Geological disaster risk point monitoring and emergency response system and method
By combining principal component analysis and K-means clustering with the LSTM model, the low efficiency of manual inspections in the geological disaster monitoring system was solved, and efficient and accurate monitoring of geological disaster risk points and pre-disaster warnings were achieved, improving the accuracy and foresight of the system.
Patent Information
- Application Number
- CN202511138340.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-08-14
- Publication Date
- 2025-09-23
AI Technical Summary
The existing geological disaster monitoring system relies on manual inspections, which has a large workload and low accuracy. It lacks pre-disaster warning capabilities and makes it difficult to achieve efficient and accurate monitoring and emergency response to geological disaster risk points.
By combining principal component analysis and K-means clustering with the LSTM model, feature extraction and classification are performed on the historical data of monitoring nodes to generate failure probability coefficients, thereby realizing the identification and early warning of risk nodes.
It achieves efficient and accurate monitoring of geological disaster risk points, reduces computing resource requirements, improves the foresight of pre-disaster warnings and the comprehensiveness of data, and simplifies the modeling process.
Smart Images
Figure CN120689983A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of geological disaster early warning technology, and in particular to a geological disaster risk point monitoring and emergency response system and method. Background Art
[0002] As an important part of natural disasters, geological disasters have had a serious impact on human society and economic development. Monitoring these geological disasters has become an especially important task in maintaining the safety of people's lives and property and social and economic development. The monitoring of geological disasters usually relies on ground monitoring stations to predict nodes where disasters are prone to occur. In fact, the occurrence of geological disasters depends on manual inspections of monitoring stations, which have problems such as large workload, low accuracy, and inconsistent standards, and cannot meet actual needs.
[0003] Artificial intelligence technology can be used to select and predict monitoring nodes to accurately identify the scope of geological disasters and reduce the computing power required for disaster monitoring. However, existing analyses of geological disaster risk points often focus on post-disaster safety risk screening and analysis, and lack foresight in pre-disaster warnings. In addition, existing disaster warning systems have difficulty extracting disaster-related information from data-disaster areas and lack comprehensiveness.
[0004] Therefore, there is an urgent need for an efficient and accurate geological disaster risk point monitoring and emergency response system and solution to realize the search for risk nodes and pre-disaster data monitoring of risk nodes. Summary of the Invention
[0005] The purpose of the present invention is to provide a geological disaster risk point monitoring and emergency response system and method to solve the following technical problems:
[0006] How to search for risk nodes and conduct pre-disaster data monitoring of risk nodes;
[0007] The purpose of the present invention can be achieved through the following technical solutions:
[0008] A method for monitoring and responding to geological disaster risk points, comprising:
[0009] Step 1: Extract the statistical features of the historical monitoring data of each monitoring node, and perform principal component analysis on the statistical features to extract the principal component data;
[0010] Step 2: Perform K-means cluster analysis on the principal component data of each monitoring node, generate k-type data points to classify each monitoring node, and label the time series data of k-type monitoring nodes;
[0011] Step 3: Input the labeled time series data of the k-type monitoring nodes after K-means clustering into the initial LSTM model for training, and output the failure probability coefficient of the k-type monitoring nodes;
[0012] Step 4: Input the real-time sequence data of k types of monitoring nodes into the trained LSTM model to output the predicted failure probability coefficient of k types of monitoring nodes. The failure risk of the current monitoring node is determined based on the predicted failure probability coefficient of k types of monitoring nodes. If it exceeds the preset failure probability coefficient range, it is determined to be a risky node.
[0013] Step 5: Monitor risk nodes for abnormalities and issue warnings for abnormal situations.
[0014] Preferably, the statistical feature is the time series data matrix of the monitoring site in, represents a set of real numbers, n is the number of time points of each monitoring node, and p is the number of monitoring indicator features corresponding to each time point; assuming that the total number of monitoring nodes is m;
[0015] The method of performing principal component analysis on statistical features to extract principal component features is as follows:
[0016] For matrix X ij Perform Z-score standardization to obtain a standardized data matrix:
[0017] Among them, (X std ) ij is the standardized data matrix; i is the time point index, and i∈{1, 2, ..., n}; j is the monitoring indicator feature index, and j∈{1, 2, ..., p}; σ j is the standard deviation of the jth monitoring indicator feature; μ j is the mean value of the jth monitoring indicator feature;
[0018] The standardized data matrix (X std ) ij Calculate the covariance matrix ∑:
[0019] Among them, n is the total time of the current monitoring node, and the covariance matrix represents a symmetric positive semidefinite matrix, Represents X std , and calculate:
[0020] Among them, S jk Represents the output element of the covariance matrix; After orthogonal matrix; x std,ij Represents the standardized matrix The element in row i and column j in x std,ik Represents the standardized matrix The element in row i and column k of ;
[0021] The element in the jth row and kth column of the covariance matrix ∑ jk is the covariance between the jth feature and the kth feature, and the off-diagonal elements Σ are calculated based on the covariance matrix jk :
[0022]
[0023] Perform eigendecomposition on the input covariance matrix ∑:
[0024] ∑×v k =λ k ×I×v k ; Where I is The identity matrix of v k is the kth eigenvector, and v k ≠0; calculate and obtain the kth eigenvalue λ k , and k={1,2,...,p}, the eigenvalue λ k Arrange in descending order: λ1≥λ2≥...≥λ p ≥0.
[0025] Preferably, the principal component analysis further includes obtaining principal component data based on the eigenvalues, in the following manner:
[0026] Calculate each eigenvalue λ k Cumulative variance contribution rate of principal component number k Among them, λ i represents the i-th eigenvalue, p represents the total number of eigenvalues; i∈[1,p];
[0027] When the cumulative variance contribution rate Con k If it is not less than the preset threshold, the number of principal components k to be retained is determined;
[0028] According to the principal component eigenvector set {v1, v2, ..., v k} Perform principal component space projection and obtain the projection matrix:
[0029] Calculate the principal component score matrix T: T = X std W T ,in, is the principal component data after dimensionality reduction.
[0030] Preferably, the method of performing K-means cluster analysis on the principal component characteristics of each monitoring node is:
[0031] The optimal number of clusters k is determined by the silhouette coefficient:
[0032] k=argmaxS(m)
[0033] Where S(m) is the silhouette coefficient when the number of clusters is m, and the number of clusters is 2≤m≤6.
[0034] Preferably, the mapping relationship between the optimal cluster number k and each monitoring node is obtained:
[0035] The Euclidean distance from each feature data point to the center is calculated according to the optimal number of clusters, k cluster centers are obtained, and the data points of k categories of each monitoring node are labeled.
[0036] Preferably, in step 3, the time series data of the k-type monitoring nodes with labels after K-means clustering are input into the initial LSTM model for training, and the method for outputting the failure probability coefficient of the k-type monitoring nodes is:
[0037] By weighting the k-dimensional probability vector output by LSTM:
[0038]
[0039] Among them, Risk is the failure probability coefficient, p i is the probability vector of the data point of the i-th category; ω i is the weight coefficient of the data point of the i-th category, i∈[1,k].
[0040] Preferably, in step 4, the failure risk of the current monitoring node is determined based on the predicted failure probability coefficients of the k-type monitoring nodes. If the predicted failure probability coefficient exceeds the preset range, the node is determined to be a risk node. The judgment logic of the risk node is as follows:
[0041] If the failure probability coefficients at N consecutive time points all satisfy:
[0042] Risk t >τ and When , an early warning is triggered;
[0043] Where N∈[3, 5]; τ is the dynamic threshold.
[0044] Preferably, the dynamic threshold is calculated as follows:
[0045] Risk distribution based on historical normal data normal The preset ratio quantile Q ith :
[0046] τ=Q ith (Risk normal )+η×σ
[0047] Where η∈[1.5, 2.0], σ is the standard deviation of the normal contribution value.
[0048] A geological disaster risk point monitoring and emergency response system is applied to a geological disaster risk point monitoring and emergency response method, including:
[0049] The principal component analysis module is used to extract the statistical characteristics of the historical monitoring data of each monitoring node, and perform principal component analysis on the statistical characteristics to extract the principal component data;
[0050] The risk extraction module is used to perform K-means clustering analysis on the principal component data of each monitoring node, generate k data points to classify each monitoring node, and label the time series data of the k-type monitoring nodes. The time series data of the k-type monitoring nodes with labels after K-means clustering are input into the initial LSTM model for training, and the failure probability coefficient of the k-type monitoring nodes is output.
[0051] The risk monitoring module is used to input the real-time sequence data of k types of monitoring nodes into the trained LSTM model to output the predicted failure probability coefficient of k types of monitoring nodes, and judge the failure risk of the current monitoring node based on the size of the predicted failure probability coefficient of k types of monitoring nodes. If it exceeds the preset failure probability coefficient range, it is judged as a risk node; monitor whether the risk node has any abnormalities and issue early warnings for abnormal situations.
[0052] Beneficial effects of the present invention:
[0053] (1) The present invention extracts principal component features by performing principal component analysis on statistical features. Principal component analysis is to achieve dimensionality reduction processing of high-risk statistical features. After the detection data of historical monitoring nodes are standardized, feature decomposition is performed through the covariance matrix to extract the principal components with the most information content, thereby optimizing the information extraction process.
[0054] (2) The present invention is based on the principal component analysis method combined with the K-means clustering analysis of the principal component data of each monitoring node to achieve rapid classification of each monitoring node and label the time series data of k types of monitoring nodes; by classifying the historical status of the monitoring nodes into multiple groups, each group represents different operating characteristics, such as normal, abnormal, and possible risks, etc., so as to automatically discover the abnormal operating mode of the monitoring nodes through K-means clustering, including sampling abnormalities or stability abnormalities of some node equipment.
[0055] (3) K-means clustering and LSTM supervised prediction are combined, and the time series data of the k-type monitoring nodes with labels after K-means clustering are input into the initial LSTM model for training, and the fault probability coefficient of the k-type monitoring nodes is output; by calculating the fault probability coefficient, the traditional method of threshold alarm for various types of data is replaced, which reduces the computing pressure of the equipment, and the calculation of the fault probability coefficient also realizes dynamic threshold calculation based on sliding window statistics, which helps to discover the potential correlation of risk areas; at the same time, it ensures that nodes of the same type share a model, which simplifies the modeling process.
[0056] Of course, any product implementing the present invention does not necessarily need to achieve all of the advantages described above at the same time. BRIEF DESCRIPTION OF THE DRAWINGS
[0057] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.
[0058] Figure 1 This is a step diagram of a geological disaster risk point monitoring and emergency response method of the present invention;
[0059] Figure 2 This is a module diagram of a geological disaster risk point monitoring and emergency response system of the present invention. DETAILED DESCRIPTION
[0060] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.
[0061] See also Figure 1 As shown, the present invention is a method for monitoring and responding to geological disaster risk points, the method comprising:
[0062] Step 1: Extract the statistical features of the historical monitoring data of each monitoring node, and perform principal component analysis on the statistical features to extract the principal component data;
[0063] Step 2: Perform K-means cluster analysis on the principal component data of each monitoring node, generate k-type data points to classify each monitoring node, and label the time series data of k-type monitoring nodes;
[0064] Step 3: Input the labeled time series data of the k-type monitoring nodes after K-means clustering into the initial LSTM model for training, and output the failure probability coefficient of the k-type monitoring nodes;
[0065] Step 4: Input the real-time sequence data of k types of monitoring nodes into the trained LSTM model to output the predicted failure probability coefficient of k types of monitoring nodes. The failure risk of the current monitoring node is determined based on the predicted failure probability coefficient of k types of monitoring nodes. If it exceeds the preset failure probability coefficient range, it is determined to be a risky node.
[0066] Step 5: Monitor risk nodes for abnormalities and issue warnings for abnormal situations.
[0067] In the above technical solution, first, step one extracts the statistical characteristics of the historical monitoring data of each monitoring node and performs principal component analysis on the statistical characteristics to extract the principal component data, thereby combining the time domain statistical characteristics with principal component analysis PCA to solve the problem that the dimensions of multi-source sensor data in geological disaster monitoring are difficult to unify; the principal component characteristics are extracted by performing principal component analysis on the statistical characteristics. Principal component analysis is to achieve dimensionality reduction processing of high-risk statistical characteristics. It is to standardize the monitoring data of the historical monitoring nodes and then perform feature decomposition through the covariance matrix to extract the most informative principal components. Compared with the disaster warning model that analyzes historical monitoring data, the key information after dimensionality reduction can be used as the main information for predicting geological disaster risks in the disaster warning model; among them, the historical monitoring data includes sensor signal data deployed at various monitoring nodes, including uniqueness, inclination, margin, water level and environment; historical time series data (time series data) statistics are extracted from it, including at least three time domain characteristics such as mean, variance, kurtosis coefficient, trend slope, data missing rate, and the sum of the absolute values of differences between adjacent sampling points.
[0068] Then, in step 2, K-means cluster analysis is performed on the principal component data of each monitoring node, and cluster analysis is performed on the historical data, mainly generating k types of data points to classify each monitoring node, and labeling the time series data of k types of monitoring nodes; the historical time series data is classified and identified, because K-means requires a preset number of clusters k and is sensitive to the initial center and noise; the optimal number of clusters is obtained by combining the principal component features obtained by principal component analysis, and the clustered data points are grouped according to the similarity of their features, and the historical status of the monitoring nodes are classified into multiple groups. Each group represents different operating characteristics, such as normal, abnormal, and possible risks, so as to automatically discover abnormal operating modes of monitoring nodes through K-means clustering, including sampling abnormalities or stability abnormalities of equipment.
[0069] Furthermore, step three is to combine K-means clustering and LSTM supervised prediction, and further use clustering labels to train the LSTM model on the basis of monitoring the equipment to ensure the dynamic prediction of fault probability; the implementation method is to input the time series data of the k-type monitoring nodes with labels after K-means clustering into the initial LSTM model for training, and output the fault probability coefficient of the k-type monitoring nodes; by calculating the fault probability coefficient, the traditional method of threshold alarm for various types of data is replaced, reducing the computing power pressure of the equipment, and the calculation of the fault probability coefficient also realizes dynamic threshold calculation based on sliding window statistics; and in the risk node monitoring, it is found that K-means clustering does not need to know the node type or risk level in advance, and automatically groups the nodes according to historical behavior patterns, which helps to discover the potential correlation between risk areas; at the same time, it ensures that nodes of the same type share a model, simplifying the modeling process.
[0070] Then, through step 4, the real-time sequence data of k-type monitoring nodes are input into the trained LSTM model to output the predicted failure probability coefficient of k-type monitoring nodes, and the failure risk of the current monitoring node is judged according to the size of the predicted failure probability coefficient of k-type monitoring nodes. If it exceeds the preset failure probability coefficient range, it is judged as a risk node; by utilizing the processing advantages of the trained LSTM model for time series data, it can effectively capture the long-term dependencies and nonlinear dynamic changes in geological disaster monitoring data; and combine the residuals of the LSTM predicted value and the actual observation value to perform more refined anomaly detection.
[0071] Finally, based on the first four steps, step five makes early warning decisions on risk nodes, monitors whether anomalies occur and issues early warnings for abnormal situations; improves the process of rapid search for risk nodes and pre-disaster data monitoring of risk nodes.
[0072] As an embodiment of the present invention, the statistical feature is the time series data matrix of the monitoring station in, represents a set of real numbers, n is the number of time points of each monitoring node, and p is the number of monitoring indicator features corresponding to each time point; assuming that the total number of monitoring nodes is m;
[0073] The method of performing principal component analysis on statistical features to extract principal component features is as follows:
[0074] For matrix X ij Perform Z-score standardization to obtain a standardized data matrix:
[0075] Among them, (X std ) ijis the standardized data matrix; i is the time point index, and i∈{1, 2, ..., n}; j is the monitoring indicator feature index, and j∈{1, 2, ..., p}; σ j is the standard deviation of the jth monitoring indicator feature; μ j is the mean value of the jth monitoring indicator feature;
[0076] The standardized data matrix (X std ) ij Calculate the covariance matrix ∑:
[0077] Among them, n is the total time of the current monitoring node, and the covariance matrix represents a symmetric positive semidefinite matrix, Represents X std , and calculate:
[0078] Among them, S jk Represents the output element of the covariance matrix; After orthogonal matrix; x std,ij Represents the standardized matrix The element in row i and column j in x std,ik Represents the standardized matrix The element in row i and column k of ;
[0079] The element in the jth row and kth column of the covariance matrix ∑ jk is the covariance between the jth feature and the kth feature, and the off-diagonal elements ∑ jk :
[0080]
[0081] Perform eigendecomposition on the input covariance matrix Σ:
[0082] ∑×v k =λ k ×I×v k ; Where I is The identity matrix of v k is the kth eigenvector, and v k ≠0; calculate and obtain the kth eigenvalue λ k , and k={1,2,...,p}, the eigenvalue λ k Arrange in descending order: λ1≥λ2≥...≥λ p ≥0.
[0083] In the above technical solution, each feature is first standardized by Z-score standardization to ensure that its mean is 0 and standard deviation is 1. The standardized data matrix is recorded as X std ; Specifically, by taking the mean μ for each feature j and standard deviation σ j The calculation of the standardized data matrix is (X std ) ij ; Then the covariance matrix is calculated. The covariance matrix can be used to facilitate statistical characterization of the data feature set. The original data is transformed into a new coordinate system (orthogonal coordinate system) through linear transformation to ensure that the variance on the first coordinate axis after transformation is the largest, thereby ensuring most of the original information and simplifying the historical data. Finally, the high-dimensional features are reduced in dimensionality by performing feature decomposition on the input covariance matrix ∑, retaining the main variation directions (such as principal components with a cumulative contribution rate of not less than 90%) to reduce noise and improve clustering efficiency.
[0084] As an embodiment of the present invention, principal component analysis further includes obtaining principal component data based on eigenvalues, and the obtaining method is:
[0085] Calculate each eigenvalue λ k Cumulative variance contribution rate of principal component number k Among them, λ i represents the i-th eigenvalue, p represents the total number of eigenvalues; i∈[1,p];
[0086] When the cumulative variance contribution rate Con k If it is not less than the preset threshold, the number of principal components k to be retained is determined;
[0087] According to the principal component eigenvector set {v1, v2, ..., v k} Perform principal component space projection and obtain the projection matrix:
[0088] Calculate the principal component score matrix T: T = X std W T ,in, is the principal component data after dimensionality reduction.
[0089] In the above technical solution, the calculation method of extracting the principal component ensures that the core features of the monitoring data are retained while significantly reducing the complexity of the data; it ensures that the calculation speed of the equipment is accelerated, and at the same time improves the operability of data processing, and ensures that K-means clustering analysis is further performed based on the principal component features to optimize the monitoring data information. Specifically, by calculating each eigenvalue λ k Cumulative variance contribution rate of principal component number k Among them, λ irepresents the i-th eigenvalue, p represents the total number of eigenvalues; i∈[1, p]; when the cumulative variance contribution rate Con k If it is not less than the preset threshold, the number of principal components k is determined to be retained; and according to the principal component feature vector set {v1, v2, ..., v k} Perform principal component space projection and obtain the projection matrix: At this time, the covariance matrix is reduced according to the projection matrix to calculate the principal component score matrix T: T = X std W T ,in, The principal component data after dimensionality reduction significantly reduces the dimension while retaining most of the original information; finally, a few of the most representative principal component features are obtained.
[0090] As an embodiment of the present invention, the method of performing K-means cluster analysis on the principal component characteristics of each monitoring node is as follows:
[0091] The optimal number of clusters k is determined by the silhouette coefficient:
[0092] k=argmaxS(m)
[0093] Where S(m) is the silhouette coefficient when the number of clusters is m, and the number of clusters is 2≤m≤6; K-means cluster analysis can discover the similarities in the behavioral patterns between monitoring nodes based on the principal component features, and quantitatively analyze these similarities to classify nodes with certain similar dynamic characteristics; the K-means algorithm is used to cluster the principal component feature vectors of all monitoring nodes; K-means divides data points into k clusters, so that the similarity of points within the same cluster is high, and the similarity of points between different clusters is low. Output: Each monitoring node is assigned a category label (1 to k) to identify the cluster to which it belongs; this means that the k-type nodes are similar in their historical behavioral feature patterns; specifically, in one embodiment, the mapping relationship between the optimal number of clusters k and each monitoring node is obtained:
[0094] The Euclidean distance from each feature data point to the center is calculated according to the optimal number of clusters, k cluster centers are obtained, and the data points of k categories of each monitoring node are labeled.
[0095] In the above technical solution, a silhouette coefficient matching the number of clusters is constructed: Where a(i) is the average distance from eigenvalue sample i to other points in the same cluster, b(i) represents the average distance from eigenvalue sample i to the nearest different cluster; n is the total number of eigenvalue samples; and the number of clusters is 2≤m≤6; by constructing a mapping relationship between the optimal number of clusters k and each monitoring node, labeling each monitoring node is achieved. The mapping relationship is obtained by calculating the Euclidean distance from each feature data point to the center according to the optimal number of clusters, obtaining k cluster centers, and labeling the data points of each monitoring node with k categories.
[0096] As an embodiment of the present invention, in step 3, the time series data of the k-type monitoring nodes with labels after K-means clustering are input into the initial LSTM model for training, and the method of outputting the failure probability coefficient of the k-type monitoring nodes is as follows:
[0097] The input structure of the initial LSTM model is:
[0098] Time window construction:
[0099] w t =[x t-L+1 ,x t-L+2 ,...,x t ]
[0100] Among them, x t-L+1 ,x t-L+2 ,...,x t Represents a collection of time series data. Represents a time window; its input layer dimension is: (batch_size, L, d), where d is the number of principal components; L is the window length, and the window length is optimized:
[0101] The test L lengths on the validation set include 6, 12, 18, and 24 hours; the window length L value with the largest F1-score is selected;
[0102] LSTM also includes a two-layer LSTM hidden unit with a unit number range of [32, 128]; a Dropout layer with a dropout rate in the range of [0.2, 0.5]; the output layer is a k-dimensional probability vector activated by Softmax; the k-dimensional probability vector output by LSTM is weighted:
[0103]
[0104] Among them, Risk is the failure probability coefficient, p i is the probability vector of the data point of the i-th category; ω i is the weight coefficient of the data point of the i-th category, i∈[1,k].
[0105] Through experiments, let k = 4; therefore, the probability calculation is: in, By the data point probability vector p i The failure probability coefficient Risk can be obtained by weighting. The weighted fusion of the state probability obtained in the experiment enables the complex equipment operation state to be converted into a scalar risk value.
[0106] As an embodiment of the present invention, in step 4, the failure risk of the current monitoring node is determined based on the predicted failure probability coefficients of the k-type monitoring nodes. If the failure probability coefficient exceeds the preset range, the node is determined to be a risk node. The judgment logic of the risk node is as follows:
[0107] If the failure probability coefficients at N consecutive time points all satisfy:
[0108] Risk t >τ and When , an early warning is triggered;
[0109] Where N∈[3, 5]; τ is the dynamic threshold.
[0110] In the above technical solution, the corresponding conditions are further set for the dynamic warning trigger mode. The failure risk of the current monitoring node is determined according to the size of the dynamically measured failure probability coefficient, and the dynamic threshold range is determined. When the dynamic threshold τ range is exceeded, the warning is triggered. There are four cases in the dynamic threshold τ range obtained by experimental calculation. If the range Risk t At 0≤Risk t <0.3 means the risk level is normal. t When 0.3≤Risk t <0.65 indicates that the risk level is low risk; if 0.65≤Risk t <0.8 means the risk level is medium risk; if Risk t ≥0.8 indicates that the risk level is high risk; and response measures are taken according to the four risk situations and corresponding early warning modes are adopted.
[0111] As an embodiment of the present invention, the dynamic threshold is calculated as follows:
[0112] Risk distribution based on historical normal data normal The preset ratio quantile Q ith :
[0113] τ=Q ith (Risk normal )+η×σ
[0114] Where η∈[1.5, 2.0], σ is the standard deviation of the normal contribution value.
[0115] In the above technical solution, after dynamic warning calculation, the dynamic threshold range is used to select the mechanism for triggering abnormalities and continuous abnormalities, and the warning information is responded to in a timely manner to complete the emergency response judgment.
[0116] A geological disaster risk point monitoring and emergency response system, please refer to Figure 2 As shown, it is applied to a geological disaster risk point monitoring and emergency response method, including:
[0117] The principal component analysis module is used to extract the statistical characteristics of the historical monitoring data of each monitoring node, and perform principal component analysis on the statistical characteristics to extract the principal component data;
[0118] The risk extraction module is used to perform K-means clustering analysis on the principal component data of each monitoring node, generate k data points to classify each monitoring node, and label the time series data of the k-type monitoring nodes. The time series data of the k-type monitoring nodes with labels after K-means clustering are input into the initial LSTM model for training, and the failure probability coefficient of the k-type monitoring nodes is output.
[0119] Risk monitoring module: inputs the real-time sequence data of k types of monitoring nodes into the trained LSTM model to output the predicted failure probability coefficient of k types of monitoring nodes, and judges the failure risk of the current monitoring node based on the size of the predicted failure probability coefficient of k types of monitoring nodes. If it exceeds the preset failure probability coefficient range, it is judged as a risk node; monitors whether the risk node has any abnormalities and issues early warnings for abnormal situations.
[0120] The various embodiments in this specification are described in a progressive manner. Similar portions between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from the other embodiments. In particular, the device, apparatus, and non-volatile computer storage medium embodiments are generally similar to the method embodiments, so their descriptions are relatively simplified. For relevant details, refer to the descriptions of the method embodiments.
[0121] The above content is merely an example and explanation of the concept of the present invention. Those skilled in the art may make various modifications or additions to the described specific embodiments or replace them in a similar manner. As long as they do not deviate from the concept of the invention or exceed the scope defined in this application, they should all fall within the scope of protection of the present invention.
Claims
1. A method for monitoring and emergency response to geological disaster risk points, characterized in that: The method comprises: Step 1: Extract the statistical features of the historical monitoring data of each monitoring node, and perform principal component analysis on the statistical features to extract the principal component data; Step 2: Perform K-means cluster analysis on the principal component data of each monitoring node, generate k-type data points to classify each monitoring node, and label the time series data of k-type monitoring nodes; Step 3: Input the labeled time series data of the k-type monitoring nodes after K-means clustering into the initial LSTM model for training, and output the failure probability coefficient of the k-type monitoring nodes; Step 4: Input the real-time sequence data of k types of monitoring nodes into the trained LSTM model to output the predicted failure probability coefficient of k types of monitoring nodes. The failure risk of the current monitoring node is determined based on the predicted failure probability coefficient of k types of monitoring nodes. If it exceeds the preset failure probability coefficient range, it is determined to be a risky node. Step 5: Monitor risk nodes to determine whether abnormal situations occur and issue alarms for abnormal situations.
2. A geological disaster risk point monitoring and emergency response method according to claim 1, characterized in that: The statistical characteristics are the time series data matrix of the monitoring site in, represents a set of real numbers, n is the number of time points of each monitoring node, and p is the number of monitoring indicator features corresponding to each time point; assuming that the total number of monitoring nodes is m; The method of performing principal component analysis on statistical features to extract principal component features is: For matrix X ij Perform Z-score standardization to obtain a standardized data matrix: Among them, (X std ) ij is the standardized data matrix; i is the time point index, and i∈{1, 2, ..., n}; j is the monitoring indicator feature index, and j∈{1, 2, ..., p}; σ j is the standard deviation of the jth monitoring indicator feature; μ j is the mean value of the jth monitoring indicator feature; The standardized data matrix (X std ) ij Calculate the covariance matrix Σ: Among them, n is the total time of the current monitoring node, and the covariance matrix represents a symmetric positive semidefinite matrix, Represents X std , and calculate: Among them, S jk Represents the output element of the covariance matrix; After orthogonal matrix; x std,ij Represents the standardized matrix The element in row i and column j in x std,ik Represents the standardized matrix The element in row i and column k of ; The element in the jth row and kth column of the covariance matrix ∑ jk is the covariance between the jth feature and the kth feature, and the off-diagonal elements ∑ jk : Perform eigendecomposition on the input covariance matrix ∑: ∑×υ k =λ k ×I×υ k ; Where I is The identity matrix of k is the kth eigenvector, and υ k ≠0; calculate and obtain the kth eigenvalue λ k , and k={1,2,...,p}, the eigenvalue λ k Arrange in descending order: λ1≥λ2≥...≥λ p ≥0.
3. A geological disaster risk point monitoring and emergency response method according to claim 2, characterized in that: The principal component analysis further includes obtaining principal component data based on the eigenvalues, in the following manner: Calculate each eigenvalue λ k Cumulative variance contribution rate of principal component number k Among them, λ i represents the i-th eigenvalue, p represents the total number of eigenvalues; i∈[1,p]; When the cumulative variance contribution rate Con k If it is not less than the preset threshold, the number of principal components k to be retained is determined; According to the principal component eigenvector set {υ1,υ2,...,υ k } Perform principal component space projection and obtain the projection matrix: Calculate the principal component score matrix T: T = X std W T ,in, is the principal component data after dimensionality reduction.
4. A geological disaster risk point monitoring and emergency response method according to claim 1, characterized in that: The method of performing K-means cluster analysis on the principal component characteristics of each monitoring node is as follows: The optimal number of clusters k is determined by the silhouette coefficient: k=argmaxS(m) Where S(m) is the silhouette coefficient when the number of clusters is m, and the number of clusters is 2≤m≤6.
5. A geological disaster risk point monitoring and emergency response method according to claim 4, characterized in that: Get the mapping relationship between the optimal cluster number k and each monitoring node: The Euclidean distance from each feature data point to the center is calculated according to the optimal number of clusters, k cluster centers are obtained, and the data points of k categories of each monitoring node are labeled.
6. A geological disaster risk point monitoring and emergency response method according to claim 5, characterized in that: In step 3, the time series data of the k-type monitoring nodes with labels after K-means clustering are input into the initial LSTM model for training. The method for outputting the failure probability coefficient of the k-type monitoring nodes is: By weighting the probability vector output by LSTM: Among them, Risk is the failure probability coefficient, p i is the probability vector of the data point of the i-th category; ω i is the weight coefficient of the data point of the i-th category, i∈[1,k].
7. A geological disaster risk point monitoring and emergency response method according to claim 6, characterized in that: In step 4, the failure risk of the current monitoring node is determined based on the predicted failure probability coefficients of the k-type monitoring nodes. If the failure probability coefficient exceeds the preset range, the node is determined to be a risk node. The judgment logic of the risk node is as follows: If the failure probability coefficients at N consecutive time points all satisfy: Risk t >τ and When , an early warning is triggered; Where N∈[3, 5]; τ is the dynamic threshold.
8. A geological disaster risk point monitoring and emergency response method according to claim 7, characterized in that: The dynamic threshold is calculated as follows: Risk distribution based on historical normal data normal Preset ratio quantile Q ith : τ=Q ith (Risk normal )+η×σ Where η∈[1.5, 2.0], σ is the standard deviation of the normal contribution value.
9. A geological disaster risk point monitoring and emergency response system, characterized in that: A geological disaster risk point monitoring and emergency response method as described in any one of claims 1 to 8, comprising: The principal component analysis module is used to extract the statistical characteristics of the historical monitoring data of each monitoring node, and perform principal component analysis on the statistical characteristics to extract the principal component data; The risk extraction module is used to perform K-means clustering analysis on the principal component data of each monitoring node, generate k data points to classify each monitoring node, and label the time series data of the k-type monitoring nodes. The time series data of the k-type monitoring nodes with labels after K-means clustering are input into the initial LSTM model for training, and the failure probability coefficient of the k-type monitoring nodes is output. The risk monitoring module is used to input the real-time sequence data of k types of monitoring nodes into the trained LSTM model to output the predicted failure probability coefficient of k types of monitoring nodes, and judge the failure risk of the current monitoring node based on the size of the predicted failure probability coefficient of k types of monitoring nodes. If it exceeds the preset failure probability coefficient range, it is judged as a risk node; monitor whether the risk node has any abnormalities and issue early warnings for abnormal situations.