Dam deformation outlier identification method and system

Abstract

The application belongs to the technical field of dam safety monitoring, and specifically discloses a dam deformation anomaly value identification method and system, which solves the technical problem of inaccurate identification results of the existing dam deformation anomaly value identification method. The identification method comprises the following steps: clustering and partitioning a plurality of measuring points on a dam, and obtaining the residual error between the predicted value and the measured value of each measuring point in each partition; determining the comprehensive residual error of each partition according to the residual error of all measuring points in each partition; and determining the anomaly value in the measured value of the measuring point in each partition in combination with the predicted value and the comprehensive residual error. The identification method and system consider the space-time correlation characteristics of dam deformation and propose a high-precision dam deformation prediction model, can more reasonably identify the abnormal measured value of the dam, improve the anomaly value identification precision of the deformation monitoring data, and provide a high-precision prediction and anomaly value identification method for dam deformation.

Description

Technical Field

[0001] This invention discloses a method and system for identifying dam deformation anomalies, specifically a method and system for identifying dam deformation anomalies based on spatiotemporal clustering and deep learning, belonging to the field of dam safety monitoring technology. Background Technology

[0002] Safety monitoring is a crucial method for quantitatively analyzing the safety status of dams. Among the many monitoring parameters for dams, deformation can truly, intuitively, and accurately reflect changes in the dam's safety status, making it the most important monitoring content. However, the collected deformation monitoring data is inevitably affected by instrument malfunctions, human factors, or other uncertainties, resulting in various outliers.

[0003] Anomalies may be normal structural responses caused by rapid load changes, or they may be gross errors caused by instrument malfunctions or human factors. Therefore, accurate identification of anomalies in dam deformation is of great significance for dam monitoring data analysis and safety evaluation research.

[0004] Existing technologies typically utilize statistical testing and monitoring models to identify outliers in dam deformation. However, statistical testing only considers the deformation data itself, neglecting the input-output relationship of dam deformation, which can easily lead to misdiagnosis of normal data as outliers. Monitoring model methods primarily consider outliers in single-point monitoring data, ignoring the spatiotemporal correlation with similar monitoring points, resulting in incorrect identification results. Summary of the Invention

[0005] The purpose of this application is to provide a method and system for identifying dam deformation anomalies, so as to solve the technical problem of inaccurate identification results in existing dam deformation anomaly identification methods.

[0006] The first aspect of the present invention provides a method for identifying anomalies in dam deformation, comprising:

[0007] Multiple measuring points on the dam are clustered and partitioned, and the residuals between the predicted and measured values ​​of each measuring point within each partition are obtained.

[0008] The comprehensive residual for each partition is determined based on the residuals of all measuring points within each partition.

[0009] By combining the predicted values ​​and the comprehensive residuals, outliers in the measured values ​​of the measurement points within each partition are identified.

[0010] Preferably, multiple measuring points on the dam are clustered and partitioned, specifically including:

[0011] Obtain the combined distance between each measuring point on the dam and other measuring points;

[0012] Based on the comprehensive distance, the multiple measurement points are clustered and partitioned using the least squares clustering method.

[0013] Preferably, the comprehensive distance between each measuring point on the dam and other measuring points is obtained, specifically including:

[0014] Obtain the absolute distance, incremental distance, and speed-up distance between each measuring point on the dam and other measuring points;

[0015] Based on the absolute distance, incremental distance, and speed-up distance, determine the comprehensive distance between each measuring point and other measuring points.

[0016] Preferably, the absolute distance is determined based on the difference between the deformation value of one measuring point and the deformation value of another measuring point;

[0017] The incremental distance is determined based on the difference between the deformation increment of one measuring point compared to the previous moment and the deformation increment of another measuring point compared to the previous moment.

[0018] The growth rate distance is determined based on the difference between the deformation growth rate of one measuring point and the deformation growth rate of another measuring point.

[0019] Preferably, obtaining the residual between the predicted value and the measured value of each measuring point within each partition specifically includes:

[0020] Optimize the hyperparameters of the prediction model for each partition;

[0021] Based on the environmental impact factors of each partition, the historical measurements of each measuring point, and the hyperparameters, the prediction model for the corresponding partition is trained.

[0022] Obtain the predicted value of the prediction model for each measurement point in each corresponding partition, and determine the residual between the predicted value and the measured value for each measurement point.

[0023] Preferably, optimizing the hyperparameters of the prediction model for each partition specifically includes:

[0024] The honey badger algorithm is used to optimize the hyperparameters of the prediction model for each partition.

[0025] Preferably, the prediction model is a bidirectional long short-term memory neural network model.

[0026] Preferably, the comprehensive residual of each partition is determined based on the residuals of all measuring points within each partition, specifically including:

[0027] Obtain the correlation coefficient between the residual of each measuring point in each partition and the residuals of the other measuring points;

[0028] The composite residual for each partition is determined based on all the correlation coefficients within the partition.

[0029] Preferably, by combining the predicted value and the comprehensive residual, outliers in the measured values ​​of each measurement point within each partition are determined, specifically including:

[0030] The baseline value for each partition is determined using the comprehensive residual;

[0031] The difference between the predicted value and the measured value of each measuring point in each partition is compared with the benchmark value to identify outliers in the measured values ​​of the measuring points in each partition.

[0032] A second aspect of the present invention provides a dam deformation anomaly identification system, comprising:

[0033] The residual module is used to cluster and partition multiple measuring points on the dam, and obtain the residual between the predicted value and the measured value of each measuring point in each partition;

[0034] The comprehensive residual module is used to determine the comprehensive residual of each partition based on the residuals of all measurement points within each partition.

[0035] An outlier determination module is used to combine the predicted value and the comprehensive residual to determine outliers in the measured values ​​of the measurement points within each partition.

[0036] The dam deformation anomaly identification method and system of the present invention have the following advantages compared with the prior art:

[0037] This invention utilizes the least squares clustering method to cluster and partition measurement point deformations, employs the honey badger algorithm to optimize the hyperparameters of the prediction model, and performs modeling and prediction based on the optimal hyperparameters. Then, based on the correlation of residuals among similar measurement points, an outlier discrimination index is constructed. When the residual of the monitoring data sequence of the studied measurement point exceeds the index standard, it can be identified as an outlier.

[0038] The method of this invention considers the spatiotemporal correlation characteristics of dam deformation and proposes a high-precision dam deformation prediction model, which can more reasonably identify abnormal dam measurements, improve the accuracy of abnormal value identification of deformation monitoring data, and provide a high-precision prediction and abnormal value identification method for dam deformation. Attached Figure Description

[0039] Figure 1 This is a flowchart of the dam deformation anomaly identification method of the present invention;

[0040] Figure 2 This is a time series process line of air temperature and upstream water level in an embodiment of the present invention;

[0041] Figure 3 This is a diagram showing the Ward clustering results of the dam body measuring points in an embodiment of the present invention.

[0042] Figure 4 These are the radial horizontal displacement process lines of measuring points in each section of the dam body in this embodiment of the invention, where (a) is... Figure 3 (a) shows the radial and horizontal displacement process lines of each measuring point within partition I; (b) shows the radial and horizontal displacement process lines of each measuring point within partition I. Figure 3 (c) is the radial horizontal displacement process line of each measuring point in partition II; Figure 3 Radial horizontal displacement process lines of each measuring point within partition III; (d) is Figure 3 Radial horizontal displacement process lines of each measuring point within partition IV; (e) is Figure 3 Radial horizontal displacement process lines of each measuring point within partition V;

[0043] Figure 5 In the above, (a) is a comparison diagram of the prediction results and actual deformation sequences of each model at the PL3-1 measuring point in the embodiment of the present invention, and (b) is a comparison diagram of the evaluation index analysis of each model at the PL3-1 measuring point in the embodiment of the present invention.

[0044] Figure 6 In the figure, (a) is a comparison of the prediction results of each model at the PL4-1 measuring point and the actual deformation sequence in the embodiment of the present invention, and (b) is a comparison of the evaluation index analysis of each model at the PL4-1 measuring point in the embodiment of the present invention.

[0045] Figure 7 In the figure, (a) is a comparison of the prediction results of each model at the PL5-1 measuring point and the actual deformation sequence in the embodiment of the present invention, and (b) is a comparison of the evaluation index analysis of each model at the PL5-1 measuring point in the embodiment of the present invention.

[0046] Figure 8 This is a diagram showing the abnormal value identification results of PL4-1 measurement point in an embodiment of the present invention;

[0047] Figure 9 This is a schematic diagram of the dam deformation anomaly identification system of the present invention. Detailed Implementation

[0048] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of the invention. However, those skilled in the art will understand that the invention can be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods are omitted so as not to obscure the description of the invention with unnecessary detail.

[0049] The first aspect of the present invention provides a method for identifying anomalies in dam deformation, such as... Figure 1 As shown, it includes:

[0050] Step 1: Cluster and partition multiple measuring points on the dam, and obtain the residual between the predicted value and the measured value of each measuring point in each partition.

[0051] Before step 1, deformation data and corresponding environmental data of multiple measuring points are obtained through the dam safety monitoring system, and the obtained dataset is preprocessed.

[0052] Step 1 of this embodiment of the invention specifically includes:

[0053] Step 1.1: Obtain the combined distance between each measuring point on the dam and other measuring points, specifically including:

[0054] Step 1.1.1: Obtain the absolute distance, incremental distance, and speed-up distance between each measuring point on the dam and other measuring points.

[0055] The absolute distance is determined based on the difference between the deformation value of one measuring point and the deformation value of another measuring point;

[0056] The incremental distance is determined based on the difference between the deformation increment of one measuring point compared to the previous moment and the deformation increment of another measuring point compared to the previous moment.

[0057] The growth rate distance is determined based on the difference between the deformation growth rate of one measuring point and the deformation growth rate of another measuring point.

[0058] The absolute distance, incremental distance, and growth rate distance mentioned above are shown in formula (1):

[0059]

[0060] In the formula: These represent the absolute distance, incremental distance, and speed-up distance between measuring point a and measuring point b, respectively; x at Indicates the magnitude of the deformation at measuring point a at time t, x bt This represents the magnitude of the deformation at measuring point b at time t; y at y represents the change in the value of measuring point a at time t. at =x at -x a,t-1 y bt y represents the change in the value of measuring point b at time t. bt =x bt -x b，t-1 ;z at This represents the magnitude of the numerical change at measuring point a at time t. z bt This represents the magnitude of the numerical change at measuring point b at time t.

[0061] Step 1.1.2: Based on the absolute distance, incremental distance, and speed-up distance, determine the comprehensive distance between each measuring point and other measuring points, as shown in formula (2):

[0062]

[0063] In the formula: This represents the combined distance between measuring points a and b, where α1, α2, and α3 are the weights of the three distances, and the formula is...

[0064] Step 1.2: Based on the comprehensive distance, use the minimum sum of squared deviations (Ward's method) to cluster and partition multiple measurement points.

[0065] The Ward method groups the N measurement points of the dam into one class. By continuously narrowing down the classes, it merges the two classes with the smallest increase in the sum of squared deviations between measurement points within each class into one class, until all similar measurement points are clustered into one class, and finally outputs the clustering results. Compared with the K-means method, this method does not require specifying the number of clusters. The sum of squared deviations of measurement points within each class is shown in equation (3):

[0066]

[0067] If divided into m classes, the total sum of squared deviations is as shown in equation (4):

[0068]

[0069] In the formula: k n Indicates category; N n Indicates k n The number of measurement points in the class; W tn Indicates k n The sum of squares of the deviations of the measuring point t across all time sections; Representing k respectively n At time t, N n The average value of parameters at each measuring point.

[0070] Step 1.3: Obtain the residual between the predicted value and the measured value for each measurement point within each partition, specifically including:

[0071] Step 1.3.1: Optimize the hyperparameters of the prediction model for each partition.

[0072] The prediction model in this embodiment of the invention is a bidirectional long short-term memory neural network model (BiLSTM), which is a bidirectional LSTM neural network that is deeply improved on the basis of LSTM. This improves the model's ability to extract sequence information on the basis of LSTM and solves the problem of insufficient information mining ability caused by LSTM only exploring data patterns in one direction.

[0073] However, the selection of model parameters has a significant impact on the prediction accuracy of deep learning, and the quality of the parameter optimization algorithm directly affects the accuracy of outlier identification. Therefore, this embodiment of the invention utilizes the honey badger algorithm to optimize the hyperparameters of the prediction model for each partition.

[0074] The Honey Badger Algorithm (HBA) is a novel swarm intelligence optimization algorithm that primarily simulates the dynamic search behavior of honey badgers in digging and searching for honey. Compared with traditional optimization algorithms, it has advantages such as strong optimization ability, high convergence accuracy, and strong stability.

[0075] The HBA algorithm's hyperparameter optimization process is divided into a "mining phase" and a "honeycomb phase." The optimization process is shown below:

[0076] Initialize the population:

[0077] The population size N and the position of individual i of the honey badger are initialized according to equation (5).

[0078] x i =lb i +r1×(ub i -lb i (5)

[0079] In the formula: x i The position of the i-th honey badger in population N; r1 ∈ (0, 1) is a random number; ub i and lb i These are the upper and lower bounds of the optimization space, respectively.

[0080] Define olfactory intensity:

[0081] The honey badger's sense of smell is related to the intensity of its concentration on prey and the distance between the prey and the i-th honey badger. i It is the intensity of the prey's scent, I i The larger the value, the faster the honey badger moves, and vice versa. The definition is as follows:

[0082]

[0083] In the formula: S is the prey concentration intensity; d i x represents the distance between the prey and the i-th honey badger; prey It is the best position for the prey; a random number within (0,1).

[0084] Update density factor:

[0085] The density factor α decreases gradually with the number of iterations, as defined below:

[0086]

[0087] In the formula: l max It represents the maximum number of iterations; C is a constant greater than 1, usually defaulting to 2.

[0088] Excavation phase:

[0089] During the digging phase, the honey badger locates its prey by digging, thus enabling it to capture the prey. Its movement pattern resembles a heart shape, and its trajectory can be described using the following formula:

[0090]

[0091] In the formula: x new The updated position of the honey badger; β is greater than or equal to 1 (generally 6 by default) and represents the honey badger's ability to obtain food; r3, r4, and r5 are three different random numbers (0,1); F is a flag for changing the search direction, determined by equation (9):

[0092]

[0093] In the formula: r6 ∈ (0,1) is a random number. During the excavation phase, the honey badger not only detects the olfactory intensity I i d, the distance between the prey and the prey i The update density factor α is highly dependent and is also affected by F, thus finding a better prey location.

[0094] Honey-finding stage:

[0095] The honey badger will continuously update its movement trajectory based on prey information and honey distance information until it completes the honey-finding task. The specific location update is shown in equation (10):

[0096] x new =x prey +F×r7×w×d i (10)

[0097] In the formula: x new This is the updated location of the honey badger; x prey This refers to the location of the prey. Honey badgers use distance information (d) when searching for prey. i In x prey Search nearby locations.

[0098] Step 1.3.2: Train the prediction model for the corresponding partition based on the environmental impact factors of each partition, the historical measurements of each measuring point, and the hyperparameters optimized using the HBA algorithm.

[0099] In this embodiment of the invention, based on the theory of dam safety monitoring, environmental impact factors are calculated from dam environmental data. Then, an environmental impact factor set for each measuring point in each zone is established as input, and the historical measured values ​​of each measuring point in each zone are used as output. The input and output data are normalized respectively, and training and testing sets are divided to train and test the prediction model.

[0100] The calculation formula for an LSTM neural network is as follows:

[0101]

[0102] In the formula: x t Input for the current time, i t f t c t o t These represent the input gate, forget gate, updated cell state, and output gate, respectively; W i W f W c W o b is the weight matrix; i b f b c b o σ is the bias vector; σ is the sigmoid function; tanh is the hyperbolic tangent activation function; * denotes the Hadamard product symbol.

[0103] The mathematical expression for a Bi-LSTM network is:

[0104]

[0105] In the formula: and These represent the hidden layer outputs of the forward and reverse LSTM networks at time t, respectively; y t This represents the final output at time t; f and g are both activation functions. and For positive bias and weight; and Here, c represents the inverse bias and weights. U represents the bias and weights.

[0106] Step 1.3.3: Obtain the predicted value of the prediction model for each measurement point in the corresponding partition, and determine the residual between the predicted value and the measured value of each measurement point.

[0107] Taking a research measurement point as an example: input the environmental impact factors of the dam at the current stage corresponding to the research measurement point into the prediction model of the corresponding partition after being trained on the training set and tested on the test set, and obtain the predicted value of the research measurement point; then calculate the residual between the predicted value and the measured value of the research measurement point.

[0108] Step 2: Based on the residuals of all measurement points within each partition, determine the comprehensive residual for each partition, specifically including:

[0109] Step 2.1: Obtain the correlation coefficient between the residual of each measuring point in each partition and the residual of the other measuring points.

[0110] In this embodiment of the invention, the correlation between residuals of different measuring points within the same partition is used as the criterion for judgment. A dam deformation anomaly discrimination index based on the comprehensive residuals of similar measuring points is constructed. Specifically, for a measuring point i within a partition, assuming that the other similar measuring points in the same partition are 1, 2, 3, ..., n, the correlation coefficient r between the residuals of the other deformation measuring points and the residual of this measuring point is calculated. i (i = 1, 2, 3, ..., n).

[0111] Step 2.2: Determine the composite residual for each partition based on all correlation coefficients within the partition.

[0112] Based on equations (13) and (14), construct the comprehensive residual index for the studied measurement points within the partition:

[0113]

[0114] s 综 = w1×s1+w2×s2+…+w n ×s n (14)

[0115] In the formula: r i w is the correlation coefficient between the residual of the i-th measuring point and the residual of the measuring point under study; i s represents the proportion of the residual at the i-th measurement point in the overall residual; 综 For the comprehensive residual; s i Let be the residual at the i-th measurement point.

[0116] Step 3: Combining the predicted values ​​and the overall residuals, identify outliers in the measured values ​​of each measurement point within each partition, specifically including:

[0117] Step 3.1: Determine the baseline value for each partition using the comprehensive residual. In this embodiment of the invention, the baseline value is 3std(S). 综合 ).

[0118] Step 3.2: Compare the difference between the predicted value and the measured value of each measuring point in each partition with the benchmark value to identify outliers in the measured values ​​of each measuring point in each partition.

[0119] By integrating the comprehensive residual index and the confidence interval method as the standard for outlier identification, when the residual of the monitoring data sequence of the studied measurement point is greater than this standard, it can be identified as an outlier. The discrimination standard is shown in Equation (15):

[0120]

[0121] In the formula: Y i Y is the model's predicted value. i ' represents the measured value; std represents the standard deviation of the residual sequence.

[0122] To verify the effectiveness of the outlier identification method, precision P, recall R, and weighted evaluation index F were selected as evaluation indicators, as shown in equation (16):

[0123]

[0124] In the formula: TP represents the number of correctly identified outliers; FP represents the number of incorrectly identified outliers; TN represents the number of correctly identified normal values; and FN represents the number of incorrectly identified normal values.

[0125] The method of this invention considers the spatiotemporal correlation characteristics of dam deformation and proposes a high-precision dam deformation prediction model, which improves the accuracy of outlier identification in deformation monitoring data and provides a high-precision prediction and outlier identification method for dam deformation.

[0126] The method and system for identifying dam deformation anomalies of the present invention will be described in detail below with more specific embodiments.

[0127] The method for identifying anomalies in dam deformation according to the present invention is used to predict and identify anomalies in the deformation monitoring data of a hydropower station dam. Specifically, it includes the following steps:

[0128] Step 1: Select 21 vertical measuring points on the dam body, and analyze the deformation measurements from September 14, 2012 to July 9, 2021. The time process curves of air temperature and upstream water level are shown below. Figure 2 As shown.

[0129] Step 2: Convert the deformation measurements of each measuring point into a comprehensive distance attribute, and use the Ward method to achieve clustering and partitioning of the measuring points. The partitioning results of the 21 measuring points are as follows: Figure 3 As shown, the radial and horizontal displacement process lines of each measuring point in each zone are as follows: Figure 4 As shown.

[0130] Step 3: Based on dam safety monitoring theory, establish a monitoring model factor set (i.e., environmental impact factor set) for each monitoring point within the zone. The output variable is the deformation sequence δ of the monitoring points, and the input variables are each environmental impact factor. The dam deformation δ is mainly composed of the water pressure component δ. H Temperature component δ T δ, time-dependent component θ Composition, namely:

[0131] δ=δH +δ T +δ θ (17)

[0132]

[0133]

[0134] δ θ =c1(θ-θ0)+c2(lnθ-lnθ0) (20)

[0135] In the formula: a i b 1i b 2i c1 and c2 are the regression coefficients of each component; H0 is the initial water depth; H is the water depth in front of the dam; t0 is the reference time of the monitoring sequence; t is the observation time of the monitoring sequence; θ0 is the time factor of the reference time; θ is the time factor of time t.

[0136] In summary, the HBA-BiLSTM model inputs a total of 10 environmental factors, namely H 1 H 2 H 3 H 4 , θ-θ0, lnθ-lnθ0. To improve the prediction accuracy of the model, the reconstructed data is normalized.

[0137] Step 4: To construct the HBA-BiLSTM model and test its prediction accuracy, three similar measurement points (PL3-1, PL4-1, and PL5-1) in partition I were selected as the research objects. 85% of the monitored data was used as the training set, and 15% as the test set. The HBA algorithm was used to automatically optimize three parameters of the model: the number of hidden nodes, the initial learning rate, and the maximum number of iterations. The optimization ranges were set to [10, 400], [0.0001, 0.005], and [0, 500], respectively, and the miniBatchSize was set to 128. The population size of the HBA algorithm was 50, and the maximum number of iterations was 100.

[0138] To verify the improvement of the HBA-BiLSTM model compared with traditional prediction models, the prediction results of the HBA-BiLSTM model were compared and analyzed with those of Support Vector Regression (SVR), LSTM, and BiLSTM models, and the applicability of the method was verified using experimental data from multiple test points within a partition. Specifically, the number of hidden nodes, initial learning rate, and maximum number of iterations for LSTM and BiLSTM were set to 300, 0.005, and 400, respectively; the penalty factor and kernel function parameters for the SVR model were set to 128 and 0.2, respectively. Figure 5 , Figure 6 , Figure 7 Table 1 compares the model prediction results with the actual deformation sequences for the three measurement points PL3-1, PL4-1, and PL5-1. The evaluation indexes of the prediction results of different models are also compared.

[0139] Table 1 Comparison of Evaluation Indicators for Prediction Results of Different Models

[0140]

[0141] As can be seen from Table 1, the HBA-BiLSTM model of the present invention has higher prediction accuracy for each measurement point.

[0142] Step 5: Calculate the model prediction residual for each measuring point within the partition output by the model, based on the deformation values ​​of each measuring point.

[0143] Step 6: Select the correlation of model prediction residuals between different measuring points within the same partition as the judgment criterion, and construct a dam deformation anomaly discrimination index based on the comprehensive residuals of similar measuring points.

[0144] Step 7: Taking the measured deformation data of PL4-1 measuring point as an example, 68 sets of data from March 20, 2020 to July 9, 2021 were selected. To verify the identification ability of different outlier identification methods, gross errors were added manually at five random locations at the PL4-1 measuring point on July 3, August 21, and December 18, 2020, and March 5 and May 14, 2021. Generally, 2-3 times the standard deviation was selected as the construction standard for the gross errors. To test the high-precision identification ability of different identification methods, 2 times the standard deviation was selected as the construction standard for the gross errors. The constructed comprehensive residual confidence interval method was used to identify outliers in the measured data of PL4-1. The outlier identification results of the PL4-1 measuring point are as follows. Figure 8 As shown in Table 2, to verify the superiority of this identification method, the detection results were compared and analyzed with those of the Rheinda criterion, the Grubbs criterion, the comprehensive residual confidence interval method based on HBA-BiLSTM, and the comprehensive residual confidence interval method based on spatiotemporal clustering of HBA-LSTM. The comparison results are shown in Table 2.

[0145] Table 2 Evaluation Indicators for Outlier Identification Results of PL4-1 Measurement Points Using Different Methods

[0146]

[0147] As can be seen from Table 2, the HBA-BiLSTM integrated residual confidence interval method based on spatiotemporal clustering of this invention has high recognition accuracy.

[0148] A second aspect of the present invention provides a system for identifying anomalies in dam deformation, such as... Figure 9As shown, it includes a residual module 101, a comprehensive residual module 102, and an outlier determination module 103.

[0149] The residual module 101 is used to cluster and partition multiple measuring points on the dam, and obtain the residual between the predicted value and the measured value of each measuring point in each partition;

[0150] The comprehensive residual module 102 is used to determine the comprehensive residual of each partition based on the residuals of all measuring points;

[0151] The outlier determination module 103 is used to combine the predicted value and the comprehensive residual to determine the outliers in the measured values ​​of the measurement points within each partition.

[0152] This invention discloses a method for identifying outliers in dam deformation based on spatiotemporal clustering and deep learning. This method addresses the inevitable occurrence of various outliers in dam deformation monitoring data, which negatively impact dam monitoring data analysis and safety evaluation research. First, the invention utilizes the Ward method to achieve spatiotemporal clustering partitioning of monitoring points. Then, based on dam safety monitoring theory, it establishes a monitoring model factor set for each monitoring point within the partition. This set is used as input to a BiLSTM model optimized by the HBA algorithm for hyperparameter prediction, yielding the predicted deformation value for the monitored point. The model prediction residual for the monitored point is calculated, and outlier identification and verification are completed by determining whether the residual of the monitoring data sequence for the monitored point exceeds the index range. The proposed HBA-BiLSTM model improves deformation prediction accuracy. The HBA-BiLSTM integrated residual confidence interval method based on spatiotemporal clustering demonstrates significant performance improvements in accuracy, reduced false negatives, and lower false positives in dam deformation outlier identification, making it crucial for dam monitoring data analysis and safety evaluation research.

[0153] The above description is merely a few embodiments of this application and is not intended to limit this application in any way. Although this application discloses preferred embodiments as described above, it is not intended to limit this application. Any changes or modifications made by those skilled in the art without departing from the scope of the technical solution of this application using the disclosed technical content are equivalent to equivalent implementation cases and fall within the scope of the technical solution.

Claims

1. A dam deformation outlier identification method, characterized in that, The method comprises the following steps: clustering and partitioning a plurality of measuring points on a dam, and obtaining the residual error between the predicted value and the measured value of each measuring point in each partition; determining the comprehensive residual error of each partition according to the residual error of all measuring points in each partition; determining the abnormal value in the measured value of the measuring point in each partition in combination with the predicted value and the comprehensive residual error; clustering and partitioning a plurality of measuring points on a dam, specifically comprising: obtaining the comprehensive distance between each measuring point on the dam and other measuring points; and clustering and partitioning the plurality of measuring points by using the minimum deviation square system clustering method according to the comprehensive distance; obtaining the comprehensive distance between each measuring point on the dam and other measuring points, specifically comprising: obtaining the absolute distance, incremental distance and acceleration distance between each measuring point on the dam and other measuring points; and determining the comprehensive distance between each measuring point and other measuring points according to the absolute distance, incremental distance and acceleration distance; the absolute distance is determined according to the difference between the deformation value of one measuring point and the deformation value of another measuring point; the incremental distance is determined according to the difference between the deformation increment of one measuring point compared with the previous time and the deformation increment of another measuring point compared with the previous time; and the acceleration distance is determined according to the difference between the deformation acceleration of one measuring point and the deformation acceleration of another measuring point; determining the comprehensive residual error of each partition according to the residual error of all measuring points in each partition, specifically comprising: obtaining the correlation coefficient between the residual error of each measuring point in each partition and the residual error of the remaining measuring points; determining the comprehensive residual error of each partition according to all the correlation coefficients in the partition.

2. The dam deformation outlier identification method according to claim 1, characterized in that, obtaining the residual error between the predicted value and the measured value of each measuring point in each partition, specifically comprising: optimizing the hyperparameters of the prediction model of each partition; training the prediction model of the corresponding partition according to the environmental influence factor of each partition, the historical measured value of each measuring point and the hyperparameters; obtaining the predicted value of each measuring point in the corresponding partition by the prediction model, and determining the residual error between the predicted value and the measured value of each measuring point.

3. The dam deformation outlier identification method of claim 2, wherein, optimizing the hyperparameters of the prediction model of each partition, specifically comprising: optimizing the hyperparameters of the prediction model of each partition by using the meerkat algorithm.

4. The dam deformation outlier identification method of claim 2, wherein The prediction model is a bidirectional long short-term memory neural network model.

5. The dam deformation outlier identification method of claim 1, wherein, determining the abnormal value in the measured value of the measuring point in each partition in combination with the predicted value and the comprehensive residual error, specifically comprising: determining the reference value of each partition by using the comprehensive residual error; comparing the difference between the predicted value and the measured value of each measuring point in each partition with the reference value to determine the abnormal value in the measured value of the measuring point in each partition.

6. A dam deformation outlier identification system based on the dam deformation outlier identification method according to any one of claims 1-5, characterized in that, The method comprises the following steps: a residual error module for clustering and partitioning a plurality of measuring points on a dam, and obtaining the residual error between the predicted value and the measured value of each measuring point in each partition; a comprehensive residual error module for determining the comprehensive residual error of each partition according to the residual error of all measuring points in each partition; an abnormal value determination module for determining the abnormal value in the measured value of the measuring point in each partition in combination with the predicted value and the comprehensive residual error.

Images