Tailing dam stability analysis method based on GCN-AGRU fusion model and related device

Abstract

The invention belongs to the field of safety and stability analysis of tailing dams, and discloses a tailing dam stability analysis method based on a GCN-AGRU fusion model and a related device, and the method comprises the steps: obtaining the monitoring data of a multi-source heterogeneous sensor of a tailing dam; preprocessing the monitoring data to obtain multi-source heterogeneous time series data; performing feature analysis on the multi-source heterogeneous time series data to obtain key factors of the stable state of the tailing dam; constructing a heterogeneous hierarchical causal graph by using the key factors; determining a key factor stable state index system according to the heterogeneous hierarchy causal diagram; a multi-source GCN-AGRU fusion space-time prediction model is utilized to obtain space information of the state of the tailing dam; performing time correlation modeling on the multi-source heterogeneous time sequence data to obtain a time sequence of the state of the tailing dam; calculating a predicted value of the multi-source heterogeneous sensor according to the spatial information and the time sequence; and determining the stability state trend of the tailing dam according to the predicted value in combination with a key factor stability state index system. The intelligent sensing of the situation of the tailing dam can be realized.

Description

Technical Field

[0001] The present invention belongs to the field of safety stability analysis of tailings dams, and relates to a tailings dam stability analysis method based on a GCN-AGRU fusion model and related devices. Background Art

[0002] Tailings dams are hydraulic sediments, and most of the factors that affect their stability are dynamic and uncertain. They are easily affected by the external environment, such as natural phenomena such as rainfall and earthquakes, and human activities such as slag accumulation, which can lead to dam failure accidents. In order to effectively curb the occurrence of such high-impact and high-risk accidents, it is necessary to gradually improve the supervision and maintenance requirements of the dam body and improve the tailings dam monitoring system. Existing sensors can automatically and continuously monitor the environment (air and water level), slope stability and ore discharge, and timely grasp the possible abnormal information of the dam body. Therefore, mining useful information from massive monitoring data is the basis for mastering the operation status of tailings dams, and preprocessing monitoring data, accurately predicting its development trend and conducting intelligent analysis are very important for safety management. At present, many achievements have been made in the research on the stability of tailings dams, mainly focusing on numerical simulation, ecological governance, risk prevention and safety evaluation, and there are few studies on the analysis and prediction of the evolution of dam performance and the spatial and temporal precursor characteristics of accidents. Summary of the invention

[0003] The purpose of the present invention is to overcome the shortcomings of the prior art, make use of the great advantages of artificial intelligence technology in dealing with highly nonlinear and uncertain factors, utilize the redundancy and complementarity of multi-dimensional spatiotemporal data, explore the relationship between factors, and predict and analyze the stability state of the dam body, and provide a tailings dam stability analysis method and related devices based on the GCN-AGRU fusion model, thereby realizing intelligent perception of the tailings dam situation.

[0004] To achieve the above objectives, the present invention adopts the following technical solutions: A tailings dam stability analysis method based on the GCN-AGRU fusion model includes the following processes: Obtain monitoring data from multi-source heterogeneous sensors at tailings dams; Preprocessing the monitoring data of the multi-source heterogeneous sensors to obtain multi-source heterogeneous time series data that conforms to a standard normal distribution; Performing feature analysis on the multi-source heterogeneous time series data to obtain key factors for the stability of the tailings dam; Using the key factors, a heterogeneous hierarchical causal diagram is constructed; According to the heterogeneous hierarchical causal diagram, the correlation of multi-source heterogeneous time series data is calculated, the connection between key factors and the stability state of the tailings dam is established, and the key factor stability state indicator system is obtained; The constructed multi-source GCN-AGRU fusion spatiotemporal prediction model is used to analyze the spatial information of the multi-source heterogeneous time series data to obtain the spatial information of the tailings dam state; the time correlation modeling of the multi-source heterogeneous time series data is performed to obtain the time series of the tailings dam state; the prediction value of the multi-source heterogeneous sensor is calculated based on the spatial information and time series of the multi-source heterogeneous time series data obtained by fusion analysis; According to the predicted values ​​of the multi-source heterogeneous sensors and in combination with the key factor stability state indicator system, the stability state trend of the tailings dam is determined.

[0005] Preferably, the monitoring data of the multi-source heterogeneous sensors include rainfall, reservoir water level, dry beach monitoring data, infiltration line, surface displacement, internal displacement, and video monitoring data.

[0006] Preferably, the monitoring data of the multi-source heterogeneous sensors are preprocessed to obtain multi-source heterogeneous time series data that conforms to a standard normal distribution, including: The monitoring data of multi-source heterogeneous sensors are processed by wavelet method based on 3δ criterion, linear interpolation method and standard deviation normalization method to obtain multi-source heterogeneous time series data that conforms to standard normal distribution.

[0007] Preferably, the multi-source heterogeneous time series data is subjected to feature analysis to obtain key factors for the stability of the tailings dam, including: The changes of multi-source heterogeneous time series data are analyzed from the time and space dimensions, the general laws and characteristics of multi-source heterogeneous time series data in the operation of tailings dams are analyzed, and the key factors of the stable state of the tailings dam are obtained by combining the internal structure and evolution laws of the tailings dam. The key factors include environmental field, seepage field and stress-strain field.

[0008] Preferably, the key factors are determined as nodes of a graph, and a graph representation learning method and prior knowledge are used to construct a heterogeneous hierarchical causal graph containing three different types of hierarchical feature nodes.

[0009] Preferably, the multi-source GCN-AGRU fusion spatiotemporal prediction model includes: Input layer: used to perform convolution operation and vectorization processing on the multi-source heterogeneous time series data to obtain the feature vector of each monitoring point; Spatial aggregation layer: used to extract and learn spatial features from monitoring data of multi-source heterogeneous sensors using graph convolutional neural network modules to obtain graph structures containing structural information; Time series prediction layer: It is used to use the recurrent neural network module to capture the dynamic changes of the feature vectors of each monitoring point and obtain the local time trend of each sensor; it uses the attention mechanism model to focus on important time series information; it performs weighted summation of feature states from a global perspective to make the network focus on key information and capture time dependencies; Output layer: used to output the predicted values ​​of each sensor monitoring point in the graph structure .

[0010] Preferably, the stability state trend of the tailings dam is determined based on the predicted values ​​of the multi-source heterogeneous sensors and in combination with the key factor stability state indicator system, including: Combining the predicted values ​​of multi-source heterogeneous sensors and according to the key factor stability state index system, the dynamic grade range of quantitative indicators is delineated, and the membership function is used to transform the fuzziness of the qualitative problem of the dam stability state into a quantitative expression, and the stability of the tailings dam is obtained. Finally, the development trend of the dam stability state is analyzed by combining the relevant theories of set pair analysis.

[0011] The present invention also provides a tailings dam stability analysis system based on the GCN-AGRU fusion model, which is used to implement the above analysis method of the present invention. The system includes: Data acquisition unit: used to acquire monitoring data from multi-source heterogeneous sensors of the tailings dam; A data preprocessing unit is used to preprocess the monitoring data of the multi-source heterogeneous sensors to obtain multi-source heterogeneous time series data that conforms to the standard normal distribution; Feature analysis unit: used to perform feature analysis on the multi-source heterogeneous time series data to obtain key factors for the stability of the tailings dam; Modeling unit: used to construct a heterogeneous hierarchical causal diagram using the key factors; The first calculation unit is used to calculate the correlation of multi-source heterogeneous time series data according to the heterogeneous hierarchical causal diagram, establish the connection between key factors and the stability state of the tailings dam, and obtain the key factor stability state indicator system; The second computing unit is used to analyze the spatial information of the multi-source heterogeneous time series data using the constructed multi-source GCN-AGRU fusion spatiotemporal prediction model to obtain the spatial information of the tailings dam state; perform time correlation modeling on the multi-source heterogeneous time series data to obtain the time series of the tailings dam state; and calculate the multi-source heterogeneous sensor prediction value based on the spatial information and time series of the multi-source heterogeneous time series data obtained by fusion analysis; Analysis unit: used to determine the stability state trend of the tailings dam according to the predicted values ​​of the multi-source heterogeneous sensors and the key factor stability state indicator system.

[0012] The present invention also provides an electronic device, comprising: one or more processors; a storage device having one or more programs stored thereon; When the one or more programs are executed by the one or more processors, the one or more processors implement the tailings dam stability analysis method based on the GCN-AGRU fusion model as described above in the present invention.

[0013] The present invention also provides a storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the tailings dam stability analysis method based on the GCN-AGRU fusion model as described above is implemented in the present invention.

[0014] The present invention has the following beneficial effects: The GCN-AGUR spatiotemporal prediction model established in the present invention explores the development trend of multi-source monitoring data and avoids the problem that the single-point time domain prediction method of feature screening ignores the correlation between monitoring point data. Accurately and quickly predict monitoring parameters. Spatiotemporal graph data processing and time correlation modeling are performed on multi-source monitoring data to describe sensor data arranged at multiple points, and the graph neural network is used to capture the complex spatial dependencies of the sensor location map. The spatial features extracted from the graph convolutional neural network at different times are fused with the time series and input into the gated recursive unit. The purpose is to use the fused features to capture the spatiotemporal correlation between the time series of different monitoring points. Combined with the time attention mechanism to predict the tailings dam stability monitoring data. Based on real data, the monitoring data prediction problem is converted into a graph node prediction problem. The proposed GCN-AGUR model can simultaneously predict all nodes based on the correlation between nodes and the time series data, and the MSE, RMSE, MAE and R 2 These four indicators are compared and analyzed with other algorithms to verify their accuracy and universality, indicating that the model of the present invention is of great significance to improving the risk prevention and control level of tailings dams. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art are briefly introduced below.

[0016] Figure 1 A distribution diagram of the locations of the monitoring points in the embodiment of the present invention; Figure 2 The monitoring data standardization processing result in the embodiment of the present invention; FIG3 (a) is a heterogeneous causal graph constructed in an embodiment of the present invention; FIG3 (b) is a heterogeneous hierarchical graph constructed in an embodiment of the present invention; Figure 4 It is the GCN-AGRU model prediction framework in the embodiment of the present invention; Figure 5 Schematic diagram of the structure of AGRU in an embodiment of the present invention; Figure 6 Schematic diagram of the structure of GRU in an embodiment of the present invention.

[0017] FIG7 (a) is a comparison diagram of dry beach line 1 prediction results in an embodiment of the present invention; FIG7 (b) is a comparison diagram of dry beach line 2 prediction results in an embodiment; FIG7 (c) is a comparison diagram of external displacement 1 prediction results in an embodiment; FIG7 (d) is a comparison diagram of external displacement 2 prediction results in an embodiment; FIG7 (e) is a comparison diagram of internal displacement 1 prediction results in an embodiment; FIG7 (f) is a comparison diagram of internal displacement 2 prediction results in an embodiment; FIG8( a ) is a comparison diagram of the prediction results of the infiltration line 2 in an embodiment of the present invention; FIG8( b ) is a comparison diagram of the prediction results of the infiltration line 3 in an embodiment of the present invention. DETAILED DESCRIPTION

[0018] To make the purpose, technical solution and advantages of the embodiments of the present invention more clear, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0019] The present invention conducts in-depth information mining on tailings dam monitoring data in order to better understand the laws and characteristics of the stability of the dam during operation, thereby establishing intelligent analysis theories and methods suitable for such data, laying a solid foundation for subsequent research.

[0020] Specifically, the tailings dam stability analysis method based on the GCN-AGRU fusion model of the present invention includes the following steps: Step 1: monitor and process the multi-source sensor data of the tailings dam to obtain multi-source heterogeneous time series data that conforms to the standard normal distribution.

[0021] Among them, according to the unified engineering and technical requirements of the online safety monitoring system of tailings dams in the Engineering and Technical Specifications for Online Safety Monitoring Systems of Tailings Dams GB51108-2015

[115] , the present invention arranges multiple sensors at the tailings dam to perform multi-source data safety monitoring, see Figure 1 The monitoring items include rainfall, reservoir water level, dry beach monitoring, infiltration line, surface displacement, internal displacement, and video monitoring.

[0022] Specifically, the present invention uses a wavelet method based on the 3δ criterion, a linear interpolation method, and a standard deviation normalization method to process noisy and misaligned multi-source heterogeneous sensor monitoring data to obtain multi-source heterogeneous time series data that conforms to the standard normal distribution. Figure 2As shown in the figure, GT represents dry beach line monitoring, GT1 and GT2 are the data monitored by sensors 1 and 2 respectively; KSW represents reservoir water level monitoring data; JYL represents rainfall monitoring data; WBWY represents external displacement, WBW-1-21, WBWY-2-21, and WBWY-3-21 are sensor data of three different monitoring points respectively; NBWY represents internal displacement, NBWY2-21-1, NBWY2-21-2, and NBWY2-21-3 are three groups of internal displacement sensor data respectively; JRX represents infiltration line, JRX-1-21, JRX-2-21, and JRX-2-31 are three groups of infiltration line sensor monitoring data respectively.

[0023] Step 2: Establish the key factors of tailings dam stability and obtain a heterogeneous hierarchical cause-effect diagram to represent the stability state of the tailings dam.

[0024] Specifically, according to the multi-source heterogeneous time series data that conforms to the standard normal distribution, feature analysis is performed on it, data changes are analyzed from the time and space dimensions, the general laws and characteristics between the multi-source heterogeneous time series data in the operation of the tailings dam are analyzed, and the key factors of the stable state of the tailings dam are obtained by combining the internal structure and evolution law of the tailings dam. The key factors include three factors: environmental field, seepage field, and stress-strain field; as shown in Figure 3 (a), the three key factors are determined as nodes of the graph, and a heterogeneous hierarchical causal graph containing three different types of hierarchical feature nodes is constructed by using the graph representation learning method and prior knowledge; as shown in Figure 3 (b), the heterogeneous hierarchical graph of the tailings dam can be regarded as composed of three heterogeneous hierarchical graphs (graphs CP, CF, PF). The heterogeneous hierarchical causal graph is used as the stable state graph of the tailings dam. By integrating factors of different levels and heterogeneity, the complex causal relationship within the system can be fully revealed, indicating the stable state of the tailings dam; Step 3: Based on the obtained heterogeneous hierarchical causal diagram, the dynamic time warping algorithm is used to calculate the correlation of multi-source heterogeneous time series data, establish the relationship between the key factors of environment, seepage field and stress field and the stability state of the tailings dam, and obtain the key factor stability state index system, which is recorded as the environment-seepage-stress stability state index system; Specifically, the ring-seepage-stress stability state index system, as shown in Table 1, divides the state levels of the tailings dam into four types to describe the stability state and dam performance during operation.

[0025] Table 1

[0026] With the help of cloud model theory, a dynamic grade interval calculation method for quantitative indicators is adopted to avoid the subjectivity of grade intervals and ensure the scientific nature of the analysis results.

[0027] Step 4: Build a multi-source GCN-AGRU fusion spatiotemporal prediction model (the multi-source GCN-AGRU fusion spatiotemporal prediction model is a deep learning model that combines graph convolutional networks (GCNs) and attention gated recurrent units (AGRUs) for joint modeling and prediction of spatiotemporal data), as shown in Figure 4 As shown, the GCN module is used to analyze the spatial information of the multi-source heterogeneous time series data to obtain the spatial information of the tailings dam status; the time correlation modeling of the multi-source heterogeneous time series data is performed to obtain the time series of the tailings dam status.

[0028] Step 5: Calculate the predicted value of multi-source heterogeneous sensors and the loss value based on the spatial information and time series of multi-source heterogeneous time series data obtained by fusion analysis, so as to provide data support for the stability situation analysis of the tailings dam; Since multi-source heterogeneous time series data usually changes over time, there is a close coupling relationship between their spatiotemporal characteristics. The present invention obtains spatial information of tailings dam status by analyzing the spatial information of multi-source heterogeneous time series data and time series based on fusion analysis, specifically, using the GCN module; and performing time correlation modeling on the multi-source heterogeneous time series data to obtain the tailings dam status time series.

[0029] like Figure 4 As shown, the overall calculation process of the GCN-AGRU model is divided into four steps: The first step, input layer: convolution operation and vectorization processing are performed on the multi-source heterogeneous time series data to obtain the feature vector of each monitoring point , used to describe the spatial characteristics of the multi-source heterogeneous sensor monitoring data; specifically, the dry beach length, reservoir water level, infiltration line, surface displacement, and internal displacement sensor monitoring data in the tailings dam monitoring system are and predicted length label As input, at each moment, multi-source heterogeneous time series data is transformed into the feature vector through convolution operation and vectorization processing , represents the corresponding values ​​of N sensors at time T; The second step is the spatial aggregation layer: using the graph convolutional neural network (GCN) module to extract and learn the spatial features in the multi-source heterogeneous sensor data to obtain a graph structure containing structural information. , where V is the eigenvector , E indicates that each monitoring point is connected by an edge E, and A is the adjacency matrix, which defines the mutual connections between nodes.

[0030] The third step is the time series prediction layer: Figure 5The following is a schematic diagram of the GRU structure. The GRU is used as the basic time-dependent network structure. In each time step, the attention mechanism is introduced to dynamically assign different weights to the feature states at different times. The recurrent neural network (AGRU) structure in the model is as follows: Figure 6 As shown, the AGRU module is used to capture the feature vector of each monitoring point. The dynamic changes of the sensor are used to obtain the local time trend of each sensor; the attention mechanism model is used to focus on important time series information; the feature state is weighted and summed from a global perspective to make the network focus on key information and capture time dependencies. The specific formula is as follows: (1) (2) (3) (4) (5) (6) (7) (8) (9) Where: h t-1 yes t -1 time output, Respectively represent t The update gate, reset gate and current state at each moment, W , b are the parameters learned during the training process. By updating the gate and resetting the gate to forget and select information, passing through the attention layer, the attention weight of the historical state to the current input is calculated , weighted summation yields the eigenvector s , and the final hidden layer state value is obtained .

[0031] Step 4: Output layer: output the predicted values ​​of each sensor monitoring point in the graph structure , and the input layer , calculate the loss. In the present invention, the goal of the GCN-AGRU prediction model is to minimize the error between the prediction result and the true value, so we add Normalization parameters, using Regularization term, using weight decay method for regularization to prevent overfitting, is a hyperparameter. The formula is as follows: (10) Since the goal of the GCN-AGRU prediction model of the present invention is to minimize the error between the prediction result and the true value, an Lreg normalization parameter is added, such as Figure 4 As shown, the L2 regularization term is used and the weight decay method is used for regularization. is a hyperparameter used to prevent overfitting.

[0032] Step 6: Determine the stability state trend of the tailings dam based on the obtained prediction values ​​of the multi-source heterogeneous sensors and in combination with the ring-seepage-stress stability state indicator system.

[0033] Example In order to verify the effectiveness of the model proposed in this paper, the monitoring data and monitoring point layout of an online safety monitoring system of a tailings pond in Luoyang are used as the verification basis. Based on real data, the monitoring data prediction problem is transformed into a graph node prediction problem. The GCN-AGUR model proposed in this paper can predict all nodes simultaneously according to the correlation between nodes and time series data. The MSE, RMSE, MAE and R 2 These four indicators are compared with other algorithms to verify their accuracy and universality.

[0034] The tailings dam is a "U"-shaped valley-shaped tailings pond with an initial dam crest elevation of 1116.5m. It adopts the tailings upstream method for accumulation. The average accumulation slope below the elevation of 1180m is designed to be 1:3.8, and above the elevation of 1180m is 1:5. The final dam crest elevation is 1375m. The total catchment area of ​​the reservoir is 1.6 square kilometers, the dry beach length is maintained at 250~300m, and the safety overheight is 4.5m.

[0035] Its monitoring items include rainfall, reservoir water level, dry beach monitoring, infiltration line, surface displacement, internal displacement, video monitoring, etc. Since multiple sensors are arranged at 1220m of the tailings dam, the collected data is relatively complete. Taking into account the integrity and comprehensiveness of the experimental data, the monitoring data from December 1, 2019 to April 13, 2021 were selected as the data set, including 3 groups of infiltration lines (JRX-1-21, JRX-2-21, JRX-3-21), 3 groups of surface displacements (WBWY-1, WBWY-2, WBWY-3), 3 groups of internal displacements (NBWY-1, NBWY-2, NBWY-3), 2 groups of dry beach monitoring (GT-1, GT-2), and 1 group of reservoir water level (KS-1), a total of 12 multi-source heterogeneous sensors. The relevant instrument layout is as follows: Figure 1 shown.

[0036] According to the graph data processing method, the monitoring data of the 12 multi-source heterogeneous sensors are modeled according to the correlation modeling method. Each group of data samples is processed by time series and standardized, mainly to fill in missing values ​​and process their continuity, and obtain multi-source heterogeneous time series data that conforms to the standard normal distribution. The processing results are as follows: Figure 2 shown.

[0037] The experimental environment was built using the TensorFlow open source deep learning framework, with the ratio of training set to test set set at 8:2. After multiple experiments and parameter tuning, the training parameters of the model were determined, and the multi-source heterogeneous time series data of the past four days was used to predict the data of the next day.

[0038] This example selects the root mean square error (RMSE), mean absolute error (MAE) and R 2 As the evaluation index of the algorithm, RMSE and MAE are used to analyze the difference between the predicted value and the actual value, and R 2 Analyze the similarity between the predicted value and the actual value, the calculation method is as follows: (11) (12) (13) Among them, m is the total number of monitoring points, x i is the true value of the i-th monitoring point, y i is the predicted value of the ith monitoring point.

[0039] This embodiment compares the prediction results of the GCN-AGUR model with the time series models HA and SVM, and uses a convolutional neural network CNN to extract spatial relationships, and deletes the attention mechanism in the AGRU module to compare the optimization effect of the construction of the graph model and the addition of the attention module on the prediction results. Due to the diversity of model structures, data preprocessing may vary to varying degrees. For example, the HA and SVM models only input each sensor measurement as a feature (or variable) and output a single sensor measurement value, the CNN-GRU model is multi-input and single output, and the GCN-AGRU uses the matrix mapping of multiple sensor measurements as input and outputs multiple sensor prediction values. Therefore, after repeated experiments, the performance of the models was averaged and compared in Table 2.

[0040] Table 2

[0041] As can be seen from Table 2, the traditional algorithm does not contain a multi-layer network, and the calculation time is usually less than 0.1 seconds. Therefore, when comparing the performance, only RMSE, MAE and R 2. The accuracy of the GCN-AGRU model proposed in this paper exceeds that of other models, reaching 99.6%. Although the traditional algorithm also has a good accuracy, its RMSE and MAE values ​​are too large, indicating that the gap between the predicted value and the actual value is large, and the information hidden in the data cannot be effectively learned by the traditional algorithm. The performance of the CNN-GRU and GCN-GRU models is significantly better than that of the traditional algorithms, indicating that the deep network can capture the nonlinear relationship in the sample data and the prediction effect is more accurate. At the same time, the GCN-GRU is also better than the CNN-GRU model, with a prediction accuracy of 0.9924, and the RMSE and MAE values ​​are significantly reduced, proving that the establishment of the graph structure can effectively utilize the implicit information in the data to achieve the effect of improving the prediction accuracy.

[0042] The GCN-AGRU model of the present invention can predict all input features at the same time, which is beneficial for managers to grasp the changing trends of factors affecting the stability of tailings dams in real time. As shown in Figure 7 (a)-Figure 7 (f), the predicted values ​​of the monitoring data from March 2 to April 11, 2021 are compared with the true values. Figure 7 (a) and Figure 7 (b) are comparisons between the predicted values ​​and the true values ​​of two sets of dry beach monitoring data; Figure 7 (c) and Figure 7 (d) are comparisons between the predicted values ​​and the true values ​​of two sets of external displacement monitoring data; Figure 7 (e) and Figure 7 (f) are comparisons between the predicted values ​​and the true values ​​of internal displacement monitoring data. The blue label data represents the changing law of the true value, and the orange GCN-AGRU represents the predicted value of this model. It can be seen from Figure 7 (a)-Figure 7 (f) that the GCN-AGRU model can accurately predict the fluctuation trend when predicting the length of the dry beach and the external displacement. Due to factors such as interference from the external environment, the collected data of the internal displacement sensor may not change. As shown in Figure 7 (e) and Figure 7 (f), the GCN-AGRU model can predict the fluctuation of the internal displacement through the relationship between other factors. As shown in Figure 8 (a) and Figure 8 (b), the predicted values ​​and true values ​​of the infiltration line monitoring data under the CNN-GRU, GCN-GRU and GCN-AGRU models are compared. The GCN-AGRU model with the attention mechanism can adjust the contribution of different time points. By focusing on the key feature information of the node, the model can identify small abnormal fluctuations, thereby capturing the change trend of the data in advance.

[0043] According to the calculation formula of membership, the grade connection degree of each single indicator is calculated, and the comprehensive connection degree matrix of the stable state of the tailings dam is calculated by combining the weights. The generalized set pair potential values ​​of the tailings dam at four levels are calculated and normalized. Finally, the stable state of the tailings dam and the calculation state grade judgment results are shown in Table 3: Table 3

[0044] According to the set pair analysis theory, the connection degree and partial connection coefficient between the stability of the tailings dam and the sensitivity index are calculated. The calculation results of this embodiment show that the stability of the current tailings is 50%~75%, and the stability level is level II. The possibility of dam break is not great, but it is necessary to pay attention to strengthening supervision. The stability level of the tailings dam is shown in Table 1.

[0045] The present invention also provides a system based on the tailings dam stability analysis method based on the GCN-AGRU fusion model as described above, the system comprising: Data acquisition unit: used to acquire monitoring data from multi-source heterogeneous sensors of the tailings dam; A data preprocessing unit is used to preprocess the monitoring data of the multi-source heterogeneous sensors to obtain multi-source heterogeneous time series data that conforms to the standard normal distribution; Feature analysis unit: used to perform feature analysis on the multi-source heterogeneous time series data to obtain key factors for the stability of the tailings dam; Modeling unit: used to construct a heterogeneous hierarchical causal diagram using the key factors; The first calculation unit is used to calculate the correlation of multi-source heterogeneous time series data according to the heterogeneous hierarchical causal diagram, establish the connection between key factors and the stability state of the tailings dam, and obtain the key factor stability state indicator system; The second computing unit is used to analyze the spatial information of the multi-source heterogeneous time series data using the constructed multi-source GCN-AGRU fusion spatiotemporal prediction model to obtain the spatial information of the tailings dam state; perform time correlation modeling on the multi-source heterogeneous time series data to obtain the time series of the tailings dam state; and calculate the multi-source heterogeneous sensor prediction value based on the spatial information and time series of the multi-source heterogeneous time series data obtained by fusion analysis; Analysis unit: used to determine the stability state trend of the tailings dam according to the predicted values ​​of the multi-source heterogeneous sensors and the key factor stability state indicator system.

[0046] The embodiments of the present invention also provide corresponding electronic devices and computer-readable storage media for implementing the solutions provided by the embodiments of the present invention.

[0047] The device includes a memory and a processor, the memory is used to store instructions or codes, and the processor is used to execute the instructions or codes so that the device executes the tailings dam stability analysis method based on the GCN-AGRU fusion model described in any embodiment of the present application.

[0048] The storage medium stores a computer program, wherein the computer program, when executed by the processor, implements the tailings dam stability analysis method based on the GCN-AGRU fusion model described in any embodiment of the present application.

[0049] The present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, readable storage media, optical storage, etc.) containing computer-usable program codes.

[0050] Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0051] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A tailings dam stability analysis method based on the GCN-AGRU fusion model, characterized in that: The process includes the following: Obtain monitoring data from multi-source heterogeneous sensors at tailings dams; Preprocessing the monitoring data of the multi-source heterogeneous sensors to obtain multi-source heterogeneous time series data that conforms to a standard normal distribution; Performing feature analysis on the multi-source heterogeneous time series data to obtain key factors for the stability of the tailings dam; Using the key factors, a heterogeneous hierarchical causal diagram is constructed; According to the heterogeneous hierarchical causal diagram, the correlation of multi-source heterogeneous time series data is calculated, the connection between key factors and the stability state of the tailings dam is established, and the key factor stability state indicator system is obtained; The constructed multi-source GCN-AGRU fusion spatiotemporal prediction model is used to analyze the spatial information of the multi-source heterogeneous time series data to obtain the spatial information of the tailings dam status; Performing time correlation modeling on the multi-source heterogeneous time series data to obtain a time series of the tailings dam status; Calculate the prediction value of multi-source heterogeneous sensors based on the spatial information and time series of multi-source heterogeneous time series data obtained by fusion analysis; According to the predicted values ​​of the multi-source heterogeneous sensors and in combination with the key factor stability state indicator system, the stability state trend of the tailings dam is determined.

2. A tailings dam stability analysis method based on the GCN-AGRU fusion model according to claim 1, characterized in that: The monitoring data of multi-source heterogeneous sensors include rainfall, reservoir water level, dry beach monitoring data, infiltration line, surface displacement, internal displacement, and video monitoring data.

3. The tailings dam stability analysis method based on the GCN-AGRU fusion model according to claim 1 is characterized in that: The monitoring data of the multi-source heterogeneous sensors are preprocessed to obtain multi-source heterogeneous time series data that conforms to the standard normal distribution, including: The monitoring data of multi-source heterogeneous sensors are processed by wavelet method based on 3δ criterion, linear interpolation method and standard deviation normalization method to obtain multi-source heterogeneous time series data that conforms to standard normal distribution.

4. The tailings dam stability analysis method based on the GCN-AGRU fusion model according to claim 1 is characterized in that: By performing feature analysis on the multi-source heterogeneous time series data, key factors for the stability of the tailings dam are obtained, including: The changes of multi-source heterogeneous time series data are analyzed from the time and space dimensions, the general laws and characteristics of multi-source heterogeneous time series data in the operation of tailings dams are analyzed, and the key factors of the stable state of the tailings dam are obtained by combining the internal structure and evolution laws of the tailings dam. The key factors include environmental field, seepage field and stress-strain field.

5. A tailings dam stability analysis method based on the GCN-AGRU fusion model according to claim 4, characterized in that: The key factors are determined as nodes of a graph, and a heterogeneous hierarchical causal graph containing three different types of hierarchical feature nodes is constructed using a graph representation learning method and prior knowledge.

6. The tailings dam stability analysis method based on the GCN-AGRU fusion model according to claim 1 is characterized in that: The multi-source GCN-AGRU fusion spatiotemporal prediction model includes: Input layer: used to perform convolution operation and vectorization processing on the multi-source heterogeneous time series data to obtain the feature vector of each monitoring point; Spatial aggregation layer: used to extract and learn spatial features from monitoring data of multi-source heterogeneous sensors using graph convolutional neural network modules to obtain graph structures containing structural information; Time series prediction layer: used to use the recurrent neural network module to capture the dynamic changes of the feature vectors of each monitoring point and obtain the local time trend of each sensor; Use the attention mechanism model to focus on important time series information; perform weighted summation of feature states from a global perspective to make the network focus on key information and capture time dependencies; Output layer: used to output the predicted values ​​of each sensor monitoring point in the graph structure .

7. The tailings dam stability analysis method based on the GCN-AGRU fusion model according to claim 1 is characterized in that: According to the predicted values ​​of the multi-source heterogeneous sensors and in combination with the key factor stability state indicator system, the stability state trend of the tailings dam is determined, including: Combining the predicted values ​​of multi-source heterogeneous sensors and according to the key factor stability state index system, the dynamic grade range of quantitative indicators is delineated, and the membership function is used to transform the fuzziness of the qualitative problem of the dam stability state into a quantitative expression, and the stability of the tailings dam is obtained. Finally, the development trend of the dam stability state is analyzed by combining the relevant theories of set pair analysis.

8. A tailings dam stability analysis system based on the GCN-AGRU fusion model, characterized in that: include: Data acquisition unit: used to acquire monitoring data from multi-source heterogeneous sensors of the tailings dam; A data preprocessing unit is used to preprocess the monitoring data of the multi-source heterogeneous sensors to obtain multi-source heterogeneous time series data that conforms to the standard normal distribution; Feature analysis unit: used to perform feature analysis on the multi-source heterogeneous time series data to obtain key factors for the stability of the tailings dam; Modeling unit: used to construct a heterogeneous hierarchical causal diagram using the key factors; The first calculation unit is used to calculate the correlation of multi-source heterogeneous time series data according to the heterogeneous hierarchical causal diagram, establish the connection between key factors and the stability state of the tailings dam, and obtain the key factor stability state indicator system; The second computing unit is used to analyze the spatial information of the multi-source heterogeneous time series data using the constructed multi-source GCN-AGRU fusion spatiotemporal prediction model to obtain the spatial information of the tailings dam status; Performing time correlation modeling on the multi-source heterogeneous time series data to obtain a time series of the tailings dam status; Calculate the prediction value of multi-source heterogeneous sensors based on the spatial information and time series of multi-source heterogeneous time series data obtained by fusion analysis; Analysis unit: used to determine the stability state trend of the tailings dam according to the predicted values ​​of the multi-source heterogeneous sensors and in combination with the key factor stability state indicator system.

9. An electronic device, characterized in that: include: one or more processors; a storage device having one or more programs stored thereon; When the one or more programs are executed by the one or more processors, the one or more processors implement the tailings dam stability analysis method based on the GCN-AGRU fusion model as described in any one of claims 1 to 7.

10. A storage medium, characterized in that: A computer program is stored thereon, wherein when the computer program is executed by a processor, the tailings dam stability analysis method based on the GCN-AGRU fusion model as described in any one of claims 1 to 7 is implemented.

Images