotation rules, quality rating, and recommended application scenarios. Each dataset is assigned a unique version number and global identifier, and iteration update logs are recorded to ensure that the assets are traceable and rollbackable.

[0129] 3. Asset archiving, storage, and management: Store the packaged high-value datasets in the data asset repository, establish an asset catalog and tag system to support rapid retrieval and business calls; at the same time, establish version management, access control, and traceability query mechanisms to achieve closed-loop management of the entire lifecycle of data assets.

[0130] 4. Scenario-based adaptation verification: For core business scenarios such as load forecasting, anomaly detection, fault diagnosis, equipment health management, and energy efficiency analysis and optimization, conduct dataset call tests to verify the business adaptability of the datasets. Based on business feedback, continuously optimize the feature system and annotation rules to form a closed loop of asset iterative optimization.

[0131] The process follows a closed-loop logic encompassing classification system construction, standardized encapsulation, full lifecycle management, and scenario-based verification and iteration. Specifically: The input for power scenario-specific dataset classification is high-value datasets and their annotations, along with business scenario classification results. The output is dataset category division results conforming to four major classification systems. Unclassified datasets cannot achieve structured encapsulation and targeted business adaptation, hindering efficient retrieval and reuse. The input for standardized asset encapsulation is the classified high-value dataset. The output is standardized data assets that have undergone structured encapsulation, include standardized metadata documentation, and are assigned unique version numbers and global identifiers. Unstandardized datasets lack interpretability and traceability, preventing the formation of reusable standardized assets. The input for asset archiving and management is standardized data assets, with the output being archived data. The power data asset repository, along with its supporting asset catalog, tagging system, and full lifecycle management rules, is the core output of this invention. It directly provides data asset access services for power business scenarios, while the management rules directly constrain asset access and version iteration behavior during the scenario-based adaptation verification process. The input for scenario-based adaptation verification is the standardized data assets archived and stored, and the output is a dataset business adaptability verification report and optimization suggestions. This output drives optimization throughout the entire process: for insufficient adaptability, feedback is provided to feature engineering in step S3 and the annotation and rating optimization strategy in step S4; for unreasonable classification, the classification system is optimized; for incomplete encapsulated information, the encapsulation specifications are optimized; and for imperfect management rules, the full lifecycle management mechanism is optimized, forming a complete closed loop for asset iteration optimization.

[0132] Step S5 is the final transformation of raw data into reusable data assets, outputting a standardized, packaged, and archived high-value power data asset library, a supporting data asset lifecycle management system and control rules, and a dataset business scenario adaptation verification report and optimization suggestions. This output is the final core achievement of this invention, and its adaptation verification feedback will drive the optimization of the entire process: for datasets with insufficient business scenario adaptability, optimization suggestions will be fed back to the feature engineering, labeling, and rating processing stages to adjust feature extraction strategies and labeling rules; for datasets with substandard data quality, feedback will be fed back to the preprocessing stage for secondary governance; and for problems with insufficient storage and scheduling efficiency, feedback will be fed back to the time series library module to optimize storage and query strategies, thereby forming a reverse iterative closed loop for the entire process.

[0133] In this optional embodiment, high-value datasets are classified and structured, and the encapsulation results are stored in a pre-defined data asset repository. Combined with a pre-deployed power industrial control alliance chain and power business knowledge graph, the data asset repository is optimized to achieve power production monitoring data governance and high-value data asset generation, including:

[0134] S51. Construct a high-value dataset classification system based on centralized monitoring of new energy, operation and management of thermal power, health of wind power equipment, and digitalization of general power, and use the high-value dataset classification system to classify high-value datasets.

[0135] It should be further noted that the high-value dataset classification system of this invention strictly follows the standards and specifications of the power industry, conforms to the business needs of the entire power production monitoring scenario, and adopts a construction method that combines top-down and bottom-up approaches. The specific implementation steps are as follows:

[0136] 1. Conduct Demand Survey and Standard Benchmarking: Conduct a full-scenario business demand survey, targeting all scenarios of power production monitoring, including thermal power, hydropower, wind power, photovoltaic, new energy centralized control, and grid dispatch. Survey the dataset requirements for digital and intelligent businesses such as load forecasting, anomaly detection, fault diagnosis, equipment health management, and energy efficiency analysis and optimization, and clarify the classification dimensions and retrieval requirements of datasets for different business scenarios; Benchmark against industry standards, benchmarking against national and industry standards and specifications such as the "Guidelines for Classification and Grading of Power Data," "Guidelines for Classification and Grading of Industrial Data," and "White Paper on Data Security Standardization in the Power Industry," to ensure that the classification system meets industry compliance requirements; Review the benchmarking results, review the existing power industry dataset construction results and public dataset classification methods, and combine the in-store computing architecture and AI governance capabilities of this invention to clarify the core design principles of the classification system: business-oriented, clear hierarchy, scalability, searchability, and compliance.

[0137] 2. Classification Dimensions and Hierarchical Architecture Design: Based on the results of requirements survey and standard benchmarking, a four-level hierarchical classification architecture is designed, consisting of first-level categories, second-level categories, third-level categories, and fourth-level tags. Among them: the first-level categories are based on major business scenarios and are the highest-level classifications; the second-level categories are based on the professional fields of power production; the third-level categories are based on equipment type and system; and the fourth-level tags are set based on refined dimensions such as business purpose, data characteristics, and operating condition type to achieve multi-dimensional retrieval.

[0138] 3. Category System Refinement and Content Filling: Based on the hierarchical architecture, the entire category system is refined and its content filled, forming the final four primary category classification systems, as follows:

[0139] (1) Category 1: Centralized monitoring of new energy; Category 2: Centralized monitoring of wind power, centralized monitoring of photovoltaic power, integrated monitoring of wind, solar and energy storage, and cluster control of new energy power plants; Category 3: Wind turbine equipment, photovoltaic inverters, energy storage systems, transformer substations, combiner boxes, AGC / AVC control systems, etc.; Core compatible scenarios: New energy power prediction, wind and solar curtailment analysis, health management of new energy power plant equipment, and coordinated control of wind, solar and energy storage, etc.

[0140] (2) Category 2, Thermal power operation and control; Category 2, Boiler system control, Steam turbine system control, Electrical system control, Thermal control system control, Environmental protection system control, Plant-level operation and control; Category 3, Boiler body, Steam turbine, Generator, Feedwater pump, Exhaust fan, Denitrification system, Desulfurization system, etc.; Core compatible scenarios, Unit load prediction, Combustion optimization, Equipment fault diagnosis, Energy efficiency analysis and optimization, Unit health management, etc.

[0141] (3) Category 3, Wind power equipment health category; Category 2, Wind turbine transmission chain health management, Wind turbine blade health management, Wind turbine control system management, Wind farm group equipment health management; Category 3, Gearbox, Generator, Main bearing, Blade, Pitch system, Yaw system, Converter, etc.; Core compatible scenarios, Wind turbine fault diagnosis, Remaining service life prediction, Equipment condition inspection, Operation and maintenance strategy optimization, etc.

[0142] (4) Category 4, General power digitalization; Category 2, Power load forecasting, power grid operation control, power industrial control security, power energy efficiency management, general equipment health management; Category 3, Transformers, circuit breakers, switch cabinets, motors, metering devices, etc.; Core adaptation scenarios, Regional power grid load forecasting, transmission and transformation equipment health management, power industrial control security testing, general equipment fault diagnosis, etc.

[0143] 4. Category Compliance and Reasonableness Verification: Compliance verification involves inviting power industry data management and security compliance experts to verify whether the classification system complies with industry standards such as the "Guideline for Classification and Grading of Power Data" and meets the compliance requirements for power security zoning and data classification protection; Business Reasonableness Verification involves inviting power production, equipment management, and digital application business experts to verify whether the classification system covers the entire power production monitoring scenario, whether it fits the actual business needs, and whether the category division is clear, without overlap, and without omissions; Optimization and Correction: Based on the expert verification opinions, the category system is optimized and corrected to ensure the compliance, reasonableness, and comprehensiveness of the classification system.

[0144] 5. Classification System Release and Dynamic Expansion Mechanism Establishment: Officially release the classification system and simultaneously formulate classification system management specifications, clarifying the processes and approval rules for adding, modifying, and abolishing categories; establish a dynamic expansion mechanism: for new scenarios, new equipment, and new businesses in new power systems (such as virtual power plants, new energy storage, hydrogen energy, etc.), the category system can be rapidly expanded according to standardized processes to ensure the forward-looking and scalable nature of the classification system; embed the classification system into the standardized packaging and asset archiving of datasets, requiring all high-value datasets to be categorized according to the classification system, laying the foundation for the subsequent construction of asset catalogs and tagging systems.

[0145] S52. The high-value dataset after classification is structured and encapsulated to generate hash values. The hash values ​​are then stored on the blockchain through a pre-deployed power industrial control alliance chain to achieve full-link trusted traceability and full lifecycle management of data assets.

[0146] In this optional embodiment, the high-value dataset after classification is structurally encapsulated to generate hash values, and the hash values ​​are stored on the blockchain through a pre-deployed power industrial control consortium blockchain to achieve end-to-end trusted traceability and full lifecycle management of data assets, including:

[0147] S521. Based on the preset power industry safety zoning principle, deploy production alliance chain nodes in the second production control area safety zone and management alliance chain nodes in the management information area to realize the deployment of the power industrial control alliance chain.

[0148] S522. According to the preset power industry dataset construction specifications, the high-value dataset after classification is structurally encapsulated, assigned a global identifier and version number, and a high-value data description document is generated simultaneously.

[0149] S523. Based on the global identifier, version number, and high-value data description document, generate a hash value and submit the hash value to the corresponding consortium chain node for on-chain storage through the power industrial control consortium chain.

[0150] S524. Based on the confirmation receipt of the power industrial control alliance chain, determine whether the hash value on-chain notarization is successful. If so, store the encapsulated high-value dataset into the preset data asset library and establish an asset catalog and tag system to achieve full-link trusted traceability and full life cycle management of data assets. If not, trigger the preset retry mechanism until the hash value on-chain notarization is successful.

[0151] It should be further explained that establishing an asset catalog and tagging system specifically includes:

[0152] 1. Asset Catalog Architecture Design: Based on the four-level classification system of the dataset, a four-level tree-shaped asset catalog architecture is designed, consisting of a master catalog, first-level category sub-catalogs, second-level category sub-catalogs, and third-level category asset details. The master catalog serves as the main entry point to the data asset repository, displaying statistical information, category distribution, and version information for all assets. The first-level category sub-catalogs correspond to four primary categories, displaying the total asset volume, second-level category distribution, and business scenario distribution for each category. The second-level category sub-catalogs correspond to the secondary categories, displaying asset details and equipment type distribution for each professional field. The third-level category asset details correspond to the tertiary categories, displaying a detailed list of dataset assets, including core metadata information such as dataset name, unique identifier, version number, quality rating, packaging time, and recommended application scenarios.

[0153] 2. Labeling System Dimension Design: Based on the needs of power business, a six-dimensional labeling system is designed to achieve multi-dimensional and refined labeling of dataset assets. The specific dimensions are as follows: Basic attribute labels, including basic information such as dataset name, unique global identifier, version number, data volume, number of measurement points, time span, and data sampling frequency; Classification and attribution labels, corresponding to the first, second, and third-level categories of the dataset classification system, clearly defining the dataset's business scenario, professional field, and equipment type attribution; Business application labels, including specific business application scenarios such as load forecasting, anomaly detection, fault diagnosis, equipment health management, energy efficiency analysis and optimization, and simulation training; Quality attribute labels, including quality scores, quality levels, and compliance status of core acceptance indicators across five dimensions: completeness, accuracy, timeliness, relevance, and security; Feature and annotation labels, including information such as the number of feature dimensions, core feature types, annotation types, annotation accuracy, and annotation scenarios; Compliance attribute labels, including information such as data security classification, security partition, access permission level, and compliance verification results.

[0154] 3. Automated generation and mounting of directories and tags: During the standardization and encapsulation of datasets, the system automatically extracts metadata information, classification information, quality rating information, feature annotation information, etc., and automatically generates corresponding tags according to the dimensions of the tag system; based on the classification of the dataset, the system automatically mounts the dataset to the corresponding level of asset directory and generates asset details entries without manual intervention; the automatically generated directories and tags support manual review and supplementation to ensure the accuracy of directory classification and the completeness of tag information.

[0155] 4. Implementation of Catalog and Tag Retrieval and Management Functions: Based on the asset catalog and tag system, multi-dimensional retrieval of datasets is realized, supporting hierarchical retrieval by catalog level, filtering retrieval by tag combination, and fuzzy retrieval by keyword, enabling rapid location of target datasets; a dynamic management mechanism for catalogs and tags is established, automatically updating the corresponding catalog ownership and tag information when a dataset completes version iteration, classification adjustment, or offline operation, ensuring the real-time and accuracy of catalogs and tags; based on the tag system, statistical analysis of dataset assets is realized, visually displaying the distribution of assets in different categories, different business scenarios, and different quality levels, providing support for data asset operation.

[0156] Version management, access control, and source tracing mechanisms specifically include:

[0157] 1. Steps for establishing and implementing a version management mechanism:

[0158] (1) Version management rules and version number specification: Formulate dataset version management specifications, clarify the triggering conditions, approval process, update rules, rollback mechanism, and offline rules for version iteration; adopt a three-stage version number specification of major version number, minor version number, and revision version number, and clarify the update rules for each stage: major version number, when the core measurement point list, feature dimension, and annotation system of the dataset undergo significant changes, or when the data time span expands by a large margin; minor version number, when the dataset completes incremental data updates, minor optimization of feature dimensions, and optimization of annotation rules, minor version number, revision version number, when the dataset completes data quality repair, metadata information correction, and error annotation correction, revision version number ...

[0159] (2) Version full-process control process design: Version generation: In the dataset encapsulation stage, the first generated dataset is assigned an initial version number V1.0.0, and a version description document is generated simultaneously to record the core information of the version, applicable scenarios, and updated content; Version iteration: When the dataset triggers the iteration condition, the system automatically generates a new version, updates the version number, and generates a version update log simultaneously to record the changed content, reasons for the change, changed personnel, and change time. After the new version is generated, it is simultaneously uploaded to the blockchain for evidence storage; Version archiving: All historical versions are fully archived and saved, supporting the query, download, and backtracking of historical versions. Historical versions are not deleted with version iteration to ensure the traceability of assets; Version rollback: When a new version has quality problems or business compatibility problems, it supports one-click rollback to a specified historical version and synchronously updates the default version information in the asset catalog; Version decommissioning: For dataset versions that are no longer applicable or have compliance risks, decommissioning operations are performed. After decommissioning, only the archived records and blockchain evidence storage information are retained, and services are no longer provided to the outside world. At the same time, the reasons for decommissioning and the approval process are recorded.

[0160] (3) Version management system function implementation and closed-loop optimization: In the high-value data asset generation and management platform, the full-process function of version management is implemented, supporting operations such as version query, version comparison, version iteration, version rollback, and version offline; a version iteration effect feedback mechanism is established, and the version management rules and iteration process are continuously optimized based on the business application effect of the new version.

[0161] 2. Steps for establishing and implementing an access control mechanism:

[0162] (1) Design of access control system to adapt to power security zoning requirements: Strictly follow the principle of least privilege and security zoning protection requirements of the power industry, design a three-level access control system of zoning isolation, hierarchical authorization and fine-grained control, and clarify that: Data set assets in the production control zone can only be authorized to be accessed within the production zone, and it is strictly prohibited to transmit core data set content to the management information zone across zones; Data set assets in the management information zone are subject to hierarchical authorization based on business needs; The role-based access control (RBAC) model is adopted, combined with the attribute-based access control (ABAC) model, to achieve fine-grained access control.

[0163] (2) Role and permission dimension division: Role division is based on the job responsibilities of the power industry, and the core roles such as system administrator, asset administrator, operation and maintenance personnel, business application personnel and auditors are divided. Each role has a default permission boundary. Permission dimension division realizes fine-grained control: Functional permissions include operation permissions such as query, preview, download, edit, iteration, offline, and approval of datasets; Data permissions include data access permissions such as accessible dataset categories, specific datasets, data time range, and measurement point range; Partition permissions clearly define the secure partitions that roles can access, and cross-partition access is strictly prohibited.

[0164] (3) Full-process control of permissions: Permission application and approval: Based on business needs, users submit permission applications, specifying the scope, purpose, and duration of the requested permissions. The corresponding permissions can only be granted after approval by both the asset administrator and the security administrator. Permission grant: Based on the approval results, users are granted corresponding roles and fine-grained permissions, following the principle of least privilege, and only the minimum permissions required to complete the business are granted. Permission audit: The entire process of user permission operation behavior is recorded, including the full-process logs of permission application, approval, grant, modification, and revocation, as well as the logs of permission-based dataset access, call, and download operations. Permission revoke: When user permissions expire, job responsibilities change, or the corresponding permissions are no longer needed, the system automatically or manually revokes the permissions to ensure closed-loop control of the entire lifecycle of permissions. Permission compliance verification: Permission compliance verification is carried out regularly to clean up unauthorized permissions, idle permissions, and permissions that do not meet compliance requirements, ensuring that permission control complies with the safety requirements of the power industry.

[0165] 3. Steps for establishing and implementing the traceability and inquiry mechanism:

[0166] (1) Design of full-link traceability information dimensions: Based on the above consortium blockchain evidence storage mechanism, the full-link traceability information dimensions of data assets are designed, covering: data source traceability information, access system of original data, source of measurement points, collection time, equipment information, access logs; governance process traceability information, full-process processing logs, algorithm parameters, processing personnel, and processing time of data preprocessing, feature engineering, annotation, and quality rating; asset generation traceability information, dataset encapsulation, metadata information, version number, encapsulation personnel, encapsulation time, and on-chain evidence storage hash value; full life cycle traceability information, full-process operation logs, approval records, and on-chain evidence storage information of dataset archiving, version iteration, permission change, access call, and offline.

[0167] (2) Traceability query process and function implementation: Unique identifier anchoring: Based on the unique global identifier of the dataset, it serves as the unique anchor point for traceability query. Users can initiate traceability query by inputting the unique identifier of the dataset; On-chain and off-chain data collaborative query: Based on the unique identifier, the system obtains the evidence storage hash value and operation log index of the dataset's entire lifecycle from the consortium chain, and obtains the corresponding detailed traceability information from the local database based on the index. At the same time, the integrity and immutability of the traceability information are verified through the hash value; Full-link visualization display: The traceability information is visualized according to the process of data access, governance, asset generation, and full lifecycle management, clearly presenting the complete process of the dataset from raw data to data assets; Multi-dimensional traceability query: Supports refined traceability query by link, by time, by operator, by operation type, etc., to meet the traceability needs of different scenarios such as compliance audit, business verification, and asset verification.

[0168] (3) Establishment of traceability and compliance application mechanism: compliance audit support, based on the traceability query results, provides a complete and tamper-proof evidence chain for data security compliance audit and industrial control security special inspection in the power industry; asset credibility verification, based on traceability information, verifies the authenticity, compliance and integrity of the dataset, and provides credible support for cross-entity sharing and trading of data assets; problem location and backtracking: when the dataset has quality problems or business adaptation problems, the problem link and root cause can be quickly located through traceability query, providing a basis for problem repair and process optimization.

[0169] S53. Construct a power business knowledge graph, take the target business scenario as input, use the power business knowledge graph to generate corresponding high-value dataset recommendation strategies, perform scenario-based adaptation verification based on the high-value dataset recommendation strategies, and establish a dynamic iteration mechanism to optimize the data asset library, so as to realize the governance of power production monitoring data and the generation of high-value data assets.

[0170] In this optional embodiment, a power business knowledge graph is constructed. Using the target business scenario as input, a corresponding high-value dataset recommendation strategy is generated using the power business knowledge graph. Scenario-based adaptation verification is performed based on the high-value dataset recommendation strategy, and a dynamic iteration mechanism is established to optimize the data asset library. This achieves power production monitoring data governance and high-value data asset generation, including:

[0171] S531. Based on high-value datasets in the data asset library, define power business scenarios, digital applications, algorithm models, datasets, feature dimensions, and equipment types as entities, and based on preset power industry standards, business rules, and historical application cases, set semantic relationships between entities to construct a power business knowledge graph.

[0172] S532. Using the target business scenario as input, and leveraging entity association reasoning and semantic similarity calculation of the power business knowledge graph, generate a corresponding high-value dataset recommendation strategy.

[0173] It should be further explained that, taking the target business scenario as input, including the equipment model of a 1.5MW doubly-fed wind turbine, the fault type of early gearbox bearing failure, and the application purpose of the supervised learning model training, the system calls upon the power business knowledge graph deployed on the Neo4j graph database. This graph contains 128,000 entities across 6 categories and 372,000 semantic relationships. Cypher path query statements are executed to achieve entity association reasoning. A depth-first traversal is performed along predefined association paths based on business scenario, digital applications, algorithm models, datasets, feature dimensions, and equipment types, automatically filtering cross-domain entities such as thermal power and photovoltaics. A total of 92,000 entities were identified, with 127 strongly relevant entities precisely targeted. Structured semantic association replaced traditional keyword retrieval, eliminating irrelevant data interference at its source. A power sector-specific semantic embedding model, fine-tuned based on BERT-base, was invoked to map the target business scenario request and the metadata of all candidate datasets—including data source, feature dimensions, annotation rules, and quality rating—to a 768-dimensional unified vector space. Cosine similarity between vectors was calculated, with a similarity threshold of 0.75. The calculated similarity for the high-value wind turbine gearbox vibration dataset V2.1 was 0.92. The similarity between the V1.3 dataset and the general wind power equipment vibration dataset is 0.53, resulting in three candidate datasets that meet the threshold requirements. A weighted ranking algorithm is then used to prioritize these candidate datasets, with weights allocated as follows: semantic similarity 0.6, dataset quality score (annotation accuracy, sample balance, feature coverage) 0.2, and historical application performance score 0.2. This generates a ranked list of candidate datasets. Combining the dataset metadata stored in the knowledge graph with the association rule mining results, the association confidence score between oil temperature features and early gearbox bearing failure is 0.89, and the support score is 0.72. A complete high-value dataset recommendation strategy is generated, identifying the optimal recommended dataset as the wind turbine gearbox vibration high-value dataset V2.1. This dataset contains 1200 normal operating condition samples and 320 early gearbox bearing failure samples. Additionally, a wind turbine gearbox oil temperature correlation feature dataset is recommended. The recommendation strategy also includes directly applicable engineering information such as dataset call interfaces, applicable model training hyperparameter ranges, and data preprocessing specifications. This can shorten the dataset preparation time in the development process of power digitalization business models, improve dataset matching accuracy, significantly reduce manual screening costs, and improve model implementation effectiveness.

[0174] S533. Based on the high-value dataset recommendation strategy, call the corresponding high-value dataset from the data asset library for scenario-based adaptation verification, and apply the verified high-value dataset to the target business scenario to obtain feedback on the effect of the business application.

[0175] S534. Based on the feedback, construct a dynamic iteration mechanism and use the dynamic iteration mechanism to optimize the high-value dataset. Based on the optimization results, update the power business knowledge graph and the recommendation strategy for the high-value dataset to optimize the data asset library.

[0176] It should be further explained that, in the process of standardized generation and full lifecycle management of high-value power data assets, this invention introduces a trusted traceability and evidence storage mechanism for the entire chain of high-value data assets based on the power industrial control consortium blockchain, and an intelligent recommendation and dynamic iteration engine for datasets based on power business knowledge graphs, as detailed below:

[0177] 1. A trusted traceability and evidence storage mechanism for high-value data assets across the entire power industrial control alliance chain:

[0178] In existing technologies, the traceability and management of power data assets generally adopt a centralized database log recording model, which suffers from problems such as log tampering, incomplete traceability, and insufficient reliability of cross-entity data sharing. Furthermore, it does not adapt to the compliance requirements of power industrial control system security partitions and cannot meet the high security and high reliability requirements of core power production data assets. This invention deeply integrates consortium blockchain technology with the full lifecycle management of power data assets, constructing a consortium blockchain evidence storage and traceability mechanism adapted to power security partitions. The specific implementation method is as follows:

[0179] (1) Compliance deployment of consortium blockchain architecture:

[0180] Strictly adhering to the power industry's security zoning principles, a two-tiered consortium blockchain architecture is constructed: Production control zone II deploys production consortium blockchain nodes, connecting only to the time-series library, AI analysis engine, and asset generation module. These nodes record the entire process of data asset management, from governance and labeling to encapsulation and archiving, ensuring tamper-proof evidence preservation of the asset generation process. Management information zone deploys management consortium blockchain nodes, achieving one-way data synchronization with the production consortium blockchain nodes through a one-way isolation device. Data can only be synchronized from the production zone and cannot be written back. These nodes record the entire lifecycle of data asset access, invocation, version iteration, sharing, and decommissioning, providing trusted traceability services for business applications. The consortium blockchain employs a permissioned access mechanism, allowing only authorized system modules, maintenance personnel, and business application entities to access it, fully complying with the security and compliance requirements of power industrial control systems.

[0181] (2) Implementation process of end-to-end evidence storage and traceability:

[0182] Asset generation process evidence storage: During the standardized packaging of the dataset, the unique global identifier, version number, metadata information, governance full-link logs, annotation records, quality rating reports, packaging personnel information and other core metadata of the dataset are simultaneously generated into hash values ​​and stored on the production consortium blockchain node to ensure that the entire asset generation process is traceable and tamper-proof.

[0183] Asset lifecycle notarization: Throughout the entire lifecycle of data assets, including archiving, version iteration, permission changes, access calls, and decommissioning, each operation generates a corresponding operation log and hash value, which is then stored on the blockchain to the corresponding consortium blockchain node, forming a complete notarization chain for the entire lifecycle of data assets.

[0184] Trusted traceability query: For any data asset, based on its unique global identifier, the complete information of the asset can be traced from the consortium blockchain, from the original data access, governance, feature extraction, labeling, encapsulation, archiving to invocation, to verify the authenticity, compliance and integrity of the asset.

[0185] (3) Compliance and security assurance: All data on the consortium blockchain is only metadata hash value and does not contain core production data and dataset content, thus avoiding core data leakage; only one-way synchronization of hash values ​​is achieved between the production area and the management area, which fully complies with the horizontal isolation security requirements of the power industry and will not lead to the illegal cross-area outflow of production data.

[0186] 2. Intelligent recommendation and dynamic iteration engine for datasets based on power business knowledge graph:

[0187] In existing technologies, the management of power data assets generally adopts a static archiving and manual retrieval model, which cannot achieve intelligent matching of datasets with business scenarios. Furthermore, the iterative optimization of datasets relies on manual feedback, failing to achieve automatic dynamic iteration based on business application effects, resulting in insufficient data asset reuse efficiency and value release capabilities. This invention constructs a power business knowledge graph to achieve intelligent recommendation and dynamic iteration of datasets. The specific implementation method is as follows:

[0188] (1) Construction of power business knowledge graph: The power business knowledge graph covers six core entities: power business scenarios, digital and intelligent applications, algorithm models, datasets, feature dimensions, and equipment types. It defines the relationships between entities. For example, the fault diagnosis scenario is associated with the turbine bearing fault diagnosis model, which is associated with the high-value dataset of turbine vibration, which is associated with the turbine vibration feature dimension and the turbine generator set equipment. Based on power industry standards, business rules, and historical application cases, the knowledge graph is continuously iterated and improved to provide a semantic foundation for intelligent recommendation and dynamic iteration.

[0189] (2) Intelligent recommendation mechanism for datasets: In the scenario-based adaptation and verification process, and in the process of providing data assets to external customers, intelligent recommendation of datasets is realized based on business knowledge graphs: For business application needs, such as wind turbine gearbox fault diagnosis model training, the highest value dataset with the highest adaptability is automatically recommended through semantic matching of knowledge graphs, and the feature dimensions, annotation information, historical application effects and recommendation reasons of the dataset are given; For business applications that have called datasets, supplementary feature datasets and optimized datasets are automatically recommended based on model training effects and business application accuracy to improve the effect of business models.

[0190] (3) Dynamic iteration closed-loop mechanism for datasets: Based on the feedback of business application effects, an automatic dynamic iteration mechanism for datasets is constructed: For datasets that perform poorly in business applications, the root cause of the problem is automatically analyzed: If the feature dimension is insufficient, it is fed back to feature engineering processing to supplement and mine relevant business features; if the annotation accuracy is insufficient, it is fed back to intelligent annotation processing to optimize the annotation rules; if the data quality is insufficient, it is fed back to preprocessing for secondary governance; after optimization, a new version of the dataset is automatically generated, the metadata information and version number are updated, and the data is synchronously stored on the blockchain to complete the automatic iteration update of the dataset; based on the application effect of the iterated dataset, the entity association relationship and recommendation strategy of the knowledge graph are continuously optimized to form a complete closed loop of dataset application, feedback, optimization and iteration.

[0191] In this optional embodiment, before constructing a storage-computing integrated underlying support environment based on a time-series database, the raw time-series data from multi-source power production monitoring is accessed, and storage management and security control are performed synchronously during the access process, and the time-series dataset is output, the following is also included:

[0192] Based on the pre-set safety protection principles of the power industry, the deployment environment is divided into several safety domains, including the first production control area safety domain, the second production control area safety domain, and the management information area.

[0193] Based on the preset safety level requirements, the time-series database and the lightweight digital twin of the power equipment are deployed in the first production control area safety zone; the artificial intelligence algorithm, intelligent labeling and multi-dimensional quality rating system are deployed in the second production control area safety zone.

[0194] A dedicated industrial firewall for power is deployed between the first production control zone security area and the second production control zone security area, and the dedicated industrial firewall for power only opens a one-way data synchronization link; the first production control zone security area and the second production control zone security area are used as production control zone security areas, and a one-way isolation device is deployed between the production control zone security area and the management information area to build a full-link security protection strategy;

[0195] After deployment, a compliance verification process is performed, including special testing of power industrial control security, full-item verification of functions and performance, and production trial operation for no less than a preset time. Based on the compliance verification results, the security protection strategy for the entire chain is optimized.

[0196] It should be further explained that, in accordance with the security protection principles of security zoning, dedicated networks, horizontal isolation, and vertical authentication in the power industry, the system was deployed and put into operation in a hierarchical and domain-specific manner to ensure that the solution complies with the security compliance requirements of power industrial control systems. Specifically, this includes:

[0197] 1. Secure partitioning deployment: The core real-time storage capabilities of the time series library and the lightweight real-time AI inference module are deployed in Security Zone I (First Production Control Zone) of the production control area to meet the low latency and high security requirements of real-time control; the batch AI analysis training, feature engineering, and high-value dataset management modules are deployed in Security Zone II (Second Production Control Zone) to achieve secure decoupling of real-time business and offline analysis.

[0198] 2. Cross-zone security interaction control: Security Zone I and II are logically isolated using a dedicated industrial firewall for power, with only a one-way data synchronization link open; the production control zone and the management information zone are physically isolated using a dedicated one-way isolation device, and the unauthorized cross-zone outflow of high-value datasets and core production data is strictly prohibited.

[0199] 3. Compliance verification before commissioning: After deployment, conduct special safety testing of power industrial control systems, full-item verification of functions and performance, and no less than 3 months of production trial operation to ensure that all indicators of the system meet the acceptance standards and comply with the safety and compliance requirements of the power industry.

[0200] The process follows a linear progression of compliance deployment, security control, and verification and acceptance, as follows: The input to security partition deployment is the security protection principles of security partitioning, dedicated networks, horizontal isolation, and vertical authentication in the power industry, as well as the functional positioning and real-time requirements of each system module. The output is a hierarchical and domain-based deployment architecture and module deployment scheme that conforms to security partitioning requirements. In other words, systems without compliant partition deployment cannot formulate targeted cross-regional security interaction control rules. The input to cross-regional security interaction control is the partition deployment architecture, and the output is cross-regional interaction between security zones I / II and between the production control zone and the management information zone. The control rules, isolation device configuration scheme, and data one-way synchronization link specifications provide rigid constraints for cross-regional data interaction throughout the entire process from steps S1 to S5, and also serve as the core basis for compliance verification. The inputs for pre-production compliance verification are the deployment architecture, security control rules, and power industry industrial control security standards and system acceptance indicators. The outputs are the system compliance verification report, functional performance acceptance results, and production permit recommendations. The verification results of this output are used to optimize the deployment scheme and security control rules in reverse, ensuring that the entire system process complies with the power industry's compliance requirements, while also providing compliance boundaries for the continuous iterative optimization of the system.

[0201] This step provides compliance assurance for the entire process of power industrial control scenarios, outputting a hierarchical and domain-based deployment architecture that conforms to the power industry's security protection principles, cross-regional security interaction control rules, and a compliance verification report and performance acceptance results before system deployment. This output provides rigid compliance constraints for the entire process execution of steps S1 to S5: from the deployment of the time-series library in step S1, the deployment of AI governance in S2, the deployment of feature engineering in S3 and the deployment of annotation and rating in S4, to the deployment of the asset management platform in S5, all must strictly follow the established security partitioning deployment rules, and the data interaction and cross-regional transmission throughout the process must follow the cross-regional security interaction control mechanism. At the same time, the compliance verification results will reversely optimize the full-link security control strategy to ensure that the entire process strictly complies with the power industrial control security compliance requirements.

[0202] Furthermore, this invention constitutes a complete closed-loop logic of linear progression and reverse iteration across the entire chain, encompassing basic support, core governance, value transformation, compliance implementation, and closed-loop optimization, forming a self-consistent, closed-loop, and iterative execution system. Breaking through the bottleneck of traditional architectures that separate storage and computing, this system leverages a distributed time-series database that integrates storage and computing to extend AI data governance capabilities to storage nodes. This ensures that computing occurs wherever the data is located, significantly improving the collaborative efficiency of massive time-series data governance and reducing computing power consumption and processing latency. It constructs a full-chain AI data governance system adapted to all scenarios of power production monitoring, achieving fully automated processing from intelligent preprocessing of raw data, power-specific feature engineering, intelligent labeling and quality rating to the standardized generation of high-value datasets, thereby improving data quality and asset value density. It deeply integrates full-chain data security control, abnormal data identification, processing and governance, and asset generation processes to meet the compliance requirements of security zoning and hierarchical protection in power industrial control systems, ensuring the security and trustworthiness of data assets throughout their entire lifecycle. Finally, it establishes a high-value dataset classification, packaging, archiving, and management system that conforms to power industry standards, enabling traceability and reusability of data assets and supporting the implementation of digital and intelligent business applications such as power load forecasting, equipment anomaly detection and fault diagnosis, equipment health management, and energy efficiency analysis and optimization.

[0203] Specifically, this invention relies on a storage-compute integrated distributed time-series database architecture to deeply integrate AI data governance capabilities into the storage layer. It constructs a complete technical system that is adapted to all scenarios of power production monitoring, integrating full-link data security control, full-process AI governance, and standardized generation of high-value data assets. It breaks through the efficiency bottleneck caused by the separation of storage and computing in traditional solutions, and solves the core problems of insufficient business adaptability, disconnection of security control, and low standardization in the process of power time-series data governance and asset generation.

[0204] From an underlying architectural perspective, traditional solutions generally employ a storage-compute separation architecture. Data governance necessitates migrating massive amounts of data from the storage layer to the compute layer for execution. This process inevitably incurs significant data transfer overhead, directly leading to high processing latency, high computational cost, and high system resource utilization. This bottleneck is amplified, especially when dealing with petabyte-scale power time-series data and high-concurrency scenarios involving tens of millions of measurement points, potentially even causing system blockage and business interruption. In contrast, this invention adopts a storage-compute integrated architecture, sinking AI governance capabilities to storage nodes. This ensures that computation occurs wherever the data is located, eliminating the need for large-scale cross-node data migration and removing the additional overhead from data transfer at the root of the architecture. Therefore, it significantly improves governance efficiency, reduces processing latency and system resource consumption, and can deeply collaborate with the high-concurrency read / write and efficient query capabilities of time-series databases, maintaining stable performance even under mixed write and governance load scenarios. From a business adaptability perspective, traditional solutions either target only a single power scenario or employ generalized governance schemes, failing to deeply integrate with the business rules and data characteristics of power production monitoring. This results in data that has only undergone basic format unification after governance, without uncovering the underlying business value, exhibiting low value density, and being unable to directly support digital and intelligent business applications, still requiring secondary processing on the business side. In contrast, this invention constructs a full-link AI governance system that deeply integrates with power business rules for all power scenarios, including centralized monitoring of thermal power, wind power, and new energy. It can accurately extract high-value features strongly correlated with power business and generate standardized labeled datasets that meet business needs. Therefore, it can significantly improve data quality and business adaptability. The generated high-value datasets can be directly used for model training and business analysis without secondary processing, greatly reducing the implementation cost of digital and intelligent power business. From a security compliance perspective, traditional solutions largely separate data security control from data governance and asset generation processes, providing only basic protection at the storage or access stages without achieving end-to-end integration. Furthermore, they fail to consider the specific compliance requirements of power industrial control systems' security zoning, making them highly susceptible to security risks such as unauthorized data access and unauthorized cross-regional transmission, thus failing to meet the high security demands of core power production scenarios. This invention, however, integrates data encryption / decryption, abnormal data identification, and fine-grained access control throughout the entire process of data access, governance, asset generation, and access. It also strictly adheres to the principles of power security zoning, designing a hierarchical deployment architecture and cross-regional interaction control mechanism. Therefore, it can fully meet the compliance requirements of power industry industrial control security, achieving full lifecycle security protection for data assets and fundamentally eliminating the risks of data leakage and unauthorized access. From a data assetization perspective, traditional solutions mostly stop at the data governance stage, failing to build a complete standardized asset generation and full lifecycle management system. This results in fragmented storage of governed data, lack of unified standards, version control, and business labeling, preventing the formation of reusable and traceable standardized data assets, and hindering the full realization of the value of massive amounts of data.This invention establishes a dataset classification system, encapsulation standards, version management, and archiving and reuse mechanism that conforms to power industry standards. It achieves a complete closed loop of transformation from raw data to high-value data assets, thus enabling standardized management and efficient reuse of power data assets. This provides technical support for the formation of new power data asset business models and can quickly adapt to the reuse needs of different business scenarios, possessing strong replicability and promotional value. The advantages of this invention are mainly reflected in high efficiency of storage-computing collaborative governance, strong adaptability to power business scenarios, good end-to-end security and compliance, high degree of data asset standardization, and strong engineering implementation. These advantages help improve the governance efficiency and asset value of massive time-series data in new power systems, meet the high-concurrency, high-security, and high-reliability business needs of power production monitoring scenarios, and provide solid technical support for the digital transformation and intelligent business implementation of the power industry.

[0205] like Figure 2 As shown, according to another embodiment of the present invention, a power production monitoring data governance and high-value data asset generation system based on an in-store computing time-series database is also provided. This system includes:

[0206] Data support and acquisition module 1 is used to access the raw time-series data of multi-source power production monitoring in the construction of an integrated storage and computing underlying support environment based on time-series database, and to simultaneously perform storage management and security control during the access process, and output time-series datasets;

[0207] Anomaly detection and management module 2 is used to call the preset lightweight digital twin of power equipment based on the time series dataset, generate a dynamic operating condition baseline, and perform anomaly detection and management on the time series dataset in combination with the dual-track detection mechanism, and output a standardized time series dataset.

[0208] The data filtering and optimization module 3 is used to match the artificial intelligence algorithms corresponding to the standardized time series dataset, perform feature mining based on the matching results, and use a two-dimensional filtering mechanism to filter and optimize the feature mining results to generate a standardized high-value feature set.

[0209] Data asset generation module 4 is used to construct an intelligent labeling and multi-dimensional quality rating system based on the preset power industry business specifications, and to use the intelligent labeling and multi-dimensional quality rating system to label and filter the standardized high-value feature set to obtain a high-value dataset.

[0210] The data asset management module 5 is used to classify and structure high-value datasets and store the encapsulation results in a preset data asset library. Combined with the pre-deployed power industrial control alliance chain and power business knowledge graph, the data asset library is optimized to achieve power production monitoring data governance and high-value data asset generation.

[0211] It should be further noted that this invention adopts a three-layer loosely coupled, highly cohesive architecture. The architecture diagram, from top to bottom, consists of: a high-value data asset generation and management platform (upper-layer application), a time-series data AI analysis engine (middle-layer core), and a storage-and-computing integrated distributed time-series database (lower-layer support). It also incorporates a built-in end-to-end security management module, running through the entire three-layer architecture process, to achieve closed-loop management of power production monitoring data from access, storage, and governance to asset generation; for example... Figure 4 and Figure 5 As shown, the detailed design of the core architecture is as follows:

[0212] 1. Support Layer: In-Storage Computing Distributed Time-Series Database (equivalent to Data Support and Acquisition Module 1): This layer provides the underlying data storage, read / write scheduling, and basic security capabilities for the entire system. It adopts a distributed multi-node domestically produced server cluster deployment architecture, providing underlying core support for in-store and compute collaborative governance. Core modules include:

[0213] (1) High-concurrency read and write module: Supports concurrent writing and reading of tens of millions of measurement points per second in a single cluster, adapts to the millisecond / second sampling frequency of the power SCADA system, is compatible with mixed writing of strictly increasing and out-of-order timestamp data, has built-in write path data preprocessing capability, and realizes format verification and anomaly screening in the data entry stage.

[0214] (2) Distributed storage management module: It adopts a cold and hot data tiered storage and adaptive dynamic compression mechanism to support the persistent storage and compression of PB-level massive time-series data; through distributed data sharding and node load balancing strategies, it realizes data storage and computation in the nearest location, providing the underlying scheduling foundation for storage and computation collaboration.

[0215] (3) High-efficiency query engine module: The access path is optimized for the three core query modes of the power scenario (real-time monitoring query of the latest value, query of the time range of historical curves, and full cross-section query at a specific moment), ensuring that the query response latency is in the second level under the scale of tens of millions of measurement points, and supporting multi-condition combination query, aggregation query and large time span data query.

[0216] (4) Data persistence and fault tolerance module: Based on the write-ahead log mechanism, the data writing persistence is guaranteed, and the data recovery is supported after node crash or process abnormal interruption, ensuring zero loss of written data; through the multi-replica redundancy mechanism, cluster-level fault tolerance and high data availability are achieved.

[0217] (5) Basic security and anomaly screening module: It has built-in basic capabilities such as static data storage encryption, transmission encryption, and fine-grained access control. At the same time, it integrates the initial screening rules for typical abnormal data in the power scenario, and performs preliminary marking of out-of-bounds, mutated, missing, and noisy data, providing a preliminary foundation for upper-level AI governance.

[0218] 2. Core Layer, Time-Series Data AI Analysis Engine (equivalent to Anomaly Detection and Governance Module 2 and Data Filtering and Optimization Module 3): This layer is the core processing unit of the system, deeply integrated into the storage nodes of the in-memory computing time-series database. It enables on-site data governance, eliminating the performance loss of cross-node data transmission in traditional architectures at its source. Core modules include:

[0219] (1) Storage and computing collaborative scheduling module: responsible for node-level scheduling and resource allocation of AI governance tasks. According to the storage location of data shards, the governance tasks are sent to the corresponding storage nodes for local execution, so that the data does not move but the computing moves. It supports dual-mode scheduling of real-time streaming AI governance and offline batch AI governance, and adapts to different governance needs of high-concurrency real-time data and PB-level historical data.

[0220] (2) Intelligent preprocessing module: It has a built-in preprocessing algorithm library specifically for power scenarios, and the core includes:

[0221] ① Intelligent missing value filling unit: Supports algorithms such as forward filling, mean filling, linear filling, and adaptive interpolation filling. It can automatically select the optimal filling strategy according to the operating conditions of power equipment, and mark and distinguish legitimate missing values ​​for business scenarios such as wind turbine shutdown and unit maintenance to avoid incorrect filling.

[0222] ② Intelligent outlier management unit: It integrates anomaly detection algorithms such as 3-Sigma, Z-Score, Isolation Forest, and KNN, and combines the rated operating threshold of power equipment and business rule base to accurately identify out-of-bounds data, mutated data, noisy data, and erroneously collected data. It supports custom configuration of three processing strategies: filtering, marking, and correction.

[0223] ③ Standardized and normalized unit: Automatically identifies differences in timestamp format, unit, and range of multi-source data, and converts them into the standard format of the power industry. Based on algorithms such as Min-Max and Z-Score, it realizes data scale normalization and eliminates the dimensional differences of data from different measuring points.

[0224] (3) Power-Specific Feature Engineering Module: This module constructs a power industry-specific feature library based on the temporal, periodic, and business-related characteristics of power production monitoring data. The core features include:

[0225] ① Basic Time Series Feature Extraction Unit: Automatically calculates basic time series features such as maximum / minimum, mean, median, variance, standard deviation, root mean square, slope, peak-to-peak value, and periodicity factor through a sliding window to capture the fluctuation patterns and trend changes of equipment operation data.

[0226] ② Business Feature Mining Unit: Based on KMeans clustering, PCA dimensionality reduction, and association rule mining algorithms, combined with business rules from scenarios such as thermal power, wind power, and new energy centralized control, it automatically mines high-value business-specific features such as equipment health features, load fluctuation features, fault warning features, and energy efficiency features.

[0227] ③ Feature Filtering and Optimization Unit: Based on Pearson correlation coefficient and Spearman correlation coefficient, the extracted features are evaluated for weight and redundant features are removed. Features that are highly relevant to business objectives are retained, feature dimensionality is reduced, and the business adaptability of the dataset is improved.

[0228] (4) Intelligent annotation and quality rating module, the core of which includes:

[0229] ① Scenario-based intelligent annotation unit: Construct a power industry annotation rule library, which automatically classifies and annotates fault data, operating condition data, alarm data, energy efficiency data, etc., by combining historical fault cases and business rules. It supports semi-supervised learning annotation optimization and manual review and correction, and core fault samples achieve 100% expert review.

[0230] ② Multi-dimensional quality rating unit: A quality scoring model is constructed from five dimensions: completeness, accuracy, timeliness, relevance, and security. AI automatically scores and rates the data, classifying it into high, medium, and low value levels, and selecting a subset of high-value data. The core thresholds are: data completeness ≥ 98%, data content accuracy ≥ 99%, and business labeling accuracy ≥ 97%.

[0231] ③ Quality closed-loop optimization unit: For low- and medium-value data, it automatically feeds back to the preprocessing module for secondary optimization, forming a closed-loop mechanism for data quality governance.

[0232] 3. Application Layer, High-Value Data Asset Generation and Management Platform (equivalent to Data Asset Generation Module 4 and Data Asset Management Module 5): This layer is the system's external service and management unit, realizing standardized generation, full lifecycle management, and business scenario adaptation of data assets. Core modules include:

[0233] (1) Dataset classification management module: Construct a new type of high-value time series dataset classification system for power systems, which is divided into four categories: centralized monitoring of new energy, operation and control of thermal power, health of wind power equipment, and general power digitalization. It supports multi-dimensional classification retrieval and management by scenario, equipment type, and business purpose.

[0234] (2) Asset standardization and encapsulation module: Strictly follow the power industry dataset construction specifications, structure and encapsulate high-value data subsets, and generate standardized dataset description documents in a synchronous manner, including metadata information such as data source, measurement point list, feature dimension, annotation rules, quality rating, version number, and recommended application scenarios, to ensure the interpretability, traceability and reusability of the dataset.

[0235] (3) Full lifecycle management module: realizes full lifecycle management of datasets from generation, archiving, version iteration, permission control, source tracing and query to decommissioning, supports batch incremental updates, version rollback and access auditing of datasets, and establishes a closed-loop iterative optimization mechanism for data assets.

[0236] (4) Scenario-based adaptation verification module: For core power business scenarios such as load forecasting, anomaly detection, fault diagnosis, equipment health management, and energy efficiency analysis and optimization, it provides dataset call testing, effect verification, and feedback optimization functions to verify the business adaptability of the dataset and continuously optimize the feature system and annotation rules based on business feedback.

[0237] (5) Visualized Operation and Maintenance Module: Provides a visual display of system running status, governance task progress, dataset asset statistics, quality indicators, and resource usage, and supports custom configuration of governance processes and operation and maintenance alarms.

[0238] 4. Full-link security management system: This system runs through the entire process of the three-layer architecture of the system. The core modules include: transmission encryption module, storage encryption and decryption module, key lifecycle management module, fine-grained access control module, cross-regional secure interaction module, and security audit module, realizing full-link security protection for data transmission, storage, governance, asset generation, and access.

[0239] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for power production monitoring data governance and high-value data asset generation based on a memory-compute integrated timing library, characterized in that, The method includes: S1. In the underlying support environment of building a storage-computing integrated system based on a time-series database, the raw time-series data of multi-source power production monitoring is accessed, and storage management and security control are performed synchronously during the access process to output the time-series dataset. S2. Based on the time-series dataset, call the preset lightweight digital twin of power equipment to generate a dynamic operating condition baseline, and combine the dual-track detection mechanism to perform anomaly detection and management on the time-series dataset, and output a standardized time-series dataset. S3. Match the artificial intelligence algorithm corresponding to the standardized time series dataset, perform feature mining based on the matching results, and use a two-dimensional screening mechanism to screen and optimize the feature mining results to generate a standardized high-value feature set. S4. Based on the preset business specifications of the power industry, construct an intelligent annotation and multi-dimensional quality rating system, and use the intelligent annotation and multi-dimensional quality rating system to annotate and filter the standardized high-value feature set to obtain a high-value dataset. S5. Classify and structure high-value datasets, and store the encapsulation results in a pre-defined data asset library. Combine the pre-deployed power industrial control alliance chain and power business knowledge graph to optimize the data asset library, so as to realize the governance of power production monitoring data and the generation of high-value data assets.

2. The method according to claim 1, wherein, The process of building a storage-computing integrated underlying support environment based on a time-series database, accessing raw time-series data from multi-source power production monitoring, and simultaneously performing storage management and security control during the access process before outputting the time-series dataset includes: Based on the preset safety protection principles of the power industry, the deployment environment is divided into several security domains, including the first production control area security domain, the second production control area security domain, and the management information area. Based on the preset safety level requirements, the time-series database and the lightweight digital twin of the power equipment are deployed in the first production control area safety zone; the artificial intelligence algorithm, intelligent labeling and multi-dimensional quality rating system are deployed in the second production control area safety zone. A dedicated industrial firewall for power is deployed between the first production control zone security area and the second production control zone security area, and the dedicated industrial firewall for power only opens a one-way data synchronization link; the first production control zone security area and the second production control zone security area are used as production control zone security areas, and a one-way isolation device is deployed between the production control zone security area and the management information area to build a full-link security protection strategy; After deployment, a compliance verification process is performed, including special testing of power industrial control security, full-item verification of functions and performance, and production trial operation for no less than a preset time. Based on the compliance verification results, the security protection strategy for the entire chain is optimized.

3. The method according to claim 1, wherein, Based on the time-series dataset, a preset lightweight digital twin of power equipment is invoked to generate a dynamic operating condition baseline. An anomaly detection and mitigation mechanism is then used to perform anomaly detection on the time-series dataset, resulting in a standardized time-series dataset including: S21. Based on the sharding information of the time series dataset, the preset lightweight digital twin of the power equipment is called locally on the storage node of the time series database, and the power equipment information and operating characteristics are combined to match the current operating conditions of the power equipment in real time and generate a dynamic operating condition baseline. S22. Based on the dynamic operating condition baseline and the preset power equipment business rules, distinguish between legal missing data and abnormal missing data in the time series dataset, and adaptively select a differentiated filling strategy according to the current operating condition type to generate a continuous time series dataset. S23. Based on the continuous time series dataset and the dynamic operating condition baseline, construct a dual-track detection mechanism based on baseline rule verification and anomaly detection. Use the dual-track detection mechanism to perform anomaly detection and governance, and output a time series dataset without anomaly interference. S24. Perform data format standardization and normalization processing on the time series dataset without abnormal interference to generate a standardized time series dataset.

4. The method for power production monitoring data governance and high-value data asset generation based on an in-store computing time-series database according to claim 3, characterized in that, The step involves constructing a dual-track detection mechanism based on baseline rule verification and anomaly detection, using a continuous time-series dataset and a dynamic operating condition baseline. This mechanism performs anomaly detection and mitigation, outputting a time-series dataset free of anomaly interference. S231. Based on the initial screening labeling results and fragmentation information in the continuous time series dataset, perform initial anomaly classification and detection on the time series data, and output the dataset to be detected. S232. Based on the dynamic operating condition baseline and combined with the preset power business rule base, perform rule-level anomaly verification on the dataset to be tested, and generate a full dataset based on the verification results. The rule-level anomaly verification includes boundary crossing anomaly verification, mutation anomaly verification, and logic anomaly verification. S233. A cascaded detection mode combining statistical algorithms and machine learning algorithms is adopted. On the local storage node, a secondary anomaly classification and detection is performed on the full dataset, and an anomaly data classification set is output. S234. Combining the abnormal data classification set with the preset power business rule base, adaptively match the corresponding governance strategy; S235. On the local storage node, perform abnormal data governance on the abnormal data classification set according to the matching governance strategy, and output a time series dataset without abnormal interference.

5. The method for power production monitoring data governance and high-value data asset generation based on an in-store computing time-series database according to claim 1, characterized in that, The artificial intelligence algorithm corresponding to the standardized time-series dataset is used to perform feature mining based on the matching results, and a two-dimensional screening mechanism is used to filter and optimize the feature mining results to generate a standardized high-value feature set, including: S31. Use the sliding window technique to extract time-series features from the standardized time-series dataset, generate a time-series feature set, and match the corresponding power feature mining template according to the power equipment type and business target scenario. S32. Based on the preset power business rules, perform operating condition identification and data segmentation on the standardized time series dataset to generate different operating condition segments, and construct the target feature dimension by combining the power feature mining template. S33. Based on the target feature dimension, match the corresponding artificial intelligence algorithm, perform feature mining on the time series feature set based on the matching result, and generate full business features; S34. Combining the preset power business rule base, historical fault case base and power industry standards and specifications, perform business validity verification on all business features, and merge the verified all business features with the time series feature set to generate a full feature set. S35. A dual-dimensional screening mechanism combining mathematical correlation screening and power business weight assignment is used to screen and optimize the full feature set and generate a standardized high-value feature set.

6. The method for power production monitoring data governance and high-value data asset generation based on an in-store computing time-series database according to claim 5, characterized in that, The dual-dimensional screening mechanism, which combines mathematical correlation screening with power business weight assignment, filters and optimizes the entire feature set to generate a standardized high-value feature set, including: S351. Based on the full feature set, calculate the Pearson correlation coefficient between the full features and the preset business target labels, and perform linear correlation screening in combination with the preset threshold. S352. Based on the linear correlation screening results, calculate the Spearman correlation coefficient and maximum information coefficient between the full set of features and the preset business target labels, and perform nonlinear correlation screening based on the calculation results to obtain the mathematical correlation screening results. S353. Based on the business objective scenario, construct a power business weight evaluation system, assign weights to all features from preset dimensions, prioritize all features based on the weight assignment results, and output a priority list of all features. S354. Combining the mathematical correlation screening results with the full feature priority list, set a business weight threshold, and use the weight threshold to perform feature screening on the full feature set, outputting a standardized high-value feature set.

7. The method for power production monitoring data governance and high-value data asset generation based on an in-store computing time-series database according to claim 1, characterized in that, The process of classifying and structuring high-value datasets, storing the encapsulation results in a pre-defined data asset repository, and optimizing the data asset repository in conjunction with a pre-deployed power industrial control alliance chain and power business knowledge graph, to achieve power production monitoring data governance and high-value data asset generation includes: S51. Construct a high-value dataset classification system based on centralized monitoring of new energy, operation and management of thermal power, health of wind power equipment, and digitalization of general power, and use the high-value dataset classification system to classify high-value datasets. S52. The high-value dataset after classification is structured and encapsulated to generate hash values. The hash values ​​are then stored on the blockchain through a pre-deployed power industrial control alliance chain to achieve full-link trusted traceability and full lifecycle management of data assets. S53. Construct a power business knowledge graph, take the target business scenario as input, use the power business knowledge graph to generate corresponding high-value dataset recommendation strategies, perform scenario-based adaptation verification based on the high-value dataset recommendation strategies, and establish a dynamic iteration mechanism to optimize the data asset library, so as to realize the governance of power production monitoring data and the generation of high-value data assets.

8. The method for power production monitoring data governance and high-value data asset generation based on an in-store computing time-series database according to claim 7, characterized in that, The process of structurally encapsulating the classified high-value dataset, generating hash values, and then storing these hash values ​​on a pre-deployed power industrial control consortium blockchain to achieve end-to-end trusted traceability and full lifecycle management of data assets includes: S521. Based on the preset power industry safety zoning principle, deploy production alliance chain nodes in the second production control area safety zone and management alliance chain nodes in the management information area to realize the deployment of the power industrial control alliance chain. S522. According to the preset power industry dataset construction specifications, the high-value dataset after classification is structurally encapsulated, assigned a global identifier and version number, and a high-value data description document is generated simultaneously. S523. Based on the global identifier, version number, and high-value data description document, generate a hash value and submit the hash value to the corresponding consortium chain node for on-chain storage through the power industrial control consortium chain. S524. Based on the confirmation receipt of the power industrial control alliance chain, determine whether the hash value on-chain notarization is successful. If so, store the encapsulated high-value dataset into the preset data asset library and establish an asset catalog and tag system to achieve full-link trusted traceability and full life cycle management of data assets. If not, trigger the preset retry mechanism until the hash value on-chain notarization is successful.

9. A method for power production monitoring data governance and high-value data asset generation based on an in-store computing time-series database according to claim 8, characterized in that, The construction of a power business knowledge graph, using the target business scenario as input, generates corresponding high-value dataset recommendation strategies using the power business knowledge graph, performs scenario-based adaptation verification based on the high-value dataset recommendation strategies, and establishes a dynamic iteration mechanism to optimize the data asset library, thereby realizing power production monitoring data governance and high-value data asset generation, includes: S531. Based on high-value datasets in the data asset library, define power business scenarios, digital applications, algorithm models, datasets, feature dimensions, and equipment types as entities, and based on preset power industry standards, business rules, and historical application cases, set semantic relationships between entities to construct a power business knowledge graph. S532. Using the target business scenario as input, and leveraging entity association reasoning and semantic similarity calculation of the power business knowledge graph, generate a corresponding high-value dataset recommendation strategy. S533. Based on the high-value dataset recommendation strategy, call the corresponding high-value dataset from the data asset library for scenario-based adaptation verification, and apply the verified high-value dataset to the target business scenario to obtain feedback on the effect of the business application. S534. Based on the feedback, construct a dynamic iteration mechanism and use the dynamic iteration mechanism to optimize the high-value dataset. Based on the optimization results, update the power business knowledge graph and the recommendation strategy for the high-value dataset to optimize the data asset library.

10. A system for power production monitoring data governance and high-value data asset generation based on an in-memory computing time-series database, used to implement the method for power production monitoring data governance and high-value data asset generation based on an in-memory computing time-series database as described in any one of claims 1-9, characterized in that, The system includes: The data support and acquisition module is used to access the raw time-series data of multi-source power production monitoring in the construction of an integrated storage and computing underlying support environment based on time-series database, and to simultaneously perform storage management and security control during the access process, and output time-series datasets. The anomaly detection and management module is used to generate a dynamic operating condition baseline by calling a preset lightweight digital twin of power equipment based on the time series dataset, and to perform anomaly detection and management on the time series dataset in combination with a dual-track detection mechanism, and output a standardized time series dataset. The data filtering and optimization module is used to match the artificial intelligence algorithms corresponding to the standardized time series datasets, perform feature mining based on the matching results, and use a two-dimensional filtering mechanism to filter and optimize the feature mining results to generate a standardized high-value feature set. The data asset generation module is used to construct an intelligent labeling and multi-dimensional quality rating system based on the preset power industry business specifications, and to use the intelligent labeling and multi-dimensional quality rating system to label and filter standardized high-value feature sets to obtain high-value datasets. The data asset management module is used to classify and structure high-value datasets and store the encapsulation results in a pre-set data asset library. Combined with a pre-deployed power industrial control alliance chain and power business knowledge graph, the data asset library is optimized to achieve power production monitoring data governance and high-value data asset generation.

Images