Multi-source heterogeneous data fusion whole plant comprehensive early warning management platform and method

Abstract

The present application relates to the technical field of early warning management platform, and specifically discloses a multi-source heterogeneous data fusion full-plant comprehensive early warning management platform and method, which comprises an image data acquisition layer, a data management layer, an early warning analysis layer, an application interaction layer and a security protection layer; the image data acquisition layer covers DCS system, NCS system, field inspection system, fire alarm system, video monitoring system and intelligent monitoring system data; the early warning analysis layer realizes alarm grading through three-level processes of weight calculation, severity evaluation and score correction; the application interaction layer is linked with the early warning analysis layer, and triggers a differentiated push strategy according to the alarm level; and the security protection layer meets the requirements of the third level and above of the national energy group network security level protection. The core technical scheme (such as data graph construction logic, early warning algorithm model and security protection mechanism) of the present application is not limited to a specific unit type, and can be flexibly adapted to different types of thermal power projects such as 300MW-1000MW capacity gas turbines, coal-fired units, etc.

Description

Technical Field

[0001] This invention relates to the field of early warning management platform technology, specifically to a plant-wide integrated early warning management platform and method that integrates multi-source heterogeneous data. Background Technology

[0002] At this critical stage of the thermal power industry's transformation towards "intelligent and unmanned" operation, industry leaders such as the State Energy Group have clearly proposed to achieve automated replacement of power generation processes through intelligent control and detection technologies, reaching the core goals of "precise control, autonomous optimization, and unmanned operation." Currently, existing units in power plants such as Beijing Gas-fired Power Plant Co., Ltd. mostly adopt a "two-to-one" combined cycle configuration, equipped with multiple control systems such as DCS (e.g., Emerson Ovation 3.5), TCS (e.g., Mitsubishi), and NCS, while also encompassing decentralized subsystems such as on-site inspection, fire alarm, and video monitoring. The production and safety data generated by these systems exhibit significant "multi-source heterogeneity" characteristics, specifically manifested in vastly different data formats (e.g., PI real-time data, JSON format inspection data), communication protocols (e.g., IEC104, ModbusTCP, API), and safety zones (Zone I control network, Zone II monitoring network, Zone III management network). Existing early warning management models face numerous technical challenges and are difficult to adapt to the needs of intelligent power plant construction.

[0003] Currently, power plant systems store data independently on dedicated platforms. For example, DCS data is only stored in local historical stations, and inspection data is scattered across maintenance personnel's mobile terminals, lacking a unified data governance and correlation mechanism. When a unit malfunctions, maintenance personnel need to query data across multiple terminals, including the DCS operator station, NCS monitoring interface, and inspection APP. This makes it impossible to achieve linked analysis of "DCS parameter anomalies - NCS electrical status - on-site video footage - historical inspection records." This not only makes it difficult to quickly locate the root cause of the fault but may also lead to delays in emergency alarm response due to data fragmentation, affecting the safety of unit operation.

[0004] Traditional early warning systems rely solely on fixed thresholds at single measurement points to trigger alarms, neglecting critical factors such as the importance of the data source and the unit's operating conditions. For example, when the unit is operating at low load, normal fluctuations in some auxiliary equipment parameters may trigger invalid alarms due to fixed threshold settings, creating an "alarm flood" that interferes with the judgment of maintenance personnel. Furthermore, abnormal signals from critical measurement points such as trip points and protection points may be missed or delayed in identification because they are not analyzed in conjunction with the status of related equipment, increasing the risk of equipment damage and unplanned unit shutdowns. This fails to meet the requirements of "accurate early warning" for smart power plants.

[0005] Power plant data needs to be transmitted across security zones. Existing solutions mostly use simple isolation devices, which have two core problems: First, the data transmission delay is too long, with some cross-zone data transmission taking 5-10 seconds, which cannot meet the real-time push requirements of alarms such as fire alarms and trip risk; Second, there is a lack of hierarchical encryption and fine-grained access control, which makes it easy for unauthorized access and data leakage to occur, and it does not meet the Level 3 requirements of the National Energy Group's network security level protection, resulting in insufficient security compliance.

[0006] The existing early warning information is scattered across different terminals, lacking a "one-screen overview" function. Maintenance personnel need to frequently switch interfaces to grasp the early warning status of the entire plant. Mobile push notifications only support simple SMS notifications and do not implement a hierarchical push strategy of "first-class alarms to the person in charge, second-class alarms to the maintenance team, and third / fourth-class alarms archived for review". It also lacks a closed-loop management function for alarm confirmation, processing and feedback, resulting in a lengthy alarm handling process that is difficult to adapt to the efficient maintenance needs of "unattended" scenarios.

[0007] In summary, the existing early warning management model cannot solve the core problems of "difficulty in integrating multi-source data, low early warning accuracy, imbalance between security and real-time performance, and poor operation and maintenance efficiency". A comprehensive management solution that can integrate multi-source heterogeneous data, realize intelligent hierarchical early warning, and balance security and real-time performance is needed to provide technical support for the construction of smart power plants. Summary of the Invention

[0008] To address the shortcomings of existing technologies, this invention provides a comprehensive plant-wide early warning management platform and method for multi-source heterogeneous data fusion, solving the problems mentioned in the background technology.

[0009] To achieve the above objectives, the present invention provides the following technical solution: a plant-wide integrated early warning management platform for multi-source heterogeneous data fusion, comprising:

[0010] The image data acquisition layer is used to collect multi-source heterogeneous production and safety data from the entire plant, and is compatible with the communication protocols and data formats of different systems.

[0011] The data governance layer performs cleaning, association, and structuring processing on multi-source heterogeneous production and safety data, constructs unit data graphs and knowledge graphs, and realizes unified management of data assets;

[0012] The early warning analysis layer, based on the unit data graph and knowledge graph, performs hierarchical calculations on the data through a multi-dimensional fusion algorithm to generate a comprehensive alarm score and hierarchical alarm information.

[0013] The application interaction layer visualizes the hierarchical alarm information, pushes alarm information to designated terminals, and provides alarm service flow management and historical data query interfaces.

[0014] The security protection layer performs full-process security control over platform data transmission, storage, and access to ensure system and data security.

[0015] The image data acquisition layer covers data from DCS system, NCS system, on-site inspection system, fire alarm system, video surveillance system and intelligent monitoring system; the early warning analysis layer realizes alarm classification through a three-level process of weight calculation, severity assessment and score correction; the application interaction layer links with the early warning analysis layer to trigger differentiated push strategies according to the alarm level; and the security protection layer meets the requirements of Level 3 or above of the National Energy Group's network security level protection.

[0016] Preferably, the image data acquisition layer includes a source acquisition unit, an interface adaptation unit, and a data transmission unit:

[0017] The source-splitting acquisition unit is configured with multiple types of acquisition modules, and the parameters and acquisition rules are as follows:

[0018] Auxiliary network data acquisition module: Uses a PI interface to acquire analog data, with an acquisition frequency of... and Measurement point scanning cycle Matching, satisfying ,in The value range is 100ms-1s to ensure data real-time performance. System synchronization;

[0019] Data acquisition module: Acquires switch inputs and alarm information via the IEC104 protocol interface, with data sampling intervals. Data frame transmission bit error rate ;

[0020] Fire alarm system data acquisition module: based on Protocol for collecting alarm information, communication baud rate The data verification method is The response time for a single communication is ≤300ms;

[0021] Intelligent monitoring / inspection system data acquisition module: through Interface for collecting diagnostic and inspection data; interface response time. The data format is

[0022] The interface adaptation unit performs format normalization processing on data from different protocols, and the output fields include data source, acquisition time, parameter name, parameter value, and data quality code; the data transmission unit implements a security zone through a unidirectional isolation gateway. Inter-channel data transmission, network gateway data throughput Transmission delay And supports Dual protocol stack.

[0023] Preferably, the data governance layer includes a data cleaning unit, a data graph construction unit, and a knowledge graph construction unit:

[0024] The data cleaning unit performs outlier removal and missing value completion, with the following specific rules:

[0025] Outlier removal: using The principle and the determination formula are as follows: ,in The original data values, The arithmetic mean of the dataset. This represents the standard deviation of the dataset. Data points that deviate from the mean by more than three times the standard deviation are considered outliers.

[0026] Missing value completion: Linear interpolation is used, and the completion formula is as follows: ,in The timestamps for collecting adjacent valid data. For the corresponding valid data values, The missing data is collected using the timestamps, and the missing values ​​are calculated based on the time ratio.

[0027] The data mapping construction unit covers over 5000 original measuring points and over 2500 alarm messages for the two-to-one generator units, automatically generating KKS codes conforming to GB / T50764-2012. The encoding rules are based on equipment type, system number, and measuring point number. It also constructs a four-level association relationship between the generator unit, system, equipment, and measuring points, achieving a high accuracy rate. ;

[0028] The knowledge graph construction unit, based on the concepts of FMEA and FTA, transforms operational experience into a triple structure and calculates the confidence level of knowledge graph inference. ,Require .

[0029] Preferably, the early warning analysis layer includes a weight calculation unit, which performs the assignment of data source weights and measurement point importance weights, as well as operating condition calibration. The specific process is as follows:

[0030] Basic weight assignment:

[0031] Data source base weights: Intelligent inventory monitoring system ubiquitous sensing system ;

[0032] Basic weighting of measurement point importance: jump points Protection point Generally speaking ;

[0033] Operating condition calibration: Introducing operating condition correction factors The calibration formula is ,in The rule for determining the value is: the unit is at full load, i.e., the load factor. hour Low load, i.e., load factor hour Other operating conditions The weights are dynamically adjusted based on the load status.

[0034] in, To calibrate the data source weights, The weighting of the importance of the measurement points after calibration, and the weighting calculation error. ,

[0035] in, The calculation error of the comprehensive weight of a single measuring point, The total weight value of a single measuring point is calculated in practice, derived from the weight of the calibrated data source. × Importance weight of measurement points after calibration get; This is the theoretical baseline value for the comprehensive weight of a single measurement point, derived from the basic weight of the data source. Basic weight of measurement point importance Theoretical value of working condition correction factor The calculated value is based on the platform's preset weighting rules;

[0036] The early warning analysis layer also includes a severity assessment unit, which calculates the severity M for different data sources, according to the following rules:

[0037] DCS / Ubiquitous Sensing System Data Source:

[0038] If the data is between the preset high and low reporting thresholds ;

[0039] If the data exceeds the high reporting threshold or the low reporting threshold ;

[0040] If the data exceeds the high-high reporting threshold or the low-low reporting threshold... ;

[0041] If the data exceeds the trip threshold ;

[0042] The intelligent inventory monitoring system uses a multi-parameter coupling algorithm for its data source, as shown in the formula below. ,in The difference between the parameter and the normal range. The value represents the impact of the associated equipment malfunction, ranging from 0 to 1. These are the weighting coefficients. ;

[0043] Single data source, single item score ,in For comprehensive weighting, This represents the severity value, and the score calculation precision. .

[0044] Preferably, the early warning analysis layer also includes a comprehensive score calculation unit. The comprehensive score calculation unit uses a dual-mode calculation and performs score correction. The specific process is as follows:

[0045] Calculation mode selection:

[0046] Mode 1: ,in for Data source score, Scoring is given to the data source for intelligent monitoring. To score generalized data sources, the importance of different data sources is distinguished by weighting coefficients.

[0047] Mode 2: The highest score from each data source is taken as the overall score;

[0048] Mode switching rule: Unit at full load, i.e., load factor When using mode one, start / stop or low load, i.e., load factor. Mode 2 will be used at that time;

[0049] Score Correction: Introduce an alarm frequency correction coefficient. The corrected formula is as follows: ,in To calculate a score for the pattern, when the alarm frequency is >10 times / minute ,otherwise ;

[0050] Alarm classification: based on the corrected comprehensive score Alarms are categorized into four types: This is a type of alarm. It is a Class II alarm. There are three types of alarms. There are four types of alarms. .

[0051] Preferably, the early warning analysis layer also includes a cold-end optimization early warning module, which performs optimal vacuum calculations and deviation early warnings. The specific process is as follows:

[0052] Optimal vacuum value calculation: ,in This represents the current vacuum value of the condenser. For circulating water flow rate, The inlet temperature of the circulating water. The outlet temperature of the circulating water. The specific heat capacity of water, For the density of circulating water, The heat exchange area of ​​the condenser;

[0053] Warning threshold setting: Vacuum deviation warning threshold ,in Design a vacuum for the condenser. For ambient atmospheric pressure; when At that time, generate cold-end optimization early warning information;

[0054] Energy consumption calculation: Energy consumption of cold end equipment ,in For the head of the circulating water pump, It is the acceleration due to gravity. For pump efficiency, energy consumption calculation error .

[0055] Preferably, the application interaction layer includes a visualization display unit and a mobile interaction unit:

[0056] The visualization unit supports a one-screen overview function, displaying a real-time overview of unit alarms, early warning information, and economic indicators. The calculation of the power supply coal consumption rate is as follows:

[0057] Basic coal consumption formula: ,in This refers to the coal consumption of the generating unit. The lower heating value of coal. For the generator unit's power generation;

[0058] Coal quality correction: Introducing coal quality correction coefficients Corrected coal consumption When the calorific value of coal is lower than the design value hour Below the design value hour Energy consumption indicators are dynamically corrected based on differences in coal quality.

[0059] Refresh rate: Refresh rate of the visual interface , The interface data delay time, measured in seconds (s), refers to the time interval between the data output from the early warning analysis layer and its display on the visualization interface. Requirements: , which is the core threshold used to measure the real-time performance of the interface;

[0060] The mobile interaction unit pushes alarm information via SMS / the group's ICE platform, with the following push rules:

[0061] Push delay: ;

[0062] Push success rate: ;

[0063] Interactive function: Support receiving, confirming, and feedback on the processing status of alarm sheets, and the response time of the mobile interface .

[0064] Preferably, the security protection layer includes a data encryption unit, an access control unit, and an anomaly monitoring unit:

[0065] Data encryption unit: Use the AES-256 algorithm to encrypt the transmitted and stored data, and the encryption key update period is days, and the decryption success rate of the encrypted data is ;

[0066] Access control unit: Allocate permissions based on the RBAC model, and refine the permission granularity to the data table level. The retention time of user operation logs is days, ;

[0067] Anomaly monitoring unit: Calculate the network load increment , where is the network load before platform access, and is the network load after platform access. It is required that ; When detecting abnormal access, trigger linkage blocking, and the blocking response time is .

[0068] Preferably, the data governance layer further includes a historical data management unit, and the historical data management unit performs data storage and backup. The specific rules are as follows:

[0069] Storage period: Class I alarm data , Class II / III / IV alarm data and regular operation data ;

[0070] Backup strategy: Adopt dual backup of local and remote locations. The local backup period is , and the remote backup period is , and the backup integrity is ;

[0071] Recovery ability: Support directional recovery according to time intervals, alarm types, and device numbers. The single-batch data recovery time is , and the accuracy rate of the recovered data is .

[0072] The whole-plant comprehensive early warning management method for multi-source heterogeneous data fusion includes the following steps:

[0073] S1. Multi-source heterogeneous data acquisition: Through the multi-source acquisition unit, different communication protocols and data formats are adapted to collect production and safety data from the entire plant's DCS system, NCS system, on-site inspection system, fire alarm system, video surveillance system and intelligent monitoring system. After being converted into a unified format by the interface adaptation unit, the data is transmitted to the data governance layer through a secure transmission channel.

[0074] S2. Data Governance and Knowledge Graph Construction: Cleaning, association, and structuring of the collected multi-source heterogeneous data to construct unit data graphs and knowledge graphs;

[0075] S3. Multi-dimensional early warning analysis and classification: Based on the unit data graph and knowledge graph, a comprehensive score is generated through weight calculation, severity assessment and score correction. Alarm information is divided into four categories according to the comprehensive score.

[0076] S4. Warning Information Display and Interaction: Visualize the graded alarm information, push information to designated terminals according to the alarm level, and support alarm business flow management and historical data query;

[0077] S5. Full-process security control: Implement security protection for the data collection, transmission, storage and access process to ensure system and data security.

[0078] Beneficial Effects: This invention, through a five-layer architecture design comprising "image data acquisition layer, data governance layer, early warning analysis layer, application interaction layer, and security protection layer," combined with a three-level early warning process of "weight calculation, severity assessment, and score correction," specifically addresses the pain points of existing technologies and possesses the following significant technical advantages and practical value:

[0079] 1. This invention utilizes a multi-source acquisition unit in the image data acquisition layer, adapting to various communication protocols such as PI, IEC104, ModbusTCP, and API, covering data from all systems including DCS, NCS, on-site inspection, fire alarm, and video surveillance. Simultaneously, an interface adaptation unit normalizes data of different formats into a unified structure of "data source, acquisition time, parameter name, parameter value, and data quality code," ensuring data format consistency. The data governance layer further constructs a four-level interconnected data map of "unit, system, equipment, and measurement point," automatically generating KKS codes conforming to national standards. Based on FMEA (Failure Mode and Effects Analysis) and FTA (Fault Tree Analysis) concepts, a knowledge graph is built, transforming operational experience into reasonable triplet logic, achieving a data management model of "physically dispersed, logically unified." When a unit malfunctions, the system can automatically correlate multi-source data, providing panoramic information support for fault diagnosis and significantly improving data utilization efficiency.

[0080] 2. The early warning analysis layer optimizes the early warning logic through multi-dimensional algorithms: First, it assigns basic weights based on the importance of the data source (such as DCS / NCS systems, intelligent monitoring systems, and general sensing systems) and the criticality of the measurement points (such as trip points, protection points, and general points). Then, it introduces operating condition correction coefficients (different coefficients for full load, low load, and start-up / shutdown conditions) to dynamically adjust the weights. Subsequently, it calculates the severity based on the characteristics of different data sources. For example, DCS data is judged based on threshold ranges, and intelligent monitoring data is evaluated for severity through multi-parameter coupling algorithms. Finally, it achieves alarm classification through dual-mode comprehensive score calculation (main cause decomposition method for full load and maximum value method for low load) and alarm frequency correction. This logic can effectively suppress invalid alarms, avoid "alarm floods," and ensure that no critical alarms are missed, significantly improving the accuracy of early warning and providing reliable decision-making basis for operation and maintenance personnel.

[0081] 3. The security protection layer employs multiple security mechanisms: Data transmission and storage are encrypted using the AES-256 algorithm, with encryption keys updated periodically; fine-grained permissions at the data table level are allocated based on the RBAC (Role-Based Access Control) model, and complete operation logs are retained; network load increments are monitored in real time, triggering millisecond-level linkage blocking for abnormal behaviors such as unauthorized IP logins and unauthorized operations, meeting the Level 3 requirements of the National Energy Group's network security protection system; simultaneously, cross-security zone data transmission is achieved through a one-way isolation gateway, ensuring that gateway throughput and transmission latency meet real-time requirements. DCS data acquisition latency and mobile alarm push latency are both controlled at industry-leading levels, achieving the dual goals of "secure transmission" and "real-time early warning."

[0082] 4. The application interaction layer supports a "one-screen overview" function, displaying a real-time overview of all plant alarms, warning details, and economic indicators such as power supply coal consumption rate. The interface refresh delay is controlled within 1 second, allowing maintenance personnel to grasp the overall status without cross-terminal queries. The mobile terminal implements hierarchical push and closed-loop management, prioritizing the push of Class I alarms to the maintenance manager and Class II alarms to the maintenance team. It also supports alarm order reception, confirmation, processing feedback, and other operations, significantly shortening the alarm handling process. In addition, the system supports targeted recovery of historical data by "time interval-alarm type-device number", providing convenience for fault tracing and maintenance review, significantly improving maintenance efficiency, and adapting to the needs of "unattended" scenarios.

[0083] In summary, the core technical solutions of this invention (such as data map construction logic, early warning algorithm model, and security protection mechanism) are not limited to specific unit types. They can be flexibly adapted to different types of thermal power projects, such as gas turbines and coal-fired units with capacities ranging from 300MW to 1000MW. This provides a replicable and scalable technical solution for the State Energy Group's "Gas-fired Smart Power Plant Demonstration Project," helping the thermal power industry as a whole to achieve the goal of intelligent transformation towards "unmanned operation and autonomous optimization." Attached Figure Description

[0084] Figure 1 This is a schematic diagram of the plant-wide integrated early warning management platform for multi-source heterogeneous data fusion as described in this invention.

[0085] Figure 2 This is a flowchart illustrating the plant-wide integrated early warning management method for multi-source heterogeneous data fusion described in this invention. Detailed Implementation

[0086] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0087] like Figure 1 As shown, this invention provides a technical solution: a plant-wide integrated early warning management platform for multi-source heterogeneous data fusion, comprising: an image data acquisition layer, a data governance layer, an early warning analysis layer, an application interaction layer, and a security protection layer; the image data acquisition layer is used to collect multi-source heterogeneous production and safety data from the entire plant, adapting to the communication protocols and data formats of different systems; the data governance layer performs cleaning, association, and structuring processing on the multi-source heterogeneous production and safety data, constructing unit data graphs and knowledge graphs to achieve unified management of data assets; the early warning analysis layer, based on the unit data graphs and knowledge graphs, performs hierarchical calculations on the data through multi-dimensional fusion algorithms to generate comprehensive alarm scores and hierarchical alarm information; the application interaction layer visualizes the hierarchical alarm information, pushes alarm information to designated terminals, and provides alarm business flow management and historical data query interfaces; the security protection layer performs full-process security control on platform data transmission, storage, and access to ensure system and data security;

[0088] The image data acquisition layer covers data from DCS system, NCS system, on-site inspection system, fire alarm system, video surveillance system and intelligent monitoring system; the early warning analysis layer realizes alarm classification through a three-level process of weight calculation, severity assessment and score correction; the application interaction layer links with the early warning analysis layer to trigger differentiated push strategies according to the alarm level; and the security protection layer meets the requirements of Level 3 or above of the National Energy Group's network security level protection.

[0089] More specifically, the image data acquisition layer includes a source acquisition unit, an interface adaptation unit, and a data transmission unit:

[0090] The source-splitting acquisition unit is configured with multiple types of acquisition modules, and the parameters and acquisition rules are as follows:

[0091] The DCS / TCS / auxiliary network data acquisition module uses the Advantech PCI-1716 data acquisition card (supporting native PI protocol access) and is paired with a Siemens S7-300 series signal isolation module to avoid analog signal interference; the acquisition frequency... With DCS measurement point scanning cycle Forced matching, execute formula ;in Dynamically set key parameters for gas turbines, steam turbines, etc., based on the measurement point type of the DCS system (Emerson Ovation 3.5). Parameters of auxiliary equipment for waste heat boiler Public system parameters This ensures that the acquired data is completely synchronized with the real-time DCS screen without any delay or deviation; the analog signal acquisition range covers 0-10V voltage signals and 4-20mA current signals, and is compatible with the output parameters of the DCS system, such as temperature (-200~1600℃), pressure (0~40MPa), and flow rate (0~1000m³ / h).

[0092] The NCS data acquisition module uses a Huawei USG6000E firewall (with a built-in IEC104 protocol parsing module) and is directly connected to the NCS system (NARI PCS-9700) switch via an RJ45 interface; data sampling interval The timeout is fixed at 500ms. The NCS database is periodically polled via the firewall to obtain switch quantities (such as circuit breaker open / close status) and electrical alarm information (such as overcurrent and overvoltage alarms). Data frame transmission uses a "CRC32 + retransmission mechanism" to ensure a low bit error rate. When a data frame loss or error is detected, three automatic retransmissions are triggered, with a retransmission interval of 100ms. If the retransmission fails, a hardware fault alarm is generated and pushed to the operation and maintenance terminal.

[0093] The fire alarm system data acquisition module uses a Honeywell XLS9200 fire alarm controller (supporting Modbus TCP protocol) and a TP-Link industrial-grade PoE switch to achieve data transmission; communication baud rate. Fixed at 115200bps, 8 data bits, 1 stop bit, and CRC16 checksum (polynomial is...). The single communication response time is controlled to ≤300ms through a dual mechanism of "controller active reporting + platform periodic query"; the collected data includes fire alarm area code (such as column A of the #1 unit turbine room), alarm type (smoke / heat / manual alarm), and alarm timestamp to ensure that fire alarm information can be accurately located.

[0094] The data acquisition module of the intelligent monitoring / inspection system uses a Dell PowerEdge R750 server (deploying RESTful API interface services) and communicates with the intelligent monitoring system (Guodian Zhishen EDPF-NT+) and the inspection APP (developed based on Android 12) via HTTPS protocol; interface response time... By controlling the transmission time to ≤1 second through "server resource reservation + data compression": the server reserves 4 CPU cores and 8GB memory dedicated resources to run the API service, and the inspection data is transmitted after GZIP compression (compression rate ≥60%); the data format is forced to be JSON, and the fields include "device ID (KKS code), diagnosis result (normal / abnormal), inspection personnel, inspection time, and data quality code (0=valid / 1=invalid)" to ensure that the data is structured and can be directly used for subsequent management.

[0095] The interface adaptation unit performs format normalization processing on data from different protocols, and the output fields include data source, acquisition time, parameter name, parameter value, and data quality code; the data transmission unit implements a security zone through a unidirectional isolation gateway. Inter-channel data transmission, network gateway data throughput Transmission delay And supports Dual protocol stack.

[0096] More specifically, the data governance layer includes a data cleaning unit, a data graph construction unit, and a knowledge graph construction unit:

[0097] The data cleaning unit performs outlier removal and missing value completion, with the following specific rules:

[0098] The data cleaning unit is deployed on a Huawei FusionServerPro2288HV5 server (CentOS 8 system, 16-core CPU + 64GB memory), and the algorithm is developed based on Python Pandas to execute two main processes:

[0099] Outlier removal (Principle): Statistical data is collected in 1-hour segments, and the arithmetic mean of each segment is calculated. with standard deviation ,in accordance with Identify and remove outliers (such as the main steam temperature of waste heat boiler #1). Data with temperatures >545.5℃ or <530.5℃ will be removed; if the outlier removal rate is ≤0.5%, a sensor fault alarm will be triggered if the outlier exceeds the threshold.

[0100] Missing value completion: Locating missing intervals Take adjacent valid data Substitute Complete data (e.g., complete missing data for #2 turbine bearing vibration); only applicable to continuous data, missing duration ≤30 minutes, completion accuracy deviation from the true value ≤2%, and completed data marked with "interpolation".

[0101] The data mapping construction unit covers over 5000 original measuring points and over 2500 alarm messages for the two-to-one generator units, automatically generating KKS codes conforming to GB / T50764-2012. The encoding rules are based on equipment type, system number, and measuring point number. It also constructs a four-level association relationship between the generator unit, system, equipment, and measuring points, achieving a high accuracy rate. ;

[0102] The data graph construction unit, developed based on the Neo4j graph database, integrates 5000+ original measurement points and 2500+ alarm information.

[0103] KKS code automatic generation: It adopts a three-segment code of "equipment type-system number-measuring point number" (e.g., PU1B01P01 represents the outlet pressure of the feedwater pump in column B of unit #1), and automatically generates the code by reading the "data source" field, with an accuracy of 100%.

[0104] Four-level relationship construction: Establish a tree-like relationship of "unit, system, equipment and measuring point" (e.g., #1 unit → #1 waste heat boiler system → feedwater pump → outlet pressure measuring point);

[0105] Association accuracy It was verified through 1,000 sets of random checks.

[0106] The knowledge graph construction unit, based on the concepts of FMEA (Failure Mode and Effects Analysis) and FTA (Fault Tree Analysis), transforms operational experience into a triple structure (entity-relationship-attribute), such as (steam turbine bearing-association-bearing temperature measurement point), and calculates the confidence level of knowledge graph inference. ,Require .

[0107] The knowledge graph construction unit is developed based on the Apache Jena framework, and its core is the transformation of triple structure and optimization of reasoning.

[0108] Triplet structure design:

[0109] Physical components: equipment (such as turbine bearings), measuring points (such as bearing temperature), fault types, and handling measures;

[0110] Relationship: Associated (equipment - measuring point), resulting in (fault - alarm), requiring (fault - solution);

[0111] Example: The maintenance experience "steam turbine bearing temperature exceeds 95℃, causing wear failure, requiring shutdown for maintenance" is converted into ternary sets such as (steam turbine bearing - associated - bearing temperature measurement point) and (temperature exceeds 95℃ - causing - wear failure), with an initial ternary set ≥ 5000 sets;

[0112] Inference confidence calculation, inference confidence The system is validated through 1,000 pre-set test cases; the rule base is optimized monthly (e.g., adding new working condition association rules) to ensure that the confidence level continues to meet the standard.

[0113] More specifically, the early warning analysis layer includes a weight calculation unit. This unit assigns weights to the data source and the importance weights of the measurement points, and performs operational condition calibration. The specific process is as follows:

[0114] Basic weight assignment:

[0115] Data source base weights: Intelligent inventory monitoring system ubiquitous sensing system ;

[0116] Basic weighting of measurement point importance: jump points Protection point Generally speaking ;

[0117] Operating condition calibration: Introducing operating condition correction factors The calibration formula is ,in The rule for determining the value is: the unit is at full load, i.e., the load factor. hour Low load, i.e., load factor hour Other operating conditions The weights are dynamically adjusted based on the load status; among them, To calibrate the data source weights, The weighting of the importance of the measurement points after calibration, and the weighting calculation error. ,in, The calculation error of the comprehensive weight of a single measuring point, The total weight value of a single measuring point is calculated in practice, derived from the weight of the calibrated data source. × Importance weight of measurement points after calibration get; This is the theoretical baseline value for the comprehensive weight of a single measurement point, derived from the basic weight of the data source. Basic weight of measurement point importance Theoretical value of working condition correction factor The calculated value is based on the platform's preset weighting rules.

[0118] The weight calculation unit is deployed on an Inspur NF5280M6 server (running Ubuntu 20.04 system, configured with 24-core CPU and 128GB memory). A real-time calculation module developed in C++ is used, which executes a four-step process: "basic weight assignment, operating condition calibration, comprehensive weight calculation, and error verification." The specific implementation is as follows:

[0119] Data source base weight ( The system assigns values ​​based on the impact of the data source on unit safety, pre-setting three basic weights and writing them directly into the system configuration file without manual intervention.

[0120] Data source type <![CDATA[Base weight K0]]> Coverage System Example DCS / NCS System 5 Gas turbine TCS, waste heat boiler DCS, electrical NCS Intelligent inventory monitoring system 4 Intelligent diagnostic system and precision inspection system for generator units ubiquitous sensing system 1 On-site video surveillance and fire alarm system

[0121] Basic weights for the importance of measurement points ( Assigning values: Based on the criticality of the measuring points to the unit's operation, three basic weights are preset and stored in conjunction with the measuring point's KKS code.

[0122] Measurement point type <![CDATA[Base weight M0]]> Example of covered measurement points Jump point 5.5 Steam turbine overspeed protection measuring point, gas turbine flameout measuring point Protection point 3.5 Boiler water level protection measuring point, generator overcurrent measuring point General 1 Auxiliary machine lubricating oil temperature measuring point, utility system pressure measuring point

[0123] Operating condition calibration, with weights dynamically adjusted based on load rate:

[0124] The operating condition determination logic reads the unit load signal (such as the total power P of Unit #1) from the DCS system in real time. ,in accordance with Determine the working condition correction factor :

[0125] Full load: ;

[0126] Low load: ;

[0127] Other operating conditions: .

[0128] The calibration formula is executed; for each measuring point, according to the formula... The following is an example of calculating the calibrated weights:

[0129] Scenario: DCS system tripping point ( ) at full load ( )hour;

[0130] calculate: .

[0131] Comprehensive weight calculation and error control; comprehensive weight calculation:

[0132] Comprehensive weight of individual measuring points Round to two decimal places, as shown in the example below:

[0133] Scenario: Protection point of intelligent inventory monitoring system Under low load hour;

[0134] calculate: .

[0135] Error verification, weight calculation error ,in ( ); The system performs error verification by randomly selecting 100 measurement points daily. When the error exceeds the threshold, it triggers a fault alarm in the computing module to investigate hardware computing power or data transmission issues.

[0136] The early warning analysis layer also includes a severity assessment unit, which calculates the severity M for different data sources, according to the following rules:

[0137] DCS / Ubiquitous Sensing System Data Source: If the data is between the preset high and low reporting thresholds. If the data exceeds the over-reporting threshold or under-reporting threshold, If the data exceeds the high-high reporting threshold or the low-low reporting threshold, If the data exceeds the trip threshold, ;

[0138] The intelligent inventory monitoring system uses a multi-parameter coupling algorithm for its data source, as shown in the formula below. ,in The difference between the parameter and the normal range (unit: ), The value represents the impact of the associated equipment malfunction, ranging from 0 to 1. These are the weighting coefficients. (Normal operating conditions) , Start-up and shutdown status , );

[0139] Single data source, single item score ,in For comprehensive weighting, This represents the severity value, and the score calculation precision. (Keep one decimal place).

[0140] The severity assessment unit and the weight calculation unit work together to calculate severity using differentiated rules based on the data source type. , and then combine Receive individual score The specific implementation is as follows:

[0141] DCS / Pervasive Sensing System data source, assigned values ​​in tiers according to threshold ranges, with preset threshold rules:

[0142] For each DCS / general sensing system measuring point, four threshold levels are preset (high / low alarm, high-high / low-low alarm, trip threshold), which are bound to the measuring point's KKS code, as shown in the example below (#1 boiler main steam pressure measuring point):

[0143] Threshold type Numerical range Severity M Normal range 10-12MPa 0 High / Low Reporting Threshold >12MPa or <10MPa 6 High / Low Reporting Threshold >13MPa or <9MPa 8 Trip threshold >14MPa or <8MPa 10

[0144] Real-time evaluation logic: The current value of the measurement point is compared with the preset threshold in real time, and an automatic matching is performed. Value, example: main steam pressure .

[0145] Intelligent monitoring system data source, multi-parameter coupled algorithm evaluation, algorithm parameter definition:

[0146] The difference between the current value and the normal range of the measuring point (e.g., the normal range of bearing temperature is 40-80℃, the current value is...). ); The degree of influence of associated equipment anomalies (derived from knowledge graph inference, such as the degree of influence of bearing temperature overheating on "increased turbine vibration") ); weighting coefficient This is under normal operating conditions (stable unit operation). Start-up and shutdown conditions (unit start-up / shutdown phase) (Prioritize the impact on related equipment).

[0147] Algorithm execution example, scenario: Intelligent monitoring system monitors the temperature of #2 turbine bearing (normal operating conditions). ), ;calculate .

[0148] Individual score calculation, quantifying the risk of abnormal measurement points, calculation logic: individual score Retain one decimal place (score calculation precision ≤ 0.1), example scenario: DCS system tripping point ( Over-jump threshold (M=10); Calculate . As the basis for subsequent comprehensive score calculation, it is directly input into the comprehensive score calculation unit for alarm classification determination.

[0149] More specifically, the early warning analysis layer also includes a comprehensive score calculation unit. The comprehensive score calculation unit uses a dual-mode calculation and performs score correction. The specific process is as follows:

[0150] Calculation mode selection:

[0151] Mode 1: ,in for Data source score, Scoring is given to the data source for intelligent monitoring. To score generalized data sources, the importance of different data sources is distinguished by weighting coefficients.

[0152] Mode 2: The highest score from each data source is taken as the overall score;

[0153] Mode switching rule: Unit at full load, i.e., load factor When using mode one, start / stop or low load, i.e., load factor. Mode 2 will be used at that time.

[0154] The comprehensive score calculation unit is the core module of the early warning analysis layer for achieving "differentiated alarm grading." Deployed on an Inspur NF5280M6 server (sharing hardware resources with the weight calculation unit and equipped with a real-time computing engine), it quantifies the comprehensive risk of multi-source data anomalies through a three-step process: "dual-mode comprehensive score calculation, score correction, and alarm grading," providing a priority basis for operation and maintenance response. The following section, using a "two-to-one" combined cycle unit (rated load 475MW) as an example, explains the implementation rules, parameter configurations, and verification standards for each process:

[0155] Calculation mode selection, adapting to dual modes based on unit load rate:

[0156] The mode-triggered logic automatically switches based on load factor, and reads the actual unit load output from the DCS system in real time. Calculate the load factor ,according to Trigger the corresponding calculation mode:

[0157] Mode 1: (Right now Triggered under full load conditions, the core is to distinguish the importance of DCS, intelligent monitoring panel and general sensing data source by weighting coefficients (1, 0.5, 0.25), highlighting the leading role of the core system (DCS) in alarm.

[0158] Mode 2: (Right now Triggered during start-up / shutdown / low-load conditions, the core is to take the maximum value of the individual scores of the three types of data sources to avoid underestimation of scores caused by parameter fluctuations under low load, and to ensure that no critical anomalies are missed.

[0159] A dual-mode calculation example, taking the scenario "#1 Steam Turbine Bearing Temperature Anomaly" as an example, shows the individual scores of three types of data sources ( DCS score, For intelligent monitoring scores, (For general perception score)

[0160] Scenario 1, Full Load (Trigger Mode 1): If ,but ;

[0161] Scenario 2, Low Load (Trigger Mode 2): If ,but .

[0162] Score Correction: Introduce an alarm frequency correction coefficient. The corrected formula is as follows: ,in For pattern scoring, an alarm frequency > 10 times / minute (e.g., frequent alarms caused by fluctuations in auxiliary machine parameters during unit start-up and shutdown) is considered a valid alarm. (Lower the overall score to suppress invalid alarms), otherwise When the alarm frequency (Normal operating conditions): To ensure the accuracy of the alarm, the score is kept constant.

[0163] Alarm classification: based on the corrected comprehensive score Alarms are categorized into four types: This is a type of alarm. It is a Class II alarm. There are three types of alarms. There are four types of alarms. .

[0164] Based on the revised composite score Alarms are categorized into four types, each with specific operational and maintenance response requirements, and are directly linked to the push strategy of the application interaction layer:

[0165] Alarm Level Score range Corresponding scenario examples Operation and maintenance response requirements Type 1 alarm DCS tripping point exceeds threshold, fire alarm Within 5 seconds, the operations and maintenance manager is contacted, an emergency report is broadcast, and the response process is initiated. Class II alarm Abnormal intelligent monitoring protection point, NCS parameter exceeding limits Push to the operations and maintenance team; confirmation and processing within 10 minutes. Three types of alarms Auxiliary machine general point parameter fluctuation Storage archives are reviewed periodically by operations and maintenance personnel (once daily). Four types of alarms Normal equipment start-up and shutdown, valve opening and closing events Automatically archived to the historical database, requiring no manual response.

[0166] For misclassified records (≤2%), analyze the reasons and optimize: if the misjudgment is caused by the threshold setting deviation, update the threshold configuration of DCS / intelligent monitoring panel synchronously; if the misjudgment is caused by the frequency statistics error, optimize the timing logic of the alarm queue.

[0167] More specifically, the early warning analysis layer also includes a cold-end optimization early warning module. This module performs optimal vacuum calculations and deviation warnings, with the specific process as follows:

[0168] Optimal vacuum value calculation: ,in The current vacuum value of the condenser (unit: ), Circulating water flow rate (unit: ), The circulating water inlet temperature (unit: °C). The circulating water outlet temperature (unit: °C). The specific heat capacity of water (value) ), The density of circulating water (value) ), Condenser heat exchange area (unit: );

[0169] Warning threshold setting: Vacuum deviation warning threshold ,in Vacuum design for condenser (unit: ), Atmospheric pressure (unit: );when At that time, generate cold-end optimization early warning information;

[0170] Energy consumption calculation: Energy consumption of cold end equipment ,in The head of the circulating water pump (unit: m). The acceleration due to gravity (value) ), For water pump efficiency (requirement) Energy consumption calculation error .

[0171] The cold-end optimization and early warning module is deployed on the Inspur NF5280M6 server. It provides precise monitoring of the unit's cold-end system (condenser, circulating water pump) through "optimal vacuum calculation, deviation early warning, and energy consumption calculation." The implementation details are explained below using a two-to-one turbine unit (condenser model N-35000):

[0172] Optimal vacuum calculation, parameter acquisition: real-time acquisition of DCS data ( Condenser vacuum, Circulating water flow rate (Circulating water inlet and outlet temperature), sampling frequency 1 time / second; preset fixed value ;

[0173] Formula execution: Press Calculate and round to two decimal places;

[0174] Example: ,have to ;

[0175] Deviation warning judgment, threshold calculation: preset (Design vacuum), real-time sampling (Ambient atmospheric pressure), according to Calculate the threshold (take the absolute value);

[0176] Example: ,have to ;

[0177] Warning triggered: If If the deviation value is not specified, an early warning will be generated (including the deviation value, suggested value, and possible cause); otherwise, the condition will be judged as normal.

[0178] Cold end energy consumption calculation, real-time sampling (Pump head, 1 cycle / second), preset (Default 88%); Press Calculate energy consumption (unit: kW), rounded to the nearest integer;

[0179] Example: ,have to Monthly calibration and measurement are performed to ensure an error of ≤2%.

[0180] More specifically, the application interaction layer includes visual display units and mobile interaction units:

[0181] The visualization unit supports a one-screen overview function, displaying a real-time overview of unit alarms, early warning information, and economic indicators. The calculation of the power supply coal consumption rate is as follows:

[0182] Basic coal consumption formula: ,in This refers to the unit's coal consumption (unit: t). The lower heating value of coal (unit: ), Power generation of the unit (unit: );

[0183] Coal quality correction: Introducing coal quality correction coefficients Corrected coal consumption When the calorific value of coal is lower than the design value hour Below the design value hour Energy consumption indicators are dynamically corrected based on differences in coal quality.

[0184] Refresh rate: Refresh rate of the visual interface , The interface data delay time, measured in seconds (s), refers to the time interval between the data output from the early warning analysis layer and its display on the visualization interface. Requirements: , which is the core threshold used to measure the real-time performance of the interface;

[0185] The mobile interaction unit pushes alarm information via SMS / the group's ICE platform, with the following push rules:

[0186] Push delay: ;

[0187] Push success rate: ;

[0188] Interactive features: Supports alarm receipt, confirmation, and processing status feedback; mobile interface response time. .

[0189] The application interaction layer includes a visualization display unit and a mobile interaction unit, focusing on efficient transmission of early warning information and convenient operation and maintenance management. The implementation is as follows, tailored to the "two-to-one" unit scenario:

[0190] Visualization unit, deployment and layout: 55-inch 4K industrial screen (central control room), connected to Inspur NF5280M6 server; the interface is divided into three areas: "Alarm Overview (red / orange / yellow / blue in four categories), Alarm Details, and Economic Indicators";

[0191] Calculation of coal consumption for power supply: based on Calculate the base value when the calorific value of the coal is 5% / 10% lower. After correction ;

[0192] Refresh control: (Refresh rate ≥ 1Hz), incremental updates ensure no lag.

[0193] Mobile interaction unit with dual-channel push: Class I / II alarms use "SMS + Group ICE Platform", while Class III / IV alarms only use ICE; (Category 1 ≤ 3s), if it fails, retransmit 3 times within 10 seconds. ;

[0194] The interactive features support receiving, confirming, and processing feedback for alarm requests. (Local caching speedup), a type of alarm is highlighted in red and vibrates.

[0195] More specifically, the security protection layer includes a data encryption unit, an access control unit, and an anomaly detection unit:

[0196] Data encryption unit: Employs AES-256 algorithm to encrypt transmitted and stored data; encryption key update cycle. Heavens, the success rate of decrypting encrypted data ;

[0197] Access control unit: Permissions are assigned based on the RBAC (Role-Based Access Control) model, with permission granularity refined to the data table level, and user operation logs are retained for a specified period. sky, ;

[0198] Anomaly monitoring unit: Calculates network load increments ,in To reduce network load before platform access, To reduce network load after platform access, requirements are needed. ; When abnormal access is detected (such as illegal IP login, operation beyond permissions), trigger associated blocking, and the blocking response time .

[0199] More specifically, the data governance layer also includes a historical data management unit, which performs data storage and backup. The specific rules are as follows:

[0200] Storage period: For type I alarm data (2 years), for type II / III / IV alarm data and regular operation data (1 year);

[0201] Backup strategy: Adopt dual backup of local and remote locations. The local backup period , the remote backup period , and the backup integrity ;

[0202] Recovery ability: Support targeted recovery according to time intervals, alarm types, and device numbers. The single-batch data recovery time , and the accuracy rate of the recovered data .

[0203] The historical data management unit is the core module of the data governance layer to achieve "long-term data archiving, secure backup, and fast traceability". It is deployed on the Huawei OceanStor Dorado 8000 all-flash storage array (configured with 16 4TB SSD hard disks, RAID5 redundancy). Through the three-level mechanisms of "hierarchical storage, dual backup, and targeted recovery", it ensures the integrity, security, and traceability of the whole plant's production and alarm data. The following combines the actual application scenario of a "two-dragging-one" combined cycle unit to illustrate the implementation parameters, execution rules, and performance verification standards of each mechanism:

[0204] Hierarchical storage, determining the cycle according to importance:

[0205] Type I alarm data (trip / fire alarm): , stored in an independent partition, and deletion requires manual approval;

[0206] Type II / III / IV alarm and regular operation data: [[ID=,]]44]] when expired, automatically archived to the remote tape library, and 90 days of data is retained locally;

[0207] Cycle control: Push a reminder 7 days before expiration to ensure compliant storage.

[0208] Dual backup strategy, local and remote redundancy:

[0209] Local backup: Independent RAID groups in the same array, (incremental daily, full backup on weekends);

[0210] Off-site backup: Disaster recovery center 30 km away, 10Gbps encrypted link. (Full backup);

[0211] Integrity Guarantee: We sample 1,000 data entries daily for verification; if they fail to meet the standards, we redo the data.

[0212] Targeted recovery, accurate and rapid tracing:

[0213] Recovery conditions: Supports querying by combination of "time interval, alarm type and device KKS code", and can be triggered manually or automatically.

[0214] Performance control: When a single batch recovery is ≤10GB, After recovery, compare the MD5 checksum to ensure 100% accuracy.

[0215] like Figure 2 As shown, the plant-wide integrated early warning management method based on multi-source heterogeneous data fusion includes the following steps:

[0216] S1. Multi-source heterogeneous data acquisition: Through the multi-source acquisition unit, different communication protocols and data formats are adapted to collect production and safety data from the entire plant's DCS system, NCS system, on-site inspection system, fire alarm system, video surveillance system and intelligent monitoring system. After being converted into a unified format by the interface adaptation unit, the data is transmitted to the data governance layer through a secure transmission channel.

[0217] S2. Data Governance and Knowledge Graph Construction: Cleaning, association, and structuring of the collected multi-source heterogeneous data to construct unit data graphs and knowledge graphs;

[0218] S3. Multi-dimensional early warning analysis and classification: Based on the unit data graph and knowledge graph, a comprehensive score is generated through weight calculation, severity assessment and score correction. Alarm information is divided into four categories according to the comprehensive score.

[0219] S4. Warning Information Display and Interaction: Visualize the graded alarm information, push information to designated terminals according to the alarm level, and support alarm business flow management and historical data query;

[0220] S5. Full-process security control: Implement security protection for the data collection, transmission, storage and access process to ensure system and data security.

[0221] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A plant-wide integrated early warning management platform for multi-source heterogeneous data fusion, characterized in that: include: The image data acquisition layer is used to collect multi-source heterogeneous production and safety data from the entire plant, and is compatible with the communication protocols and data formats of different systems. The data governance layer performs cleaning, association, and structuring processing on multi-source heterogeneous production and safety data, constructs unit data graphs and knowledge graphs, and realizes unified management of data assets; The early warning analysis layer, based on the unit data graph and knowledge graph, performs hierarchical calculations on the data through a multi-dimensional fusion algorithm to generate a comprehensive alarm score and hierarchical alarm information. The application interaction layer visualizes the hierarchical alarm information, pushes alarm information to designated terminals, and provides alarm service flow management and historical data query interfaces. The security protection layer performs full-process security control over platform data transmission, storage, and access to ensure system and data security. The image data acquisition layer covers data from DCS system, NCS system, on-site inspection system, fire alarm system, video surveillance system and intelligent monitoring system; the early warning analysis layer realizes alarm classification through a three-level process of weight calculation, severity assessment and score correction; the application interaction layer links with the early warning analysis layer to trigger differentiated push strategies according to the alarm level; and the security protection layer meets the requirements of Level 3 or above of the National Energy Group's network security level protection. The data governance layer includes a data cleaning unit, a data graph construction unit, and a knowledge graph construction unit. The data cleaning unit performs outlier removal and missing value completion, with the following specific rules: Outlier removal: using The principle and the determination formula are as follows: ,in The original data values, The arithmetic mean of the dataset. This represents the standard deviation of the dataset. Data points that deviate from the mean by more than three times the standard deviation are considered outliers. Missing value completion: Linear interpolation is used, and the completion formula is as follows: ,in The timestamps for collecting adjacent valid data. For the corresponding valid data values, The missing data is collected using the timestamps, and the missing values ​​are calculated based on the time ratio. The data mapping construction unit covers over 5000 original measuring points and over 2500 alarm messages for the two-to-one generator units, automatically generating KKS codes conforming to GB / T50764-2012. The encoding rules are based on equipment type, system number, and measuring point number. It also constructs a four-level association relationship between the generator unit, system, equipment, and measuring points, achieving a high accuracy rate. ; The knowledge graph construction unit, based on the concepts of FMEA and FTA, transforms operational experience into a triple structure and calculates the confidence level of knowledge graph inference. ,Require ; The early warning analysis layer also includes a comprehensive score calculation unit. The comprehensive score calculation unit uses a dual-mode calculation and performs score correction. The specific process is as follows: Calculation mode selection: Mode 1: ,in for Data source score, Scoring is given to the data source for intelligent monitoring. To score generalized data sources, the importance of different data sources is distinguished by weighting coefficients. Mode 2: The highest score from each data source is taken as the overall score; Mode switching rule: Unit at full load, i.e., load factor When using mode one, start / stop or low load, i.e., load factor. Mode 2 will be used at that time; Score Correction: Introduce an alarm frequency correction coefficient. The corrected formula is as follows: ,in To calculate a score for the pattern, when the alarm frequency is >10 times / minute ,otherwise ; Alarm classification: based on the corrected comprehensive score Alarms are categorized into four types: This is a type of alarm. It is a Class II alarm. There are three types of alarms. There are four types of alarms. .

2. The plant-wide integrated early warning management platform for multi-source heterogeneous data fusion according to claim 1, characterized in that, The image data acquisition layer includes a source acquisition unit, an interface adaptation unit, and a data transmission unit. The source-splitting acquisition unit is configured with multiple types of acquisition modules, and the parameters and acquisition rules are as follows: Auxiliary network data acquisition module: Uses a PI interface to acquire analog data, with an acquisition frequency of... and Measurement point scanning cycle Matching, satisfying ,in The value range is 100ms-1s to ensure data real-time performance. System synchronization; Data acquisition module: Acquires switch inputs and alarm information via the IEC104 protocol interface, with data sampling intervals. Data frame transmission bit error rate ; Fire alarm system data acquisition module: based on Protocol for collecting alarm information, communication baud rate The data verification method is The response time for a single communication is ≤300ms; Intelligent monitoring / inspection system data acquisition module: through Interface for collecting diagnostic and inspection data; interface response time. The data format is ; The interface adaptation unit performs format normalization processing on data from different protocols, and the output fields include data source, collection time, parameter name, parameter value and data quality code; The data transmission unit achieves a secure zone through a one-way isolation gateway. Inter-channel data transmission, network gateway data throughput Transmission delay And supports Dual protocol stack.

3. The plant-wide integrated early warning management platform for multi-source heterogeneous data fusion according to claim 2, characterized in that, The early warning analysis layer includes a weight calculation unit, which performs the assignment of data source weights and measurement point importance weights, as well as operating condition calibration. The specific process is as follows: Basic weight assignment: Data source base weights: Intelligent inventory monitoring system ubiquitous sensing system ; Basic weighting of measurement point importance: jump points Protection point Generally speaking ; Operating condition calibration: Introducing operating condition correction factors The calibration formula is ,in The rule for determining the value is: the unit is at full load, i.e., the load factor. hour Low load, i.e., load factor hour Other operating conditions The weights are dynamically adjusted based on the load status. in, To calibrate the data source weights, The weighting of the importance of the measurement points after calibration, and the weighting calculation error. , in, The calculation error of the comprehensive weight of a single measuring point, The total weight value of a single measuring point is calculated in practice, derived from the weight of the calibrated data source. × Importance weight of measurement points after calibration get; This is the theoretical baseline value for the comprehensive weight of a single measurement point, derived from the basic weight of the data source. Basic weight of measurement point importance Theoretical value of working condition correction factor The calculated value is based on the platform's preset weighting rules; The early warning analysis layer also includes a severity assessment unit, which calculates the severity M for different data sources, according to the following rules: DCS / Ubiquitous Sensing System Data Source: If the data is between the preset high and low reporting thresholds ; If the data exceeds the high reporting threshold or the low reporting threshold ; If the data exceeds the high-high reporting threshold or the low-low reporting threshold... ; If the data exceeds the trip threshold ; The intelligent inventory monitoring system uses a multi-parameter coupling algorithm for its data source, as shown in the formula below. ,in The difference between the parameter and the normal range. The value represents the impact of the associated equipment malfunction, ranging from 0 to 1. These are the weighting coefficients. ; Single data source, single item score ,in For comprehensive weighting, This represents the severity value, and the score calculation precision. .

4. The plant-wide integrated early warning management platform for multi-source heterogeneous data fusion according to claim 3, characterized in that, The early warning analysis layer also includes a cold-end optimization early warning module. This module performs optimal vacuum calculations and deviation warnings, and the specific process is as follows: Optimal vacuum value calculation: ,in This represents the current vacuum value of the condenser. For circulating water flow rate, The inlet temperature of the circulating water. The outlet temperature of the circulating water. The specific heat capacity of water, For the density of circulating water, The heat exchange area of ​​the condenser; Warning threshold setting: Vacuum deviation warning threshold ,in Design a vacuum for the condenser. For ambient atmospheric pressure; when At that time, generate cold-end optimization early warning information; Energy consumption calculation: Energy consumption of cold end equipment ,in For the head of the circulating water pump, It is the acceleration due to gravity. For pump efficiency, energy consumption calculation error .

5. The plant-wide integrated early warning management platform for multi-source heterogeneous data fusion according to claim 4, characterized in that, The application interaction layer includes visualization display units and mobile interaction units: The visualization unit supports a one-screen overview function, displaying a real-time overview of unit alarms, early warning information, and economic indicators. The calculation of the power supply coal consumption rate is as follows: Basic coal consumption formula: ,in This refers to the coal consumption of the generating unit. The lower heating value of coal. For the generator unit's power generation; Coal quality correction: Introducing coal quality correction coefficients Corrected coal consumption When the calorific value of coal is lower than the design value hour Below the design value hour Energy consumption indicators are dynamically corrected based on differences in coal quality. Refresh rate: Refresh rate of the visual interface , The interface data delay time, in seconds, refers to the time interval between the data output from the early warning analysis layer and its display on the visualization interface. Requirements: , which is the core threshold used to measure the real-time performance of the interface; The mobile interaction unit pushes alarm information via SMS / the group's ICE platform, with the following push rules: Push delay: ; Push success rate: ; Interactive features: Supports alarm receipt, confirmation, and processing status feedback; mobile interface response time. .

6. The plant-wide integrated early warning management platform for multi-source heterogeneous data fusion according to claim 5, characterized in that, The security protection layer includes a data encryption unit, an access control unit, and an anomaly detection unit: Data encryption unit: Employs AES-256 algorithm to encrypt transmitted and stored data; encryption key update cycle. Heavens, the success rate of decrypting encrypted data ; Access control unit: Permissions are assigned based on the RBAC model, with permission granularity refined to the data table level, and user operation logs are retained for a specified period. sky, ; Abnormal monitoring unit: Calculate the network load increment , where is the network load before platform access, is the network load after platform access, and it is required that ; When abnormal access is detected, trigger linkage blocking, and the blocking response time .

7. The plant-wide integrated early warning management platform for multi-source heterogeneous data fusion according to claim 6, characterized in that, The data governance layer also includes a historical data management unit, which performs data storage and backup, with the following specific rules: Storage period: One type of alarm data Class II / III / IV alarm data and routine operation data ; Backup strategy: Employ dual backups, both local and off-site, with local backups occurring on a specific period. Off-site backup cycle Backup integrity ; Recovery capabilities: Supports targeted recovery by time interval, alarm type, and device number; single batch data recovery time... To restore data accuracy .

8. A plant-wide integrated early warning management method based on multi-source heterogeneous data fusion, applied in the plant-wide integrated early warning management platform based on multi-source heterogeneous data fusion as described in claim 7, characterized in that... Includes the following steps: S1. Multi-source heterogeneous data acquisition: Through the multi-source acquisition unit, different communication protocols and data formats are adapted to collect production and safety data from the entire plant's DCS system, NCS system, on-site inspection system, fire alarm system, video surveillance system and intelligent monitoring system. After being converted into a unified format by the interface adaptation unit, the data is transmitted to the data governance layer through a secure transmission channel. S2. Data Governance and Knowledge Graph Construction: Cleaning, association, and structuring of the collected multi-source heterogeneous data to construct unit data graphs and knowledge graphs; S3. Multi-dimensional early warning analysis and classification: Based on the unit data graph and knowledge graph, a comprehensive score is generated through weight calculation, severity assessment and score correction. Alarm information is divided into four categories according to the comprehensive score. S4. Warning Information Display and Interaction: Visualize the graded alarm information, push information to designated terminals according to the alarm level, and support alarm business flow management and historical data query; S5. Full-process security control: Implement security protection for the data collection, transmission, storage and access process to ensure system and data security.

Images