A method for distinguishing power station data anomalies from equipment state anomalies

Abstract

The application discloses a power station data anomaly and equipment state anomaly distinguishing method, and belongs to the technical field of power station intelligent operation and maintenance and data management. The application establishes a judgment foundation system by constructing an equipment physical mechanism and parameter linkage rule library, a whole life cycle health file of a measuring point, and a multi-dimensional health working condition benchmark library; excludes pure link data anomalies through pre-checking and screening of a transmission link; locks dynamic benchmark thresholds based on subdivided working conditions, and removes normal working condition fluctuation interference; divides two types of indexes according to the number of associated measuring points, and executes differentiated anomaly preliminary judgment; verifies the physical rationality of anomalies through cross-parameter strong association mechanism linkage checking; and finally generates a judgment result with a confidence rating through multi-model cross verification. The application solves the industry pain point that the prior art cannot accurately distinguish power station data anomalies and equipment state anomalies, greatly reduces the false positive and false negative rates, and provides reliable data support for power station intelligent operation and safety management and control.

Description

Technical Field

[0001] This invention belongs to the field of intelligent operation and maintenance and data governance technology of power plants, and specifically relates to a method for distinguishing between abnormal power plant data and abnormal equipment status. Background Technology

[0002] With the comprehensive advancement of intelligent construction in hydropower stations and pumped storage power stations, power station data platforms have integrated data from multiple systems, including computer monitoring systems, unit condition monitoring systems, and hydrological forecasting systems. Developing intelligent applications based on these data platforms, such as operation assistance, maintenance decision-making, and safety management, has become the industry's development direction. Data quality directly determines the reliability of intelligent applications, and the core pain point currently facing the industry is the inability to accurately distinguish between data anomalies caused by sensor / acquisition / transmission failures and status anomalies caused by equipment degradation / failure.

[0003] The existing technology has the following core defects: 1) Existing data quality control solutions can only determine whether data is abnormal, but cannot distinguish whether the root cause of the abnormality is a data link / sensor failure or an abnormal state of the equipment itself. This makes it impossible for maintenance personnel to accurately locate the object to be dealt with, which greatly reduces maintenance efficiency and may even lead to unplanned downtime caused by false alarms or the expansion of equipment failure due to missed alarms.

[0004] 2) Existing data anomaly detection methods mostly use fixed threshold schemes, which cannot adapt to the operating characteristics of hydropower stations / pumped storage power stations with multiple operating conditions switching between power generation / pumping / phase adjustment and large fluctuations in head / load. They are prone to misjudging normal fluctuations in operating conditions as anomalies, and cannot identify the slow deterioration trend of equipment under operating conditions.

[0005] 3) For scenarios with no redundancy at a single measurement point, the existing solution can only identify extreme anomalies through range verification and jump verification. It cannot distinguish between data anomalies caused by slow sensor drift and state anomalies caused by slow equipment degradation, resulting in a large blind spot.

[0006] 4) Existing discrimination schemes are mostly based on pure data statistical analysis, without taking into account the physical mechanism and operating characteristics of hydropower / pumped storage power station equipment. The discrimination results lack physical basis, have low accuracy, and cannot meet the stringent requirements for safe production of power stations.

[0007] Prior art document CN120723754B discloses a method for monitoring and judging the data quality of hydropower equipment, which only identifies data anomalies and does not address the distinction between data anomalies and equipment status anomalies. Prior art document CN121326893A discloses a method for real-time assessment of power production data quality, which only focuses on the quantitative scoring of data quality and does not solve the problem of distinguishing the root causes of anomalies. Prior art document CN108319649B discloses a method for improving the quality of hydrological and water dispatch data, which only handles anomalies in hydrological data and does not cover the scenario of distinguishing between status anomalies and data anomalies of the main power station equipment. In summary, none of the existing technologies have solved the core industry pain point of accurately distinguishing between data anomalies and equipment status anomalies in hydropower / pumped storage power stations. Summary of the Invention

[0008] The purpose of this invention is to address the shortcomings of existing technologies by providing a method for distinguishing between data anomalies and equipment status anomalies in hydropower stations / pumped storage power stations. By constructing a mechanism-driven, hierarchical, and scenario-adaptive discrimination system, the invention can accurately distinguish between data anomalies and equipment status anomalies, significantly reducing the false alarm and missed alarm rates of the monitoring system and providing reliable support for the intelligent operation and maintenance of power stations.

[0009] To achieve the above objectives, the present invention adopts the following technical solution: A method for distinguishing between abnormal power plant data and abnormal equipment status includes the following steps: S1. Construct a three-in-one basic support system for judgment by establishing a physical mechanism of equipment, a parameter linkage rule library, a full life-cycle health record of measurement points, and a multi-dimensional health condition benchmark library; S2. Real-time data and related operating parameters of the measurement points to be judged are collected in real time, and the three-layer pre-verification of the data acquisition and transmission link is performed to screen out abnormal data due to link failure. S3. Based on a multi-dimensional health condition benchmark library, match the current subdivided operating condition scenario, lock the corresponding benchmark threshold, and shield the interference of transient fluctuations in operating conditions. S4. Based on the number of measurement points associated with the indicators, the indicators to be judged are divided into L1 multi-measurement point redundant indicators and L2 single / double measurement point indicators. Differentiated anomaly initial judgment is performed on each, and single-point data anomalies and transient interference data anomalies are quickly identified. S5. For measurement points that are initially determined to be non-data abnormal by step S4, retrieve the equipment physical mechanism and parameter linkage rule library, perform cross-parameter strong correlation mechanism linkage verification, and determine whether the measurement point abnormality conforms to the equipment physical evolution law. S6. For anomalies verified in step S5, perform multi-model cross-validation to generate the final judgment result and confidence rating, thus distinguishing between data anomalies and equipment status anomalies.

[0010] Furthermore, the equipment physical mechanism and parameter linkage rule library mentioned in S1 constructs a four-dimensional rule set for core power plant equipment such as water pumps and turbines, generators and motors, main transformers, and high-voltage switchgear. This set includes core parameters, strongly correlated parameters, physical constraints, and fault evolution logic. Each rule is also attached with an operating condition adaptation tag (power generation / pumping / phase adjustment / shutdown, load range, head range, etc.) to ensure the adaptability of the rules under different operating scenarios.

[0011] The full lifecycle health record of the measurement point includes the basic attributes of each measurement point (sensor model, range, installation location, associated acquisition device / link), operation and maintenance records (installation time, calibration cycle, historical fault records), and quantified degradation characteristics (sensor degradation coefficient, historical fluctuation characteristics, rated refresh frequency), enabling the traceability and quantification of the sensor's own health status.

[0012] The multi-dimensional health condition benchmark library is divided into five subdivisions based on five dimensions: operating mode, load rate range, working head range, ambient temperature range, and equipment operation stage. The operating mode is divided into four categories: power generation, pumping, phase adjustment, and standby. The load rate is divided into 5% increments, the working head into 2m increments, and the ambient temperature into 5℃ ranges. The equipment operation stage is divided into the break-in period after major overhaul, the stable operation period, and the deterioration period before major overhaul. For each subdivision, historical operating data with no alarms and no operations are selected to automatically calibrate the normal value range, change rate limit, and parameter linkage deviation threshold for each measuring point. At the same time, a seasonal correction coefficient is introduced to dynamically adjust the benchmark.

[0013] Furthermore, in step S2, the pre-verification of the transmission link specifically includes: S21 Fast Link Consistency Check: If the current test point satisfies... A i = A i-a ( A i This represents the measured value at the current measurement point. A i-a The measurement value at the measuring point at the previous moment. a It is a positive integer, 1≤ a If the value remains unchanged for a continuous period (≤60), it is considered a visible communication interruption fault. S22 Comprehensive Quiet Index Verification: Based on measurement point data, monitor the state transition frequency of switch quantities, the refresh cycle of state quantities, and the numerical fluctuation range of analog quantities within a preset time period. Calculate the quiet factor (quiet factor = 1 - actual value / rated value) for the three types of data, and obtain the comprehensive quiet index by weighting. When the comprehensive quiet index is greater than the preset threshold and the duration exceeds the limit, it is determined to be an implicit transmission anomaly. S23 Full-Link Synchronization Verification: Based on NTP system clock synchronization, it verifies the deviation between the measurement point time scale and the system clock, checks whether all measurement points on the same acquisition link / device are abnormally synchronized, and locates batch failures in links such as acquisition cards and forward isolation devices.

[0014] Furthermore, S3 specifically includes: S31. Extract the current operating mode, load rate, working head, equipment operating status, and ambient temperature core operating parameters. S32. Match the optimal subdivided working condition scenario in the multi-dimensional health working condition benchmark library using the cosine similarity algorithm, and lock the corresponding benchmark threshold under the scenario. S33. Set a transient shielding window to shield the transient processes of unit start-up and shutdown, operating condition switching, and maintenance tests. After the operating conditions stabilize, execute the subsequent judgment steps to avoid normal transient fluctuations being misjudged as abnormal.

[0015] Furthermore, in step S4, the L1 type index is an index with more than 2 measurement points associated with a single index (such as multi-point temperature of generator stator, vibration / temperature of multi-shaft bearing of turbine), and the L2 type index is an index with less than or equal to 2 measurement points associated with a single index (such as main transformer oil level, circuit breaker opening and closing position, unit speed).

[0016] The consistency verification of multiple measurement points for the same L1 type indicator is specifically as follows: Z-score standardization is performed on the real-time data of all measurement points under the same indicator, the covariance matrix between measurement points is constructed, and the Mahalanobis distance of all measurement points is calculated. If the Mahalanobis distance of the current measurement point is greater than the preset alarm threshold, it is marked as an outlier measurement point; if only a single measurement point is an isolated outlier point, and the other measurement points of the same indicator are all stable within the benchmark range, it is determined to be a single point data anomaly.

[0017] The time-series trend rationality verification of the L2 category indicators is specifically as follows: extract continuous time-series data of abnormal measurement points and compare them with historical healthy time-series data under the matching working conditions. If the abnormality is an instantaneous and irregular jump, and then quickly returns to the baseline range without a continuous evolution trend, it is determined to be an instantaneous interference data abnormality. If the abnormality is a continuous and monotonous trend change, and the rate of change conforms to the physical evolution law of the equipment, it proceeds to the subsequent mechanism verification step.

[0018] Further, in step S5, the cross-parameter strong correlation mechanism linkage verification specifically involves: retrieving all strongly correlated parameters corresponding to the abnormal measurement points in the equipment physical mechanism and parameter linkage rule base, and verifying whether the strongly correlated parameters show synchronous changes in the same direction as the abnormal measurement points and conform to physical constraints; if the changes of the abnormal measurement points are synchronously matched with the strongly correlated parameters and conform to the equipment fault evolution logic, it is determined to conform to physical laws and proceeds to the subsequent cross-verification step; if the changes of the abnormal measurement points do not have corresponding synchronous changes in the strongly correlated parameters and do not conform to the equipment physical mechanism at all, it is determined to be data abnormal.

[0019] Furthermore, in step S6, the multi-model cross-validation includes: Validation of the time series prediction model: Based on the LSTM time series prediction model, historical health data of measurement points are combined with the current working conditions to predict the normal value range at the current moment, verify whether the abnormal values ​​exceed the prediction range, and predict the future evolution trend. Equipment deterioration coefficient matching verification: Calculate the comprehensive deterioration coefficient of the equipment based on the equipment running time and overhaul cycle. If the abnormal trend is highly matched with the overall deterioration trend of the equipment, it is determined that the equipment is in an abnormal state. Horizontal comparison verification of similar equipment: Horizontal comparison is carried out for equipment of the same power station, model and operating conditions. If only the corresponding measuring point of a single device is abnormal and the other devices are stable, the judgment is made in combination with the mechanism verification results; if multiple devices of the same model are synchronously abnormal at the same measuring point, it is judged as data abnormality.

[0020] The confidence level is divided into three levels: high confidence (≥90%), medium confidence (60%-90%), and low confidence (<60%), each corresponding to a graded operation and maintenance process.

[0021] Furthermore, this method also includes a closed-loop self-learning optimization step: feeding back the abnormal results confirmed by the operation and maintenance personnel on-site to the system, updating the equipment physical mechanism and parameter linkage rule library, the multi-dimensional health condition benchmark library, and the full life cycle health record of the measurement point, while optimizing the core threshold and weight of the discrimination model through machine learning algorithms to achieve continuous iteration of discrimination capability.

[0022] Furthermore, this method also includes special scenario adaptation steps: for transient operating conditions of the unit, maintenance and testing scenarios, and extreme hydrological / weather scenarios, special shielding rules and scenario-based adaptation benchmarks are set to avoid normal fluctuations in the scenario being misjudged as abnormal.

[0023] A system for distinguishing between power plant data anomalies and equipment status anomalies includes: a basic library construction module for constructing the three-in-one discrimination foundation support system; a link pre-verification module for verifying the three-layer acquisition and transmission link; a working condition matching and threshold locking module for performing detailed working condition matching and transient shielding; a differential preliminary judgment module for performing preliminary judgment of L1 / L2 class index anomalies; a mechanism linkage verification module for performing cross-parameter strong correlation verification; and a cross-validation and discrimination output module for performing multi-model verification and confidence rating. The system is used to execute the aforementioned discrimination method.

[0024] The embodiments of the present invention bring the following beneficial effects: This invention constructs a parameter linkage rule library based on the physical mechanisms of core equipment in hydropower / pumped storage power stations, providing a solid physical basis for anomaly identification and solving the problems of low accuracy and lack of mechanistic support in existing pure data statistical analysis. This invention adopts a hierarchical and progressive identification process: first, pre-verification filters out clear pure data anomalies; then, preliminary classification narrows the identification range; and finally, mechanism-linked verification and cross-validation complete accurate identification, balancing identification efficiency and accuracy, while perfectly adapting to all scenarios with multiple redundant measuring points and single non-redundant measuring points. This invention constructs a multi-dimensional, subdivided health benchmark library, fully adapting to the operating characteristics of hydropower / pumped storage power stations with multiple operating condition switching and significant fluctuations in head / load, solving the problems of misjudgment and missed judgment in existing fixed threshold schemes. This invention introduces a full lifecycle health record for measuring points and a dual communication link verification model, filtering out pure data anomalies caused by data source failures and link faults from two dimensions: sensor health itself and the entire acquisition and transmission link, significantly narrowing the subsequent identification range and improving the accuracy of anomaly root cause location. This invention establishes a three-level confidence rating system that matches the power plant's graded operation and maintenance process, which can reduce the interference of invalid alarms on production operations. At the same time, it constructs a closed-loop self-learning optimization system to achieve continuous iteration of the discrimination capability and adapt to changes in equipment operating status over a long period of time. Attached Figure Description

[0025] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0026] Figure 1 This is a schematic diagram of the overall process of the method described in the embodiments of the present invention; Figure 2 This is a schematic diagram of the hierarchical progressive discrimination process described in an embodiment of the present invention. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions 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, 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.

[0028] This embodiment uses a 300MW pumped storage power station as an application scenario. The power station includes four pumped storage units. The data platform is connected to a computer monitoring system, a unit status monitoring system, a hydrological monitoring system, and a high-voltage equipment online monitoring system, covering the production control area and the management information area, with a total of over 20,000 monitoring points. Based on the above scenario, this embodiment details the specific implementation of the method of the present invention to ensure that those skilled in the art can fully implement the present invention.

[0029] S1. Determine the construction of the basic support system. This step provides the basis for judgment and needs to be initially constructed during the construction phase of the power plant data platform, and continuously iterated and updated during operation.

[0030] 1. Construction of a rule base for linking equipment physical mechanisms and parameters For the core equipment of the power plant, four-dimensional rule sets are constructed respectively. Examples of rules for core equipment are as follows:

[0031] 2. Construction of a full life-cycle health record based on measurement points Health records were created for each of the more than 20,000 monitoring points connected to the platform, as shown in the example below: Basic attributes: Measurement point name: #1 water guide bearing temperature 1; Sensor model: PT100; Measurement range: 0-200℃; Installation location: #1 water turbine water guide bearing; Acquisition device: #1 LCU cabinet AI acquisition card; Safety zone: Production Control Zone I. Maintenance records: Installation date: May 2020; Last calibration date: March 2024; Calibration cycle: 12 months; Historical fault records: Sensor drift fault occurred in August 2023. Quantitative degradation characteristics: sensor degradation coefficient 0.12, historical normal fluctuation range ±2℃, rated refresh cycle 1s.

[0032] 3. Construction of a multi-dimensional health condition benchmark library The operating scenarios were broken down into 864 effective sub-scenarios based on five dimensions. For each scenario, historical operating data of the unit's equipment health, no alarms, and no operations over the past two years were extracted. The normal value range, rate of change limit, and linkage deviation threshold of each measuring point were automatically calibrated. Example: In the scenario of #1 unit generating mode, load 80%-85%, head 200-202m, ambient temperature 20-25℃, and stable operation period, the normal value range of water guide bearing temperature is 35-55℃, and the maximum rate of change is ≤2℃ / min.

[0033] S2, Pre-verification of Data Acquisition and Transmission Link The data platform collects data from all measurement points in real time at a frequency of 1 second, and first performs a link pre-verification: 1. Rapid Link Consistency Verification: For the temperature measurement point 1 of the water guide bearing of Unit #1, the measured value is 45℃ without change for 60 consecutive cycles (60s), triggering a rapid verification alarm. It is determined to be an explicit communication interruption fault, directly classified as data anomaly, and pushes a data acquisition link maintenance alarm.

[0034] 2. Comprehensive silence index verification: For the oil level measuring point of the main transformer of Unit #2, the monitoring window is preset to 5 minutes. The calculated silence factors are 0.9 for switch quantity, 0.85 for status quantity, and 0.92 for analog quantity. The comprehensive silence index is 0.89, which is greater than the preset threshold of 0.8 and lasts for more than 10 minutes. It is judged as a hidden transmission anomaly and classified as a data anomaly.

[0035] 3. Synchronization verification: The time scale of all measuring points in the LCU cabinet of Unit #3 deviated from the system clock by more than 10 seconds, and the values ​​of all measuring points were stuck synchronously. This was determined to be a batch failure of the acquisition device and classified as data abnormality.

[0036] S3, Sub-condition Matching and Interference Shielding For test points that have passed the link verification, perform operating condition matching: Extract the operating parameters of Unit #1 at the current moment: operating mode is power generation, load rate is 82%, operating head is 201m, ambient temperature is 22℃, and the equipment is in a stable operating period; By matching the corresponding sub-operating scenarios in the benchmark library using the cosine similarity algorithm, the normal temperature range of the water guide bearing is locked at 35-55℃, with a change rate limit of 2℃ / min. The 30-minute transient window for the start-up, shutdown, and operation mode switching of the shielded unit only performs subsequent judgments on data under stable operating conditions.

[0037] S4. Preliminary Judgment of Differential Anomalies (Step S4) Based on the number of associated measurement points, indicators are categorized into L1 and L2 classes, and initial judgments are performed for each. Example 1: L1 index discrimination (#1 turbine bearing temperature index)

[0038] This indicator is associated with a total of 6 temperature measuring points, including the upper conductor, lower conductor, and water conductor. The number of measuring points is greater than 2, so it belongs to the L1 category of indicators.

[0039] Real-time data shows that the temperature of the #1 unit water guide bearing at measuring point 1 is 72℃, which exceeds the reference range of 35-55℃, triggering an abnormality warning. Z-score standardization was performed on the real-time data of 6 bearing temperature measurement points, and the covariance matrix was constructed to calculate the Mahalanobis distance. The results showed that only the Mahalanobis distance of the water guide bearing temperature measurement point 1 was 12.6, which was much greater than the preset threshold of 3. The Mahalanobis distances of the other 5 measurement points were all less than 1, which were isolated outliers. Preliminary assessment: Single-point data anomaly. The assessment process is terminated, and a sensor calibration / replacement alarm is pushed.

[0040] Example 2: L2 Index Judgment (#2 Main Transformer Oil Level Measurement Point)

[0041] This indicator has only one measurement point and belongs to the L2 category.

[0042] Real-time data shows that the oil level of the main transformer of Unit #2 has been slowly decreasing for 72 hours, from 1.2m to 0.8m, without any instantaneous jumps, showing a continuous monotonic downward trend; Time-series trend comparison: Matching historical health data under the same operating conditions, the normal fluctuation range of oil level is ±0.1m / 72h. The current decline is far greater than the normal fluctuation and there is no reverse regression trend. Preliminary assessment: Non-transient interference data anomaly, proceeding to subsequent mechanism linkage verification steps.

[0043] S5, Cross-parameter strong correlation mechanism linkage verification In response to the abnormal oil level in the main transformer of Unit #2 in Example 2, a mechanism linkage verification was performed: Retrieve the strongly correlated parameters corresponding to the main transformer oil level from the rule base: load rate, ambient temperature, cooler operating status, main body leakage alarm signal, gas relay signal, oil temperature / winding temperature; Verification results: The load rate of the #2 main transformer has been stable at 60%-70% for the past 72 hours, the ambient temperature has been stable at 20-25℃, the cooler is operating normally, the oil temperature / winding temperature has been stable at 45-50℃ without any abnormalities, and there is no gas relay activation signal, but the main body leakage alarm continues to be triggered. Logical matching: The continuous drop in oil level + leakage alarm + stable oil temperature fully conforms to the fault evolution logic of leakage in the main transformer body and is consistent with the physical laws of the equipment. Validation result: No data anomaly. Proceed to the next cross-validation step.

[0044] If the verification result is: the oil level continues to drop, there is no leakage alarm, the oil temperature, ambient temperature, and load rate are completely stable, there is no synchronous change of any strongly correlated parameters, and it does not conform to the physical mechanism of oil level change, then it is directly judged as data abnormal (oil level sensor failure).

[0045] S6. Multi-model cross-validation and confidence leveling Cross-validation was performed to address the abnormal oil level in the main transformer of Unit #2 in Example 2: Validation of the time series prediction model: Based on the LSTM model and combined with the current operating conditions, the normal range of the oil level is predicted to be 1.1-1.3m. The current value of 0.8m is far beyond the prediction range, and the model predicts that the oil level will continue to decline in the future, which is consistent with the trend of equipment failure evolution. Equipment deterioration coefficient matching verification: The main transformer of Unit #2 has been in operation for 8 years and is nearing its overhaul cycle. The comprehensive deterioration coefficient is 0.28, and the oil level decline trend matches the overall equipment deterioration trend. Horizontal comparison verification with similar equipment: Under the same operating conditions, the oil levels of the main transformers of the same model in the same power station (#1, #3, and #4) were all stable at 1.1-1.3m, except for the #2 unit where the oil level continued to drop; Cross-validation results: All validation items point to abnormal device status, with a confidence level of 95%, which is high confidence. Final assessment result: Equipment status abnormal (main transformer leakage fault), high-priority special inspection and maintenance work order is pushed.

[0046] S7, Closed-loop self-learning optimization Maintenance personnel conducted an on-site inspection and confirmed that the leak was caused by a faulty oil drain valve on the main transformer of Unit #2, which was consistent with the initial assessment. The inspection results were then fed back to the system. Update the equipment mechanism rule base and add fault evolution logic for leakage of the main transformer oil drain valve; After the overhaul is completed, the health condition benchmark library for the main transformer oil level is updated based on the new health data. Update the health record of the oil level measuring point of the main transformer of Unit #2 and supplement the maintenance record; The model threshold was optimized by adjusting the limit of the main transformer oil level change rate from ±0.1m / 72h to ±0.08m / 72h, thereby improving the discrimination sensitivity.

[0047] S8, Special Scene Adaptation For unit maintenance and testing scenarios, the system automatically sets Unit #3 to maintenance status based on the maintenance work order, and blocks the abnormality detection of all measuring points of the unit to avoid human operation data during maintenance and testing being misjudged as abnormal; for extreme head fluctuation scenarios during the flood season, the system automatically activates the special benchmark for the flood season to avoid normal parameter fluctuations caused by large changes in head being misjudged.

[0048] Finally, it should be noted that the above-described embodiments are merely specific implementations of the present invention, used to illustrate the technical solutions of the present invention, and not to limit it. The scope of protection of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments within the technical scope disclosed in the present invention, or make equivalent substitutions for some of the technical features; and these modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for distinguishing between abnormal power plant data and abnormal equipment status, characterized in that, Includes the following steps: S1. Construct a three-in-one basic support system for judgment by establishing a physical mechanism of equipment, a parameter linkage rule library, a full life-cycle health record of measurement points, and a multi-dimensional health condition benchmark library; S2. Real-time data and related operating parameters of the measurement points to be judged are collected in real time, and the three-layer pre-verification of the data acquisition and transmission link is performed to screen out abnormal data due to link failure. S3. Based on a multi-dimensional health condition benchmark library, match the current subdivided operating condition scenario, lock the corresponding benchmark threshold, and shield the interference of transient fluctuations in operating conditions. S4. Based on the number of measurement points associated with the indicators, the indicators to be judged are divided into L1 multi-measurement point redundant indicators and L2 single / double measurement point indicators. Differentiated anomaly initial judgment is performed on each, and single-point data anomalies and transient interference data anomalies are quickly identified. S5. For measurement points that are initially determined to be non-data abnormal by step S4, retrieve the equipment physical mechanism and parameter linkage rule library, perform cross-parameter strong correlation mechanism linkage verification, and determine whether the measurement point abnormality conforms to the equipment physical evolution law. S6. For anomalies verified in step S5, perform multi-model cross-validation to generate the final judgment result and confidence rating, thus distinguishing between data anomalies and equipment status anomalies.

2. The method according to claim 1, characterized in that, The equipment physical mechanism and parameter linkage rule library described in S1 constructs a four-dimensional rule set for each type of core equipment in the power plant, which includes core parameters, strongly correlated parameters, physical constraint relationships, and fault evolution logic, and adds operating condition adaptation tags to each rule; the full life cycle health record of the measuring point includes three types of information: basic attributes, operation and maintenance records, and quantitative deterioration characteristics of each measuring point. The multi-dimensional health condition benchmark library is divided into subdivided operating condition scenarios according to operating mode, load rate range, working head range, ambient temperature range, and equipment operation stage. For each scenario, the normal value range, change rate limit, and parameter linkage deviation threshold of the corresponding measuring point are automatically calibrated.

3. The method according to claim 1, characterized in that, The pre-verification of the data acquisition and transmission link mentioned in S2 specifically includes: S21 Fast Link Consistency Check: Determines whether the measured value of the current measurement point remains unchanged within a continuous period, and identifies explicit communication interruption faults; S22 Comprehensive Quiet Index Verification: Calculate the quiet factor for switch quantities, status quantities, and analog quantities respectively, and obtain the comprehensive quiet index by weighting them to identify hidden transmission anomalies where the link is connected but the data is not refreshed. S23 Full-Link Synchronization Verification: Based on system clock synchronization, it verifies the deviation between the measurement point time stamp and the system clock, checks whether the measurement points on the same acquisition link / device are abnormally synchronized, and locates batch failures of acquisition units; if any of the above verifications are abnormal, it is directly determined to be a data anomaly of link failure type, and the judgment process ends.

4. The method according to claim 1, characterized in that, S3 specifically includes: S31. Extract the current operating mode, load rate, working head, equipment operating status, and ambient temperature core operating parameters. S32. Match the optimal subdivided working condition scenario in the multi-dimensional health working condition benchmark library using the cosine similarity algorithm, and lock the corresponding benchmark threshold under the scenario. S33. Set a transient shielding window to shield the transient processes of unit start-up, shutdown, operating condition switching, and maintenance tests, and perform subsequent judgments only after the operating conditions stabilize.

5. The method according to claim 1, characterized in that, S4 The L1 category index is an index that is associated with more than 2 measurement points, and the L2 category index is an index that is associated with 2 or less measurement points. The consistency verification of multiple measurement points of the same L1 type index is specifically as follows: Z-score standardization is performed on the real-time data of all measurement points under the same index, the matrix between measurement points is constructed to calculate the Mahalanobis distance, and outlier measurement points are screened; if only a single measurement point is an isolated outlier point, and the other measurement points of the same index are stable within the benchmark range, it is determined to be a single point data anomaly. The time-series trend rationality verification of the L2 type indicators is specifically as follows: extract continuous time-series data of abnormal measurement points and compare them with historical healthy time-series data under the matching working conditions. If the abnormality is an instantaneous and irregular jump, and the jump quickly returns to the baseline range without a continuous evolution trend, it is determined to be an instantaneous interference type of data abnormality.

6. The method according to claim 1, characterized in that, The cross-parameter strong correlation mechanism linkage verification described in S5 specifically refers to: Retrieve all strongly correlated parameters corresponding to abnormal measurement points from the equipment physical mechanism and parameter linkage rule base, and verify whether the strongly correlated parameters show synchronous changes in the same direction as the abnormal measurement points and conform to the physical constraint relationship; If the changes at abnormal measurement points are synchronously matched with strongly correlated parameters and conform to the equipment fault evolution logic, it is determined to conform to physical laws and proceeds to the subsequent cross-validation steps. If the changes at abnormal measurement points do not have a corresponding strong correlation parameter change synchronously and are completely inconsistent with the physical mechanism of the equipment, the data is judged to be abnormal and the judgment process ends.

7. The method according to claim 1, characterized in that, The multi-model cross-validation mentioned in S6 includes time-series prediction model validation, equipment degradation coefficient matching validation, and horizontal comparison validation of similar equipment; the confidence level is divided into three levels: high confidence, medium confidence, and low confidence, which are respectively matched with the corresponding graded operation and maintenance handling process.

8. The method according to claim 1, characterized in that, It also includes a closed-loop self-learning optimization step: feeding back the abnormal results confirmed by the operation and maintenance personnel on-site to the system, updating the equipment physical mechanism and parameter linkage rule library, multi-dimensional health condition benchmark library, and full life cycle health record of measurement points, while optimizing the core parameters of the discrimination model to achieve continuous iteration of discrimination capabilities.

9. The method according to claim 1, characterized in that, It also includes special scenario adaptation steps: for transient operating conditions of the unit, maintenance and testing scenarios, and extreme hydrological / weather scenarios, special shielding rules and scenario-based adaptation benchmarks are set to avoid normal fluctuations in the scenario being misjudged as abnormal.

10. A system for distinguishing between power plant data anomalies and equipment status anomalies, characterized in that, include: The basic library construction module is used to execute the construction of the three-in-one discrimination basic support system as described in claim 2; The link pre-verification module is used to perform the three-layer acquisition and transmission link verification as described in claim 3; The system includes a working condition matching and threshold locking module for performing the detailed working condition matching and transient masking as described in claim 4; a differential initial judgment module for performing the initial judgment of L1 / L2 class index anomalies as described in claim 5; a mechanism linkage verification module for performing the cross-parameter strong correlation verification as described in claim 6; and a cross-validation and discriminant output module for performing the multi-model verification and confidence rating as described in claim 7. The system is used to perform the differentiation method described in any one of claims 1 to 9.

Images