Slope real-time monitoring and early warning device, equipment, medium and product
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- HUANENG LANCANG RIVER HYDROPOWER CO LTD
- Filing Date
- 2026-02-03
- Publication Date
- 2026-05-29
AI Technical Summary
The existing slope safety monitoring system mainly relies on manual interpretation, which makes it difficult to achieve real-time, automated and intelligent monitoring and early warning. Furthermore, it cannot perform quantitative analysis of multi-source monitoring data, resulting in inconsistent judgment standards and lagging data processing.
A Gaussian process regression (GPR) model was constructed using multi-source monitoring data. By screening influencing parameters, early warning levels were generated. Data fitting and cluster analysis were performed using radial basis function kernel function to achieve intelligent identification and early warning of abnormal slope monitoring points.
It enables intelligent, efficient and accurate early warning for real-time slope monitoring, supports safety management throughout the entire life cycle of slopes, adapts to multi-source data input, and improves monitoring accuracy and emergency response capabilities.
Smart Images
Figure CN122116560A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of engineering safety monitoring technology, and in particular to a real-time monitoring and early warning device, equipment, medium and product for slopes. Background Technology
[0002] Slope safety monitoring is a crucial aspect of geological disaster prevention and safety management during engineering construction and operation. During construction, slopes are constantly changing due to excavation, support installation, and other engineering activities. During operation, slopes are further affected by multiple factors, including rainfall, reservoir water level changes, long-term creep effects, earthquakes, and operational loads, exhibiting complex spatiotemporal evolution characteristics. Achieving continuous, accurate, and real-time identification and early warning of slope deformation during both construction and operation is one of the core challenges in the safety management of major projects such as water conservancy, hydropower, and transportation infrastructure.
[0003] Existing slope safety monitoring systems mostly rely on manual interpretation, typically involving manual comparison of monitoring curves and expert experience to identify anomalies and issue risk warnings. This approach has significant shortcomings: First, it depends on individual expert experience, making it difficult to standardize judgment criteria and leading to significant discrepancies in conclusions among different personnel. Second, with the widespread adoption of multi-source monitoring equipment, the amount of data is growing exponentially, making it difficult to complete manual processing and comparison in a timely manner, and failing to meet the real-time and automated requirements of modern engineering, especially slope digital twin systems. Third, traditional methods are mostly "static" data interpretations, unable to quantitatively analyze fluctuations and potential anomalies in monitoring data, and thus unable to support the needs for intelligent and lightweight early warning systems. Summary of the Invention
[0004] This invention provides a method, device, equipment, and storage medium for real-time monitoring and early warning of slopes, enabling intelligent, efficient, and accurate real-time monitoring and early warning of slopes. The technical solution includes at least the following components: Firstly, a method for real-time monitoring and early warning of slopes is provided, comprising: acquiring multi-source monitoring data of various measuring points in a target slope to obtain a monitoring dataset, wherein the monitoring dataset includes a historical monitoring dataset and a real-time monitoring dataset; filtering out influencing parameters of the target monitoring parameters from the monitoring dataset, wherein the influencing parameters are parameters that affect the target monitoring parameters; constructing a GPR model based on the historical influencing parameters and historical target monitoring parameters in the historical monitoring dataset; inputting the real-time influencing parameters in the real-time monitoring dataset into the GPR model to obtain a reference mean fitted by the GPR model; determining multiple abnormal measuring points in the target slope at the current time based on the difference between the real-time target monitoring parameters of various measuring points in the target slope and the corresponding reference mean, wherein the real-time target monitoring parameters are the target monitoring parameters in the real-time monitoring dataset; clustering the multiple abnormal measuring points according to their spatial location to obtain multiple clusters, and generating an early warning level for each cluster.
[0005] Optionally, the step of filtering the impact parameters of the target monitoring parameter from the monitoring dataset includes: using a feature selection algorithm to filter the impact parameters of the target monitoring parameter from the monitoring dataset, wherein the feature selection algorithm includes a correlation analysis algorithm, a principal component analysis algorithm, and a mutual information algorithm; and when the number of impact parameter types is greater than 1, using an automatic correlation dimension scaling mechanism to assign an independent length scale parameter to each type of impact parameter.
[0006] Optionally, generating an early warning level for each cluster includes: obtaining a quality score corresponding to each sample in the monitoring dataset, wherein the quality score is obtained based on short-time SNR, short-term drift rate, and missing rate; calculating the confidence level of each cluster based on the average quality score of each abnormal measurement point within each cluster, the average standardized residual, and the spatial density; and generating an early warning level for each cluster based on the confidence level of each cluster.
[0007] Optionally, the kernel function of the GPR is a radial basis function kernel, which is expressed by the following formula:
[0008] in, This refers to the real-time impact parameter of a measuring point in the real-time monitoring dataset. The historical impact parameter of a measuring point in the historical monitoring dataset. The output of the radial basis function kernel is represented by the covariance matrix. for The corresponding length scale parameter, For signal variance, This refers to the Kronecker delta function.
[0009] Optionally, the reference mean fitted by the GPR model is expressed by the following formula:
[0010] in, This is the reference mean fitted by the GPR model. This is a feature vector composed of the real-time influence parameters of each measuring point in the real-time monitoring dataset. This is a feature vector composed of the historical impact parameters of each measuring point at different times in the historical monitoring dataset. This is a feature vector composed of the historical target monitoring parameters of each measuring point in the historical monitoring dataset. It is an identity matrix. This represents the noise variance.
[0011] Optionally, the method further includes: displaying the warning level of each abnormal measuring point in the digital twin model of the target slope, wherein the warning level of any abnormal measuring point is the warning level of the cluster to which the abnormal measuring point belongs.
[0012] Secondly, a real-time slope monitoring and early warning device is also provided, comprising: an acquisition module for acquiring multi-source monitoring data of various measuring points in a target slope to obtain a monitoring dataset, wherein the monitoring dataset includes a historical monitoring dataset and a real-time monitoring dataset; a filtering module for filtering out the impact parameters of the target monitoring parameters from the monitoring dataset, wherein the impact parameters are parameters that affect the target monitoring parameters; a modeling module for constructing a GPR model based on the historical impact parameters and historical target monitoring parameters in the historical monitoring dataset; a fitting module for inputting the real-time impact parameters in the real-time monitoring dataset into the GPR model to obtain a reference mean fitted by the GPR model; an abnormal measuring point determination module for determining multiple abnormal measuring points in the target slope at the current time based on the difference between the real-time target monitoring parameters of each measuring point in the target slope and the corresponding reference mean, wherein the real-time target monitoring parameters are the target monitoring parameters in the real-time monitoring dataset; and an early warning module for clustering the multiple abnormal measuring points according to their spatial location to obtain multiple clusters and generating an early warning level for each cluster.
[0013] Optionally, the filtering module is further configured to use a feature selection algorithm to filter out the influencing parameters of the target monitoring parameters from the monitoring dataset. The feature selection algorithm includes a correlation analysis algorithm, a principal component analysis algorithm, and a mutual information algorithm. When the number of types of influencing parameters is greater than 1, an automatic correlation dimension scaling mechanism is used to assign an independent length scale parameter to each type of influencing parameter.
[0014] Optionally, the early warning module is further configured to obtain the quality score corresponding to each sample in the monitoring dataset, the quality score being obtained based on short-time SNR, short-term drift rate, and missing rate; calculate the confidence level of each cluster based on the average quality score of each abnormal measurement point within each cluster, the average standardized residual, and the spatial density; and generate an early warning level for each cluster based on the confidence level of each cluster.
[0015] Optionally, in the modeling module, the kernel function of the GPR is a radial basis function kernel, which is expressed by the following formula:
[0016] in, This refers to the real-time impact parameter of a measuring point in the real-time monitoring dataset. The historical impact parameter of a measuring point in the historical monitoring dataset. The output of the radial basis function kernel is represented by the covariance matrix. for The corresponding length scale parameter, For signal variance, This refers to the Kronecker delta function.
[0017] Optionally, in the modeling module, the reference mean fitted by the GPR model is represented by the following formula:
[0018] in, This is the reference mean fitted by the GPR model. This is a feature vector composed of the real-time influence parameters of each measuring point in the real-time monitoring dataset. This is a feature vector composed of the historical impact parameters of each measuring point at different times in the historical monitoring dataset. This is a feature vector composed of the historical target monitoring parameters of each measuring point in the historical monitoring dataset. It is an identity matrix. This represents the noise variance.
[0019] Optionally, the early warning module is also used to display the early warning level of each abnormal measuring point in the digital twin model of the target slope, wherein the early warning level of any abnormal measuring point is the early warning level of the cluster to which the abnormal measuring point belongs.
[0020] Thirdly, a computer device is also provided, comprising: a memory and a processor, wherein the memory stores at least one computer program, the at least one computer program being loaded and executed by the processor to perform the slope real-time monitoring and early warning method described in the above embodiments.
[0021] Fourthly, a computer-readable storage medium is also provided, wherein at least one computer program is stored in the computer-readable storage medium, and the at least one computer program is loaded and executed by a processor to perform the slope real-time monitoring and early warning method described in the above embodiments.
[0022] Fifthly, a computer program product is provided, including a computer program / instructions that, when executed by a processor, implement the method described in the first aspect.
[0023] The beneficial effects of the technical solution provided by this invention include at least the following: In this embodiment, a monitoring dataset is obtained by acquiring multi-source monitoring data from various measuring points in the target slope. This dataset includes historical and real-time monitoring datasets. Influence parameters on the target monitoring parameters are selected from the monitoring dataset; these parameters are those that affect the target monitoring parameters. A Geometric Parameter Recognition (GPR) model is constructed based on the historical influence parameters and historical target monitoring parameters in the historical monitoring dataset. The real-time influence parameters from the real-time monitoring dataset are input into the GPR model to obtain the reference mean fitted by the GPR model. Based on the difference between the real-time target monitoring parameters and the corresponding reference mean at each measuring point in the target slope, multiple abnormal measuring points in the target slope at the current moment are identified. The real-time target monitoring parameters are the target monitoring parameters in the real-time monitoring dataset. These abnormal measuring points are clustered according to their spatial location to obtain multiple clusters, and an early warning level is generated for each cluster. This achieves intelligent, efficient, and accurate real-time slope monitoring and early warning, providing a comprehensive, real-time, and intelligent monitoring method for the safety management of slopes throughout their entire lifecycle. Attached Figure Description
[0024] To more clearly illustrate the technical solutions in this embodiment, the accompanying drawings used in the description of the embodiment will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0025] Figure 1 A flowchart of a real-time slope monitoring and early warning method provided by an exemplary embodiment of the present invention is shown; Figure 2 This is a schematic diagram of the real-time monitoring and early warning results of the target slope when the target monitoring parameter is the slope displacement collected by the surface point displacement gauge; Figure 3 This is a visualization of the real-time slope monitoring and early warning method provided by an exemplary embodiment of the present invention in a digital twin model of a slope in a water conservancy and hydropower project; Figure 4 This diagram illustrates the structure of a real-time slope monitoring and early warning device provided in an exemplary embodiment of the present invention. Figure 5 This is a schematic diagram of the structure of a computer device provided in an exemplary embodiment of the present invention. Detailed Implementation
[0026] Unless otherwise defined, the technical or scientific terms used herein shall have the ordinary meaning as understood by one of ordinary skill in the art to which this invention pertains. The terms “first,” “second,” “third,” and similar terms used in this patent application specification and claims do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Similarly, the terms “an” or “a” and similar terms do not indicate a quantity limitation, but rather indicate the presence of at least one. The terms “comprising” or “including” and similar terms mean that the elements or objects preceding “comprising” or “including” encompass the elements or objects listed following “comprising” or “including” and their equivalents, but do not exclude other elements or objects.
[0027] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.
[0028] Example 1.
[0029] Figure 1 A flowchart illustrating a real-time slope monitoring and early warning method provided by an exemplary embodiment of the present invention is shown. This method can be executed by a computer device. See also Figure 1 The method includes: In step 101, multi-source monitoring data of each measuring point in the target slope are obtained to obtain a monitoring dataset.
[0030] The monitoring dataset includes historical monitoring data and real-time monitoring data.
[0031] In this embodiment, the types of multi-source monitoring data include, but are not limited to: settlement, displacement and tilt data of each measuring point on the slope; external environmental data; slope construction data; and overall geometric changes and displacement field data of the slope.
[0032] The settlement, displacement and tilt data of each measuring point on the slope can be obtained by sensors such as surface displacement gauges, multi-point displacement gauges, inclinometers, fiber optic sensors, and internal borehole displacement gauges installed at each measuring point.
[0033] External environmental data includes rainfall, reservoir water level changes, temperature, etc., and can be obtained through weather statistics websites.
[0034] Slope construction data includes construction logs and other data.
[0035] Data on the overall geometric changes and displacement fields of slopes can be obtained through UAV aerial imagery, laser point cloud scanning, InSAR, and satellite remote sensing data.
[0036] If the current time is time t, then we need to obtain multi-source monitoring data from the start of slope construction (t=0) to time t-1 as the historical monitoring dataset, and use the multi-source monitoring data obtained at time t as the real-time monitoring dataset.
[0037] Historical monitoring datasets can be used to train GPR (Gaussian Process Regression) models, while real-time monitoring datasets are used to identify abnormal measuring points on the slope at the current moment.
[0038] Therefore, each sample in this monitoring dataset represents the multi-source monitoring data corresponding to a single measuring point on the target slope at a given time. For the same measuring point, the monitoring dataset should include multi-source monitoring data from multiple consecutive times at that measuring point. In other words, for the same measuring point, the multi-source monitoring data from different times in the monitoring dataset should be considered as time-series data.
[0039] Before performing step 102, the samples in the monitoring dataset need to be preprocessed.
[0040] Optionally, preprocessing includes: timestamp alignment, noise reduction, and normalization.
[0041] Timestamp alignment is used to ensure temporal consistency of data from different devices. Denoising processes, such as moving average filtering, median filtering, wavelet denoising, or empirical mode decomposition (EMD), are used to remove sensor noise and isolated outliers. Standardization processes, such as Z-score normalization, are used to eliminate dimensional differences between different types of monitoring data.
[0042] Optionally, a Quality Score (QS) can be calculated for any sample in the monitoring dataset. This Quality Score can be used to reflect the observation confidence in the output of subsequent GPR models as a weighted or heteroscedastic observation variance.
[0043] In this embodiment, the quality score of any sample is obtained by weighted summation of the sample's short-term SNR, short-term drift rate, and missing rate. In implementation, different weights are pre-assigned to the short-term SNR, short-term drift rate, and missing rate. Then, the sample's short-term SNR, short-term drift rate, and missing rate are calculated, and then weighted summation is performed according to the preset weights to obtain the sample's quality score. The quality score ranges from [0,1].
[0044] In this embodiment, the historical monitoring dataset is used to train the GPR model. Data augmentation processing can also be performed on the historical monitoring dataset to improve the training effect of the GPR model. For example, a synthetic injection experiment is conducted on the historical monitoring dataset. This synthetic injection experiment includes, but is not limited to, artificially injecting linear accelerations, abrupt changes, or impulse anomalies into the historical monitoring dataset according to a preset pattern. This sets special samples in the historical monitoring dataset, achieving data augmentation. Such special samples can evaluate and optimize the anomaly detection sensitivity, false alarm rate, and average detection latency of the GPR model during training.
[0045] In step 102, the influencing parameters of the target monitoring parameters are selected from the multi-source monitoring data. The influencing parameters are the parameters that affect the target monitoring parameters.
[0046] Here, the target monitoring parameters are manually set and can be configured according to needs. For example, if it is necessary to monitor the settlement data of a target slope, the target monitoring parameters can be set to the settlement data of the target slope.
[0047] It should be noted that target monitoring parameters can be parameters collected from the target slope, such as settlement, displacement, and tilt data at various measuring points on the slope, slope construction data, geometric changes at the measuring point locations, and displacement field data. However, target monitoring parameters cannot be external environmental data or overall geometric changes and displacement field data of areas other than the measuring point locations. This is because external environmental data cannot directly indicate whether there is a safety issue at a specific measuring point on the slope. For overall geometric changes and displacement field data, only data at the measuring point locations can directly reflect the true state of the measuring point. Geometric changes and displacement field data of areas other than the measuring point locations are part of the overall assessment of the target slope and cannot directly determine whether there is an anomaly at a specific measuring point. Therefore, external environmental data and overall geometric changes and displacement field data of the slope outside the measuring point locations can only exist as influencing parameters.
[0048] Optionally, step 102 includes: using a feature selection algorithm to filter out the influencing parameters of the target monitoring parameters from the monitoring dataset. The feature selection algorithm includes correlation analysis algorithms, principal component analysis algorithms, and mutual information algorithms. Implementation methods for feature selection algorithms are widely available in related technologies and will not be detailed here.
[0049] When the number of influence parameter types is greater than 1, an independent length scale parameter is assigned to each type of influence parameter using the Automatic Relevance Determination (ARD) mechanism.
[0050] Here, if the number of influencing parameter types is greater than one, it indicates the existence of at least two types of influencing parameters. Since different types of influencing parameters have varying degrees of impact on the target monitoring parameters, it is necessary to assign an independent length scale parameter to each type of influencing parameter. In this embodiment, the ARD mechanism is used to assign an independent length scale parameter to each type of influencing parameter.
[0051] The ARD mechanism allows the model to learn to assign an independent length scale parameter to different input features, automatically reflecting the importance of the feature to the model's prediction. A larger length scale parameter value indicates a smaller impact of the feature on the prediction result, thus allowing less important features to be ignored. Detailed explanations of the implementation of the ARD mechanism are available in various related technologies and will not be elaborated upon here.
[0052] In step 103, a GPR model is constructed based on the historical impact parameters and historical target monitoring parameters in the historical monitoring dataset.
[0053] In this embodiment, when constructing the GPR model, it is necessary to pre-set the hyperparameters of the GPR model, which include the length scale parameter and the observation noise level parameter. After setting the hyperparameters, it is also necessary to perform hyperparameter boundary checks and uncertainty calibration on the GPR model.
[0054] Optionally, hyperparameter boundary checks include: determining whether the kernel function length-scale parameter and observation noise parameter of the GPR are within a preset boundary range. If the kernel function length-scale parameter and observation noise parameter of the GPR are outside the preset boundary range, boundary expansion, kernel function replacement, or data reprocessing are performed.
[0055] If the kernel function length scale parameter and observation noise parameter of GPR are outside the preset boundary range, it indicates that there may be three situations: (1) the preset boundary range is not set reasonably; (2) the GPR kernel function is not selected reasonably; (3) the training data is not processed reasonably (e.g., there is noise). Different methods need to be used to deal with different situations. For situation (1), boundary expansion processing can be performed; for situation (2), the kernel function of the GPR model can be replaced; for situation (3), the training data (historical monitoring dataset) can be reprocessed to optimize the unreasonable parts of the training data (such as noise).
[0056] In this embodiment, uncertainty calibration is used to calibrate the uncertainty of the GPR model. In implementation, the Prediction Interval Coverage Probability (PICP) can first be estimated based on the retention set or time series cross-validation. Then, the PICP value is used to determine whether the uncertainty of the GPR model needs calibration. If the PICP value indicates that the uncertainty of the GPR model needs calibration, the uncertainty can be calibrated by scaling the standard deviation or using a heteroscedasticity model, thereby making the uncertainty of the GPR model more accurately reflect the variability of the measurement points.
[0057] There are many related technologies regarding methods for determining whether the uncertainty of a GPR model needs calibration based on the PICP value, and for calibrating the uncertainty of a GPR model by scaling the standard deviation or using a heteroscedasticity model. These methods will not be detailed here.
[0058] Then, the GPR model can be trained using the aforementioned historical monitoring dataset. Since the types of target monitoring parameters and impact parameters have been selected in step 102, historical target monitoring parameters and historical impact parameters can be directly obtained from the historical monitoring dataset. Then, the historical target monitoring parameters and historical impact parameters at the same time are taken as a sample, and the historical impact parameters in the sample are input into the GPR model to train the GPR model. The historical target monitoring parameters in the sample are stored as labels.
[0059] For example, the target monitoring parameters and historical impact parameters of measuring point a in the historical monitoring dataset at time t-1 constitute a sample. The historical impact parameters of this sample need to be input into the GPR model, and the target monitoring parameters of this sample exist as labels.
[0060] The kernel function of the GPR model in this embodiment includes, but is not limited to, the radial basis function (RBF). Kernel functions, such as Rational Quadratic, can also be combined with constant terms, linear trend terms, and white noise terms.
[0061] Optionally, when the kernel function of the GPR model is the RBF kernel function, the radial basis function kernel is represented by the following formula (1).
[0062] (1) In company (1), To monitor the real-time impact parameters of a measurement point in the dataset, For a historical impact parameter of a measuring point in the historical monitoring dataset, The output of the radial basis function kernel is represented by the covariance matrix. for The corresponding length scale parameter, For signal variance, This refers to the Kronecker delta function.
[0063] In this case, the reference mean fitted by the GPR model is represented by formula (2), and the uncertainty of the reference mean is represented by formula (3).
[0064] (2) In formula (2), This is the reference mean fitted by the GPR model. To monitor the feature vector formed by the real-time influence parameters of each measuring point in the dataset, This is a feature vector composed of the historical impact parameters of each measuring point at different times in the historical monitoring dataset. This is a feature vector composed of the historical target monitoring parameters of each measuring point in the historical monitoring dataset. It is an identity matrix. Let V be the noise variance. The meanings of the other parameters in formula (2) are the same as those in formula (1), and will not be elaborated here.
[0065] (3) In formula (3), Is with The corresponding uncertainty, This is the uncertainty reduction term. The meanings of the other parameters in formula (3) are the same as those in formula (2), and will not be elaborated here.
[0066] The input to the GPR model is the influencing parameter, and the output of the GPR model is the reference mean of the target monitoring parameter under the influencing parameter and the uncertainty (standard deviation) of the reference mean.
[0067] During training, the uncertainty output by the GPR model can be verified using a retention set or PICP to further calibrate the uncertainty of the reference mean.
[0068] The samples at multiple different times corresponding to the same measuring point in the historical monitoring dataset are essentially a set of time series data. Therefore, after training the GPR model with the historical monitoring dataset, the GPR model can learn and capture the nonlinear relationship changes between the influencing parameters and the target monitoring parameters in the time series data, so as to accurately fit the reference mean of the target monitoring parameters according to the input influencing parameters.
[0069] In step 104, the real-time impact parameters in the real-time monitoring dataset are input into the GPR model to obtain the reference mean fitted by the GPR model.
[0070] The GPR model in step 104 is the trained GPR model, which can accurately fit the reference mean of the target monitoring parameters based on the input influence parameters. Therefore, when the real-time influence parameters in the real-time monitoring dataset are input into the GPR model, the reference mean fitted by the GPR model is equivalent to an accurate and reliable reference mean of the target monitoring parameters at each measuring point at the current time.
[0071] It should be noted that, to simplify calculations, in practical applications, the real-time impact parameters of each measuring point in the real-time monitoring dataset are integrated into an input matrix and input into the GPR model. One dimension of this input matrix corresponds to the real-time impact parameter of a measuring point. Similarly, the reference mean output by the GPR model is also a reference mean matrix, where one dimension corresponds to the reference mean of the target monitoring parameter at the current time for a measuring point. The uncertainty is handled similarly; the uncertainty output by the GPR model is an uncertainty matrix, where each dimension represents the uncertainty of the corresponding dimension in the reference mean matrix. In other words, the GPR model actually fits a reference mean for each measuring point.
[0072] In step 105, based on the difference between the real-time target monitoring parameters of each measuring point in the target slope and the corresponding reference mean, multiple abnormal measuring points in the target slope are identified.
[0073] Real-time target monitoring parameters are the target monitoring parameters in the real-time monitoring dataset.
[0074] The real-time target monitoring parameters at each measuring point are the actual measured values in the slope, while the reference mean for each measuring point is the ideal value of the target monitoring parameter corresponding to the current influencing parameter under safe conditions, fitted by the GPR model. If the difference between the real-time target monitoring parameter and the reference mean at a certain measuring point is small, it indicates that there is no anomaly at that measuring point; if the difference between the real-time target monitoring parameter and the reference mean at a certain measuring point is large, it indicates that there is anomaly at that measuring point.
[0075] The process of determining the difference between the real-time target monitoring parameters and the corresponding reference mean at each measuring point in the target slope is essentially a significant deviation detection process. When performing significant deviation detection, a standard statistic for deviation detection needs to be set. In this embodiment, standardized residual values are used as the standard statistic for significant deviation detection. If the standardized residual value of a measuring point is greater than or equal to the residual threshold, it indicates that the measuring point is an abnormal measuring point; if the standardized residual value of a measuring point is less than the residual threshold, it indicates that the measuring point is not an abnormal measuring point but a normal measuring point. The residual threshold is taken as an empirical value, such as 2 or 3.
[0076] The standardized residual of any measuring point is calculated using formula (4).
[0077] (4) In formula (4), Let i be the standardized residual of the i-th measurement point. Let i be the real-time target monitoring parameters for the i-th measuring point. The reference mean of the i-th measurement point in the reference mean matrix output by the GPR model. Let represent the uncertainty of the reference mean at the i-th measurement point in the uncertainty matrix output by the GPR model. The reference mean matrix output by the GPR model is calculated using formula (2), and the uncertainty matrix output by the GPR model is calculated using formula (3), which will be omitted here.
[0078] Furthermore, since slope monitoring typically involves multiple measuring points (i.e., multi-point scenarios), multiple validation control methods are needed to control the overall false detection rate of significant deviations. For example, the Benjamini–Hochberg FDR method can be used to control the overall false detection rate of significant deviations.
[0079] To streamline the slope safety monitoring process, the following methods can be used to improve efficiency: single-point GPR parallel computation for each monitoring point; limiting the sliding training window size to restrict the amount of data required for each retraining session; and using sparse sample / induced point approximation on resource-constrained equipment. This reduces computational overhead and achieves a streamlined slope safety monitoring process.
[0080] Step 105 enables rapid identification of single-point or short-term abnormal displacement trends, providing immediate early warning of potential slope risks.
[0081] In step 106, multiple abnormal measurement points are clustered according to their spatial location to obtain multiple clusters, and an early warning level is generated for each cluster.
[0082] Optionally, the clustering algorithm used when clustering multiple outlier measurement points includes, but is not limited to, density-based clustering algorithms such as DBSCAN and HDBSCAN.
[0083] Optionally, step 106 includes steps 1061 to 1063 as follows.
[0084] Step 1061: Obtain the quality score corresponding to each sample in the monitoring dataset.
[0085] For details regarding the mass fraction, please refer to step 101 above, which will not be elaborated here.
[0086] Step 1062: Calculate the confidence level of each cluster based on the average quality score of each outlier measurement point within each cluster, the average standardized residual, and the spatial density.
[0087] For example, if there are M outlier measurement points in the j-th cluster, then we need to calculate the average quality score of these M outlier measurement points, the average standardized residual of these M outlier measurement points (the absolute value of each standardized residual needs to be taken before calculating the average of the standardized residuals), and the spatial density of the j-th cluster. Then, we perform a weighted summation of these three indicators to obtain the confidence score of the j-th cluster. The weights of each indicator can be preset based on experience when performing the weighted summation.
[0088] For all clusters except the j-th cluster, the confidence level can be calculated using the above method, thus obtaining the confidence level of each cluster.
[0089] Step 1063: Generate an early warning level for each cluster based on the confidence level of each cluster.
[0090] In this embodiment, different warning levels correspond to different confidence level ranges. For example, if the confidence level range is [a, b], and there are a total of N warning levels (N is a positive integer), then [a, b] can be divided into N equal ranges, each range corresponding to one warning level. The warning level of any cluster is the warning level corresponding to the confidence level range of that cluster.
[0091] The essence of step 106 is that if there are many anomalous measuring points in a certain area, since these anomalous measuring points are spatially close (belonging to the same area), they will cluster together. The confidence level of this cluster reflects the reliability of the multiple anomalous measuring points in the area. The higher the confidence level, the more reliable these anomalous measuring points are, meaning that these anomalous measuring points are all locations with slope instability risks, and the corresponding warning level will be higher.
[0092] Optionally, the method further includes displaying the warning level of each abnormal measuring point in a digital twin model of the target slope.
[0093] When implementing steps 101 to 106 above, the relevant parameters of each abnormal measuring point can be encapsulated as abnormal events, and then the abnormal events can be pushed to the digital twin model of the target slope through a standardized interface, thereby displaying the warning level of each abnormal measuring point in the digital twin model of the target slope.
[0094] For example, the interfaces used to access abnormal events in the digital twin model include, but are not limited to, RESTful APIs or message queue interfaces (such as MQTT, Kafka, etc.).
[0095] The content of any abnormal event includes, but is not limited to, fields such as abnormal event ID, timestamp, list of abnormal measurement points, coordinates of each abnormal measurement point, standardized residual value of each abnormal measurement point, quality score, ID of the cluster to which each abnormal measurement point belongs, confidence level of each cluster, and suggested handling.
[0096] Abnormal events can be encapsulated in a standardized file format and sent to the digital twin model. This standardized file format is, for example, JSON.
[0097] Upon receiving anomaly events, the digital twin model of the target slope can display various anomalous measuring points in real time through views such as 3D heatmaps, profile diagrams, time-series curves, and event lists. It supports manual verification and write-back of handling records for subsequent model improvement and system closed-loop management. For example, after the anomalous measuring points displayed by the digital twin model have been manually confirmed or handled, verification results can be obtained. These verification results can be used as tags or feedback and written back to the monitoring dataset for subsequent model performance evaluation and synthetic injection test parameter adjustment.
[0098] During the operation of the digital twin model, steps 101 to 106 are executed cyclically based on newly collected real-time monitoring data. The kernel function parameters and noise terms of the Gaussian process regression model are adjusted according to real-time data feedback (such as verification results) to ensure accuracy. This further enables continuous operation of rolling updates and anomaly detection, thus achieving closed-loop optimization of slope monitoring. This closed-loop optimization, while maintaining lightweight computation, meets the requirements of the slope digital twin system for real-time monitoring and rapid response, realizing intelligent monitoring and early warning throughout the entire lifecycle of construction and operation. The kernel function parameters and noise terms of the Gaussian process regression model are adjusted based on real-time data feedback.
[0099] Figure 2 This diagram illustrates the real-time monitoring and early warning results of a target slope, where the target monitoring parameter is the slope displacement collected by a surface point displacement gauge. Figure 2Part (a) shows the normal real-time monitoring and early warning results for the target slope. Figure 2 Part (b) describes the abnormal situation in the real-time early warning results of the target slope.
[0100] Figure 2 In the diagram, the circular dots represent target monitoring data, specifically the data collected by the surface point displacement gauge at a given measuring point. Gray circles represent historical target monitoring parameters, while red circles represent real-time target monitoring parameters. The curve represents the reference mean value at the corresponding time point fitted by the GPR model. The horizontal axis represents time, and the vertical axis represents the slope displacement measured by the surface point displacement gauge.
[0101] It can be seen that, Figure 2 In part (a), most of the target monitoring parameters are located near the reference mean and deviate little from the reference mean, indicating that the real-time monitoring and early warning results of the target slope are normal. Figure 2 In part (b), there are target monitoring parameters that deviate significantly from the reference mean (such as the real-time target monitoring parameters indicated by point 201). Point 201 indicates that the real-time early warning result of the target slope is abnormal, that is, under the current circumstances, the measuring point where the surface displacement meter is located is an abnormal measuring point.
[0102] Figure 3 This is a visualization of the real-time slope monitoring and early warning method provided by an exemplary embodiment of the present invention in a digital twin model of a slope in a water conservancy and hydropower project. For example... Figure 3 As shown, in the digital twin model, measuring points with different warning levels are visualized using different colors. Red and yellow points are abnormal measuring points, while blue points are normal measuring points. The warning level indicated by red is higher than that indicated by yellow.
[0103] In this embodiment, through steps 101 to 106, dynamic monitoring and anomaly identification of the slope are achieved. This method can continuously respond to real-time monitoring changes in the slope during construction and operation, and monitor changes in slope displacement and environmental factors. It can promptly identify potential risk points, provide early warning information, and support real-time slope monitoring in the digital twin system of water conservancy and hydropower projects, thereby providing a comprehensive, real-time, and intelligent monitoring method for the safety management of slopes throughout their entire lifecycle.
[0104] Furthermore, this real-time slope monitoring and early warning method is adaptable to multi-source data input, including traditional sensor data, UAV imagery, laser point clouds, and remote sensing data. It enables cross-device and cross-scale information fusion, improving monitoring accuracy and reliability, and significantly enhancing the scientific nature of slope safety management and emergency response capabilities. Through the real-time slope monitoring and early warning method in this embodiment, managers can dynamically grasp the slope status during construction and operation phases, optimize support and maintenance plans, reduce engineering risks, and improve the safety and operational reliability of major slopes in water conservancy and hydropower projects.
[0105] The following are device embodiments of this application. For details not described in detail in the device embodiments, please refer to the above method embodiments.
[0106] Example 2.
[0107] Figure 4 A schematic diagram of a real-time slope monitoring and early warning device provided in an exemplary embodiment of the present invention is shown. See also... Figure 4 The slope real-time monitoring and early warning device 400 includes: an acquisition module 401, a screening module 402, a modeling module 403, a fitting module 404, an abnormal measuring point determination module 405, and an early warning module 406.
[0108] The acquisition module 401 is used to acquire multi-source monitoring data of various measuring points in the target slope to obtain a monitoring dataset, which includes historical monitoring dataset and real-time monitoring dataset. The filtering module 402 is used to filter out the influencing parameters of the target monitoring parameters from the monitoring dataset. The influencing parameters are the parameters that affect the target monitoring parameters. Modeling module 403 is used to build a GPR model based on historical impact parameters and historical target monitoring parameters in the historical monitoring dataset; The fitting module 404 is used to input the real-time impact parameters in the real-time monitoring dataset into the GPR model to obtain the reference mean fitted by the GPR model. The abnormal measuring point determination module 405 is used to determine multiple abnormal measuring points in the target slope at the current time based on the difference between the real-time target monitoring parameters of each measuring point in the target slope and the corresponding reference mean. The real-time target monitoring parameters are the target monitoring parameters in the real-time monitoring dataset. The early warning module 406 is used to cluster multiple abnormal measurement points according to their spatial location, obtain multiple clusters, and generate an early warning level for each cluster.
[0109] Optionally, the screening module 402 is also used to screen out the influencing parameters of the target monitoring parameters from the monitoring dataset using a feature selection algorithm, including correlation analysis algorithm, principal component analysis algorithm, and mutual information algorithm; when the number of types of influencing parameters is greater than 1, an automatic correlation dimension scaling mechanism is used to assign an independent length scale parameter to each type of influencing parameter.
[0110] Optionally, the early warning module 406 is also used to obtain the quality score corresponding to each sample in the monitoring dataset. The quality score is obtained based on the short-term SNR, short-term drift rate, and missing rate. The confidence level of each cluster is calculated based on the average quality score of each abnormal measurement point in each cluster, the average standardized residual, and the spatial density. An early warning level is generated for each cluster based on the confidence level of each cluster.
[0111] Optionally, in modeling module 403, the kernel function of GPR is a radial basis function kernel, which is expressed by the following formula:
[0112] in, To monitor the real-time impact parameters of a measurement point in the dataset, For a historical impact parameter of a measuring point in the historical monitoring dataset, The output of the radial basis function kernel is represented by the covariance matrix. for The corresponding length scale parameter, For signal variance, This refers to the Kronecker delta function.
[0113] Optionally, in modeling module 403, the reference mean fitted by the GPR model is expressed using the following formula:
[0114] in, This is the reference mean fitted by the GPR model. To monitor the feature vector formed by the real-time influence parameters of each measuring point in the dataset, This is a feature vector composed of the historical impact parameters of each measuring point at different times in the historical monitoring dataset. This is a feature vector composed of the historical target monitoring parameters of each measuring point in the historical monitoring dataset. It is an identity matrix. This represents the noise variance.
[0115] Optionally, the early warning module 406 is also used to display the early warning level of each abnormal measuring point in the digital twin model of the target slope, wherein the early warning level of any abnormal measuring point is the early warning level of the cluster to which the abnormal measuring point belongs.
[0116] It should be noted that the slope real-time monitoring and early warning device provided in the above embodiments is only illustrated by the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. In addition, the slope real-time monitoring and early warning device and the slope real-time monitoring and early warning method embodiments provided in the above embodiments belong to the same concept, and their specific implementation process can be found in the method embodiments, which will not be repeated here.
[0117] The module division in this embodiment of the invention is illustrative and represents only one logical functional division. In actual implementation, other division methods are possible. Furthermore, the functional modules in each embodiment of the invention can be integrated into a single processor, exist as separate physical entities, or consist of two or more modules integrated into one module. The integrated modules described above can be implemented in hardware or as software functional modules.
[0118] If the integrated module is implemented as a software functional module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a terminal device (which may be a personal computer, mobile phone, or communication device, etc.) or processor to execute all or part of the steps of the method of the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.
[0119] Figure 5 This is a schematic diagram of the structure of a computer device provided in an exemplary embodiment of the present invention. For example... Figure 5 As shown, the computer device 500 includes a processor 501 and a memory 502.
[0120] Processor 501 may include one or more processing cores, such as a quad-core processor, an octa-core processor, etc. Processor 501 may be implemented using at least one hardware form selected from DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), and PLA (Programmable Logic Array). Processor 501 may also include a main processor and a coprocessor. The main processor, also known as a CPU (Central Processing Unit), is used to process data in the wake-up state; the coprocessor is a low-power processor used to process data in the standby state. In some embodiments, processor 501 may integrate a GPU (Graphics Processing Unit), which is responsible for rendering and drawing the content to be displayed on the screen. In some embodiments, processor 501 may also include an AI (Artificial Intelligence) processor, which is used to handle computational operations related to machine learning.
[0121] The memory 502 may include one or more computer-readable storage media, which may be non-transitory. The memory 502 may also include high-speed random access memory and non-volatile memory, such as one or more disk storage devices or flash memory devices. In some embodiments, the non-transitory computer-readable storage media in the memory 502 is used to store at least one instruction, which is executed by the processor 501 to implement the slope real-time monitoring and early warning method provided in this embodiment of the invention.
[0122] Those skilled in the art will understand that Figure 5 The structure shown does not constitute a limitation on the computer device 500, and may include more or fewer components than shown, or combine certain components, or use different component arrangements.
[0123] This invention also provides a non-transitory computer-readable storage medium, which, when the instructions in the storage medium are executed by the processor of a computer device, enables the computer device to execute the slope real-time monitoring and early warning method provided in this invention.
[0124] This invention also provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the slope real-time monitoring and early warning method provided in this invention.
[0125] The above description is merely an optional embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A method for real-time monitoring and early warning of slopes, characterized in that, The method includes: Multi-source monitoring data from various measuring points in the target slope are acquired to obtain a monitoring dataset, which includes historical monitoring datasets and real-time monitoring datasets. The influencing parameters of the target monitoring parameters are selected from the monitoring dataset. The influencing parameters are the parameters that affect the target monitoring parameters. A GPR model is constructed based on the historical impact parameters and historical target monitoring parameters in the historical monitoring dataset. The real-time impact parameters in the real-time monitoring dataset are input into the GPR model to obtain the reference mean fitted by the GPR model. Based on the difference between the real-time target monitoring parameters of each measuring point in the target slope and the corresponding reference mean, multiple abnormal measuring points in the target slope at the current time are determined, and the real-time target monitoring parameters are the target monitoring parameters in the real-time monitoring dataset. The multiple abnormal measurement points are clustered according to their spatial location to obtain multiple clusters, and an early warning level is generated for each cluster.
2. The method according to claim 1, characterized in that, The step of filtering the impact parameters of the target monitoring parameter from the monitoring dataset includes: The feature selection algorithm is used to filter out the influencing parameters of the target monitoring parameters from the monitoring dataset. The feature selection algorithm includes correlation analysis algorithm, principal component analysis algorithm, and mutual information algorithm. When the number of types of influencing parameters is greater than 1, an automatic correlation dimension scaling mechanism is used to assign an independent length scale parameter to each type of influencing parameter.
3. The method according to claim 1, characterized in that, The generation of the early warning level for each cluster includes: The quality score corresponding to each sample in the monitoring dataset is obtained, and the quality score is based on short-time SNR, short-term drift rate, and missing rate. The confidence level of each cluster is calculated based on the average quality score of each outlier measurement point within each cluster, the average standardized residual, and the spatial density. An early warning level is generated for each cluster based on the confidence level of each cluster.
4. The method according to claim 1, characterized in that, The kernel function of the GPR is a radial basis function kernel, which is expressed by the following formula: in, This refers to the real-time impact parameter of a measuring point in the real-time monitoring dataset. The historical impact parameter of a measuring point in the historical monitoring dataset. The output of the radial basis function kernel is represented by the covariance matrix. for The corresponding length scale parameter, For signal variance, This refers to the Kronecker delta function.
5. The method according to claim 4, characterized in that, The reference mean fitted by the GPR model is expressed by the following formula: in, This is the reference mean fitted by the GPR model. This is a feature vector composed of the real-time influence parameters of each measuring point in the real-time monitoring dataset. This is a feature vector composed of the historical impact parameters of each measuring point at different times in the historical monitoring dataset. This is a feature vector composed of the historical target monitoring parameters of each measuring point in the historical monitoring dataset. It is an identity matrix. This represents the noise variance.
6. The method according to any one of claims 1 to 5, characterized in that, The method further includes: The warning level of each abnormal measuring point is displayed in the digital twin model of the target slope. The warning level of any abnormal measuring point is the warning level of the cluster to which the abnormal measuring point belongs.
7. A real-time slope monitoring and early warning device, characterized in that, The device includes: The acquisition module is used to acquire multi-source monitoring data from various measuring points in the target slope to obtain a monitoring dataset, which includes historical monitoring datasets and real-time monitoring datasets. A filtering module is used to filter out the influencing parameters of the target monitoring parameter from the monitoring dataset. The influencing parameters are the parameters that affect the target monitoring parameter. The modeling module is used to construct a GPR model based on the historical impact parameters and historical target monitoring parameters in the historical monitoring dataset. The fitting module is used to input the real-time impact parameters in the real-time monitoring dataset into the GPR model to obtain the reference mean fitted by the GPR model. An abnormal measuring point determination module is used to determine multiple abnormal measuring points in the target slope at the current time based on the difference between the real-time target monitoring parameters of each measuring point in the target slope and the corresponding reference mean. The real-time target monitoring parameters are the target monitoring parameters in the real-time monitoring dataset. The early warning module is used to cluster the multiple abnormal measurement points according to their spatial location to obtain multiple clusters and generate an early warning level for each cluster.
8. A computer device, characterized in that, The computer device includes a memory and a processor, wherein the memory stores at least one computer program, which is loaded and executed by the processor to implement the method according to any one of claims 1 to 6.
9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores at least one computer program, which is loaded and executed by a processor to implement the method according to any one of claims 1 to 6.
10. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instructions are executed by the processor, they implement the method described in any one of claims 1 to 6.