Hydroelectric equipment library intelligent classification and dynamic updating method and system based on deep neural network

Abstract

The invention discloses a deep neural network-based hydroelectric equipment library intelligent classification and dynamic updating method and system, and relates to the technical field of industrial internet platforms and hydraulic engineering monitoring. The method comprises the steps of mathematical model establishment and multi-dimensional discrimination method integration, monitoring data preliminary screening and effectiveness preliminary discrimination, monitoring data retest and secondary verification, test point importance grading and multi-physical quantity joint discrimination, key test point continuous tracking and field comparison test, inspection result feedback and defect management, and manual recheck driving model optimization. And setting a measuring point state. According to the invention, through accurate data validity identification and efficient processing, monitoring equipment faults and dam potential hidden dangers can be found in advance, safety accident risks caused by data distortion are reduced, manual inspection frequency is reduced through an automatic process, the equipment monitoring full life cycle cost is reduced, and the system has high practicability. And core technical support is provided for intelligent management of water conservancy projects.

Description

TECHNICAL FIELD

[0001] The application relates to the fields of industrial internet platforms and water conservancy engineering monitoring technology, and in particular to a method and system for intelligent classification and dynamic updating of a hydropower equipment library based on a deep neural network. BACKGROUND

[0002] In the fields of industrial internet platforms and water conservancy engineering safety monitoring technology, the following defects exist in the effectiveness identification of traditional dam monitoring data: A single discriminant model is difficult to adapt to the differences in characteristics of different monitoring projects, resulting in a high misjudgment rate; The invalid data processing flow is fixed and rigid, and a unified retest flow is used for all invalid data, lacking a hierarchical disposal mechanism based on the importance of the measuring point, and the disposal efficiency is low; There is a lack of a continuous tracking mechanism for the state of the measuring point, which cannot dynamically reflect the state evolution of the measuring point, and it is difficult to realize the whole life cycle management of data quality.

[0003] In addition, the massive monitoring data generated during the operation of the hydropower station equipment and facilities are influenced by factors such as instrument failure, system error, and environmental interference, and contain a large amount of invalid data. If directly used for safety analysis, it will lead to decision bias. Therefore, an intelligent identification method that integrates multiple discriminant methods and dynamically manages is urgently needed to improve the data quality and the intelligent level of the monitoring system. SUMMARY

[0004] The purpose of the present application is to provide a method and system for intelligent classification and dynamic updating of a hydropower equipment library based on a deep neural network to solve the problems raised in the background art.

[0005] To achieve the above purpose, the present application provides the following technical solution: a method and system for intelligent classification and dynamic updating of a hydropower equipment library based on a deep neural network, comprising the following steps: S1, comprehensive discriminant model construction: construct a mathematical model integrating logical discriminant method, statistical discriminant method and clustering analysis method, combine data time sequence characteristics, physical quantity inherent characteristics and measuring point importance, and match the identification model for different physical quantities and measuring points; S2, monitoring data preliminary screening: automatically pull the new data of the previous 24 hours every day and store them by measuring point ID, after completing the missing data processing, call the model constructed in S1 to preliminarily identify the data effectiveness, store the valid data directly, and the invalid data enters the next step; S3, data retest verification: push the retest instruction to the station, and call the S1 model again based on the retest data; if the retest is valid, record the gross error and end the process, and the invalid data enters the next step; S4, point classification and joint discrimination: read the preset important measurement point list, and directly touch the scene comparison measurement instruction for important measurement points; for general measurement points, the same dam section and the same type of measurement point data are combined to determine the risk classification; S5, key measurement point tracking and comparison: low-risk measurement points are included in the attention list and the monitoring frequency is increased, and after the period expires, the S1 model is called for re-determination, and if there is an exception, the scene comparison measurement is triggered, and instrument failure prompts or inspection instructions are pushed synchronously; S6, inspection result feedback: if the inspection confirms the defect, a repair suggestion is generated and included in the defect management, and if there is no exception, the corresponding part of the inspection plan is adjusted; S7, model dynamic optimization: periodically extract historical invalid data cases for manual review, and according to the gross error identification rate and the abnormal error rate, re-match the algorithm for the model that is not suitable; S8, measurement point state control: mark the measurement point as normal, retest, attention, fault, and verification five states, and when starting data checking, access the corresponding processing branch of S2 to S6 according to the corresponding state.

[0006] Further, in the above step S2, the completion processing method for missing data is: when the missing value is ≤3 data points, linear interpolation method is used for completion, and when the missing value is >3 data points, historical same period average method is used for completion; in the above step S3, if the response is not received within 2 hours after pushing the retest instruction, the instruction is automatically upgraded to the upper management node; in the above step S4, the risk classification standard of general measurement points is: ≥80% of the same type / same part measurement point data in the same dam section is qualified, which is determined as low risk, and ≥50% of the same type / same part measurement point data is abnormal, which is determined as high risk; in the above step S5, the measurement points included in the above attention list are monitored at a frequency of 1 time / 2 hours, and the monitoring period is 5 days; in the above step S6, the adjustment method of the inspection plan when there is no exception is: the corresponding part / monitoring quantity is supplemented to the monthly inspection plan, and the inspection frequency is increased to 2 times / month; in the above step S7, the specific parameters of periodically extracting historical invalid data cases for manual review are: 10% of the invalid data cases of the last month are extracted every month, and the gross error identification rate ≥95% and the real abnormal error rate ≤5% are set as the model adaptability determination threshold; in the above step S8, the trigger logic of the measurement point state is: the normal measurement point directly enters the S2 preliminary screening process, the verification state measurement point is directly pushed to the abnormal identification module, and the retest, attention, and fault state measurement points are respectively accessed to S3 retest, S5 tracking, and S6 defect management branch.

[0007] Further, in the above S1 step, the logical discrimination method specifically includes: based on the monitoring instrument range, the data exceeding the limit is removed for the instrument range, for the current measurement value, , The data exceeding the preset threshold value is removed based on engineering specifications.

[0008] Further, in the S1 step, the statistical discrimination method specifically includes: selecting historical data (sample size ≥ 30 groups) of similar working conditions in the last three hydrological years, taking the water level , temperature , time effect as influencing factors to establish a multiple regression equation . is the theoretical displacement, and is the regression coefficient, and the residual error and the residual standard deviation are calculated. ; If the residual error exceeds 6 times the standard deviation , it is determined as invalid data.

[0009] Further, in the S1 step, the clustering analysis method specifically includes: using K-means clustering, the number of clusters , distinguishing normal clusters / abnormal clusters, and calculating the Euclidean distance of each data point to the center of the normal cluster : ; If the distance exceeds 2 times the average distance of the normal cluster : ; Then it is determined as an outlier gross error and triggers S3.

[0010] Further, in the S2 step, in combination with the time sequence characteristics of the data, the inherent characteristics of the physical quantity and the importance of the measuring point, the identification model matched and adapted for different physical quantities and measuring points is specifically: a multi-factor coupled physical quantity matching statistical discrimination method + a logic discrimination method for displacement, stress and the like, a logic discrimination method for physical meaning single quantity such as seepage flow and water level, and a forced superposition clustering analysis method for collaborative data of multiple measuring points in the same dam section.

[0011] Further, in the S4 step, the difference between the statistical discrimination threshold values of important measuring points and general measuring points is specifically: Important measuring points: the statistical discrimination threshold value is tightened to 4 times the standard deviation ; When the condition is met, the on-site ratio measurement instruction is directly triggered; General measuring points: the statistical discrimination threshold value is relaxed to 8 times the standard deviation, and only when the residual error is 6 ; When the interval, i.e., exceeding the conventional 6σ threshold but not reaching the 8σ extreme value, starts the joint discrimination of multiple physical quantities of the same dam section to exclude the error interference of a single measuring point.

[0012] Further, in the S8 step, the branch control details of the five states further include: the retest state measuring point preferentially performs the S3 retest process, the attention state measuring point continuously performs the 1 time / 2 hours high-frequency monitoring of S5, the fault state measuring point does not collect new data before being repaired, and the verification state measuring point skips S2-S7 and directly pushes the abnormality identification module.

[0013] Further, in the S7 step, the model optimization iteration rule is specifically: when the algorithm library is called to re-match the marked unsuitable model after rechecking, the original model can be replaced only when the backtest error identification rate of the historical data backtest is greater than or equal to 96%, and the optimization result is archived to the model-measuring point matching rule library.

[0014] Further, in the S5 step, the determination standard of the field comparison is specifically: using the standard instrument authenticated by measurement, if the difference between the monitoring values of the standard instrument and the original instrument is greater than 2 times the precision of the original instrument, it is determined that the instrument is faulty; if the difference is less than or equal to 2 times the precision of the original instrument, it is determined that the monitoring value is a real abnormality.

[0015] Compared with the prior art, the beneficial effects of the present application are: 1. Improve data identification accuracy and reliability: integrate logical discrimination, statistical discrimination and clustering analysis and other mathematical models to form a multi-level discrimination system, effectively identify invalid data such as exceeding the instrument range, sign error, statistical anomaly and outlier gross error, avoid the misjudgment risk of a single model, greatly improve the data identification accuracy, dynamically match the discrimination model according to the importance of the measuring point and the time sequence characteristics of the physical quantity, and effectively distinguish between instrument failure and real data anomaly through multi-measuring point and multi-physical quantity joint analysis, reduce the misjudgment caused by single measuring point data fluctuation, especially suitable for complex monitoring scenes of water mechanical and electrical equipment in multiple working conditions and multiple physical field coupling; 2. Improve monitoring efficiency and response speed: classify the measuring points according to importance to avoid "one-size-fits-all" disposal, reduce unnecessary field intervention through correlation analysis for general measuring points, improve monitoring disposal efficiency and reduce labor costs. From data preliminary screening, retest verification to hierarchical disposal, field comparison, the whole process realizes the closed-loop management of "model discrimination-automatic shunting-instruction pushing", compared with the traditional manual inspection mode, the abnormal response time is shortened from hours to minutes, ensuring timely risk disposal; 3. Construct a self-evolution system to realize long-term accurate monitoring: Through regular manual verification of data effectiveness, the automatic labeling of inapplicable data is pushed and a message is sent, and the algorithm is matched again to form a self-evolution closed loop of "determination - review - optimization". The state of the measuring point is divided into "normal, retest, attention, fault, and verification" five categories, combined with the feedback of the inspection results (such as crack and water seepage defect repair), to realize the seamless linkage of monitoring data anomaly and engineering maintenance, ensure the two-way tracking of data anomaly and engineering hidden danger, and improve the systematization of dam safety management; 4. Improve the intelligent and standardized level: unify the definition of measuring point state and disposal rules, provide a standard framework for the automation programming and modular development of the monitoring system, can be quickly integrated into the dam safety monitoring system, realize the digitalization of the whole process of data processing, abnormal early warning, and maintenance scheduling, reduce the development and maintenance cost of the system. Continuous high-frequency monitoring of key measuring points and on-site instrument comparison can accurately identify the essential cause of data anomaly, avoid missing or false alarm of safety hidden danger caused by data misjudgment, and provide more reliable decision basis for safe operation of equipment and facilities; 5. Through accurate data effectiveness identification and efficient disposal, the present application can discover monitoring equipment failure and potential dam hidden danger in advance, reduce the risk of safety accidents caused by data distortion, and reduce the frequency of manual inspection through automatic process, reduce the life cycle cost of equipment monitoring, and provide core technical support for intelligent management of water conservancy projects. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 The flowchart of the method and system for intelligent classification and dynamic updating of the water and electricity equipment library based on the deep neural network provided by the embodiment of the present application; Figure 2 The construction method schematic diagram of the multi-dimensional determination model of the method and system for intelligent classification and dynamic updating of the water and electricity equipment library based on the deep neural network provided by the embodiment of the present application; Figure 3 The artificial review optimization flowchart of the method and system for intelligent classification and dynamic updating of the water and electricity equipment library based on the deep neural network provided by the embodiment of the present application; Figure 4 The measuring point state conversion mechanism schematic diagram of the method and system for intelligent classification and dynamic updating of the water and electricity equipment library based on the deep neural network provided by the embodiment of the present application; DETAILED DESCRIPTION The combination mode described can be freely combined or combined in any way, unless contradictory to each other.

[0017] It should be noted that the above enumeration is only the specific implementation of the present application. The technical solutions in the embodiments of the present application will be described clearly and completely in conjunction with the drawings of the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor belong to the scope of protection of the present application.

[0018] I. Working principle: The present application will be further described in detail below in conjunction with specific embodiments: The intelligent classification and dynamic updating method and system for the hydropower equipment library based on deep neural network include mathematical model establishment and multi-dimensional discrimination method integration, monitoring data preliminary screening and effectiveness preliminary discrimination, monitoring data retest and secondary verification, measurement point importance grading and multi-physical quantity joint discrimination, key measurement point continuous tracking and field comparison, inspection result feedback and defect management, artificial review driven model optimization, and measurement point state setting.

[0019] Reference Figure 2 , the mathematical model establishment and multi-dimensional discrimination method integration, that is, the mathematical model including logical discrimination, statistical discrimination and clustering analysis is constructed for monitoring data effectiveness discrimination; Specifically, the logical discrimination method judges the data effectiveness according to the monitoring instrument range, monitoring accuracy and physical meaning of the monitoring data; the statistical discrimination method selects the historical monitoring data under similar working conditions to establish a multiple regression model to predict the expected range of the current measurement value, and if the monitoring effect quantity and the measured value deviation exceeds the range with 6 times the standard deviation as the threshold, it is determined as invalid; the clustering analysis method detects outliers for time series data of the same measurement point, and marks the isolated points as gross errors, and simultaneously performs multi-dimensional joint clustering combined with external variables such as temperature and water level to avoid misjudgment caused by environmental interference; according to the time sequence characteristics of the data, different physical quantities and different measurement points are matched with the appropriate identification model to ensure the applicability of the model, such as displacement monitoring data, which is characterized by "being influenced by multiple factors cooperatively": water level change, temperature fluctuation, time-dependent deformation, etc.; the data presents "trend + periodicity" characteristics, and the statistical discrimination model is preferred; for seepage flow / seepage pressure data of dam foundation seepage flow and around dam seepage pressure, the characteristic is "strongly related to water level", that is, the water level rises usually with the increase of seepage pressure, but the physical meaning is clear, such as the seepage flow cannot be negative, and the data mutation is caused by instrument failure, so the logical discrimination model is preferred; for temperature monitoring data, the characteristic is "affected by environmental temperature, hydration heat, etc., with diurnal / seasonal periodicity", and the instrument accuracy is high, with error usually ≤0.1℃, so the statistical discrimination model is preferred.

[0020] Specifically, the establishment of each mathematical model: • Core judgment rule of logical discrimination model Range over-limit discrimination: For any monitored physical quantity, such as displacement , osmotic pressure , temperature , if it exceeds the factory range of the instrument , it is directly determined as invalid: or

[0021] is the current detection value, , is the minimum / maximum range of the instrument, which can be obtained from the device parameter table; • Quantitative calculation of statistical discrimination model Select historical monitoring data under similar working conditions to establish a multiple regression model to predict the expected range of the current measurement value, and use "6 times the standard deviation" as the extreme abnormal threshold, and the specific calculation is as follows: Multiple regression benchmark model (taking dam body horizontal displacement as an example): For displacement affected by multiple factors, water level , temperature , time , establish a regression equation: ; Among them, is the theoretically calculated displacement, ~ is the regression coefficient, which is obtained by least squares fitting of historical effective data for nearly 3 hydrological years, and the sample size should be ≥ 30 groups; Residual and standard deviation calculation: calculate the residual of the measured displacement and the theoretical displacement , and calculate the standard deviation based on the historical residual ; Among them, is the number of historical residual samples; Abnormality determination rule: if the absolute value of the residual exceeds 6 times the standard deviation, it is determined as invalid data (gross error); if it exceeds 3 times but does not exceed 6 times the standard deviation, it is marked as data to be retested, and the formula is: ; In the scheme, the standard deviation of the residual of the monitoring data is specifically referred to.

[0022] • Cluster analysis model construction K-means clustering of the same period data of the same dam section, same type of measuring point , distinguish "normal cluster" and "abnormal cluster", and identify outliers and gross errors through distance calculation:​ Euclidean distance calculation: assuming that there are m seepage pressure gauges in the same dam section, and the same period monitoring data are ; the Euclidean distance of each data point to the center of the normal cluster is calculated: ; wherein, is the mean value of all data in the normal cluster.

[0023] Outlier determination threshold: the average distance of all data in the normal cluster is calculated , and if the distance of a data point exceeds 2 times the average distance, it is determined to be invalid data: ; wherein, is the number of data points in the normal cluster.

[0024] The monitoring data preliminary screening and preliminary validity identification is that the newly generated monitoring data of each day is preliminarily screened by a mathematical model, the effective data exits the validity identification process, and the invalid data enters the retest process.

[0025] Specifically, when performing preliminary validity identification, the new data of the previous 24 hours is automatically pulled each day and stored in a temporary database according to the measurement point ID, and linear interpolation or historical mean value is used to fill in the missing values to ensure data continuity; the system automatically matches the preset model according to the measurement point attributes, such as type, location, and physical quantity, and performs validity identification through the matched model to trigger the subsequent retest process; the model determines that the data is invalid to trigger the flag; after processing, the effective data exits the data validity identification process, and the invalid data enters the monitoring data retest and secondary verification.

[0026] The monitoring data retest and secondary verification is that the invalid measurement point retest reminder is triggered, and if it is still invalid after retest, it enters the multi-physical quantity or on-site inspection process.

[0027] Specifically, during retest and secondary identification, the system sends a retest instruction to the station side, and automatically upgrades to the superior management department if it does not respond after a timeout; the station side uploads the data after retest, and the system automatically performs data identification; if the retest value is valid, the original invalid data is marked as "gross error" and recorded, and if the retest is still invalid, it is pushed to the measurement point importance classification and multi-physical quantity joint identification.

[0028] The measurement point importance classification and multi-physical quantity joint identification is to distinguish important measurement points and general measurement points, and to dispose them according to the risk level.

[0029] Specifically, important measurement points and general measurement points are distinguished, and disposed according to the risk level, as follows: Measurement point type and importance determination: Read the preset "important measurement point list", which includes key measurement points such as dam crest plumb line, dam foundation water measuring weir, and spillway stress gauge, and the remaining measurement points are classified as general measurement points.

[0030] Specifically, when risk levels are stratified, the list of important monitoring points is first read. According to the stratification logic, on-site comparison instructions are directly sent to the power plant for important monitoring points. For general monitoring points, multi-monitoring and multi-physical quantity discrimination is initiated. Spatially, all monitoring point data within the same dam section are searched, and related monitoring quantities are checked in terms of physical quantities. According to the risk judgment rules, if the surrounding monitoring points are normal, they are judged as low risk, added to the attention list, and the monitoring frequency is increased. They are then continuously tracked and compared on-site at key monitoring points. If similar monitoring points are abnormal, they are judged as high risk, and on-site comparison instructions are immediately sent to the power plant.

[0031] Key measurement points are continuously tracked and compared on-site. The measurement points on the list of concerns are continuously tracked for 5 days and evaluated again using mathematical models. If the problem is not resolved, instrument checks are initiated. If the true value is confirmed, the process moves to anomaly identification.

[0032] Specifically, for the monitoring points listed in the key attention list, a 5-day closed-loop monitoring mechanism is implemented: On the 5th day, the system automatically calls the discrimination model to perform a secondary validity assessment of the measured values. If the measured values ​​return to normal, the previous gross error data is saved, and the analysis process exits. If they are still unqualified, an instruction is issued to the plant / station side, reminding staff to conduct on-site inspection and comparison testing of the monitoring instruments. After on-site comparison testing, if the comparison data deviation is large, it is determined to be an instrument malfunction; if the comparison data match, the measured value is determined to be a genuine anomaly, and the data is automatically pushed to the anomaly identification module for analysis. At the same time, an inspection instruction is issued to the plant / station side, and the inspection result feedback and defect management process begins.

[0033] Inspection result feedback and defect management, that is, to handle defects or incorporate them into the inspection management module based on the inspection results, and to retrieve the gross error data of the cusp.

[0034] Specifically, during on-site inspections and defect handling, the inspection personnel will provide feedback on the inspection results. If the inspection results do indeed show defects such as cracks or water seepage, repair suggestions will be given and the defect will be included in the defect management. If the inspection results show no abnormalities, the location or monitoring quantity will be added as an inspection location or inspection content and included in the inspection management module.

[0035] Reference Figure 3 Manual review drives model optimization, which means regularly reviewing the model's performance manually, marking inapplicable methods, and re-matching algorithms.

[0036] Specifically, when the model review and optimization is carried out, a certain amount of invalid data is extracted every month to determine the case for manual review; according to the manual review results, the data model method used in the validity discrimination is checked, if an inappropriate model is used, it is automatically marked and a message is pushed, and the algorithm library is called again to match the algorithm, and the model optimization iteration is completed by selecting the model algorithm version with lower misjudgment rate.

[0037] Referring to Figure 4 , the measurement point state setting marks the measurement point as 5 states: normal, retest, attention, fault, and verification, and the processing flow is divided according to the state.

[0038] Specifically, the monitoring point is marked as 5 states according to the validity recognition result of the measured value, as follows: Normal - measurement point discriminated by mathematical model; Re-test - measurement point waiting for retest result which does not pass mathematical model discrimination; Attention - retest value of the measurement point is still unqualified, but no major risk is identified by multiple measurement points and multiple physical quantity discrimination; Fault - the measured value cannot pass the validity check continuously, and the instrument is determined to be faulty after on-site comparison; Verification - the measured value cannot pass the validity check continuously, but the measured value is determined to be the true value after on-site comparison.

[0039] Each time the monitoring data check is started, for the measurement point in normal state, it starts from the beginning and enters the data analysis flow, for the measurement point in verification state, the measured value can directly enter the abnormal identification link, and for the measurement points in other states, the branch flow according to their respective states is processed.

[0040] II. Example Application of data validity identification of a concrete gravity dam (dam height 100m) 1. Basic parameter setting Monitoring object: The dam 1# section contains 3 important measurement points (dam crest vertical displacement meter W1, dam foundation water weir seepage flow meter S1, spillway stress meter Y0) and 5 general measurement points (dam shoulder stress meter Y1-Y5, dam foundation seepage pressure meter P1-P3); Historical data processing: select the historical data of 2022-2024 (3 hydrological years) (n=1095 groups), and establish a multiple regression model for W1 displacement data: ; Calculate the residual standard deviation = 0.25mm ; Threshold application: Important measurement point W1: 4 = 4 x 0.25 = 1.0 mm, when > 1.0 mm trigger direct trigger than the measurement; General test point Y1: 8 = 8 x 0.25 = 2.0 mm, when

[0041] start multi-physical quantity evaluation.

[0042] 2. Monitoring data preliminary screening Pull W1 24-hour data on August 10, 2025, find 3 consecutive missing data points: 8:00-10:00, fill in with linear interpolation; find 5 consecutive missing data points of P1: 12:00-16:00, fill in with 2024 same period: August 10, 12:00-16:00 average 0.32 MPa; After logical discrimination: P1 seepage pressure has no negative sign, not out of range 0-1.6 MPa and statistical discrimination W1 residual = 0.8 mm≤4 , the proportion of valid data in preliminary screening is 92%.

[0043] 3. Test point grading discrimination Perform "1 time / 2 hours" high-frequency monitoring on Y1 (12 groups of data per day) for 5 days (August 11-15): General test point Y1: residual = 1.8 mm (satisfies 1.5 mm < 1.8 mm ≤ 2.0 mm), retrieve the same dam section Y2-Y5 (residuals are 0.9 mm, 1.2 mm, 1.1 mm, 1.3 mm, all ≤3 = 0.75 mm, here correction: 3 = 0.75 mm, Y2-Y5 residual ≤0.75 mm, the proportion of qualified test points is 5 / 5 = 100% ≥80%), determine as low risk, included in the list of attention; General test point P2: residual = 2.1 mm > 2.0 mm, retrieve P1, P3: residuals are 2.2 mm, 2.3 mm, the proportion of abnormal test points is 3 / 3 = 100% ≥50%, determine as high risk, directly trigger on-site measurement instruction.

[0044] 4. Key test point tracking Perform "1 time / 2 hours" high-frequency monitoring on Y1: 12 groups of data per day for 5 days: August 11-15: August 11 data: residual average = 1.6 mm; August 15 summary: 5-day residual mean ≤1.5mm (6 ), determine "return to normal", stop high-frequency monitoring, and resume "1 time / day" regular frequency.

[0045] 5. On-site comparison The P2 was measured by a standard osmotic pressure calibrator with an accuracy of ±0.05MPa: The original instrument measured value = 0.85MPa, the standard instrument measured value = 0.72MPa, the difference = 0.13MPa; The original instrument accuracy = ±0.05MPa, 2 times accuracy = 0.10MPa, because 0.13MPa > 0.10MPa, determine "instrument failure", remind to replace within 24 hours.

[0046] 6. Model optimization Manual review: In August 2025, 10% of invalid data cases in July were extracted: a total of 50 groups, including 48 groups of gross errors and 2 groups of real anomalies. Manual review found that the original model misjudged 2 groups of gross errors, and the gross error identification rate = (48-2) / 48x100%=95.8%=96%, meeting the verification index in S7 "gross error identification rate ≥95%, real anomaly misjudgment rate ≤5%"; Model backtest: based on instrument range, remove out-of-limit data, based on physical meaning, remove sign error data, based on engineering specification, remove data exceeding the preset threshold, backtest 500 cases from January to December 2024: 480 groups of gross errors, gross error types include: 220 groups of instrument range out-of-limit, 160 groups of physical meaning contradiction, such as negative seepage flow, water level sudden drop more than 10m in a single day, 100 groups of exceeding engineering specification threshold, such as seepage flow single-day increase exceeding 30%, dam water level fluctuation exceeding 8m. After identification, 461 groups of known gross errors were correctly identified among the 480 groups, and 19 groups of unidentified gross errors (all near the engineering specification threshold, such as seepage flow single-day increase 29.8%). The gross error identification rate = 461 / 480x100%=96.04%≥96%, meeting the rule "backtest gross error identification rate ≥96% to replace the original model", so the original model is replaced and the specific application standards of logical discrimination method, including the range of each monitoring physical quantity and engineering specification threshold, are archived to "model - measurement point matching rule library".

[0047] 7. Inspection management After replacing the instrument at P2, the seepage pressure gauges P1-P3 of this dam section were supplemented to the monthly inspection plan, and the inspection frequency was increased from "1 time / month" to "2 times / month". The first inspection time was set to August 20.

[0048] Those skilled in the art can clearly understand that the modules and method steps of each example described in combination with the embodiments disclosed herein can be realized by electronic hardware or a combination of computer software and electronic hardware. Whether the functions are realized in hardware or software manner depends on the specific application and design constraints of the technical solution. The skilled person can use different methods to realize the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0049] Those skilled in the art can clearly understand that, for the convenience and brevity of the description, the specific working processes of the above-described devices, apparatuses and modules can refer to the corresponding processes in the foregoing method embodiments, which will not be repeated here.

[0050] In several embodiments provided in the present application, it should be understood that the disclosed devices, apparatuses and methods can be implemented in other ways. For example, the above-described device embodiments are only schematic, for example, the division of the above-described modules is only a logical function division, and actual implementation can have another division manner, for example, a plurality of modules or units can be combined or integrated into another device, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the shown or discussed each other can be indirect coupling or communication connection through some interfaces, devices or apparatuses, which can be electrical, mechanical or other forms.

[0051] The above-described modules described as separate components can or can not be physically separated, and the components shown as modules can or can not be physical units, i.e. can be located in one place, or can be distributed on a plurality of network units. Part or all of the units can be selected according to actual needs to achieve the purpose of the embodiment.

[0052] In addition, each functional module in each embodiment of the present application can be integrated into a processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0053] If the above functions are realized in the form of software function modules and sold or used as independent products, they can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application or parts of the present application that essentially contribute to the prior art or the technical solutions of the present application can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a number of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the above-mentioned method steps of the embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory server, a random access memory server, a magnetic disk or an optical disk, and various storage media that can store program instructions.

[0054] In addition, it should be noted that the combination of various technical features in the present case is not limited to the combination or specific embodiments described in the claims of the present case. Obviously, the present application is not limited to the above embodiments, and there are many similar changes. All modifications directly derived or inferred from the disclosure of the present application by those skilled in the art shall fall within the scope of the present application.

[0055] The above is only a preferred embodiment of the present application, and is not intended to limit the protection scope of the present application. Any modification, equivalent replacement, improvement, etc. within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A method for intelligent classification and dynamic updating of a hydropower equipment library based on a deep neural network, characterized in that, Includes the following steps: S1. Construction of integrated discrimination model: Construct a mathematical model that integrates logical discrimination method, statistical discrimination method and cluster analysis method, and combine data time series characteristics, inherent characteristics of physical quantities and importance of measurement points to create an identification model that matches different physical quantities with measurement points. S2. Initial screening of monitoring data: The system automatically retrieves new data from the previous 24 hours each day and stores it in groups according to the monitoring point ID. After completing the missing data, the model built in S1 is called to initially determine the validity of the data. Valid data is stored directly, and invalid data proceeds to the next step. S3. Data Retesting and Verification: Push retesting instructions to the plant and station, and call the S1 model again for judgment based on the retesting data; if the retest is valid, record the gross error and end the process; invalid data proceeds to the next step. S4. Measurement point classification and joint judgment: Read the preset list of important measurement points and directly trigger the on-site comparison command for important measurement points; For general measuring points, risk classification is determined by combining data from similar measuring points in the same dam section; S5. Key Measurement Point Tracking and Comparison: Low-risk measurement points are included in the attention list and the monitoring frequency is increased. After the period expires, the S1 model is called for re-judgment. If there is an abnormality, on-site comparison is triggered, and instrument failure reminders or inspection instructions are pushed simultaneously. S6. Inspection Result Feedback: If defects are confirmed during inspection, repair suggestions are generated and included in defect management; if no abnormalities are found, the inspection plan for the corresponding part is adjusted. S7. Dynamic Model Optimization: Periodically extract invalid historical data cases for manual review, and re-match the algorithm to models with insufficient adaptability based on gross error recognition rate and anomaly misjudgment rate; S8. Measurement point status control: Mark measurement points into five states: normal, retest, attention, fault, and verification. When data inspection is started, connect to the corresponding processing branches of S2 to S6 according to the corresponding state.

2. The deep neural network-based method and system for intelligent classification and dynamic updating of a hydropower equipment library according to claim 1, characterized in that, In step S2, the missing data completion method is as follows: when the missing value is ≤3 data points, linear interpolation is used for completion; when the missing value is >3 data points, historical average is used for completion. In step S3, after pushing the retest command, if no response is received within 2 hours, the command is automatically escalated to the superior management node. In step S4, the risk classification standard for general monitoring points is as follows: if ≥80% of the data from similar / same location monitoring points within the same dam section are qualified, it is judged as low risk; if ≥50% of the data from similar / same location monitoring points are abnormal, it is judged as high risk. In step S5, monitoring points included in the attention list are monitored at a high frequency of once every 2 hours, with a monitoring cycle of 5 days. In step S6, the method for adjusting the inspection plan when there are no abnormalities is to add the corresponding parts / monitoring quantities to the monthly inspection plan and increase the inspection frequency to 2 times / month. In step S7, the specific parameters for regularly extracting historical invalid data cases for manual review are as follows: extract 10% of the invalid data cases from the previous month each month, and set the gross error identification rate ≥95% and the false positive rate of real anomalies ≤5% as the model adaptability judgment threshold. In step S8, the triggering logic for the measurement point status is as follows: normal measurement points directly enter the S2 initial screening process, verification status measurement points are directly pushed to the anomaly identification module, and retest, attention, and fault status measurement points are respectively connected to the S3 retest, S5 tracking, and S6 defect management branches.

3. The intelligent classification and dynamic updating method and system for hydropower equipment database based on deep neural networks according to claim 1, characterized in that, The logical discrimination method in the S1 step specifically includes: removing out-of-limit data based on the range of the monitoring instrument is the current measurement value, is the range of the instrument, removing positive and negative sign error data based on physical meaning, and removing data exceeding the preset threshold based on engineering specifications.

4. The intelligent classification and dynamic updating method and system for hydropower equipment database based on deep neural networks according to claim 1, characterized in that, In step S1, the statistical discrimination method specifically includes: selecting historical data of similar operating conditions from the past three hydrological years, based on water level... ,temperature ,aging Establish a multiple regression equation for the impact factor ； for the theoretical displacement, for the regression coefficient, calculating the residual ; and residual standard deviation: ； Exceeding 6 standard deviations The data was deemed invalid.

5. The intelligent classification and dynamic updating method and system for hydropower equipment database based on deep neural networks according to claim 1, characterized in that, In step S1, the clustering analysis method specifically includes: using K-means clustering, with a certain number of clusters. Distinguish between normal and abnormal clusters, and calculate the distance from each data point to the center of the normal cluster. Euclidean distance: ； If the distance exceeds the average distance of a normal cluster 2 times: ； It is then identified as an outlier and triggers S3.

6. The intelligent classification and dynamic updating method and system for hydropower equipment database based on deep neural networks according to claim 1, characterized in that, In step S2, the identification model for matching different physical quantities with measuring points is specifically adapted by combining the time sequence characteristics of data, the inherent characteristics of physical quantities, and the importance of measuring points: statistical discrimination method + logical discrimination method for matching physical quantities coupled by multiple factors such as displacement and stress; logical discrimination method for matching single physical quantities with physical meaning such as seepage flow and water level; and forced superposition clustering analysis method for collaborative data of multiple measuring points in the same dam section.

7. The intelligent classification and dynamic updating method and system for hydropower equipment database based on deep neural networks according to claim 2, characterized in that, In step S4, the specific difference in the statistical discrimination threshold between important measuring points and general measuring points is as follows: Key measurement point: The statistical discrimination threshold is tightened to 4 times the standard deviation. ； When this condition is met, the on-site comparison test command is triggered directly; For general measurement points: the statistical discrimination threshold is relaxed to 8 times the standard deviation, only when the residual is within 6... ; When the range exceeds the conventional 6σ threshold but does not reach the extreme value of 8σ, the joint discrimination of multiple physical quantities in the same dam section is initiated to eliminate the interference of single measurement point error.

8. The intelligent classification and dynamic updating method and system for hydropower equipment database based on deep neural networks according to claim 2, characterized in that, In step S8, the branch control details for the five states also include: retesting state test points prioritize the execution of the S3 retesting process, monitoring state test points continuously execute the S5 high-frequency monitoring once every 2 hours, no new data is collected for fault state test points before they are repaired, and verification state test points skip S2-S7 and directly push to the anomaly identification module.

9. The intelligent classification and dynamic updating method and system for hydropower equipment database based on deep neural networks according to claim 2, characterized in that, In step S7, the model optimization iteration rules are as follows: for unsuitable models marked after review, when calling the algorithm library for rematching, it is necessary to backtest with historical data. The original model can only be replaced if the gross error recognition rate of the backtest is ≥96%. The optimization results are archived to the model-test point matching rule library.

10. The intelligent classification and dynamic updating method and system for hydropower equipment database based on deep neural networks according to claim 2, characterized in that, In step S5, the specific criteria for on-site comparison testing are as follows: using a metrologically certified standard instrument for comparison testing, if the difference between the monitoring value of the standard instrument and the original instrument is greater than twice the accuracy of the original instrument, it is determined to be an instrument malfunction; if the difference is less than or equal to twice the accuracy of the original instrument, it is determined to be an abnormal measurement.

Images