Collapsible loess area rainfall geological disaster early warning method and device and readable medium

Abstract

The invention discloses a collapsible loess area rainfall geological disaster early warning method and device and a readable medium, and belongs to the technical field of geological disaster early warning, and the method comprises the steps: firstly, interpreting a suspected landslide area through the surface deformation rate result of a collapsible loess area in combination with an optical image and topographic data; for the suspected landslide area, automatically detecting an accelerated deformation stage through a deep learning time sequence analysis model; according to the rainfall information corresponding to the accelerated deformation stage, the number of earlier-stage rainfall days and an earlier-stage effective rainfall attenuation coefficient are determined; and finally, determining the effective rainfall capacity in the collapsible loess area, establishing E-D, I-D and E-I rainfall threshold value models by combining a power function, and constructing a rainfall type geological disaster early warning model according with the characteristics of the collapsible loess area. According to the method, the multi-temporal InSAR technology and the deep learning time sequence analysis model are combined, and the rainfall information is introduced, so that accurate early warning of the rainfall-type geological disaster in the collapsible loess area is realized, and the capability of preventing the rainfall-type geological disaster is remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of geological disaster early warning, and in particular to a method, a device and a readable medium for early warning of geological disasters caused by rainfall in collapsible loess areas. Background Art

[0002] Collapsible loess regions are prone to geological disasters such as landslides under the influence of factors such as rainfall, posing a serious threat to people's lives and property. Traditional geological disaster early warning methods rely on susceptibility assessments or rainfall data. However, these methods suffer from inaccurate detection of accelerated deformation stages and difficulty in integrating multiple factors into early warning models. Consequently, they cannot meet the demand for accurate early warning of rainfall-induced geological disasters in collapsible loess regions. Summary of the Invention

[0003] In view of this, the present invention addresses the shortcomings of the existing technology and provides a rainfall geological disaster warning method, device and readable medium for collapsible loess areas. By combining multi-phase InSAR technology with a deep learning time series analysis model and introducing rainfall information, a complete rainfall geological disaster warning model for collapsible loess areas is constructed, significantly improving the ability to prevent rainfall geological disasters.

[0004] To solve the above technical problems, the technical solution adopted by the present invention is: a method for early warning of rainfall geological disasters in collapsible loess areas, comprising the following steps:

[0005] S1. Use multi-temporal InSAR technology to obtain surface deformation rate data within the collapsible loess area, and combine optical images and topographic data to interpret the suspected landslide area;

[0006] S2. Extracting time-series deformation data based on the surface deformation rate data of the suspected landslide area, and using a deep learning model to interpolate and filter the time-series deformation data, and analyzing the accelerated deformation stage in the time-series deformation curve;

[0007] S3. Determine the number of rainfall days in the early stage and the effective rainfall attenuation coefficient in the early stage based on the rainfall data corresponding to the accelerated deformation stage;

[0008] S4. Determine the calculation method of effective rainfall in collapsible loess areas, and establish rainfall threshold models of rainfall and deformation, rainfall intensity and deformation, and rainfall and rainfall intensity, and construct a rainfall geological disaster early warning model in collapsible loess areas.

[0009] Furthermore, in S1, first, multi-temporal InSAR technology was used to continuously monitor the collapsible loess area to obtain high-precision surface deformation rate data; then, combined with optical image data, the surface deformation area was identified through image interpretation technology; finally, the digital elevation model was used to analyze the topographic characteristics of the surface deformation area to determine the location and scope of the suspected landslide area.

[0010] Furthermore, in S2, first, the long short-term memory network (LSTM) is used to interpolate and filter the temporal deformation data to remove noise and smooth the curve; then, the processed temporal deformation curve is analyzed to identify the starting and ending points of the accelerated deformation stage; finally, according to the characteristics of the accelerated deformation stage, the corresponding deformation rate and time series information are extracted.

[0011] Furthermore, in S3, rainfall data is obtained through a meteorological station in the target area or satellite remote sensing, and the rainfall data includes rainfall amount, rainfall intensity, and rainfall duration;

[0012] The rainfall data is cleaned and normalized to remove outliers and noise; the normalization formula is as follows:

[0013]

[0014] Among them, R is the original data, R min and R max are the minimum and maximum values ​​of the data series respectively;

[0015] The method for determining the number of days with rainfall in the previous period is:

[0016] (1) By using the multivariate time series analysis method and combining the deformation data and rainfall data monitored by InSAR, a statistical model between rainfall and deformation was established, and a multivariate linear regression model was used:

[0017] ln(y t )=α+β1ln(x t―1 )+β2ln(x t―2 )+...+β n ln(x t―n )+∈ t

[0018] Among them, y t is the deformation at time point t, x t―i is the rainfall at time point t-i, α is the intercept term, β i is the regression coefficient, ∈ t is the error term;

[0019] (2) Through stepwise regression analysis, the number of days with antecedent rainfall that is significantly correlated with deformation is screened out;

[0020] When the R 2 The value reaches the set value, at this time, the number of rainfall days in the early stage is determined;

[0021] The method for determining the attenuation coefficient of the previous effective rainfall is:

[0022] (a) Based on the determined number of previous rainfall days, a rainfall attenuation model is established:

[0023] R eff (t) = R(t)·e ―λt

[0024] Among them, R eff (t) is the effective rainfall at time point t, time point t is the number of days of rainfall in the previous period, R(t) is the actual rainfall, and λ is the attenuation coefficient;

[0025] (b) Estimate the attenuation coefficient λ by the least squares method;

[0026] Construct time series t, actual rainfall R(t) and observed deformation D obs The linear relationship between:

[0027]

[0028] make The above formula is simplified to:

[0029] y≈―λt

[0030] Minimize according to the least squares rule:

[0031]

[0032] The attenuation coefficient λ can be calculated by taking the derivative of the loss function MSE and setting the derivative to zero.

[0033] Furthermore, in S4, the calculation method of effective rainfall in collapsible loess areas is determined by combining historical rainfall data and geological characteristics:

[0034] R eff (t) = R(t)·e ―λt ·(1―e ―μt )

[0035] Where: R eff (t) is the effective rainfall at time point t; R(t) is the actual rainfall at time point t; λ is the rainfall attenuation coefficient, which reflects the attenuation of rainfall intensity over time; μ is the permeability coefficient, which reflects the speed at which rainfall penetrates into the loess layer;

[0036] Rainfall threshold model of rainfall and deformation: A power function model is used to establish the relationship between effective rainfall and deformation. The model formula is as follows:

[0037]

[0038] Where D is the deformation; R eff is the effective rainfall; a and b are model parameters, determined by fitting historical data;

[0039] Rainfall threshold model of rainfall intensity and deformation: A power function model is used to establish the relationship between rainfall intensity and deformation. The model formula is as follows:

[0040] D=c·I d

[0041] Where I is the rainfall intensity; c and d are model parameters, which are determined by fitting historical data.

[0042] Rainfall threshold model of rainfall amount and rainfall intensity: A power function model is used to establish the relationship between effective rainfall amount and rainfall intensity. The model formula is as follows:

[0043]

[0044] Among them, e and f are model parameters, which are determined by fitting historical data;

[0045] The method for constructing a rainfall geological disaster early warning model for collapsible loess areas is as follows:

[0046] Step 1: Rainfall threshold model fitting and validation

[0047] Historical rainfall data and corresponding deformation data were collected as samples for model fitting and verification. The data included rainfall amount, rainfall intensity, and deformation variables. The parameters of the rainfall threshold model were fitted using the least squares method. The fitted model was validated using some data that was not used in model training to evaluate the model's prediction accuracy and generalization ability.

[0048] Step 2: Determine the parameters of the rainfall threshold model based on the fitting and verification results;

[0049] Step 3: The rainfall threshold model of rainfall amount and deformation, rainfall intensity and deformation, and rainfall amount and rainfall intensity after parameter determination is used as a rainfall geological disaster warning model for collapsible loess areas.

[0050] Another object of the present invention is to provide an early warning device for rainfall geological disasters in collapsible loess areas, comprising:

[0051] The acquisition and interpretation module is configured to acquire surface deformation rate data within the collapsible loess area using multi-temporal InSAR technology and interpret suspected landslide areas in combination with optical images and topographic data;

[0052] A preprocessing module is configured to extract time-series deformation data from the surface deformation rate data of the suspected landslide area, interpolate and filter the time-series deformation data using a deep learning model, and analyze the accelerated deformation stage in the time-series deformation curve;

[0053] an analysis module configured to determine the number of previous rainfall days and the previous effective rainfall attenuation coefficient based on rainfall data corresponding to the accelerated deformation stage;

[0054] The model building module is configured to determine the effective rainfall calculation method that is suitable for the collapsible loess area, and to establish rainfall threshold models of rainfall and deformation, rainfall intensity and deformation, and rainfall and rainfall intensity, so as to construct a rainfall geological disaster early warning model for the collapsible loess area.

[0055] Another object of the present invention is to provide a computer-readable medium storing a computer program, which, when executed, implements the above-mentioned method for early warning of rainfall geological disasters in collapsible loess areas.

[0056] Another object of the present invention is to provide a computer device comprising a processor storing a program, which, when executed by the processor, implements the above-mentioned method for warning of rainfall geological disasters in collapsible loess areas.

[0057] Compared with the prior art, the present invention has the following beneficial effects:

[0058] This invention combines multi-temporal InSAR technology with a deep learning time series analysis model and incorporates rainfall information to construct a comprehensive early warning model for rainfall-induced geological hazards in collapsible loess regions. Its advantages are as follows: First, it uses multi-temporal InSAR technology to acquire high-precision surface deformation data, combined with a deep learning time series analysis model to automatically detect accelerated deformation stages, improving the ability to identify potential landslide areas. Second, it comprehensively considers rainfall information, determines the number of days of previous rainfall and the previous effective rainfall attenuation coefficient, and uses power functions to establish rainfall threshold models for ED, ID, and EI. This enables accurate early warning of rainfall-induced geological hazards in collapsible loess regions, significantly improving the ability to prevent rainfall-induced geological hazards. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 It is a flow chart of a method according to an embodiment of the present invention. DETAILED DESCRIPTION

[0060] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions of the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the described embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field are within the scope of protection of the present invention.

[0061] Example 1: Figure 1 As shown, a method for early warning of rainfall geological disasters in collapsible loess areas includes the following steps:

[0062] S1. Use multi-temporal InSAR technology to obtain surface deformation rate data within the collapsible loess area, and combine optical images and topographic data to interpret the suspected landslide area;

[0063] S2. Extracting time-series deformation data (deformation displacement values ​​in a time series) from the surface deformation rate data of the suspected landslide area, interpolating and filtering the time-series deformation data using a deep learning model, and analyzing the accelerated deformation stage in the time-series deformation curve;

[0064] S3. Determine the number of rainfall days in the early stage and the effective rainfall attenuation coefficient in the early stage based on the rainfall data corresponding to the accelerated deformation stage;

[0065] S4. Determine the calculation method of effective rainfall in collapsible loess areas, and establish rainfall threshold models of rainfall and deformation, rainfall intensity and deformation, and rainfall and rainfall intensity, and construct a rainfall geological disaster early warning model in collapsible loess areas.

[0066] Specifically, in step S1, first, the collapsible loess area is continuously monitored using multi-temporal InSAR technology to obtain high-precision surface deformation rate data; then, combined with optical image data, the surface deformation area is identified through image interpretation technology; finally, the digital elevation model (DEM) is used to analyze the topographic characteristics of the surface deformation area to determine the location and scope of the suspected landslide area.

[0067] In step S2, first, the time series deformation data is normalized to convert the deformation displacement values ​​into a uniform range (such as [0, 1]) for subsequent processing. The normalization formula is as follows:

[0068]

[0069] Among them, x is the original deformation displacement value, x min and x max are the minimum and maximum values ​​of the data sequence respectively; then, the long short-term memory network LSTM is used to interpolate and filter the time series deformation data to remove noise and smooth the curve; then, the processed time series deformation curve is analyzed to identify the starting and ending points of the accelerated deformation stage; finally, according to the characteristics of the accelerated deformation stage, the corresponding deformation rate and time series information are extracted.

[0070] A long short-term memory (LSTM) network is used as a deep learning model. LSTM is a variant of recurrent neural networks (RNNs) suitable for processing time series data, effectively capturing long-term dependencies in time series. When building an LSTM model, the input layer receives normalized time series deformed data, the hidden layer contains multiple LSTM units, and the output layer outputs the processed time series data.

[0071] Train the LSTM model using historical time series deformation data. During training, optimize model parameters by minimizing the mean squared error (MSE) between predicted and actual values. The training dataset should contain multiple examples from known accelerated deformation phases so that the model can learn the characteristics of accelerated deformation.

[0072] The normalized time series deformation data is fed into the trained LSTM model, which then outputs a time series curve that has undergone interpolation and filtering. Interpolation is used to fill in missing data, while filtering is used to remove noise and smooth the curve, thereby improving data quality and usability.

[0073] Extract key features from the time series deformation curve after LSTM processing, including the rate of change of deformation rate and deformation acceleration. The rate of change of deformation rate can be obtained by calculating the difference in deformation rate between adjacent time points, and the deformation acceleration can be calculated through quadratic difference.

[0074] Thresholds are set to identify the start and end points of the accelerated deformation phase. For example, when the rate of change of the deformation rate exceeds a certain threshold, the accelerated deformation phase is considered to have begun; when the rate of change of the deformation rate falls below another threshold, the accelerated deformation phase is considered to have ended. The specific thresholds can be adjusted based on historical data and experience.

[0075] Verification and Optimization: We used some data not used in model training to verify the detection results during the accelerated deformation phase and evaluate the accuracy and reliability of the model. Based on the verification results, we optimized the parameters and thresholds of the LSTM model to improve the accuracy of detection during the accelerated deformation phase.

[0076] For the identified accelerated deformation stage, the average deformation rate within that stage is calculated. The deformation rate can be calculated by taking the difference of the deformation displacement values ​​in the time series.

[0077] Record the time range of the accelerated deformation phase, including the start and end time points. At the same time, extract the time series data within this phase, including the displacement value and deformation rate value at each time point, for subsequent analysis and modeling.

[0078] In step S3, rainfall data, including rainfall amount, rainfall intensity, and rainfall duration, is acquired from meteorological stations or satellite remote sensing in the target area. Statistical analysis is then used to determine the number of days of pre-accelerated rainfall associated with the accelerated deformation phase. Finally, the pre-accelerated effective rainfall attenuation coefficient is calculated based on correlation analysis between the rainfall and deformation data.

[0079] First, the rainfall data is cleaned and normalized to remove outliers and noise; the normalization formula is as follows:

[0080]

[0081] Among them, R is the original data, R min and R max are the minimum and maximum values ​​of the data series respectively;

[0082] The method for determining the number of days with rainfall in the previous period is:

[0083] (1) By using the multivariate time series analysis method and combining the deformation data and rainfall data monitored by InSAR, a statistical model between rainfall and deformation was established, and a multivariate linear regression model was used:

[0084] ln(y t )=α+β1ln(x t―1 )+β2ln(x t―2 )+...+β n ln(x t―n )+∈ t

[0085] Among them, y t is the deformation at time point t, x t―i is the rainfall at time point t-i, α is the intercept term, β i is the regression coefficient, ∈ t is the error term;

[0086] (2) Through stepwise regression analysis, the number of days with antecedent rainfall that is significantly correlated with deformation is screened out;

[0087] When the R 2 The value reaches the set value, such as 0.8 (R 2 As the proportion of the variability of the target variable explained by the regression model, when R 2 When the value reaches 0.8, it indicates that the hysteresis rainfall term (x t-1 ,x t-2 ,…,x t-n ) together explain the shape variable y t About 80% of the variation is considered reasonable), at this time, the number of days of rainfall in the early stage is determined;

[0088] The method for determining the attenuation coefficient of the previous effective rainfall is:

[0089] (a) Based on the determined number of rainfall days in the previous period, a rainfall attenuation model is established, using an exponential attenuation model:

[0090] R eff (t) = R(t)·e ―λt

[0091] Among them, R eff(t) is the effective rainfall at time point t, time point t is the number of days of rainfall in the previous period, R(t) is the actual rainfall, and λ is the attenuation coefficient;

[0092] (b) Estimate the attenuation coefficient λ by the least squares method;

[0093] Construct time series t, actual rainfall R(t) and observed deformation D obs The linear relationship between:

[0094]

[0095] make The above formula is simplified to:

[0096] y≈―λt

[0097] Minimize according to the least squares rule:

[0098]

[0099] The attenuation coefficient λ can be calculated by taking the derivative of the loss function MSE and setting the derivative to zero.

[0100] (c) Model validation

[0101] The rainfall attenuation model is validated using data that was not used in model training to evaluate the model's prediction accuracy and generalization ability. Common validation metrics include mean square error and coefficient of determination.

[0102] In step S4, the response characteristics of collapsible loess to rainfall are analyzed based on the geological characteristics of the area. Collapsible loess has high porosity and permeability, allowing rainfall to easily penetrate and cause surface deformation. Therefore, the penetration depth and time delay effect of rainfall need to be considered.

[0103] Combining historical rainfall data and geological characteristics, the calculation method of effective rainfall in collapsible loess areas is determined:

[0104] R eff (t) = R(t)·e ―λt ·(1―e ―μt )

[0105] Where: R eff (t) is the effective rainfall at time point t; R(t) is the actual rainfall at time point t; λ is the rainfall attenuation coefficient, which reflects the attenuation of rainfall intensity over time; μ is the permeability coefficient, which reflects the speed at which rainfall penetrates into the loess layer;

[0106] Rainfall threshold model (ED) of rainfall and deformation: A power function model is used to establish the relationship between effective rainfall and deformation. The model formula is as follows:

[0107]

[0108] Where D is the deformation; R eff is the effective rainfall; a and b are model parameters, determined by fitting historical data;

[0109] Rainfall threshold model (ID) of rainfall intensity and deformation: A power function model is used to establish the relationship between rainfall intensity and deformation. The model formula is as follows:

[0110] D=c·I d

[0111] Where I is the rainfall intensity; c and d are model parameters, which are determined by fitting historical data.

[0112] Rainfall threshold model (EI) of rainfall amount and rainfall intensity: A power function model is used to establish the relationship between effective rainfall amount and rainfall intensity. The model formula is as follows:

[0113]

[0114] Among them, e and f are model parameters, which are determined by fitting historical data.

[0115] The method for constructing a rainfall geological disaster early warning model for collapsible loess areas is as follows:

[0116] Step 1: Rainfall threshold model fitting and validation

[0117] Historical rainfall data and corresponding deformation data were collected as samples for model fitting and verification. The data included rainfall amount, rainfall intensity, and deformation variables. The parameters of the rainfall threshold model were fitted using the least squares method. The fitted model was validated using some data that was not used in model training to evaluate the model's prediction accuracy and generalization ability.

[0118] Step 2: Determine the parameters of the rainfall threshold model based on the fitting and verification results;

[0119] Step 3: The rainfall and deformation, rainfall intensity and deformation, and rainfall threshold models of rainfall and rainfall intensity after parameter determination are used as the rainfall geological disaster warning model for collapsible loess areas. That is, if the effective rainfall, rainfall intensity, and deformation variables meet the rainfall and deformation, rainfall intensity and deformation, and rainfall threshold models of rainfall and rainfall intensity, it is considered that a geological disaster may occur and an early warning is issued.

[0120] Example 2

[0121] An embodiment of the present invention provides an early warning device for rainfall geological disasters in collapsible loess areas, comprising:

[0122] The acquisition and interpretation module is configured to acquire surface deformation rate data within the collapsible loess area using multi-temporal InSAR technology and interpret suspected landslide areas in combination with optical images and topographic data;

[0123] A preprocessing module is configured to extract time-series deformation data from the surface deformation rate data of the suspected landslide area, interpolate and filter the time-series deformation data using a deep learning model, and analyze the accelerated deformation stage in the time-series deformation curve;

[0124] an analysis module configured to determine the number of previous rainfall days and the previous effective rainfall attenuation coefficient based on rainfall data corresponding to the accelerated deformation stage;

[0125] The model building module is configured to determine the effective rainfall calculation method that is suitable for the collapsible loess area, and to establish rainfall threshold models of rainfall and deformation, rainfall intensity and deformation, and rainfall and rainfall intensity, so as to construct a rainfall geological disaster early warning model for the collapsible loess area.

[0126] The specific processing methods of each module are as described in Example 1.

[0127] Example 3

[0128] An embodiment of the present invention provides a computer-readable medium storing a computer program. When the computer program is executed, the method for early warning of rainfall geological disasters in collapsible loess areas in Example 1 is implemented.

[0129] Example 4

[0130] An embodiment of the present invention provides a computer device, including a processor, wherein the processor stores a program. When the program is executed by the processor, the rainfall geological disaster warning method in the collapsible loess area in Example 1 is implemented.

[0131] The above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit the same. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the field can still modify or replace the specific implementation methods of the present invention with equivalents. Any modifications or equivalent replacements that do not depart from the spirit and scope of the present invention are within the scope of protection of the claims of the present invention.

Claims

1. A method for early warning of rainfall geological disasters in collapsible loess areas, characterized in that: The following steps are involved: S1. Use multi-temporal InSAR technology to obtain surface deformation rate data within the collapsible loess area, and combine optical images and topographic data to interpret the suspected landslide area; S2. Extracting time-series deformation data based on the surface deformation rate data of the suspected landslide area, and using a deep learning model to interpolate and filter the time-series deformation data, and analyzing the accelerated deformation stage in the time-series deformation curve; S3. Determine the number of rainfall days in the early stage and the effective rainfall attenuation coefficient in the early stage based on the rainfall data corresponding to the accelerated deformation stage; S4. Determine the calculation method of effective rainfall in collapsible loess areas, and establish rainfall threshold models of rainfall and deformation, rainfall intensity and deformation, and rainfall and rainfall intensity, and construct a rainfall geological disaster early warning model in collapsible loess areas.

2. The method for early warning of rainfall geological disasters in collapsible loess areas according to claim 1, characterized in that: In S1, first, multi-temporal InSAR technology is used to continuously monitor the collapsible loess area to obtain high-precision surface deformation rate data; then, combined with optical image data, surface deformation areas are identified through image interpretation technology; finally, digital elevation models are used to analyze the topographic characteristics of the surface deformation area and determine the location and extent of the suspected landslide area.

3. The method for early warning of geological disasters caused by rainfall in collapsible loess areas according to claim 1, characterized in that: In S2, first, the long short-term memory network (LSTM) is used to interpolate and filter the temporal deformation data to remove noise and smooth the curve; then, the processed temporal deformation curve is analyzed to identify the starting and ending points of the accelerated deformation stage; finally, based on the characteristics of the accelerated deformation stage, the corresponding deformation rate and time series information are extracted.

4. The method for early warning of rainfall geological disasters in collapsible loess areas according to claim 1, characterized in that: In S3, rainfall data are obtained through meteorological stations in the target area or satellite remote sensing. The rainfall data include rainfall amount, rainfall intensity and rainfall duration; Clean and normalize rainfall data to remove outliers and noise; The normalization formula is as follows: Among them, R is the original data, R min and R max are the minimum and maximum values ​​of the data series respectively; The method for determining the number of days with rainfall in the previous period is: (1) By using the multivariate time series analysis method and combining the deformation data and rainfall data monitored by InSAR, a statistical model between rainfall and deformation was established, and a multivariate linear regression model was used: ln(y t )=α+β1ln(x t―1 )+β2ln(x t―2 )+...+b n ln(x t―n )+∈ t Among them, y t is the deformation at time point t, x t―i is the rainfall at time point t-i, α is the intercept term, β i is the regression coefficient, ∈ t is the error term; (2) Through stepwise regression analysis, the number of days with antecedent rainfall that is significantly correlated with deformation is screened out; When the R 2 The value reaches the set value, at this time, the number of rainfall days in the early stage is determined; The method for determining the attenuation coefficient of the previous effective rainfall is: (a) Based on the determined number of previous rainfall days, a rainfall attenuation model is established: R eff (t)=R(t)·e ―λt Among them, R eff (t) is the effective rainfall at time point t, time point t is the number of days of rainfall in the previous period, R(t) is the actual rainfall, and λ is the attenuation coefficient; (b) Estimate the attenuation coefficient λ by the least squares method; Construct time series t, actual rainfall R(t) and observed deformation D obs The linear relationship between: make The above formula is simplified to: y≈―λt Minimize according to the least squares rule: The attenuation coefficient λ can be calculated by taking the derivative of the loss function MSE and setting the derivative to zero.

5. The method for early warning of rainfall geological disasters in collapsible loess areas according to claim 1, characterized in that: In S4, the calculation method of effective rainfall in collapsible loess areas is determined by combining historical rainfall data and geological characteristics: R eff (t)=R(t)·e ―λt ·(1―e ―μt ) Where: R eff (t) is the effective rainfall at time point t; R(t) is the actual rainfall at time point t; λ is the rainfall attenuation coefficient, which reflects the attenuation of rainfall intensity over time; μ is the permeability coefficient, which reflects the speed at which rainfall penetrates into the loess layer; Rainfall threshold model of rainfall and deformation: A power function model is used to establish the relationship between effective rainfall and deformation. The model formula is as follows: Where D is the deformation; R eff is the effective rainfall; a and b are model parameters, determined by fitting historical data; Rainfall threshold model of rainfall intensity and deformation: A power function model is used to establish the relationship between rainfall intensity and deformation. The model formula is as follows: D=c·I d Where I is the rainfall intensity; c and d are model parameters, which are determined by fitting historical data. Rainfall threshold model of rainfall amount and rainfall intensity: A power function model is used to establish the relationship between effective rainfall amount and rainfall intensity. The model formula is as follows: Among them, e and f are model parameters, which are determined by fitting historical data; The method for constructing a rainfall geological disaster early warning model for collapsible loess areas is as follows: Step 1: Rainfall threshold model fitting and validation Historical rainfall data and corresponding deformation data were collected as samples for model fitting and verification. The data included rainfall amount, rainfall intensity, and deformation variables. The parameters of the rainfall threshold model were fitted using the least squares method. The fitted model was validated using some data that was not used in model training to evaluate the model's prediction accuracy and generalization ability. Step 2: Determine the parameters of the rainfall threshold model based on the fitting and verification results; Step 3: The rainfall threshold model of rainfall amount and deformation, rainfall intensity and deformation, and rainfall amount and rainfall intensity after parameter determination is used as a rainfall geological disaster warning model for collapsible loess areas.

6. A rainfall geological disaster early warning device for collapsible loess areas, characterized by: include: The acquisition and interpretation module is configured to acquire surface deformation rate data within the collapsible loess area using multi-temporal InSAR technology and interpret suspected landslide areas in combination with optical images and topographic data; A preprocessing module is configured to extract time-series deformation data from the surface deformation rate data of the suspected landslide area, interpolate and filter the time-series deformation data using a deep learning model, and analyze the accelerated deformation stage in the time-series deformation curve; an analysis module configured to determine the number of previous rainfall days and the previous effective rainfall attenuation coefficient based on rainfall data corresponding to the accelerated deformation stage; The model building module is configured to determine the effective rainfall calculation method that is suitable for the collapsible loess area, and to establish rainfall threshold models of rainfall and deformation, rainfall intensity and deformation, and rainfall and rainfall intensity, so as to construct a rainfall geological disaster early warning model for the collapsible loess area.

7. A computer-readable medium storing a computer program, characterized in that: When the computer program is executed, the method for early warning of rainfall geological disasters in collapsible loess areas as described in any one of claims 1 to 5 is implemented.

8. A computer device comprising a processor storing a program, wherein: When the program is executed by the processor, the method for early warning of rainfall geological disasters in collapsible loess areas as described in any one of claims 1 to 5 is implemented.

Images