Data analysis methods for mine ecological restoration

Abstract

This invention relates to the field of data analysis, and more particularly to a data analysis method for mine ecological restoration. The method includes: acquiring monitoring data from collection points within a mining area, treating these collection points as nodes in a graph network; calculating the spatial influence coefficient representing the strength of physical associations between nodes, and the time validity weight representing data timeliness; constructing a spatiotemporal spectrum aggregation state vector for each node, the construction method including: weighted fusion of the effective observations of a node and the cumulative sum of the effective observations of neighboring nodes, wherein the effective observation value of a node is the product of the node's latest observation value and its corresponding time validity weight, and the effective observation value of a neighboring node is the product of the neighboring node's effective observation value and its corresponding spatial influence coefficient; and predicting the mine's health status based on the spatiotemporal spectrum aggregation state vector. This invention achieves accurate prediction of mine health status.

Description

Technical Field

[0001] This invention relates to the field of data analysis, and in particular to data analysis methods for mine ecological restoration. Background Technology

[0002] Mining activities drastically alter the original geological structure, creating numerous unstable geological units such as steep slopes and spoil heaps. Triggered by factors like rainfall and groundwater activity, these areas are highly susceptible to geological disasters such as landslides and debris flows, which may also be accompanied by the spread of heavy metal pollutants, posing a serious threat to the ecological restoration of mining areas and the safety of personnel. Therefore, establishing a system capable of providing early and accurate warnings of geological and environmental risks is crucial.

[0003] Long Short-Term Memory (LSTM) network models, as a deep learning model adept at capturing long-term dependencies in time series data, have been applied to geological risk prediction. They predict future risk states by learning the temporal evolution of historical monitoring data (such as displacement, rainfall, and pore water pressure). However, existing LSTM-based early warning methods have inherent technical limitations when processing multi-source, heterogeneous monitoring data from mines. Specifically, mine monitoring systems include various types of sensors, and their data exhibit typical characteristics of being multi-source, multi-rate, and spatially heterogeneous. For example, GPS displacement monitoring can reach minute-level sampling frequencies, while rainfall monitoring is typically hourly, and remote sensing data such as vegetation cover may be at the daily or weekly level. Standard LSTM models require a fixed-dimensional feature vector as input at each time step.

[0004] To meet this requirement, existing technologies typically employ resampling or interpolation to forcibly align data of different frequencies onto a unified time axis and flatten sensor readings from different spatial locations into a long vector.

[0005] Interpolation or forward padding during clock synchronization introduces false data, distorting the true instantaneous dynamics of risk factors. Secondly, flattening ignores the inherent physical spatial correlation between monitoring points, such as the relationship between upstream displacement and downstream stress response of a slope. This prevents the model from learning risk evolution patterns based on physical transmission mechanisms and allows only surface statistical fitting, thus significantly reducing the accuracy and reliability of LSTM model predictions. Summary of the Invention

[0006] To address the issue of low accuracy and reliability in LSTM model predictions, this invention provides a data analysis method for mine ecological restoration.

[0007] This invention provides a data analysis method for mine ecological restoration, employing the following technical solution:

[0008] Acquire monitoring data from collection points within the mining area, and treat these collection points as nodes in a graph network;

[0009] Calculate the spatial influence coefficient representing the strength of physical association between nodes, and the time validity weight representing the timeliness of data;

[0010] A spatiotemporal spectrum aggregation state vector for each node is constructed. The construction method includes: weighted fusion of the effective observations of a node and the cumulative sum of the effective observations of its neighboring nodes. The effective observation of a node is the product of the node's latest observation and the corresponding temporal validity weight, and the effective observation of a neighboring node is the product of the neighboring node's effective observation and the corresponding spatial influence coefficient. An attribute graph is constructed based on the spatiotemporal spectrum aggregation state vector, and a risk fingerprint vector is obtained based on the attribute graph to predict the health status of the mine.

[0011] By constructing monitoring points as a graph network and introducing time validity weights and spatial influence coefficients, the fusion of multi-source, heterogeneous, and asynchronous data is achieved, effectively avoiding data distortion. Furthermore, the spatial influence coefficients model the intrinsic correlations between monitoring points based on physical mechanisms such as geological structures and hydrological pathways, enabling the model to learn risk evolution patterns based on real physical transmission paths, rather than simple statistical fitting. Therefore, this invention can fundamentally improve the accuracy and reliability of mine health status prediction.

[0012] Preferably, the time validity weight is calculated as follows: calculate the difference between the analysis time and the most recent data collection timestamp, and use the difference as the exponent of the natural number e to obtain the time validity weight.

[0013] By employing an exponential decay function to quantify the timeliness of historical data, reasonable weights are assigned to data collected at different times, ensuring that the timeliness of information is accurately quantified when constructing state vectors, thereby improving the accuracy of the analysis results.

[0014] The preferred expression for the spatial influence coefficient is:

[0015]

[0016] In the formula, Let be the spatial influence coefficient from node j to node i. Geological and hydrological connectivity factors, d ( i , j Let be the spatial Euclidean distance between nodes i and j. Let exp represent the characteristic influence length of the slope, and let e represent an exponential function.

[0017] Existing technologies only consider Euclidean distance when analyzing spatial relationships. This scheme constructs a spatial influence model by combining geological and hydrological connectivity factors with a distance-based exponential decay function. This model can distinguish between two cases: even if two monitoring points are very close, their influence coefficient will be very low if they are blocked by aquitards or other obstructions; conversely, two distant points located on the same fault zone may have a high influence coefficient, enabling the model to accurately capture risk transmission paths that conform to geological realities.

[0018] Preferably, the method for determining the geological and hydrological connectivity factor is as follows: if node i and node j are located on the same sliding surface, the same dominant joint fracture zone, or the hydraulic conduction path, then Φ(i,j) is 1; otherwise, it is 0.

[0019] Based on whether the monitoring points are located on the same key geological structural unit such as the slip surface, fracture zone, or hydraulic path, their connectivity is directly assigned a value of 1 or 0. This binary approach directly and efficiently integrates the prior knowledge of geological experts into the model, simplifies the complexity of the model, ensures strong constraints on key physical connections, and improves the robustness of spatial influence judgment.

[0020] Preferably, the method for determining the geological-hydrological connectivity factor is as follows: If node i and node j are located on the same slip surface, the same dominant joint fracture zone, or the hydraulic conduction path, then the multi-channel parameter vectors at nodes i and j are obtained. The elements in the multi-channel parameter vectors include: rock dip angle, slope, water content, and rock mass stress. The elements in the multi-channel parameter vectors are normalized using a standard normalization algorithm. The cosine similarity between the multi-channel parameter vectors of node i and node j is calculated. The difference between 1 and the cosine similarity is taken as the geological-hydrological connectivity factor. If node i and node j are blocked by different rock layers or aquitards, the value is 0.

[0021] Preferably, the expression for the spatiotemporal spectrum aggregation state vector is:

[0022]

[0023]

[0024]

[0025] In the formula, This represents the spatiotemporal spectral aggregation state vector of node i at time t. This represents the effective observation value of node j at time t. Let i and j represent the latest observations, respectively. These are the time validity weights for nodes i and j at analysis time t, respectively. Let N(i) be the spatial influence coefficient from node j to node i, and let N(i) be the set of neighboring nodes of node i. This represents the balance coefficient of node i at time t.

[0026] This expression uses a balance coefficient α to weight and fuse the node's own information with the aggregated information from its neighborhood, enabling the model to dynamically adjust its dependence on its own information and neighborhood information according to specific circumstances. This allows for more effective integration of spatiotemporal information and the generation of more expressive node state representations.

[0027] Preferably, the method for calculating the balance coefficient is as follows: calculate the rate of change of the node observations, normalize the rate of change, and use the normalized rate of change to adjust the time validity weight to obtain the self-information importance score.

[0028] Calculate the sum of the products of the spatial influence coefficients and the temporal effectiveness weights of all neighboring nodes on the corresponding nodes, and use the sum as the neighborhood information importance score; calculate the sum of the self-information importance score and the neighborhood information importance score; use the ratio of the self-information importance score to the sum as the balance coefficient.

[0029] The balance coefficients are dynamically adjusted based on the rate of change of a node's own observations (importance of its own information) and the combined influence of neighborhood information (importance of neighborhood information). When a node's own data changes drastically, the model places more emphasis on its own information; when the node's own data is stable but its neighborhood fluctuates, it places more emphasis on neighborhood information. Compared to a fixed weight allocation strategy, the adaptive attention mechanism allows the model to dynamically allocate focus to the most critical information sources, enhancing the model's sensitivity to local anomalies and its adaptability to complex scenarios.

[0030] Preferred method for predicting the health status of a mine is as follows: construct an attribute graph by aggregating state vectors using a spatiotemporal spectrum, encode the state of the attribute graph into a risk fingerprint vector using a graph convolutional network, construct a risk fingerprint sequence using risk fingerprint vectors at multiple time points, and input the risk fingerprint sequence into an LSTM model to obtain the health status of the mine.

[0031] Preferably, the monitoring data includes: surface displacement, deep displacement, pore water pressure, soil moisture content, hourly rainfall, and air temperature.

[0032] By integrating various types of monitoring data, such as surface and deep displacement, pore water pressure, soil moisture content, rainfall, and temperature, this multi-source data fusion strategy can more comprehensively and three-dimensionally reflect the internal and external driving factors and response status of mine slopes compared to analysis methods that rely on only a single data type (such as displacement data alone), thereby establishing a more complete and robust health status assessment system.

[0033] The present invention has the following technical effects:

[0034] By constructing a spatiotemporal graph network model, this method introduces temporal validity weights and spatial influence coefficients, addressing the issues of data distortion and neglect of physical correlations between monitoring points in existing technologies when processing multi-source asynchronous monitoring data. This approach effectively integrates multi-source heterogeneous data and combines graph convolutional networks and long short-term memory networks to deeply learn the evolutionary patterns of risk in the spatiotemporal dimension, thereby achieving more accurate and reliable dynamic predictions of mine health status. Attached Figure Description

[0035] Figure 1 This is a flowchart of the data analysis method for mine ecological restoration proposed in this invention. Detailed Implementation

[0036] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0037] This invention discloses a data analysis method for mine ecological restoration, referring to... Figure 1 This includes the following steps:

[0038] S1: Acquire multi-source heterogeneous data.

[0039] Acquire monitoring data for mine slopes, including: real-time dynamic data collected by sensors deployed at different locations on the slope, such as surface displacement, deep displacement, pore water pressure, and soil moisture content; dynamic data collected by environmental monitoring stations, such as hourly rainfall and air temperature; and low-frequency dynamic data acquired through remote sensing satellites or drones, such as normalized vegetation index and surface temperature. Simultaneously, acquire static data describing the slope's physical properties, such as digital elevation models, soil and rock stratification, and fault zone distribution.

[0040] Each data collection point is defined as node i in a graph network, and the set of all nodes is denoted as V. The observation value of each node i at any time t is denoted as... And record its precise collection timestamp.

[0041] S2: Time validity weight for node construction.

[0042] The validity of sensor readings decays over time until the next new data update. To quantify this decay process, a node time validity weight is constructed. At any analysis time t, for node i, its latest observation data was acquired at time t. Then the expression for its time validity weight is:

[0043]

[0044] in, Let i be the time validity weight of node i at analysis time t. Let be the timestamp of the most recent data collection at node i before time t, and exp represent the exponential function with base e. Let be the information attenuation coefficient of the physical quantity represented by node i. This coefficient is determined according to physical laws. For example, for the rainfall node, its... The value is relatively small because the impact of a single rainfall event on slope stability (such as increased soil saturation) can last for a considerable period; while for displacement monitoring nodes, its The value is relatively large because the instantaneous changes in displacement state are more critical, and historical information has a shorter timeliness. The value is determined based on soil mechanics or hydrogeological models, analyzing the influence of specific physical quantities on the half-life. The acquisition of this value is based on existing technology, and the specific steps will not be elaborated here.

[0045] At analysis time t and the latest data time The larger the time difference, the lower the weight of time validity, indicating that the historical data contributes less to the analysis of the current state.

[0046] S3: Calculate the spatial influence coefficient between nodes.

[0047] In one embodiment, during mine ecological restoration, the mutual influence between any two nodes on the slope is not solely determined by spatial distance, but is also constrained by the geological structure and hydraulic connections between the nodes. Therefore, the influence coefficient between any two nodes is calculated using the following expression:

[0048]

[0049] In the formula, Let be the spatial influence coefficient from node j to node i. This is a geological and hydrological connectivity factor. If node i and node j are located on the same slip surface, the same dominant joint fracture zone, or the same hydraulic conduction path, then Φ(i,j) takes the value of 1; otherwise, if node i and node j are blocked by different rock layers or aquitards, then the value takes the value of 0. d ( i , j ) represents the spatial Euclidean distance between nodes i and j, calculated using the coordinates of the nodes, and exp represents an exponential function with base e. The characteristic influence length of the slope represents the characteristic scale of the physical influence attenuation in space. It can be determined by the geometric dimensions of the slope. For example, the slope height is taken as the characteristic influence length of the slope. Nodes with spatial influence coefficients greater than the preset influence threshold are taken as the neighbor nodes of node i. The influence threshold is set manually according to the actual situation. For example, the influence threshold is 0.6.

[0050] The spatial influence intensity between two nodes is affected by both geological connectivity Φ(i,j) and spatial distance d(i,j). Even if two nodes are very close, their influence coefficient is extremely low if there is no geological connection. Conversely, two distant points located on the same fault may have a high influence coefficient. This allows the model to learn risk evolution patterns based on real physical transmission paths. The calculation method for geological connectivity in this scheme is simple and suitable for scenarios with relatively simple geomorphological structures.

[0051] In another embodiment, the geological-hydrological connectivity factor is calculated as follows: if node i and node j are located on the same slip surface, the same dominant joint fracture zone, or the hydraulic conduction path, then the multi-channel parameter vectors at nodes i and j are obtained. The elements in the multi-channel parameter vectors include: rock dip angle, slope, water content, and rock mass stress. The elements in the multi-channel parameter vectors are normalized using a standard normalization algorithm. The cosine similarity between the multi-channel parameter vectors of node i and node j is calculated. The difference between 1 and the cosine similarity is taken as the geological-hydrological connectivity factor. If nodes i and j are blocked by different rock layers or aquitards, the value is 0.

[0052] When the cosine similarity of the multi-channel parameter vectors between two nodes is small, it indicates that there is a large difference in geological features between the two nodes, the value of the geological and hydrological connectivity factor is large, and the change of node j has a large impact on node i. This calculation method is suitable for scenarios with relatively complex geomorphological structures and can improve the accuracy of the spatial influence coefficient.

[0053] S4: Construct the spatiotemporal spectrum aggregation state vector.

[0054] Each node i generates a spatiotemporal spectrum aggregation state vector at analysis time t. This vector not only contains the latest information of the node itself, but also incorporates the information of its neighboring nodes weighted in time and space dimensions.

[0055]

[0056]

[0057]

[0058] In the formula, This represents the spatiotemporal spectral aggregation state vector of node i at time t. This represents the effective observation value of node j at time t. This represents the latest observation value of node j. Let j be the time validity weight at analysis time t. This represents the latest observation value of node i. Let i be the time validity weight of node i at analysis time t. Let N(i) be the spatial influence coefficient from node j to node i, and let N(i) be the set of neighboring nodes of node i. This represents the balance coefficient of node i at time t, used to adjust the relative importance of the node's own information and the information aggregated from its neighborhood. It should be noted here that... and The calculation method is the same. , and All are vectors that include the current valid value, velocity, and acceleration.

[0059] Current state of each node It reflects its own latest and most effective information, and is also the result of the effective information of its physically related neighbors propagating and converging through spatial influence paths. The newer the data of neighboring node j (…), the more… The larger the value, the stronger the physical association with node i. The larger the value, the greater its contribution to the state of node i.

[0060] In one embodiment, the balance coefficient The calculation method includes the following steps:

[0061] S41: Calculate the local fluctuation factor of node i at time t.

[0062] The expression is:

[0063]

[0064] In the formula, This represents the local fluctuation factor of node i at time t. These represent the latest and second-newest observations of node i, respectively. This indicates the timestamp corresponding to the latest observation. This represents the timestamp corresponding to the newest observation, and tanh represents the hyperbolic tangent function used for normalization. The local volatility factor represents the volatility of the information of node i itself; a value close to 1 indicates that the volatility of the information is more severe and the change is greater.

[0065] S42: Calculate the importance score of node i's own information.

[0066]

[0067] In the formula, This represents the score of the importance of node i's own information at time t. Let i be the time validity weight of node i at analysis time t. This represents the local fluctuation factor of node i at time t.

[0068] When the node is stable The value approaches 0. The value is approximately equal to When the node fluctuates drastically, The value approaches 1. The value is approximately equal to Its importance is thus highlighted.

[0069] S42: Calculate the importance score of the neighborhood information of node i.

[0070]

[0071] In the formula, This represents the importance score of the neighborhood information of node i at time t. Let be the spatial influence coefficient from node j to node i. Let be the time validity weight of node i at analysis time t.

[0072] The fresher the data of a neighboring node and the stronger its physical connection with the central node i, the greater its contribution to the overall importance of the neighborhood information. This indicates the degree to which the information of neighboring nodes contributes to the corresponding node. The larger the value, the greater the contribution of the information of neighboring nodes to the corresponding node, and vice versa.

[0073] S43: Calculate the balance coefficient.

[0074]

[0075] In the formula, This represents the balance coefficient of node i at time t. This represents the score of the importance of node i's own information at time t. This represents the importance score of the neighborhood information of node i at time t.

[0076] By calculating the balance coefficient, the model can dynamically allocate attention to key information sources when facing different situations, thereby enhancing the model's expressive power and adaptability to complex scenarios.

[0077] S5: Use spatiotemporal spectrum to aggregate state vectors to construct an attribute graph, and obtain risk fingerprint vectors based on the attribute graph.

[0078] At time t, the spatiotemporal spectrum aggregation state vectors of all nodes together constitute an attribute graph. A graph convolutional network is used to encode the state of the attribute graph into a fixed-dimensional vector, namely the risk fingerprint vector.

[0079] Graph convolutional networks iteratively update the state representation of each node through a multi-layered information transfer mechanism, ensuring that the final representation of each node contains the structural information of the entire graph. After L layers of graph convolution, processing such as mean pooling is performed to obtain the risk fingerprint at time t. The discrete, multi-dimensional set of node states is condensed into a unique and dense risk fingerprint vector that can macroscopically characterize the cooperative state of the entire slope system at a specific moment.

[0080] S6: Predict the health status of mines using LSTM models.

[0081] Construct a risk fingerprint sequence E using risk fingerprints generated continuously over time:

[0082] The risk fingerprint sequence E is used as the input to the LSTM model to obtain the health status of the mine, which includes: good, average, and poor.

[0083] By replacing the input of the LSTM model from a flattened raw data vector to a risk fingerprint sequence containing rich spatiotemporal spectrum information, the LSTM model can learn in a higher-dimensional feature space that is more in line with physical laws, thereby achieving earlier and more accurate dynamic early warning of geological risks and monitoring of the mine ecological environment.

[0084] The above are all preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape and principle of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A data analysis method for mine ecological restoration, characterized in that, Including the following steps: Acquire monitoring data from collection points within the mining area, and treat these collection points as nodes in a graph network; Calculate the spatial influence coefficient representing the strength of physical association between nodes, and the time validity weight representing the timeliness of data; A spatiotemporal spectrum aggregation state vector for each node is constructed. The construction method includes: weighted fusion of the effective observations of a node and the cumulative sum of the effective observations of its neighboring nodes. The effective observation of a node is the product of the node's latest observation and its corresponding temporal validity weight, and the effective observation of a neighboring node is the product of its effective observation and its corresponding spatial influence coefficient. An attribute graph is constructed based on the spatiotemporal spectrum aggregation state vector, and a risk fingerprint vector is obtained based on the attribute graph to predict the health status of the mine. The calculation method for the time validity weight is as follows: at any analysis time t, for node i, its latest monitoring data was collected at time t. Then the expression for its time validity weight is: ； in, Let i be the time validity weight of node i at analysis time t. Let be the timestamp of the most recent data collection at node i before time t, and exp represent the exponential function with base e. The information attenuation coefficient of the physical quantity represented by node i; The expression for the spatial influence coefficient is: ； In the formula, Let be the spatial influence coefficient from node j to node i. Geological and hydrological connectivity factors, d ( i , j Let be the spatial Euclidean distance between nodes i and j. Let exp represent the characteristic influence length of the slope, and let exp denote an exponential function with base e.

2. The data analysis method for mine ecological restoration according to claim 1, characterized in that, The method for determining the geological and hydrological connectivity factor is as follows: if node i and node j are located on the same sliding surface, the same dominant joint fracture zone, or the hydraulic conduction path, then Φ(i,j) is 1; otherwise, it is 0.

3. The data analysis method for mine ecological restoration according to claim 1, characterized in that, The method for determining the geological-hydrological connectivity factor is as follows: If node i and node j are located on the same slip surface, the same dominant joint fracture zone, or the hydraulic conduction path, then the multi-channel parameter vectors at nodes i and j are obtained. The elements in the multi-channel parameter vectors include: rock dip angle, slope, water content, and rock mass stress. The elements in the multi-channel parameter vectors are normalized using a standard normalization algorithm. The cosine similarity between the multi-channel parameter vectors of nodes i and j is calculated. The difference between 1 and the cosine similarity is taken as the geological-hydrological connectivity factor. If nodes i and j are blocked by different rock layers or aquitards, the value is 0.

4. The data analysis method for mine ecological restoration according to claim 1, characterized in that, The expression for the spatiotemporal spectrum aggregated state vector is: ； ； ； In the formula, This represents the spatiotemporal spectral aggregation state vector of node i at time t. This represents the effective observation value of node j at time t. Let i and j represent the latest observations, respectively. These are the time validity weights for nodes i and j at analysis time t, respectively. Let N(i) be the spatial influence coefficient from node j to node i, and let N(i) be the set of neighboring nodes of node i. This represents the balance coefficient of node i at time t.

5. The data analysis method for mine ecological restoration according to claim 4, characterized in that, The balance coefficient is calculated as follows: calculate the rate of change of the nodal observations, normalize the rate of change, and use the normalized rate of change to adjust the time validity weight to obtain the self-information importance score. Calculate the cumulative sum of the products of the spatial influence coefficients of all neighboring nodes on the corresponding nodes and the temporal effectiveness weights, and use the cumulative sum as the neighborhood information importance score; Calculate the sum of the importance score of self-information and the importance score of neighborhood information; use the ratio of the importance score of self-information to the sum as a balancing coefficient.

6. The data analysis method for mine ecological restoration according to claim 1, characterized in that, The method for predicting the health status of a mine is as follows: an attribute graph is constructed by aggregating state vectors using a spatiotemporal spectrum; the state of the attribute graph is encoded into a risk fingerprint vector using a graph convolutional network; a risk fingerprint sequence is constructed using risk fingerprint vectors at multiple time points; and the risk fingerprint sequence is input into an LSTM model to obtain the health status of the mine.

7. The data analysis method for mine ecological restoration according to claim 1, characterized in that, The monitoring data includes: surface displacement, deep displacement, pore water pressure, soil moisture content, hourly rainfall, and air temperature.

Images