A landslide disaster precision early warning method based on multi-element monitoring data fusion

Abstract

The application provides a landslide disaster precision early warning method based on multi-element monitoring data fusion, and belongs to the technical field of landslide disaster early warning. In order to solve the technical problems of low early warning accuracy and weak anti-interference ability caused by single data of the existing slope early warning method, the technical scheme is that: the target monitoring data is determined, which is deep displacement data, surface displacement data and rockfall occurrence; physical monitoring equipment is laid out to collect the deep displacement data, surface displacement data and rockfall occurrence of the open pit slope; the grading early warning criterion is established, according to the response sequence of different data, the yellow early warning criterion based on the deep displacement data, the orange early warning criterion based on the surface displacement data and the red early warning criterion based on the rockfall monitoring data are respectively constructed; the landslide disaster precision early warning method based on multi-element monitoring data fusion is constructed to realize the open pit landslide precision grading early warning; the application is applied to the early warning of the open pit slope landslide disaster.

Description

Technical Field

[0001] This invention provides a precise early warning method for landslide disasters based on the fusion of multi-source monitoring data, belonging to the field of landslide disaster early warning technology. Background Technology

[0002] Landslides caused by slope instability in open-pit mines are characterized by their suddenness and destructive power, seriously threatening normal mining operations and the safety of personnel and equipment. Furthermore, the stability of the slope's rock and soil is significantly affected by multiple factors, including mining disturbances, rainfall infiltration, and blasting vibrations. As mining progresses deeper, slope height and angle increase, exacerbating the risk of disasters. Therefore, there is an urgent need to construct an all-weather, high-precision monitoring and early warning system to shift from "post-disaster response" to "pre-disaster prevention." Against this backdrop, to ensure safe production and efficient resource development, slope monitoring and early warning solutions tailored to complex operating conditions must be adopted in relevant open-pit mining areas.

[0003] In recent years, significant progress has been made in slope monitoring technology for open-pit mines. Real-time displacement monitoring systems based on BeiDou / GNSS have been widely used in mining slopes, enabling millimeter-level capture of three-dimensional rock mass displacement. Fiber optic sensing technology, through the installation of distributed sensors, has enabled continuous monitoring of internal stress, temperature, and crack development on slopes. The combination of UAV aerial photography and three-dimensional laser scanning technology has enabled the rapid generation of high-precision point cloud models of slope surfaces, identifying crack propagation and topographic changes.

[0004] However, existing monitoring systems still have significant limitations: on the one hand, single data sources are susceptible to environmental interference (such as sensors being blocked by vegetation or optical images being affected by clouds and fog), leading to insufficient reliability of monitoring results; on the other hand, isolated analysis of multi-source data makes it difficult to fully explore the complementary value of different monitoring methods, resulting in problems such as weak generalization ability of early warning models and high false alarm and false negative rates. Fundamentally, existing monitoring technologies lack a deep fusion mechanism for multi-scale and multi-modal monitoring data, which restricts the improvement of early warning accuracy.

[0005] Therefore, developing a precise early warning method for landslide disasters based on the fusion of multi-source monitoring data is of great significance for solving problems such as inconsistent data interfaces of monitoring and early warning systems, limited types of monitoring and early warning data, weak integration capabilities of monitoring and early warning systems, and low early warning accuracy. It can also help reduce personnel and equipment investment costs and improve the economic benefits of mines. Summary of the Invention

[0006] To address the technical problems existing in the background art, the present invention adopts the following technical solution: providing a precise landslide disaster early warning method based on multi-source monitoring data fusion, comprising the following landslide disaster early warning steps:

[0007] Step 1: Determine the target monitoring data, including: deep displacement data, surface displacement data, and data on the occurrence of rockfalls;

[0008] Step 2: Deploy physical monitoring modules for collecting deep displacement data, surface displacement data, and rockfall occurrence data of open-pit mine slopes. The physical monitoring modules include multiple monitoring devices with different functions.

[0009] Step 3: Based on the response order of different data, construct corresponding hierarchical early warning criteria, including: yellow early warning criteria based on deep displacement data, orange early warning criteria based on surface displacement data, and red early warning criteria based on rockfall occurrence data;

[0010] Step 4: Based on the data obtained by the monitoring equipment and different early warning criteria, conduct precise early warning of landslide disasters based on the fusion of multi-source monitoring data, determine whether the corresponding data currently monitored needs to be issued as an early warning, and determine the early warning level according to different situations.

[0011] The monitoring equipment used in step 2 is set up in corresponding locations for different monitoring objects and areas, including:

[0012] GNSS receivers and radar receivers used for monitoring slope surface displacement data;

[0013] Inclinometers, strain gauges, and fixed inclinometers used to monitor deep displacement data of slopes;

[0014] High-definition network cameras and panoramic stitching cameras used to monitor data on rockfalls on slopes;

[0015] Each monitoring device generates monitoring signals during monitoring, including: analog voltage signals, resistance change signals, digital pulse signals, and digital video streams, which ultimately generate corresponding deep displacement data, surface displacement data, and data on the occurrence of falling rocks.

[0016] The specific method for constructing the yellow warning criterion based on deep displacement data in step 3 is as follows:

[0017] The timing of landslide warnings is identified by detecting outliers in the distribution characteristics of displacement monitoring data, including:

[0018] The displacement displacement of adjacent measuring points within a specific monitoring well at a given monitoring time is sorted from bottom to top to form a random variable, denoted as S. i-1 The expression is:

[0019] ;

[0020] Where: M i Here is the displacement data at time i;

[0021] Perform a normality test on the random variables obtained in real time:

[0022] According to the 3σ law, the range [-2σ, 2σ] is used to determine the confidence level of a random variable following a normal distribution. The test method uses a confidence level of 0.95 and a significance level of 0.05. When the significance test p-value is greater than 0.05, the random variable is considered to follow a normal distribution.

[0023] When a certain monitoring time T0 is included, a certain displacement displacement S is... j When the test result is false, it means that S j For abnormal data points, time T0 can be used as the landslide initiation level warning time point, and a yellow landslide warning can be issued, indicating that the landslide is highly likely.

[0024] If the random variable of displacement at monitoring time T0 follows a normal distribution, then the displacement at the next monitoring time T1 will be tested.

[0025] The specific method for constructing the orange warning criterion based on surface displacement data in step 3 is as follows:

[0026] By acquiring slope surface displacement data, the state transition matrix and transition probabilities are solved using weighted Markov chain theory. The future slope condition is predicted based on existing data. If the conditions for the current day, the previous day, and the next day are all abnormal, an orange alert is issued, indicating a high probability of landslides in the short term, including:

[0027] Calculate the state transition frequency matrix Q for discrete time steps of size k (k=1,2,...,r). k The expression is:

[0028] ;

[0029] The state transition probability matrix is ​​obtained from the state transition frequency matrix, and then row sum normalization is performed to obtain the state transition probability P. k The expression is:

[0030] ;

[0031] Where: p ij Let q be any element in the state transition probability matrix ij Divide by the sum of the rows containing that element, i.e.:

[0032] ;

[0033] Given a Markov chain with a step size of k, the initial probability vector for calculating state transition probabilities is expressed as:

[0034] ;

[0035] in, ;

[0036] Calculate the autocorrelation coefficient for different time series data. The autocorrelation coefficient for the same time series data is calculated using the following formula:

[0037] ;

[0038] The weights of the z-order weighted matrices with different step lengths are calculated using the following formula:

[0039] ;

[0040] The expression for the z-order weighted matrix formed by the weights of each step size is:

[0041] ;

[0042] The state distribution probability matrix, composed of the state distribution vectors for different step sizes, is calculated using the following formula:

[0043] ;

[0044] To calculate the final prediction result of the Markov chain, in the final result vector, the state corresponding to the data with the highest probability is defined as the state predicted by the Markov chain for day n+1, expressed as:

[0045] ;

[0046] Based on the final Markov chain prediction results, the status of the current day and the previous day is judged. If both judgments are abnormal, an orange landslide warning is issued, indicating that the landslide is highly likely.

[0047] If the current day, the next day, and the following day are all in a normal state, then the next monitoring data will be introduced to dynamically verify the data.

[0048] The specific method for constructing the red alert criterion based on the stone rolling event data in step 3 is as follows:

[0049] The frame difference method in the moving target tracking algorithm is used to identify rockfall occurrences. The previous frame image is used as the background model for the current frame to identify and segment moving targets, calculate rockfall frequency, and generate a time-series curve of the target. Based on the rockfall frequency and time-series curve, the abrupt change in the rockfall frequency curve before a landslide is characterized and enhanced using the long-short time window mean ratio method. Long and short windows are continuously slid along the time axis to calculate a time series of the two-window mean ratio. The calculation formula is as follows:

[0050] ;

[0051] Where: STA and LTA are the short-term and long-term window averages, respectively; n s and n l These represent the short and long time window lengths, respectively; F is the time series of rockfall frequency.

[0052] A red alert is issued when the threshold is exceeded for the first time.

[0053] The specific method for precise landslide disaster early warning based on multi-source monitoring data fusion in step 4 is as follows:

[0054] Based on the constructed early warning criteria and the data currently acquired by the monitoring equipment, it is determined whether an early warning needs to be issued, and the corresponding early warning level is determined and issued according to different situations. The rules for determination and issuance are as follows:

[0055] When the deep displacement data reaches the warning criteria, a yellow warning is issued, indicating that there is a high probability of a landslide.

[0056] When the surface displacement reaches the warning criteria, an orange warning is issued, indicating that there is a high probability of a landslide in the short term.

[0057] When the rockfall data reaches the warning threshold, a red alert is issued, indicating that there is a very high probability of a landslide in the short term.

[0058] The advantages of this invention compared to existing technologies are as follows: This invention provides a precise early warning method for landslide disasters based on the fusion of multiple monitoring data. By integrating various monitoring data from open-pit mines and employing multiple prediction algorithms based on early warning value criteria, it achieves precise and graded early warning of landslide disasters in open-pit mines. It can provide different early warning information at different stages of landslide disasters. Once a dangerous situation occurs, remote personnel can promptly grasp the situation and quickly organize subsequent handling work. It effectively solves the problems of inconsistent data interfaces, single types of monitoring and early warning data, and weak integration capabilities of monitoring and early warning systems in existing methods, thereby reducing personnel and equipment investment and improving economic efficiency. Attached Figure Description

[0059] The present invention will be further described below with reference to the accompanying drawings:

[0060] Figure 1 This is a flowchart illustrating the steps of the landslide disaster precision early warning method of the present invention;

[0061] Figure 2 This is a graph showing the typical slope displacement-time characteristics in an embodiment of the present invention.

[0062] Figure 3 This is a graph showing the cumulative displacement versus depth of the monitoring hole at different monitoring time points in embodiment 3-2 of the present invention.

[0063] Figure 4 This is a graph showing the relationship between the displacement and depth of the monitoring hole at different monitoring time points in embodiment 3-2 of the present invention.

[0064] Figure 5 This is a schematic diagram illustrating the timing of the rolling stone displacement warning in an embodiment of the present invention;

[0065] Figure 6 This is a time-series curve of the rockfall frequency on the slope of an open-pit mine in an embodiment of the present invention. Detailed Implementation

[0066] like Figures 1 to 6 As shown, this invention provides a precise early warning method for landslide disasters based on the fusion of multi-source monitoring data, applied to slope disaster early warning in open-pit mines, mainly including:

[0067] The target monitoring data are established, namely deep displacement data, surface displacement data, and rockfall occurrence. Physical monitoring equipment is deployed to collect deep displacement data, surface displacement data, and rockfall occurrence data of open-pit mine slopes. A graded early warning criterion is established, and based on the response order of different data, a yellow early warning criterion based on deep displacement data, an orange early warning criterion based on surface displacement data, and a red early warning criterion based on rockfall monitoring data are constructed. A precise early warning method for landslide disasters based on the fusion of multi-source monitoring data is constructed to achieve precise graded early warning for open-pit mine landslides.

[0068] Furthermore, to address the technical problems of low accuracy and weak anti-interference ability in existing slope early warning methods due to the reliance on single data sources, this invention provides a precise landslide disaster early warning method based on the fusion of multi-source monitoring data, specifically including the following steps:

[0069] Step 1: Determine the target monitoring data, namely deep displacement data, surface displacement data, and the occurrence of rolling stones.

[0070] Step 2: Deploy physical monitoring equipment to collect deep displacement data, surface displacement data, and rockfall occurrence information of the open-pit mine slope. Deep displacement monitoring equipment includes displacement gauges and strain gauges; surface displacement monitoring equipment includes drones, radar, and GNSS automated monitoring points; rockfall occurrence monitoring equipment consists of various models of high-definition surveillance cameras.

[0071] Step 3: Establish multi-dimensional data hierarchical early warning criteria. Based on the response order of different data, construct separate criteria for landslide yellow warning based on deep displacement data, landslide orange warning based on surface displacement data, and landslide red warning based on rockfall monitoring data.

[0072] Step 3.1: Construct a yellow warning criterion for landslides based on deep displacement data;

[0073] Deep displacement is a direct response to disturbances and deformations within the slope, exhibiting a much higher sensitivity than surface displacement changes. Based on normal distribution theory and the distribution characteristics of displacement monitoring data, when the slope enters the accelerated deformation stage, the displacement velocity increases significantly, and the data distribution characteristics show significant differences. Therefore, it can be assumed that when the slope deformation data characteristics undergo abrupt changes, the random variable introducing measurement error deviates from the normal distribution. That is, by performing distribution tests on the distribution characteristics of displacement monitoring data to capture anomalies, the timing of landslide warnings can be identified. The displacement displacement of adjacent measuring points within a specific monitoring borehole at a given monitoring time is sorted from bottom to top to form a random variable, denoted as S. i-1 :

[0074] (1);

[0075] Where: M i Here is the displacement data at time i;

[0076] The random variables acquired in real time are tested for normality. According to the 3σ law, the range [-2σ, 2σ] is selected to judge the confidence level of the random variables following a normal distribution. That is, the test method takes a confidence level of 0.95 and a significance level of 0.05. When the significance test P value is greater than 0.05, the random variables are considered to follow a normal distribution.

[0077] When a certain monitoring time T0 is included, a certain displacement displacement S is... j When the test result is false, it means that S j For abnormal data points, time T0 can be used as the landslide initiation level warning time point, and a yellow landslide warning can be issued, indicating a high probability of landslide;

[0078] If the random variable of displacement at time T0 follows a normal distribution, then the displacement at the next monitoring time T1 will be tested.

[0079] Step 3.2: Constructing an orange landslide early warning criterion based on surface displacement data;

[0080] By acquiring slope surface displacement data, the state transition matrix and transition probability are solved using weighted Markov chain theory. The future slope condition is predicted based on existing data. If the condition is abnormal on the current day, the previous day, and the next day, an orange alert is issued, indicating a high probability of landslide in the short term.

[0081] First, calculate the state transition frequency matrix Q for discrete time steps of k (k=1,2,...,r). k (STFM):

[0082] (2);

[0083] The state transition probability matrix (STPM) is derived from the state transition frequency matrix and then subjected to row sum normalization to obtain the state transition probability P. k :

[0084] (3);

[0085] Where: p ij Let q be any element in the State Transition Probability Matrix (STPM). ij Divide by the sum of the rows containing that element, i.e.:

[0086] .

[0087] Determine the initial probability vector for calculating the state transition probabilities of a Markov chain with a step size of k:

[0088] (4);

[0089] in, .

[0090] Calculate the autocorrelation coefficients for different time series data, and perform autocorrelation coefficient calculations for the same time series data:

[0091] (5);

[0092] Calculate the weight expression for a z-order weighted matrix of different lengths:

[0093] (6);

[0094] The z-order weighted matrix formed by the weights of each step size is:

[0095] (7);

[0096] Calculate the state distribution probability matrix (SDPM) composed of the state distribution vectors for different step sizes:

[0097] (8);

[0098] Calculate the final prediction result of the Markov chain. In the final result vector, the state corresponding to the data with the highest probability is the state predicted by the Markov chain for day n+1:

[0099] (9);

[0100] Based on the final Markov chain prediction results, the status of the current day and the previous day is judged. If both are abnormal, an orange landslide warning is issued, indicating a high probability of a landslide.

[0101] If the current day, the next day, and the following day are all in a normal state, then the next monitoring data will be introduced to dynamically verify the data.

[0102] Step 3.3: Construct a red alert criterion for landslides based on rockfall occurrence data;

[0103] Rockfall identification utilizes the frame difference method in moving target tracking algorithms, using the previous frame image as the background model for the current frame to identify and segment moving targets, calculate rockfall frequency, and generate its time-series curve. Based on the rockfall frequency and its time-series curve, the abrupt change in the rockfall frequency curve before a landslide is characterized and enhanced using the long-short time window mean ratio method. By continuously sliding long and short windows along the time axis, a time series of two-window mean ratios is calculated.

[0104] (10);

[0105] Where: STA and LTA are the short-term and long-term window averages, respectively; n s and n l , respectively, represent the short and long time window lengths; F is the time series of rockfall frequency.

[0106] A red alert is issued when the threshold is exceeded for the first time.

[0107] Step 4: Construct a precise early warning method for landslide disasters based on the fusion of multi-source monitoring data. Based on the data obtained from monitoring equipment and different early warning criteria, determine whether an early warning needs to be issued and determine the early warning level according to different situations. When the deep displacement data reaches the early warning threshold, a yellow warning is issued, indicating a high probability of landslide; when the surface displacement reaches the early warning criteria, an orange warning is issued, indicating a high probability of landslide in the short term; when the rockfall data reaches the early warning threshold, a red warning is issued, indicating a very high probability of landslide in the short term.

[0108] Furthermore, embodiments of the present invention also provide a method for precise early warning of landslide disasters based on the fusion of multi-source monitoring data, specifically including the following steps:

[0109] Step 1: Determine the target monitoring data, namely deep displacement data, surface displacement data, and the occurrence of rockfalls;

[0110] Step 2: Deploy physical monitoring equipment to collect surface displacement data, deep displacement data, and rockfall occurrence information of the open-pit mine slope. For different monitoring objects and areas, the monitoring module may include several monitoring devices set in corresponding locations. For example, in this embodiment, slope monitoring equipment such as GNSS receiver R30 and radar receiver are used to monitor the surface displacement of the slope; inclinometers, strain gauges, and fixed inclinometers are used to monitor the deep displacement information of the slope; and high-definition network cameras and panoramic stitching cameras are used to monitor the rockfall occurrence information of the slope. Each monitoring device can generate monitoring signals during monitoring, including analog voltage signals, resistance change signals, digital pulse signals, and digital video streams, and finally generate corresponding deep displacement, surface displacement, and rockfall data.

[0111] Step 3: Establish hierarchical early warning criteria. Based on the response order of different data, construct yellow early warning criteria based on deep displacement data, orange early warning criteria based on surface displacement data, and red early warning criteria based on rolling stone monitoring data.

[0112] Step 3.1: Construct a yellow warning criterion for landslides based on deep displacement data;

[0113] The displacement displacements of adjacent measuring points within a specific monitoring well at a given monitoring time are sorted from bottom to top to form a random variable, denoted as:

[0114] (11);

[0115] Where: M i Let be the displacement data at time i.

[0116] In this embodiment, displacement monitoring of the deep eastern flank of a certain mine was conducted from January 7, 2020 to March 4, 2020. Statistical analysis was performed on the data from monitoring borehole 3-2. Figure 3 To obtain the cumulative displacement versus depth curves at different monitoring time points of the monitoring borehole, the attached curves can be obtained. Figure 4 The curve showing the relationship between the fault displacement of the monitoring hole and its depth is shown.

[0117] The obtained deep displacement and fault amount were subjected to a normal distribution test. The range [-2σ, 2σ] was used to judge the confidence level of the random variable following a normal distribution. That is, the test method took a confidence level of 0.95 and a significance level of 0.05. When the significance test P value was greater than 0.05, the random variable was considered to follow a normal distribution.

[0118] The data from monitoring well 3-2 were examined. The monitoring data from January 7, 2020, January 15, 2020, January 20, 2020, February 8, 2020, February 13, 2020, February 19, 2020, February 25, 2020, and March 4, 2020 were examined. When the random variable generated after including the displacement at a depth of 27-28m on February 19, 2020, did not follow a normal distribution, it can be determined that the slope had entered the landslide initiation stage on February 19, 2020. At this time, a yellow landslide warning should be issued, and the probability of a landslide is relatively high.

[0119] Step 3.2: Constructing an orange landslide early warning criterion based on surface displacement data;

[0120] By acquiring slope surface displacement data, the state transition matrix and transition probability are solved using weighted Markov chain theory. The future slope displacement state is predicted based on existing data. If the predicted state for the current day, the previous day, and the next day are all abnormal, an orange alert is issued, indicating a high probability of landslide in the short term.

[0121] With a sample size of 20 and a standard deviation multiple of 0.4, the prediction process is explained. For the first 20 monitoring data, the mean is calculated using equations (12) and (13). =1.0215, standard deviation is s=0.6982, and k is 0.4. The dividing point D=1.3008 is determined according to formula (14).

[0122] (12);

[0123] (13);

[0124] (14);

[0125] The 20 displacement velocity data have a minimum value of -0.12 and a maximum value of 2.39. The data states are divided into two categories with D as the dividing point. Data with values ​​less than D are classified as normal states, represented by 1, and the rest are classified as abnormal states, represented by 2. The monitoring data and state sequence are shown in Table 1 below.

[0126] Table 1. Surface displacement monitoring point data and status classification

[0127]

[0128] The state transition frequency matrix and state transition probability matrix for step sizes 1, 2, 3, 4, and 5 are calculated as follows:

[0129] (15);

[0130] For the selected 20 data points, the 20th data point has a state of 2, therefore the initial state vector A1 with a step size of 1 is obtained. T (n)=(0,1), similarly, the states of data points 19, 18, 17, and 16 are 1, 1, 1, 2 respectively. Therefore, the initial state vector for each step is A2. T (n)=(1,0), A3 T (n)=(1,0), A4 T (n)=(1,0), A5 T (n)=(0,1).

[0131] Calculate the autocorrelation coefficients and weight vectors for each step size. The step autocorrelation coefficients are r1=-0.5176, r2=0.2228, r3=-0.5072, r4=0.2757, and r5=-0.2833.

[0132] After converting the above data into a matrix, the final prediction result of the Markov chain is calculated. For each vector, the state corresponding to the column with the largest element is the most likely state of the future displacement velocity predicted by the weighted Markov chain of that order. The results are shown in Table 2 below.

[0133] Table 2 Weight vectors of each order and state probability distribution

[0134]

[0135] The calculation results show that the predicted value for the displacement status of the next day is 1, meaning the status of the next day is normal. Since there are instances of normal status among the previous day, the current day, and the next day, no landslide warning is issued. After dynamically updating the data, the aforementioned weighted Markov prediction process is executed to obtain the status prediction sequence. Based on the actual status information, if the status of the current day, the previous day, and the predicted next day are all abnormal, an orange warning is issued, indicating a high probability of a landslide in the short term.

[0136] Step 3.3: Construct red alert criteria based on rolling stone occurrence data;

[0137] Based on the red landslide warning issued by Rolling Stone, the frame difference method in the moving target tracking algorithm is used to identify and segment moving targets by using the previous frame image as the background model of the current frame, calculating the rockfall frequency, and generating its time series curve. Based on the rockfall frequency and its time series curve, the abrupt change in the rockfall frequency curve before the landslide is characterized and enhanced by the long-short time window mean ratio method. The long window and short window are continuously slid along the time axis to calculate a time series of the two-window mean ratio. The first time the threshold is exceeded is used as the identification mark, and a red warning is issued.

[0138] (16);

[0139] Where: STA and LTA are the short-term and long-term window averages, respectively; n s and n l , respectively, represent the short and long time window lengths; F is the time series of rockfall frequency.

[0140] Taking a certain landslide as an example, small-scale rockfalls occurred in the early stage of the overall landslide. Within 2 minutes before the overall landslide, the frequency of rockfalls began to rise significantly, and the time-series curve was significantly different from the previous stage. It reached its maximum value during the overall landslide, which lasted for about 8 seconds. During the landslide, the frequency of rockfalls rose rapidly to its maximum value. After the landslide, the remaining rocks continued to fall, and the frequency of rockfalls returned to a low level.

[0141] As attached Figure 6 As shown, before the landslide occurs, the rockfall frequency time series curve rises sharply, which is very beneficial for setting the landslide warning threshold. If the warning threshold is set to 5, the warning lead time for the rockfall frequency is 113 seconds.

[0142] Step 4: Construct a precise early warning method for landslide disasters based on the fusion of multi-source monitoring data. Based on the data obtained from monitoring equipment and different early warning criteria, determine whether an early warning is needed and determine the early warning level according to different situations. When the deep displacement data reaches the early warning threshold, a yellow warning is issued, indicating a high probability of landslide; when the surface displacement reaches the early warning criteria, an orange warning is issued, indicating a high probability of landslide in the short term; when the rockfall data reaches the early warning threshold, a red warning is issued, indicating a very high probability of landslide in the short term.

[0143] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A landslide disaster precision early warning method based on multi-element monitoring data fusion, characterized in that: The following are the steps for landslide disaster early warning: Step 1: Determine the target monitoring data, including: deep displacement data, surface displacement data, and data on the occurrence of rockfalls; Step 2: Deploy physical monitoring modules for collecting deep displacement data, surface displacement data, and rockfall occurrence data of open-pit mine slopes. The physical monitoring modules include multiple monitoring devices with different functions. Step 3: Based on the response order of different data, construct corresponding hierarchical early warning criteria, including: yellow early warning criteria based on deep displacement data, orange early warning criteria based on surface displacement data, and red early warning criteria based on rockfall occurrence data; The specific method for constructing the yellow warning criterion based on deep displacement data is as follows: The timing of landslide warnings is identified by detecting outliers in the distribution characteristics of displacement monitoring data, including: The displacement dislocation amount of adjacent measuring points in a certain monitoring hole at a certain monitoring time is sorted from bottom to top to form a random variable, which is denoted as S i-1 , and the expression is: ； wherein: M i is displacement data at time i; Perform a normality test on the random variables obtained in real time: According to the 3σ law, the range [-2σ, 2σ] is used to determine the confidence level of a random variable following a normal distribution. The test method uses a confidence level of 0.95 and a significance level of 0.

05. When the significance test p-value is greater than 0.05, the random variable is considered to follow a normal distribution. When a certain monitoring time T0 incorporates a certain displacement dislocation amount S j , the test result is false, indicating that S j is an abnormal data point, and the time T0 can be used as a landslide starting stage warning time point, and a landslide yellow warning is issued, indicating that the landslide is likely to occur; If the random variable of displacement displacement at monitoring time T0 follows a normal distribution, then the displacement displacement at the next monitoring time T1 will be tested. Step 4: Based on the data obtained by the monitoring equipment and different early warning criteria, conduct precise early warning of landslide disasters based on the fusion of multi-source monitoring data, determine whether the corresponding data currently monitored needs to be issued as an early warning, and determine the early warning level according to different situations.

2. The landslide disaster precision early warning method based on multi-element monitoring data fusion according to claim 1, characterized in that: The monitoring equipment used in step 2 is set up in corresponding locations for different monitoring objects and areas, including: GNSS receivers and radar receivers used for monitoring slope surface displacement data; Inclinometers, strain gauges, and fixed inclinometers used to monitor deep displacement data of slopes; High-definition network cameras and panoramic stitching cameras used to monitor data on rockfalls on slopes; Each monitoring device generates monitoring signals during monitoring, including: analog voltage signals, resistance change signals, digital pulse signals, and digital video streams, which ultimately generate corresponding deep displacement data, surface displacement data, and data on the occurrence of falling rocks.

3. The landslide disaster precision early warning method based on multi-element monitoring data fusion according to claim 2, characterized in that: The specific method for constructing the orange warning criterion based on surface displacement data in step 3 is as follows: By acquiring slope surface displacement data, the state transition matrix and transition probabilities are solved using weighted Markov chain theory. The future slope condition is predicted based on existing data. If the conditions for the current day, the previous day, and the next day are all abnormal, an orange alert is issued, indicating a high probability of landslides in the short term, including: A state transition frequency matrix Q at a step size k (k = 1, 2,..., r) at discrete time is calculated k , and the expression is ； According to the state transition frequency matrix, a state transition probability matrix is obtained, and a row sum normalization processing is performed to obtain a state transition probability P k , and the expression is: ； where: p ij is the sum of the elements in the row of the state transition probability matrix in which the element q ij is located, i.e.: ； Given a Markov chain with a step size of k, the initial probability vector for calculating state transition probabilities is expressed as: ； wherein ; Calculate the autocorrelation coefficient for different time series data. The autocorrelation coefficient for the same time series data is calculated using the following formula: ； The weights of the z-order weighted matrices with different step lengths are calculated using the following formula: ； The expression for the z-order weighted matrix formed by the weights of each step size is: ； The state distribution probability matrix, composed of the state distribution vectors for different step sizes, is calculated using the following formula: ； To calculate the final prediction result of the Markov chain, in the final result vector, the state corresponding to the data with the highest probability is defined as the state predicted by the Markov chain for day n+1, expressed as: ； Based on the final Markov chain prediction results, the status of the current day and the previous day is judged. If both judgments are abnormal, an orange landslide warning is issued, indicating that the landslide is highly likely. If the current day, the next day, and the following day are all in a normal state, then the next monitoring data will be introduced to dynamically verify the data.

4. The landslide disaster precision early warning method based on multi-element monitoring data fusion according to claim 3, characterized in that: The specific method for constructing the red alert criterion based on the stone rolling event data in step 3 is as follows: The frame difference method in the moving target tracking algorithm is used to identify rockfall occurrences. The previous frame image is used as the background model for the current frame to identify and segment moving targets, calculate rockfall frequency, and generate a time-series curve of the target. Based on the rockfall frequency and time-series curve, the abrupt change in the rockfall frequency curve before a landslide is characterized and enhanced using the long-short time window mean ratio method. Long and short windows are continuously slid along the time axis to calculate a time series of the two-window mean ratio. The calculation formula is as follows: ； where: STA, LTA are short and long time window mean values; n s and n l are short and long time window lengths; F is the rockfall frequency time series; A red alert is issued when the threshold is exceeded for the first time.

5. The landslide disaster precision early warning method based on multi-element monitoring data fusion according to claim 4, characterized in that: The specific method for precise landslide disaster early warning based on multi-source monitoring data fusion in step 4 is as follows: Based on the constructed early warning criteria and the data currently acquired by the monitoring equipment, it is determined whether an early warning needs to be issued, and the corresponding early warning level is determined and issued according to different situations. The rules for determination and issuance are as follows: When the deep displacement data reaches the warning criteria, a yellow warning is issued, indicating that there is a high probability of a landslide. When the surface displacement reaches the warning criteria, an orange warning is issued, indicating that there is a high probability of a landslide in the short term. When the rockfall data reaches the warning threshold, a red alert is issued, indicating that there is a very high probability of a landslide in the short term.

Images