Real-time health evaluation method and system for DCS system of nuclear power plant

Abstract

The application belongs to the technical field of DCS system health evaluation, aims to solve the problems of equipment fault monitoring and diagnosis deficiency and difficulty in identifying hidden faults, and discloses a nuclear power plant DCS system real-time health evaluation method and system. The method constructs a health evaluation standard through offline data, inputs online data into a local health evaluation module, calculates the health degree of a single device by using the local health evaluation module, inputs the result output by the local health evaluation module into a whole health evaluation module, and outputs the result of the whole health evaluation by the whole health evaluation module to realize real-time health evaluation. The system comprises an offline standard construction module, a real-time data acquisition module, a local health evaluation module, a whole health evaluation module and a visualization module. Through the cooperative processing of offline and online data, the combined detection of hidden and obvious faults and the multi-level weight fusion mechanism, the application realizes real-time and accurate evaluation of the health state of the nuclear power DCS system and provides dynamic support for system maintenance decision-making.

Description

Technical Field

[0001] This application belongs to the field of DCS system health assessment technology, and in particular relates to a real-time health assessment method and system for a nuclear power plant DCS system. Background Technology

[0002] With the development of digital technology, all newly built power plants and most operating nuclear power plants now use DCS (Digital Control System) technology, and some older power plants that used analog instruments have also gradually undergone digital control transformation.

[0003] As the nervous system and brain of a nuclear power plant, the DCS (Distributed Control System) has a global and systemic impact on the unit. If any part of the DCS deviates during operation, it may cause the relevant equipment in the nuclear power plant to be downgraded. In severe cases, it may even lead to reactor shutdown, turbine shutdown, and reduced unit power, posing a great challenge to the safe, reliable, and economical operation of the nuclear power plant. Therefore, carrying out research on DCS health management technology and gradually improving the reliability and intelligent operation and maintenance level of DCS plays a vital role in the safe and stable operation of the unit.

[0004] Existing methods for assessing the health status of nuclear power plants have the following drawbacks:

[0005] First, there is a lack of equipment fault monitoring and diagnosis. In nuclear power plants, equipment faults can affect safety and efficiency. Overt faults are easy to detect, but covert faults are often difficult to identify.

[0006] Second, health status assessment lacks quantitative standards. Traditional health status assessment methods often lack systematicity and quantitative basis, making it difficult for maintenance personnel to accurately determine whether equipment is in normal working condition.

[0007] Third, insufficient basis for decision-making. In complex equipment environments, a lack of scientific data support may lead to wasted resources and flawed decision-making.

[0008] Fourth, preventative maintenance is not effective enough. Equipment failures in high-risk areas can have serious consequences, and traditional maintenance methods often fail to identify and address potential problems in a timely manner.

[0009] Fifth, insufficient overall system reliability assessment. In nuclear power plants, the failure of a single piece of equipment can lead to a decline in the overall system performance. If the health status of each component cannot be comprehensively assessed, the reliability of the entire system will be threatened. Summary of the Invention

[0010] The main purpose of this application is to provide a real-time health assessment method and system for a nuclear power plant DCS system, which solves the problems of insufficient equipment fault monitoring and diagnosis and difficulty in identifying latent faults. It can monitor and effectively distinguish between these two types of faults in real time, thereby reducing potential risks.

[0011] Another objective of this application is to provide a real-time health assessment method and system for nuclear power plant DCS systems, which solves the problem of lack of quantitative standards for health status assessment and difficulty in accurately determining whether equipment is in normal working condition. This application provides a set of quantitative health calculation standards to make equipment status assessment more explicit and reliable.

[0012] Another objective of this application is to provide a real-time health assessment method and system for a nuclear power plant's DCS system, addressing the problem of resource waste and decision-making errors caused by insufficient decision-making basis. This application provides clear decision-making basis through real-time data analysis, helping management to optimize maintenance strategies to reduce resource waste and operating costs.

[0013] Another objective of this application is to provide a real-time health assessment method and system for a nuclear power plant DCS system, which addresses the insufficient effectiveness of preventive maintenance and the inability to identify and address potential problems in a timely manner. This application enhances the effectiveness of preventive maintenance by monitoring the health status of equipment in real time, ensuring that potential risks can be addressed promptly.

[0014] Another objective of this application is to provide a real-time health assessment method and system for a nuclear power plant DCS system, which solves the problem of insufficient overall system reliability assessment and can improve the stability and safety of the overall system.

[0015] To achieve the above objectives, this application provides the following technical solution:

[0016] Firstly, this application provides a method for real-time health assessment of a nuclear power plant's DCS system, including:

[0017] Step 1: Construct health evaluation standards using offline data, which includes DCS device information, DCS system configuration, and DCS system expert knowledge.

[0018] Step 2: Input the online data into the local health assessment module. The online data includes monitoring data and log data. Use the local health assessment module to calculate the health of a single device.

[0019] Step 3: Input the results output by the local health assessment module into the overall health assessment module. The overall health assessment module outputs the overall health assessment results to achieve real-time health assessment.

[0020] In some embodiments, the health evaluation criteria are constructed based on equipment classification, functional correlation between equipment, and expert knowledge analysis. By integrating DCS equipment information, system configuration, and industry experience, a deduction mechanism with 100 points as the benchmark is established.

[0021] In some embodiments, the input data of the local health assessment module includes abnormal information in the data. After the online data is cleaned and preprocessed, real-time anomalies in the monitoring data and log data are identified through feature extraction and anomaly detection models. The degree of deviation between the device's abnormality and normal state is measured, and the degree of abnormal deviation is converted into the device's health status.

[0022] In some embodiments, equipment health is affected by two factors: equipment failure and equipment anomaly. The input to the equipment health assessment model is the health level when the equipment is only failured and the health level when the equipment is only anomaly. The output is the equipment health level, calculated as follows:

[0023] H dev =max(H fault +H anomaly -100,0)

[0024] In the formula, H dev For equipment health, H fault H represents the health status of the equipment when it is only malfunctioning. anomaly This refers to the health status of the device when it is only malfunctioning.

[0025] In some embodiments, rack health is affected by environment, faults, anomalies, and workload. The input to the rack health assessment model is the equipment health under the influence of these factors, and the output is the rack health score, calculated as follows:

[0026] H = αH e +βH c +γHA s +δH l

[0027] In the formula, H e H is the health score calculated from the environment. c The health score calculated from the faults, HA s H is the health score calculated from the anomalies. l The health score is calculated from the workload, with α, β, γ, and δ being weighting coefficients.

[0028] In some embodiments, the health of the DCS system is affected by the health of each cabinet contained in the system. In this case, the input of the DCS system health evaluation model is various parameters related to the health of the cabinets, and the output is the health of the DCS system.

[0029] The Analytic Hierarchy Process (AHP) is used to assign weights to the CC1 cabinet, CC2 cabinet, NC cabinet, RTM cabinet, DCS Layer 2 system, and network bus. The calculation model is as follows:

[0030]

[0031] In the formula, H s The system health score is represented by β1, β2, β3, and β4, which are the weight values ​​of the four cabinet types (CC1, CC2, NC, and RTM) calculated using the analytic hierarchy process (AHP). β5 and β6 represent the weights of the DCS Layer 2 and network bus, respectively. CC1 N CC2 N NC N RTM The number of CC1, CC2, NC, and RTM racks; i1, i2, i3, and i4 are intermediate variables; For the health status of each DCS cabinet at different levels; H 二层 For DCS Layer 2 health; H LAN 'b' represents the network health score; 'b' represents the DCS system health score calculation coefficient.

[0032] In some embodiments, a real-time multi-window PCA anomaly detection model is used to detect anomalies in log data, using the endpoints of vector y to the normal subspace S. d The distance to y is used to determine whether y is abnormal. The real-time multi-window PCA anomaly detection model is as follows:

[0033] For the original data matrix Standardize (m samples, n features):

[0034]

[0035] In the formula, The mean vector for each feature, The standard deviation vector for each feature;

[0036] Calculate the covariance matrix:

[0037]

[0038] Perform eigenvalue decomposition on Σ:

[0039] Σ=VΛV T

[0040] In the formula, For the eigenvector matrix, according to the corresponding eigenvalues ​​λ1≥λ2≥…≥λ n Sorting, Λ=diag(λ1,λ2,…,λ n ) is a diagonal matrix;

[0041] Select the first k principal components to construct the normal subspace S. d The remaining nk principal components construct the anomalous subspace S. a :

[0042] P = [v1, v2, ..., v k ]

[0043] Q = [v k+1 ,v k+2 ,…,v n ]

[0044] In the formula, P is the normal subspace projection matrix and Q is the abnormal subspace projection matrix;

[0045] For test samples After standardization, we get x std The projections are respectively projected into two subspaces. The projection in the normal subspace (preserving the main variation information) is as follows:

[0046] y d =x std PP T

[0047] The projection into the anomalous subspace (the residual part reflects the anomalous information) is as follows:

[0048] y a =x std -y d =x std (I-PP T )

[0049] Calculate the residual vector y a The square norm is used as an outlier score:

[0050]

[0051] Set threshold θ anomaly If the abnormal score exceeds the threshold, it is judged as abnormal, where θ anomaly It can be determined in the following ways:

[0052] Take the 95th percentile of the score of the normal sample.

[0053] In some embodiments, for each malfunctioning device, the health deduction D is calculated as follows:

[0054] D = A × B

[0055] In the formula, A is the abnormality coefficient of the equipment, and B is the fault deduction value of the equipment;

[0056] The device's health score (HD) is:

[0057] HD = 100 - D

[0058] Health H of the computer cabinet total At that time, a weighted average is calculated by combining the importance weights of the equipment:

[0059]

[0060] In the formula, HD ij N represents the health status of the j-th device in the i-th classification. i W represents the number of devices of the i-th type of classification. i This represents the importance weight of the i-th type of graded equipment.

[0061] In some embodiments, the calculation of superimposed faults of equipment at different levels is as follows:

[0062] If the highest fault level is CC1, then the calculation formula is:

[0063] H c =H1-0.59n2-0.42n3-0.33n4

[0064] If the highest fault level is CC2, then the calculation formula is:

[0065] H c =H2-0.42n3-0.33n4

[0066] If the highest fault level is NC, then the calculation formula is:

[0067] H c =H3-0.33n4

[0068] If the highest fault level is RTM, then the calculation formula is:

[0069] H c =H4

[0070] In the formula, H c For rack health, when H c When the calculation result is less than 0, H = 0;

[0071] When the DCS control cabinet is disconnected from the network, the calculation formula is as follows:

[0072]

[0073] In the formula, α1, α2, α3, and α4 are the cabinet weight values ​​of different levels calculated by the analytic hierarchy process.

[0074] In some embodiments, a workload of 40% or less is set as the maximum score of 100 points; a workload of 60% is set as the score of 60 points; and a workload of 100% is set as the score of 0 points. The formula is as follows:

[0075]

[0076] In the formula, X load H represents the load factor. l This indicates the corresponding score.

[0077] In some embodiments,

[0078] H = αH e +βH c +γHA s +δH l

[0079] In the formula, H e H represents the health score calculated from the environment. c HA represents the health score calculated from the faults. s H represents the health score calculated from the anomalies. l This represents the health score calculated from the workload, where α, β, γ, and δ are weighting coefficients.

[0080] In some embodiments, the analytic hierarchy process (AHP) is used to calculate system health, as follows:

[0081] The Analytic Hierarchy Process (AHP) is used to assign weights to the CC1 cabinet, CC2 cabinet, NC cabinet, RTM cabinet, DCS Layer 2 system, and network bus. The calculation model is as follows:

[0082]

[0083] In the formula, H s This represents the system health status; β1, β2, β3, and β4 are the weight values ​​of the four types of cabinets (CC1, CC2, NC, and RTM) calculated by the analytic hierarchy process; β5 and β6 are the weights of the DCS Layer 2 and network bus; N CC1 N CC2 N NC N RTM This indicates the number of CC1, CC2, NC, and RTM racks; i1, i2, i3, and i4 represent intermediate variables. This indicates the health status of each DCS cabinet at different levels; H 二层 Indicates the second-level health status of DCS; H LAN 'b' represents the network health status; 'b' represents the DCS system health status calculation coefficient.

[0084] Secondly, this application provides a real-time health assessment system for a nuclear power plant's DCS system, comprising:

[0085] The offline standard construction module is used to establish equipment health evaluation standards based on historical offline data, which includes DCS equipment information, DCS system configuration, and DCS system expert knowledge.

[0086] The real-time data acquisition module is used to collect online data in real time, including monitoring data and log data.

[0087] The local health assessment module receives input data from the real-time data acquisition module, has a built-in device health assessment model, and outputs the health status of a single device.

[0088] The overall health assessment module is used to receive the health status of multiple devices and assign weights. It has a built-in cabinet health assessment model and system health assessment model, and calculates and outputs the overall health assessment results.

[0089] The visualization module is used to visualize the output data of the overall health assessment module.

[0090] Compared with existing technologies, the real-time health assessment method and system for nuclear power plant DCS systems provided in this application have the following advantages:

[0091] This application enables real-time health assessment, which can comprehensively and accurately evaluate the operating status and health level of equipment, ensuring the stable and reliable operation of the DCS system.

[0092] This application can improve the fault identification rate. It can monitor and distinguish between explicit and implicit faults in real time, significantly improving the timeliness of fault detection and reducing the impact of equipment failures on safety and performance.

[0093] This application provides quantifiable health status and decision support. It introduces scientific health calculation standards to quantify equipment health status, improve management transparency, and utilizes real-time data analysis to provide reliable data for decision-making, optimize resource allocation, and reduce operating costs.

[0094] This application enhances preventative maintenance capabilities. By monitoring equipment health in real time, it can promptly identify potential problems, improve the effectiveness of preventative maintenance, reduce the risk of sudden failures, and enhance operational safety.

[0095] This application improves overall system reliability. It comprehensively assesses the health status of the DCS system, ensuring efficient collaboration among components and significantly reducing the risk of system crashes due to single points of failure.

[0096] This application can optimize information management and sustainable development, integrate equipment and system data, reduce information fragmentation, improve information processing efficiency, support rapid decision-making, ensure the long-term safe and efficient operation of nuclear power plants, and promote the sustainable development of the industry.

[0097] This application constructs a health assessment standard based on offline data from the DCS system. Offline data includes equipment information, system configuration, and expert knowledge. By extracting equipment classification and evaluation indicators, equipment correlation, equipment weights, and health thresholds, multi-level health calculation rules and weight allocation criteria are formed for equipment, cabinets, and the system. For online data, sensor monitoring data and process control station logs are cleaned and preprocessed. Log features are extracted using a multi-window feature extraction method, and explicit and implicit faults are identified using a fault detection module. Explicit faults are detected through conditional / rule constraints, while implicit faults are detected using a principal component analysis (PCA) model. Based on the detection results and equipment weights, equipment health is calculated to complete a local health assessment. Cabinet health is further calculated by combining cabinet weights, and finally, the weights of other influencing factors are integrated to achieve an overall health assessment of the DCS system. This application, through the collaborative processing of offline and online data, joint detection of explicit and implicit faults, and a multi-level weight fusion mechanism, achieves real-time and accurate assessment of the health status of nuclear power DCS systems, providing dynamic support for system maintenance decisions. Attached Figure Description

[0098] To more clearly illustrate the technical solution of this application, the accompanying drawings used in the technical description will be briefly introduced below.

[0099] Figure 1 A flowchart of a real-time health assessment method for a nuclear power plant DCS system provided in this application embodiment;

[0100] Figure 2 A complete flowchart of DCS real-time health assessment provided for embodiments of this application;

[0101] Figure 3 A schematic diagram illustrating the data cleaning and preprocessing process provided in the embodiments of this application;

[0102] Figure 4 A schematic diagram of three types of windows provided in the embodiments of this application;

[0103] Figure 5 A schematic diagram illustrating online data feature extraction provided in an embodiment of this application;

[0104] Figure 6 A schematic diagram illustrating the principle of PCA detection using simplified data, provided for embodiments of this application;

[0105] Figure 7 A flowchart of real-time local health assessment provided in the embodiments of this application;

[0106] Figure 8 This is a schematic diagram of the structure of a real-time health assessment system for a nuclear power plant DCS system provided in an embodiment of this application. Detailed Implementation

[0107] The following detailed description provides further details on specific implementation methods.

[0108] like Figures 1 to 7 As shown in the illustration, this application provides a real-time health assessment method for a nuclear power plant's DCS system, including local health assessment and overall health assessment. The local health assessment is the health assessment of the DCS equipment, and the overall health assessment is the health assessment of the DCS cabinets and system. The overall health assessment is a summary of multiple local health assessments. After the local health status of the DCS system is assessed, the results of multiple local health assessments are weighted and calculated to obtain the overall health assessment. The method includes the following steps:

[0109] Step 1: Construct a health evaluation standard. The data source used to construct this standard is offline data, and a health score is used to measure whether the device is healthy.

[0110] Step 2: Input the online data into the local health assessment module. The online data includes monitoring data and log data. Use the local health assessment module to calculate the health of a single device.

[0111] Step 3: Input the results (equipment health status) output by the local health assessment module into the overall health assessment module. The overall health assessment module then outputs the overall health assessment results to achieve real-time health assessment.

[0112] By following the steps above, real-time health assessment can be achieved, enabling a comprehensive and accurate evaluation of the equipment's operating status and health level, thus ensuring the stable and reliable operation of the DCS system.

[0113] The real-time health assessment method for a nuclear power plant's DCS system is designed based on DCS system data, which is divided into offline and online data. Offline data includes DCS equipment information, system configuration, and expert knowledge. This offline data can be further used to extract equipment classification and evaluation indicators, equipment correlation, equipment weight, and health threshold information, which together constitute the health assessment criteria for the DCS system.

[0114] The health assessment criteria for DCS systems define how to calculate the health of equipment, cabinets, and systems, and also determine the weights for various factors in the health calculation process. These weights include equipment weights, cabinet weights, and weights for other influencing factors.

[0115] Online data includes sensor monitoring data and DCS system process control station logs. After data cleaning and preprocessing, multi-window feature extraction methods are used to extract features from the log data. This data is then sent to the fault detection module to detect explicit and implicit faults. Explicit faults are detected using conditional / rule constraints, while implicit faults are detected using PCA. After a fault is detected, the equipment's fault status is combined with equipment weights to calculate the equipment's health score. This health score calculation is a partial health assessment.

[0116] The overall health assessment includes rack health assessment and DCS system health assessment. After the equipment health is calculated, the rack health is calculated by combining the rack weight. After the rack health is calculated, the DCS system health is calculated by combining the weights of other influencing factors. This completes the real-time health assessment of the DCS system.

[0117] In this embodiment, the health of a single device is calculated by the device health evaluation model in the local health evaluation module. Then, the health of these devices, together with the weighting factors, is input into the overall health evaluation module, and the cabinet health and DCS system health are calculated by weighting, thus achieving real-time health evaluation.

[0118] In Step 1, the construction of health evaluation standards is based on equipment classification, functional correlation between equipment, and expert knowledge analysis. By integrating DCS equipment information, system configuration, and industry experience, a deduction mechanism with a baseline of 100 points is established. Quantitative rules are formulated based on dynamic factors such as environmental conditions (temperature / humidity), abnormal workload, and equipment failures to reflect the system's health status and ensure compliance with industry standards. The methodology for constructing health evaluation standards includes:

[0119] Offline data is generated based on DCS equipment information, system configuration, and expert knowledge. Equipment classification rules, functional correlation rules between equipment, and health calculation rules are defined.

[0120] The equipment classification rules prioritize equipment failures based on their impact on the overall health of the system.

[0121] The functional correlation rules between devices establish a device association model through a collaborative working mechanism of fieldbus or Ethernet communication;

[0122] The health score calculation rule adopts an initial 100-point deduction system. By analyzing DCS operating environment parameters, equipment operating load thresholds and failure mode data, abnormal threshold ranges for temperature, humidity and workload are set, and the deduction value for different equipment failures or abnormal states is determined based on Failure Mode and Effects Analysis (FMEA).

[0123] By combining equipment classification weights, equipment-related failure propagation paths, and expert experience knowledge bases, the health assessment model is dynamically revised to ensure that the assessment standards comply with nuclear power plant safety regulations and industry best practices.

[0124] Specifically, the construction of health assessment standards needs to consider factors such as equipment classification, equipment relevance, expert experience and knowledge, temperature, humidity, workload, equipment malfunctions and abnormalities. The health score mechanism adopts a deduction system, with an initial health score of 100 points. If the temperature, humidity, or workload exceeds certain rules, or if the equipment malfunctions or becomes abnormal, the health score will be deducted accordingly.

[0125] In this embodiment of the application, the offline data used to construct the health evaluation criteria includes DCS device information, DCS system configuration, and DCS system expert knowledge, as shown in Table 1:

[0126] Table 1. Offline Data for Real-Time Health Assessment

[0127]

[0128] When constructing health assessment criteria, various factors affecting DCS operation must be considered, including but not limited to: DCS equipment failure modes, DCS system operating environment conditions, DCS equipment operating load, DCS system and equipment design requirements, and DCS system and equipment technical parameters. The process of constructing health assessment criteria using offline data consists of three parts: equipment classification, equipment relevance, and expert knowledge.

[0129] In nuclear power plants, the impact of DCS equipment failures varies depending on its level, leading to different degrees of health degradation. Therefore, equipment classification is a crucial component in constructing health assessment standards. Secondly, DCS system equipment communicates and collaborates via fieldbus or Ethernet, exhibiting functional connections and inter-device correlations. The collaborative operation and functional connections between these devices directly influence system health and performance. Finally, expert knowledge is essential for developing health calculation rules. Through expert knowledge, failure mode and impact analysis (FMEA) can be performed to identify potential failure modes and their effects on system health, ensuring that standards comply with industry norms and best practices.

[0130] In step 2, monitoring data includes, for example, temperature, humidity, and workload. This data can be directly input into the local health assessment module. Log data records all equipment events, and fault information from these logs can be directly input into the local health assessment module.

[0131] In addition to the data directly input into the local health assessment module, abnormal information from the data also needs to be input. This abnormal information is obtained through anomaly detection of the online data. The process includes: first, the online data undergoes data cleaning and preprocessing; then, feature extraction and anomaly detection models are used to identify real-time anomalies in the monitoring and log data, measuring the degree of deviation between the equipment's abnormal state and its normal state, and inputting this deviation into the local health assessment module.

[0132] In step 2, online data is used to extract useful information for input into the local health assessment module, including monitoring data and log data, as shown in Table 2:

[0133] Table 2 Online Data for Real-Time Health Assessment

[0134]

[0135]

[0136] Various forms of noise may be introduced during data acquisition, such as sensor errors and environmental interference. Furthermore, the actual collected data may contain missing values, for example, due to network problems during transmission. Therefore, data cleaning and preprocessing of online data are crucial for improving data quality and contribute to the performance of subsequent models. Even after improving data quality, this data cannot be directly fed into the model because it often contains highly redundant information, which can obscure relatively limited discriminative features. Therefore, feature extraction is necessary. The extracted features are then fed into the PCA (Principal Component Analysis) anomaly detection algorithm model to detect anomalies in the data in real time. Real-time faults, however, are bypassed by the anomaly detection step and directly impact the health assessment model.

[0137] The impact of offline and online data on the health assessment method is reflected in the weight allocation. The weight adjustment of offline data for the local health assessment module is coordinated by expert knowledge. Specifically, the role of expert knowledge here is mainly reflected in the following aspects: First, experts can determine the importance of each device in the overall system based on historical data and experience, thus influencing the weight allocation; second, experts can identify which features or indicators are more critical for device health assessment and prioritize these factors. On the other hand, after anomaly detection, the weight of the local health assessment module is adjusted according to the degree of deviation from the normal range. When online data deviates significantly from the normal range, the model will adjust the weights accordingly to reflect the actual health status of the device. Finally, the health result output by the local health assessment module will be used as input to the overall health assessment module, outputting the overall health assessment result, thereby achieving real-time health assessment of the DCS.

[0138] In this embodiment, online data is divided into monitoring data and log data. Monitoring data includes temperature, humidity, and load data of the environment where the DCS equipment is located, while log data is instrumentation log data. Data cleaning identifies and eliminates noise in the dirty data. The process includes:

[0139] (1) Given a database instance I with schema R and data quality requirements;

[0140] (2) Find database instance I′ that meets all data quality requirements while minimizing cleaning costs.

[0141] To improve the quality of data content, especially in areas such as missing data completion, data deduplication, and error correction, the indicators guiding data cleaning can be specifically represented as integrity constraints and data cleaning rules.

[0142] Integrity constraints require that primary keys in a data table cannot be null or duplicated, and that columns in the table must satisfy certain data type constraints. In addition, there are user-defined integrity constraints. These constraints mostly reflect the interdependencies and constraints between attributes or groups of attributes; clean data must satisfy these constraints. Common integrity constraints include conditional functional dependencies, inclusion dependencies, metric dependencies, and differential dependencies. Data cleaning rules identify data noise and its corresponding true values. When the attribute values ​​in a data table match the true values ​​indicated by the rules, the data meets data quality requirements. Some cleaning rules directly encode the true values ​​within the rules, such as repair rules; others require establishing a correspondence between external data sources and database instances to obtain correct data, such as edit rules, Sherlock rules, and probe rules. Commonly used external data sources include master data and knowledge bases. Data quality standards and specifications can be formalized into data cleaning rules.

[0143] The monitoring data is structured continuous time series data, such as Figure 3 As shown, the data cleaning and preprocessing process for monitoring data includes:

[0144] 1) Conduct a preliminary check on the collected data to identify whether the data format is correct and whether the fields are complete;

[0145] 2) Determine that all data points are aligned at fixed time intervals. If the data collection intervals are inconsistent, resampling or interpolation is required.

[0146] 3) Check for missing values ​​in the data and choose an appropriate method to process them, such as deletion, interpolation, or filling.

[0147] The log data is semi-structured discrete sequence data, with fields including date, time, device name, device information, system display, message code, and message details. The data cleaning and preprocessing process for the log data includes:

[0148] 1) Parse and structure the logs, converting semi-structured log entries into structured data;

[0149] 2) Remove irrelevant information. The log file may contain many irrelevant lines, such as debugging information or duplicate log entries, which need to be removed.

[0150] 3) Log merging and sorting: Merge log data from different sources and sort them by timestamp for easier subsequent analysis and processing.

[0151] In this embodiment of the application, the feature extraction methods for the time series composed of monitoring data include: a time-domain-based feature extraction method, a frequency-domain-based feature extraction method, and a time-frequency-domain-based feature extraction method.

[0152] In one embodiment, the time-domain-based feature extraction method involves calculating a series of statistical indicators, including mean, standard deviation, root mean square, skewness, kurtosis, peak factor, and signal-to-noise ratio, as shown in the following formulas:

[0153] The formula for calculating the mean is:

[0154]

[0155] The formula for calculating standard deviation is:

[0156]

[0157] The formula for calculating the root mean square is:

[0158]

[0159] The formula for calculating skewness is:

[0160]

[0161] The formula for calculating kurtosis is:

[0162]

[0163] The formula for calculating the peak factor is:

[0164]

[0165] The formula for calculating the signal-to-noise ratio is:

[0166]

[0167] In the formula, x i Let N be the i-th data point, N be the total number of data points, and S be the ith data point. i It is the i-th signal data point, N i It is the i-th noisy data point.

[0168] Existing signal processing techniques, such as autocorrelation and cross-correlation, convolution, fractal analysis, and correlation dimension, are also available. The calculation method is as follows:

[0169] The formula for calculating autocorrelation is:

[0170]

[0171] The formula for calculating cross-correlation is:

[0172]

[0173] The formula for calculating the convolution value is:

[0174]

[0175] In the formula, R(τ) is the autocorrelation function value with lag τ, τ is the lag time, and R... xy (τ) is the cross-correlation function value of lag τ, y i+τ Let x(t) be the i-th data point of signal y at time lag τ, (x*h)(t) be the convolution value of signals x and h at time t, x(τ) be the value of signal x at time τ, and h(t-τ) be the value of signal h at time t-τ. i x represents the value at time i. i+τ This represents the value at time i+τ, where τ is the time lag (or time delay).

[0176] Fractal analysis and correlation dimension are solved using existing iterative algorithms.

[0177] In one embodiment, model-based time-domain feature extraction methods, such as autoregressive (AR, ARMA) or probability distribution models, have features composed of model parameters. The autoregressive model is calculated as follows:

[0178]

[0179] In the formula, x t φ represents the data value for time t. i It is the i-th parameter of the autoregressive model, x t-i It is the data value of time ti, ε t is the noise term at time t, and p is the order of the autoregressive model.

[0180] The autoregressive moving average model is calculated as follows:

[0181]

[0182] In the formula, θ j q is the j-th parameter of the moving average model, and p and q are the orders of the autoregressive and moving average models, respectively.

[0183] In one embodiment, in a frequency-domain-based feature extraction method, the power spectrum is obtained through spectral analysis of the data. The Fast Fourier Transform (FFT) is applied to the original data to extract the power spectrum and obtain information about characteristic frequencies. The FFT is calculated as follows:

[0184]

[0185] In the formula, X k It is the result of the Fourier transform with frequency k, x n Here are the data values ​​at time n, where N is the total number of data points. The power spectral density is:

[0186]

[0187] In the formula, P(f) is the power spectral density at frequency f, X(f) is the Fourier transform result at frequency f, and N is the total number of data points.

[0188] In one embodiment, the time-frequency domain-based feature extraction method employs the short-time Fourier transform, which is calculated as follows:

[0189]

[0190] In the formula, X(t,f) is the short-time Fourier transform result of time t and frequency f, x(τ) is the data value at time τ, and w(τ-t) is the value of the window function at time τ-t.

[0191] In one embodiment, the time-frequency domain-based feature extraction method employs wavelet transform, and the wavelet transform is calculated as follows:

[0192]

[0193] In the formula, W x (a,b) represents the wavelet transform result of scale a and displacement b, x(t) represents the data value at time t, and ψ(t) represents the mother wavelet function. It is the complex conjugate of the mother wavelet function.

[0194] In one embodiment, the time-frequency domain-based feature extraction method employs empirical mode decomposition, which is an adaptive method that decomposes a signal into intrinsic mode functions. The specific mathematical expression depends on the iterative algorithm.

[0195] These methods aim to capture how the frequency components of data change over time, and are particularly suitable for non-stationary time series analysis.

[0196] In this embodiment, log data has two characteristics that differentiate its processing from that of monitoring data. First, while log data has timestamps, it is not continuous because logs are only recorded when a corresponding event occurs. Therefore, continuous data processing methods suitable for monitoring data are unsuitable for log data. Second, log data contains textual information, while monitoring data is numerical data. This textual content cannot be directly input into the algorithm model; therefore, information needs to be extracted from the text and converted into numerical features so that it can be directly input into the algorithm model.

[0197] Based on the above analysis, the input for log data feature extraction is the log events generated in the log parsing step, and the output is an event count matrix. To extract features, the log data needs to be divided into different groups, each representing a log sequence. Therefore, existing windowing techniques are applied to divide the log dataset into finite blocks. For example... Figure 4 As shown, there are three different types of windows available: fixed windows, sliding windows, and session windows.

[0198] (1) Fixed window

[0199] Both fixed and sliding windows are timestamp-based, recording the time each log entry occurs. Each fixed window has a size, i.e., a time span or duration. For example... Figure 4 As shown, the window size is Δ t This is a constant value, such as an hour or a day. Therefore, the number of fixed windows depends on the predefined window size. Logs occurring within the same window are considered a log sequence.

[0200] (2) Sliding window

[0201] Unlike fixed windows, sliding windows have two attributes: window size and step size. For example, a window might slide every five minutes for every hour. Typically, the step size is smaller than the window size, thus causing different windows to overlap. Figure 4 The display window size is ΔT, and the step size is the distance moved forward. The number of sliding windows is typically greater than that of fixed windows, depending primarily on the window size and step size. Logs occurring within the same sliding window are also grouped into a log sequence, although logs may repeat in multiple sliding windows due to overlap.

[0202] (3) Session window

[0203] Compared to the two window types mentioned above, session windows are based on identifiers rather than timestamps. Identifiers are used to mark different execution paths within certain log data. For example, logs for the same device record device online / offline status, port enabling / disabling, communication, etc. Therefore, logs are grouped according to identifiers, and each session window has a unique identifier.

[0204] After constructing the log sequences using windowing techniques, an event counting matrix is ​​generated. Within each log sequence, the occurrence count of each log event is recorded to form an event count vector. For example, if the event count vector is [0,0,2,3,0,1,0], it means that event 3 occurred twice and event 4 occurred three times in this log sequence. Finally, a large number of event count vectors are constructed to form an event counting matrix.

[0205] In summary, the feature extraction step of the data adopts, such as Figure 5 The proposed schemes extract relevant features containing important information from both monitoring data and log data.

[0206] In this embodiment, a real-time multi-window PCA anomaly detection model is used. This anomaly detection algorithm model achieves efficient anomaly detection of log data through multi-window feature extraction, PCA dimensionality reduction, and distance metric methods. The details are as follows:

[0207] For the original data matrix Standardize (m samples, n features):

[0208]

[0209] In the formula, The mean vector for each feature, The standard deviation vector for each feature;

[0210] Calculate the covariance matrix:

[0211]

[0212] Perform eigenvalue decomposition on Σ:

[0213] Σ=VΛV T

[0214] In the formula, For the eigenvector matrix, according to the corresponding eigenvalues ​​λ1≥λ2≥…≥λ n Sorting, Λ=diag(λ1,λ2,…,λ n ) is a diagonal matrix;

[0215] Select the first k principal components to construct the normal subspace S. d The remaining nk principal components construct the anomalous subspace S.a :

[0216] P = [v1, v2, ..., v k ]

[0217] Q = [v k+1 ,v k+2 ,…,v n ]

[0218] In the formula, P is the normal subspace projection matrix and Q is the abnormal subspace projection matrix;

[0219] For test samples After standardization, we get x std The projections are respectively projected into two subspaces. The projection in the normal subspace (preserving the main variation information) is as follows:

[0220] y d =x std PP T

[0221] The projection into the anomalous subspace (the residual part reflects the anomalous information) is as follows:

[0222] y a =x std -y d =x std (I-PP T )

[0223] Calculate the residual vector y a The square norm is used as an outlier score:

[0224]

[0225] Set threshold θ anomaly If the abnormal score exceeds the threshold, it is judged as abnormal, where θ anomaly It can be determined in the following ways:

[0226] Take the 95th percentile of the score of the normal sample.

[0227] The task of anomaly detection is to identify log data exhibiting abnormal patterns of system behavior that do not conform to expectations. Anomalies can arise from various causes, including input errors, measurement errors, unusual events or rare occurrences, and even potential faults. Anomaly detection technology can automatically identify the data segments most likely indicating problems. When an anomaly occurs in the DCS system, the health score will decrease accordingly; after the anomaly is resolved, the health score will recover, thus reflecting the real-time health status of the DCS system.

[0228] Principal Component Analysis (PCA) is a statistical method that captures patterns in high-dimensional data by automatically selecting a set of coordinates—principal components—that reflect the covariant relationships between the original coordinates. PCA is used to separate recurring patterns in feature vectors, making it easier to detect anomalous message patterns. The running time of PCA is linearly related to the number of feature vectors, thus detection can be scaled to large-scale data.

[0229] Because the dataset used for anomaly detection exhibits strong correlations, the dimensions of the constructed feature vectors are highly correlated. The goal is to identify anomalous vectors that deviate from this correlation pattern. For example, such as... Figure 6 As shown, most data points are close to a straight line (one-dimensional subspace), indicating that the data has a low effective dimensionality. Axis S d A strong correlation between the two dimensions was captured. In the figure, one dimension is far from S. d The data point (e.g., point A) shows an abnormal correlation, thus representing an anomaly; in contrast, point B, while far from most other points, is close to S. d Therefore, it is normal.

[0230] The parameter k required for PCA can be defined as the number of dimensions needed to capture 95% of the data variance. For monitoring data, when the system is in a steady state, the ratios between different state variable values ​​remain roughly constant. For the message count vector of the log, since each dimension corresponds to a specific stage in the program, and these stages are determined by the program logic, messages within a group are correlated. The correlation between messages determined by normal program execution leads to a high correlation between these two feature dimensions. In summary, PCA constructs a k-dimensional (lower) normal subspace S in the original n-dimensional space by capturing the main patterns in the data. d The remaining (nk) dimensions form the anomalous subspace S. a By projecting the vector y onto S d Above (separate it from S) d The components on the vector make it easier to identify anomaly vectors, forming the basis of anomaly detection.

[0231] This embodiment uses the endpoints of vector y to the normal subspace S. d The distance to determine whether y is abnormal is calculated. That is, by calculating the squared prediction error, which is the vector y. a The square length is used to determine this, as follows:

[0232] SPE≡||y a || 2

[0233] In the formula, y a Is y in the abnormal subspace S? a The projection on, ya =(I-PP) T )y, where P is the transformation matrix composed of the first k principal components selected by the PCA algorithm.

[0234] Therefore, the detection rule is:

[0235] If SPE > Q a Then the vector y is marked as an anomaly, where Q a This represents the threshold statistic of the SPE residual function at the (1-a) confidence level.

[0236] To calculate Q a The Q statistic, a well-known test statistic for the SPE residual function, is used. Under the assumption that the data matrix follows a multivariate Gaussian distribution, the threshold Q is calculated. a This ensures that the false alarm probability does not exceed α. Even if the fundamental distribution of the data differs significantly from the Gaussian distribution, the Q statistic remains robust. This is verified by performing several rounds of bootstrapping on the data matrix to obtain the Q statistic. a The empirical distribution if Q a If it roughly follows a Gaussian distribution and has a small variance, this can confirm the existence of Q. a This is a stable hypothesis in the dataset. Using α as the confidence parameter for anomaly detection, we can choose α = 0.001. The specific value should be determined experimentally to determine whether the detection results are sensitive to the choice of this parameter.

[0237] In this embodiment, the local health evaluation module evaluates the health of the DCS equipment. By comprehensively analyzing monitoring data, real-time fault information, and anomaly detection results, it quantitatively assesses the health status of the DCS system equipment and finally outputs the health score of the equipment.

[0238] The inputs to the local health assessment module are monitoring data, real-time fault information, and anomalies detected by the anomaly detection steps. The output is the local health score of the DCS system (the health score of an individual device). The processing of the inputs in the model is mainly based on the health assessment standards constructed by expert experience and knowledge, which are used to apply weighted deductions.

[0239] In Step 1, when constructing the health rating criteria for the DCS system, the equipment classification of the DCS system must be considered. Equipment classification helps to more accurately assess the health status of the system, determine the importance and priority of different equipment, and thus optimize the system's maintenance and operation strategies. DCS equipment can be divided into CC1 equipment, CC2 equipment, NC equipment, and RTM equipment. Since the classification of DCS equipment is based on the classification of the corresponding process equipment, the failure of CC1 equipment in the DCS system has a higher probability of causing transients in the unit. That is, about 0.8% of the equipment health status is the decisive factor for the overall health status of the DCS system. In addition, although the failure of CC2 equipment will not directly cause the unit to generate more than 10% (rated) power disturbance, its failure may cause the loss of control functions of important equipment. If there is a superposition of failures, it may cause a large transient disturbance to the unit's operating status. Generally speaking, the failure of NC equipment has a relatively small threat to the controlled equipment and the unit. The failure of RTM equipment has basically no impact on the stable operation of the system or the unit. It can be seen that the importance of equipment classification and its impact on the assessment of system health status are important when constructing the health evaluation criteria for the DCS system.

[0240] Given a defined DCS equipment classification, the impact of a DCS system equipment failure is deterministic. Therefore, equipment classification should carry significant weight in the assessment of real-time health. Consequently, corresponding to the DCS equipment classification, the real-time health of the DCS is divided into four levels: healthy, mildly sub-healthy, severely sub-healthy, and unhealthy.

[0241] The thresholds for the four levels of DCS real-time health are defined as follows:

[0242] (1) Unhealthy: Health level is in the range of 0 ≤ health level < 60;

[0243] (2) Deep sub-health: Real-time health score is in the range of 60 ≤ health score < 80;

[0244] (3) Mild sub-health: Real-time health score is in the range of 80 ≤ health score < 90;

[0245] (4) Health: Real-time health is in the range of 90≤health≤100.

[0246] When considering the relationships between devices, the impact of overlapping faults is taken into account, and then the weight settings for health calculation are determined based on expert knowledge.

[0247] In summary, the calculation of equipment health adopts a deduction method, using different deduction calculation models for different levels of equipment failure. Finally, a comprehensive health score is obtained by weighting the equipment levels, and the DCS health level is further obtained based on the aforementioned thresholds.

[0248] Specifically, the health score of DCS system equipment during normal operation is 100%. Two factors can cause a decline in real-time health score: faults and anomalies. Faults can be directly obtained from online data, while anomalies cannot be directly reflected in online data and must be detected by a real-time multi-window PCA anomaly detection model. Fault information is directly transmitted to the local health assessment module, but anomalies cannot be directly transmitted. It is necessary to statistically analyze anomaly characteristics such as frequency, magnitude, and duration. These anomaly characteristics are then transmitted to the local health assessment module to calculate the equipment's health score. The process is as follows: Figure 7 As shown.

[0249] In cases where a fault causes a decrease in real-time health, according to expert experience, if the system load shedding is caused by a DCS equipment fault, the health score is directly defined as 59 points. A single CC1 equipment fault can be calculated as an equivalent load shedding of 5%, so a single CC1 equipment fault deducts 2.95 points (59 × 5% = 2.95).

[0250] Compare the importance of CC2, NC, and RTM devices with that of CC1 device. Calculate the deduction value for CC2, NC, and RTM devices when they malfunction by multiplying the ratio of their importance by the deduction value of CC1 device. Assume the deduction values ​​are as shown in Table 3, or for example, based on expert experience.

[0251] Table 3 Equivalent Deduction Values ​​(Assumptions)

[0252] CC1 CC2 NC RTM Importance 1 1 / 5 1 / 7 1 / 9 Deduction value 2.95 0.59 0.42 0.33

[0253] When anomalies cause a decline in real-time health, not every anomaly detected by a device is a fault. For monitoring data, anomalies refer to changes in patterns or signs of faults; they do not include faults themselves. Therefore, after detecting an anomaly, it is necessary to select appropriate health indicators, that is, to define which aspects of the anomaly will affect the health and how they will affect it. For example:

[0254] (1) Abnormal frequency: The number of abnormalities detected within a given time period;

[0255] (2) Abnormal range: The degree to which the abnormal value deviates from the normal range;

[0256] (3) Duration of abnormality: The length of time that the abnormal phenomenon lasts;

[0257] Based on the analysis of the health impact of the above abnormalities, the following is a summary:

[0258] (1) Abnormal frequency

[0259] 0-1 times / day: 0 points deducted; 2-3 times / day: 0.2 points deducted; 4-5 times / day: 0.4 points deducted; 6 times or more / day: 0.8 points deducted.

[0260] (2) Abnormal amplitude

[0261] Less than 1: Deduction of 0 points; 1-2: Deduction of 0.2 points; 2-3: Deduction of 0.4 points; Greater than 3: Deduction of 0.8 points.

[0262] (3) Abnormal duration

[0263] Less than 1 minute: Deduction of 0 points; 1-5 minutes: Deduction of 0.2 points; 5-10 minutes: Deduction of 0.4 points; More than 10 minutes: Deduction of 0.8 points.

[0264] In some embodiments, the calculation of DCS equipment faults and anomalies at different levels is as follows:

[0265] (1) When CC1 device malfunctions:

[0266] H1 = 59 - (2.95) × n1

[0267] A single CC1 equipment failure is calculated as equivalent to a 5% load shedding, and after conversion, a single CC1 equipment failure will result in a deduction of 2.95 points.

[0268] (2) When CC2 equipment malfunctions:

[0269] Threshold ω, CC2 device failure:

[0270]

[0271] (3) When the NC equipment malfunctions:

[0272] Thresholds θ1, θ2, and NC device malfunction:

[0273]

[0274] (4) The total number of RTM devices is less than or equal to the threshold. When RTM equipment fails:

[0275]

[0276] The threshold parameters may differ for different DCS cabinets. Here, n1, n2, n3, and n4 represent the number of faulty CC1, CC2, NC, and RTM devices in the cabinet, respectively.

[0277] When an anomaly is detected by the equipment, an anomaly severity coefficient A is defined based on the degree of the anomaly, ranging from 0 to 1, where 0 represents no anomaly and 1 represents the most severe anomaly. This anomaly severity coefficient can be directly output from the aforementioned real-time multi-window PCA anomaly detection model, i.e., the impact of the anomaly on health. Based on the known fault deduction values ​​(Table 3), assuming it to be B, the impact of the anomaly severity can be set in the following way: For each anomalous device, its health deduction D is calculated:

[0278] D = A × B

[0279] In the formula, A is the abnormality coefficient of the equipment, and B is the fault deduction value of the equipment. Therefore, the health score HD of the equipment is:

[0280] HD = 100 - D

[0281] Health H of the computer cabinet total In this case, a weighted average can be calculated by combining the importance weights of the equipment:

[0282]

[0283] Among them, HD ij N represents the health status of the j-th device in the i-th classification. i W represents the number of devices of the i-th type of classification. i This represents the importance weight of the i-th type of graded equipment.

[0284] By calculating faults and anomalies in DCS equipment at different levels, it is possible to achieve refined assessment, risk control, optimized maintenance strategies, and rational allocation of maintenance resources. Specifically:

[0285] Refined Assessment. Different Equipment Levels: Equipment in a DCS system is categorized into different levels (e.g., CC1, CC2, NC, RTM) based on its importance and impact on system operation. Different levels of equipment have different roles and importance within the system; therefore, different deduction values ​​and weightings are required in the health assessment. Distinguishing Between Faults and Anomalies: Faults typically refer to obvious equipment failures, while anomalies refer to abnormal operating conditions that do not reach the level of a fault. By using different deduction values ​​and weightings, the health status of the equipment can be more accurately reflected.

[0286] Risk Control. High-Risk Equipment: For high-risk equipment (such as CC1 equipment), even minor anomalies can have a serious impact on the system, therefore higher deduction values ​​and weights are required. Low-Risk Equipment: For low-risk equipment (such as RTM equipment), even if a fault or anomaly occurs, the impact on the system is smaller, therefore lower deduction values ​​and weights are acceptable.

[0287] Optimize maintenance strategies. Priority management: By assigning different penalty values ​​and weights, the maintenance priority of equipment can be clearly defined. The health status of high-risk equipment requires more frequent and rigorous monitoring, while the maintenance frequency of low-risk equipment can be appropriately reduced.

[0288] Resource allocation. Allocate maintenance resources rationally to ensure critical equipment receives timely maintenance and improves the overall reliability of the system.

[0289] After calculating faults and anomalies of DCS equipment at different levels, the health status of the equipment is the output result of the local health evaluation module. Subsequently, the health status of the equipment will be used to assess the health status of computer cabinets and the system.

[0290] In some embodiments, the calculation of superimposed faults of equipment at different levels is as follows:

[0291] If the highest fault level is CC1, then the calculation formula is:

[0292] H c =H1-0.59n2-0.42n3-0.33n4

[0293] If the highest fault level is CC2, then the calculation formula is:

[0294] H c =H2-0.42n3-0.33n4

[0295] If the highest fault level is NC, then the calculation formula is:

[0296] H c =H3-0.33n4

[0297] If the highest fault level is RTM, then the calculation formula is:

[0298] H c =H4

[0299] In the formula, H c For rack health, when H c When the calculation result is less than 0, H = 0.

[0300] In special circumstances, such as when the DCS control cabinet is disconnected from the network, the calculation formula is as follows:

[0301]

[0302] In the formula, α1, α2, α3, and α4 are the cabinet weight values ​​of different levels calculated by the analytic hierarchy process.

[0303] By taking into account the impact of different levels of equipment and failures, a more comprehensive understanding of the overall health status of the cabinet can be obtained, facilitating refined health assessments. This calculation method represents a holistic risk management approach for the cabinet, with overlaying failure calculations helping to identify and quantify overall risks, especially when facing multiple failures; it can better predict cabinet vulnerabilities. Traditional health assessments may only focus on the status of individual devices, while overlaying failure calculations consider the interactive effects of multiple devices, emphasizing system complexity. Furthermore, this approach aligns perfectly with the safety and reliability requirements of the nuclear power industry, making it highly practical.

[0304] In some embodiments, the real-time health of the environment and load is calculated as follows:

[0305] When the temperature and humidity are above or below the normal operating range, they are included in the DCS cabinet health calculation. Examples of temperature and humidity ranges are shown in Table 4, which can be modified according to power plant guidelines and regulations.

[0306] Table 4 Temperature and Humidity Ranges

[0307]

[0308] Normal range: 100 points, not included in health score calculation, weight a is 0%;

[0309] General range: 80-89 points, included in the health score calculation, with a weight of 20%;

[0310] Severe range: 60-79 points, included in health score calculation, with a weight of 50% (a).

[0311] Unacceptable range: 0-59 points, included in the health score calculation, with a weight of 80%.

[0312] The weight 'a' is currently an assumed value and will be revised based on actual conditions and the experience of power plant experts.

[0313] In the calculation, the relationship between health status and duration t (in hours) is considered:

[0314] (1) The functional relationship between health status, temperature, and duration is as follows:

[0315]

[0316] (2) The functional relationship between health status, humidity, and duration is as follows:

[0317]

[0318] In the formula, t0 is the upper limit of time. If this upper limit is exceeded, t will not continue to increase.

[0319] The environmental health status is as follows:

[0320] H e =min(H T H W )

[0321] If H e If the calculation result is less than 0, the result is 0. The corresponding health level is calculated based on the real-time monitored temperature and humidity, and a comprehensive health level calculation is performed using appropriate weights.

[0322] In addition, workload needs to be considered. The existing "Guidelines for Negotiation of Imported Large Generating Units" clearly stipulates the performance requirements for DCS. The basic indicators for import are that the load rate of operator stations and data processing stations should not exceed 40%, the load rate of control processing units should not exceed 60%, and the network communication load rate should not exceed 50%. The existing "Acceptance Test Procedures for Distributed Control Systems in Thermal Power Plants" stipulates that the load rate of the central processing unit of all control stations should not exceed 60% under severe operating conditions; the load rate of the central processing units of computing stations, operator stations, data management stations, etc., should not exceed 40% under severe operating conditions; the load rate should be tested five times under different operating conditions, and the highest value should be taken, with each test lasting 10 seconds; the memory margin should be greater than 40% of the memory capacity, and the external storage margin should be greater than 60% of the memory capacity.

[0323] When the rack load rate exceeds the standard of 50%, and some rack load rates reach over 60%, the high load rate increases the probability of momentary blockage during data communication and data exchange. Once a momentary failure occurs, it often leads to system disturbances such as shutdown and redundancy switching. In this process, there have been instances where the entire rack has failed and taken out of operation due to switching problems and communication problems, threatening the safe operation of the unit.

[0324] In summary, the load limits set by all parties are shown in Table 5:

[0325] Table 5 Recommended and Upper Limits for Load

[0326] Suggestioner Recommended value Load limit Manufacturer <40％ / Relevant procedures <40％ 60％ power plant <50％ 60％

[0327] The recommended range and upper limit of load rate are based on the combined information from DCS manufacturers, relevant regulations, and owners.

[0328] A workload of 40% or less is awarded a score of 100; a workload of 60% is awarded 60 points; and a workload of 100% is awarded 0 points. For other workloads, the score follows a direct proportional relationship to the line connecting the two boundary points, as follows:

[0329]

[0330] In the formula, X load H represents the load factor. l This indicates the corresponding score.

[0331] In some embodiments, the final calculation result of the overall health of the server rack is:

[0332] H = αH e +βH c +γHA s +δH l

[0333] In the formula, H e H represents the health score calculated from the environment. c HA represents the health score calculated from the faults. s H represents the health score calculated from the anomalies. l This represents the health score calculated from the workload, where α, β, γ, and δ are weighting coefficients.

[0334] By using the above-mentioned comprehensive health calculation of the cabinet, multiple influencing factors can be introduced to comprehensively assess the health status, rather than relying solely on a single indicator.

[0335] In some embodiments, system health is calculated as follows:

[0336] The weights of the indicators can be calculated using two methods:

[0337] (1) Expert experience method: Experts directly assign weights to each rack;

[0338] (2) Analytic Hierarchy Process (AHP): The AHP is used to assign weights to the CC1 cabinet, CC2 cabinet, NC cabinet, RTM cabinet, DCS Layer 2 system, and network bus. β1, β2, β3, and β4 are the weight values ​​of the four cabinet types (CC1, CC2, NC, and RTM) calculated by the AHP, while β5 and β6 are the weights of the DCS Layer 2 system and the network bus. For individual special DCS cabinets, the weights can be adjusted separately after being calculated using this method.

[0339] The calculation model is as follows:

[0340]

[0341] In the formula, H s β1, β2, β3, β4, β5, and β6 represent the weights of the evaluation indicators; N represents the system health level. CC1 N CC2 N NC N RTM This indicates the number of CC1, CC2, NC, and RTM racks; i1, i2, i3, and i4 represent intermediate variables. This indicates the health status of each DCS cabinet at different levels; H 二层 This indicates the DCS Layer 2 health status, calculated using the same method as the DCS cabinet health status, based on the DCS equipment classification and real-time status of the DCS equipment in the KIC system; H LAN 'b' represents the network health status, calculated using the same method as the DCS cabinet health status calculation, based on the DCS device hierarchy and real-time status of the DCS devices in the network bus; 'b' represents the DCS system health status calculation coefficient.

[0342] If any of the following cabinets—CC1, CC2, NC, RTM, DCS Layer 2, or network bus—has a minimum health level of unhealthy, the DCS system is directly defined as unhealthy, starting from a score of 59, with b = 0.59.

[0343] When the lowest health level is deep sub-health, the DCS system directly defines it as deep sub-health, starting from 79 points, with b = 0.79;

[0344] When the lowest health level is mild sub-health, the DCS system directly defines it as mild sub-health, starting from 89 points, with b = 0.89;

[0345] When there is a minimum health level of "healthy", the DCS system directly defines it as "healthy", starting from 100 points, with b=1.

[0346] In some embodiments, after performing health monitoring and evaluation on DCS equipment and systems, this application performs auxiliary instrumentation and control maintenance work, specifically including:

[0347] (1) By identifying faulty or abnormal equipment and related fault information, the expert knowledge base can be called in the health assessment module to find the right equipment fault or maintenance plan to assist in the troubleshooting and maintenance of equipment problems.

[0348] (2) If the expert knowledge base does not cover the fault or maintenance data of the relevant equipment, users can call the fault simulation function in the health evaluation module, rely on the DCS parallel system to perform digital system simulation on the operation events, and interpret the fault development trend of the equipment in order to study the impact of the fault and determine the potential defects of the system.

[0349] (3) For equipment failures that have occurred, users can automatically generate equipment work orders based on the expert knowledge base;

[0350] (4) Users can use the card diagnostic device provided with the platform to diagnose faulty or abnormal equipment, locate faulty components, and facilitate subsequent card repair.

[0351] (5) Every six months or one year, the platform will provide optimization suggestions for periodic testing or preventive maintenance cycles based on the health evaluation report, including the health status, trends and number of failures of the DCS system and each cabinet; at the same time, the platform will provide corresponding optimization suggestions based on various health evaluation indicators (e.g., working environment, spare parts status, equipment aging status, etc.).

[0352] In addition, such as Figure 8 As shown in the embodiments of this application, a real-time health assessment system for a nuclear power plant's DCS system is also provided, comprising:

[0353] The offline standard construction module is used to establish equipment health evaluation standards based on historical offline data, which includes DCS equipment information, DCS system configuration, and DCS system expert knowledge.

[0354] The real-time data acquisition module is used to collect online data in real time, including monitoring data and log data.

[0355] The local health assessment module receives input data from the real-time data acquisition module, has a built-in device health assessment model, and outputs the health status of a single device.

[0356] The overall health assessment module is used to receive the health status of multiple devices and assign weights. It has a built-in cabinet health assessment model and system health assessment model, and calculates and outputs the overall health assessment results.

[0357] The visualization module is used to visualize the output data of the overall health assessment module.

[0358] The real-time health evaluation system for nuclear power plant DCS provided in the above embodiments can realize the technical solutions described in the above method embodiments. The specific implementation principles of each module can be found in the corresponding content in the above method embodiments, and will not be repeated here.

[0359] In addition, this application also provides an electronic device, including a memory and a processor, wherein the memory stores computer-readable instructions, and the processor executes the computer-readable instructions to implement the above-mentioned real-time health assessment method for a nuclear power plant DCS system.

[0360] In addition, this application also provides a computer-readable storage medium storing computer-readable instructions, which, when executed, implement the above-described real-time health assessment method for a nuclear power plant DCS system.

[0361] Computer-readable media, including both permanent and non-permanent, removable and non-removable media, can store information using any method or technology. Information can be computer-readable instructions, data structures, program modules, or other data.

[0362] The above description is only a specific embodiment of this application, but the protection scope of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the protection scope of this application.

Claims

1. A method for real-time health assessment of a nuclear power plant's DCS system, characterized in that, include: Step 1: Construct health evaluation standards using offline data, which includes DCS device information, DCS system configuration, and DCS system expert knowledge. Step 2: Input the online data into the local health assessment module. The online data includes monitoring data and log data. Calculate the health of a single device using the device health assessment model in the local health assessment module. Step 3: Input the results output by the local health assessment module into the overall health assessment module. The overall health assessment module outputs the overall health assessment results to achieve real-time health assessment.

2. The real-time health assessment method for a nuclear power plant DCS system according to claim 1, characterized in that, In step 1, the construction of health evaluation standards is based on equipment classification, functional correlation between equipment and expert knowledge analysis. By integrating DCS equipment information, system configuration and industry experience, a deduction mechanism with 100 points as the benchmark is established.

3. The real-time health assessment method for a nuclear power plant DCS system according to claim 1, characterized in that, The input data for the local health assessment module includes abnormal information in the data. After online data is cleaned and preprocessed, real-time anomalies in monitoring data and log data are identified through feature extraction and anomaly detection models. The degree of deviation between the abnormal and normal state of the equipment is measured, and the degree of deviation is input into the local health assessment module.

4. The real-time health assessment method for a nuclear power plant DCS system according to claim 1, characterized in that, The equipment health assessment model takes as input the equipment's health level when it only fails and the equipment's health level when it only malfunctions, and outputs the equipment's health level. The calculation formula is as follows: H dev = max(H fault +H anomaly -100.0) In the formula, H dev For equipment health, H fault H represents the health status of the equipment when it is only malfunctioning. anomaly This refers to the health status of the device when it is only malfunctioning.

5. The real-time health assessment method for a nuclear power plant DCS system according to claim 1, characterized in that, For each malfunctioning device, the health deduction D is calculated as follows: D = A × B In the formula, A is the abnormality coefficient of the equipment, and B is the fault deduction value of the equipment; The device's health score (HD) is: HD = 100 - D Health H of the computer cabinet total At that time, a weighted average is calculated by combining the importance weights of the equipment: In the formula, HD ij For the health status of the j-th device in the i-th classification, N i Let W be the number of the i-th type of grading equipment. i Let be the importance weight of the i-th type of graded equipment.

6. The real-time health assessment method for a nuclear power plant DCS system according to claim 1, characterized in that, The calculation of superimposed faults in equipment of different levels is as follows: If the highest fault level is CC1, then the calculation formula is: H c =H1-0.59n2-0.42n3-0.33n4 If the highest fault level is CC2, then the calculation formula is: H c =H2-0.42n3-0.33n4 If the highest fault level is NC, then the calculation formula is: H c =H3-0.33n4 If the highest fault level is RTM, then the calculation formula is: H c =H4 In the formula, H c For rack health, when H c When the calculation result is less than 0, H = 0; When the DCS control cabinet is disconnected from the network, the calculation formula is as follows: In the formula, α1, α2, α3, and α4 are the cabinet weight values ​​of different levels calculated by the analytic hierarchy process.

7. The real-time health assessment method for a nuclear power plant DCS system according to claim 1, characterized in that, A workload of 40% or less is awarded 100 points; a workload of 60% is awarded 60 points; and a workload of 100% is awarded 0 points. The formula is as follows: In the formula, X load H represents the load factor. l This corresponds to the score.

8. The real-time health assessment method for a nuclear power plant DCS system according to claim 1, characterized in that, The overall health score of the server rack is calculated as follows: H=αH e +βH c +γHA s +δH l In the formula, H e H is the health score calculated from the environment. c The health score calculated from the faults, HA s H is the health score calculated from the anomalies. l The health score is calculated from the workload, with α, β, γ, and δ being weighting coefficients.

9. The real-time health assessment method for a nuclear power plant DCS system according to claim 1, characterized in that, The system health is calculated using the analytic hierarchy process (AHP) as follows: The Analytic Hierarchy Process (AHP) is used to assign weights to the CC1 cabinet, CC2 cabinet, NC cabinet, RTM cabinet, DCS Layer 2 system, and network bus. The calculation model is as follows: In the formula, H s The system health score is represented by β1, β2, β3, and β4, which are the weight values ​​of the four cabinet types (CC1, CC2, NC, and RTM) calculated using the analytic hierarchy process (AHP). β5 and β6 represent the weights of the DCS Layer 2 and network bus, respectively. CC1 N CC2 N NC N RTM The number of CC1, CC2, NC, and RTM racks; i1, i2, i3, and i4 are intermediate variables; For the health status of each DCS cabinet at different levels; H 二层 For DCS Layer 2 health; H LAN 'b' represents the network health score; 'b' represents the DCS system health score calculation coefficient.

10. A real-time health assessment system for a nuclear power plant's DCS system, characterized in that, include: The offline standard construction module is used to establish equipment health evaluation standards based on historical offline data, which includes DCS equipment information, DCS system configuration, and DCS system expert knowledge. The real-time data acquisition module is used to collect online data in real time, including monitoring data and log data. The local health assessment module receives input data from the real-time data acquisition module, has a built-in device health assessment model, and outputs the health status of a single device. The overall health assessment module is used to receive the health status of multiple devices and assign weights. It has a built-in cabinet health assessment model and system health assessment model, and calculates and outputs the overall health assessment results. The visualization module is used to visualize the output data of the overall health assessment module.

Images