A reservoir dam safety monitoring method and system based on edge computing

Abstract

This invention discloses a method and system for monitoring the safety of reservoir dams based on edge computing, belonging to the field of water conservancy engineering safety monitoring technology. The method includes: each edge computing node acquiring monitoring data and performing preprocessing to generate a standardized data matrix; establishing a benchmark database and performing anomaly detection to generate a labeled time-series data matrix; performing multi-scale decomposition on the labeled time-series data matrix to construct a node response feature matrix; a central processing unit receiving data from each edge computing node, analyzing multi-parameter spatial propagation patterns by constructing a parameter-spatial correlation matrix to obtain a spatial correlation feature matrix; establishing a Bayesian risk assessment model to predict the dam risk level and output the risk evolution trend; and issuing differentiated early warning commands based on the risk assessment results. This invention, through a distributed architecture combining edge computing and central processing, achieves real-time monitoring, intelligent analysis, and accurate early warning of dam safety status, improving monitoring efficiency and risk identification accuracy.

Description

Technical Field

[0001] This invention relates to the field of water conservancy project safety monitoring technology, and in particular to a reservoir dam safety monitoring method and system based on edge computing. Background Technology

[0002] With the continuous expansion of water conservancy projects, dam safety monitoring has become a crucial technical means to ensure the safe operation of reservoirs. Traditional dam safety monitoring systems deploy various sensors in key areas such as the dam body and foundation to collect real-time monitoring data on displacement, seepage pressure, stress, and strain. This data is then transmitted to a monitoring center via a data transmission network for analysis and processing. This centralized monitoring architecture has developed a relatively mature technical system for data acquisition, transmission, and storage, providing fundamental data support for dam safety assessment and early warning.

[0003] Existing dam safety monitoring technologies face challenges in practical applications, including insufficient data processing timeliness and limited spatial correlation analysis capabilities. On one hand, the massive amounts of monitoring data in traditional centralized architectures must be transmitted to a central server for processing. Limited network bandwidth or communication interruptions can easily cause data processing delays, affecting the timely detection of abnormal states. On the other hand, existing monitoring methods primarily focus on the temporal characteristics of individual monitoring points, lacking in-depth analysis of the spatial correlations between multiple monitoring points. This makes it difficult to accurately capture the overall propagation patterns of dam structural responses, limiting the accuracy of safety status assessments and the foresight of early warnings. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a reservoir dam safety monitoring method and system based on edge computing to solve the technical problems of insufficient timeliness of data processing and limited spatial correlation analysis capabilities of existing monitoring technologies.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides a method for monitoring the safety of a reservoir dam based on edge computing, comprising:

[0008] Each edge computing node acquires dam monitoring data for its corresponding monitoring area and performs preprocessing to generate a standardized data matrix.

[0009] A benchmark database is established based on a standardized data matrix, and anomaly detection is performed to generate an annotated time series data matrix.

[0010] The labeled time-series data matrix is ​​decomposed into multi-scale features to extract multi-scale response features and construct a node response feature matrix.

[0011] The node response feature matrix and monitoring point location information of each edge computing node are uploaded to the central processing unit;

[0012] The central processing unit receives data from each edge computing node, analyzes the multi-parameter spatial propagation mode by constructing a parameter-spatial correlation matrix, and obtains the spatial correlation feature matrix.

[0013] A Bayesian risk assessment model is established to predict the dam risk level by integrating the nodal response feature matrix and the spatial correlation feature matrix, and the risk evolution trend is output.

[0014] Differentiated early warning instructions are issued to each edge computing node based on the risk assessment results.

[0015] As a preferred embodiment of the reservoir dam safety monitoring method based on edge computing of the present invention, wherein: each edge computing node acquires dam monitoring data for its corresponding monitoring area and performs preprocessing, including:

[0016] Each edge computing node acquires dam monitoring data from the corresponding sensor array and records the time stamp to generate an initial dataset;

[0017] Perform data quality checks on the initial dataset, identify and repair invalid data, and generate a cleaned dataset.

[0018] The cleaned dataset is time-aligned and numerically normalized to generate a standardized data matrix for each edge computing node.

[0019] As a preferred embodiment of the reservoir dam safety monitoring method based on edge computing of the present invention, the generation of the labeled time-series data matrix includes:

[0020] Based on the statistical analysis of historical monitoring data, the correlation characteristics of each monitoring parameter are analyzed to establish the normal behavior benchmark of the parameter under different working conditions and to build an adaptive benchmark database.

[0021] Based on the adaptive benchmark database, the statistical quantile method is used to calculate the grading threshold and rate of change threshold of each monitoring parameter, and generate a dual threshold parameter vector;

[0022] Based on the dual threshold parameter vector, single-parameter anomaly detection and multi-parameter correlation verification are performed to generate anomaly identification vector;

[0023] Annotated time-series data matrix is ​​generated by labeling the standardized data matrix based on anomaly identifier vectors.

[0024] As a preferred embodiment of the reservoir dam safety monitoring method based on edge computing of the present invention, the construction of the node response feature matrix includes:

[0025] Multi-scale time series decomposition is performed on each monitoring parameter in the labeled time series data matrix. The decomposition method and decomposition parameters are adaptively selected according to the data change characteristics to generate multi-level decomposition components.

[0026] The multi-level decomposition components of each monitoring parameter are classified into transient response components, periodic change components, and long-term evolution components according to their time scale characteristics.

[0027] Correlation verification is performed on the same type of components of different monitoring parameters, and the consistency of component changes is verified by using the statistical correlation between components.

[0028] Features are extracted from the transient response components, periodic variation components, and long-term evolution components of each monitoring parameter, and multi-scale response feature vectors of each monitoring parameter are generated by combining anomaly identification information.

[0029] Construct node response feature matrices based on multi-scale response feature vectors.

[0030] As a preferred embodiment of the reservoir dam safety monitoring method based on edge computing of the present invention, the method includes: analyzing the multi-parameter spatial propagation mode by constructing a parameter-spatial correlation matrix.

[0031] Based on the location information of the monitoring points of each edge computing node, a spatial topology structure of the monitoring points is constructed, and a spatial adjacency matrix is ​​generated.

[0032] By combining the node response feature matrices of each edge computing node, spatial correlation analysis is used to identify the correlation characteristics between different monitoring points and generate a parameter-spatial correlation matrix.

[0033] Based on the parameter-spatial correlation matrix, we analyze multi-parameter spatial correlation patterns and identify correlation propagation paths and influence decay patterns.

[0034] Feature quantification is performed on the correlation propagation path and influence attenuation mode to extract spatial correlation feature parameters and construct a spatial correlation feature matrix.

[0035] As a preferred embodiment of the edge computing-based reservoir dam safety monitoring method of the present invention, the establishment of a Bayesian risk assessment model includes:

[0036] Establish a prior knowledge base for the Bayesian risk assessment model, convert prior knowledge into prior probability distributions, and generate a multi-level prior parameter set;

[0037] The node response feature matrix and spatial correlation feature matrix are used as observation evidence input into the Bayesian network to construct a likelihood function model and calculate the likelihood value of the observation evidence under different risk levels.

[0038] The Bayesian inference method is used to fuse prior probability distributions and likelihood values ​​to calculate the posterior probability distributions of structural safety level, seepage safety level, and stability safety level, generating a comprehensive risk assessment result.

[0039] A risk state evolution model is established based on the comprehensive risk assessment results to predict the risk evolution trend and output the risk evolution trend and uncertainty assessment results.

[0040] As a preferred embodiment of the reservoir dam safety monitoring method based on edge computing of the present invention, the method includes: issuing differentiated early warning instructions to each edge computing node based on risk assessment results, including:

[0041] Based on the risk assessment results, establish mapping rules from risk level to early warning level, and generate basic early warning instructions;

[0042] Based on the prediction results of risk evolution trends, basic early warning instructions are configured differently to generate node-specific early warning instruction sets;

[0043] It issues node-specific early warning instruction sets to each edge computing node and receives instruction confirmation feedback from each node;

[0044] Each edge computing node executes the corresponding early warning strategy based on the received early warning instruction set and reports the early warning response status to the central processing unit.

[0045] Secondly, the present invention provides a reservoir dam safety monitoring system based on edge computing, comprising:

[0046] The data preprocessing module is used to acquire dam monitoring data of the corresponding monitoring area and perform preprocessing to generate a standardized data matrix;

[0047] The anomaly detection module is used to establish a benchmark database based on a standardized data matrix and perform anomaly detection to generate an annotated time series data matrix.

[0048] The feature extraction module is used to perform multi-scale decomposition on the labeled time-series data matrix, extract multi-scale response features, and construct a node response feature matrix.

[0049] The data upload module is used to upload the node response feature matrix and monitoring point location information of each edge computing node to the central processing unit;

[0050] The spatial correlation analysis module is used to receive data from each edge computing node, analyze multi-parameter spatial propagation patterns by constructing a parameter-spatial correlation matrix, and obtain a spatial correlation feature matrix.

[0051] The risk assessment module is used to build a Bayesian risk assessment model, which combines the nodal response feature matrix and the spatial correlation feature matrix to predict the dam risk level and output the risk evolution trend.

[0052] The instruction issuance module is used to issue differentiated early warning instructions to each edge computing node based on the risk assessment results.

[0053] Thirdly, the present invention provides a computer device including a memory and a processor, wherein the memory stores a computer program, wherein when the computer program is executed by the processor, it implements any step of the edge computing-based reservoir dam safety monitoring method of the first aspect of the present invention.

[0054] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the edge computing-based reservoir dam safety monitoring method of the first aspect of the present invention.

[0055] The beneficial effects of this invention are as follows: By employing adaptive multi-scale decomposition technology, this invention effectively solves the problem of decomposition distortion caused by differences in the time-frequency characteristics of different monitoring parameters, achieving accurate extraction of response information across multiple time scales. Through spatial correlation analysis, a quantitative model of the propagation law of monitoring parameters is established, overcoming the limitations of traditional monitoring systems that focus only on single-point features while ignoring spatial correlation, significantly improving the overall perception capability of dam safety status. Through a risk state evolution model, a predictive capability from the current risk state to future evolution trends is established, and combined with uncertainty quantification, it provides forward-looking decision support for dam safety management, enhancing the intelligence level and early warning capability of the dam monitoring system. Attached Figure Description

[0056] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the 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.

[0057] Figure 1 This is a flowchart of a reservoir dam safety monitoring method based on edge computing.

[0058] Figure 2 A flowchart for generating labeled time-series data matrices for a reservoir dam safety monitoring method based on edge computing.

[0059] Figure 3 A flowchart for constructing the node response feature matrix of a reservoir dam safety monitoring method based on edge computing.

[0060] Figure 4 This is a module connection diagram of a reservoir dam safety monitoring system based on edge computing. Detailed Implementation

[0061] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0062] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0063] Secondly, the term "an embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places throughout this specification does not necessarily refer to the same embodiment, nor is it a single embodiment or an embodiment selectively excluded from other embodiments.

[0064] Reference Figures 1-4 As one embodiment of the present invention, this embodiment provides a reservoir dam safety monitoring method based on edge computing, the flowchart of which is shown below. Figure 1 As shown, the method includes the following steps:

[0065] S1: Each edge computing node acquires the dam monitoring data of its corresponding monitoring area and performs preprocessing to generate a standardized data matrix.

[0066] In one embodiment, each edge computing node acquires dam monitoring data from the corresponding sensor array and records time stamps to generate an initial dataset; data quality checks are performed on the initial dataset to identify and repair invalid data, generating a cleaned dataset.

[0067] It should be noted that the sensor array is selected from displacement gauges, piezometers, uplift gauges, flow meters, strain gauges, stress gauges, inclinometers, thermometers, water level gauges, accelerometers, and crack gauges, depending on the requirements of the monitored location. One or more of each type of sensor can be configured according to monitoring needs. Invalid data includes, but is not limited to, missing data, out-of-range data, communication anomaly data, gradient anomaly data, duplicate data, timestamp anomaly data, and sensor zero-point drift data.

[0068] Furthermore, the cleaned dataset undergoes time alignment and numerical normalization transformation to generate a standardized data matrix corresponding to each edge computing node. The standardized data matrix is ​​constructed with a unified time grid as rows and monitoring parameters (in this embodiment, monitoring point parameters) as columns, providing a standardized data foundation for subsequent collaborative analysis.

[0069] S2: Establish a benchmark database based on a standardized data matrix and perform anomaly detection to generate an annotated time series data matrix.

[0070] Specifically, the flowchart for generating the labeled time-series data matrix is ​​as follows: Figure 2 As shown, it includes the following steps:

[0071] S2.1: Based on the statistical analysis of historical monitoring data, establish the correlation characteristics of each monitoring parameter, establish the normal behavior benchmark of the parameter under different working conditions, and construct an adaptive benchmark database.

[0072] In one embodiment, time-series data for each monitoring parameter are extracted from historical monitoring data, and the data is grouped according to reservoir water level intervals to generate sub-condition datasets. Pearson correlation coefficients are calculated on the sub-condition datasets to identify the correlation strength between parameters, generating a parameter correlation coefficient matrix. A sliding window statistical method is used to calculate the statistical distribution characteristics of the mean, standard deviation, and variation gradient of the monitoring parameters under each condition, establishing a parameter statistical benchmark set. The parameter correlation coefficient matrix and the parameter statistical benchmark set are integrated and stored to construct an adaptive benchmark database.

[0073] S2.2: Based on the adaptive benchmark database, the statistical quantile method is used to calculate the grading threshold and rate of change threshold of each monitoring parameter, and generate a dual threshold parameter vector.

[0074] Specifically, the statistical benchmark set of parameters corresponding to the current operating condition is retrieved from the adaptive benchmark database to obtain the statistical distribution data of the monitored parameters. Based on the statistical distribution data, a static graded threshold set and a dynamic threshold set of change rate are generated using the quantile calculation method. The static graded threshold set establishes multi-level early warning boundaries by calculating the statistical quantiles of parameter values, while the dynamic threshold set of change rate determines the anomaly discrimination criteria for the change rate by analyzing the distribution characteristics of the parameter change rate within a preset time window.

[0075] Furthermore, the static threshold set and the dynamic threshold set of change rate are combined into a vector, and the threshold weights of strongly correlated parameters are adjusted according to the parameter correlation coefficient matrix to generate an adaptive dual threshold parameter vector.

[0076] S2.3: Based on the dual threshold parameter vector, perform single-parameter anomaly detection and multi-parameter correlation verification to generate anomaly identification vector.

[0077] The multi-parameter correlation verification includes verification of the physical coupling relationship of parameters and verification of the spatial coordination of measurement points. Specifically, the current monitoring data is compared with the dual threshold parameter vector to determine whether each monitoring parameter exceeds the static threshold boundary and the dynamic change rate threshold, generating a single-parameter abnormal state matrix; based on the parameter correlation coefficient matrix, multi-parameter correlation verification is performed on the abnormal parameters to identify whether the correlated parameters show coordinated abnormal responses within the same time window; combining the results of single-parameter anomaly detection and multi-parameter correlation verification, an anomaly identifier vector containing anomaly identification and anomaly type is generated.

[0078] S2.4: Annotate the standardized data matrix based on the anomaly identification vector to generate an annotated time-series data matrix.

[0079] It should be noted that the labeled time series data matrix maintains a structure with a unified time grid as rows and monitoring parameters as columns, and adds anomaly identification and anomaly type information.

[0080] Preferably, this invention effectively solves the problem of misjudgment caused by differences in the normal behavior benchmarks of monitoring parameters under different operating conditions through an adaptive benchmark database and a dual threshold detection mechanism, significantly improving the accuracy of anomaly detection. The multi-parameter correlation verification mechanism, through analysis of the physical coupling relationship of parameters, effectively distinguishes between structural anomalies and sensor faults, reducing the system's false alarm rate and providing a reliable data foundation for subsequent risk assessment.

[0081] S3: Perform multi-scale decomposition on the labeled time-series data matrix, extract multi-scale response features, and construct a node response feature matrix.

[0082] Specifically, the flowchart for constructing the node response feature matrix is ​​as follows: Figure 3 As shown, it includes the following steps:

[0083] S3.1: Perform multi-scale time series decomposition on each monitoring parameter in the labeled time series data matrix, and adaptively select the decomposition method and decomposition parameters according to the data change characteristics to generate multi-level decomposition components.

[0084] In one embodiment, the variation period and amplitude characteristics of each monitoring parameter in the labeled time-series data matrix are analyzed to determine the time-frequency characteristic parameters of each monitoring parameter. Based on the time-frequency characteristic parameters, a corresponding decomposition method is selected for each monitoring parameter, and the number of decomposition levels is determined to generate a decomposition parameter configuration. Based on the decomposition parameter configuration, multi-scale decomposition processing is performed on each monitoring parameter to generate multi-level decomposition components of the corresponding number of levels. The decomposition quality of the multi-level decomposition components is verified, and the decomposition parameters are adjusted and re-decomposed based on the verification results until qualified multi-level decomposition components are obtained.

[0085] S3.2: Classify the multi-layer decomposition components of each monitoring parameter into transient response components, periodic change components, and long-term evolution components according to their time scale characteristics.

[0086] It should be noted that classification refers to labeling each decomposition component according to its physical meaning based on its time-scale characteristics. Specifically, decomposition components with timescales within the hour are labeled as transient response components, those with timescales from hours to days are labeled as periodically changing components, and those with timescales above days are labeled as long-term evolution components. This classification provides a unified processing framework for subsequent cross-parameter correlation verification and feature extraction, enabling effective correlation analysis of similar time-scale components of different monitoring parameters.

[0087] S3.3: Perform correlation verification on the same type of components of different monitoring parameters, and use the statistical correlation between components to verify the consistency of component changes.

[0088] Furthermore, decomposed components belonging to the same time scale type are identified among the monitoring parameters, and the correspondence between parameters of the same type is established. Statistical correlation indicators among similar components of different monitoring parameters are calculated, including correlation coefficients and time lag relationships. A correlation verification benchmark is established based on the physical correlation of the monitoring parameters, and correlation judgment thresholds for each type of component are determined. The statistical correlation indicators are compared with the corresponding correlation judgment thresholds to generate correlation verification results for each type of component. Reliability weights are assigned to each type of component based on the correlation verification results, providing a basis for adjusting component weights for subsequent feature extraction.

[0089] S3.4: Extract features from the transient response components, periodic variation components, and long-term evolution components of each monitoring parameter, and generate multi-scale response feature vectors for each monitoring parameter by combining them with anomaly identification information.

[0090] Furthermore, peak intensity and rate of change features are extracted from the transient response components of each monitoring parameter, and then weighted across parameters using component reliability weights to generate transient response intensity features. Periodic stability and amplitude variability features are extracted from the periodic variation components of each monitoring parameter, and then weighted across parameters using component reliability weights to generate periodic stability features. Trend intensity and consistency of change features are extracted from the long-term evolution components of each monitoring parameter, and then weighted across parameters using component reliability weights to generate long-term evolution intensity features. Based on anomaly identification information, the frequency and severity of anomalous events in various components are statistically analyzed, and anomaly statistical features are generated through weighted calculation. The transient response intensity features, periodic stability features, long-term evolution intensity features, and anomaly statistical features are then combined sequentially to construct a multi-scale response feature vector.

[0091] S3.5: Construct the node response feature matrix based on multi-scale response feature vectors.

[0092] The node response feature matrix consists of monitoring points as rows and multi-scale response features as columns.

[0093] Preferably, this invention effectively solves the problem of decomposition distortion caused by differences in the time-frequency characteristics of different monitoring parameters through adaptive multi-scale decomposition technology, and realizes accurate extraction of response information at multiple time scales. Cross-parameter correlation verification and reliability weighting mechanisms ensure the reliability of decomposition quality, establish an effective conversion system from time-domain signals to structural response features, and significantly improve data processing efficiency and feature representation capabilities in edge computing environments.

[0094] S4: Upload the node response feature matrix and monitoring point location information of each edge computing node to the central processing unit.

[0095] It should be noted that the monitoring point location information includes the monitoring point code, monitoring location identifier, sensor type identifier, and spatial coordinate information. The monitoring point code is a unique identifier for each monitoring point, used to distinguish and identify different monitoring points. The monitoring location identifier identifies the structural part of the dam where the monitoring point is located, including the dam body, dam foundation, left bank, right bank, and dam crest. The sensor type identifier identifies the type of sensor used at the monitoring point, including displacement gauges, piezometers, etc. The spatial coordinate information includes the three-dimensional coordinates of the monitoring point, used to calculate the spatial distance relationship between monitoring points.

[0096] S5: The central processing unit receives data from each edge computing node, analyzes the multi-parameter spatial propagation mode by constructing a parameter-spatial correlation matrix, and obtains the spatial correlation feature matrix.

[0097] Specifically, S5 includes the following steps:

[0098] S5.1: Based on the location information of each edge computing node, construct the spatial topology of the monitoring points and generate a spatial adjacency matrix.

[0099] In one embodiment, the steps for constructing the spatial topology of monitoring points include: calculating the three-dimensional Euclidean distance between each monitoring point based on the spatial coordinate information in the monitoring point location information, and constructing a distance matrix; combining the monitoring location identifier and sensor type identifier in the monitoring point location information to establish a monitoring point attribute classification and generate an attribute label matrix; and using an adaptive neighborhood determination algorithm based on the distance matrix and the attribute label matrix to identify the spatial adjacency relationships of each monitoring point and generate an initial adjacency matrix. The adaptive neighborhood determination algorithm comprehensively considers spatial distance thresholds, structural connectivity constraints, and sensor type compatibility.

[0100] Furthermore, the initial adjacency matrix is ​​subjected to connectivity verification. Isolated nodes are identified through connected component analysis, and adjacency relationships are corrected based on physical constraints. Based on the corrected adjacency relationships and the measurement point coding information of each monitoring point, a spatial adjacency matrix is ​​generated.

[0101] The construction of the spatial topology of the monitoring points adopts an adaptive neighborhood determination algorithm, which can dynamically adjust the adjacency relationship according to the physical constraints of the dam structure and the characteristics of sensor configuration. This ensures the accuracy of the adjacency relationship under complex geometric structures and lays a reliable topological foundation for subsequent spatial correlation analysis.

[0102] S5.2: Combining the node response feature matrix of each edge computing node, the spatial correlation analysis method is used to identify the correlation characteristics between different monitoring points and generate a parameter-spatial correlation matrix.

[0103] Specifically, the node response feature matrix of each edge computing node is obtained, and the multi-scale response feature vector corresponding to each monitoring point is extracted. Combined with the spatial adjacency matrix, the spatial adjacency relationship and adjacency weight of each monitoring point are determined. Using spatial correlation analysis, spatial correlation coefficients between monitoring points are calculated for transient response intensity characteristics, periodic stability characteristics, long-term evolution intensity characteristics, and anomaly statistical characteristics, respectively. The spatial correlation coefficients are organized into a matrix form according to monitoring points and feature types to generate a parameter-spatial correlation matrix, and correlation propagation patterns are identified through spatial correlation network analysis.

[0104] S5.3: Based on the parameter-spatial correlation matrix, analyze the multi-parameter spatial correlation pattern, identify the correlation propagation path and the influence decay mode.

[0105] Preferably, based on the parameter-spatial correlation matrix, a correlation strength threshold screening method is used to identify significantly correlated monitoring point pairs and construct an effective correlation network. Path search is performed on the effective correlation network to identify the correlation propagation path from the source monitoring point to the target monitoring point, and the path length and propagation hop count are calculated. Distance attenuation analysis is used to calculate the influence attenuation coefficient on each correlation propagation path, establishing a model of the attenuation relationship between spatial distance and correlation strength. Based on the correlation propagation path and attenuation relationship model, multi-parameter spatial correlation patterns are identified, including the dominant propagation direction, the range of influence propagation, and the attenuation rate distribution.

[0106] S5.4: Perform feature quantization on the correlation propagation path and influence decay mode, extract spatial correlation feature parameters, and construct a spatial correlation feature matrix.

[0107] Furthermore, based on the associated propagation paths, the path length and propagation hop count of each propagation path are calculated to extract propagation distance features. Combining the spatial coordinate information of each monitoring point, the spatial azimuth of the associated propagation paths is calculated to extract propagation azimuth features. Based on the attenuation relationship model in the influence attenuation mode, attenuation function parameters are fitted to extract attenuation rate parameters. The correlation coefficients of each monitoring point pair are obtained from the parameter-spatial correlation matrix as influence intensity coefficients. The propagation distance features, propagation azimuth features, attenuation rate parameters, and influence intensity coefficients are combined according to the monitoring point pairs to construct a spatial correlation feature matrix.

[0108] Preferably, this invention effectively solves the technical problem of traditional monitoring systems neglecting the spatial propagation laws of parameters by constructing a spatial correlation analysis framework. The adaptive neighborhood determination algorithm ensures the accuracy of spatial topology construction under complex geometric structures, and the spatial correlation analysis of multi-scale response features establishes the correlation mapping relationship between monitoring points. By extracting key feature parameters such as propagation distance, propagation orientation, attenuation rate, and influence intensity, the invention achieves a quantitative characterization of complex spatial correlation relationships, significantly improving the dam monitoring system's ability to perceive and analyze spatial propagation laws.

[0109] S6: Establish a Bayesian risk assessment model, combine the nodal response feature matrix and spatial correlation feature matrix to predict the dam risk level, and output the risk evolution trend.

[0110] Specifically, S6 includes the following steps:

[0111] S6.1: Establish a prior knowledge base for the Bayesian risk assessment model, convert prior knowledge into prior probability distributions, and generate multi-level prior parameter sets.

[0112] In one embodiment, the prior knowledge base includes three sub-bases: structural safety priors, seepage safety priors, and stability safety priors. The structural safety priors are constructed based on dam safety evaluation criteria, the seepage safety priors are constructed based on expert experience rules, and the stability safety priors are constructed based on historical accident cases. The prior probability distribution uses a Beta distribution to parametrically model the probability of risk levels.

[0113] S6.2: Input the node response feature matrix and spatial correlation feature matrix as observation evidence into the Bayesian network, construct the likelihood function model, and calculate the likelihood value of the observation evidence under different risk levels.

[0114] Preferably, the node response feature matrix and spatial correlation feature matrix are obtained, and the response feature vector and spatial correlation feature vector of each monitoring point are extracted to form an observation evidence set. Based on historical risk sample data, a likelihood function is constructed using statistical modeling methods to describe the probability distribution of the observation evidence under a given risk level. The observation evidence set is input into the likelihood function model to calculate the likelihood values ​​of the current observation evidence at each risk level of structural safety, seepage safety, and stability safety.

[0115] S6.3: The Bayesian inference method is used to fuse the prior probability distribution and likelihood value to calculate the posterior probability distribution of the structural safety level, seepage safety level and stability safety level, and generate a comprehensive risk assessment result.

[0116] Furthermore, a multi-level prior parameter set and likelihood values ​​for each risk level are obtained to construct a Bayesian inference computation framework. Based on Bayes' theorem, a variational approximation algorithm is used to calculate the posterior probability distribution, yielding posterior probability distributions for structural safety, seepage safety, and stability safety, respectively. Each safety category includes probability values ​​for four levels: safe, alert, warning, and hazardous. Based on the posterior probability distributions of the three safety categories, a weighted fusion method is used to calculate the overall risk level probability. The current risk level is determined according to the principle of maximizing the overall risk level probability, generating a comprehensive risk assessment result.

[0117] S6.4: Based on the comprehensive risk assessment results, establish a risk state evolution model, predict the risk evolution trend, and output the risk evolution trend and uncertainty assessment results.

[0118] Specifically, based on historical comprehensive risk assessment results, a risk state space is constructed, comprising four levels: safe, alert, early warning, and dangerous. By statistically analyzing historical risk state change patterns, a state transition model is established to determine the probabilistic relationships between risk levels. Using the current comprehensive risk level as the starting state, multi-step predictions are performed based on the state transition model to generate a risk state evolution sequence. The probability propagation method is used to calculate the uncertainty of the prediction results, quantifying the reliability of the risk evolution trend. Based on the risk state evolution sequence and uncertainty assessment, a risk evolution trend prediction result is generated.

[0119] Preferably, this invention effectively integrates multi-source knowledge from engineering principles, expert experience, and historical cases by constructing a multi-level prior knowledge base, providing a reliable prior foundation for Bayesian risk assessment. Bayesian inference based on variational approximation algorithms enables efficient calculation of complex posterior probability distributions, significantly improving the computational efficiency and accuracy of risk assessment. A weighted fusion strategy unifies structural safety, seepage safety, and stability safety into a comprehensive risk assessment result, overcoming the limitations of single-dimensional assessment.

[0120] S7: Issue differentiated early warning instructions to each edge computing node based on the risk assessment results.

[0121] Specifically, based on the comprehensive risk assessment results, a mapping rule is established from risk level to warning level, generating basic warning instructions. The mapping rule is based on dam safety management standards, mapping the four risk levels—safe, caution, warning, and danger—to green, yellow, orange, and red warning levels, respectively. The basic warning instructions set general warning levels and standard response frameworks for structural safety, seepage safety, and stability safety.

[0122] Furthermore, by combining the risk evolution trend prediction results and the monitoring status of each edge computing node, basic early warning instructions are configured differently to generate node-specific early warning instruction sets. The differentiated early warning strategy is determined based on the abnormal state, spatial location, and monitoring importance of each node; the node-specific early warning instruction set includes monitoring frequency adjustment instructions, data transmission priority instructions, and local response execution instructions.

[0123] Furthermore, a node-specific early warning instruction set is issued to each edge computing node, and confirmation feedback from each node is received. Each edge computing node executes the corresponding early warning strategy based on the received node-specific early warning instruction set and reports the early warning response status back to the central processing unit.

[0124] The instruction confirmation feedback mechanism requires each node to return confirmation information and execution status after receiving an instruction; the execution of the early warning strategy includes adjusting monitoring parameters, outputting early warning signals, and preparing for emergency response; and the early warning response status feedback is used to evaluate the early warning effect and optimize the early warning strategy.

[0125] This embodiment also provides a reservoir dam safety monitoring system based on edge computing, as shown in the module connection diagram below. Figure 4 As shown, the system includes:

[0126] The data preprocessing module is used to acquire dam monitoring data of the corresponding monitoring area and perform preprocessing to generate a standardized data matrix;

[0127] The anomaly detection module is used to establish a benchmark database based on a standardized data matrix and perform anomaly detection to generate an annotated time series data matrix.

[0128] The feature extraction module is used to perform multi-scale decomposition on the labeled time-series data matrix, extract multi-scale response features, and construct a node response feature matrix.

[0129] The data upload module is used to upload the node response feature matrix and monitoring point location information of each edge computing node to the central processing unit;

[0130] The spatial correlation analysis module is used to receive data from each edge computing node, analyze multi-parameter spatial propagation patterns by constructing a parameter-spatial correlation matrix, and obtain a spatial correlation feature matrix.

[0131] The risk assessment module is used to build a Bayesian risk assessment model, which combines the nodal response feature matrix and the spatial correlation feature matrix to predict the dam risk level and output the risk evolution trend.

[0132] The instruction issuance module is used to issue differentiated early warning instructions to each edge computing node based on the risk assessment results.

[0133] This embodiment also provides a computer device applicable to the edge computing-based reservoir dam safety monitoring method, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the edge computing-based reservoir dam safety monitoring method proposed in the above embodiment.

[0134] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0135] This embodiment also provides a storage medium storing a computer program, which, when executed by a processor, implements the edge computing-based reservoir dam safety monitoring method proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0136] In summary, this invention effectively solves the problem of decomposition distortion caused by differences in the time-frequency characteristics of different monitoring parameters through adaptive multi-scale decomposition technology, achieving accurate extraction of response information across multiple time scales. By establishing a quantitative model of the propagation law of monitoring parameters through spatial correlation analysis, it overcomes the limitation of traditional monitoring systems that only focus on single-point features while ignoring spatial correlation, significantly improving the overall perception capability of dam safety status. Furthermore, by establishing a risk state evolution model, it provides predictive capabilities from the current risk state to future evolution trends, and combined with uncertainty quantification, offers forward-looking decision support for dam safety management, enhancing the intelligence level and early warning capability of the dam monitoring system.

[0137] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for monitoring the safety of reservoir dams based on edge computing, characterized in that: include: Each edge computing node acquires dam monitoring data for its corresponding monitoring area and performs preprocessing to generate a standardized data matrix. A benchmark database is established based on the standardized data matrix, and anomaly detection is performed to generate a labeled time series data matrix. The labeled time-series data matrix is ​​decomposed into multiple scales to extract multi-scale response features and construct a node response feature matrix. The node response feature matrix and monitoring point location information of each edge computing node are uploaded to the central processing unit; The central processing unit receives data from each edge computing node, analyzes the multi-parameter spatial propagation mode by constructing a parameter-spatial correlation matrix, and obtains a spatial correlation feature matrix. A Bayesian risk assessment model is established to predict the dam risk level by combining the node response feature matrix and the spatial correlation feature matrix, and the risk evolution trend is output. Differentiated early warning instructions are issued to each edge computing node based on the risk assessment results; The generated labeled time-series data matrix includes: Based on the statistical analysis of historical monitoring data, the correlation characteristics of each monitoring parameter are analyzed to establish the normal behavior benchmark of the parameter under different working conditions and to build an adaptive benchmark database. Based on the adaptive benchmark database, the statistical quantile method is used to calculate the grading threshold and rate of change threshold of each monitoring parameter, and generate a dual threshold parameter vector; Based on the dual threshold parameter vector, single-parameter anomaly detection and multi-parameter correlation verification are performed to generate an anomaly identification vector. The standardized data matrix is ​​labeled based on the anomaly identification vector to generate a labeled time-series data matrix. The constructed node response feature matrix includes: Multi-scale time series decomposition is performed on each monitoring parameter in the labeled time series data matrix. The decomposition method and decomposition parameters are adaptively selected according to the data change characteristics to generate multi-level decomposition components. The multi-level decomposition components of each monitoring parameter are classified into transient response components, periodic change components, and long-term evolution components according to their time scale characteristics. Correlation verification is performed on the same type of components of different monitoring parameters, and the consistency of component changes is verified by using the statistical correlation between components. Features are extracted from the transient response components, periodic variation components, and long-term evolution components of each monitoring parameter, and multi-scale response feature vectors of each monitoring parameter are generated by combining anomaly identification information. The node response feature matrix is ​​constructed based on the multi-scale response feature vector.

2. The reservoir dam safety monitoring method based on edge computing as described in claim 1, characterized in that: Each edge computing node acquires dam monitoring data for its corresponding monitoring area and performs preprocessing, including: Each edge computing node acquires dam monitoring data from the corresponding sensor array and records the time stamp to generate an initial dataset; Perform data quality checks on the initial dataset to identify and repair invalid data, generating a cleaned dataset; The cleaned dataset is subjected to time alignment and numerical normalization transformation to generate a standardized data matrix corresponding to each edge computing node.

3. The reservoir dam safety monitoring method based on edge computing as described in claim 1, characterized in that: The analysis of multi-parameter spatial propagation modes by constructing a parameter-spatial correlation matrix includes: Based on the location information of the monitoring points of each edge computing node, a spatial topology structure of the monitoring points is constructed, and a spatial adjacency matrix is ​​generated. By combining the node response feature matrices of each edge computing node, spatial correlation analysis is used to identify the correlation characteristics between different monitoring points and generate a parameter-spatial correlation matrix. Based on the parameter-spatial correlation matrix, multi-parameter spatial correlation patterns are analyzed to identify correlation propagation paths and influence attenuation patterns. The correlation propagation path and influence attenuation mode are quantized to extract spatial correlation feature parameters and construct a spatial correlation feature matrix.

4. The reservoir dam safety monitoring method based on edge computing as described in claim 1, characterized in that: The establishment of the Bayesian risk assessment model includes: Establish a prior knowledge base for the Bayesian risk assessment model, convert prior knowledge into prior probability distributions, and generate a multi-level prior parameter set; The node response feature matrix and spatial correlation feature matrix are used as observation evidence input into the Bayesian network to construct a likelihood function model and calculate the likelihood value of the observation evidence under different risk levels. The prior probability distribution and likelihood value are fused using Bayesian inference method to calculate the posterior probability distribution of structural safety level, seepage safety level and stability safety level, and generate comprehensive risk assessment results. Based on the comprehensive risk assessment results, a risk state evolution model is established to predict the risk evolution trend and output the risk evolution trend and uncertainty assessment results.

5. The reservoir dam safety monitoring method based on edge computing as described in claim 1, characterized in that: The step of issuing differentiated early warning instructions to each edge computing node based on the risk assessment results includes: Based on the risk assessment results, establish mapping rules from risk level to early warning level, and generate basic early warning instructions; Based on the risk evolution trend prediction results, the basic early warning instructions are configured differently to generate a node-specific early warning instruction set; The system issues the node-specific early warning instruction set to each edge computing node and receives instruction confirmation feedback from each node. Each edge computing node executes the corresponding early warning strategy based on the received early warning instruction set and reports the early warning response status to the central processing unit.

6. A reservoir dam safety monitoring system based on edge computing, based on the reservoir dam safety monitoring method based on edge computing as described in any one of claims 1 to 5, characterized in that: include: The data preprocessing module is used to acquire dam monitoring data of the corresponding monitoring area and perform preprocessing to generate a standardized data matrix; Anomaly detection module is used to establish a benchmark database based on the standardized data matrix and perform anomaly detection to generate an annotated time series data matrix; The feature extraction module is used to perform multi-scale decomposition on the labeled time-series data matrix, extract multi-scale response features, and construct a node response feature matrix; The data upload module is used to upload the node response feature matrix and monitoring point location information of each edge computing node to the central processing unit; The spatial correlation analysis module is used to receive data from each edge computing node, analyze multi-parameter spatial propagation patterns by constructing a parameter-spatial correlation matrix, and obtain a spatial correlation feature matrix. The risk assessment module is used to establish a Bayesian risk assessment model, which combines the node response feature matrix and the spatial correlation feature matrix to predict the dam risk level and output the risk evolution trend. The instruction issuance module is used to issue differentiated early warning instructions to each edge computing node based on the risk assessment results; The generated labeled time-series data matrix includes: Based on the statistical analysis of historical monitoring data, the correlation characteristics of each monitoring parameter are analyzed to establish the normal behavior benchmark of the parameter under different working conditions and to build an adaptive benchmark database. Based on the adaptive benchmark database, the statistical quantile method is used to calculate the grading threshold and rate of change threshold of each monitoring parameter, and generate a dual threshold parameter vector; Based on the dual threshold parameter vector, single-parameter anomaly detection and multi-parameter correlation verification are performed to generate an anomaly identification vector. The standardized data matrix is ​​labeled based on the anomaly identification vector to generate a labeled time-series data matrix. The constructed node response feature matrix includes: Multi-scale time series decomposition is performed on each monitoring parameter in the labeled time series data matrix. The decomposition method and decomposition parameters are adaptively selected according to the data change characteristics to generate multi-level decomposition components. The multi-level decomposition components of each monitoring parameter are classified into transient response components, periodic change components, and long-term evolution components according to their time scale characteristics. Correlation verification is performed on the same type of components of different monitoring parameters, and the consistency of component changes is verified by using the statistical correlation between components. Features are extracted from the transient response components, periodic variation components, and long-term evolution components of each monitoring parameter, and multi-scale response feature vectors of each monitoring parameter are generated by combining anomaly identification information. The node response feature matrix is ​​constructed based on the multi-scale response feature vector.

7. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the edge computing-based reservoir dam safety monitoring method according to any one of claims 1 to 5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the edge computing-based reservoir dam safety monitoring method according to any one of claims 1 to 5.

Images