Safety monitoring data reliability evaluation method based on multistage pre-screening and feature fusion

Abstract

The invention relates to the technical field of safety monitoring of water conservancy and hydropower engineering, in particular to a safety monitoring data reliability evaluation method based on multi-stage pre-screening and feature fusion, and the method comprises the steps: sequentially checking recent data missing, an overall data missing rate and a preliminary abnormal value ratio of an instrument through a three-stage pre-screening mechanism; invalid instruments or instruments with extremely poor data quality can be quickly judged and screened out; and for the pre-screened instrument, further extracting the multi-dimensional depth features of the pre-screened instrument, carrying out comprehensive intelligent evaluation by fusing an expert rule engine and a machine learning model, and finally outputting a reliability conclusion. According to the method, efficient, automatic and intelligent evaluation of the reliability of historical monitoring data is realized, the evaluation result is objective and accurate, continuous self-optimization can be realized through a man-machine cooperation mechanism, the efficiency and the intelligent level of dam safety monitoring management are greatly improved, and the system is suitable for popularization and application. And reliable technical support is provided for scientific management and maintenance decision of a dam safety monitoring system.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of safety monitoring of water conservancy and hydropower projects, and in particular to a safety monitoring data reliability evaluation method based on multi-level pre-screening and feature fusion. BACKGROUND

[0002] A large number of monitoring instruments of various types are deployed in a dam safety monitoring system to continuously collect data reflecting the operating conditions of the dam and its associated structures. The reliability of these instruments is directly related to the effectiveness of the monitoring data, which in turn affects the accuracy of dam safety assessment. Currently, the evaluation of the reliability of monitoring instruments mainly relies on human experience, with visual inspection and judgment of historical data. This approach has the following disadvantages: Low efficiency: The number of dam monitoring instruments is large, and the amount of historical data is vast, making the workload of manual evaluation enormous.

[0003] High subjectivity: The evaluation criteria are difficult to unify, and highly dependent on individual experience of experts, different experts may draw different conclusions.

[0004] Not comprehensive: Manual analysis cannot deeply mine the time series rules, mutation characteristics and other multi-dimensional information implied in the data.

[0005] Strong hysteresis: It is not possible to achieve automatic real-time evaluation, and it is difficult to discover instrument failure problems in a timely manner.

[0006] Therefore, there is an urgent need for an automatic, intelligent, objective and efficient method to evaluate the reliability of monitoring instruments. The purpose of the present application is to solve the above problems in the prior art and provide a method and system for automatically, intelligently, efficiently and accurately evaluating the reliability of safety monitoring data. The method quickly locks the failed instruments through a multi-level pre-screening mechanism, and uses multi-feature fusion and intelligent algorithms to deeply analyze the instrument data screened by the pre-screening, finally outputs the reliability conclusion, and can realize human-computer collaboration and self-iterative optimization. SUMMARY

[0007] To achieve the above purpose, the present application provides the following technical solutions: According to the first aspect of the present application, a safety monitoring data reliability evaluation method based on multi-level pre-screening and feature fusion is claimed, comprising the following steps: S1: Perform data acquisition and preprocessing, acquire historical time series data of the target monitoring instrument and its corresponding time stamp, and acquire the current time of the system; S2: Three-level unreliable pre-screening judgment, the historical time series data is sequentially subjected to recent data missing check, overall data missing rate check and preliminary abnormal value ratio check according to priority order, if any check meets the unreliable condition, the target monitoring instrument is directly judged as unreliable and the process is terminated; S3: Multi-dimensional deep feature extraction, for the historical time series data judged by S2, multi-dimensional deep features including statistical features, time series regularity features and mutation point features are extracted to construct a feature vector; S4: Reliability comprehensive intelligent evaluation, based on the feature vector, evaluation is performed by combining a rule engine and a machine learning model to generate a reliability conclusion, the rule engine is based on a pre-set expert rule library for judgment, and the machine learning model is a pre-trained multi-classification model, and an output probability distribution is output; S5: The machine learning model is retrained based on artificial feedback, and is iteratively optimized to continuously optimize the model performance.

[0008] Further, the recent data missing check in step S2 specifically includes: calculating the time interval between the last valid data timestamp and the current system time; judging whether the time interval is greater than a pre-set threshold, the pre-set threshold being 5 to 10 days; if yes, it is determined that the target monitoring instrument is unreliable.

[0009] Further, the preliminary abnormal value ratio check in step S2 specifically includes: using the interquartile range method to detect abnormal values of the historical time series data, and calculating the ratio of the number of abnormal values to the total number of data points; judging whether the ratio is greater than a pre-set threshold, the pre-set threshold being 0.4; if yes, it is determined that the target monitoring instrument is unreliable.

[0010] Further, the time series regularity features extracted in step S3 include: trend strength, periodicity strength and noise level; the trend strength is obtained by fitting a nonlinear model and calculating the ratio of residual variance to original data variance; the periodicity strength is obtained by calculating the first significant peak value of the autocorrelation function in the non-zero lag region; the noise level is obtained by calculating the standard deviation of the detrended residual sequence.

[0011] Further, the mutation point features extracted in step S3 include the number of mutation points and the average mutation amplitude; the number of mutation points is detected using the PELT algorithm; the average mutation amplitude is obtained by calculating the average value of the data change at the mutation point.

[0012] Further, the evaluation performed by the rule engine in step S4 specifically includes: Based on the conditional statements in the expert rule base, the relationship between each feature value in the feature vector and the preset threshold is combined and contrasted, and the reliability conclusion of the rule engine is output; The expert rule base is constructed based on the dam safety monitoring system evaluation specification.

[0013] Further, the machine learning model evaluation in step S4 specifically includes: The feature vector is input into a pre-trained LightGBM classification model to obtain the probability distribution of the target monitoring instrument belonging to the "reliable", "basically reliable" and "unreliable" categories; According to the comparison of the highest probability value of the probability distribution and the confidence threshold, it is determined whether to adopt the conclusion of the machine learning model.

[0014] Further, the step S4 further includes: If the highest probability value output by the machine learning model is greater than the confidence threshold, the category is directly used as the final reliability conclusion; If the highest probability value is lower than the confidence threshold, or the rule engine conclusion conflicts with the machine learning conclusion, the artificial research and judgment process is started, and the result is stored in the sample database by experts.

[0015] Further, the step S5 further includes: Periodically obtain the newly added adjudication results from the artificial research and judgment process as labeled samples; The labeled samples are used to update the training data set, and the machine learning model is retrained and parameter optimized to realize self-adaptive learning of the model.

[0016] The present application relates to the technical field of water conservancy and hydropower engineering safety monitoring, and particularly relates to a safety monitoring data reliability evaluation method based on multi-level pre-screening and feature fusion, which quickly determines and screens out instruments that have failed or have poor data quality through a three-level pre-screening mechanism, sequentially checks recent data missing, overall data missing rate and preliminary abnormal value ratio of the instruments, and further extracts multi-dimensional deep features of the instruments that pass the pre-screening, and integrates an expert rule engine and a machine learning model for comprehensive intelligent evaluation, and finally outputs a reliability conclusion. The present application realizes efficient, automatic and intelligent evaluation of the reliability of historical monitoring data, the evaluation result is objective and accurate, and can be continuously self-optimized through a man-machine cooperation mechanism, greatly improving the efficiency and intelligent level of dam safety monitoring management, and providing reliable technical support for scientific management and maintenance decision of the dam safety monitoring system. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 A work flow chart of a safety monitoring data reliability evaluation method based on multi-level pre-screening and feature fusion according to the present application. Figure 2 A multi-dimensional deep feature extraction and comprehensive intelligent evaluation process schematic diagram of a safety monitoring data reliability evaluation method based on multi-level pre-screening and feature fusion according to the present application; Figure 3 A test data schematic diagram of a safety monitoring data reliability evaluation method based on multi-level pre-screening and feature fusion according to the present application; Figures 4-6 A running result schematic diagram of a safety monitoring data reliability evaluation method based on multi-level pre-screening and feature fusion according to the present application. DETAILED DESCRIPTION

[0018] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative efforts fall within the scope of the present application.

[0019] The terms "first", "second", "third" in the present application are only for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined with "first", "second", "third" can explicitly or implicitly include at least one of the features. In the description of the present application, the meaning of "a plurality of" is at least two, such as two, three, etc., unless otherwise explicitly specified. All directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of the present application are only used to explain the relative positional relationship, movement condition, etc. between the components in a certain posture (as shown in the drawings), and if the certain posture changes, the directional indications also change accordingly. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device including a series of steps or units is not limited to the listed steps or units, but can optionally include steps or units not listed, or can optionally include other steps or units inherent to the process, method, product or device.

[0020] In this document, the reference to "embodiment" means that the specific feature, structure or property described in connection with the embodiment can be included in at least one embodiment of the present application. The appearance of this phrase in various places in the specification does not necessarily mean the same embodiment is referred to, nor does it mean that independent or alternative embodiments are mutually exclusive or alternative to other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0021] According to the first embodiment of the present invention, referring to Figure 1 This invention claims protection for a method for evaluating the reliability of safety monitoring data based on multi-level pre-screening and feature fusion, comprising the following steps: S1: Perform data acquisition and preprocessing, acquire historical time series data of the target monitoring instrument and its corresponding timestamps, and acquire the current system time; S2: Three-level unreliability pre-screening judgment: The historical time series data is sequentially checked for recent data missing, overall data missing rate, and preliminary outlier ratio according to priority. If any check meets the unreliability condition, the target monitoring instrument is directly determined to be unreliable and the process is terminated. S3: Multi-dimensional deep feature extraction. For historical time series data that has been pre-screened by S2, multi-dimensional deep features including statistical features, temporal regularity features, and abrupt change features are extracted to construct feature vectors. S4: Comprehensive intelligent reliability evaluation, based on the feature vector, uses a combination of rule engine and machine learning model to evaluate and generate reliability conclusions. The rule engine makes judgments based on a preset expert rule base, and the machine learning model is a pre-trained multi-classification model that outputs a probability distribution. S5: Retrain the machine learning model based on human feedback and iteratively optimize it to continuously improve model performance.

[0022] In this embodiment, historical time-series data of the target monitoring instrument is acquired. and its corresponding timestamp and obtain the current system time. ; Three judgment conditions are set in descending order of priority. If any condition is met, the instrument is directly determined to be unreliable and the process is terminated. The number of measurement points to be calculated by the instrument within a historical time period. and the actual number of measurement points Calculate the overall missing rate :

[0023] judge Is it greater than the preset threshold? ( (Values ​​range from 0.3 to 0.4). If so, the instrument is deemed unreliable because historical data is severely lacking and cannot accurately reflect the building's operational status.

[0024] Furthermore, the recent data missing check in step S2 specifically includes: Calculate the time interval between the last valid data timestamp and the current system time; Determine whether the time interval is greater than a preset threshold, wherein the preset threshold is 5 to 10 days; If so, the target monitoring instrument is deemed unreliable.

[0025] In this embodiment, the last valid data timestamp is calculated. With current time time interval :

[0026] judge Is it greater than the preset threshold? ( (The value is 5-10 days). If so, the instrument is considered unreliable because there is no recent data, and it is considered to have failed.

[0027] Furthermore, the preliminary outlier ratio check in step S2 specifically includes: The interquartile range method is used to detect outliers in the historical time series data, and the ratio of the number of outlier points to the total number of data points is calculated. Determine whether the ratio is greater than a preset threshold, wherein the preset threshold is 0.4; If so, the target monitoring instrument is deemed unreliable.

[0028] In this embodiment, the interquartile range (IQR) method is used to analyze the actual data. Perform preliminary outlier detection:

[0029]

[0030]

[0031]

[0032]

[0033] In the above formula, The function is used to calculate quantiles. express 25% of the values ​​are less than , express 75% of the values ​​are less than , Interquartile range, and These represent the lower and upper limits of the normal value, respectively.

[0034] Statistics are not available number of data points within a range and calculate the ratio :

[0035] determine whether it is greater than a preset threshold ( 0.4). If yes, it is determined that the instrument is unreliable, and the basis is that the proportion of data gross errors is too high and cannot correctly reflect the building operation state.

[0036] Further, the time series regularity features extracted in the step S3 include: trend strength, periodicity strength, and noise level; The trend strength is obtained by fitting a nonlinear model and calculating the ratio of residual variance to original data variance; The periodicity strength is obtained by calculating the first significant peak value of the autocorrelation function in the non-zero lag region; The noise level is obtained by calculating the standard deviation of the detrended residual sequence.

[0037] Further, the mutation point features extracted in the step S3 include the number of mutation points and the average mutation amplitude; The number of mutation points is detected using the PELT algorithm; The average mutation amplitude is obtained by calculating the average value of the data change at the mutation point.

[0038] In this embodiment, for the data pre-screened by the S2 stage, the feature vector for deep analysis is extracted, including: statistical features: variance , skewness , kurtosis .

[0039] Time series regularity features: trend strength : , wherein represents the variance of the residual after nonlinear fitting of the original data , and the selection of the fitting model is related to the type of instrument, such as Gaussian model, logarithmic model, trigonometric function, multiple regression, etc.

[0040] Periodicity strength : Calculate the first significant peak value of the autocorrelation function in the non-zero lag region.

[0041] Noise level : Calculate the standard deviation of the detrended residual sequence.

[0042] Mutation point feature: the number of mutation points in the sequence is detected using the mutation point detection algorithm PELT and the average mutation amplitude ; Referring Figure 2 to the schematic diagram of the multi-dimensional deep feature extraction and comprehensive intelligent evaluation process in this embodiment.

[0043] Further, the rule engine evaluation in step S4 specifically includes: Based on the conditional statements in the expert rule base, the relationship between each feature value in the feature vector and the preset threshold is combined and compared, and the reliability conclusion of the rule engine is output; The expert rule base is constructed based on the dam safety monitoring system evaluation specification.

[0044] Further, the machine learning model evaluation in step S4 specifically includes: The feature vector is input into the pre-trained LightGBM classification model to obtain the probability distribution of the target monitoring instrument belonging to the "reliable", "basically reliable" and "unreliable" categories; According to the comparison of the highest probability value of the probability distribution and the confidence threshold, it is determined whether to adopt the conclusion of the machine learning model.

[0045] Further, the step S4 further includes: If the highest probability value output by the machine learning model is greater than the confidence threshold, the category is directly used as the final reliability conclusion; If the highest probability value is lower than the confidence threshold, or the rule engine conclusion and the machine learning conclusion conflict, the artificial research and judgment process is started, and the expert makes a decision and stores the result in the sample database.

[0046] In this embodiment, based on the feature vector , the rule engine and the machine learning model are combined for evaluation: Rule engine evaluation: based on the preset expert rule base for judgment. The rule is composed of a series of conditional statements, and according to the relevant specifications of the dam safety monitoring system evaluation, the relationship between each feature value and its limit threshold is combined and compared, and then the reliability of the data is judged, such as: "if and , the data is evaluated as reliable".

[0047] Machine learning model evaluation: the feature vector is input into the pre-trained classification model LightGBM to obtain the probability distribution of its belonging to the "reliable", "basically reliable" and "unreliable" three categories .

[0048] If the highest probability output by the machine learning model is greater than the confidence threshold , then directly adopt the category as the final conclusion.

[0049] If the highest probability is lower than the confidence threshold, or the rule engine conclusion conflicts with the machine learning conclusion, then push the instrument data and all feature analysis results to the artificial research and judgment interface for final adjudication by an expert, and store the adjudication result in the sample database.

[0050] Further, the step S5 further includes: periodically obtaining newly added adjudication results from the artificial research and judgment process as labeled samples; updating the training data set using the labeled samples, retraining and parameter optimization of the machine learning model, to realize self-adaptive learning of the model.

[0051] In this embodiment, the machine learning model is retrained periodically using newly added artificial adjudication sample data, so that the model performance is continuously optimized.

[0052] The algorithm described in the application is programmed and implemented in MATLAB, a total of 5 groups of data are tested, the 5 groups of data are as shown in Figure 3 , and the running results are as shown in Figures 4-6 ; From the above running results, it can be seen that the method described in the application can accurately perform reliability evaluation on various types of data, and detailed evaluation basis is given.

[0053] In the several embodiments of the application, it should be understood that the disclosed system, device and method can be implemented in other ways. For example, the device embodiments described above are only schematic. The division of units is only a logical function division. Other division manners can be used in actual implementation. For example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the units shown or discussed can be indirect coupling or communication connection through some interfaces, devices or units, and can be electrical, mechanical or other forms.

[0054] ​In addition, each function unit in each embodiment of the present application can be integrated in one processing unit, or each unit can exist physically separately, or two or more units can be integrated in one unit. The integrated unit can be realized in the form of hardware or in the form of a software function unit. The above is only an embodiment of the present application, and does not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made by using the content of the specification and drawings, or directly or indirectly applied to other related technical fields, is also included in the patent protection scope of the present application.

[0055] The specific embodiments of the application have been described above, but they are only examples, and the application is not limited to the above-described specific embodiments. Any equivalent modification or substitution made to the application by those skilled in the art is also within the scope of the application, and therefore, equivalent transformations and modifications, improvements, etc. made without departing from the spirit and principle range of the application should be included in the scope of the application.

Claims

1. A method for evaluating the reliability of safety monitoring data based on multi-stage pre-screening and feature fusion, characterized in that, The method comprises the following steps: S1: performing data acquisition and preprocessing, acquiring historical time series data of a target monitoring instrument and its corresponding time stamp, and acquiring the current time of the system; S2: three-level unreliable pre-screening judgment, sequentially performing recent data missing check, overall data missing rate check and preliminary abnormal value ratio check on the historical time series data according to priority order, if any check meets the unreliable condition, directly judging that the target monitoring instrument is unreliable and terminating the process; S3: multi-dimensional deep feature extraction, for the historical time series data pre-screened by S2, extracting multi-dimensional deep features including statistical features, time series regularity features and mutation point features, and constructing a feature vector; S4: reliable comprehensive intelligent evaluation, based on the feature vector, combining rule engine and machine learning model for evaluation, generating a reliability conclusion, the rule engine is judged based on a pre-set expert rule base, and the machine learning model is a pre-trained multi-classification model, outputting a probability distribution; S5: retraining the machine learning model based on artificial feedback, iterative optimization to continuously optimize the model performance.

2. The method according to claim 1, characterized in that, The recent data missing check in step S2 specifically includes: calculating the time interval between the last valid data timestamp and the current time of the system; determining whether the time interval is greater than a pre-set threshold, the pre-set threshold is 5 to 10 days; if yes, it is judged that the target monitoring instrument is unreliable.

3. The method according to claim 1, characterized in that, The preliminary abnormal value ratio check in step S2 specifically includes: detecting abnormal values of the historical time series data using the interquartile range method, calculating the ratio of the number of abnormal values to the total number of data points; determining whether the ratio is greater than a pre-set threshold, the pre-set threshold is 0.4; if yes, it is judged that the target monitoring instrument is unreliable.

4. The method according to claim 1, characterized in that, The time series regularity features extracted in step S3 include: trend intensity, periodicity intensity and noise level; the trend intensity is obtained by fitting a nonlinear model and calculating the ratio of residual variance to original data variance; the periodicity intensity is obtained by calculating the first significant peak value of the autocorrelation function in the non-zero lag region; the noise level is obtained by calculating the standard deviation of the detrended residual sequence.

5. The method of claim 1, wherein the method is characterized by, The mutation point features extracted in step S3 include the number of mutation points and the average mutation amplitude; the number of mutation points is detected using the PELT algorithm; the average mutation amplitude is obtained by calculating the average value of data changes at the mutation points.

6. The method of claim 1, wherein the method is characterized by, The rule engine evaluation in step S4 specifically includes: based on the conditional statements in the expert rule base, combining and comparing the relationship between each feature value in the feature vector and the pre-set threshold, outputting the reliability conclusion of the rule engine; the expert rule base is constructed based on the dam safety monitoring system evaluation specification.

7. The method according to claim 1, characterized in that, The machine learning model evaluation in step S4 specifically includes: inputting the feature vector into the pre-trained LightGBM classification model to obtain the probability distribution of the target monitoring instrument belonging to the "reliable", "basically reliable" and "unreliable" categories; According to a comparison between a highest probability value of the probability distribution and a confidence threshold, it is determined whether to adopt a conclusion of the machine learning model.

8. The method according to claim 7, wherein, The step S4 further includes: If the highest probability value output by the machine learning model is greater than the confidence threshold, the category is directly adopted as a final reliable conclusion; If the highest probability value is lower than the confidence threshold, or the rule engine conclusion conflicts with the machine learning conclusion, an artificial research and judgment process is started, an expert makes a ruling, and the result is stored in a sample database.

9. The method according to claim 1, characterized in that, The step S5 further includes: Periodically, new ruling results from the artificial research and judgment process are obtained as labeled samples; The labeled samples are used to update a training data set, and the machine learning model is retrained and parameter-optimized to realize self-adaptive learning of the model.

Images