CN-121978683-A - GNSS and radar data fusion-based large-scale landslide mass displacement early warning method

CN121978683ACN 121978683 ACN121978683 ACN 121978683ACN-121978683-A

Abstract

The invention discloses a large-range landslide body displacement early warning method based on fusion of GNSS and radar data, which comprises the steps of acquiring GNSS monitoring sequence data in a landslide area based on position information of a reference station and a GNSS monitoring station in a GNSS monitoring system to further obtain landslide area displacement related information, acquiring data through a satellite InSAR, generating deformation field data of the landslide area through radar images at different time points, carrying out time-space alignment on the acquired data, carrying out two types of data fusion S by adopting an adaptive Kalman filtering method after the alignment, decomposing time sequence data of the fused landslide body displacement by utilizing wavelet transformation to extract multi-scale features of a time domain and a frequency domain, inputting the extracted features into a landslide body displacement trend prediction model for constructing an LSTM (least square) method by adding a memory fading factor, setting a multi-level threshold, judging the change trend of the landslide body displacement according to the displacement rate, acceleration and a prediction result, and generating a multi-level early warning signal.

Inventors

ZHANG XUEJIE
Bi Yadi
WANG LONGBAO
ZHU YUN
GUO JIAJUN
WU SHUBIN
FAN HAO
LIU SHUANG

Assignees

河海大学

Dates

Publication Date: 20260505
Application Date: 20260121

Claims (7)

1. A Global Navigation Satellite System (GNSS) and radar data fusion-based large-range landslide mass displacement early warning method is characterized by comprising the following steps: s1, acquiring GNSS monitoring sequence data in a landslide area based on position information of a reference station and a GNSS monitoring station in a GNSS monitoring system to further obtain landslide area displacement related information; s2, acquiring data through a satellite InSAR, and generating deformation field data of a landslide region through comparing radar images at different time points; S3, carrying out space-time registration on the GNSS displacement information obtained in the step S1 and the radar deformation field data obtained in the step S2, and fusing the acquired GNSS displacement information and the radar deformation field data by using a self-adaptive double-weight Kalman filtering method; s4, decomposing time sequence data of the landslide body displacement by utilizing wavelet transformation to extract multi-scale characteristics of a time domain and a frequency domain; S5, constructing a landslide body displacement trend prediction model by using an LSTM method added with a memory decay factor according to the time-frequency characteristics extracted in the S4 and combining with historical data of landslide displacement, and predicting the future trend of the landslide displacement; And S6, setting a multistage threshold value, judging the change trend of the displacement of the landslide body according to the displacement rate, the acceleration and the prediction result, and generating a multistage early warning signal.
2. The method for early warning of displacement of a large-scale landslide body based on fusion of GNSS and radar data according to claim 1, wherein the specific steps of acquiring GNSS monitoring sequence data in a landslide region in step S1 to further obtain the related information of displacement of the landslide region are as follows: Al, converting the data coordinates monitored by GNSS from longitude, latitude and elevation into ENU coordinates, selecting a reference point (usually coordinates of GNSS stations) as an origin of an ENU coordinate system, and assuming the longitude and latitude coordinates (), for the longitude and latitude of a point, converting the longitude, latitude and elevation into ECEF coordinates by using the following formula: X=(N(φ)+h)cosφcosλ (1) Y=(N(φ)+h)cosφsinλ (2) Z=((1-e 2 )N(φ)+h)sinφ (3) Wherein phi is latitude, lambda is longitude, and h is elevation; the method comprises the steps of calculating ECEF coordinates X 0 ,Y 0 ,Z 0 of a reference point by using the same formula; A2, calculating a difference value between the target point and the reference point, converting the difference value from an ECEF coordinate system to an ENU coordinate through a rotation matrix, wherein the rotation matrix of the ENU coordinate system is based on longitude and latitude of the reference point, the coordinates of the target point in the ECEF coordinate system are (X, Y and Z), the ECEF coordinate system of the reference point is, and then the local ENU coordinate is calculated as follows: wherein the rotation matrix R (phi 0 ,λ 0 ) is used to convert ECEF coordinates into ENU coordinates, and the formula is as follows: The east, north and high of the target point in the ENU coordinate system can be obtained; A3 after obtaining the ENU coordinates, we can perform displacement calculation, assuming that there are a plurality of observations in successive time steps, where taking k 1 and k 2 as examples, the ENU coordinates at each time are E (k 1 ),N(k 1 ),U(k 1 ) corresponding to the ENU coordinates at time k 1 , E (k 2 ),N(k 2 ),U(k 2 ) corresponding to the ENU coordinates at time k 2 , and Δd GNSS,k represents the displacement difference from the last time, calculated by the following formula:
3. The method for early warning of large-scale landslide body displacement based on fusion of GNSS and radar data according to claim 1, wherein the method for generating a landslide region deformation field by satellite InSAR in the step S2 is as follows: B1, acquiring radar images through a satellite platform, wherein the radar images generally contain phase information of radar waves, reflect the deformation of the earth surface at different times, and the phase information phi in each SAR image is related to a radar wave propagation path R of a target and can be expressed as follows: where φ is the phase, R is the path length of radar wave propagation, λ is the radar wavelength; B2, calculating an interference pattern between two SAR images by adopting a differential interference technology, wherein the differential interference can reveal tiny deformation of the earth surface, and the phase change phi ps (k) at a moment t is related to the displacement d (k): and calculates a displacement difference by the phase difference according to the following formula: Δφ=φ 2 -φ 1 (9) Wherein Δφ is the differential phase of the two images, φ 2 and φ 1 are the phases of the two images at times k 2 and k 1 , respectively, and the relationship between the phase difference and the displacement difference can be obtained according to formula (8): From this equation (10), the displacement difference Δd can be calculated: and B3, calculating phase differences at a plurality of time points, performing projection conversion after obtaining corresponding displacement differences, and generating a deformation field of earth surface displacement, wherein in order to obtain deformation data consistent with the actual landslide direction, the displacement data is required to be projected to the direction under the slope from the radar sight direction, and the calculation formula is as follows: Δd downslop =Δd·cosθ (12) where Δd is the displacement difference calculated from the phase difference, Δd downslop is the converted displacement, and θ is the angle between the LOS direction and the ground perpendicular direction.
4. The method for early warning of large-scale landslide body displacement based on fusion of GNSS and radar data according to claim 1, wherein the step S3 performs space-time registration on GNSS data and obtained deformation field data, and performs data fusion by adopting a self-adaptive double-weight Kalman filtering method, and the specific method comprises the following steps: C1. since the time intervals of GNSS data and InSAR data are different, the acquisition frequency of GNSS data is generally higher (e.g. provided every hour or every time), and the acquisition frequency of InSAR data is lower (e.g. once every few days), so, in order to ensure that the two data can be consistent in time, interpolation processing needs to be performed on the InSAR data, where GNSS data is denoted as d GNSS (k) = [ E (k), N (k), U (k) ], and InSAR data is denoted as Δd downslop (k), where k is a time point for interpolation, we interpolate the InSAR data according to the time point k of GNSS data, assuming that the InSAR data is denoted as Δd downslop (k 1 ),Δd downslop (k 2 at the time point k 1 ,k 2 ), and in order to calculate the InSAR data corresponding to the time point k, we can use the following interpolation formula: C2. in order to ensure spatial consistency of GNSS data and radar data, it is necessary to map them to the same coordinate system, spatially align the GNSS data with the InSAR data, since the InSAR data is measured based on the line of sight (LOS) of the satellite, and we wish to compare it with the GNSS data in the same coordinate system, it is necessary to convert the displacement in LOS direction to the displacement in ENU coordinate system, assuming that the angular relationship between the satellite and the target point is β, the LOS direction displacement can be converted to the displacement in ENU coordinate system using a rotation matrix, assuming that the LOS displacement Δd LOS in the InSAR data can be converted to the displacement in ENU coordinate system by the following formula: wherein R (beta) is a rotation matrix which converts LOS displacement into three components in an ENU coordinate system, namely displacement in the east, north and sky directions respectively, The GNSS data are already represented in the ENU coordinate system and the converted InSAR data are also represented in the ENU coordinate system, so that their spatial positions are already aligned; C3, after space-time alignment, obtaining displacement data of GNSS data and satellite InSAR data in the same time and space position in each time step, adopting a self-adaptive double-weight Kalman filtering method to fuse the GNSS data and the satellite InSAR data, introducing a time weight coefficient w t in the time dimension, dynamically adjusting the occupation ratio of the GNSS and the InSAR data in Kalman updating according to different observation time intervals, introducing a space weight coefficient w s in the space dimension, dynamically distributing weights according to the complexity of the topography, crushing or shielding more areas in mountain bodies, improving the weight of GNSS point positions in areas with relatively gentle topography and better sight conditions, and improving the weight of an InSAR network; the GNSS and satellite InSAR data are respectively used as measurement inputs, and the measured noise covariance of the GNSS and satellite InSAR data is used for calculating Kalman gain, and then state estimation is updated based on respective gain weights, and the method is as follows: updating GNSS data: Representing an updated GNSS state estimate, representing a displacement estimate at time k; The GNSS state estimate representing the previous time k-1, i.e. the displacement estimate of the previous time, is here given that the initial reference position displacement d0 is required to obtain the displacement estimate of the moment since before this we have obtained the displacement data as the displacement difference from the previous time, as follows: Wherein the method comprises the steps of For the space-time integrated weights of the GNSS, For time weights, the smaller the observation interval, the larger the weight, The method is characterized by comprising the steps of obtaining space weight, obtaining more complex terrain and more GNSS weight, obtaining GNSS Kalman gain by K GNSS,k , obtaining the response degree to observation errors in Kalman filtering, obtaining more weight to observation data when updating, obtaining displacement difference between Deltad GNSS,k and the last moment, obtaining a GNSS observation matrix by H GNSS , and mapping state estimation to the observation space, wherein K GNSS,k is the GNSS Kalman gain, and obtaining the response degree to the observation errors in Kalman filtering, wherein Deltad GNSS,k is the displacement difference between the GNSS observation matrix and the last moment; Updating InSAR data: representing an updated InSAR state estimate, representing a displacement estimate at time k; InSAR state estimation representing the previous moment k-1, namely the displacement estimation value of the previous moment, wherein the displacement calculation method of InSAR data at a certain moment is the same as the GNSS data method, and the method comprises the following steps of Is the space-time comprehensive weight of InSAR, For time weights, the smaller the observation interval, the larger the weight, For spatial weight, the flatter the terrain, the greater the InSAR weight, where Θ is gradient, ε is coherence of InSAR data, typically ε is (0, 1), K InSAR,k represents InSAR Kalman gain, the larger the gain is, the larger the weight to the observed data is in updating, Δd InSAR,k represents displacement difference from the last moment, H InSAR represents an observed matrix for mapping state estimation to an observed space, typically in the form of a unit matrix; Finally, combining the estimation results of the two to obtain the fused state estimation: The fused state of the moment k is represented, and the fused state of the moment k is represented by a displacement estimated value obtained by fusing GNSS data and InSAR data, wherein alpha g and alpha i are weight factors of the GNSS data and the InSAR data respectively, and the contribution of each data source to final estimation is controlled; representing an updated GNSS state estimate, Representing the updated InSAR state estimate; Wherein R GNSS is a GNSS observation noise covariance matrix, and the prediction error covariance matrix at the last moment of P k|k-1 ; Wherein R InSAR is an InSAR observation noise covariance matrix, and K InSAR,k is the same.
5. The large-range landslide body displacement early warning method based on GNSS and radar data fusion is characterized in that the fused data in the step S4 are decomposed by utilizing wavelet transformation to extract multi-scale characteristics of time domain and frequency domain, in the time domain characteristics, not only mean value and standard deviation are extracted, but also first derivative (speed) and second derivative (acceleration) of a displacement sequence are further calculated and used for representing the dynamic evolution process of the landslide body displacement, wherein the displacement speed reflects the speed of displacement change, the displacement acceleration reflects the change trend of the displacement speed, when the acceleration is continuously positive and gradually increased, the landslide body enters an acceleration deformation stage and is a precursor characteristic with obviously increased instability occurrence probability, and the whole process from stable acceleration to temporary sliding of the landslide body can be more comprehensively revealed through the joint analysis of the speed and acceleration characteristic and the original displacement characteristic, and the method comprises the following steps: d1, calculating the speed and acceleration time sequence of the landslide body according to the displacement time sequence of the landslide body, wherein the formulas are respectively as follows Acceleration of displacement D2, the Morlet wavelet has signals with obvious frequency characteristics and is commonly used for frequency analysis of landslide data, so the Morlet wavelet is used as a mother wavelet for decomposing a basic function of the signals, the time sequence data x (t) of the landslide body displacement is subjected to wavelet transformation, and Continuous Wavelet Transformation (CWT) is selected for decomposing the signals; morlet wavelet is defined as: Wherein ψ (t) is the time domain representation of the Morlet wavelet describing the morphology of the wavelet in the time domain for generating a multi-scale transformation of the signal, w 0 is the center frequency, the value of which depends on the frequency range of the signal we wish to analyze; Equation for Continuous Wavelet Transform (CWT): wherein x (t) represents the time series data of the displacement of the landslide body, i.e. as mentioned above The psi * is the complex conjugate of the mother wavelet, a is a scale factor, the time-frequency resolution of the wavelet function is determined, a larger a corresponds to a wider wavelet (analyzing low-frequency components), a smaller a corresponds to a narrower wavelet (analyzing high-frequency components), b is a translation factor, the position of the wavelet function on a time axis is controlled, the time of the wavelet is unknown and is used for analyzing the local characteristics of the signal at different time points, and W (a, b) is the result of wavelet transformation and represents the characteristics of the signal on the scale a and the time position b; D3, through multiple times of wavelet transformation, the landslide mass displacement data can be finally decomposed into low-frequency and high-frequency parts under multiple scales, the low-frequency coefficients and the high-frequency coefficients are respectively represented by C a ,C d , the characteristics of the data on a time domain can be extracted through the low-frequency parts, the mean value and the standard deviation can be extracted, and the high-frequency components can help to analyze local changes in the landslide displacement; and carrying out multi-scale decomposition on landslide displacement data through CWT to obtain frequency components of different frequency bands, analyzing the frequency range of decomposition of each layer, extracting the frequency characteristics of the landslide displacement data under different scales, and obtaining the frequency domain characteristic instantaneous frequency, the main frequency and the power spectral density of the landslide displacement data.
6. The method for early warning of large-scale landslide body displacement based on GNSS and radar data fusion according to claim 1, wherein the landslide body displacement trend prediction model in the step S5 introduces a memory attenuation factor gamma t into a forgetting gate of an LSTM network for dynamically regulating and controlling memory retention time, and performs trend prediction according to historical data and time-frequency characteristics; E1, constructing an input sequence by using the data fused in the previous step, namely cleaning and preprocessing the data and the obtained time-frequency characteristics, determining an input time window of a model, segmenting training data and target prediction data according to the time window with a fixed length, and constructing a training set and a testing set; E2, improving a forgetting gate on the basis of an LSTM network, introducing a memory attenuation factor gamma t , adaptively adjusting the factor along with time and the fluctuation degree of data, when the sliding mass displacement fluctuates severely at a certain stage, the gamma t has larger value, promoting the forgetting gate to accelerate the forgetting speed so as to inhibit the disturbance of sudden abnormality on long-term memory, when the sliding mass displacement is in a relatively stable state, the gamma t has smaller value, and the forgetting gate prolongs the memory retention time to capture long-term dependence, wherein the gamma t can be adaptively adjusted based on the standard deviation of a sliding window, and the standard deviation of the sliding window is as follows: Where W is the window size, x is the time series, μ is the mean of the data in the sliding window, and m is a small constant for controlling the magnitude of the variance of the decay factor; E3, in the improved network structure, updating a calculation formula of the forgetting gate as follows: f t ＝σ(W f ·[h t-1 ,x t ]+b f )·γ t (23) wherein f t is the output of the forgetting gate, sigma is a Sigmoid function, h t-1 is the hidden state of the last moment, x t is the current input, W f and b f are the weight and bias parameters of the forgetting gate respectively, and gamma t is the adaptive memory attenuation factor; And E4, training the improved model after constructing a training set and a testing set, wherein the improved LSTM network after finishing training can adaptively adjust the memory half-life according to different period characteristics of landslide mass displacement data, can quickly respond during severe fluctuation and can also keep long-term dependence in a stable stage, so that a predicted result more conforming to the actual displacement trend is output, and can provide a more accurate displacement change trend for a landslide early warning system through further analysis of the predicted result, thereby early warning the occurrence of landslide disasters.
7. The method is characterized in that different thresholds are set in the step S6 to generate multi-stage early warning signals, the multi-stage thresholds comprise a displacement rate threshold and a displacement acceleration threshold, the displacement rate threshold and the displacement acceleration threshold are used for judging that the displacement is in a normal state when the displacement is in a safety range but the rate or acceleration is obviously increased but the dangerous level is not reached yet, the displacement is close to the dangerous level and the rate is continuously increased, the displacement is judged to be in a concerned level state, the displacement rate and the acceleration are simultaneously over the thresholds to indicate that the landslide body enters a critical sliding stage, the alarm state is judged, and the system can recognize the instability trend in advance when the accumulated displacement does not reach the dangerous threshold through introducing the rate and the acceleration index, so that more sensitive and timely early warning is realized.

Description

GNSS and radar data fusion-based large-scale landslide mass displacement early warning method Technical Field The invention belongs to the technical field of landslide prediction and early warning, and particularly relates to a large-range landslide body displacement early warning method based on GNSS and radar data fusion. Background Landslide is one of the common geological disasters in mountainous and hilly areas, and the formation process is complex and is influenced by various factors, including geological structures, rainfall, earthquake, ergonomic activities and the like. Once landslide disasters occur, serious life and property loss and environmental damage are often caused, so that the development of real-time monitoring and trend prediction of landslide body displacement has important scientific significance and social value. The current landslide monitoring and predicting method mainly comprises a detection technology based on GNSS, the technology can obtain high-precision three-dimensional point displacement information by arranging a reference station and a monitoring station on a landslide body, the measurement precision is high and the method can be suitable for various complex terrain conditions, especially for areas with less vegetation coverage, the satellite InSAR can obtain earth surface deformation field information in a large range, the displacement of the earth surface in the sight direction can be calculated by comparing SAR images of different time phases, the method has wide space coverage, deformation data of a large area can be obtained, planar deformation field information can be provided, the defect of sparse GNSS point positions is overcome, but the method receives the atmosphere, the earth surface covering change influence is large, the measurement result precision is low, and the reliability of the point position level is lacking relative to the GNSS. In the comprehensive view, the prior art still has the problems that point position data and planar data are difficult to fuse in the landslide displacement prediction field, time-frequency characteristic analysis is insufficient, the prior prediction method is mostly remained in a time-domain analysis layer, potential multi-scale frequency characteristics are difficult to identify, so that understanding of a landslide body deformation rule is not deep enough, and prediction and early warning accuracy is limited. Therefore, it is necessary to provide a landslide displacement prediction technology capable of integrating the advantages of GNSS and InSAR and combining a time-frequency analysis method and a deep learning method, so as to improve the comprehensiveness of monitoring, the accuracy of prediction and the timeliness of early warning. Disclosure of Invention The invention aims to overcome the defects of the prior art and provides a large-range landslide body displacement early warning method based on GNSS and radar data fusion. According to the method, space-time registration is carried out on GNSS point location data and satellite InSAR deformation field data, a self-adaptive double-weight Kalman filtering method is used for data fusion, time-frequency characteristics are extracted through wavelet transformation, and trend prediction is carried out by combining an improved LSTM deep learning model added with a memory decay factor, so that accurate prediction and multistage early warning of landslide body displacement are realized. The technical scheme is that in order to achieve the purpose, the invention provides a large-range landslide body displacement early warning method based on GNSS and radar data fusion, which comprises the following steps: s1, acquiring GNSS monitoring sequence data in a landslide area based on position information of a reference station and a GNSS monitoring station in a GNSS monitoring system to further obtain landslide area displacement related information; s2, acquiring data through a satellite InSAR, and generating deformation field data of a landslide region through comparing radar images at different time points; S3, carrying out space-time registration on the GNSS displacement information obtained in the step S1 and the radar deformation field data obtained in the step S2, and fusing the acquired GNSS displacement information and the radar deformation field data by using a self-adaptive double-weight Kalman filtering method; s4, decomposing time sequence data of the landslide body displacement by utilizing wavelet transformation to extract multi-scale characteristics of a time domain and a frequency domain; S5, constructing a landslide body displacement trend prediction model by using an LSTM method added with a memory decay factor according to the time-frequency characteristics extracted in the S4 and combining with historical data of landslide displacement, and predicting the future trend of the landslide displacement; And S6, setting a multistage threshold value, judging the change tr