Mine slope landslide disaster early warning method and system based on multi-modal data

Abstract

The invention relates to the technical field of slope disaster early warning, in particular to a mine slope landslide disaster early warning method and system based on multi-modal data, and the method comprises the following steps: monitoring a surface mine slope through radar equipment, and obtaining a radar data sequence corresponding to slope rockfall and rock-soil loosening conditions; extracting a meteorological data sequence at the same moment as the radar data sequence from meteorological equipment; and calling a pre-established slope disaster early warning model through the cloud server, and predicting a landslide occurrence area and landslide occurrence time of the surface mine slope by using the radar data sequence and the meteorological data sequence. According to the method, radar data and meteorological data of rockfall and rock-soil loosening in the initial stage of the slope disaster are utilized to predict the occurrence area and occurrence time of the landslide in the outbreak period of the slope disaster, the early warning timeliness is met, and the neural network structure model is utilized to predict the occurrence area and occurrence time of the landslide on the basis of multi-modal data. And the prediction accuracy of the landslide disaster is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of slope disaster early warning, and in particular to a method and system for early warning of slope landslide disasters in open-pit mines. Background Art

[0002] Slope hazards refer to disasters in the form of rupture, landslides, and collapses of geological bodies and soil and rock bodies such as slopes and cliffs caused by various factors (such as earthquakes, rainfall, and human activities). Slope hazards usually occur rapidly, are highly destructive, and are difficult to predict. Slope hazards are mainly divided into two stages: the initial stage of the disaster and the outbreak stage. In the initial stage of slope hazards, rockfall and loosening of rock and soil are the main manifestations, while in the outbreak stage, landslides and mudslides are the main manifestations. Most of them are affected by factors such as rainfall and earthquakes, and have a certain degree of seasonality and regularity. Abnormal conditions can be monitored and identified by monitoring meteorological indicators such as rainfall, soil moisture, and water levels.

[0003] In the existing technology, early warning of slope landslide disasters is usually carried out in real time by monitoring meteorological indicators such as rainfall, soil moisture, and water level. The indicator data is single, resulting in limited accuracy in landslide disaster identification. Moreover, the impact of meteorological indicators on landslide occurrence is cumulative. Therefore, the sensitivity of meteorological indicators in warning of landslide occurrence is insufficient. When used for real-time monitoring of landslide disasters, it is easy to cause a lag in landslide disaster warning and reduce the timeliness of warning. Summary of the Invention

[0004] The purpose of the present invention is to provide a mine slope landslide disaster early warning method and system based on multimodal data to solve the technical problems of the existing technology such as the single monitoring data and limited accuracy and sensitivity.

[0005] In order to solve the above technical problems, the present invention specifically provides the following technical solutions:

[0006] A mine slope landslide disaster early warning method based on multimodal data includes the following steps:

[0007] Use radar equipment to monitor open-pit mine slopes and obtain radar data sequences corresponding to slope rockfall and loose soil conditions;

[0008] Extracting a meteorological data sequence at the same time as the radar data sequence in the meteorological device, and transmitting the meteorological data sequence and the radar data sequence to a cloud server;

[0009] The pre-established slope disaster warning model is called through the cloud server, and the radar data sequence and the meteorological data sequence are used to predict the landslide occurrence area and landslide occurrence time of the open-pit mine slope.

[0010] As a preferred solution of the present invention, the radar device obtains radar data corresponding to the slope rockfall and rock and soil loosening conditions through a set deformation rate threshold;

[0011] When the radar equipment detects that the slope deformation rate exceeds the deformation rate threshold, a radar data sequence with this time as the starting point is marked as the radar data corresponding to the slope rockfall and rock loosening conditions;

[0012] When the radar device detects that the slope deformation rate does not exceed the deformation rate threshold, the radar data will not be marked.

[0013] As a preferred solution of the present invention, the slope disaster warning model includes a landslide prediction structure, a Verhulst grey model and an error correction structure;

[0014] The landslide prediction structure includes a radar data landslide prediction model and a meteorological data landslide early warning model, wherein the radar data landslide prediction model is used to predict the landslide occurrence area and landslide occurrence time of the open-pit mine slope based on the radar data sequence;

[0015] The meteorological data landslide prediction model is used to predict the landslide occurrence area of ​​the open-pit mine slope based on the meteorological data sequence;

[0016] The Verhulst grey model is used to predict the occurrence time of landslide on the slope of the open-pit mine based on the radar data sequence;

[0017] The error correction structure includes a regional error correction model and a time error correction model, wherein the regional error correction model is used to perform error correction on the landslide occurrence area predicted by the radar data landslide prediction model using the landslide occurrence area predicted by the meteorological data landslide prediction model;

[0018] The time error correction model is used to perform error correction on the landslide occurrence time predicted by the radar data landslide prediction model using the landslide occurrence time predicted by the Verhulst grey model.

[0019] As a preferred embodiment of the present invention, the radar data landslide prediction model is:

[0020] {Pr,tr}=CNN(dataList_radar);

[0021] Where Pr is the landslide occurrence area output by the radar data landslide prediction model, tr is the landslide occurrence time output by the radar data landslide prediction model, dataList_radar is the radar data sequence, and CNN is the convolutional neural network.

[0022] The meteorological data landslide early warning model is:

[0023] {Pw}=CNN(dataList_weather);

[0024] Where Pw is the landslide occurrence area output by the meteorological data landslide prediction model, dataList_weather is the meteorological data sequence, and CNN is the convolutional neural network.

[0025] As a preferred solution of the present invention, the differential equation of the Verhulst grey model is: Where x(t) is the slope shape variable corresponding to the radar data at the tth time sequence in the radar data sequence, is the slope deformation rate corresponding to the tth time sequence in the radar data sequence, a is the development coefficient reflecting the slope deformation trend, and b is the gray action of external influence;

[0026] The slope shape variable sequence {x(t)|t∈[1,n]} corresponding to the radar data sequence is accumulated to generate a new sequence {x(t) (1) |t∈[1,n]}, and then estimate the parameters a and b based on the new sequence by the least squares method, where: Where n is the total number of time series of radar data sequence, x(1) (1) ,x(2) (1) ,x(n-1) (1) ,x(n) (1) They are {x(t) (1) The data at the 1st, 2nd, n-1st, and nth time series in |t∈[1,n]}, x(2) and x(n) are the data at the 2nd and nth time series in {x(t)|t∈[1,n]} respectively;

[0027] The differential equation of the Verhulst grey model is Solve the maximum value and get the time when the landslide occurred where tv is the landslide occurrence time output by the Verhulst grey model, t0 is the monitoring time of the radar data at the first time sequence of the radar data sequence, and x(1) is the data at the first time sequence in {x(t)|t∈[1,n]}.

[0028] As a preferred solution of the present invention, the regional error correction model is: Pfixed=Add(Unet(Pr, Pw), Pr), where Pfixed is the corrected landslide occurrence area, Unet is the Unet network structure, Add is the pixel layer addition operation, Pr is the landslide occurrence area output by the radar data landslide prediction model, and Pw is the landslide occurrence area output by the meteorological data landslide prediction model;

[0029] The time error correction model is tfixed=wr*tr+(1-wr)*tv, where where tfixed is the corrected landslide occurrence time, wr is the weight of tr, tr is the landslide occurrence time output by the radar data landslide prediction model, tv is the landslide occurrence time output by the Verhulst grey model, tGT is the true value of the landslide occurrence time, MSE(tr,tGT) is the mean square error between tr and tGT, and MSE(tv,tGT) is the mean square error between tv and tGT.

[0030] As a preferred embodiment of the present invention, the present invention provides a mine slope landslide disaster early warning system based on multimodal data, which is applied to a mine slope landslide disaster early warning method based on multimodal data. The system includes:

[0031] Radar equipment is used to monitor open-pit mine slopes and obtain radar data sequences corresponding to slope rockfall and loosening of rock and soil;

[0032] Meteorological equipment for monitoring meteorological data on open-pit mine slopes;

[0033] The cloud server is used to call a pre-established slope disaster warning model and use the radar data sequence and meteorological data sequence to predict the landslide occurrence area and landslide occurrence time of the open-pit mine slope.

[0034] As a preferred solution of the present invention, the slope disaster warning model includes a landslide prediction structure, a Verhulst grey model and an error correction structure;

[0035] Wherein, the landslide prediction structure includes a radar data landslide prediction model and a meteorological data landslide early warning model;

[0036] The error correction structure includes a regional error correction model and a temporal error correction model;

[0037] The output ends of the radar data landslide prediction model and the meteorological data landslide early warning model are connected to the input end of the regional error correction model;

[0038] The output ends of the radar data landslide prediction model and the Verhulst grey model are connected to the input end of the time error correction model.

[0039] As a preferred embodiment of the present invention, the method for constructing the radar data landslide prediction model includes:

[0040] Obtain radar data series and meteorological data series of multiple slopes where landslide disasters occurred;

[0041] Extract the time points of rockfall and soil loosening on the slopes, as well as the time points and areas of landslide occurrence, from the radar data series of the slopes where landslide disasters occurred;

[0042] In the radar data sequence of the slope where the landslide occurred, a time sequence following the time point when the slope rockfall and rock and soil loosening occurred was intercepted as the input of the CNN network, and the time point and area where the landslide occurred were used as the output of the CNN network. The CNN network was trained to obtain the radar data landslide prediction model.

[0043] As a preferred solution of the present invention, the method for constructing the meteorological data landslide early warning model includes:

[0044] A time sequence subsequent to the time point when slope rockfall and rock and soil loosening occurred is intercepted from the meteorological data sequence as the input of the CNN neural network, and the area where the landslide occurred is used as the output of the CNN network. The CNN network is trained to obtain the meteorological data landslide prediction model.

[0045] Compared with the prior art, the present invention has the following beneficial effects:

[0046] The present invention uses radar data and meteorological data of rockfall and rock and soil loosening in the early stage of slope disasters to predict the occurrence area and time of landslides during the outbreak period of slope disasters, thereby realizing prediction and early warning of slope landslides in the early stage of slope disasters, meeting the timeliness of early warning. In addition, a neural network structure model is used to predict the occurrence area and time of landslides based on radar data and meteorological data, and data multi-modality is used to improve the accuracy of landslide disaster prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for the embodiments or the description of the prior art. Obviously, the drawings described below are merely exemplary, and those skilled in the art can derive other implementation drawings based on the provided drawings without inventive effort.

[0048] Figure 1A flow chart of a method for early warning of landslide disasters on open-pit mine slopes provided by an embodiment of the present invention;

[0049] Figure 2 A block diagram of an open-pit mine slope landslide disaster early warning system provided by an embodiment of the present invention;

[0050] Figure 3 This is a structural diagram of the slope disaster warning model provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0051] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0052] like Figure 1 As shown, the present invention provides a mine slope landslide disaster early warning method based on multimodal data, comprising the following steps:

[0053] Use radar equipment to monitor open-pit mine slopes and obtain radar data sequences corresponding to slope rockfall and loose soil conditions;

[0054] Extracting the meteorological data sequence that is simultaneous with the radar data sequence from the meteorological equipment, and transmitting the meteorological data sequence and the radar data sequence to the cloud server;

[0055] The pre-established slope disaster warning model is called through the cloud server, and the radar data series and meteorological data series are used to predict the landslide occurrence area and time of the open-pit mine slope.

[0056] In order to provide early warning of slope landslides, the present invention obtains radar data and meteorological data of falling rocks and loosening of rock and soil in the early stage of slope disasters, which will be used to predict slope landslide disasters, so as to provide early warning of slope landslide disasters in the early stage of occurrence, improve the timeliness of warning, achieve the effect of early warning and early prevention, and help reduce disaster losses.

[0057] The present invention constructs a slope disaster early warning model to predict the landslide occurrence area and time of open-pit mine slopes based on radar data and meteorological data. It uses multimodal data for prediction, more comprehensively understands and processes data characteristics, and thus obtains accurate prediction results.

[0058] The slope disaster early warning model constructed by the present invention is composed of a landslide prediction structure, a Verhulst grey model and an error correction structure. Among them, the landslide prediction structure is composed of two neural network models, one is a radar data landslide prediction model that uses radar data to predict the time and area of ​​landslide occurrence, and the other is a meteorological data landslide prediction model that uses meteorological data to predict the landslide occurrence area. In order to further correct the accuracy of the landslide occurrence area result obtained by radar data prediction alone, the present invention uses the landslide occurrence area result obtained by meteorological data, extracts the error value between the landslide occurrence area result obtained by radar data prediction and the landslide occurrence area result obtained by meteorological data prediction, combines the error value with the landslide occurrence area result obtained by radar data prediction, corrects the prediction error in the landslide occurrence area result obtained by radar data prediction, and obtains the landslide occurrence area result obtained by radar data prediction corrected by the landslide occurrence area result obtained by meteorological data prediction, that is, ultimately obtains a more accurate landslide occurrence area result.

[0059] This correction process corresponds to the regional error correction model, using one modal data to correct the prediction results of another modal data. It is equivalent to using radar data and meteorological data simultaneously in the prediction of landslide occurrence areas, achieving the effect of multimodal prediction, using more data features, and improving the accuracy of model prediction.

[0060] The present invention also utilizes the Verhulst grey model, a fast and low-cost tool for landslide early warning, to calculate the landslide warning time. The present invention combines the landslide occurrence time calculated by the Verhulst grey model with the landslide occurrence time result obtained by radar data prediction, and corrects the landslide occurrence time result obtained by radar data prediction. The landslide occurrence time calculated by the Verhulst grey model and the landslide occurrence time result obtained by radar data prediction are weighted averaged, wherein the weight depends on the model performance of the Verhulst grey model and the radar data landslide prediction model. The better the performance of the model, the higher the weight of the time prediction result. Therefore, the landslide occurrence time results predicted by the two models are complementary to each other, thereby obtaining a more accurate time prediction result.

[0061] This correction process corresponds to a temporal error correction model, combining the performance of gray models, which are suitable for small samples but may be inaccurate in the long term, with machine learning models, which require large amounts of data but can handle nonlinear relationships. By combining their respective strengths, they compensate for their shortcomings and improve the accuracy of model predictions.

[0062] In summary, the slope disaster warning model constructed by the present invention not only sets a neural network structure for realizing landslide occurrence area and time prediction, but also sets a structure for correcting the predicted value of landslide occurrence area and time, which can achieve high-precision prediction performance.

[0063] In order to provide early warning of slope landslides, the present invention obtains radar data and meteorological data of rockfall and rock and soil loosening in the early stages of slope disasters, specifically as follows:

[0064] The radar equipment obtains radar data corresponding to the slope rockfall and rock and soil loosening conditions through the set deformation rate threshold;

[0065] When the radar equipment detects that the slope deformation rate exceeds the deformation rate threshold, a radar data sequence with a certain period of time (usually 4-5 consecutive time points, which can also be set according to actual needs) is marked with this time as the starting point, and used as the radar data corresponding to the slope rockfall and rock loosening conditions;

[0066] When the radar device detects that the slope deformation rate does not exceed the deformation rate threshold, the radar data will not be marked.

[0067] like Figure 3 As shown in Figure 1, the slope disaster warning model includes a landslide prediction structure, a Verhulst grey model, and an error correction structure;

[0068] The landslide prediction structure includes a radar data landslide prediction model and a meteorological data landslide early warning model. The radar data landslide prediction model is used to predict the landslide occurrence area and landslide occurrence time of the open-pit mine slope based on the radar data sequence.

[0069] The meteorological data landslide prediction model is used to predict the landslide occurrence area of ​​the open-pit mine slope based on the meteorological data sequence;

[0070] The Verhulst grey model is used to predict the occurrence time of landslides on open-pit mine slopes based on radar data sequences;

[0071] The error correction structure includes a regional error correction model and a temporal error correction model, wherein the regional error correction model is used to perform error correction on the landslide occurrence area predicted by the radar data landslide prediction model using the landslide occurrence area predicted by the meteorological data landslide prediction model;

[0072] The time error correction model is used to correct the error of the landslide occurrence time predicted by the radar data landslide prediction model using the landslide occurrence time predicted by the Verhulst grey model.

[0073] The landslide prediction model based on radar data is:

[0074] {Pr,tr}=CNN(dataList_radar);

[0075] Where Pr is the landslide occurrence area output by the radar data landslide prediction model, tr is the landslide occurrence time output by the radar data landslide prediction model, dataList_radar is the radar data sequence, and CNN is the convolutional neural network.

[0076] The meteorological data landslide warning model is:

[0077] {Pw}=CNN(dataList_weather);

[0078] Where Pw is the landslide occurrence area output by the meteorological data landslide prediction model, dataList_weather is the meteorological data sequence, and CNN is the convolutional neural network.

[0079] The regional error correction model is: Pfixed = Add(Unet(Pr, Pw), Pr), where Pfixed is the corrected landslide occurrence area, Unet is the Unet network structure, Add is the pixel-level addition operation, Pr is the landslide occurrence area output by the radar data landslide prediction model, and Pw is the landslide occurrence area output by the meteorological data landslide prediction model.

[0080] The original U-Net network structure is designed for segmentation tasks, and the present invention makes slight modifications to make it suitable for error extraction. Among them, the input is set to two region images as input, and the output is set to the difference map between the two region images. The U-Net architecture is modified, and the input layer: two images are used as input channels. For example, if each image is RGB (3 channels), the input layer should have 6 channels (3 corresponding to the first image and the other 3 corresponding to the second image). Encoder (contraction path): The same as the original U-Net, using convolution layers and pooling layers. Bottleneck: The same as the original U-Net, this is the deepest layer of the network. Decoder (expansion path): In the decoder, an additional convolution layer can be added to merge the feature maps of the two images, and then the output is reconstructed through a series of convolution and upsampling layers. Output layer: A convolution layer is used to output the error map, and its number of channels is usually 1 (single channel). Therefore, the present invention can use Unet (Pr, Pw) to achieve the extraction of two region errors.

[0081] In order to further correct the accuracy of the landslide occurrence area results predicted by radar data alone, the present invention utilizes the landslide occurrence area results obtained by meteorological data, extracts the error value between the landslide occurrence area results predicted by radar data and the landslide occurrence area results predicted by meteorological data, combines the error value with the landslide occurrence area results predicted by radar data, corrects the prediction error in the landslide occurrence area results predicted by radar data, and obtains the landslide occurrence area results predicted by radar data corrected by the landslide occurrence area results predicted by meteorological data, that is, ultimately obtains a more accurate landslide occurrence area result.

[0082] The differential equation of the Verhulst grey model is: Where x(t) is the slope shape variable corresponding to the radar data at the t-th time sequence in the radar data sequence, is the slope deformation rate corresponding to the tth time sequence in the radar data sequence, a is the development coefficient reflecting the slope deformation trend, and b is the gray action of external influence;

[0083] The slope shape variable sequence {x(t)|t∈[1,n]} corresponding to the radar data sequence is accumulated to generate a new sequence {x(t) (1) |t∈[1,n]}, and then estimate the parameters a and b based on the new sequence using the least squares method, where: Where n is the total number of time series of radar data sequence, x(1) (1) ,x(2) (1) ,x(n-1) (1) ,x(n) (1) They are {x(t) (1) The data at the 1st, 2nd, n-1st, and nth time series in |t∈[1,n]}, x(2) and x(n) are the data at the 2nd and nth time series in {x(t)|t∈[1,n]} respectively;

[0084] The differential equation of the Verhulst grey model is Solve the maximum value and get the time when the landslide occurred where tv is the landslide occurrence time output by the Verhulst grey model, t0 is the monitoring time of the radar data at the first time sequence of the radar data sequence, and x(1) is the data at the first time sequence in {x(t)|t∈[1,n]}.

[0085] The time error correction model is tfixed = wr*tr+(1-wr)*tv, where where tfixed is the corrected landslide occurrence time, wr is the weight of tr, tr is the landslide occurrence time output by the radar data landslide prediction model, tv is the landslide occurrence time output by the Verhulst grey model, tGT is the true value of the landslide occurrence time, MSE(tr,tGT) is the mean square error between tr and tGT, and MSE(tv,tGT) is the mean square error between tv and tGT.

[0086] The present invention also utilizes the Verhulst grey model, a fast and low-cost tool for landslide early warning, to calculate the landslide warning time. The present invention combines the landslide occurrence time calculated by the Verhulst grey model with the landslide occurrence time result obtained by radar data prediction, and corrects the landslide occurrence time result obtained by radar data prediction. The landslide occurrence time calculated by the Verhulst grey model and the landslide occurrence time result obtained by radar data prediction are weighted averaged, wherein the weight depends on the model performance of the Verhulst grey model and the radar data landslide prediction model. The better the performance of the model, the higher the weight of the time prediction result. Therefore, the landslide occurrence time results predicted by the two models are complementary to each other, thereby obtaining a more accurate time prediction result.

[0087] like Figure 2 As shown, the present invention provides a mine slope landslide disaster early warning system based on multimodal data, which is applied to a mine slope landslide disaster early warning method based on multimodal data. The system includes:

[0088] Radar equipment is used to monitor open-pit mine slopes and obtain radar data sequences corresponding to slope rockfall and loosening of rock and soil;

[0089] Meteorological equipment for monitoring meteorological data on open-pit mine slopes;

[0090] The cloud server is used to call the pre-established slope disaster warning model and use radar data sequences and meteorological data sequences to predict the landslide occurrence area and time of the open-pit mine slope.

[0091] The slope hazard warning model includes a landslide prediction structure, a Verhulst grey model, and an error correction structure;

[0092] Among them, the landslide prediction structure includes the radar data landslide prediction model and the meteorological data landslide early warning model;

[0093] The error correction structure includes a regional error correction model and a temporal error correction model;

[0094] The output ends of the radar data landslide prediction model and the meteorological data landslide early warning model are connected to the input end of the regional error correction model;

[0095] The outputs of the radar data landslide prediction model and the Verhulst grey model are connected to the input of the time error correction model.

[0096] The construction method of radar data landslide prediction model includes:

[0097] Obtain radar data series and meteorological data series of multiple slopes where landslide disasters occurred;

[0098] Extract the time points of rockfall and soil loosening on the slopes, as well as the time points and areas of landslide occurrence, from the radar data series of the slopes where landslide disasters occurred;

[0099] In the radar data sequence of the slope where the landslide occurred, a time sequence following the time point when the slope rockfall and rock and soil loosening occurred was intercepted as the input of the CNN network. The time point and area where the landslide occurred were used as the output of the CNN network. The CNN network was trained to obtain a radar data landslide prediction model.

[0100] The method for constructing a meteorological data landslide early warning model includes:

[0101] In the meteorological data sequence, a time sequence following the time point when slope rockfall and rock and soil loosening occurred is intercepted as the input of the CNN neural network, and the area where landslide occurred is used as the output of the CNN network. The CNN network is trained to obtain a meteorological data landslide prediction model.

[0102] The present invention uses radar data and meteorological data of rockfall and rock and soil loosening in the early stage of slope disasters to predict the occurrence area and time of landslides during the outbreak period of slope disasters, thereby realizing prediction and early warning of slope landslides in the early stage of slope disasters, meeting the timeliness of early warning. In addition, a neural network structure model is used to predict the occurrence area and time of landslides based on radar data and meteorological data, and data multi-modality is used to improve the accuracy of landslide disaster prediction.

[0103] The above embodiments are merely exemplary embodiments of the present application and are not intended to limit the scope of the present application. The scope of protection of the present application is defined by the claims. Those skilled in the art may make various modifications or equivalent substitutions to the present application within the essence and scope of protection of the present application, and such modifications or equivalent substitutions shall also be deemed to fall within the scope of protection of the present application.

Claims

1. A mine slope landslide disaster early warning method based on multimodal data, characterized in that: The following steps are involved: Use radar equipment to monitor open-pit mine slopes and obtain radar data sequences corresponding to slope rockfall and loose rock and soil conditions; extracting a meteorological data sequence at the same time as the radar data sequence in the meteorological device, and transmitting the meteorological data sequence and the radar data sequence to a cloud server; The pre-established slope disaster warning model is called through the cloud server, and the radar data sequence and the meteorological data sequence are used to predict the landslide occurrence area and landslide occurrence time of the open-pit mine slope.

2. The mine slope landslide disaster early warning method based on multimodal data according to claim 1, characterized in that: The radar device acquires radar data corresponding to the slope rockfall and rock and soil loosening conditions through a set deformation rate threshold; When the radar equipment detects that the slope deformation rate exceeds the deformation rate threshold, a radar data sequence with this time as the starting point is marked as the radar data corresponding to the slope rockfall and rock loosening conditions; When the radar device detects that the slope deformation rate does not exceed the deformation rate threshold, the radar data will not be marked.

3. The mine slope landslide disaster early warning method based on multimodal data according to claim 1, characterized in that: The slope disaster warning model includes a landslide prediction structure, a Verhulst grey model and an error correction structure; The landslide prediction structure includes a radar data landslide prediction model and a meteorological data landslide early warning model, wherein the radar data landslide prediction model is used to predict the landslide occurrence area and landslide occurrence time of the open-pit mine slope based on the radar data sequence; The meteorological data landslide prediction model is used to predict the landslide occurrence area of ​​the open-pit mine slope based on the meteorological data sequence; The Verhulst grey model is used to predict the occurrence time of landslide on the slope of the open-pit mine based on the radar data sequence; The error correction structure includes a regional error correction model and a time error correction model, wherein the regional error correction model is used to perform error correction on the landslide occurrence area predicted by the radar data landslide prediction model using the landslide occurrence area predicted by the meteorological data landslide prediction model; The time error correction model is used to perform error correction on the landslide occurrence time predicted by the radar data landslide prediction model using the landslide occurrence time predicted by the Verhulst grey model.

4. The mine slope landslide disaster early warning method based on multimodal data according to claim 3 is characterized by: The radar data landslide prediction model is: {Pr,tr}=CNN(dataList_radar); Where Pr is the landslide occurrence area output by the radar data landslide prediction model, tr is the landslide occurrence time output by the radar data landslide prediction model, dataList_radar is the radar data sequence, and CNN is the convolutional neural network. The meteorological data landslide early warning model is: {Pw}=CNN(dataList_weather); Where Pw is the landslide occurrence area output by the meteorological data landslide prediction model, dataList_weather is the meteorological data sequence, and CNN is the convolutional neural network.

5. The mine slope landslide disaster early warning method based on multimodal data according to claim 4 is characterized by: The differential equation of the Verhulst grey model is: Where x(t) is the slope shape variable corresponding to the radar data at the tth time sequence in the radar data sequence, is the slope deformation rate corresponding to the tth time sequence in the radar data sequence, a is the development coefficient reflecting the slope deformation trend, and b is the gray action of external influence; The slope shape variable sequence {x(t)|t∈[1,n]} corresponding to the radar data sequence is accumulated to generate a new sequence {x(t)( 1 )|t∈[1,n]}, and then estimate the parameters a and b based on the new sequence by the least squares method, where: Where n is the total number of time series of radar data sequence, x(1) (1) ,x(2) (1) ,x(n-1) (1) ,x(n) (1) They are {x(t) (1) The data at the 1st, 2nd, n-1st, and nth time series in |t∈[1,n]}, x(2) and x(n) are the data at the 2nd and nth time series in {x(t)|t∈[1,n]} respectively; The differential equation of the Verhulst grey model is Solve the maximum value and get the time when the landslide occurred where tv is the landslide occurrence time output by the Verhulst grey model, t0 is the monitoring time of the radar data at the first time sequence of the radar data sequence, and x(1) is the data at the first time sequence in {x(t)|t∈[1,n]}.

6. The mine slope landslide disaster early warning method based on multimodal data according to claim 5, characterized in that: The regional error correction model is: Pfixed = Add(Unet(Pr, Pw), Pr), where Pfixed is the corrected landslide occurrence area, Unet is the Unet network structure, Add is the pixel layer addition operation, Pr is the landslide occurrence area output by the radar data landslide prediction model, and Pw is the landslide occurrence area output by the meteorological data landslide prediction model; The time error correction model is tfixed=wr*tr+(1-wr)*tv, where where tfixed is the corrected landslide occurrence time, wr is the weight of tr, tr is the landslide occurrence time output by the radar data landslide prediction model, tv is the landslide occurrence time output by the Verhulst grey model, tGT is the true value of the landslide occurrence time, MSE(tr,tGT) is the mean square error between tr and tGT, and MSE(tv,tGT) is the mean square error between tv and tGT.

7. A mine slope landslide disaster early warning system based on multimodal data, characterized in that: A method for warning mine slope landslide disasters based on multimodal data as described in any one of claims 1 to 6, the system comprising: Radar equipment is used to monitor open-pit mine slopes and obtain radar data sequences corresponding to slope rockfall and loosening of rock and soil; Meteorological equipment for monitoring meteorological data on open-pit mine slopes; The cloud server is used to call a pre-established slope disaster warning model and use the radar data sequence and meteorological data sequence to predict the landslide occurrence area and landslide occurrence time of the open-pit mine slope.

8. The mine slope landslide disaster early warning system based on multimodal data according to claim 7, characterized in that: The slope disaster warning model includes a landslide prediction structure, a Verhulst grey model and an error correction structure; Wherein, the landslide prediction structure includes a radar data landslide prediction model and a meteorological data landslide early warning model; The error correction structure includes a regional error correction model and a temporal error correction model; The output ends of the radar data landslide prediction model and the meteorological data landslide early warning model are connected to the input end of the regional error correction model; The output ends of the radar data landslide prediction model and the Verhulst grey model are connected to the input end of the time error correction model.

9. The mine slope landslide disaster early warning system based on multimodal data according to claim 8, characterized in that: The method for constructing the radar data landslide prediction model includes: Obtain radar data series and meteorological data series of multiple slopes where landslide disasters occurred; Extract the time points of rockfall and soil loosening on the slopes, as well as the time points and areas of landslide occurrence, from the radar data series of the slopes where landslide disasters occurred; In the radar data sequence of the slope where the landslide occurred, a time sequence following the time point when the slope rockfall and rock and soil loosening occurred was intercepted as the input of the CNN network, and the time point and area where the landslide occurred were used as the output of the CNN network. The CNN network was trained to obtain the radar data landslide prediction model.

10. The mine slope landslide disaster early warning system based on multimodal data according to claim 9, characterized in that: The method for constructing the meteorological data landslide early warning model includes: A time sequence subsequent to the time point when slope rockfall and rock and soil loosening occurred is intercepted from the meteorological data sequence as the input of the CNN neural network, and the area where the landslide occurred is used as the output of the CNN network. The CNN network is trained to obtain the meteorological data landslide prediction model.

Images