Reservoir landslide deformation prediction method based on InSAR technology and wavelet analysis

Abstract

The invention discloses a reservoir area landslide deformation prediction method based on an InSAR technology and wavelet analysis. The method comprises the following steps: S1, collecting a reservoir area image data set; s2, collecting reservoir water level and reservoir area rainfall data in the same time interval; s3, the accumulated deformation quantity of each period is obtained through processing in the step S1; s4, fitting by using a least square method; S6, analyzing the periodicity of displacement, reservoir water level and rainfall sequence by using continuous wavelet transform; step S7, using cross wavelet transform and wavelet coherence to extract hysteresis of deformation point displacement; step S8, the data are substituted into an XGBoost model for training, and a landslide prediction model is obtained; and carrying out landslide prediction by using the landslide prediction model. According to the reservoir area landslide deformation prediction method based on the InSAR technology and the wavelet analysis, the wavelet analysis and the XGBoost model are combined, the reservoir area landslide characteristics are fully utilized, and the reservoir area landslide displacement prediction precision can be remarkably improved.

Description

Technical Field

[0001] The present invention belongs to the field of geological technology, and relates to deformation prediction of water-related landslide geological disasters in reservoir areas, and specifically to a method for predicting landslide deformation in reservoir areas based on InSAR technology and wavelet analysis. Background Art

[0002] Landslides are a common geological disaster, typically occurring on steep slopes or unstable rock and soil. Landslides occur when factors such as rainfall, earthquakes, or human activity alter the stability of a slope. Landslides are highly sudden and destructive, especially in mountainous and hilly areas, posing a significant threat to life, property, and infrastructure. Landslides can not only destroy important infrastructure such as houses, roads, and bridges, but can also lead to severe secondary disasters such as mudslides and floods, further exacerbating the severity of the disaster. Landslide displacement is a key physical manifestation of a landslide and is often a precursor to an impending landslide. Therefore, accurately predicting changes in landslide displacement can provide timely and scientific decision-making for disaster prevention and emergency response.

[0003] Landslide displacement prediction in reservoir areas has always been a hot topic in the field of geological disaster research due to its fragile geological conditions, seasonal rainfall, human activities and large-scale reservoir storage. For landslide displacement with a hysteresis property, common time series prediction modeling often sets a time step and uses the data of the previous few periods to construct features to predict the current value. However, since it is impossible to determine the exact lag time, the constructed lag features may have deviations in describing the displacement changes; in particular, the influence of factors such as reservoir water level and rainfall on displacement usually has a certain lag effect, that is, the influence of changes in these external factors on landslide displacement may not appear until several periods later. In response to this situation, the present invention focuses on the lag of displacement to external factors. When constructing lag features using the lag time, external factor data that is ahead of the lag period of the displacement series is intentionally selected and applied to landslide displacement prediction. Summary of the Invention

[0004] In response to the defects of the above-mentioned existing technologies, the present invention discloses a reservoir area landslide deformation prediction method based on InSAR technology and wavelet analysis. By combining wavelet analysis and the XGBoost model, the exact time that the displacement time series lags behind external factors is extracted through wavelet analysis. A lag feature selection scheme is formulated based on the specific lag time, and the lag time is then provided to the XGBoost model to enhance the model's fitting ability.

[0005] The method for predicting landslide deformation in a reservoir area based on InSAR technology and wavelet analysis of the present invention comprises the following steps:

[0006] Step S1. Collecting a reservoir area SAR image dataset within a time interval TX acquired by a satellite; the reservoir area SAR image dataset is in the form of an n-period installment dataset with a time interval TA;

[0007] Step S2. Collect reservoir water level and reservoir area rainfall data within the same time interval TX;

[0008] The reservoir water level and rainfall data in the reservoir area are divided into periods according to the time interval TA described in step 1 to obtain the reservoir water level sequence l i and rainfall time series r i ; i=1,2,...n;

[0009] Step S3. The reservoir area SAR image data obtained in step S1 is processed by small baseline subset interferometric synthetic aperture radar (SBAS-InSAR) to obtain the cumulative deformation of each period; the period number and the corresponding original displacement of each period are t i and d i ; i=1,2,...n;

[0010] Step S4. Use the least squares method to fit,

[0011] For the t given in step S3 i and d i , assuming t i and d i The relationship between can be fitted by the linear function formula 1

[0012] d i =f(t i )=a0+*a1*t i ---Formula 1

[0013] Among them, a0 is the intercept, a1 is the slope, and the error square sum R1 is defined to represent the fitting error:

[0014]

[0015] When the R1 value is the smallest, the best fitting parameters a0 and a1 are obtained. i Substitute into formula 1 to calculate the displacement of the fitted trend term;

[0016] Reusing the original displacement d i Subtract the fitted trend term displacement to obtain the period term displacement d′ i =d i -(a0+a1·t i );

[0017] Step S6. Use continuous wavelet transform to analyze the periodicity of displacement, reservoir water level, and rainfall series. The continuous wavelet transform is defined as:

[0018]

[0019] where ψ is the sub-wavelet, ψ * is the complex conjugate of ψ, R represents a set of real numbers, τ is a translation parameter, and s is a scaling factor with a value greater than 0; f(t) is a function of displacement, reservoir water level, and rainfall relative to time, respectively, where the function of displacement relative to time adopts the periodic term displacement obtained in step S4, and the functions of reservoir water level and rainfall relative to time adopt the functions obtained in step S2;

[0020] The continuous wavelet values ​​W of displacement, reservoir water level, rainfall and other parameters can be obtained from formula 3: i (τ,s), where the subscript i represents different parameters such as displacement, reservoir level, or rainfall;

[0021] Step S7. Using cross wavelet transform (XWT) and wavelet coherence (WTC) to extract the hysteresis of deformation point displacement to reservoir water level and rainfall;

[0022] The cross wavelet transform W of two time series variables x and y xy (τ, s) is defined as:

[0023]

[0024] in It's W y The complex conjugate of W x (τ,s) represents the continuous wavelet value on the displacement sequence, W y (τ,s) represents the continuous wavelet value of external factors such as reservoir water level and rainfall;

[0025] By using formula 4 and substituting the continuous wavelet value obtained by formula 3, we can calculate the cross wavelet transform of the displacement sequence variable and the reservoir water level and rainfall sequence variables.

[0026] Using the calculation results of formula 4, calculate the wavelet coherence;

[0027] Wavelet coherence WTC is:

[0028]

[0029] In formula 5, S() represents the smoothing operator;

[0030] The cross wavelet transform and wavelet coherence obtained by formula 4 and formula 5 are used to generate the wavelet coherence spectrum;

[0031] The wavelet coherence spectrum was analyzed to obtain the hysteresis value of landslide occurrence;

[0032] Step S8. The predicted lag value obtained in step 7, the actual occurrence time of the landslide obtained in step 1, and the factors causing the landslide are brought into the XGBoost model for training to obtain a landslide prediction model; and the landslide prediction model is used to predict the landslide.

[0033] Preferably, between the steps S4 and S6, there is also a step S5. Using grey correlation degree to screen the influencing factors, the grey correlation degree ξ i The calculation formula of (t) can be expressed as:

[0034]

[0035] In formula 2: x0(t) is the reference sequence, i.e., the displacement sequence, x i (t) is the sequence to be compared, i.e., the reservoir water level or rainfall sequence; ρ is the resolution coefficient used to control the system discrimination, ρ∈[0,1]; |x0(t)=x i (t)| is called x0(t) and x at time t i The absolute difference of (t);

[0036] is the minimum difference between the two levels,

[0037] is the maximum difference between the two poles;

[0038] In this step, the displacement sequence x0(t) uses the periodic term displacement obtained in step 4;

[0039] The two factors of reservoir water level and rainfall data in the reservoir area obtained in step 2 are used as the sequences to be compared in formula 2 to calculate the grey correlation degree ξ. i (t), whether it is greater than the set grey correlation threshold, if so, the factor enters the subsequent step, otherwise it does not enter the subsequent step.

[0040] Preferably, the step S1 specifically includes:

[0041] Step S1.1. Preprocessing the data, including radiometric calibration, geometric correction, filtering, and registration;

[0042] Step S1.2 uses the pre-processed SAR image to perform interferometric processing to generate an initial interferometric phase map containing information such as terrain phase, deformation phase, and atmospheric phase;

[0043] The terrain phase is simulated using an external digital elevation model, and the terrain phase is subtracted from the initial interferometric phase map to eliminate the influence of the terrain phase and obtain a differential interferometric phase map containing information such as deformation phase and atmospheric phase.

[0044] Step S1.3. Perform phase unwrapping on the differential interferometer phase image to convert the wrapped phase value into the real phase value;

[0045] Step S1.4. Filter the unwrapped phase image to remove atmospheric noise and surface cover variation errors;

[0046] Step S1.5 converts the phase values ​​of the filtered differential interferometry phase image into distance values ​​to obtain the surface shape variables, thereby forming the reservoir area SAR image dataset.

[0047] Preferably, the step S8 is specifically as follows:

[0048] Step S8.1. Create a supervised learning dataset based on the predicted lag values ​​obtained in step S7;

[0049] Step S8.2. Obtain data on the actual occurrence time of the landslide from step 1, and factors causing the landslide, such as reservoir water level and rainfall data;

[0050] Step S8.3. Select reservoir water level lag data, rainfall lag data, and displacement lag data based on the lag time;

[0051] Step S8.4. Build the initial XGBoost regression model and set hyperparameters such as n_estimators, learning_rate, and max_depth.

[0052] Step S8.5. Divide the data selected in step S8.3 into a training set and a test set according to the time sequence, and perform fitting training on the training set;

[0053] Step S8.6. Perform predictions on the test set step by step, initialize the lagged feature values ​​of the test set, and then perform predictions by updating the lagged features period by period.

[0054] Step S8.7. Calculate the error between the predicted result and the actual value, and continue training until the training target is reached.

[0055] Preferably, in step S8.7, mean square error, root mean square error and coefficient of determination are used as training targets.

[0056] The reservoir area landslide deformation prediction method based on InSAR technology and wavelet analysis described in the present invention utilizes the combination of wavelet analysis and XGBoost model to fully utilize the reservoir area landslide characteristics and can significantly improve the reservoir area landslide displacement prediction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 This is a flow chart of a specific implementation of the evaluation method of the present invention;

[0058] Figure 2 1 is a schematic diagram of removing trend items using the least squares method in a specific embodiment of the present invention;

[0059] Figure 3 This is a schematic diagram of the calculation results of grey relational degree in a specific embodiment of the present invention;

[0060] Figure 4 This is a graph of the continuous wavelet results of displacement points in a specific embodiment of the present invention;

[0061] Figure 5 Schematic diagram of wavelet coherence spectrum of reservoir water level and displacement in a specific embodiment of the present invention;

[0062] Figure 6 is a schematic diagram of a wavelet coherence spectrum of rainfall and displacement in a specific embodiment of the present invention;

[0063] Figure 7 It is a schematic diagram of a prediction result curve in a specific embodiment of the present invention. DETAILED DESCRIPTION

[0064] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely explained below in conjunction with the specific embodiments of the present invention and the corresponding drawings. Obviously, the embodiments described 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.

[0065] The present invention will be further described below in conjunction with the accompanying drawings and specific preferred embodiments, but the scope of protection of the present invention is not limited thereby.

[0066] like Figure 1 The method for predicting landslide deformation in a reservoir area based on InSAR technology and wavelet analysis is shown in the figure. It is necessary to first obtain the data source, including the following steps:

[0067] Step S1. Download the SLC (single-look complex) data of the Sentinel-1A satellite, with VV (vertical-vertical) polarization, an azimuth angle of 347.43°, an incidence angle of 33.92°, and acquisition time from 2020 / 01 / 16 to 2023 / 11 / 08, a total of n=111 periods, each period interval TA=12 days, and all 111 periods of data are used as the raw data set.

[0068] In the raw data set, the first 101 periods can be used for wavelet analysis, and periods 7-111 can be used for model training and prediction;

[0069] Based on satellite data analysis, the actual time of landslide occurrence was obtained;

[0070] Step S2. Collect reservoir water level and rainfall data in the study area to form tabular data. Select the reservoir water level corresponding to the InSAR data period and accumulate it every 12 days as the rainfall corresponding to the deformation period. The data type, source, range, and processing method of this step are shown in Table 1.

[0071] Table 1 Basic data information table

[0072]

[0073] Finally, the reservoir water level sequence l is obtained i and rainfall time series r i ; i=1,2,...n; n is the total number of periods;

[0074] Step S3. The reservoir area SAR image dataset obtained in step S1 is processed by small baseline subset interferometric synthetic aperture radar (SBAS-InSAR) to obtain the cumulative deformation of each period in the reservoir area. Assume that the period number and the corresponding original displacement of each period are t i and d i ; i = 1, 2, ... n, n is the total number of periods, n = 111;

[0075] Step S4. Fitting the displacement linear trend using the least squares method,

[0076] t given by step S3 i and d i , assuming t i and d i The relationship between can be fitted by the linear function formula 1

[0077] d i =f(t i )=a0+*a1*t i ---Formula 1

[0078] , where a0 is the intercept and a1 is the slope. Then define the error square sum R1 to represent the fitting error:

[0079]

[0080] When the R1 value is the smallest, the best fitting parameters a0 and a1 are obtained. i Substitute into formula 1 to calculate the displacement of the fitted trend term;

[0081] Reusing the original displacement d i Subtract the fitted trend term displacement to obtain the period term displacement d′ i =d i -(a0+a1·t i ).

[0082] A specific implementation of the displacement data before and after processing in this step is as follows Figure 2 As shown, from Figure 2 It can be seen that the trend of the displacement data after processing is significantly weakened.

[0083] Step S5. Use grey correlation to determine the primary and secondary influencing factors, provide the ranking of influencing factors for the subsequent wavelet analysis step, and provide a basis for the selection of input features for the XGBoost model. i The calculation formula of (t) can be expressed as:

[0084]

[0085] In formula 2: x0(t) is the reference sequence, i.e., the displacement sequence, x i (t) is the sequence to be compared, i.e., the reservoir water level or rainfall sequence; ρ is the resolution coefficient used to control the discrimination of the system, ρ∈[0,1], the larger the ρ value, the greater the discrimination, and the default value is 0.5; |x0(t)-x i (t)| is called x0(t) and x at time t i The absolute difference of (t);

[0086] is the minimum difference between the two levels,

[0087] is the maximum difference between the two poles;

[0088] In this step, the displacement sequence x0(t) uses the periodic item displacement obtained in step 4; the results of calculating the grey relational degree for different time periods are as follows: Figure 3 As shown, Figure 3 The horizontal axis is time, and the vertical axis is the grey correlation degree.

[0089] For example, if the grey correlation threshold is set to 0.01, the calculation shows that the grey correlation between displacement and reservoir water level and rainfall are 0.9604 and 0.8241 respectively, both of which are greater than the grey correlation threshold. In this case, both reservoir water level and rainfall factors enter the subsequent steps, and it can be determined that reservoir water level is the main influencing factor of landslide displacement.

[0090] This step is used to perform a preliminary screening of factors based on the grey correlation degree and set the grey correlation degree threshold. When the grey correlation degree is found to be lower than the grey management degree threshold, the corresponding factors are not considered and the subsequent steps are not entered.

[0091] Step S6. Use continuous wavelet transform (CWT) to analyze the periodicity of displacement, reservoir water level, and rainfall series. The continuous wavelet transform is defined as:

[0092]

[0093] where ψ is the sub-wavelet, ψ * is the complex conjugate of ψ, R represents the set of real numbers, τ is the translation parameter, which can be any real number, and s is the scaling factor with a value greater than 0.

[0094] f(t) is the function of displacement, reservoir water level and rainfall relative to time, where the function of displacement relative to time adopts the periodic displacement obtained in step S4.

[0095] d′ i =d i -(a0+a1·t i ),

[0096] The function of reservoir water level and rainfall relative to time adopts the function obtained in step S2;

[0097] The continuous wavelet results W of displacement, reservoir water level and rainfall can be obtained from formula 3 d′ (τ,s),W l (τ,s) and W r Obtaining the cycle length T is necessary for the subsequent calculation of the lag time. The y-axis on the left side of the continuous wavelet spectrum reflects the cycle length. The closed curve in the figure corresponds to approximately 30 periods on the left y-axis, each 12 days. Therefore, it is concluded that the reservoir water level exhibited a cycle of approximately one year (30 × 12 days) from February 2020 to January 2023.

[0098] Step S7. Use cross wavelet transform (XWT) and wavelet coherence (WTC) to extract the hysteresis of deformation point displacement to reservoir water level and rainfall. xy (τ, s) is defined as:

[0099]

[0100] in It's W y The complex conjugate of W is used in the present invention. x (τ,s) represents the continuous wavelet on the displacement sequence, W y (τ,s) represents the continuous wavelet of external factors such as reservoir water level and rainfall. The cross wavelet transform provides the interaction strength between the displacement sequence and the external factor sequence at different scales and times. The cross wavelet power spectrum can be obtained by |W xy (τ,s)| 2 This value represents the mutual phase and intensity of the two time series at different times and scales, and can reveal the hysteresis and periodicity characteristics.

[0101] The cross wavelet transform of the displacement sequence variable and the reservoir water level and rainfall sequence variables is calculated using formula 4.

[0102] Reservoir water level and rainfall, due to their influence on the hydrogeological conditions of reservoir bank landslides, are considered important external factors in predicting landslide displacements in reservoir areas. However, due to the correlation between reservoir water level and rainfall, the independence of individual external factors must be ensured when studying their respective effects on displacement. Wavelet coherence can be further calculated using a cross-wavelet transform. Wavelet coherence clearly identifies the consistency between the displacement time series and the time series of a single external influencing factor in the time-frequency space, while eliminating high-frequency noise from other external factors. Furthermore, wavelet coherence is normalized by dividing by the respective wavelet power spectra. This effectively removes the influence of differences in the power spectra of the sequences themselves, ensuring that the analysis results focus on the relative coherence between the sequences.

[0103] Because the two external factors in this invention are both hydrological factors, there is a certain correlation between them. There is a relatively fixed relationship between the reservoir water level and the rainfall in the reservoir area. For example, in this invention, the rainy season from July to September coincides with the reservoir impoundment period, and the reservoir water level will rise during this period. The correlation between external factors can affect correlation analysis. Without using cross-wavelet or similar independent external factor analysis methods, the response of displacement to a single factor cannot be clearly explained, which in turn affects the accuracy of the lag time extraction for a specific factor.

[0104] When performing wavelet analysis, different sequences may exhibit differences in their power spectra due to differences in units, dimensions, or numerical ranges. For example, one sequence may have a large numerical range, while another may have a smaller numerical range. This difference can lead to differences in their power spectra, which in turn affects the results of coherence analysis. To eliminate this effect, wavelet coherence normalizes the power spectra of each sequence by dividing them by their respective power spectra. This approach ensures that the coherence analysis focuses on the relative changes between the sequences, regardless of factors such as the numerical range and units of the sequences themselves, thus providing a unified metric.

[0105] Wavelet coherence WTC is:

[0106]

[0107] In formula 5, S() represents the smoothing operator, and the result R xy The value of (τ,s) is between 0 and 1. The closer the value is to 1, the stronger the correlation between the two time series at that time and scale.

[0108] The obtained cross wavelet transform XWT and wavelet coherence WTC results can be brought into existing computer software such as MATLAB to generate wavelet coherence maps using the software.

[0109] like Figure 4 As shown, the deformation point diagram of the displacement is given, Figure 5 Give Figure 4 The coherent wavelet spectrum of displacement and reservoir water level is shown in the figure. Figure 6 Give Figure 4 The coherent wavelet spectrum of displacement and rainfall is shown. Figure 4 、 Figure 5 and Figure 6 In the figure, the X-axis is time and the Y-axis is the period unit (the unit is 12 days). The inverted cone curve in the figure is the cone of influence (COI). The data within the cone of influence curve is reliable, while the data outside the cone of influence curve is affected by boundary effects and has low reliability, so it is usually not considered.

[0110] The thick black line represents the 95% confidence level. The periodic signal within the area enclosed by the confidence level and the influence cone is significantly stronger than the noise, indicating that the data in these areas have strong periodicity on the time scale.

[0111] The arrows in the graph are the direction of the phase angle, reflecting the time Δt of the displacement lag, which is calculated as follows:

[0112]

[0113] Where Δt is the lag time, is the phase angle indicated by the arrow, T is the period, T = 360 days;

[0114] The impact of changes in external factors such as precipitation and reservoir water levels on landslide displacement may not manifest until several periods have passed. To address this issue, this paper focuses on the lag in displacement response to external factors. By using this lag time to construct hysteresis features, we intentionally select external factor data that precedes the displacement series by several lag periods and apply this data to landslide displacement prediction.

[0115] from Figure 5 and Figure 6 It can be seen that within the area below the influence cone and enclosed by the confidence level, the phase angle of the displacement-reservoir level wavelet arrow is approximately 0.137π, and the phase angle of the displacement-rainfall arrow is approximately 0.375π. Substituting this into Formula 6 to calculate the lag time Δt, the calculated results show that displacement lags behind reservoir level by approximately 24.7 days and behind rainfall by approximately 67.5 days. In practice, because the displacement prediction period is after the period used in the wavelet analysis, the phase angle at the end of the wavelet result better represents the lag characteristics of the prediction period. In other words, the phase angle is preferentially selected within the area below the influence cone and enclosed by the confidence level. The calculated displacement lags behind reservoir level by approximately 24 days and behind rainfall by approximately 69 days.

[0116] Step S8.

[0117] The predicted lag value obtained in step 7, the actual occurrence time of the landslide obtained in step 1, and the factors causing the landslide, such as reservoir water level and rainfall data, are brought into the XGBoost model for training to obtain a landslide prediction model.

[0118] Landslide prediction is carried out using a landslide prediction model. Reservoir water level data and rainfall data are input into the landslide prediction model to obtain the predicted lag value of landslide occurrence, thereby realizing landslide prediction.

[0119] The XGBoost model has the ability to process nonlinear relationships and time series data. Based on XGBoost, a landslide displacement prediction model is built, which integrates reservoir water level, rainfall and historical displacement for data-driven modeling, and extracts hysteresis features to capture the time series characteristics of displacement. After model training, a step-by-step prediction strategy is used to generate test set displacement values, and the mean square error (MSE), root mean square error (RMSE) and coefficient of determination (R 2 ) and other indicators to evaluate the performance and verify the effectiveness of the model.

[0120] When using XGBoost to predict landslide displacement, the usual approach is to use hysteresis features, including the hysteresis of the displacement itself and the hysteresis of external factors such as rainfall and water level. For example, the method of using two-period hysteresis features to predict displacement is to use the displacement of periods t-1 and t-2, the reservoir water level of periods t-1 and t-2, and the rainfall data of periods t-1 and t-2 as input variables to predict the displacement of period t.

[0121] For the hysteresis of displacement itself, we only need to consider how many periods of displacement preceding the current period's displacement can describe the approximate displacement trend. However, the selection of hysteresis for external factors is more complex. First, the displacement sequence and the external factor sequence may change out of sync, especially for landslides in reservoir areas affected by hydrological factors. For example, in the present invention, displacement will not respond until a certain number of periods have passed since the reservoir water level and rainfall changed. Second, there is often more than one external factor, and combining the hysteresis periods of multiple external factors consumes human and computational resources.

[0122] The present invention can provide specific lag periods of external factors by using wavelet analysis, thereby predicting the comprehensive lag results of the above-mentioned asynchronous influences, greatly improving the complexity and uncertainty of lag feature selection in the original method, and ensuring the accuracy of model prediction.

[0123] Step S8. A typical implementation of model training includes the following steps:

[0124] Step S8.1. Create a supervised learning dataset based on the predicted lag values ​​obtained in step S7;

[0125] Step S8.2. Obtain data on the actual occurrence time of the landslide from step 1, and factors causing the landslide, such as reservoir water level and rainfall data;

[0126] Step S8.3. Select reservoir water level lag data, rainfall lag data, and displacement lag data as characteristic data of the present invention based on the lag time;

[0127] A specific implementation method is as follows: reservoir water level lag data, rainfall lag data, and displacement lag data at the same monitoring location as the characteristic data of the present invention are selected as the characteristic data of the existing model; the data amount of the reservoir water level lag data and the rainfall lag data in the characteristic data of the existing model are the same, and the total amount of the data of the two is equal to the total amount of the data of the two in the characteristic data of the present invention; the data amount of the displacement lag data in the characteristic data of the existing model and the characteristic data of the present invention are the same;

[0128] For example, 10 lag data, including the 1st and 2nd period reservoir water level lag data, the 1st, 2nd, 3rd, 4th, 5th and 6th period rainfall lag data and the 1st and 2nd period displacement lag data, are selected as the characteristic data of the present invention; while the control group is the same monitoring point, but the accurate lag time is unknown by default, so the same number of uniform reservoir water level and rainfall lag characteristics are selected, that is, 1st, 2nd, 3rd and 4th period reservoir water level lag, 1st, 2nd, 3rd and 4th period rainfall lag, and 1st and 2nd period displacement lag, a total of 10 lag characteristics are selected as the characteristic data of the existing model.

[0129] The specific hysteresis characteristics and the control group hysteresis characteristics are set according to the wavelet analysis results. The hysteresis characteristics are actually a dislocated sequence, for example, there is a sequence containing displacement and an external factor:

[0130] For example, in a specific embodiment, the displacement lags behind the reservoir water level by about 24 days and lags behind the rainfall by about 69 days, which translates to approximately 2 and 6 periods, respectively. Therefore, the reservoir water level in the first two periods and the rainfall in the first six periods are used to construct the "displacement" sequence and predict the current displacement.

[0131] For the setting of the existing model characteristic data as the control group, the displacement lag of 1 and 2 periods, which are consistent with the experimental group, is adopted. The setting of the external lag characteristics adopts the general processing method, that is, the same reservoir water level and rainfall lag characteristics are used, which are all 1, 2, 3, and 4 periods. This ensures that the total lag characteristic number of the experimental group and the control group is consistent at 10, and better controls the variables.

[0132] Step S8.4. Build the initial XGBoost regression model and set hyperparameters such as n_estimators, learning_rate, and max_depth.

[0133] Step S8.5. Divide the data into a training set and a test set according to the time sequence, and use a rolling update prediction method to fit the model to the training set;

[0134] For example, after obtaining the predicted value using the data from one period in the test set, the predicted value is incorporated into the original training set to make it the new training set data, and then the new test set is predicted, and so on.

[0135] Step S8.6: Perform predictions on the test set step by step. To avoid using future true values ​​for predictions, initialize the lagged feature values ​​of the test set, and then perform predictions by updating the lagged features period by period.

[0136] Step S8.7. Calculate the error between the predicted result and the actual value using the mean square error (MSE), root mean square error (RMSE) and coefficient of determination (R 2 ) and other indicators to evaluate the prediction effect of the model.

[0137] Table 2 Accuracy test table

[0138]

[0139] As shown in Table 2, the MSE of the prediction results using the accurate lag period and not using it are 5.650 and 16.114 respectively, and the RMSE are 2.377 and 4.014 respectively. 2 The values ​​of 0.928 and 0.795 are respectively 0.928 and 0.795. This shows that the model using accurate lag features has a certain improvement in both prediction error and fitting effect compared with the model not using accurate lag features.

[0140] Table 3 Comparison of model prediction results

[0141]

[0142] Extract some data from Table 3 to draw the prediction curve, such as Figure 7 As shown, it can be seen that the prediction results of the present invention are more consistent with the true value than those of the prior art.

[0143] The foregoing are the preferred embodiments of the present invention. Unless the preferred implementation modes in each preferred embodiment are obviously self-contradictory or based on a certain preferred implementation mode, each preferred implementation mode can be arbitrarily superimposed and used in combination. The embodiments and the specific parameters in the embodiments are only for the purpose of clearly describing the inventor's invention verification process, and are not intended to limit the patent protection scope of the present invention. The patent protection scope of the present invention shall still be based on its claims. Any equivalent structural changes made using the contents of the description and drawings of the present invention should also be included in the protection scope of the present invention.

Claims

1. The landslide deformation prediction method in reservoir area based on InSAR technology and wavelet analysis is characterized by , including the following steps: Step S1. Collecting a reservoir area SAR image dataset within a time interval TX acquired by a satellite; the reservoir area SAR image dataset is in the form of an n-period installment dataset with a time interval TA; Step S2. Collect reservoir water level and reservoir area rainfall data within the same time interval TX; The reservoir water level and rainfall data in the reservoir area are divided into periods according to the time interval TA described in step 1 to obtain the reservoir water level sequence l i and rainfall time series r i ; i=1,2,...n; Step S3. The reservoir area SAR image data obtained in step S1 is processed by small baseline subset interferometric synthetic aperture radar (SBAS-InSAR) to obtain the cumulative deformation of each period; the period number and the corresponding original displacement of each period are t i and d i ; i=1,2,...n; Step S4. Use the least squares method to fit, For the t given in step S3 i and d i , assuming t i and d i The relationship between can be fitted by the linear function formula 1 d i =f(t i )=a0+*a1*t i ---Formula 1 Among them, a0 is the intercept, a1 is the slope, and the error square sum R1 is defined to represent the fitting error: When the R1 value is the smallest, the best fitting parameters a0 and a1 are obtained. i Substitute into formula 1 to calculate the displacement of the fitted trend term; Reusing the original displacement d i Subtract the fitted trend term displacement to obtain the period term displacement d i =d i -(a0+a1·t i ); Step S6. Use continuous wavelet transform to analyze the periodicity of displacement, reservoir water level, and rainfall series. The continuous wavelet transform is defined as: ---Formula 3 Where ψ is the sub-wavelet, ψ * is the complex conjugate of ψ, R represents a set of real numbers, τ is a translation parameter, and s is a scaling factor with a value greater than 0; f(t) is a function of displacement, reservoir water level, and rainfall relative to time, respectively, where the function of displacement relative to time adopts the periodic term displacement obtained in step S4, and the functions of reservoir water level and rainfall relative to time adopt the functions obtained in step S2; The continuous wavelet values ​​W of displacement, reservoir water level, rainfall and other parameters can be obtained from formula 3: i (τ,s), where the subscript i represents different parameters such as displacement, reservoir level, or rainfall; Step S7. Using cross wavelet transform (XWT) and wavelet coherence (WTC) to extract the hysteresis of deformation point displacement to reservoir water level and rainfall; Cross wavelet transform of two time series variables x and y Defined as: ---Formula 4 in It's W y The complex conjugate of W x (τ,s) represents the continuous wavelet value on the displacement sequence, W y (τ,s) represents the continuous wavelet value of external factors such as reservoir water level and rainfall; By using formula 4 and substituting the continuous wavelet value obtained by formula 3, we can calculate the cross wavelet transform of the displacement sequence variable and the reservoir water level and rainfall sequence variables. Using the calculation results of formula 4, calculate the wavelet coherence; Wavelet coherence WTC is: ---Formula 5 In formula 5, S() represents the smoothing operator; The cross wavelet transform and wavelet coherence obtained by formula 4 and formula 5 are used to generate the wavelet coherence spectrum; The wavelet coherence spectrum was analyzed to obtain the hysteresis value of landslide occurrence; Step S8. The predicted lag value obtained in step 7, the actual occurrence time of the landslide obtained in step 1, and the factors causing the landslide are brought into the XGBoost model for training to obtain a landslide prediction model; and the landslide prediction model is used to predict the landslide.

2. The reservoir area landslide deformation prediction method based on InSAR technology and wavelet analysis as claimed in claim 1 is characterized in that: Between the steps S4 and S6, step S5 is also included. The grey correlation degree is used to screen the influencing factors. The grey correlation degree ξ i The calculation formula of (t) can be expressed as: ---Formula 2 In formula 2: x0(t) is the reference sequence, i.e., the displacement sequence, x i (t) is the sequence to be compared, i.e., the reservoir water level or rainfall sequence; ρ is the resolution coefficient used to control the system discrimination, ρ∈[0,1]; |x0(t)-x i (t)| is called x0(t) and x at time t i The absolute difference of (t); is the minimum difference between the two levels, is the maximum difference between the two poles; In this step, the displacement sequence x0(t) uses the periodic term displacement obtained in step 4; The two factors of reservoir water level and reservoir area rainfall data obtained in step 2 are used as the comparison sequence in formula 2 to calculate the grey correlation degree ξ i (t), whether it is greater than the set grey correlation threshold, if so, the factor enters the subsequent steps, otherwise it does not enter the subsequent steps.

3. The reservoir area landslide deformation prediction method based on InSAR technology and wavelet analysis as claimed in claim 1 is characterized in that: The S1 step specifically includes: Step S1.

1. Preprocessing the data, including radiometric calibration, geometric correction, filtering, and registration; Step S1.2 uses the pre-processed SAR image to perform interferometric processing to generate an initial interferometric phase map containing information such as terrain phase, deformation phase, and atmospheric phase; The terrain phase is simulated using an external digital elevation model, and the terrain phase is subtracted from the initial interferometric phase map to eliminate the influence of the terrain phase and obtain a differential interferometric phase map containing information such as deformation phase and atmospheric phase. Step S1.

3. Perform phase unwrapping on the differential interferometer phase image to convert the wrapped phase value into the real phase value; Step S1.

4. Filter the unwrapped phase image to remove atmospheric noise and surface cover variation errors; Step S1.5 converts the phase values ​​of the filtered differential interferometry phase image into distance values ​​to obtain the surface shape variables, thereby forming the reservoir area SAR image dataset.

4. The reservoir area landslide deformation prediction method based on InSAR technology and wavelet analysis as claimed in claim 1 is characterized in that: The step S8 is specifically as follows: Step S8.

1. Create a supervised learning dataset based on the predicted lag values ​​obtained in step S7; Step S8.

2. Obtain data on the actual occurrence time of the landslide from step 1, and factors causing the landslide, such as reservoir water level and rainfall data; Step S8.

3. Select reservoir water level lag data, rainfall lag data, and displacement lag data based on the lag time; Step S8.

4. Build the initial XGBoost regression model and set hyperparameters such as n_estimators, learning_rate, and max_depth. Step S8.

5. Divide the data selected in step S8.3 into a training set and a test set according to the time sequence, and perform fitting training on the training set; Step S8.

6. Perform predictions on the test set step by step, initialize the lagged feature values ​​of the test set, and then perform predictions by updating the lagged features period by period. Step S8.

7. Calculate the error between the predicted result and the actual value, and continue training until the training target is reached.

5. The reservoir area landslide deformation prediction method based on InSAR technology and wavelet analysis as claimed in claim 4 is characterized in that: In step S8.7, mean square error, root mean square error and coefficient of determination are used as training targets.

Images