A mine groundwater level monitoring and disaster early warning system

Abstract

The invention relates to the field of water level monitoring and disaster early warning, and specifically is a mine groundwater level monitoring and disaster early warning system, comprising a data acquisition and processing module, a groundwater level intelligent analysis module, a surface deformation intelligent analysis module, a disaster early warning module and a user control module; the invention creatively adopts a step polymorphic graph convolution network based on a dynamic Laplace matrix, combines time series water level data of multiple sampling points, constructs a dynamic Laplace matrix with global characteristics, highlights the global spatiotemporal characteristics of the groundwater level, and uses the spatiotemporal graph convolution network to predict and analyze the water level, so as to obtain accurate groundwater level monitoring results with global characteristics; the invention creatively performs step heterogeneous transformation on the surface deformation data, and adopts a prediction algorithm based on empirical mode decomposition and autoregressive moving average to predict and analyze the surface deformation, so as to realize the joint monitoring of water level data and surface deformation, and make accurate disaster early warning in time.

Description

Technical Field

[0001] The invention belongs to the field of water level monitoring and disaster early warning, and in particular to a mine underground water level monitoring and disaster early warning system. Background Art

[0002] The groundwater level is one of the common safety factors in mining projects. Abnormal fluctuations in the groundwater level in mines may lead to safety accidents such as mine water inrush, collapse, and geological disasters. Accurate monitoring of the groundwater level is crucial to the stable and efficient operation of mine production. By monitoring the groundwater level, the drainage system of the mine can be adjusted in time to ensure the normal operation of the mine production, provide early warning signals for the occurrence of geological disasters, and help take corresponding preventive measures. However, existing mine groundwater level monitoring usually relies on limited monitoring point data and cannot predict and analyze the global water level. Existing monitoring technology usually requires manual intervention and data processing, and the real-time and timeliness of the monitoring results need to be improved. Summary of the invention

[0003] In view of the above situation, in order to overcome the defects of the prior art, the existing mine groundwater level monitoring usually relies on limited monitoring point data, cannot predict and analyze the global water level, and the real-time and timeliness of the monitoring results need to be improved. The present invention creatively adopts a step polymorphic graph convolution network based on a dynamic Laplace matrix, and constructs a dynamic graph and a static graph with global characteristics by combining the time series water level data of multiple sampling points. After the dynamic graph is subjected to step conversion, multi-head attention processing and matrix diagonalization processing, a step heterogeneous spatiotemporal matrix is ​​obtained and a dynamic Laplace matrix is ​​constructed, which highlights The temporal evolution law and characteristics of the graph structure, that is, highlighting the global spatiotemporal characteristics of the groundwater level, and using the spatiotemporal graph convolutional network T-GCN model to extract the spatiotemporal characteristics of water level data and predict and analyze the water level in real time, so as to obtain accurate groundwater level monitoring results with global characteristics; the present invention creatively performs step heterogeneous transformation on the surface deformation data, and adopts a prediction algorithm based on empirical mode decomposition and autoregressive moving average to predict and analyze the surface deformation, so as to efficiently realize the joint monitoring of water level data and surface deformation, and finally make accurate disaster warnings in time according to all real-time monitoring results.

[0004] The present invention provides a mine groundwater level monitoring and disaster early warning system, comprising a data acquisition and processing module, a groundwater level intelligent analysis module, a surface deformation intelligent analysis module, a disaster early warning module and a user control module;

[0005] The data acquisition and processing module uses a water level acquisition instrument to collect underground historical water level data of multiple sampling points of the mine, and uses SBAS-InSAR technology to collect surface historical deformation data of multiple sampling points of the mine, to obtain time-series water level data and deformation data of all sampling points, and simultaneously collects real-time underground water level data and real-time surface deformation data of multiple sampling points of the mine in real time, sends the time-series water level data and deformation data of all sampling points to the underground water level intelligent analysis module and the surface deformation intelligent analysis module, and sends the real-time underground water level data and real-time surface deformation data of multiple sampling points to the disaster warning module;

[0006] The underground water level intelligent analysis module adopts a step polymorphic graph convolution network based on a dynamic Laplace matrix to construct dynamic and static graphs of the time series water level data of all sampling points, and after performing step conversion, multi-head attention processing and matrix diagonalization processing on the dynamic graph, constructs a dynamic Laplace matrix, and uses a spatiotemporal graph convolution network T-GCN model to extract spatiotemporal features and perform water level prediction analysis to obtain underground historical water level monitoring results. Finally, according to the underground historical water level monitoring results, the step polymorphic graph convolution network based on the dynamic Laplace matrix is ​​trained to obtain an optimized graph convolution network, and the optimized graph convolution network is sent to the surface deformation intelligent analysis module and the disaster warning module;

[0007] The surface deformation intelligent analysis module constructs a step heterogeneous spatiotemporal matrix of the mine surface deformation data through the optimized graph convolutional network, and finally predicts and analyzes the surface deformation through the prediction algorithm EMD-ARIMA based on empirical mode decomposition and autoregressive moving average to obtain the surface historical deformation monitoring results, and finally optimizes the EMD-ARIMA algorithm according to the surface historical deformation monitoring results to obtain the optimized EMD-ARIMA algorithm, and sends the optimized EMD-ARIMA algorithm to the disaster warning module;

[0008] The disaster warning module uses the optimized graph convolutional network and the optimized EMD-ARIMA algorithm to analyze and monitor the real-time groundwater level and real-time surface deformation data and set safety thresholds. When the monitoring results exceed the safety threshold, an alarm is issued;

[0009] The user control module allows users to view monitoring information, receive real-time updates and alarms, and also allows users to control all modules and store data.

[0010] Furthermore, in the groundwater level intelligent analysis module, a step polymorphic graph convolutional network based on a dynamic Laplace matrix specifically includes the following steps:

[0011] Step S1: Construct a heterogeneous dynamic graph based on the time series water level data of all sampling points:

[0012] ;

[0013] ;

[0014] In the formula, Represents a dynamic point, Represents a dynamic graph, represents the order number of the sampling point, represents the total number of sampling points, represents the sampling timing, Representative The sampling points are Water level data at the moment, Indicates The edge set of sampling points and adjacent sampling points, The dimension is * The weighted adjacency matrix is ​​used to represent the connectivity relationship of dynamic points in the dynamic graph;

[0015] Step S2: Construct a heterogeneous space-time matrix. Construct a heterogeneous space-time matrix based on the heterogeneous dynamic graph:

[0016] ;

[0017] ;

[0018] In the formula, Representatives through Convolution Moment The sampling point feature vectors, is the total number of features, Indicates The sampling points are The feature vector set at time, Represents the heterogeneous spatiotemporal matrix composed of all sampling points;

[0019] Step S3: Construct a step heterogeneous space-time matrix. According to the historical time series water level data of the sampling points, construct a step heterogeneous space-time matrix { },in is the preset sampling threshold, is the sampling frequency;

[0020] Step S4: Construct a heterogeneous static graph, and construct a heterogeneous static graph based on the relative geographical locations between the sampling points and the heterogeneous dynamic graph at a specific time:

[0021] ;

[0022] ;

[0023] In the formula, and Represents the sampling points With sampling point The coordinates between Represents the sampling points calculated by the haversine function With sampling point The distance between and To adjust the preset parameters of the matrix distribution and sparsity, is an exponential function, Representative sampling point With sampling point The edge weights between is the matrix of edge weights between all sampling points, represents the edge set, For all sampling points at a specific time Water level data at the moment;

[0024] Step S5: Multi-head attention processing, multi-head attention processing is performed on the step-heterogeneous spatiotemporal matrix:

[0025] ;

[0026] In the formula, represents multi-head attention processing, represents the step-heterogeneous spatiotemporal matrix, where the three step-heterogeneous spatiotemporal matrices are used for query, key and value feature extraction respectively. Represents the weight matrix processed by multi-head attention;

[0027] Step S6: Diagonalization of weight matrix:

[0028] ;

[0029] In the formula, Represents the sampling point The weight matrix obtained by sampling is Represents the sampling point The weight matrix obtained by sampling is i ， j ， k ∈ [1 ， c ] , represents the diagonalized weight matrix;

[0030] Step S7: Dynamic Laplacian matrix construction:

[0031] ;

[0032] In the formula, represents the diagonalized weight matrix, The matrix representing the edge weights between all sample points, The representative dimension is The identity matrix of represents the dynamic Laplacian matrix;

[0033] Step S8: spatiotemporal feature learning, normalizing the dynamic Laplace matrix and using the T-GCN model to learn spatiotemporal features to obtain the spatiotemporal features of the groundwater level data;

[0034] Step S9: Activation function processing, using the activation function to activate the spatiotemporal features to obtain the historical underground water level monitoring results.

[0035] Furthermore, in the intelligent analysis module of surface deformation, the optimized graph convolutional network is used to construct a step heterogeneous spatiotemporal matrix of the mine surface deformation data. Finally, the surface deformation is predicted and analyzed by the prediction algorithm EMD-ARIMA based on empirical mode decomposition technology and autoregressive moving average to obtain the monitoring results of surface historical deformation. The specific steps include:

[0036] Step Q1: Construction of deformation step heterogeneous spatiotemporal matrix. Taking the temporal deformation data of all sampling points as input, using the optimized graph convolutional network, repeating steps S1 to S3, and obtaining the deformation step heterogeneous spatiotemporal matrix.

[0037] Step Q2: EMD decomposition, perform empirical mode decomposition on the deformation step heterogeneous space-time matrix to obtain m decomposition components and a residual;

[0038] Step Q3: ARIMA prediction: perform autoregressive moving average prediction on the m decomposition components and one residual respectively, and perform weighted summation on the m+1 prediction results to obtain the historical surface deformation monitoring results.

[0039] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0040] (1) In view of the problem that the existing underground water level monitoring in mines usually relies on limited monitoring point data and cannot predict and analyze the global water level, and the real-time and timeliness of the monitoring results need to be improved, the present invention creatively adopts a step polymorphic graph convolutional network based on a dynamic Laplace matrix. By combining the time series water level data of multiple sampling points, a dynamic graph and a static graph with global characteristics are constructed. After the dynamic graph is subjected to step transformation, multi-head attention processing and matrix diagonalization processing, a step heterogeneous spatiotemporal matrix is ​​obtained and a dynamic Laplace matrix is ​​constructed, highlighting the time evolution law and characteristics of the graph structure, that is, highlighting the spatiotemporal characteristics of the global groundwater level. The spatiotemporal graph convolutional network T-GCN model is used to extract the spatiotemporal characteristics of the water level data in real time and predict and analyze the water level, so as to obtain accurate groundwater level monitoring results with global characteristics.

[0041] (2) The present invention creatively performs step heterogeneous transformation on the surface deformation data, and adopts a prediction algorithm based on empirical mode decomposition and autoregressive moving average to predict and analyze the surface deformation, thereby efficiently realizing the joint monitoring of water level data and surface deformation, and finally making timely and accurate disaster warnings based on all real-time monitoring results. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 A module diagram of a mine groundwater level monitoring and disaster early warning system provided by the present invention;

[0043] Figure 2 This is a flow chart of the EMD-ARIMA algorithm.

[0044] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0045] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0046] Example 1, see Figure 1 , a mine groundwater level monitoring and disaster early warning system provided by the present invention includes a data acquisition and processing module, a groundwater level intelligent analysis module, a surface deformation intelligent analysis module, a disaster early warning module and a user control module;

[0047] The data acquisition and processing module uses a water level acquisition instrument to collect underground historical water level data of 11 sampling points of the mine, and uses SBAS-InSAR technology to collect surface historical deformation data of multiple sampling points of the mine, to obtain time-series water level data and deformation data of all sampling points, and simultaneously collects real-time underground water level data and real-time surface deformation data of the 11 sampling points of the mine in real time, sends the time-series water level data and deformation data of all sampling points to the underground water level intelligent analysis module and the surface deformation intelligent analysis module, and sends the real-time underground water level data and real-time surface deformation data of the 11 sampling points to the disaster warning module;

[0048] The underground water level intelligent analysis module adopts a step polymorphic graph convolution network based on a dynamic Laplace matrix to construct dynamic and static graphs of the time series water level data of all sampling points, and after performing step conversion, multi-head attention processing and matrix diagonalization processing on the dynamic graph, constructs a dynamic Laplace matrix, and uses a spatiotemporal graph convolution network T-GCN model to extract spatiotemporal features and perform water level prediction analysis to obtain underground historical water level monitoring results. Finally, according to the underground historical water level monitoring results, the step polymorphic graph convolution network based on the dynamic Laplace matrix is ​​trained to obtain an optimized graph convolution network, and the optimized graph convolution network is sent to the surface deformation intelligent analysis module and the disaster warning module;

[0049] The surface deformation intelligent analysis module constructs a step heterogeneous spatiotemporal matrix of the mine surface deformation data through the optimized graph convolutional network, and finally predicts and analyzes the surface deformation through the prediction algorithm EMD-ARIMA based on empirical mode decomposition and autoregressive moving average to obtain the surface historical deformation monitoring results, and finally optimizes the EMD-ARIMA algorithm according to the surface historical deformation monitoring results to obtain the optimized EMD-ARIMA algorithm, and sends the optimized EMD-ARIMA algorithm to the disaster warning module;

[0050] The disaster warning module uses the optimized graph convolutional network and the optimized EMD-ARIMA algorithm to analyze and monitor the real-time groundwater level and real-time surface deformation data and set safety thresholds. When the monitoring results exceed the safety threshold, an alarm is issued;

[0051] The user control module allows users to view monitoring information, receive real-time updates and alarms, and also allows users to control all modules and store data.

[0052] Embodiment 2: This embodiment is based on the above embodiment. In the groundwater level intelligent analysis module, a step polymorphic graph convolution network based on a dynamic Laplacian matrix specifically includes the following steps:

[0053] Step S1: Construct a heterogeneous dynamic graph based on the time series water level data of all sampling points:

[0054] ;

[0055] ;

[0056] In the formula, Represents a dynamic point, Represents a dynamic graph, represents the order number of the sampling point, represents the total number of sampling points, represents the sampling timing, Representative The sampling points are Water level data at the moment, Indicates The edge set of sampling points and adjacent sampling points, The dimension is * The weighted adjacency matrix is ​​used to represent the connectivity relationship of dynamic points in the dynamic graph;

[0057] Step S2: Construct a heterogeneous space-time matrix. Construct a heterogeneous space-time matrix based on the heterogeneous dynamic graph:

[0058] ;

[0059] ;

[0060] In the formula, Representatives through Convolution Moment The sampling point feature vectors, is the total number of features, Indicates The sampling points are The feature vector set at time, Represents the heterogeneous spatiotemporal matrix composed of all sampling points;

[0061] Step S3: Construct a step heterogeneous space-time matrix. According to the historical time series water level data of the sampling points, construct a step heterogeneous space-time matrix { },in The preset sampling threshold is 5 days. is the sampling frequency, which is preset to 1 day;

[0062] Step S4: Construct a heterogeneous static graph, and construct a heterogeneous static graph based on the relative geographical locations between the sampling points and the heterogeneous dynamic graph at a specific time:

[0063] ;

[0064] ;

[0065] In the formula, and Represents the sampling points With sampling point The coordinates between Represents the sampling points calculated by the haversine function With sampling point The distance between and To adjust the preset parameters of the matrix distribution and sparsity, is an exponential function, Representative sampling point With sampling point The edge weights between is the matrix of edge weights between all sampling points, represents the edge set, For all sampling points at a specific time Water level data at the moment;

[0066] Step S5: Multi-head attention processing, multi-head attention processing is performed on the step-heterogeneous spatiotemporal matrix:

[0067] ;

[0068] In the formula, represents multi-head attention processing, represents the step-heterogeneous spatiotemporal matrix, where the three step-heterogeneous spatiotemporal matrices are used for query, key and value feature extraction respectively. Represents the weight matrix processed by multi-head attention;

[0069] Step S6: Diagonalization of weight matrix:

[0070] ;

[0071] In the formula, Represents the sampling point The weight matrix obtained by sampling is Represents the sampling point The weight matrix obtained by sampling is i ， j ， k ∈ [1 ， 11 ] , represents the diagonalized weight matrix;

[0072] Step S7: Dynamic Laplacian matrix construction:

[0073] ;

[0074] In the formula, represents the diagonalized weight matrix, The matrix representing the edge weights between all sample points, The representative dimension is The identity matrix of represents the dynamic Laplacian matrix;

[0075] Step S8: spatiotemporal feature learning, normalizing the dynamic Laplace matrix and using the T-GCN model to learn spatiotemporal features to obtain the spatiotemporal features of the groundwater level data;

[0076] Step S9: Activation function processing, using the activation function to activate the spatiotemporal features to obtain the historical underground water level monitoring results.

[0077] By performing the above operations, in view of the problem that the existing mine groundwater level monitoring usually relies on limited monitoring point data, cannot predict and analyze the global water level, and the real-time and timeliness of the monitoring results need to be improved, the present invention creatively adopts a step polymorphic graph convolutional network based on a dynamic Laplace matrix, and constructs a dynamic graph and a static graph with global characteristics by combining the time series water level data of 11 sampling points. After step conversion, multi-head attention processing and matrix diagonalization processing are performed on the dynamic graph, a step heterogeneous spatiotemporal matrix is ​​obtained and a dynamic Laplace matrix is ​​constructed, highlighting the time evolution law and characteristics of the graph structure, that is, highlighting the spatiotemporal characteristics of the global groundwater level, and using the spatiotemporal graph convolutional network T-GCN model to extract the spatiotemporal characteristics of the water level data in real time and predict and analyze the water level, so as to obtain accurate groundwater level monitoring results with global characteristics.

[0078] Example 3, see Figure 2 This embodiment is based on the above embodiment. In the surface deformation intelligent analysis module, the prediction algorithm EMD-ARIMA based on empirical mode decomposition and autoregressive moving average specifically includes the following steps:

[0079] Step Q1: Construction of deformation step heterogeneous spatiotemporal matrix. Taking the temporal deformation data of all sampling points as input, using the optimized graph convolutional network, repeating steps S1 to S3, and obtaining the deformation step heterogeneous spatiotemporal matrix.

[0080] Step Q2: EMD decomposition, perform empirical mode decomposition on the deformation step heterogeneous space-time matrix to obtain four decomposition components IMF and a residual;

[0081] Step Q3: ARIMA prediction: perform autoregressive moving average prediction on the four decomposition components and one residual, and perform weighted summation on the five prediction results to obtain the surface historical deformation monitoring results.

[0082] By executing the above operations, the present invention creatively performs step heterogeneous transformation on the surface deformation data, and adopts a prediction algorithm based on empirical mode decomposition and autoregressive moving average to predict and analyze the surface deformation, so as to efficiently realize the joint monitoring of water level data and surface deformation, and finally make accurate disaster warnings in time according to all real-time monitoring results.

[0083] Embodiment 4: This embodiment is based on the above embodiment and performs multi-head attention processing, specifically including the following steps:

[0084] Step S51: linear transformation, linearly transforming the step heterogeneous spatiotemporal matrix through the learned weight matrix to obtain query, key and value vectors;

[0085] Step S52: Divide the head, divide the linearly transformed query, key and value vectors into multiple heads and perform multiple groups of attention calculations in parallel;

[0086] Step S53: Calculate the attention score, calculate the attention score for each head by the dot product of the query vector and the key vector and perform a scaling operation to stabilize the gradient;

[0087] Step S54: Calculate the attention weight, and obtain the attention weight by passing the attention score through the softmax function, which represents the importance of each position to the current position;

[0088] Step S55: weighted summation, multiplying the value vector calculated for each head by the attention weight, and then performing weighted summation on the results of all heads;

[0089] Step S56: Linear transformation output, the weighted summation result is linearly transformed to obtain the final output.

[0090] Embodiment 5, this embodiment is based on the above embodiment, haversine function, and specifically includes the following steps:

[0091] Obtain longitude and latitude information and determine the longitude and latitude coordinates of the two sampling points;

[0092] Convert longitude and latitude to radians, radians = longitude and latitude * π / 180;

[0093] Calculate the Haversine formula:

[0094] ;

[0095] In the formula, is an auxiliary variable in the Haversine formula, is the latitude of the second sampling point minus the latitude of the first sampling point. is the longitude of the second point minus the longitude of the first point in radians, and are the latitude arcs of the first and second points respectively;

[0096] Calculate the distance. The final distance calculation formula is as follows:

[0097] ;

[0098] In the formula, represents the radius of the Earth, The desired distance.

[0099] Embodiment 6: This embodiment is based on the above embodiment, and the EMD decomposition specifically includes the following steps:

[0100] Step Q21: determine the extreme points and detect the extreme points of the deformation step heterogeneous space-time matrix;

[0101] Step Q22: Connect the mean values, interpolate the extreme points, and obtain the upper and lower envelopes of the data by connecting the extreme points as the mean curve of the current signal;

[0102] Step Q23: Extract the intrinsic mode function IMF, subtract the mean curve from the original signal, and obtain a new signal with fewer local extreme points, which is called the first-order intrinsic mode function IMF).

[0103] Step Q24: repeat step Q21 to step Q23 until 4 IMF components, i.e. 4 decomposition components, are obtained;

[0104] Step Q25: Extract the residual. The last IMF obtained is considered to be the last modal function of the signal, and the remaining signal extracted is the residual.

[0105] The historical water level data and surface data of Benxi Iron Ore Area in Anshan, Liaoning Province were collected as test sets, and the prediction capabilities corresponding to different numbers of intrinsic mode components (IMFs) were trained:

[0106] Number: 4; RMSE: 0.548; MAPE (%): 0.009; MAE: 0.561;

[0107] Number: 8; RMSE: 0.674; MAPE (%): 0.023; MAE: 0.687;

[0108] Number: 12; RMSE: 0.821; MAPE (%): 0.054; MAE: 0.901.

[0109] Embodiment 7, this embodiment is based on the above embodiment, ARIMA forecasting, specifically includes the following steps:

[0110] Step Q31: Determine the difference order, and determine the difference order that makes the sequence stable through the autocorrelation diagram and partial autocorrelation diagram of the four decomposition components and one residual;

[0111] Step Q32: Determine the parameters of the ARIMA model by observing the ACF and PACF graphs to determine the order (p-value) of the autoregressive term (AR) and the order (q-value) of the moving average term (MA);

[0112] Step Q33: Fitting the ARIMA model, using the selected (p, d, q) parameters to fit the ARIMA model;

[0113] Step Q34: Model diagnosis, check the autocorrelation and normality of the residuals to ensure that the model fits well;

[0114] Step Q35: Model prediction, using the fitted ARIMA model to predict future time points.

[0115] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0116] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

[0117] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.

Claims

1. A mine groundwater level monitoring and disaster early warning system, characterized in that: It includes data acquisition and processing module, groundwater level intelligent analysis module, surface deformation intelligent analysis module, disaster warning module and user control module; The data acquisition and processing module collects the underground historical water level data of multiple sampling points of the mine, collects the surface historical deformation data of multiple sampling points of the mine, obtains the time series water level data and deformation data of all sampling points, and collects the real-time underground water level data and real-time surface deformation data of multiple sampling points of the mine in real time, sends the time series water level data and deformation data of all sampling points to the underground water level intelligent analysis module and the surface deformation intelligent analysis module, and sends the real-time underground water level data and real-time surface deformation data of multiple sampling points to the disaster warning module; The underground water level intelligent analysis module uses a step polymorphic graph convolutional network based on a dynamic Laplace matrix to perform feature extraction and water level prediction analysis on the time series water level data of all sampling points to obtain the underground historical water level monitoring results. Finally, the step polymorphic graph convolutional network based on a dynamic Laplace matrix is ​​trained according to the underground historical water level monitoring results to obtain an optimized graph convolutional network, and the optimized graph convolutional network is sent to the surface deformation intelligent analysis module and the disaster warning module. The step polymorphic graph convolutional network based on the dynamic Laplacian matrix specifically includes the following steps: Step S1: Construct a heterogeneous dynamic graph based on the time series water level data of all sampling points: ； ； In the formula, Represents a dynamic point, Represents a dynamic graph, represents the order number of the sampling point, represents the total number of sampling points, represents the sampling timing, Representative The sampling points are Water level data at the moment, Indicates The edge set of sampling points and adjacent sampling points, The dimension is * The weighted adjacency matrix is ​​used to represent the connectivity relationship of dynamic points in the dynamic graph; Step S2: Construct a heterogeneous space-time matrix. Construct a heterogeneous space-time matrix based on the heterogeneous dynamic graph: ； ； In the formula, Representatives through Convolution Moment The sampling point feature vectors, is the total number of features, Indicates The sampling points are The feature vector set at time, Represents the heterogeneous spatiotemporal matrix composed of all sampling points; Step S3: Construct a step heterogeneous space-time matrix. According to the historical time series water level data of the sampling points, construct a step heterogeneous space-time matrix { },in is the preset sampling threshold, is the sampling frequency; Step S4: Construct a heterogeneous static graph, and construct a heterogeneous static graph based on the relative geographical locations between the sampling points and the heterogeneous dynamic graph at a specific time: ； ； In the formula, and Represents the sampling points With sampling point The coordinates between Represents the sampling points calculated by the haversine function With sampling point The distance between and To adjust the preset parameters of the matrix distribution and sparsity, is an exponential function, Representative sampling points With sampling point The edge weights between is the matrix of edge weights between all sampling points, represents the edge set, For all sampling points at a specific time Water level data at the moment; Step S5: Multi-head attention processing, multi-head attention processing is performed on the step-heterogeneous spatiotemporal matrix: ； In the formula, represents multi-head attention processing, represents the step-heterogeneous spatiotemporal matrix, where the three step-heterogeneous spatiotemporal matrices are used for query, key and value feature extraction respectively. Represents the weight matrix processed by multi-head attention; Step S6: Diagonalization of weight matrix: ； In the formula, Represents the sampling point The weight matrix obtained by sampling is Represents the sampling point The weight matrix obtained by sampling is , represents the diagonalized weight matrix; Step S7: Dynamic Laplacian matrix construction: ； In the formula, represents the diagonalized weight matrix, The matrix representing the edge weights between all sample points, The representative dimension is The identity matrix of represents the dynamic Laplacian matrix; Step S8: spatiotemporal feature learning, normalizing the dynamic Laplace matrix and using the T-GCN model to learn spatiotemporal features to obtain the spatiotemporal features of the groundwater level data; Step S9: Activation function processing, using the activation function to activate the spatiotemporal features to obtain the underground historical water level monitoring results; The surface deformation intelligent analysis module constructs a step heterogeneous spatiotemporal matrix of the mine surface deformation data through the optimized graph convolutional network, and finally predicts and analyzes the surface deformation through the prediction algorithm EMD-ARIMA based on empirical mode decomposition and autoregressive moving average to obtain the surface historical deformation monitoring results, and finally optimizes the EMD-ARIMA algorithm according to the surface historical deformation monitoring results to obtain the optimized EMD-ARIMA algorithm, and sends the optimized EMD-ARIMA algorithm to the disaster warning module; The prediction and analysis of surface deformation is carried out to obtain the monitoring results of historical surface deformation, which specifically includes the following steps: Step Q1: Construction of deformation step heterogeneous spatiotemporal matrix. Taking the temporal deformation data of all sampling points as input, using the optimized graph convolutional network, repeating steps S1 to S3, and obtaining the deformation step heterogeneous spatiotemporal matrix. Step Q2: EMD decomposition, perform empirical mode decomposition on the deformation step heterogeneous space-time matrix to obtain m decomposition components and a residual; Step Q3: ARIMA prediction, perform autoregressive moving average prediction on a decomposition components and a residual, and perform weighted summation on a+1 prediction results to obtain the surface historical deformation monitoring results; The disaster warning module uses the optimized graph convolutional network and the optimized EMD-ARIMA algorithm to analyze and monitor the real-time groundwater level and real-time surface deformation data and set safety thresholds. When the monitoring results exceed the safety threshold, an alarm is issued; The user control module allows users to view monitoring information, receive real-time updates and alarms, and also allows users to control all modules and store data.

Images