A method and system for identifying abnormal monitoring data of a hydraulic structure

Abstract

The application discloses a kind of hydraulic structure monitoring data anomaly identification method and system, belong to hydraulic construction detection technical field, comprising the following steps: S1, deformation monitoring prediction model construction and fitting section residual sequence calculation;S2, fitting section residual mean sequence definition;S3, residual mean probability distribution estimation and test interval determination;S4, obtain test section residual mean alarm rate;S5, online monitoring parameter output;S6, data identification result output.Through the above mode, the application is by intercepting the prediction residual of monitoring effect quantity several subsequences, calculate the mean of residual in each subsequence, and fit the probability distribution of residual mean, determine monitoring threshold based on the confidence interval of residual mean.Compared with traditional confidence interval method, the application method can effectively weaken the influence of random observation error, significantly reduce the false positive rate and missed rate of monitoring data anomaly identification, identify abnormal data in time and continuously early warning.

Description

Technical Field

[0001] This invention relates to the field of hydraulic structure inspection technology, specifically to a method and system for identifying anomalies in monitoring data of hydraulic structures. Background Technology

[0002] Hydraulic structures serve multiple functions, including flood control, power generation, irrigation, and water supply, and are crucial infrastructure for the national economy. Deformation is a primary aspect of safety monitoring for hydraulic structures, as deformation monitoring data is relatively reliable and can reflect structural damage and other anomalies. This study establishes a deformation monitoring model for hydraulic structures using historical data under normal operating conditions, proposes a method for identifying anomalies in deformation measurements, and combines multi-point monitoring information to comprehensively evaluate the overall deformation behavior of hydraulic structures and provide early warnings of safety risks. This is of great significance for ensuring the safe and stable operation of hydraulic structures.

[0003] The traditional confidence interval method, which is commonly used in the identification of anomalies in existing hydraulic structure monitoring data, estimates the confidence interval of the monitoring effect based on the probability distribution of the residual predicted by the monitoring effect model, and determines the monitoring warning value accordingly.

[0004] The traditional confidence interval method is susceptible to random observation errors in its anomaly identification results. When the variance of the observation error increases, the fluctuation of random error may cause the monitoring data to be misjudged as structural anomalies, thus generating false warnings. At the same time, large observation errors may also overshadow the deviation between the monitored values ​​and the predicted values ​​caused by the actual structural anomalies, resulting in the failure to identify structural anomalies in a timely manner and causing missed judgments. Summary of the Invention

[0005] In view of the above-mentioned shortcomings of the existing technology, the present invention provides a method and system for identifying anomalies in monitoring data of hydraulic structures.

[0006] To achieve the above objectives, the present invention provides the following technical solution: A method for identifying anomalies in monitoring data of hydraulic structures, characterized by comprising the following steps: S1. Obtain monitoring data, establish a deformation monitoring and prediction model, and at the same time obtain the fitted value, calculate the difference between the fitted value and the measured value of the monitoring data to obtain the residual value sequence of the fitted segment. S2. Construction of the mean residual sequence of the fitted segment, specifically including the following steps: S201. Set the length of the residual subsequence. Based on the length of the residual subsequence, slide and truncate the residual value sequence of the fitted segment according to the step size to construct the residual subsequence. S202. Calculate the mean residual of the residual subsequence to obtain the mean residual sequence of the fitted segment; S3. Based on the residual mean sequence of the fitted segment, perform kernel density estimation, obtain the probability density estimation function and calculate the cumulative distribution function. Through the inverse function of the cumulative distribution function and the confidence level, calculate the monitoring warning value and form the residual mean test interval. S4. Input the monitoring effect of the test section into the deformation monitoring prediction model, calculate the residual value sequence of the test section, construct a sliding subsequence for the residual value sequence of the test section, and calculate the mean of the residual subsequence of the test section. Define the alarm exponential function based on the test interval of the mean of the residual subsequence of the test section and the mean of the residual, and calculate the alarm rate based on the alarm exponential function. S5. Select the residual subsequence length with the lowest alarm rate as the optimal residual subsequence length, and at the same time, use the upper and lower monitoring warning values ​​corresponding to the optimal residual subsequence length as the optimal monitoring warning values. S6. Input the measured environmental factor data into the deformation monitoring and prediction model to obtain the deformation prediction value at the monitoring time. At the same time, calculate the residual value by combining the measured deformation value at the monitoring time. Based on the optimal residual subsequence length and residual value, construct the optimal length residual subsequence. Select the residual value in the optimal length residual subsequence to calculate the online residual mean. By comparing the online residual mean with the optimal monitoring warning value, obtain the anomaly identification result.

[0007] Furthermore, in step S1, the calculation formula for the deformation monitoring model is: in, Let be the fitted value at time t. The value calculated by the deformation model at time t. Let be the water pressure component at time t. Let be the temperature component at time t. Let be the time-dependent component at time t. A constant term that takes into account the influence of the initial state. For monitoring time.

[0008] Furthermore, in step S201, the length of the residual subsequence is defined as: in, For the first Length of each candidate residual subsequence The increment of the residual subsequence length ( ≥10), The index of the candidate residual subsequence length. The total number of candidate residual subsequence length indices; Based on the length of the candidate residual subsequence, the residual value sequence of the fitted segment is truncated by step size to construct the residual subsequence; The formula for calculating residual subsequences is: in, For the candidate residual subsequence length index The residual subsequence of the j-th fitted segment, This represents the residual value within the subsequence. This is the starting position number of the sliding window. The candidate residual subsequence length index The total number of residual subsequences that can be formed at any time.

[0009] Furthermore, in step S202, the formula for calculating the mean residual of the subsequence is: in, The candidate residual subsequence length index The mean residual of the j-th subsequence. For the first The length of each candidate residual subsequence, where r is the index of the residual term within the subsequence; The mean sequence of the fitted segment residuals is as follows: in, The candidate residual subsequence length index The mean sequence of the residuals of the fitted segments.

[0010] Furthermore, in step S3, the probability density estimation function is calculated using the following formula: in, The candidate residual subsequence length index The probability density estimation function of the mean of the lower residuals. The residual mean variable, The candidate residual subsequence length index The number of residual mean samples under the given conditions For kernel density estimation bandwidth, For kernel function, Let j be the mean of the j-th residual subsequence; The formula for calculating the cumulative distribution function is: in, Let be the cumulative distribution function of the residual mean. For integration variables, For an integral infinitesimal element, L is the length index of the candidate residual subsequence k The probability density estimation function for the mean of the lower residuals.

[0011] Furthermore, in step S4, the alarm rate is calculated using the following formula: in, The candidate residual subsequence length index The corresponding alarm rate The candidate residual subsequence length index The alarm indication value for the j-th test segment sequence.

[0012] Furthermore, in S5, the formula for selecting the optimal length is: in, The index of the candidate residual subsequence length corresponding to the lowest alarm rate. The candidate residual subsequence length index The corresponding alarm rate To determine the optimal residual subsequence length, This is the set of indices representing the lengths of candidate residual subsequences. The length of the k-th candidate residual subsequence; The formulas for determining the optimal lower monitoring alert value and the optimal upper monitoring alert value are as follows: in, To set the optimal monitoring alert value, To achieve the optimal monitoring and alert value, and These are the lower and upper monitoring warning values ​​corresponding to the optimal residual subsequence length, respectively.

[0013] Furthermore, in step S6, the formula for calculating the mean online residual is: in, Let be the mean of the residuals at time t during online monitoring. Let be the residual value at time j within the residual subsequence; The anomaly detection result is defined as: in, The result of anomaly identification at online monitoring time t. To set the optimal monitoring alert value, To achieve the optimal monitoring and alert value, Let be the mean residual value at time t during online monitoring.

[0014] A system for identifying anomalies in monitoring data of hydraulic structures, the system comprising: Model building preprocessing module: Input monitoring data, establish a deformation monitoring and prediction model through semi-separate modeling technology, and obtain the fitted value. The difference between the measured value of the monitoring data and the fitted value is used to obtain the sequence of residual values ​​of the fitted segment. Sliding residual subsequence mean extraction module: Based on the fitted segment residual value sequence and with a set length, calculate the residual subsequence to obtain the fitted segment residual mean sequence; Test segment simulation module: Based on the residual mean sequence of the fitted segment, kernel density estimation is performed to obtain the probability density estimation function and calculate the cumulative distribution function. Through the inverse function of the cumulative distribution function and the confidence level, the monitoring warning value is calculated and the residual mean test interval is obtained. Alarm data definition module: Input the monitoring effect of the test segment into the deformation monitoring prediction model, calculate the residual value sequence of the test segment, construct a sliding subsequence for the residual value sequence of the test segment, calculate the mean of the residual subsequence of the test segment, define the alarm exponential function based on the test interval of the mean of the residual subsequence of the test segment and the mean of the residual, and calculate the alarm rate based on the alarm exponential function; Online monitoring parameter optimal filtering module: Selects the optimal residual subsequence length based on the alarm rate, and simultaneously selects the optimal monitoring warning value based on the monitoring warning value; The identification result output module inputs the measured environmental factor data into the deformation monitoring and prediction model to obtain the deformation prediction value at the monitoring time. At the same time, it calculates the residual value by combining the measured deformation value at the monitoring time. Based on the optimal residual subsequence length and residual value, it constructs the optimal length residual subsequence, selects the residual values ​​in the optimal length residual subsequence to calculate the online residual mean, and obtains the anomaly identification result by comparing the online residual mean with the optimal monitoring warning value.

[0015] Compared with the prior art, the beneficial effects of this invention are as follows: 1. This invention extracts several subsequences of the predicted residuals of the monitoring effect, calculates the mean of the residuals within each subsequence, fits the probability distribution of the mean residuals, and determines the monitoring threshold based on the confidence interval of the mean residuals. Compared with the traditional confidence interval method, this invention effectively weakens the impact of random observation errors, significantly reduces the false alarm rate and false negative rate of anomaly identification in monitoring data, promptly identifies abnormal data and provides continuous early warning, thereby effectively reducing the false positive rate of anomaly identification in hydraulic structure monitoring data;

[0016] 2. This invention determines the confidence interval and monitoring threshold based on the probability distribution of the residual mean, and identifies anomalies in monitoring data by verifying whether the mean of the residual subsequence falls within the verification interval determined by the optimal monitoring warning value. Since monitoring values ​​deviate significantly from predicted values ​​when structural anomalies occur, this scheme enhances the sensitivity to changes in structural behavior, avoids large observation errors from overshadowing deviations caused by actual structural anomalies, effectively reduces the missed detection rate of structural anomalies, and enables timely identification and continuous early warning of abnormal data. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0018] Figure 1 This is a flowchart illustrating the method structure of the present invention; Figure 2 This is a diagram showing the arrangement of vertical measuring points on the dam body according to the present invention; Figure 3 This is a graph showing the prediction results of the SSMT-s model at the PL6-1 measurement point of this invention. Figure 4 This is a graph showing the prediction results of the SSMT-s model at the PL10-1 measurement point of this invention. Figure 5 This is a graph showing the prediction results of the SSMT-s model at the PL14-1 measurement point of the present invention. Figure 6 This is a graph showing the prediction results of the SSMT-s model at the PL19-1 measurement point of the present invention. Figure 7 This is a graph showing the prediction results of the SSMT-s model at the PL24-1 measuring point of this invention. Figure 8 This is a graph showing the test results of the traditional confidence interval method for the predicted segment of the PL6-1 measuring point in this invention; Figure 9 This is a graph showing the test results of the traditional confidence interval method for the predicted segment of the PL10-1 measuring point in this invention; Figure 10 This is a graph showing the test results of the traditional confidence interval method for the predicted segment of the PL14-1 measuring point in this invention; Figure 11 This is a graph showing the test results of the traditional confidence interval method for the predicted segment of the PL19-1 measuring point in this invention; Figure 12 This is a graph showing the test results of the traditional confidence interval method for the predicted segment of the PL24-1 measuring point in this invention; Figure 13 This is a diagram showing the results of identifying anomalies in the measurement data of dam section 6 of this invention using the confidence interval method based on the residual mean. Figure 14 This is a diagram showing the results of identifying anomalies in the measurement data of dam section #10 based on the confidence interval method of the residual mean. Figure 15 This is a diagram showing the results of identifying anomalies in the measurement data of dam section 14 of this invention using the confidence interval method based on the residual mean. Figure 16 This is a diagram showing the results of identifying anomalies in the measurement data of dam section 19 of this invention using the confidence interval method based on the residual mean. Figure 17 This is a diagram showing the results of identifying anomalies in the measurement data of dam section #24 of this invention using the confidence interval method based on the residual mean. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0020] Example 1: A method for identifying anomalies in monitoring data of hydraulic structures, comprising the following steps: S1. Construction of Deformation Monitoring and Prediction Model and Calculation of Residual Sequence of Fitted Segment Acquire monitoring data, establish a deformation monitoring and prediction model based on semi-separate modeling technology, obtain fitted values, and construct expressions for water pressure, temperature, and aging components respectively. Calculate the difference between the fitted values ​​and the measured monitoring data to obtain the fitted segment residual value sequence. The specific steps include: S101. Based on the monitoring data during normal operation, a deformation monitoring model is established using semi-separated modeling technology. In the deformation monitoring statistical model, the deformation of hydraulic structures is divided into three parts: water pressure component, temperature component, and time-dependent component. Furthermore, the deformation monitoring model is represented as: in, Let be the fitted value at time t. The value calculated by the deformation model at time t. Let be the water pressure component at time t. Let be the temperature component at time t. Let be the time-dependent component at time t. A constant term that takes into account the influence of the initial state. For monitoring time; In the statistical model for deformation prediction, the water pressure component is generally expressed as a power polynomial of the reservoir water depth. The expression for the water pressure component is: in, Let be the water pressure component at time t. These are the regression coefficients (i=1, 2, ..., m); For normalized reservoir water levels, Let t be the reservoir water level at time t. This is the highest water level. The lowest water level, The term number is the polynomial number for the water pressure term; The temperature component can be expressed in the form of a lagged air temperature, and its expression is as follows: in, Let be the temperature component at time t. (i=1, 2, 3, 4) are the regression coefficients of the temperature component. Let J represent the average air temperature on the observation day and the previous j days (the lag effect of ambient temperature on deformation, j=1,3,5,10,15,30,60). The time-dependent component is an irreversible component that develops over time. It can be expressed as a logarithmic function, and its expression is as follows: in, Let be the time-dependent component at time t. , The regression coefficient for the timeliness factor is... It is a dimensionless time factor. It is the natural logarithm function. For monitoring time, The date of the initial monitoring at the monitoring point; S102. Calculate the fitted values ​​under different loads. And calculate its value compared with the measured value of the monitoring data. The difference between them yields the sequence of residual values ​​for the fitted segment: in, Let be the residual value of the fitted segment at time t. The measured value of the monitoring data at time t. Let be the fitted value at time t. For the fitted segment residual value sequence, This represents the total length of the fitted segment monitoring data; The residual value sequence of the fitted segment is used to characterize the deviation information between the fitted value of the deformation monitoring model and the measured value of the monitoring data. It can accurately reflect the influence of factors such as water pressure, temperature and time on the structure, and provide accurate benchmark data to ensure the subsequent acquisition of a reliable residual value sequence of the fitted segment. S2, Definition of the Mean Sequence of Residuals of Fitted Segments The length of the residual subsequence is set, and the residual value sequence of the fitted segment is truncated by a step size to construct the residual subsequence. The mean residual of the constructed residual subsequence is calculated to obtain the mean residual sequence of the fitted segment. The specific steps include: S201. Set the length of the residual subsequence. Based on the length of the residual subsequence, slide and truncate the residual value sequence of the fitted segment according to the step size to construct the residual subsequence. The length of the residual subsequence is defined as: in, For the first The length of each candidate residual subsequence. The increment of the residual subsequence length ( ≥10), The index of the candidate residual subsequence length. The total number of candidate residual subsequence length indices; Based on the length of the candidate residual subsequence, the residual value sequence of the fitted segment is truncated by step size to construct the residual subsequence; The formula for calculating residual subsequences is: in, For the candidate residual subsequence length index The residual subsequence of the j-th fitted segment, This represents the residual value within the subsequence. This is the starting position number of the sliding window. The candidate residual subsequence length index The total number of residual subsequences that can be formed at any time; S202. Calculate the mean residual of the residual subsequence to obtain the mean residual sequence of the fitted segment: The formula for calculating the mean residual of a subsequence is: in, The candidate residual subsequence length index The mean residual of the j-th subsequence. For the first The length of each candidate residual subsequence, where r is the index of the residual term within the subsequence; The mean sequence of the fitted segment residuals is as follows: in, The candidate residual subsequence length index The mean residual sequence of the fitted segment; The fitted segment residual mean sequence is used to characterize the distribution characteristics of the residual mean under different candidate residual subsequence lengths. By calculating the residual mean of the sliding residual subsequence, the influence of random observation error can be effectively weakened, overcoming the defect in traditional methods where large observation errors easily overwhelm the actual structural anomaly deviations, making the residual mean more sensitive to changes in the actual structural behavior of hydraulic structures. S3. Estimation of the probability distribution of the residual mean and determination of the test interval. Kernel density estimation is performed based on the residual mean sequence of the fitted segment. The probability density estimation function is obtained and the cumulative distribution function is calculated. Using the inverse function of the cumulative distribution function and the confidence level, the monitoring and warning values ​​below and above the residual mean are calculated, and the residual mean test interval is formed. The specific steps include: S301. Perform kernel density estimation on the mean residual sequence of the fitted segment to obtain the probability density estimation function; The formula for calculating the probability density estimation function is: in, The candidate residual subsequence length index The probability density estimation function of the mean of the lower residuals. The residual mean variable, The candidate residual subsequence length index The number of residual mean samples under the given conditions For kernel density estimation bandwidth, For kernel function, Let j be the mean of the j-th residual subsequence; S302. Calculate the cumulative distribution function from the probability density estimation function; in, Let be the cumulative distribution function of the residual mean. For integration variables, For an integral infinitesimal element, L is the length index of the candidate residual subsequence kThe probability density estimation function for the mean of the lower residuals; S303. Based on the confidence level and the inverse function of the cumulative distribution function, calculate the monitoring warning value below the residual mean and the monitoring warning value above the residual mean, and obtain the test interval of the residual mean. in, The candidate residual subsequence length index The corresponding monitoring and warning value under the residual mean, The candidate residual subsequence length index The corresponding residual mean is monitored as a warning value. At the significance level, For confidence level, It is the inverse function of the cumulative distribution function (quantile function); Further, the test interval for the residual mean is obtained: in, The candidate residual subsequence length index The corresponding residual mean test interval; The monitoring warning value and test interval corresponding to the length index of each candidate residual subsequence are used to characterize the normal distribution range of the residual mean under different candidate residual subsequence lengths, providing a basis for subsequent alarm rate calculation, optimal residual subsequence length selection and online anomaly identification. The kernel density estimation method is used to fit the probability distribution of the residual mean and determine the monitoring warning value accordingly. This can establish a confidence interval that is more consistent with the characteristics of the residual mean data. Compared with the traditional confidence interval method, this method can effectively reduce the fluctuation caused by random error and significantly reduce the misjudgment rate of anomaly identification in hydraulic structure monitoring data. S4. Obtain the alarm rate of the average residual value of the test segment. The monitored effect of the test segment is input into the deformation monitoring prediction model to calculate the residual value sequence of the test segment. Simultaneously, a sliding subsequence is constructed from the residual value sequence of the test segment, and the mean of the residual subsequence is calculated. An alarm exponential function is defined based on the test interval between the mean of the residual subsequence and the residual mean, and the alarm rate is calculated according to the alarm exponential function. The specific steps include: S401. Input the monitored effect of the test section into the deformation monitoring and prediction model, and calculate the residual value sequence of the test section based on the model calculation results and the corresponding measured deformation values. The formula for calculating the residual value sequence of the test segment is: in, Let be the residual value at time t of the test segment. Let be the measured deformation value at time t of the test segment. The result of the model calculation at time t is... The sequence of residual values ​​for the test segment is given by m, where m is the total length of the monitoring data for the test segment. S402. Construct a sliding subsequence for the residual value sequence of the test segment, and calculate the mean of the residual subsequence of the test segment; The formula for calculating the sliding subsequence is: in, The candidate residual subsequence length index The residual subsequence of the j-th test segment, This represents the total number of residual subsequences that can be formed from the test segment; The mean of the residual subsequence of the test segment is: in, The candidate residual subsequence length index The mean of the residual subsequence of the j-th test segment. This refers to the (j+r)th residual value in the residual subsequence of the test segment; S403. Define an alarm exponential function based on the mean of the residual subsequence of the test segment and the test interval of the residual mean, and calculate the alarm rate based on the alarm exponential function; (1) Determine whether the mean is within the test interval: If the above conditions are met, the monitoring data corresponding to the subsequence is normal; otherwise, it is determined to be an alarm. (2) Define the alarm exponential function: in, The candidate residual subsequence length index The alarm indication value of the j-th test segment sequence ( =0 indicates normal. =1 indicates an abnormal alarm); (3) Calculate the candidate residual subsequence length index Alarm rate below: in, The candidate residual subsequence length index The corresponding alarm rate The candidate residual subsequence length index The alarm indication value for the j-th test segment sequence; The alarm rate corresponding to each candidate residual subsequence length is used to evaluate the alarm performance of different candidate residual subsequence lengths, and to provide an evaluation basis for the subsequent selection of the optimal residual subsequence length. By calculating the alarm rate under different candidate residual subsequence lengths in the test segment, the monitoring performance under different time windows can be objectively and quantitatively evaluated, and test indicators can be provided to eliminate false alarms caused by fitting errors and monitoring system errors. S5, Online monitoring parameter output The residual subsequence length with the lowest alarm rate is selected as the optimal residual subsequence length. Simultaneously, the upper and lower monitoring alert values ​​corresponding to this optimal residual subsequence length are selected as the optimal monitoring alert values. The specific steps include: S501. Select the optimal residual subsequence length based on the alarm rate; The formula for selecting the optimal length is: in, The index of the candidate residual subsequence length corresponding to the lowest alarm rate. The candidate residual subsequence length index The corresponding alarm rate To determine the optimal residual subsequence length, This is the set of indices representing the lengths of candidate residual subsequences. The length of the k-th candidate residual subsequence; If multiple candidate residual subsequence lengths correspond to the same minimum alarm rate, then the smallest residual subsequence length is selected. in, To determine the optimal residual subsequence length, To obtain the minimum value; S502. Select the optimal lower monitoring warning value and the optimal upper monitoring warning value based on the lower monitoring warning value and the upper monitoring warning value corresponding to the optimal residual subsequence length: in, To set the optimal monitoring alert value, To achieve the optimal monitoring and alert value, and These are the lower and upper monitoring warning values ​​corresponding to the optimal residual subsequence length, respectively. Based on the principle of minimizing the alarm rate of the test segment, the optimal residual subsequence length and the optimal lower and upper monitoring warning values ​​are selected to ensure that the final output online monitoring discrimination parameters have the best anomaly recognition performance, and to maximize the protection of the extremely low false positive and false negative rates in the subsequent monitoring process. S6. Data recognition result output The measured environmental factor data is input into the deformation monitoring and prediction model to obtain the predicted deformation value at the monitoring time. Simultaneously, the residual value is calculated by combining the measured deformation value at the monitoring time. Based on the optimal residual subsequence length and residual value, an optimal length residual subsequence is constructed. The residual values ​​in the optimal length residual subsequence are selected to calculate the online residual mean. The anomaly identification result is obtained by comparing the online residual mean with the optimal lower monitoring warning value and the optimal upper monitoring warning value. The specific steps include: S601. Input the measured environmental factor data into the deformation monitoring and prediction model to obtain the predicted deformation value at the monitoring time, and calculate the residual value by combining it with the measured deformation value at the monitoring time: in, The residual value at time t during online monitoring. The measured deformation value at monitoring time t, The predicted deformation value at monitoring time t; S602. Construct the optimal length residual subsequence based on the optimal residual subsequence length and residual value: in, The optimal length residual subsequence corresponding to online monitoring time t. For the most recent Each residual value, The optimal residual subsequence length; The optimal length residual subsequence is used to characterize the overall deviation level of the residual over a continuous period of time near the current monitoring time, and provides a data basis for subsequent calculation of the mean of online residuals and judgment of whether the deformation monitoring data is abnormal. S603. Calculate the mean online residual value based on the residual value, and obtain the mean online residual value by comparing it with the optimal lower monitoring warning value and the optimal upper monitoring warning value: in, The mean of the online residuals at monitoring time t, Let be the residual value at time j within the optimal length residual subsequence; The anomaly identification result is defined as follows: The mean residual value at online monitoring time t is checked to see if it falls within the test interval determined by the optimal monitoring warning value. in, The result of anomaly identification at online monitoring time t. To set the optimal monitoring alert value, To achieve the optimal monitoring and alert value, The mean of the residuals at time t during online monitoring; By comparing the mean online residual with the optimal inspection interval, when an anomaly occurs in the hydraulic structure, the monitored value will deviate from the predicted value as a whole, causing the mean online residual to increase significantly and exceed the optimal monitoring warning value. Through this identification result output, abnormal data can be identified in a timely manner and continuous early warning can be achieved, effectively avoiding the omission of structural anomalies and ensuring the safe and stable operation of hydraulic structures.

[0021] Example 2: A system for identifying anomalies in monitoring data of hydraulic structures, comprising: Model building preprocessing module: Input monitoring data, establish a deformation monitoring and prediction model through semi-separate modeling technology, and obtain the fitted value. The difference between the measured value of the monitoring data and the fitted value is used to obtain the sequence of residual values ​​of the fitted segment. Sliding residual subsequence mean extraction module: Based on the fitted segment residual value sequence and with a set length, calculate the residual subsequence to obtain the fitted segment residual mean sequence; Test segment simulation module: Based on the residual mean sequence of the fitted segment, kernel density estimation is performed to obtain the probability density estimation function and calculate the cumulative distribution function. Through the inverse function of the cumulative distribution function and the confidence level, the monitoring warning value below the residual mean and the monitoring warning value above the residual mean are calculated, and the residual mean test interval is obtained. Alarm data definition module: Input the monitoring effect of the test segment into the deformation monitoring prediction model, calculate the residual value sequence of the test segment, construct a sliding subsequence for the residual value sequence of the test segment, calculate the mean of the residual subsequence of the test segment, define the alarm exponential function based on the test interval of the mean of the residual subsequence of the test segment and the mean of the residual, and calculate the alarm rate based on the alarm exponential function; The online monitoring parameter optimal filtering module selects the optimal residual subsequence length based on the alarm rate, and selects the optimal lower monitoring warning value and the optimal upper monitoring warning value based on the monitoring warning value below the residual mean and the monitoring warning value above the residual mean. The identification result output module inputs the measured environmental factor data into the deformation monitoring and prediction model to obtain the deformation prediction value. At the same time, it calculates the residual value by combining the measured deformation value. Based on the optimal residual subsequence length and residual value, it constructs the optimal length residual subsequence, selects the residual value in the optimal length residual subsequence to calculate the online residual mean, and obtains the anomaly identification result by comparing the online residual mean with the optimal lower monitoring warning value and the optimal upper monitoring warning value.

[0022] Experiment Example 1: Anomaly Identification Experiment of Pre- and Post-Earthquake Displacement Monitoring Data of Arch Dams The GP Hydropower Station is a key project that primarily generates electricity while also controlling floods. The river-blocking hydraulic structure is a concrete double-curvature arch dam with a maximum dam height of 210m. The normal water level of the reservoir is 1130.00m, and the dead water level is 1120.00m. At 12:52 PM (Beijing time) on September 5, 2022, a magnitude 6.8 earthquake occurred near the dam site, 21 km from the epicenter. To examine whether the deformation of the GP arch dam was adversely affected by the earthquake, a semi-separated modeling technique (SSMT) was applied to construct a deformation monitoring model for the vertical deformation measuring points of the GP arch dam before and after the earthquake, based on the monitoring sequences of the vertical deformation measuring points before and after the earthquake. The optimal monitoring model was selected, and the monitoring data sequences of the vertical measuring points before and after the earthquake were identified using the method of this invention. At the same time, based on the association rule mining results of the vertical displacement monitoring data of the arch dam, a multi-measuring point deformation monitoring index system for the GP arch dam was constructed. The deformation risk of the dam body was assessed by combining the alarm status of the vertical deformation measuring points of the dam. This experiment selected the vertical radial displacement of dam sections 6#, 10#, 14#, 19#, and 24# for study, and established displacement prediction models for the corresponding measuring points. The monitoring data sequences of the prediction sections were identified by combining the model fitting values ​​and the actual monitoring values. The horizontal displacement of the GP arch dam is monitored using both positive and negative plumb lines. Seven positive plumb lines (23 positive plumb lines measuring points) are installed in dam sections 6#, 10#, 14# (arch crown beam), 19#, and 24#, as well as the grouting galleries on both banks. Negative plumb lines are also installed at corresponding locations along these seven positive plumb lines. Two negative plumb lines are installed at the dam foundation of section 14#, for a total of eight negative plumb lines and eight negative plumb lines measuring points. All are integrated into an automated monitoring system to monitor the horizontal displacement of the dam body. The specific layout of the plumb line measuring points is shown in [details omitted]. Figure 2 .

[0023] The method of this invention is used to establish a monitoring model for the vertical radial displacement of dam sections 6#, 10#, 14#, 19#, and 24#. The specific steps are as follows: (1) Multiple modeling methods were used to construct and optimize the vertical measurement point prediction model. Taking five dam crest vertical measurement points as examples, the dam crest vertical measurement points PL6-1, PL10-1, PL14-1, PL19-1 and PL24-1 were modeled based on the semi-separate modeling technique. First, the ICEEMDAN algorithm was used to extract the time-dependent arch dam deformation (i.e., the trend term after decomposition and reconstruction) from the original monitoring data sequence. The time-dependent component of the deformation monitoring model was constructed using the composite exponential function to express the time-dependent deformation of the arch dam. The remaining periodic terms included water pressure deformation and temperature deformation. Then, expressions were constructed for the water pressure component and the temperature component. When fitting the temperature component, they were called SSMT-t and SSMT-s models, respectively. At the same time, machine learning algorithms such as ELM (Extreme Learning Machine), CNN (Convolutional Neural Network), and LS_SVM (Least Squares Support Vector Machine) were used to construct the model. The coefficients of each component of the SSMT-s model of the dam crest measurement point are shown in Table 1.

[0024] Table 1. Coefficients of the SSMT-S model for deformation monitoring at dam crest measuring points ; The evaluation indicators for different monitoring models at each measuring point were calculated, as shown in Table 2.

[0025] Table 2 Evaluation Indicators of the Deformation Monitoring Model at Dam Crest Measuring Points ; The process line of the SSMT-s model prediction results at the dam crest measuring point is shown in the figure. Figures 3-7 .

[0026] The calculation formula for the deformation monitoring model is: in, This is a data sequence for dam deformation monitoring. The estimated value of the time-dependent component at time ( , , (The date of the initial monitoring at the monitoring point). Here is the estimated value of the water pressure component at time t. The estimated temperature component at time t is called the SSMT-s model or SSMT-t model. For constant terms; According to the R2-OC criterion, the SSMT-s model with five measuring points not only has high fitting accuracy but also high prediction accuracy. Therefore, SSMT-s is the optimal model for constructing dam deformation measuring point monitoring.

[0027] (2) Based on the monitoring effect and measured value of each measuring point calculated by the deformation monitoring model, the monitoring warning value is determined by the traditional confidence interval method and the confidence interval method based on the residual mean, and the test interval is determined. The alarm rate of the model pre-earthquake prediction segment of each measuring point is calculated according to the amount of data in the prediction segment that exceeds the test interval. The length of the residual subsequence in the confidence interval method based on the residual mean is 10, 20, 30, 40, 50 and 60 respectively. The specific alarm rate results are shown in Table 3. Table 3 False alarm rate of anomaly identification in the pre-earthquake monitoring sequence at the dam crest monitoring point ; The results of the traditional confidence interval method test of the monitoring sequence of each measuring point on the dam crest are shown below. Figures 8-12 Test results based on the confidence interval method of the residual mean Figures 13-17 .

[0028] (3) Based on the residual sequence of the SSMT-s model fitting value and the measured value at each measuring point, the monitoring warning value is proposed by the confidence interval method based on the residual mean, and the monitoring sequence of the fitting segment is identified. The length of the residual subsequence is 10, 20, 30, 40, 50 and 60. The false alarm rate of the pre-earthquake segment (September 5, 2021 to September 4, 2022) in the fitting segment of each measuring point is calculated respectively. The appropriate interval length for each measuring point is determined according to the pre-earthquake segment results. The false alarm rate of the post-earthquake segment (September 5, 2022 to September 4, 2023) is calculated according to the corresponding interval length. The corresponding results are shown in Table 4. Table 4. Anomaly Identification Alarm Rate of Monitoring Sequence Fitted at Measurement Points ; The test results for the predicted segment based on the confidence interval method of the residual mean are shown below. Figures 13-17 .

[0029] Comparing the false alarm rates of the traditional confidence interval method and the method of this invention, for the best-performing SSMT-s model, the false alarm rate of anomaly identification in displacement monitoring data caused by using the traditional confidence interval method to formulate monitoring warning values ​​ranges from 7.67% to 13.97%. However, when using the method of this invention to formulate monitoring warning values, the maximum false alarm rate of anomaly identification in monitoring data at each measuring point is only 3.01% when the residual subsequence length L = 60 days, significantly lower than that of the traditional confidence interval method. Therefore, compared with the traditional confidence interval method, the method of this invention can effectively reduce the false alarm rate of safety monitoring.

[0030] The method of this invention was used to identify anomalies in the post-earthquake displacement monitoring data sequences of 18 monitoring points on the dam body. The results showed that 11 monitoring points had an anomaly alarm rate exceeding 10%. Among them, the average alarm rate for monitoring point #6 was 22.49%, with a maximum alarm rate of 46.26%; the average alarm rate for monitoring point #10 was 16.79%, with a maximum alarm rate of 28.25%; the average alarm rate for monitoring point #14 was 26.405%, with a maximum alarm rate of 55.96%; and the average alarm rate for monitoring point #24 was 31.76%, with a maximum alarm rate of 93.91%.

[0031] The alarm rate of monitoring data at each monitoring point increased significantly after the earthquake compared to before the earthquake, indicating that the method of the present invention can promptly identify anomalies in the displacement monitoring data sequence caused by the earthquake.

[0032] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions will 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.

Claims

1. A method for identifying anomalies in monitoring data of hydraulic structures, characterized in that, Includes the following steps: S1. Obtain monitoring data, establish a deformation monitoring and prediction model, and at the same time obtain the fitted value, calculate the difference between the fitted value and the measured value of the monitoring data to obtain the residual value sequence of the fitted segment. S2. Construction of the mean residual sequence of the fitted segment, specifically including the following steps: S201. Set the length of the residual subsequence. Based on the length of the residual subsequence, slide and truncate the residual value sequence of the fitted segment according to the step size to construct the residual subsequence. S202. Calculate the mean residual of the residual subsequence to obtain the mean residual sequence of the fitted segment; S3. Based on the residual mean sequence of the fitted segment, perform kernel density estimation, obtain the probability density estimation function and calculate the cumulative distribution function. Through the inverse function of the cumulative distribution function and the confidence level, calculate the monitoring warning value and form the residual mean test interval. S4. Input the monitoring effect of the test section into the deformation monitoring prediction model, calculate the residual value sequence of the test section, construct a sliding subsequence for the residual value sequence of the test section, and calculate the mean of the residual subsequence of the test section. Define the alarm exponential function based on the test interval of the mean of the residual subsequence of the test section and the mean of the residual, and calculate the alarm rate based on the alarm exponential function. S5. Select the residual subsequence length with the lowest alarm rate as the optimal residual subsequence length, and at the same time, use the upper and lower monitoring warning values ​​corresponding to the optimal residual subsequence length as the optimal monitoring warning values. S6. Input the measured environmental factor data into the deformation monitoring and prediction model to obtain the deformation prediction value at the monitoring time. At the same time, calculate the residual value by combining the measured deformation value at the monitoring time. Based on the optimal residual subsequence length and residual value, construct the optimal length residual subsequence. Select the residual value in the optimal length residual subsequence to calculate the online residual mean. By comparing the online residual mean with the optimal monitoring warning value, obtain the anomaly identification result.

2. The method for identifying anomalies in monitoring data of hydraulic structures according to claim 1, characterized in that, In step S1, the calculation formula for the deformation monitoring model is: in, Let be the fitted value at time t. The value calculated by the deformation model at time t. Let be the water pressure component at time t. Let be the temperature component at time t. Let be the time-dependent component at time t. A constant term that takes into account the influence of the initial state. For monitoring time.

3. The method for identifying anomalies in monitoring data of hydraulic structures according to claim 1, characterized in that, In step S201, the length of the residual subsequence is defined as: in, For the first Length of each candidate residual subsequence The increment of the residual subsequence length ( ≥10), The index of the candidate residual subsequence length. The total number of candidate residual subsequence length indices; Based on the length of the candidate residual subsequence, the residual value sequence of the fitted segment is truncated by step size to construct the residual subsequence; The formula for calculating residual subsequences is: in, For the candidate residual subsequence length index The residual subsequence of the j-th fitted segment, This represents the residual value within the subsequence. This is the starting position number of the sliding window. The candidate residual subsequence length index The total number of residual subsequences that can be formed at any time.

4. The method for identifying anomalies in monitoring data of hydraulic structures according to claim 3, characterized in that, In step S202, the formula for calculating the mean residual of the subsequence is: in, The candidate residual subsequence length index The mean residual of the j-th subsequence. For the first The length of each candidate residual subsequence, where r is the index of the residual term within the subsequence; The mean sequence of the fitted segment residuals is as follows: in, The candidate residual subsequence length index The mean sequence of the residuals of the fitted segments.

5. The method for identifying anomalies in monitoring data of hydraulic structures according to claim 1, characterized in that, In step S3, the probability density estimation function is calculated using the following formula: in, The candidate residual subsequence length index The probability density estimation function of the mean of the lower residuals. The residual mean variable, The candidate residual subsequence length index The number of residual mean samples under the given conditions For kernel density estimation bandwidth, For kernel function, Let j be the mean of the j-th residual subsequence; The formula for calculating the cumulative distribution function is: in, Let be the cumulative distribution function of the residual mean. For integration variables, For an integral infinitesimal element, L is the length index of the candidate residual subsequence k The probability density estimation function for the mean of the lower residuals.

6. The method for identifying anomalies in monitoring data of hydraulic structures according to claim 1, characterized in that, In step S4, the alarm rate is calculated using the following formula: in, The candidate residual subsequence length index The corresponding alarm rate The candidate residual subsequence length index The alarm indication value for the j-th test segment sequence.

7. The method for identifying anomalies in monitoring data of hydraulic structures according to claim 6, characterized in that, In S5, the formula for selecting the optimal length is: in, The index of the candidate residual subsequence length corresponding to the lowest alarm rate. The candidate residual subsequence length index The corresponding alarm rate To determine the optimal residual subsequence length, This is the set of indices representing the lengths of candidate residual subsequences. The length of the k-th candidate residual subsequence; The formulas for determining the optimal lower monitoring alert value and the optimal upper monitoring alert value are as follows: in, To set the optimal monitoring alert value, To achieve the optimal monitoring and alert value, and These are the lower and upper monitoring warning values ​​corresponding to the optimal residual subsequence length, respectively.

8. The method for identifying anomalies in monitoring data of hydraulic structures according to claim 1, characterized in that, In step S6, the formula for calculating the mean online residual is: in, Let be the mean of the residuals at time t during online monitoring. Let be the residual value at time j within the residual subsequence; The anomaly detection result is defined as: in, The result of anomaly identification at online monitoring time t. To set the optimal monitoring alert value, To achieve the optimal monitoring and alert value, Let be the mean residual value at time t during online monitoring.

9. A system for identifying anomalies in monitoring data of hydraulic structures, characterized in that, The system is used in the method for identifying anomalies in monitoring data of hydraulic structures according to any one of claims 1-8, and the system comprises: Model building preprocessing module: Input monitoring data, establish a deformation monitoring and prediction model through semi-separate modeling technology, and obtain the fitted value. The difference between the measured value of the monitoring data and the fitted value is used to obtain the sequence of residual values ​​of the fitted segment. Sliding residual subsequence mean extraction module: Based on the fitted segment residual value sequence and with a set length, calculate the residual subsequence to obtain the fitted segment residual mean sequence; Test segment simulation module: Based on the residual mean sequence of the fitted segment, kernel density estimation is performed to obtain the probability density estimation function and calculate the cumulative distribution function. Through the inverse function of the cumulative distribution function and the confidence level, the monitoring warning value is calculated and the residual mean test interval is obtained. Alarm data definition module: Input the monitoring effect of the test segment into the deformation monitoring prediction model, calculate the residual value sequence of the test segment, construct a sliding subsequence for the residual value sequence of the test segment, calculate the mean of the residual subsequence of the test segment, define the alarm exponential function based on the test interval of the mean of the residual subsequence of the test segment and the mean of the residual, and calculate the alarm rate based on the alarm exponential function; Online monitoring parameter optimal filtering module: Selects the optimal residual subsequence length based on the alarm rate, and simultaneously selects the optimal monitoring warning value based on the monitoring warning value; The identification result output module inputs the measured environmental factor data into the deformation monitoring and prediction model to obtain the deformation prediction value at the monitoring time. At the same time, it calculates the residual value by combining the measured deformation value at the monitoring time. Based on the optimal residual subsequence length and residual value, it constructs the optimal length residual subsequence, selects the residual values ​​in the optimal length residual subsequence to calculate the online residual mean, and obtains the anomaly identification result by comparing the online residual mean with the optimal monitoring warning value.

Images