A mine surface subsidence early warning method based on space-time correlation information

Abstract

The present application relates to a kind of mine surface subsidence early warning method based on space-time correlation information, belong to surface subsidence early warning technical field.The present application collects the surface deformation monitoring data in the deformation area of mine surface subsidence and its periphery;Through clustering method, the deformation data of each monitoring point of mine surface is clustered, and the surface deformation area showing different deformation characteristics is divided;Through space-time correlation rule mining method, the space-time correlation information between the surface deformation monitoring data of adjacent deformation area is mined;Based on the obtained space-time correlation information, determine the early warning level and early warning level critical standard, combined with the advanced prediction of surface deformation, implement advanced dynamic early warning to mine surface subsidence.The present application method can realize the intelligent early warning of surface deformation and subsidence, and provide scientific basis for formulating targeted surface subsidence prevention and control measures.

Description

Technical Field

[0001] This invention relates to a method for early warning of mine surface subsidence based on spatiotemporal correlation information, belonging to the field of surface subsidence early warning technology. Background Technology

[0002] Surface subsidence refers to a geological phenomenon in which surface rock and soil collapse downwards under natural or human influence, forming sinkholes (caves) on the ground. As underground mining extends deeper, the problem of surface subsidence becomes increasingly prominent. Surface subsidence often brings harm to mining projects, including casualties, economic losses, and project delays. Scientific and effective early warning systems for surface subsidence are crucial for avoiding or mitigating these hazards.

[0003] Currently, commonly used surface subsidence early warning methods can be mainly divided into two categories: one is risk early warning based on methods such as fuzzy mathematics; the other is threshold-based monitoring and early warning. For risk early warning based on methods such as fuzzy mathematics, fuzzy comprehensive evaluation methods are often used for risk assessment and then to achieve risk early warning. However, this type of method has certain limitations. Not only is the evaluation process somewhat subjective, but it is also difficult to achieve dynamic evaluation of disasters. Generally, it can only give a qualitative conclusion on whether a mine threatened by surface subsidence disaster is stable at a certain state. As for threshold-based monitoring and early warning methods, on the one hand, surface subsidence caused by underground excavation in mines is very complex, and various threshold methods are difficult to provide accurate and unified effective early warnings for various types of surface subsidence disasters. On the other hand, for mines, monitoring methods are relatively simple, and the monitoring cycle is short, making it impossible to achieve early warning and forecasting based on multiple parameters. This leads to problems such as low early warning reliability, high error rate, and untimely early warning due to the short monitoring cycle and limited data collection at the same monitoring location. Summary of the Invention

[0004] To address the frequent occurrence of surface subsidence during underground mining operations and to reduce the severity of such disasters through proactive dynamic early warning, this invention proposes a mine surface subsidence early warning method based on spatiotemporal correlation information. This method involves collecting surface deformation monitoring data from the mine surface subsidence area and its surrounding deformation zones; clustering the deformation data from various monitoring points to identify surface deformation zones exhibiting different deformation characteristics; mining the spatiotemporal correlation information between surface deformation monitoring data from adjacent deformation zones using spatiotemporal correlation rule mining methods; analyzing the risk evolution status of each zone based on this information; determining early warning indicators and critical level standards; and enabling proactive dynamic early warning for each deformation zone through real-time updates of monitoring data and proactive prediction of surface deformation. This invention provides intelligent early warning for surface deformation and subsidence, offering a scientific basis for developing targeted surface subsidence prevention and control measures.

[0005] A method for early warning of mine surface subsidence based on spatiotemporal correlation information, the specific steps of which are as follows:

[0006] S1. Monitor surface deformation during underground mining to obtain time-series data of surface deformation during the formation of surface subsidence;

[0007] S2. Dynamic time curvature is used as the distance metric for time series in the k-means clustering algorithm to cluster the displacement time series data of all monitoring points in the surface deformation area and divide the deformation area;

[0008] S3. Using a combination of association rules and grey relational analysis, spatiotemporal association rules are mined from the deformation time series data between adjacent deformation areas to obtain spatiotemporal association information;

[0009] S4. Based on the obtained spatiotemporal correlation information, determine the early warning level and the critical standard for the early warning level, and combine it with the advanced prediction of surface deformation to implement advanced dynamic early warning for mine surface subsidence.

[0010] S2. The method for clustering the displacement time series data of all monitoring points within the surface deformation area includes the following steps:

[0011] S201. Randomly select k data points from the deformation time series data of n measurement points as the initial cluster centers;

[0012] S202. Based on the mean of the measurement point data contained in each cluster, i.e. the cluster center, calculate the dynamic time bending distance between the deformation time series data of each measurement point and the corresponding cluster center, and re-cluster the measurement point data according to the minimum dynamic time bending distance;

[0013] S203. Calculate the mean of the clusters after each update to obtain the updated cluster centers;

[0014] S204. Repeat steps S202 and S203 until each cluster no longer changes, to obtain the time series clusters of measurement point deformation. All measurement points in the same cluster show similar deformation characteristics, while the deformation characteristics of measurement points in different clusters are significantly different.

[0015] The method for dividing the deformation region is as follows:

[0016] The surface deformation regions are divided according to the spatial location of the measurement points of each cluster. Between two adjacent regions, the region closer to the deformation center develops ahead and is called the advanced development region; the region relatively far from the deformation center develops behind and is called the lagging development region. The deformation regions are divided according to the displacement-time curve.

[0017] S3. The method for mining spatiotemporal association rules for the deformation time series data between adjacent deformation regions is as follows:

[0018] S301. Input the deformation time series data of the measurement points in the lagging development area and the advanced development area. A fixed window is generated on the time series of the measurement points in the lagging development area. The time series data of the measurement points in the advanced development area are traversed by the sliding window. The gray correlation degree between the fixed window and the sliding window is calculated. The correlation window is determined according to the set gray correlation degree threshold.

[0019] S302. The time series that may have spatiotemporal correlation are determined by the correlation window. That is, the time series formed by the earliest monitoring time of the monitoring point in the advanced development area to the end time of the correlation window, and the time series formed by the monitoring point in the adjacent lagging development area starting from the end time of the correlation window and extending in the direction of the earliest monitoring time, may have spatiotemporal correlation.

[0020] S303. Perform time series expansion operations on the measurement points and their time series that may have spatiotemporal correlations, and calculate the gray correlation degree between the formed time windows. By comparing with the gray correlation degree threshold, obtain the similarity state of the deformation trend in each time window; calculate according to the set support and confidence, and compare with the minimum support and minimum confidence. Measurement points that meet the threshold are retained, and measurement points that do not meet the threshold are removed; obtain the measurement points and their corresponding time series that have spatiotemporal correlations between adjacent advanced development regions and lagging development regions.

[0021] Specifically, spatiotemporal correlation rule mining includes the following:

[0022] (1) Fixed window

[0023] In the deformation time series of measuring points in the lagging development area, the time window consisting of the time series with a time interval of length t from the latest monitoring time to the earliest monitoring time is a fixed window.

[0024] (2) Sliding window

[0025] In the deformation time series of the monitoring points in the advanced development area, the time window consisting of the time series with a time interval of length t from the earliest monitoring time to the latest monitoring time is the first sliding window. After that, the time window slides for a time interval of Δt each time to form a new sliding window until the latest monitoring time.

[0026] (3) Related windows

[0027] The time series data contained in the fixed window serves as the reference sequence for calculating the grey relational degree when determining the associated window; the time series data contained in all sliding windows serve as the comparison sequences for calculating the grey relational degree when determining the associated window. Among the grey relational degrees calculated between the reference sequence of the fixed window and the comparison sequences of all sliding windows, the sliding window closest to the latest monitoring time that meets the set grey relational degree threshold forms the associated window with the fixed window.

[0028] (4) Extension of time series

[0029] The time series expansion is based on a time series consisting of the earliest monitoring time to the end time of the association window for monitoring points in the advanced development area, and a time series of the same length extending from the end time of the association window to the earliest monitoring time for monitoring points in the lagging development area. Starting from the earliest monitoring time of these time series, time series with a time interval of length t form the first group of time windows, and a grey relational degree is calculated once. Then, based on the first group of time windows, time series with a time interval of length Δt are successively expanded. After each expansion, a grey relational degree is calculated once within the newly formed time windows, until the end time of the association window.

[0030] (5) Support

[0031] In the gray relational degree calculated between time windows formed during the expansion of the time series, the ratio of the number of time windows that meet the gray relational degree threshold, i.e. the number of time windows with similar deformation trends between measurement points, to the total number of time windows, is the support.

[0032] (6) Confidence

[0033] In the gray relational degree calculated between time windows formed during the expansion of the time series of the measurement points that meet the minimum support, the ratio of the gray relational degree between time windows after the time series expansion to the gray relational degree between time windows before the time series expansion is the confidence level.

[0034] (7) Spatiotemporal correlation rules In this article, X = [E] i ,T i ], Y = [H j ,T j ]. H j Point j and E represent measurement points in lagging development areas. i This represents measurement point i in the advanced development region, characterizing spatial attributes; T j and T i The time interval is represented by T, which is the total length of the time series in the time series extension. Since the lengths of the time series extensions at measurement points i and j are equal, but their starting times are different, T represents the time interval. j and Ti Distinguish them and present time attributes; spatiotemporal association rules It should be noted that: the measuring point i in the advanced region is at T i The deformation trend that appears within the time period, and the lagging area measurement point j in time period T j Similar deformation trends will appear within the region, thus reflecting the deformation evolution patterns between adjacent areas.

[0035] The method for calculating the grey relational degree is as follows:

[0036] Let the reference sequence and the comparison sequence be:

[0037] x0(k)={x0(k)|k=1,2,3,...,n} (1)

[0038] x i (k)={x i (k)|k=1,2,3,...,n; i=1,2,3,...,m} (2)

[0039] For ρ∈(0,1), let:

[0040]

[0041]

[0042] Where ρ is the resolution coefficient; γ(x0,x i ) is x i Grey relational degree relative to x0.

[0043] The method for early warning of surface subsidence in mines, as described in S4, is as follows:

[0044] S401. The cumulative displacement and deformation rate of surface deformation, as well as the number of monitoring points with spatiotemporal correlation between adjacent deformation areas and the associated deformation stages, are used as early warning indicators for the risk level of each deformation area.

[0045] S402. Based on the obtained spatiotemporal correlation information, the evolution state between adjacent deformation regions is obtained, and the support vector machine predicts the future deformation of the correlated measuring points, so as to implement early warning for mine surface collapse.

[0046] S403. When all the displacement time series of measuring points in the lagging development area are correlated with the displacement time series of measuring points in the advanced development area, and the deformation stage of the correlated advanced development area measuring points has reached the latter half of the displacement data of the advanced development area, and the predicted cumulative deformation and deformation rate of the lagging area have exceeded the critical value of the original warning level, then the lagging development area has the deformation characteristics of the advanced development area, evolves into an advanced development area, and the warning level is changed to the warning level of the advanced development area.

[0047] The method for predicting the future deformation of associated measurement points using support vector machines is as follows:

[0048] Displacement time series with spatiotemporal correlation between advanced and lagging development regions are used as training samples, and subsequent data of the associated time series of advanced development regions are used as input for test samples to predict the future deformation displacement of lagging development regions. The prediction algorithm uses radial basis functions as kernel functions, and grid search method is used to determine the penalty coefficient C and parameter g of support vector regression.

[0049] The beneficial effects of this invention are:

[0050] (1) This invention explores the spatiotemporal correlation rules between surface deformation areas with similar geological environmental conditions, which can fully reflect the impact of all risk factors. Based on the spatiotemporal correlation rules, the critical value of the early warning index for the lagging development area with spatiotemporal correlation with the early warning index critical value of the development-advanced deformation area is determined from the critical value of the early warning index critical value of the development-advanced deformation area. This can solve the problem that in the complex mining construction environment, the impact and correlation of risk factors are constantly changing, and traditional methods are difficult to determine the evaluation index and critical value.

[0051] (2) The spatiotemporal correlation information mined from different deformation regions in this invention can reveal the spatiotemporal evolution law of multiple stages of surface deformation, and thus determine the critical value of each stage of deformation evolution in the process of surface deformation, thereby determining the different evolution states of risk.

[0052] (3) The early warning method based on spatiotemporal correlation information of the present invention not only focuses on the occurrence of disasters, but also enables dynamic early warning of the entire risk evolution process from surface deformation to the occurrence of disasters, including real-time dynamic early warning of the impact of surface deformation on structures and safe construction before the occurrence of collapse disasters. Attached Figure Description

[0053] Figure 1 Flowchart for early warning of surface subsidence in mines;

[0054] Figure 2 The cumulative displacement-time curve is a "broken line" type.

[0055] Figure 3 The cumulative displacement-time curve is "S-shaped".

[0056] Figure 4 The cumulative displacement-time curve is linear.

[0057] Figure 5 The cumulative displacement-time curve is of the "fluctuation type".

[0058] Figure 6 This diagram illustrates fixed windows, sliding windows, and associated windows.

[0059] Figure 7 This is an extended graph of the time series. Detailed Implementation

[0060] The present invention will be further described in detail below with reference to specific embodiments, but the scope of protection of the present invention is not limited to the content described.

[0061] Example 1: A method for early warning of mine surface subsidence based on spatiotemporal correlation information (see Example 2) Figure 1 The specific steps are as follows:

[0062] S1. Monitor surface deformation during underground mining to obtain time-series data of surface deformation during the formation of surface subsidence;

[0063] S2. Dynamic time curvature is used as the distance metric for time series in the k-means clustering algorithm to cluster the displacement time series data of all monitoring points in the surface deformation area and divide the deformation area;

[0064] Similarity measurement and clustering analysis are both important research areas in time series data mining, affecting the quality of data mining. Euclidean distance is a simple and commonly used similarity measurement method with linear time complexity; however, it is very sensitive to outliers in the time series and requires the time series to be of equal length. Therefore, the traditional Euclidean distance method is not an accurate similarity measurement method for time series. Dynamic time bending differs from traditional Euclidean distance in that it matches data points in the time series by bending the time domain, which not only provides better morphological measurement results but also measures the similarity between two time series of unequal length. Therefore, in order to obtain more accurate time series clustering results by better measuring the similarity of time series, dynamic time bending is adopted to replace the traditional Euclidean distance method as the distance (similarity) measurement method for time series in the k-means clustering algorithm, thereby improving the clustering effect of the k-means method on time series.

[0065] A method for clustering displacement time series data of all monitoring points within the surface deformation area:

[0066] Input: number of clusters k, and a database containing deformation time series data of n measurement points;

[0067] Output: k clusters that meet the minimum variance criterion;

[0068] The specific steps are as follows:

[0069] S201. Randomly select k data points from the deformation time series data of n measurement points as the initial cluster centers;

[0070] S202. Based on the mean of the measurement point data contained in each cluster, i.e. the cluster center, calculate the dynamic time bending distance between the deformation time series data of each measurement point and the corresponding cluster center, and re-cluster the measurement point data according to the minimum dynamic time bending distance;

[0071] S203. Calculate the mean of the clusters after each update to obtain the updated cluster centers;

[0072] S204. Repeat steps S202 and S203 until each cluster no longer changes, to obtain the time series clusters of measurement point deformation. All measurement points in the same cluster show similar deformation characteristics, while the deformation characteristics of measurement points in different clusters are significantly different.

[0073] The method for dividing the deformation region is as follows:

[0074] The surface deformation is divided according to the location of the measuring points in each cluster. That is, different surface deformation areas are determined by the spatial location of the measuring points in different clusters. Between two adjacent areas, the area closer to the deformation center develops faster and is called the advanced development area. The area relatively far away from the deformation center develops slower and is called the lagging development area. Different surface deformation areas have different deformation characteristics, which can represent different stages or states of surface deformation. This lays the foundation for the subsequent mining of spatiotemporal association rules of surface deformation data between adjacent areas and the early warning of risk evolution caused by the transformation of deformation stages or states.

[0075] S3. Using a combination of association rules and grey relational analysis, spatiotemporal association rules are mined from the deformation time series data between adjacent deformation areas to obtain spatiotemporal association information;

[0076] Mining spatiotemporal correlation rules for surface deformation in mines aims to obtain the spatiotemporal correlation rules between deformation data of certain monitoring points in advanced development areas over certain time periods and the most recently monitored deformation data of certain monitoring points in adjacent lagging development areas over certain time periods, for different regions of surface deformation in the same mine case. To achieve the mining of spatiotemporal correlation rules between surface deformation data in mines, grey relational analysis and correlation rule methods are combined. Grey relational analysis can measure the similarity of development trends between surface deformation of various monitoring points over time, while correlation rule methods can reveal the spatiotemporal evolution law of this similar deformation trend between adjacent regions.

[0077] The approach to mining the spatiotemporal correlation rules between adjacent areas of surface deformation in mines is as follows:

[0078] (1) Determine the measurement points and their time series that may have spatiotemporal correlation: Based on the surface deformation areas divided by time series clustering, determine the measurement points in adjacent deformation areas; determine the measurement points that may have spatiotemporal correlation between adjacent advanced development areas and lagging development intervals; filter by the gray correlation degree calculated between the fixed window on the time series of measurement points in the lagging development area and all sliding windows formed when traversing the time series of measurement points in the advanced development area, determine the correlation window, and determine the time series that may have correlation based on the correlation window;

[0079] (2) Calculate the similarity between deformation time series of measuring points that may have spatiotemporal correlation: On the deformation time series of measuring points that may have spatiotemporal correlation between adjacent areas, the time series is extended one by one on the basis of the time series contained in the first time window until the correlation window is reached. The gray correlation degree is calculated between multiple time windows formed during this period to measure the similarity of the surface deformation development trend of measuring points in adjacent areas that may have spatiotemporal correlation at each time interval.

[0080] (3) Discovering spatiotemporal correlation rules: Based on the deformation time series of measuring points that may have spatiotemporal correlations between adjacent areas, the time series is extended to obtain the gray correlation degree between each time window. The gray correlation degree threshold is set to determine the deformation similarity state between each time window. The support and confidence of the correlation rules are set, and the minimum support and minimum confidence thresholds that the rules must meet are specified. The time series of measuring points that meet the minimum support and minimum confidence requirements for the deformation development trend between adjacent areas are determined to obtain the spatiotemporal correlation rules.

[0081] The grey relational analysis method determines the relationships between elements in a system by measuring the similarity or dissimilarity of their development trends over time. It determines the closeness of multiple factors in the system by comparing the geometric relationships between a reference sequence and a comparison sequence. The closer the geometric shapes of the sequence curves, the more similar the development trends of the elements, and the greater the grey relational degree. Conversely, if the curves are significantly different, the development trends of the elements differ significantly, and the grey relational degree is smaller. The calculation method for grey relational degree is as follows:

[0082] Let the reference sequence and the comparison sequence be:

[0083] x0(k)={x0(k)|k=1,2,3,...,n} (1)

[0084] x i (k)={x i (k)|k=1,2,3,...,n; i=1,2,3,...,m} (2)

[0085] For ρ∈(0,1), let:

[0086]

[0087]

[0088] Where ρ is the resolution coefficient; γ(x0,x i ) is x i The degree of gray relational relationship relative to x0;

[0089] The fixed window: In the deformation time series of measuring points in the lagging development area, the time series consisting of time intervals of length t from the latest monitoring time to the earliest monitoring time is called the fixed window (see...). Figure 6 );

[0090] The sliding window: In the deformation time series of the monitoring points in the advanced development area, the first sliding window is a time series with a time interval of length t, starting from the earliest monitoring time and moving towards the latest monitoring time. This time window then slides for a time interval Δt each time to form a new sliding window, until the latest monitoring time (see...). Figure 6 );

[0091] The associated window consists of: a fixed window containing time-series data, which serves as a reference sequence for calculating gray correlation when determining the associated window; and all sliding windows containing time-series data, which serve as comparison sequences for calculating gray correlation when determining the associated window. Among the gray correlations calculated between the reference sequence of the fixed window and the comparison sequences of all sliding windows, the sliding window closest to the latest monitoring time that satisfies the set gray correlation threshold forms the associated window with the fixed window (see...). Figure 6 );

[0092] The time series expansion involves using a time series consisting of the earliest monitoring time to the end time of the associated window for monitoring points in the advanced development area, and a time series of the same length extending from the end time of the associated window to the earliest monitoring time for monitoring points in the lagging development area. Starting from the earliest monitoring time of these time series, time series with a time interval of length t form the first set of time windows, and a grey relational degree is calculated once. Then, based on the first set of time windows, time series with a time interval of length Δt are successively expanded. After each expansion, a grey relational degree is calculated once within the newly formed time windows, until the end time of the associated window (see...). Figure 7 );

[0093] The support is the ratio of the number of time windows that meet the gray relational degree threshold (i.e., the number of time windows with similar deformation trends between measurement points) to the total number of time windows in the gray relational degree calculation formed during the expansion of the time series.

[0094] The confidence level is the ratio of the grey relational degree between time windows formed during the expansion of the time series of the measurement points that meet the minimum support to the grey relational degree between time windows before the expansion of the time series, which is calculated in the process of expanding the time series.

[0095] The spatiotemporal correlation rules Let X = [E] i ,T i ], Y = [H j ,T j ], H j Point j and E represent measurement points in lagging development areas. i This represents measurement point i in the advanced development region, characterizing spatial attributes; T j and T i The time interval is represented by T, which is the total length of the time series in the time series extension. Since the lengths of the time series extensions at measurement points i and j are equal, but their starting times are different, the extension length for measurement point i is T. i Therefore, the measuring point j is T. j Presenting time attributes; spatiotemporal association rules Indicates: the leading measurement point i in the T region i The deformation trend that appears within the time period, and the lagging area measurement point j in time period T j Similar deformation trends will appear within it, reflecting the deformation evolution patterns between adjacent regions;

[0096] Therefore, the specific steps for mining spatiotemporal correlation rules between adjacent areas in the same mining case are as follows:

[0097] S301. Input the deformation time series data of the measurement points in the lagging development area and the advanced development area. A fixed window is generated on the time series of the measurement points in the lagging development area. The time series data of the measurement points in the advanced development area are traversed by the sliding window. The gray correlation degree between the fixed window and the sliding window is calculated. The correlation window is determined according to the set gray correlation degree threshold.

[0098] S302. The time series that may have spatiotemporal correlations are determined by the correlation window. That is, the time series consisting of the earliest monitoring time of the monitoring point in the advanced development area to the end time of the correlation window and the time series consisting of the same length from the end time of the correlation window of the monitoring point in the adjacent lagging development area to the earliest monitoring time may have spatiotemporal correlations.

[0099] S303. Perform time series expansion operations on the measurement points and their time series that may have spatiotemporal correlations, and calculate the gray correlation degree between the formed time windows. By comparing with the gray correlation degree threshold, obtain the similarity state of the deformation trend in each time window; calculate according to the set support and confidence, and compare with the minimum support and minimum confidence. Measurement points that meet the threshold are retained, and measurement points that do not meet the threshold are removed; obtain the measurement points and their corresponding time series that have spatiotemporal correlations between adjacent advanced development regions and lagging development regions.

[0100] S4. A method for early warning of mine surface subsidence based on spatiotemporal correlation information, the specific steps of which are as follows:

[0101] S401. The cumulative displacement and deformation rate of surface deformation, as well as the number of monitoring points with spatiotemporal correlation between adjacent deformation areas and the associated deformation stages, are used as early warning indicators for the risk level of each deformation area.

[0102] S402. Based on the obtained spatiotemporal correlation information, the evolution state between adjacent deformation areas is obtained, and the future deformation of the associated measuring points is predicted by support vector machine, so as to implement advanced dynamic early warning for mine surface collapse;

[0103] S403. When all the displacement time series of measuring points in the lagging development area are correlated with the displacement time series of measuring points in the advanced development area, and the deformation stage of the correlated advanced development area measuring points has reached the latter half of the displacement data of the advanced development area, and the predicted cumulative deformation and deformation rate of the lagging area have exceeded the critical value of the original warning level, then the lagging development area has the deformation characteristics of the advanced development area, evolves into an advanced development area, and the warning level is changed to the warning level of the advanced development area.

[0104] The method for predicting the future deformation of associated measurement points using support vector machines is as follows:

[0105] Displacement time series with spatiotemporal correlation between advanced and lagging development regions are used as training samples, and subsequent data of the associated time series of advanced development regions are used as input for test samples to predict the future deformation displacement of lagging development regions. The prediction algorithm uses radial basis functions as kernel functions, and grid search method is used to determine the penalty coefficient C and parameter g of support vector regression.

[0106] Example 2: Using the mine surface subsidence early warning method based on spatiotemporal correlation information from Example 1, an early warning of surface subsidence during the excavation of an underground engineering project in a mine is provided. The specific steps are as follows:

[0107] S1. Monitor surface deformation during underground mining to obtain time-series data of surface deformation during the formation of surface subsidence;

[0108] S2. Dynamic time curvature is used as the distance metric for time series in the k-means clustering algorithm to cluster the displacement time series data of all monitoring points in the surface deformation area and divide the deformation area;

[0109] The method for clustering displacement time series data of all monitoring points within the surface deformation area.

[0110] Input: number of clusters k, and a database containing deformation time series data of n measurement points;

[0111] Output: k clusters that meet the minimum variance criterion;

[0112] The specific steps are as follows:

[0113] S201. Randomly select k data points from the deformation time series data of n measurement points as the initial cluster centers;

[0114] S202. Based on the mean of the measurement point data contained in each cluster, i.e. the cluster center, calculate the dynamic time bending distance between the deformation time series data of each measurement point and the corresponding cluster center, and re-cluster the measurement point data according to the minimum dynamic time bending distance;

[0115] S203. Calculate the mean of the clusters after each update to obtain the updated cluster centers;

[0116] S204. Repeat steps S202 and S203 until each cluster no longer changes, to obtain the time series clusters of measurement point deformation. All measurement points in the same cluster show similar deformation characteristics, while the deformation characteristics of measurement points in different clusters are significantly different.

[0117] The method for dividing the deformation region is as follows:

[0118] The surface deformation is divided according to the location of the measuring points in each cluster. That is, different surface deformation areas are determined by the spatial location of the measuring points in different clusters. Between two adjacent areas, the area closer to the deformation center develops faster and is called the advanced development area. The area relatively far away from the deformation center develops slower and is called the lagging development area. Different surface deformation areas have different deformation characteristics, which can represent different stages or states of surface deformation. This lays the foundation for the subsequent mining of spatiotemporal association rules of surface deformation data between adjacent areas and the early warning of risk evolution caused by the transformation of deformation stages or states.

[0119] In this embodiment, the deformation time series data of GPS monitoring points within a certain range of the surface deformation center of a mine are clustered to obtain the following results: Figure 2-5 The four types of displacement-time curves shown, and the four surface deformation zones divided according to them:

[0120] "Broken Line" Cumulative Displacement-Time Curve (see) Figure 2 The cumulative displacement shows a deformation trend of initially gradual then steepening over time. The deformation area is close to the subsidence pit and located at the deformation center, exhibiting a "zigzag" deformation pattern. This area has the largest cumulative deformation; as of October 2012, the minimum cumulative displacement value at these three monitoring points reached 2593 mm, and the maximum reached 6116 mm, corresponding to a maximum deformation rate of 7.8 mm / d and a minimum of 3.5 mm / d. The surface cracks in this area are relatively large, with the widest crack reaching 800 mm.

[0121] "S-shaped" cumulative displacement-time curve (see...) Figure 3 The deformation trend of cumulative displacement over time is characterized by an initial gradual increase, followed by a steeper increase, and then a slow increase again. The area where the monitoring point of the deformation curve is located is separated from the deformation center by a "broken line" deformation region, meaning that this deformation region is adjacent to the "broken line" deformation region and belongs to the "S-shaped" deformation region. Although the cumulative deformation in this region is not as large as that in the "broken line" deformation region, it is still relatively large, with the largest cumulative displacement value being 1045 mm and the smallest being 124 mm. Correspondingly, the deformation rate during the slow increase process in the latter part of the deformation curve is basically in the range of 0.2-0.9 mm / d. The scale of the surface cracks in this region is still relatively large, but much smaller than that in the "broken line" deformation region, with a width of basically between 5.0-30 mm.

[0122] "Linear" cumulative displacement-time curve (see) Figure 4 The cumulative displacement shows a linear trend over time. The deformation curve monitoring point is far from the deformation center on the ground surface and is close to the "S-shaped" deformation area, belonging to the "linear" deformation area. The cumulative deformation in this area is relatively small. Until October 2012, the maximum cumulative displacement value was 528 mm and the minimum value was 82 mm. The corresponding deformation rate is about 0.1-0.3 mm / d. No cracks or a small number of closed cracks appeared on the ground surface in this area.

[0123] "Fluctuating" cumulative displacement-time curve (see...) Figure 5 The cumulative displacement fluctuates significantly over time. This deformation area is adjacent to the "linear" deformation area and is located at the outermost edge of the deformation area, belonging to the "fluctuating" deformation area. The cumulative deformation in this area is very small. The maximum cumulative displacement value in October 2012 was only 84 mm, and the deformation rate was basically less than 0.1 mm / d. No cracks appeared on the surface of this area.

[0124] S3. Using a combination of association rules and grey relational analysis, spatiotemporal association rules are mined from the deformation time series data between adjacent deformation areas to obtain spatiotemporal association information;

[0125] Mining spatiotemporal correlation rules for surface deformation in mines aims to obtain the spatiotemporal correlation rules between deformation data of certain monitoring points in advanced development areas over certain time periods and the most recently monitored deformation data of certain monitoring points in adjacent lagging development areas over certain time periods, for different regions of surface deformation in the same mine case. To achieve the mining of spatiotemporal correlation rules between surface deformation data in mines, grey relational analysis and correlation rule methods are combined. Grey relational analysis can measure the similarity of development trends between surface deformation of various monitoring points over time, while correlation rule methods can reveal the spatiotemporal evolution law of this similar deformation trend between adjacent regions.

[0126] The approach to mining the spatiotemporal correlation rules between adjacent areas of surface deformation in mines is as follows:

[0127] (1) Determine the measurement points and their time series that may have spatiotemporal correlation: Based on the surface deformation areas divided by time series clustering, determine the measurement points in adjacent deformation areas; determine the measurement points that may have spatiotemporal correlation between adjacent advanced development areas and lagging development intervals; filter by the gray correlation degree calculated between the fixed window on the time series of measurement points in the lagging development area and all sliding windows formed when traversing the time series of measurement points in the advanced development area, determine the correlation window, and determine the time series that may have correlation based on the correlation window;

[0128] (2) Calculate the similarity between deformation time series of measuring points that may have spatiotemporal correlation: On the deformation time series of measuring points that may have spatiotemporal correlation between adjacent areas, the time series is extended one by one on the basis of the time series contained in the first time window until the correlation window is reached. The gray correlation degree is calculated between multiple time windows formed during this period to measure the similarity of the surface deformation development trend of measuring points in adjacent areas that may have spatiotemporal correlation at each time interval.

[0129] (3) Discovering spatiotemporal correlation rules: Based on the deformation time series of measuring points that may have spatiotemporal correlations between adjacent areas, the time series is extended to obtain the gray correlation degree between each time window. The gray correlation degree threshold is set to determine the deformation similarity state between each time window. The support and confidence of the correlation rules are set, and the minimum support and minimum confidence thresholds that the rules must meet are specified. The time series of measuring points that meet the minimum support and minimum confidence requirements for the deformation development trend between adjacent areas are determined to obtain the spatiotemporal correlation rules.

[0130] The grey relational analysis method determines the relationships between elements in a system by measuring the similarity or dissimilarity of their development trends over time. It determines the closeness of multiple factors in the system by comparing the geometric relationships between a reference sequence and a comparison sequence. The closer the geometric shapes of the sequence curves, the more similar the development trends of the elements, and the greater the grey relational degree. Conversely, if the curves are significantly different, the development trends of the elements differ significantly, and the grey relational degree is smaller. The calculation method for grey relational degree is as follows:

[0131] Let the reference sequence and the comparison sequence be:

[0132] x0(k)={x0(k)|k=1,2,3,...,n} (1)

[0133] x i (k)={x i (k)|k=1,2,3,...,n; i=1,2,3,...,m} (2)

[0134] For ρ∈(0,1), let:

[0135]

[0136]

[0137] Where ρ is the resolution coefficient; γ(x0,x i ) is x i The degree of gray relational relationship relative to x0;

[0138] The fixed window: In the deformation time series of measuring points in the lagging development area, the time series consisting of time intervals of length t from the latest monitoring time to the earliest monitoring time is called the fixed window (see...). Figure 6 );

[0139] The sliding window: In the deformation time series of the monitoring points in the advanced development area, the first sliding window is a time series with a time interval of length t, starting from the earliest monitoring time and moving towards the latest monitoring time. This time window then slides for a time interval Δt each time to form a new sliding window, until the latest monitoring time (see...). Figure 6 );

[0140] The associated window consists of: a fixed window containing time-series data, which serves as a reference sequence for calculating gray correlation when determining the associated window; and all sliding windows containing time-series data, which serve as comparison sequences for calculating gray correlation when determining the associated window. Among the gray correlations calculated between the reference sequence of the fixed window and the comparison sequences of all sliding windows, the sliding window closest to the latest monitoring time that satisfies the set gray correlation threshold forms the associated window with the fixed window (see...). Figure 6 );

[0141] The time series expansion involves using a time series consisting of the earliest monitoring time to the end time of the associated window for monitoring points in the advanced development area, and a time series of the same length extending from the end time of the associated window to the earliest monitoring time for monitoring points in the lagging development area. Starting from the earliest monitoring time of these time series, time series with a time interval of length t form the first set of time windows, and a grey relational degree is calculated once. Then, based on the first set of time windows, time series with a time interval of length Δt are successively expanded. After each expansion, a grey relational degree is calculated once within the newly formed time windows, until the end time of the associated window (see...). Figure 7 );

[0142] The support is the ratio of the number of time windows that meet the gray relational degree threshold (i.e., the number of time windows with similar deformation trends between measurement points) to the total number of time windows in the gray relational degree calculation formed during the expansion of the time series.

[0143] The confidence level is the ratio of the grey relational degree between time windows formed during the expansion of the time series of the measurement points that meet the minimum support to the grey relational degree between time windows before the expansion of the time series, which is calculated in the process of expanding the time series.

[0144] The spatiotemporal correlation rules Let X = [E] i ,T i ], Y = [H j ,T j ], H j Point j and E represent measurement points in lagging development areas. i This represents measurement point i in the advanced development region, characterizing spatial attributes; T j and T i The time interval is represented by T, which is the total length of the time series in the time series extension. Since the lengths of the time series extensions at measurement points i and j are equal, but their starting times are different, the extension length for measurement point i is T. i Therefore, the measuring point j is T. j Presenting time attributes; spatiotemporal association rules Indicates: the leading measurement point i in the T region i The deformation trend that appears within the time period, and the lagging area measurement point j in time period T j Similar deformation trends will appear within it, reflecting the deformation evolution patterns between adjacent regions;

[0145] Therefore, the specific steps for mining spatiotemporal correlation rules between adjacent areas in the same mining case are as follows:

[0146] S301. Input the deformation time series data of the measurement points in the lagging development area and the advanced development area. A fixed window is generated on the time series of the measurement points in the lagging development area. The time series data of the measurement points in the advanced development area are traversed by the sliding window. The gray correlation degree between the fixed window and the sliding window is calculated. The correlation window is determined according to the set gray correlation degree threshold.

[0147] S302. The time series that may have spatiotemporal correlations are determined by the correlation window. That is, the time series consisting of the earliest monitoring time of the monitoring point in the advanced development area to the end time of the correlation window and the time series consisting of the same length from the end time of the correlation window of the monitoring point in the adjacent lagging development area to the earliest monitoring time may have spatiotemporal correlations.

[0148] S303. Perform time series expansion operations on the measurement points and their time series that may have spatiotemporal correlations, and calculate the gray correlation degree between the formed time windows. By comparing with the gray correlation degree threshold, obtain the similarity state of the deformation trend in each time window; calculate according to the set support and confidence, and compare with the minimum support and minimum confidence. Measurement points that meet the threshold are retained, and measurement points that do not meet the threshold are removed; obtain the measurement points and their corresponding time series that have spatiotemporal correlations between adjacent advanced development regions and lagging development regions.

[0149] In this embodiment, the method of combining association rules with grey relational analysis is used to mine spatiotemporal association rules for the deformation time series data between adjacent deformation regions, and the spatiotemporal association information is shown in Table 1 below.

[0150] Table 1 shows the measuring points and their time series between adjacent deformation regions that exhibit spatiotemporal correlation.

[0151]

[0152] Depend on Figure 2-5As shown in Table 1, in the adjacent "linear" deformation region and "S-shaped" deformation region, except for the E11 measuring point which is relatively far from the "S-shaped" deformation region, the other seven measuring points in the "linear" deformation region all showed deformation time series with spatiotemporal correlation in the "S-shaped" deformation region. Furthermore, these time series were correlated with the middle segment of the displacement time series of the monitoring points in the "S-shaped" deformation region, revealing the evolution of the deformation pattern (state) in adjacent regions. The "linear" deformation region exhibits some deformation characteristics of the "S-shaped" deformation region. This is sufficient to demonstrate that the "linear" deformation region has evolved into the "S-shaped" deformation region, and it is highly likely that it will continue to develop according to the deformation trend of the "S-shaped" deformation region. For adjacent "S-shaped" deformation regions and "broken line" deformation regions, in the "S-shaped" deformation region with seven measuring points, there is a correlation between the displacement time series of three monitoring points and the displacement time series of the measuring points in the "broken line" deformation region. All three are correlated to the initial stage of the displacement time series of the "broken line" deformation region, indicating that the "S-shaped" deformation region has just begun to evolve into the "broken line" deformation region. Half of the measuring points do not yet show the deformation characteristics of the "broken line" deformation region and continue to develop according to the original deformation trend.

[0153] S4. Based on the obtained spatiotemporal correlation information, determine the early warning level and the critical standard for the early warning level, and implement dynamic early warning for mine surface subsidence;

[0154] Based on the spatial location of surface deformation zones according to the characteristics of deformation curves: "Bokeh-shaped" deformation zones are located at the deformation center, close to existing subsidence pits. As the underground excavation area and depth continue to expand, "Bokeh-shaped" deformation zones are highly likely to collapse, corresponding to the highest warning level, Level IV; "S-shaped" deformation zones are adjacent to the deformation center, with the risk of collapse second only to "Bokeh-shaped" deformation zones, resulting in a lower warning level, Level III; "Straight-line" deformation zones are adjacent to "S-shaped" deformation zones but farther from the deformation center, therefore their warning level is lower than that of "S-shaped" deformation zones, Level II; "Wave-shaped" deformation zones are located at the outermost edge of the deformation zone, making them the safest, thus receiving the lowest warning level, Level I.

[0155] The time dimension of the entire life cycle from the beginning of surface deformation in a mine to the occurrence of collapse: Since displacement time series data is a direct representation of the evolution process of surface deformation and collapse caused by mine excavation, different regions divided according to different types of displacement-time curves can also represent different evolution stages from surface deformation to collapse. That is, affected by mine excavation, the surface generally goes through several stages from the beginning of deformation to the occurrence of collapse, from "fluctuation" deformation, "linear" deformation, "S" deformation to "zigzag" deformation and finally collapse. Among them, the "zigzag" deformation stage is closest to the collapse stage, and the warning level of this stage is naturally the highest, with the other stages in order.

[0156] The different deformation characteristics presented by the cumulative displacement-time curves of various types: The cumulative displacement, deformation rate, and surface crack width of the "broken line" deformation area are relatively the highest, and the surface deformation is the most severe. If the accelerated deformation trend shown by this type of curve continues, it is very likely to collapse, so the warning level is the highest; The cumulative displacement, deformation rate, and surface crack width of the "S-shaped" deformation area are lower than those of the "broken line" deformation area, but the deformation is more obvious than that of the "straight line" and "wave" deformation areas. Therefore, the warning level is second only to the "broken line" deformation area. The warning levels of the "straight line" and "wave" deformation areas are similarly lower in that order.

[0157] Therefore, the warning level and the critical criteria for the warning level (see Table 2) are as follows:

[0158] The "wave-type" deformation zone is adjacent to the "straight-type" deformation zone and is located at the outermost edge of the deformation zone. The cumulative displacement S is less than 100 mm, the deformation rate v is less than 0.1 mm / d, and no cracks have appeared. The warning level is I.

[0159] In the "straight-line" deformation zone, which is adjacent to the "S-shaped" deformation zone, the cumulative displacement S is 100mm≤S≤600mm, the deformation rate v is 0.1mm / d≤v≤0.3mm / d, and a closed crack or crack width L<5.0mm appears, the warning level is II.

[0160] The “S-shaped” deformation zone is adjacent to the “zigzag” deformation zone. The cumulative displacement S is 600mm < S ≤ 2600mm, the deformation rate v is 0.3mm / d < v ≤ 3.5mm / d, and the crack width L is 5.0mm ≤ L ≤ 30mm. The warning level is Level III.

[0161] The "zigzag" deformation zone is located at the deformation center, with a cumulative displacement S greater than 2600 mm, a deformation rate v greater than 3.5 mm / d, and a crack width L of 30 mm < L ≤ 800 mm, indicating a warning level of IV.

[0162] Table 2 Warning Levels and Classification Standards for Surface Deformation Areas in Various Regions

[0163]

[0164] The aforementioned warning levels I, II, III, and IV are as follows: Level I indicates a very low risk of surface subsidence, with minimal impact from surface deformation caused by mining on infrastructure, requiring no immediate action; Level II indicates a potential for surface subsidence, with potential cracks in infrastructure due to mining excavation, requiring vigilance; Level III indicates a high probability of surface subsidence, with infrastructure likely to experience uneven deformation and collapse, requiring preventative measures; and Level IV indicates a very high likelihood of surface subsidence, with all infrastructure potentially collapsing, requiring immediate preventative measures.

[0165] The warning levels for surface deformation and eventual subsidence caused by mine excavation were classified, and critical level standards for cumulative displacement and deformation rate evaluation indicators that reflect the degree of surface deformation were determined. However, as the scope and depth of underground excavation in the mine continue to extend, the deformation areas gradually develop towards the deformation center, and the corresponding risk status also evolves dynamically: the "zigzag" deformation area is currently at the highest warning level, Level IV. This area is located at the deformation center, close to the subsidence pit, and is the most advanced in development, and is highly likely to collapse in the future; the "S-shaped" deformation area is currently at the warning level III. It is spatially adjacent but lags slightly behind the "zigzag" deformation area in time, and is very likely to evolve into a "zigzag" deformation area in the future, and its warning level will also evolve to the higher Level IV; similarly, the "straight-line" deformation area at Level II warning may develop into an "S-shaped" deformation area over time, and the warning level will advance to Level III; at the same time, the Level II warning for the "straight-line" deformation area is also the future development direction of the "fluctuating" deformation area at Level I warning.

[0166] In order to dynamically reveal the evolution of risk status, the evolution between adjacent deformation areas is taken into account. The spatiotemporal correlation information mined from the displacement time series of adjacent deformation areas is used to construct a surface collapse early warning method. That is, in addition to using the cumulative displacement and deformation rate, which reflect the degree of surface deformation, as early warning indicators, the number of spatiotemporal correlation measurement points and the spatiotemporal correlation stage between adjacent deformation areas are added as indicators reflecting the degree of evolution of adjacent deformation areas. Thus, the early warning indicators of the mine surface collapse early warning method based on spatiotemporal correlation information are formed.

[0167] The method for implementing early warning of mine surface subsidence based on spatiotemporal correlation information includes the following specific steps:

[0168] S401. The cumulative displacement and deformation rate of surface deformation, as well as the number of monitoring points with spatiotemporal correlation between adjacent deformation areas and the associated deformation stages, are used as early warning indicators for the risk evolution of each deformation area.

[0169] S402. Based on the obtained spatiotemporal correlation information, the evolution states between the "linear" deformation region and the "S-shaped" deformation region, and between the "S-shaped" deformation region and the "polyline" deformation region are obtained. Support vector machines are then used to predict the future deformation displacement of the correlated measuring points. The method for predicting the future deformation displacement of the correlated measuring points using support vector machines is as follows:

[0170] Displacement time series with spatiotemporal correlation between advanced and lagging development regions are used as training samples, and subsequent data of the time series associated with the advanced development region are used as input for test samples to predict the future deformation displacement of the lagging development region. The prediction algorithm uses radial basis function as kernel function, and grid search method is used to determine the penalty coefficient C and parameter g of support vector regression.

[0171] S403. When all the displacement time series of the measuring points in the lagging development area are correlated with the displacement time series of the measuring points in the advanced development area, and the deformation stage of the correlated advanced development area measuring points has reached the latter half of the displacement data of the advanced development area, and the predicted cumulative deformation and deformation rate of the lagging area have exceeded the critical value of the original warning level, then the lagging development area has the deformation characteristics of the advanced development area, evolves into an advanced development area, and the warning level is changed to the warning level of the advanced development area.

[0172] In this embodiment, after analyzing the spatiotemporal association rule mining results between adjacent deformation regions as described above, and obtaining the evolution state between the "straight line" deformation region and the "S-shaped" deformation region, and between the "S-shaped" deformation region and the "broken line" deformation region, and then combining the displacement prediction results of the associated measurement points using support vector machines, the early warning index values ​​shown in Table 3 below can be obtained.

[0173] Table 3. Index values ​​in the spatiotemporal association rule mining and prediction results.

[0174]

[0175] From the critical level standards of the cumulative displacement and deformation rate early warning indicators in Table 2, and the early warning indicator values ​​in the spatiotemporal correlation rule mining and prediction results of the displacement data of the monitoring points in the "straight-line" deformation area and the "S-shaped" deformation area in Table 3, it can be concluded that:

[0176] (1) The results of spatiotemporal correlation rule mining of displacement data of monitoring points in the "straight-line" deformation area and the "S-shaped" deformation area show that there are eight monitoring points in the "straight-line" deformation area. Although seven monitoring points have spatiotemporally correlated time series data in the "S-shaped" deformation area, they are only correlated to the middle segment of the displacement time series of the monitoring points in the "S-shaped" deformation area. In other words, the deformation development trend of the monitoring points in the "straight-line" deformation area is related to the deformation development trend of the middle segment of the displacement time series of the monitoring points in the "S-shaped" deformation area, and has not yet developed to the final segment of the displacement time series of the monitoring points in the "S-shaped" deformation area. That is, the "straight-line" deformation area does not yet fully possess the deformation characteristics of the "S-shaped" deformation area, and it will still take some time before it fully evolves into the "S-shaped" deformation area. Furthermore, the displacement prediction results of the correlated monitoring points in the "straight-line" deformation area show that the predicted cumulative displacement and deformation rate are still within the range of change of the "straight-line" deformation area given in Table 2. Therefore, it can be concluded that the "straight-line" deformation area still belongs to its original warning level and has not undergone an advanced evolution of the warning level.

[0177] (2) The results of spatiotemporal correlation rule mining of the displacement data of monitoring points in the "S-shaped" deformation area and the "broken-line" deformation area show that the displacement data of three of the seven monitoring points in the "S-shaped" deformation area are spatiotemporally correlated with the displacement data of the monitoring points in the "broken-line" deformation area. Furthermore, the correlated time periods are all the initial segments of the time series of the displacement data of the monitoring points in the "broken-line" deformation area. It will still take some time before all monitoring points in the "S-shaped" deformation area fully exhibit the deformation characteristics of the "broken-line" deformation area. The displacement prediction results of the correlated monitoring points in the "S-shaped" deformation area are still within the range of cumulative displacement and deformation rate changes given in Table 2. Currently, the "S-shaped" deformation area is still far from the advanced evolution of the warning level. Therefore, it is considered that the "S-shaped" deformation area remains at its original warning level.

[0178] Therefore, based on the analysis of the spatiotemporal correlation rule mining and prediction results of displacement data of adjacent deformation areas in the embodiment, and by adopting the established mine surface subsidence early warning method based on spatiotemporal correlation information, it can be seen that although the surface deformation near the subsidence pit in the embodiment still maintains an increasing trend, each deformation area still belongs to its original early warning level, and there has been no progressive evolution of the early warning level. The mine can maintain attention and vigilance on the basis of existing risk avoidance measures.

[0179] The specific embodiments of the present invention have been described in detail above. However, the present invention is not limited to the above embodiments. Within the scope of knowledge possessed by those skilled in the art, various changes can be made without departing from the spirit of the present invention.

Claims

1. A method for early warning of mine surface subsidence based on spatiotemporal correlation information, characterized in that, The specific steps are as follows: S1. Monitor surface deformation during underground mining operations to obtain time-series data on surface deformation during the formation of surface subsidence. S2. Using dynamic time bending as... k - The distance metric method for time series in the -means clustering algorithm is used to cluster the displacement time series data of all monitoring points in the surface deformation area and divide the deformation area; S3. Using a method combining association rules and grey relational analysis, spatiotemporal association rule mining is performed on the deformation time series data between adjacent deformation regions to obtain spatiotemporal association information; the specific method for performing spatiotemporal association rule mining on the deformation time series data between adjacent deformation regions is as follows: S301. Input the deformation time series data of the measurement points in the lagging development area and the advanced development area. A fixed window is generated on the time series of the measurement points in the lagging development area. The time series data of the measurement points in the advanced development area are traversed by the sliding window. The gray correlation degree between the fixed window and the sliding window is calculated. The correlation window is determined according to the set gray correlation degree threshold. S302. The time series that may have spatiotemporal correlations are determined by the correlation window. That is, the time series formed by the earliest monitoring time of the monitoring point in the advanced development area to the end time of the correlation window, and the time series formed by the monitoring point in the adjacent lagging development area starting from the end time of the correlation window and extending in the direction of the earliest monitoring time, may have spatiotemporal correlations. S303. Perform time series expansion operations on the measurement points and their time series that may have spatiotemporal correlations, and calculate the grey correlation degree between the formed time windows. By comparing with the grey correlation degree threshold, obtain the similarity state of the deformation trend in each time window; calculate according to the set support and confidence, and compare with the minimum support and minimum confidence. Measurement points that meet the threshold are retained, and measurement points that do not meet the threshold are removed; obtain the measurement points and their corresponding time series that have spatiotemporal correlations between adjacent advanced development regions and lagging development regions. The method for calculating the grey relational degree is as follows: Let the reference sequence and the comparison sequence be: (1) (2) for ,make: (3) (4) in, ρ The resolution coefficient; for Compared to Grey relational degree; S4. Based on the obtained spatiotemporal correlation information, determine the early warning level and the critical standard for the early warning level, and combine it with the advanced prediction of surface deformation to implement advanced dynamic early warning for mine surface subsidence; the specific steps of the method for implementing early warning for mine surface subsidence are as follows: S401. The cumulative displacement and deformation rate of surface deformation, as well as the number of monitoring points with spatiotemporal correlation between adjacent deformation areas and the associated deformation stages, are used as early warning indicators for the risk level of each deformation area. S402. Based on the evolution state between adjacent deformation regions obtained from the spatiotemporal correlation information, and the prediction of future deformation of the correlated measuring points by the support vector machine, an early warning is implemented for mine surface collapse. S403. When all the displacement time series of measuring points in the lagging development area are correlated with the displacement time series of measuring points in the advanced development area, and the deformation stage of the correlated advanced development area measuring points has reached the latter half of the displacement data of the advanced development area, and the predicted cumulative deformation and deformation rate of the lagging area have exceeded the critical value of the original warning level, then the lagging development area has the deformation characteristics of the advanced development area, evolves into an advanced development area, and the warning level is changed to the warning level of the advanced development area.

2. The mine surface subsidence early warning method based on spatiotemporal correlation information according to claim 1, characterized in that: Step S2. The method for clustering the displacement time series data of all monitoring points within the surface deformation area is as follows: S201. From n Randomly select from the deformation time series data of each measuring point k The data from each measurement point are used as the initial cluster centers; S202. Based on the mean of the measurement point data contained in each cluster, i.e. the cluster center, calculate the dynamic time bending distance between the deformation time series data of each measurement point and the corresponding cluster center, and re-cluster the measurement point data according to the minimum dynamic time bending distance; S203. Calculate the mean of the clusters after each update to obtain the updated cluster centers; S204. Repeat steps S202 and S203 until each cluster no longer changes, to obtain the time series clusters of measurement point deformation. All measurement points in the same cluster show similar deformation characteristics, while the deformation characteristics of measurement points in different clusters are significantly different.

3. The mine surface subsidence early warning method based on spatiotemporal correlation information according to claim 2, characterized in that: The method for dividing the deformation region is as follows: The surface deformation areas are divided according to the spatial location of the measurement points of each cluster. Between two adjacent areas, the area closer to the deformation center develops more advancedly and is called the advanced development area. Areas relatively far from the deformation center are experiencing lagging development and are thus classified as lagging development areas.

4. The mine surface subsidence early warning method based on spatiotemporal correlation information according to claim 1, characterized in that: The method of using support vector machines to predict the future deformation of associated measurement points is as follows: Displacement time series with spatiotemporal correlation between advanced and lagging development regions are used as training samples, and subsequent data from the correlated time series of the advanced development region are used as input for test samples to predict future deformation of the lagging development region. The prediction algorithm uses a radial basis function as the kernel function, and a grid search method is used to determine the penalty coefficient for support vector regression. C and parameters g .

Images