Landslide prediction method based on digital twinborn technology

Abstract

The invention discloses a landslide prediction method based on a digital twinborn technology. The method comprises the following prediction steps: establishment of a landslide earth surface three-dimensional model: setting a flight path to ensure that the whole area is covered, utilizing an unmanned aerial vehicle to be equipped with an aerial camera to fly according to a specified planned route, shooting area images at multiple angles and at multiple heights, and utilizing software to generate a three-dimensional terrain model; establishing a landslide geological three-dimensional model: collecting and preprocessing drilling data, importing the preprocessed drilling data into three-dimensional geological modeling software, setting a grid range and grid precision, obtaining a lithologic model, and smoothing the lithologic model to obtain a landslide deep three-dimensional model; coupling the landslide surface three-dimensional model and the landslide deep three-dimensional model to generate a digital twinborn model; selecting a plurality of factors influencing landslide deformation, and performing correlation detection on the influence factors. According to the method, the prediction precision is improved, and important technical support is provided for promoting landslide prediction.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of landslide prediction methods, in particular to a landslide prediction method based on digital twinning technology BACKGROUND

[0002] Digital twinning is a dynamic mapping system between a physical entity and its virtual digital equivalent, emphasizing the synchronization optimization of the two through real-time data connection throughout the life cycle. Digital twinning technology has developed rapidly in the fields of aerospace, military industry, petroleum and chemical industry, etc. In recent years, in the field of smart cities, digital twinning technology has developed rapidly, realizes the collection of city all-element data based on remote sensing, sensor technology, constructs a city digital twinning body, and improves its accuracy by using big data, artificial intelligence, cloud computing, etc. Finally, it helps to realize the effective monitoring and management of traffic, warehousing logistics, water conservancy, environment, etc. Digital twinning technology has become a key force to promote the digital transformation of the water conservancy industry, and is widely used in water resources management, flood control and disaster reduction, and water conservancy engineering operation.

[0003] Landslides have gradually become one of the most serious and most difficult to prevent natural disasters due to their occurrence in a short time and suddenness, which seriously threatens people's life and property safety, so the importance and necessity of timely and accurate monitoring and early warning of landslides are self-evident. The traditional landslide monitoring and early warning method mainly relies on manual patrol, various single sensor monitoring methods, etc. Although certain results have been achieved, there are still great limitations in real-time performance, accuracy and reliability. Digital twinning technology, as a new technology emerging in recent years, has shown great potential and application value in various fields. How to integrate digital twinning technology into the field of landslide prediction and realize the "dangerous state in advance" in the digital world is a difficult problem that needs to be solved urgently.

[0004] Therefore, the application provides a landslide prediction method based on digital twinning technology. SUMMARY

[0005] The application aims to provide a landslide prediction method based on digital twinning technology, and the prediction step comprises:

[0006] Step 1: Establishment of a three-dimensional model of the landslide surface: set a flight trajectory to ensure coverage of the entire area, use a UAV equipped with a camera to fly according to the specified planning route, take images of the area from multiple angles and heights, and generate a three-dimensional terrain model using software;

[0007] The step 1 specifically comprises:

[0008] Step 11: Determine the flight trajectory according to the topographic features of the landslide, use a UAV equipped with a camera to take multiple shots at different heights and angles, and obtain multi-view terrain images;

[0009] Step 12: use image stitching software to stitch multiple images into a complete terrain image, and import the complete terrain image data into Pix4D three-dimensional modeling software;

[0010] Step 13: use the multi-view stereo matching algorithm of the three-dimensional modeling software to stitch the image data into a complete three-dimensional terrain model.

[0011] Step 2: Establishment of landslide geological three-dimensional model: collect and preprocess drilling data, import the preprocessed drilling data into three-dimensional geological modeling software, set grid range and grid accuracy, obtain lithology model, and perform smoothing processing on the lithology model to obtain a landslide deep three-dimensional model;

[0012] The step 2 specifically comprises:

[0013] Step 21: obtain drilling coordinates, drilling number, hole diameter, layer depth, layer lithology, and porosity, and convert the obtained data to PGF format after standardization processing;

[0014] Step 22: determine the boundary line of the target area, take the grid intersecting the boundary line as the grid to be interpolated, and obtain the functional relationship between stratum depth and drilling data according to the drilling coordinates;

[0015] Step 23: take the grid with drilling as the initial value, combine the functional relationship to obtain the stratum depth of several grids in the target area, and couple to obtain the lithology model.

[0016] Step 3: coupling the landslide surface three-dimensional model and the landslide deep three-dimensional model to generate a digital twin model:

[0017] Step 4: select multiple factors affecting landslide deformation, and perform correlation detection on the influencing factors to eliminate weakly correlated influencing factors;

[0018] The step 4 in.

[0019] The step 4 specifically comprises:

[0020] Step 41: select factors affecting landslide displacement including annual average rainfall, reservoir water level fluctuation rate, tangent angle, groundwater flow rate, displacement amount, and number of landslide cracks, form a data set and put it into excel, and use SPSS software to perform normalization processing on the data set;

[0021] Step 42: after normalization processing, use SPSS software to perform correlation analysis on the data set to obtain correlation results, select influencing factors with a correlation coefficient greater than 0.3, and eliminate influencing factors with a correlation coefficient less than 0.3.

[0022] Step 5: regression fitting of each influence factor is carried out by using a multivariate time series model, each influence factor estimation sequence is generated, then historical data and the obtained each influence factor estimation sequence are combined to form complete input data, the complete input data is input into a prediction model, and scene parameters are set, and a final estimation sequence of the target object in the target period is obtained;

[0023] The step 5 specifically comprises:

[0024] Step 51: the annual average rainfall, the reservoir water level rising rate, the tangent angle, the underground water flow rate, the displacement amount and the landslide crack quantity data are successively introduced into a prophet model for regression fitting to realize data expansion, and corresponding time sequences containing historical and to-be-predicted time periods are respectively generated;

[0025] Step 52: the length of a sliding window is predefined, a certain selected influence factor time sequence, including historical data of the influence factor and the generated influence factor estimation sequence, passes through the sliding window to derive additional influence factors;

[0026] Step 53: the obtained historical data of the deformation variable, the historical data of the deformation variable influence factor, the influence factor estimation sequence and the derived additional influence factor sequence are combined to form complete input data.

[0027] Step 54: the influence factor data is taken as a data set, a prediction model is trained by using the data set, and a prediction value is obtained.

[0028] Step 6: the multivariate time series model of the final estimation sequence is fused with a digital twin model to obtain a dynamic digital twin model, and landslide prediction is realized through monitoring data of each sensor on the landslide.

[0029] Compared with the prior art, the beneficial effects of the present application are as follows: the present application respectively establishes a landslide surface three-dimensional model and a landslide geological three-dimensional model, couples the landslide surface three-dimensional model and the landslide geological three-dimensional model to obtain a digital twin model, selects factors influencing landslide deformation, carries out correlation screening on the influence factors, removes weakly correlated factors, sends the removed factors into a multivariate time series, carries out regression fitting on each influence factor, constructs an estimation sequence of multiple influence factors, outputs a prediction value, combines the data with the digital twin model to obtain a dynamic digital twin model, realizes landslide prediction, improves prediction accuracy, and provides important technical support for promoting landslide prediction. BRIEF DESCRIPTION OF DRAWINGS

[0030] Figure 1 It is a prediction method flowchart of the present application. DETAILED DESCRIPTION

[0031] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative effort belong to the scope of the present application.

[0032] As shown in the accompanying drawings Figure 1 The present application provides a landslide prediction method based on digital twin technology, characterized by the following prediction steps.

[0033] Step 1: Establishment of a three-dimensional model of the landslide surface: Set the flight trajectory to ensure coverage of the entire area, use a drone equipped with a camera to fly according to the specified planning route, take multiple-angle and multiple-height images of the area, and use software to generate a three-dimensional terrain model.

[0034] The step 1 specifically includes:

[0035] Step 11: Determine the flight trajectory according to the topographic features of the landslide, use a drone equipped with a camera to take multiple images at different heights and angles, and obtain multi-view terrain images;

[0036] Step 12: Use image stitching software to stitch multiple images into a complete terrain image, and import the complete terrain image data into Pix4D three-dimensional modeling software;

[0037] Step 13: Use the multi-view stereo matching algorithm of the three-dimensional modeling software to stitch the image data into a complete three-dimensional terrain model.

[0038] Step 2: Establishment of a three-dimensional model of the landslide geology: Collect and preprocess the drilling data, import the preprocessed drilling data into three-dimensional geological modeling software, set the grid range and grid accuracy, obtain the lithology model, and perform smoothing processing on the lithology model to obtain a three-dimensional model of the landslide deep part.

[0039] The step 2 specifically includes:

[0040] Step 21: Obtain drilling coordinates, drilling numbers, hole diameters, layer depths, layer lithologies, and porosity data, make a data set, and convert the obtained data into PGF format after standardization processing, the standardization formula is:

[0041]

[0042] Where x is the value of a single data point, μ is the average value of the data set, and σ is the standard deviation of the data set.

[0043] Step 22: determine the boundary line of the target area, take the grid intersecting with the boundary line as the grid to be interpolated, and obtain the function relationship between the stratum depth and the drilling data according to the drilling coordinates;

[0044] Step 23: taking the grid provided with the drilling as the initial value, combining the function relationship to obtain the stratum depth of several grids in the target area, and coupling to obtain the lithology model.

[0045] Step 3: coupling the landslide surface three-dimensional model and the landslide deep three-dimensional model to generate a digital twin model:

[0046] Step 4: selecting a plurality of factors affecting the deformation of the landslide, detecting the correlation of the influencing factors, and eliminating the influencing factors with weak correlation, wherein the factors affecting the displacement of the landslide include annual average rainfall, reservoir water level fluctuation rate, tangent angle, groundwater flow rate, displacement amount, and number of landslide cracks;

[0047] The step 4 specifically includes:

[0048] Step 41: selecting the data set affecting the deformation of the landslide and putting it into excel, and using SPSS software to normalize the data set;

[0049] Step 42: after normalization, using SPSS software to analyze the correlation of the data set, obtaining the correlation result, selecting the influencing factors with a correlation coefficient greater than 0.3, and eliminating the influencing factors with a correlation coefficient less than 0.3.

[0050] Step 5: using a multivariate time series model to regress and fit each influencing factor to generate a predicted sequence of each influencing factor, then combining the historical data and the predicted sequence of each influencing factor to form complete input data, inputting the complete input data into the prediction model, setting the scene parameters, and obtaining the final predicted sequence of the target object in the target period;

[0051] Step 5 specifically includes:

[0052] Step 51: since the prophet regression model needs to provide both historical data of the influencing factor sequence and data of the time period to be predicted, for the influencing factor data lacking the time period to be predicted, prophet model needs to be called additionally for fitting in advance;

[0053] The annual average rainfall, reservoir water level fluctuation rate, tangent angle, groundwater flow rate, displacement amount, and number of landslide cracks data are imported into the prophet model for regression fitting to realize data expansion, and the corresponding time series containing historical and predicted time periods are generated respectively;

[0054] Step 52: the length of the sliding window is predefined, and a certain selected influence factor time series, including historical data of the influence factor and the generated influence factor prediction sequence, passes through the sliding window to derive additional influence factors. For example, if the window length is set to 30d, the smoothing calculation is performed at a time granularity of 30d. When the first 30 time series values are input, the sequence from 1-30d is used for smoothing calculation to obtain a value. The subsequent input data continues to pass through the window, and in turn, the sequence from 2-31d, 3-32d, 4-33d, and so on is used for smoothing calculation. These values constitute the derived additional influence factor sequence.

[0055] The smoothing calculation method can be any one of accumulation, average, maximum, minimum, or finding the intermediate value.

[0056] Step 53: the obtained historical data of the deformation variable, historical data of the deformation variable influence factor, and the derived additional influence factor sequence are combined to form complete input data.

[0057] Step 54: the complete input data contains multiple time series, forming a high-dimensional vector X. The high-dimensional vector is used as the independent variable, and the target to be predicted is the target function f(X). In this way, f(X) can be regarded as a prediction model, and the model parameters X are known. The influence factor data is used as the data set. The prediction model, i.e., the parameter X of the target function f(X), is converted into a model training problem using the data set. The gradient gn and Hessian matrix Hn are calculated, and according to the formula:

[0058]

[0059] The optimal model parameters X are obtained, where a is the calculation step size, x n is the model parameter vector. After obtaining the optimal model parameters, the target value to be predicted can be calculated from the value of the influence factor in the prediction period.

[0060] Step 6: the multivariate time series model of the final prediction sequence is fused with the digital twin model to obtain a dynamic digital twin model. Through the monitoring data of various sensors on the landslide, the landslide prediction is realized.

[0061] The technical solutions described in the present application or inspired by the technical solutions of the present application by those skilled in the art can be used to design similar technical solutions to achieve the above technical effects, which are within the scope of protection of the present application.

Claims

1. A landslide prediction method based on digital twin technology, characterized in that: The prediction steps include: Step 1: Establishing a 3D model of the landslide surface: Set the flight trajectory to ensure coverage of the entire area, use a drone equipped with an aerial camera to fly along the designated route, take images of the area from multiple angles and heights, and use software to generate a 3D terrain model; Step 2: Establishment of the 3D geological model of the landslide: Collect borehole data and preprocess it. Import the preprocessed borehole data into the 3D geological modeling software, set the grid range and grid accuracy, obtain the lithology model, smooth the lithology model, and obtain the deep 3D model of the landslide. Step 3: Couple the 3D surface model and the 3D deep model of the landslide to generate a digital twin model: Step 4: Select multiple factors that affect landslide deformation, perform correlation tests on the influencing factors, and eliminate influencing factors with weak correlation. Step 5: Use a multivariate time series model to perform regression fitting on each influencing factor to generate the predicted sequence of each influencing factor. Then, combine the historical data and the obtained predicted sequence of each influencing factor to form complete input data. Input the complete input data into the prediction model and set the scenario parameters to obtain the final predicted sequence of the target object in the target time period. Step 6: Fuse the multivariate time series model of the final predicted sequence with the digital twin model to obtain a dynamic digital twin model, and realize landslide prediction by monitoring data from various sensors on the landslide.

2. The landslide prediction method based on digital twin technology according to claim 1, characterized in that: Step 1 specifically includes: Step 11: Determine the flight trajectory based on the terrain features of the landslide, and use a drone equipped with an aerial camera to take multiple shots at different heights and angles to obtain multi-view terrain images; Step 12: Use image stitching software to stitch multiple images into a complete terrain image, and import the complete terrain image data into Pix4D 3D modeling software; Step 13: Use the multi-view stereo matching algorithm of 3D modeling software to stitch the image data into a complete 3D terrain model.

3. The landslide prediction method based on digital twin technology according to claim 1, characterized in that: Step 2 specifically includes: Step 21: Obtain borehole coordinates, borehole number, borehole diameter, depth of each layer, lithology of each layer, and porosity. Standardize the obtained data and convert it to PGF format. Step 22: Determine the boundary line of the target area, use the grid that intersects with the boundary line as the grid to be interpolated, and obtain the functional relationship between the formation depth and the borehole data based on the borehole coordinates; Step 23: Using the grid with boreholes as the initial value, the formation depth of several grids in the target area is obtained by combining the function relationship inversion, and then coupled to obtain the lithology model.

4. The landslide prediction method based on digital twin technology according to claim 1, characterized in that: The factors affecting the sliding displacement in step 4 include the annual average rainfall, the rate of rise and fall of the reservoir water level, the tangent angle, the groundwater flow rate, the displacement, and the number of landslide cracks.

5. The landslide prediction method based on digital twin technology according to claim 1, characterized in that: Step 4 specifically includes: Step 41: Select the dataset that affects landslide deformation and put it into an Excel file. Then, use SPSS software to normalize the dataset. Step 42: After normalization, use SPSS software to perform correlation analysis on the dataset to obtain the correlation results and remove weakly correlated influencing factors.

6. The landslide prediction method based on digital twin technology according to claim 5, characterized in that: The correlation results selected the influencing factors with a correlation coefficient greater than 0.3 and removed the influencing factors with a correlation coefficient less than 0.

3.

7. The landslide prediction method based on digital twin technology according to claim 1, characterized in that: Step 5 specifically includes: Step 51: Import the data of annual average rainfall, reservoir water level rise and fall rate, tangent angle, groundwater flow rate, displacement, and number of landslide cracks into the Prophet model for regression fitting to expand the data and generate corresponding time series containing historical and predicted time periods respectively. Step 52: Predefine the length of the sliding window, and let a selected impact factor time series, including the historical data of the impact factor and the generated impact factor prediction series, pass through the sliding window to derive additional impact factors; Step 53: Combine the obtained historical data of deformation variables, historical data of deformation variable impact factors, and the predicted impact factor sequence and derived additional impact factor sequence to form complete input data. Step 54: Use the impact factor data as a dataset, train the prediction model using the dataset, and obtain the predicted values.

Images