CN117236518B - Prediction method for geological deformation of subway along line - Google Patents
Prediction method for geological deformation of subway along line Download PDFInfo
- Publication number
- CN117236518B CN117236518B CN202311489535.5A CN202311489535A CN117236518B CN 117236518 B CN117236518 B CN 117236518B CN 202311489535 A CN202311489535 A CN 202311489535A CN 117236518 B CN117236518 B CN 117236518B
- Authority
- CN
- China
- Prior art keywords
- deformation
- subway
- lstm
- cnn
- prediction
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 62
- 238000012544 monitoring process Methods 0.000 claims abstract description 35
- 238000000605 extraction Methods 0.000 claims abstract description 7
- 238000013527 convolutional neural network Methods 0.000 claims description 40
- 238000005516 engineering process Methods 0.000 claims description 23
- 238000012545 processing Methods 0.000 claims description 23
- 238000004062 sedimentation Methods 0.000 claims description 23
- 238000004364 calculation method Methods 0.000 claims description 18
- 238000010276 construction Methods 0.000 claims description 17
- 230000007246 mechanism Effects 0.000 claims description 17
- 238000004458 analytical method Methods 0.000 claims description 15
- 230000006870 function Effects 0.000 claims description 14
- 238000004422 calculation algorithm Methods 0.000 claims description 12
- 230000015654 memory Effects 0.000 claims description 12
- 238000011160 research Methods 0.000 claims description 12
- 238000005457 optimization Methods 0.000 claims description 10
- 230000008569 process Effects 0.000 claims description 10
- 230000004913 activation Effects 0.000 claims description 9
- 239000006185 dispersion Substances 0.000 claims description 9
- 230000006403 short-term memory Effects 0.000 claims description 6
- 230000007787 long-term memory Effects 0.000 claims description 5
- 238000012876 topography Methods 0.000 claims description 5
- 238000012937 correction Methods 0.000 claims description 3
- 239000011159 matrix material Substances 0.000 claims description 3
- 230000006855 networking Effects 0.000 claims description 3
- 238000003672 processing method Methods 0.000 claims description 3
- 238000000926 separation method Methods 0.000 claims description 3
- 230000001186 cumulative effect Effects 0.000 claims description 2
- 101001095088 Homo sapiens Melanoma antigen preferentially expressed in tumors Proteins 0.000 claims 1
- 102100037020 Melanoma antigen preferentially expressed in tumors Human genes 0.000 claims 1
- 210000004027 cell Anatomy 0.000 description 16
- 238000010586 diagram Methods 0.000 description 8
- 238000012549 training Methods 0.000 description 8
- 238000013528 artificial neural network Methods 0.000 description 7
- 238000011156 evaluation Methods 0.000 description 7
- 238000012360 testing method Methods 0.000 description 7
- 230000008859 change Effects 0.000 description 6
- 238000011176 pooling Methods 0.000 description 6
- 230000000694 effects Effects 0.000 description 5
- 239000000284 extract Substances 0.000 description 5
- 238000004088 simulation Methods 0.000 description 4
- 238000013135 deep learning Methods 0.000 description 3
- 230000001413 cellular effect Effects 0.000 description 2
- 230000008034 disappearance Effects 0.000 description 2
- 238000004880 explosion Methods 0.000 description 2
- 238000003384 imaging method Methods 0.000 description 2
- 238000011835 investigation Methods 0.000 description 2
- 230000007774 longterm Effects 0.000 description 2
- 238000005065 mining Methods 0.000 description 2
- 210000002569 neuron Anatomy 0.000 description 2
- 230000000306 recurrent effect Effects 0.000 description 2
- 239000004575 stone Substances 0.000 description 2
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 210000004556 brain Anatomy 0.000 description 1
- 230000001149 cognitive effect Effects 0.000 description 1
- 238000007596 consolidation process Methods 0.000 description 1
- 230000002596 correlated effect Effects 0.000 description 1
- 238000013434 data augmentation Methods 0.000 description 1
- 230000009849 deactivation Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 230000002068 genetic effect Effects 0.000 description 1
- 238000011478 gradient descent method Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000005305 interferometry Methods 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- YHXISWVBGDMDLQ-UHFFFAOYSA-N moclobemide Chemical compound C1=CC(Cl)=CC=C1C(=O)NCCN1CCOCC1 YHXISWVBGDMDLQ-UHFFFAOYSA-N 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 239000002245 particle Substances 0.000 description 1
- 230000002093 peripheral effect Effects 0.000 description 1
- 238000005191 phase separation Methods 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 238000010187 selection method Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 1
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention provides a prediction method for geological deformation of a subway along a line, which relates to the technical field of deformation prediction and comprises the following steps: sentinel-1A image data of different strips of the same track in multiple views are collected and processed, settlement time sequence data of a subway line in a deformation period is obtained through inversion, space-time evolution characteristics and uneven deformation degrees of ground surface settlement are analyzed, the mutual interference between monitoring points and surrounding points is used as influence characteristic input, and a multi-characteristic CNN-LSTM-attribute prediction model considering deformation relevance of adjacent points is constructed. According to the method, the PS-InSAR monitoring result and the CNN-LSTM-Attention prediction model are combined to obtain the prediction result of the model, so that the identification, extraction and prediction of the deformation information of the subway along the line can be effectively realized, and data and technical support can be provided for the monitoring of the subway along the line and the establishment of a safety plan.
Description
Technical Field
The invention relates to the technical field of deformation prediction, in particular to a prediction method for geological deformation of a subway along a line.
Background
At present, urban construction suddenly advances, and subways become important transportation hubs of cities. Active underground engineering can damage the stability of loose layers, causing local overall subsidence or road and building damage. Ground subsidence has become the most dominant disaster of subway construction in major cities at home and abroad, and the phenomenon is more likely to occur in stone families due to the fragile geological structure and the long-term super-mining underground water activity background. Therefore, the method can be used for timely and effectively monitoring ground subsidence, analyzing subsidence distribution characteristics and constructing a deformation prediction model, and has important practical significance for preventing secondary disasters caused by subway construction.
In the aspect of subway line deformation monitoring, the conventional measuring means (precision level, GPS and the like) have the advantage of high monitoring precision, but the monitoring space resolution is low, and the long-time sequence fine monitoring of the ground surfaces of a plurality of lines is difficult to realize at the same time. The development of synthetic aperture radar interferometry (Interferometric Synthetic Aperture Radar, inSAR), particularly time-series InSAR techniques (e.g., PS, SBAS), has enabled a wide range of sophisticated earth surface deformation monitoring. The method comprises the steps of obtaining deformation information of Guangzhou Datansha subways, mountain and island main lines of hong Kong Dayu mountain and Shenzhen subways No. 4 lines by using IPTA technology with a small amount of multi-source SAR data, which shows that large-scale linear ground object deformation is influenced by natural and human activities; the PS-InSAR technology is adopted to obtain 2007-2016 sedimentation characteristics of the Hexi region, and factors such as subway sedimentation values, sedimentation rates and the like are comprehensively considered, and influence of sedimentation on subway structure damage is quantitatively evaluated by utilizing entropy values; sedimentation information was obtained in the martial arts 2012-2019 using COSMO-SkyMed data, and route sedimentation factors were analyzed. In addition, a plurality of scholars also apply the time sequence InSAR technology to the deformation monitoring of the linear facilities, and the feasibility of the technology in subway deformation monitoring is verified.
In the face of the problem of settlement prediction and early warning, the traditional theoretical analysis and numerical simulation methods, such as a finite element method, a finite difference method, a finite volume method and the like, have the defects of complicated model construction, large calculation amount and incapability of considering deformation mechanisms and complex influence factors. The deep learning algorithm analyzes the characteristics of sedimentation influencing factors through strong learning capability, can effectively filter redundant data, excavates the time sequence change rule of sedimentation, and realizes accurate prediction of surface sedimentation. At present, the deep learning algorithm is widely applied to various settlement predictions. Such as: taking a second line of the Shanghai subway as a research object, optimizing a BP neural network model by using a genetic algorithm and a particle swarm algorithm to carry out settlement prediction, and verifying that the convergence rate and the prediction effect of the optimization algorithm are superior to those of the traditional BP neural network; the time-varying parameters of the optimized background value are introduced to solve the problem that the fitting and prediction precision of the GM (1, 1) model is not ideal, taking the settlement observation point from He Haote to quaigher railway engineering as an example, and verifying the prediction precision of the model after optimization; after denoising time sequence data by wavelet analysis by taking subway foundation pit engineering in Beijing city as a research object, experiments prove that the ARIMA model is more suitable for short-term prediction than the BP neural network.
At present, a certain research result exists in the time sequence monitoring and predictive analysis of subway sedimentation by applying an InSAR technology and a deep learning algorithm, but certain problems exist: ① Most research results are limited to deformation analysis of the ground surface in the vertical direction or the visual line direction, but few analysis methods are used for analyzing the non-uniformity degree of subway deformation by correlating sedimentation and distance; ② Most of the existing researches only consider the time sequence change of the monitoring point position when deformation prediction is carried out, and do not consider the space neighborhood characteristics, namely, the additional influence of the peripheral measuring points on the monitoring point position is ignored.
Disclosure of Invention
The invention provides a prediction method for geological deformation of a subway along a line, and aims to solve the problem that deformation analysis of the vertical direction or the sight direction of the ground surface is limited in the prior art, and the problem that the degree of non-uniformity of deformation of the subway is analyzed by correlating sedimentation and distance is rarely generated.
In order to achieve the above object, the present invention provides the following technical solutions: a prediction method for geological deformation of a subway along a line comprises the following steps:
collecting subway Sentinel-1A image data of different strips of a plurality of scenes on the same track;
processing the subway Sentinel-1A image data by using a time sequence PS-InSAR technology, and inverting to obtain surface subsidence time sequence data of a plurality of subway lines in a period;
constructing a CNN-LSTM-attribute prediction model considering the deformation relevance of adjacent points;
Inputting the earth surface subsidence time sequence data into a CNN-LSTM-Attention prediction model, extracting high-dimensional features by using CNN, excavating time sequence rules of the high-dimensional features by using LSTM, processing feature importance differences by using Attention mechanism Attention, and finally extracting line deformation gradient distribution features along the railway;
and obtaining a prediction result of the deformation of the subway along the line through the line deformation gradient distribution characteristics.
Preferably, the processing the subway Sentinel-1A image data by using the time sequence PS-InSAR technology includes the following steps:
Registering a public main image in the subway Sentinel-1A image data with the rest images by using a public main image algorithm to obtain N interference pairs;
processing the N interference pairs by using a differential interference processing method to obtain an interference pattern after flattening, and obtaining an oblique distance DEM and a synthetic phase;
And carrying out optimization selection of PS points on the interference pattern after the flattening, obtaining amplitude dispersion index distribution, and carrying out PS networking modeling.
Preferably, the inversion obtains the surface subsidence time sequence data of a plurality of subway lines in a period, and the inversion comprises the following steps:
performing a first inversion operation on the model obtained by the PS network construction to obtain a residual terrain phase and a deformation rate;
Performing a second inversion operation on the model obtained by the PS network construction based on the residual topography phase and the deformation rate to obtain a final deformation rate, a coherence coefficient and a final height, and obtaining the separation of the atmospheric delay phases by using an atmospheric correction algorithm;
And performing geocoding on the final deformation rate, the coherence coefficient and the height to obtain final sedimentation time sequence data.
Preferably, the optimization selection of PS points is performed on the interference pattern after the flattening, including the following steps:
setting a threshold value by using an amplitude dispersion method to obtain a PS point;
And optimizing and selecting PS points by a phase stability analysis least square estimation method.
Preferably, when the processing is performed on the multi-scene Sentinel-1A image data by using the time sequence PS-InSAR technology, relevant parameters are required to be set for processing, including the following steps:
setting a time base line, a multi-vision ratio in an interference treatment process, an amplitude deviation index threshold value, a linear rate and an elevation residual error, and setting a space base line to be 45% of a critical base line;
acquiring Sentinel-1A image settlement information within a certain time range by using a time sequence PS-InSAR technology with related parameters;
And splicing the sedimentation results of different strip data of the research area, wherein the specific expression is as follows:
In the method, in the process of the invention, The number of PS points for the common overlap region of the two strips; /(I)And/>The accumulated deformation value of the same name PS points of the two strips; /(I)I.e. the overall deviation of the two strips.
Preferably, the constructing a CNN-LSTM-Attention prediction model considering the deformation relevance of adjacent points comprises the following steps:
The convolutional neural network CNN is used as an input end of a prediction model, so that input network parameters are reduced, and the complexity of data reconstruction in the feature extraction process is reduced;
the LSTM network with memory cells is combined with the convolutional neural network CNN, and the cell state is enhanced through a gate structure;
The outputs of the CNN and LSTM networks are processed in a probability distribution weight way by using an attention mechanism.
Preferably, the method for enhancing cell status through portal structure by combining LSTM network with memory cells and convolutional neural network CNN comprises the following steps:
Constructing an LSTM network through a forget gate, an input gate and an output gate;
The calculation formula of the forgetting gate is as follows:
the input gate calculation formula is as follows:
the output gate calculation formula is as follows:
the cell state is obtained through the outputs of the forgetting gate, the input gate and the output gate, and the specific expression is as follows:
Where h is a hidden state, representing short-term memory, Is the hidden state of the current moment,/>Representing the hidden state at the previous moment; c is the cellular state, representing long-term memory,/>Representing the state of the memory cell at the current moment/>Then representing the cell state at the previous time; /(I)Representing a forgetful door; x represents input,/>Representing the current time input value; /(I)Representing the input gate and the output gate of the device,Representing an output gate; sigma is a Sigmoid activation function; w f、Wi、Wo and W c refer to the weight matrix of the forgetting gate, the input gate, the output gate and the cell state, respectively; b f、bi、bo are respectively forgetting gate, input gate and output gate offset, and Tanh is an activation function.
Preferably, stability along the subway is analyzed through the uneven landmark deformation in the surface subsidence time sequence data, and the specific expression is:
wherein, The deformation gradient is i point; /(I)The deformation value at the point i; /(I)And/>Deformation values of points before and after the point i are respectively; /(I)I is the distance between the front and rear points.
Preferably, other monitoring points around the site P are searched in the CNN-LSTM-Attention prediction model by taking the site P as a center, and the correlation r between the adjacent point and the site P is measured by acquiring a Pearson correlation coefficient, wherein the specific expression is as follows:
wherein, 、/>For two variables,/>And/>For its mean value,/>Is the number of samples.
Preferably, after the prediction result of the deformation of the subway along the line is obtained through the line deformation gradient distribution characteristics, the root mean square error RMSE, the mean absolute percentage error MAPE and the mean absolute error MAE are adopted as indexes for evaluating the prediction performance of the model, and the specific expression is as follows:
wherein, Is a true value,/>Is the predicted value and n is the number of samples.
Compared with the prior art, the invention has the following beneficial effects:
according to the invention, a PS-InSAR monitoring result and a CNN-LSTM-Attention prediction model are combined, a series of settlement time sequence data is obtained by utilizing a PS-InSAR technology, the key of settlement prediction by utilizing the prediction model is that high-dimensional characteristics are extracted by means of CNN, the LSTM is used for mining time sequence rules, the ignored characteristic importance difference is processed by an Attention mechanism, and salient characteristics are extracted, so that the prediction precision and superiority of the model are greatly improved, the identification, extraction and prediction of deformation information along the subway can be effectively realized, and data and technical support can be provided for the monitoring along the subway and the establishment of a safety plan.
Drawings
FIG. 1 is a graph showing the geographical location and subway line distribution of a research area provided by the invention; wherein: (a) A geographical position diagram of a research area, (b) a subway line distribution diagram;
FIG. 2 is a subway construction time and space base line diagram provided by the invention;
FIG. 3 is a diagram of a CNN-LSTM model structure based on an attention mechanism provided by the invention;
FIG. 4 is a process flow diagram of an integrated PS-InSAR and prediction model provided by the present invention;
FIG. 5 is a graph of deformation rate of a subway line provided by the invention;
FIG. 6 is a graph of deformation gradients along the subway provided by the invention;
FIG. 7 is a graph of the results of the evaluation index provided by the present invention; wherein, (a) is a training set and (b) is a test set;
FIG. 8 is a graph of predicted results of different methods according to the present invention;
fig. 9 is a diagram of prediction results of different methods after interpolation provided by the present invention.
Detailed Description
The following describes the embodiments of the present invention further with reference to the accompanying drawings. The following examples are only for more clearly illustrating the technical aspects of the present invention, and are not intended to limit the scope of the present invention.
For understanding and explanation, a method for predicting geological deformation of a subway along a line according to an embodiment of the present invention is described in detail below.
The specific embodiment is as follows:
The urban subway construction affects the stability of the ground surface, and has urgent performance in monitoring and predicting the subsidence of the ground surface along the subway. Based on a time sequence PS-InSAR technology, 31-period Sentinel-1A image data of different strips on the same track are utilized to invert to obtain space-time evolution characteristics of earth surface subsidence of 3 subway lines in Shijialian in 2017-3-14 to 2019-9-24, namely deformation rate and accumulated subsidence information, and the degree of deformation non-uniformity along the line (line deformation gradient distribution characteristics) is analyzed; the interaction of the monitoring points in the space neighborhood is considered, a CNN-LSTM-Attention model which considers the deformation relevance of adjacent points is constructed, multi-feature settlement time sequence prediction is carried out, the mutual interference between the monitoring points and surrounding points is used as influence feature input, and the accuracy and the reliability of the model are verified. Experimental results show that the constructed model has better RMSE, MAPE and MAE values. The PSInSAR combined network model can effectively recognize, extract and predict deformation information along the subway line, and can provide data and technical support for monitoring along the subway line and making a safety plan.
The geographical coordinates of the research area Shijia city are as follows: 113 deg. 30 'to 115 deg. 20' e,37 deg. 27 'to 38 deg. 47' n, as shown in fig. 1. The fourth loose layer with huge thickness is distributed in the region, the consolidation degree is lower, and the risk of surface subsidence is high. At present, three subway lines (partial road sections) of the operation of the Shijia are covered in the main urban area and the south of the Zhengqing county, 60 sites are arranged in total, and the total mileage of the line reaches 78.2km. The time and space base lines of each line are constructed as shown in figure 2.
As shown in fig. 4, the following describes a method for predicting geological deformation of a subway along a line, which includes the following steps:
S1: and collecting subway Sentinel-1A image data of different strips of the same rail with multiple views.
In the step, in order to realize the full coverage of remote sensing images of 3 subway lines of a Shijia, experimental data selected by the invention are two-scene sentel-1A images from different strips on the same track as one period, imaging time is 2017-3-14 to 2019-9-24 (62-scene data in 31 period in total), and the ground resolution of the images is 5m multiplied by 20m. The orbit parameters of the Sentinel-1A image are downloaded according to the imaging time of the image, and the European air office provides Precise Orbit Ephemerides (POD precise orbit determination ephemeris data) of the sentry image. In addition, to eliminate the terrain phase, the external DEM uses the U.S. space agency 30m SRTM-DEM data.
S2: and processing the multi-scene subway Sentinel-1A image data by using a time sequence PS-InSAR technology, inverting to obtain surface subsidence time sequence data of a plurality of subway lines in a period, and analyzing to obtain time-space evolution characteristics and uneven deformation degrees.
The core idea of the PS-InSAR technology is to search a permanent scatterer (namely, PS point) which keeps high coherence on an SAR image of the same area according to amplitude and phase information, build a model based on the PS point, and further obtain deformation information of a target area by removing residual phases such as atmospheric noise, land leveling effect and the like. The method mainly comprises the following steps: differential interference processing, PS point identification, PS network modeling, residual topography phase and deformation rate estimation, atmospheric delay phase separation and the like. The key technology of PS-InSAR data processing comprises public main image optimization selection and PS point optimization selection.
In the step, the method for processing the image data of the multi-scene subway Sentinel-1A by utilizing the time sequence PS-InSAR technology comprises the following steps:
s21: and registering the public main image and the rest images in the multi-scene Sentinel-1A subway image data by using a public main image algorithm to obtain N interference pairs.
S22: and processing the N interference pairs by using a differential interference processing method to obtain an interference pattern after flattening, and obtaining an oblique DEM and a synthetic phase.
S23: and optimizing and selecting PS points of the interference pattern after the flattening by using an amplitude deviation index threshold method, obtaining amplitude deviation index distribution, and carrying out PS networking modeling.
The public main image optimization selection method specifically comprises the following steps: the common main image is registered with the rest images, and the selection principle is to comprehensively consider the influences of a time base line, a space base line and Doppler centroid frequency. The concrete steps are as follows: the sum of the total sum of the time base line and the total sum of the space base line of the interferograms generated after the interference processing of the main image and other auxiliary images is minimum.
The optimization selection of the PS point is specifically as follows: the key point of the PS-InSAR technology is to extract enough PS points with high backscattering coefficients, and the selection of the PS points directly influences the quality of the generated interference pattern. At present, various methods for optimizing and selecting PS points appear: a time-series coherence thresholding method, a phase dispersion exponential thresholding method, an amplitude dispersion exponential method, and the like. However, the above method is limited by factors such as space-time baseline, atmospheric delay, terrain error and the like, and is complex in calculation and time-consuming. Therefore, a learner proposes to set a threshold value to obtain a PS point by an amplitude dispersion method, and then optimize and select the PS point by a phase stability analysis least square estimation method.
In the step, inversion is carried out to obtain surface subsidence time sequence data of a plurality of subway lines in a period so as to obtain the time-space evolution characteristics and the non-uniformity degree of subsidence, and the method comprises the following steps:
s24: and performing first inversion operation on the model obtained by the PS network construction to obtain the residual topography phase and the deformation rate.
S25: and performing a second inversion operation on the model obtained by the PS network construction based on the residual topography phase and the deformation rate to obtain a final deformation rate, a coherence coefficient and a final height, and obtaining the separation of the atmospheric delay phases by using an atmospheric correction algorithm.
S26: and performing geocoding on the final deformation rate, the coherence coefficient and the height to obtain final sedimentation time sequence data.
In this step, when the sequential PS-InSAR technique is used to process the multi-view Sentinel-1A image data, relevant parameters are required to be set for processing, including the following steps:
S27: setting a time base line, a multi-view ratio during the interference treatment, an amplitude dispersion index threshold, a linear rate and an elevation residual, and setting a spatial base line to 45% of a critical base line.
S28: acquiring Sentinel-1A image settlement information within a certain time range by using a time sequence PS-InSAR technology with related parameters.
S29: and carrying out splicing treatment on sedimentation results of different strip data of the research area, wherein the specific expression is as follows:
In the method, in the process of the invention, The number of PS points for the common overlap region of the two strips; /(I)And/>The accumulated deformation value of the same name PS points of the two strips; /(I)I.e. the overall deviation of the two strips.
S3: and constructing a CNN-LSTM-Attention prediction model considering the deformation relevance of the adjacent points.
In the step, the artificial neural network simulates a processing mechanism of the human brain on complex information, and has strong nonlinear mapping capability. The characteristic of the representative Convolutional Neural Network (CNN) that the local connection and the weight share the CNN not only reduces network parameters, but also reduces the complexity of data reconstruction in the characteristic extraction process. Single CNNs do not have memory and thus, in combination with LSTM networks with memory cells, enhance cell status through the gate structure. The CNN-LSTM network random weight distribution mode can cause feature importance difference to be ignored, and the change rule of long-term time sequence data cannot be mined. The invention improves the influence of important features of the CNN-LSTM model, reduces the influence of non-important features and improves the accuracy of the model in a probability distribution weight mode based on an attention mechanism. The method comprises the following specific steps:
S31: the convolutional neural network CNN is used as an input end of the prediction model, so that input network parameters are reduced, and the complexity of data reconstruction in the feature extraction process is reduced.
S32: the LSTM network with memory cells is combined with convolutional neural network CNN to enhance cell status through portal structure.
S321: and constructing the LSTM network through the forget gate, the input gate and the output gate.
The calculation formula of the forgetting gate is as follows:
The input gate calculation formula is as follows:
S322: the output gate calculation formula is as follows:
The cell state is obtained through the outputs of the forgetting gate, the input gate and the output gate, and the specific expression is as follows:
Where h is a hidden state, representing short-term memory, Is the hidden state of the current moment,/>Representing the hidden state at the previous moment; c is the cellular state, representing long-term memory,/>Representing the state of the memory cell at the current moment/>Then representing the cell state at the previous time; /(I)Representing a forgetful door; x represents input,/>Representing the current time input value; /(I)Representing the input gate and the output gate of the device,Representing an output gate; sigma is a Sigmoid activation function; w f、Wi、Wo and W c refer to the weight matrix of the forgetting gate, the input gate, the output gate and the cell state, respectively; b f、bi、bo are respectively forgetting gate, input gate and output gate offset, and Tanh is an activation function.
Wherein CNN (Convolutional Neural Networks) is one of the neural networks, and mainly comprises 5 network layers of a data input layer, a convolution layer, a pooling layer, a full connection layer and an output layer. The convolution layer (convolution) performs feature extraction on the original data, which is equivalent to a filter of signal processing, and the first layer convolution layer may only extract some low-level features, such as edges, lines, angles and other levels, and the network of more layers can iteratively extract more complex features from the low-level features. After convolution, relu activation functions are usually introduced, and the functions have sparsity, small calculation amount and rapid convergence, so that the neural network can be arbitrarily approximated to a nonlinear function. The pooling layer (pooling) compresses the feature data, simplifies network complexity, and extracts the main features. The redundancy of information in the feature space extracted by the convolution operation can increase the calculation difficulty, and the pooling layer can continuously reduce the data space and the calculation amount and control the over-fitting phenomenon.
Long and short term memory networks (Long Short Term Memory, LSTM) were developed from recurrent neural networks (recurrent neural network, RNN). Compared with CNNs insensitive to time series, RNNs have memory on input data, but as the number of network layers gets deeper, gradient updates increase or decrease exponentially, i.e. gradient explosion and gradient disappearance occur, so RNNs have only short-term memory. LSTM is an improvement of the traditional RNN, and the introduced gate control structure can better store and access information, and the dependent length is controlled by learning to obtain the weight, so that the problems of gradient disappearance and explosion are relieved. The LSTM consists of a forget gate, an input gate and an output gate. The input gate decides which new information to add, here Tanh is chosen as the activation function. The calculation expression of each gate is as shown in step S321.
S33: and the outputs of the CNN and the LSTM network are processed in a probability weight distribution mode by using an attention mechanism, so that the model accuracy is improved.
In the step, the current attention mechanism is mostly the focusing attention in the simulation cognitive neurology, focuses on the information which is more critical to the current task in a plurality of input information, reduces the attention degree of other information, thereby extracting the characteristic which has the greatest influence on the model, effectively solving the information overload problem and finally improving the model precision and efficiency.
The CNN-LSTM based on the Attention mechanism forms a CNN-LSTM-Attention model, and mainly comprises a data input layer, a CNN layer, an LSTM layer, an Attention layer and an output layer, and FIG. 3 is a network model structure diagram.
S4: inputting the earth surface subsidence time sequence data into a CNN-LSTM-Attention prediction model, extracting high-dimensional features by using CNN, excavating time sequence rules of the high-dimensional features by using LSTM, processing feature importance difference by using an Attention mechanism, and finally extracting line deformation gradient distribution features, thereby improving prediction precision and efficiency.
In the step, a PS-InSAR monitoring result and a CNN-LSTM-Attention prediction model are combined, a series of settlement time sequence data is obtained by utilizing a PS-InSAR technology, the key of settlement prediction by utilizing the prediction model is to extract high-dimensional features by means of CNN, the LSTM digs a time sequence rule, and the ignored feature importance difference is processed by a Attention mechanism, so as to extract salient features. The prediction model consists of four parts, namely data preprocessing, model training, model testing and outputting. PS-InSAR integrated prediction model flow is shown in figure 4.
The method comprises the following steps:
stability along the subway is analyzed through uneven landmark deformation in the settlement time sequence data, and the specific expression is:
wherein, The deformation gradient is i point; /(I)The deformation value at the point i; /(I)And/>Deformation values of points before and after the point i are respectively; /(I)I is the distance between the front and rear points.
Setting up a settlement prediction model (CNN-LSTM-Attention prediction model) based on a site P, searching other surrounding monitoring points by taking the site P as a center, and acquiring the correlation r of the Pearson correlation coefficient measurement adjacent point and the site P, wherein the specific expression is as follows:
wherein, 、/>For two variables,/>And/>For its mean value,/>Is the number of samples.
S5: and acquiring points serving as influence characteristic input by setting a Pearson correlation coefficient threshold value, and inputting settlement time sequence data of the points into a prediction model to obtain a prediction result of deformation of the subway along the line.
In the step, after the prediction result of the deformation of the subway along the line is obtained through the line deformation gradient distribution characteristics, the root mean square error RMSE, the average absolute percentage error MAPE and the average absolute error MAE are adopted as indexes for evaluating the prediction performance of the model, and the specific expression is as follows:
wherein, Is a true value,/>Is the predicted value and n is the number of samples.
Examples:
1. detection results and analysis:
1. InSAR deformation interpretation.
By using the PS-InSAR data processing flow and method, relevant parameters are set as follows: ① A time base line of 96 days, a space base line of 45% of a critical base line; the longest spatial baseline was 110.15m and the longest spatial baseline was 504 days; ② The multi-view ratio in the interference treatment process is set to be 5:1, and the amplitude dispersion index threshold value is set to be 0.75; ③ The parameter range is that the linear speed is-40 mm/year to 40mm/year, and the elevation residual is-250 m to 250m. The above parameters were used to obtain Sentinel-1A image settlement information during two bands 31, periods 2017-3-14 through 2019-9-24. As shown in fig. 1, since the study area is in the overlapping area of the two strips, the two sets of sedimentation results need to be subjected to a stitching process. The invention calculates the integral deformation deviation among the strips by using the following formula, unifies the coordinate reference standard of one strip to the other strip, and solves the problem of splicing different strip data by unifying the coordinate system and the reference standard. The calculation expression is as shown in step S29. The total of 1527605 PS spots in the study area were detected, with target spots having an average rate standard deviation below 3.5mm/a accounting for 85.6% of the total. The monitoring result is used for analyzing the deformation of the subway along the ground surface, and the reliability is shown.
2. And analyzing the deformation interpretation result.
The deformation of the ground surface in the buffer area 500 m along the subway line is analyzed to obtain the deformation rate of the subway line, as shown in fig. 5. From the deformation results and frequency distribution diagrams, it can be seen that: the settlement of the stone house area is unevenly distributed, the overall settlement of the ground is relatively stable, and partial areas are slightly lifted. The main settlement area is located from Shijiazhuang East Station to Fuze Station on Line 1, with a maximum settlement rate of about -20mm/a, occurring at Xiaohe Avenue Station, with a maximum cumulative settlement of -53mm; the sedimentation rate of other areas is mainly-10 to +5mm/a, the areas with more obvious sedimentation are mainly around the Changan park station, the talk-fixation station and the West-face tomb station, and the maximum sedimentation rates are-10 mm/a, -9 mm/a and-8 mm/a respectively.
3. Stability analysis along subway line
Uneven surface deformation can effectively reflect the stability state of the subway along the line. Therefore, the concept of deformation gradient is introduced, which refers to the magnitude of the deformation change rate of the earth surface in a certain direction, is generally used for describing the spatial uniformity of deformation, and can represent the deformation rate or the degree of deformation change in a certain section along the subway. The calculation formula is shown in step S4.
Calculating deformation gradient of each line based on accumulated settlement of each line in 2017/3-2019/9 period, and extractingSubway centerline points greater than 0.15mm/m are plotted in FIG. 6.
Based on the deformation gradient map of the ground surface along the subway, the following steps are as follows:
1) The maximum value of the linear deformation gradient of No. 1 is 0.18mm/m, which occurs at the West village station, where the deformation rate is only-2.0 mm/a. The settlement change of the central line of the road section from the Jianzhong east station to Fuze station with larger deformation rate is uniform, and no obvious gradient exists.
2) Points with a linear deformation gradient of No. 2 greater than 0.1mm/m are substantially uniformly distributed over the entire line, with a maximum deformation gradient of 0.69mm/m, occurring in the section of the Yuan village-Shi-Jiang station and near the Shi-Jiang station 400 m; through investigation, the position is in the construction period in the research interval. In addition, points of the line gradient values in the interval of 0.3-0.5mm/m are distributed at Liu Xinzhuang standing positions of 90m, 3.8km, 4.7km, 6.7km, 7.1km, 10.5km and 13.2km, so that the sections need to be closely focused in the operation process.
3) The maximum value of the linear deformation gradient No.3 is 0.29mm/m, which occurs in the village station, and the deformation rate of the point is-4.82 mm/a. The station is a joining station for two-stage construction, and is jointly influenced by the north-stage construction and the first-stage operation according to investigation.
Through the analysis, the operation and construction sections of the subway line can be effectively distinguished by taking the surface deformation gradient as an index, and the method can be used as an important reference basis for subway construction and line stability in the operation period.
2. And (5) establishing a sedimentation prediction model.
The interpretation results show that the maximum surface subsidence rate occurs at Xiaohe Avenue Station. It is crucial to study and predict the subsidence situation of the station and its surrounding areas. Taking this as an example (defined as point P), a subsidence prediction model is established.
The deformation reasons of the monitoring points are influenced by external environment factors and human factors, are also influenced by other surrounding monitoring points, and also influence other monitoring points. Taking the second law of geography that the correlation between monitoring points is larger as the distance between features is closer, other surrounding monitoring points are searched by taking the P point as the center. Because the earth surface is mostly bare land and vegetation, the number of PS monitoring points formed is small, 5 monitoring points are searched only in the range of P point 25m, and other points with larger distances do not have the condition of being used as characteristic input. The correlation of the nearby point and the P point is measured using the Pearson correlation coefficient calculated as the formula in step S4. The Pearson coefficients for each monitoring point and P point are shown in table 1.
TABLE 1 correlation coefficient results
The calculation results show that r of P1 to P5 and P point is larger than 0.8, which shows that the influence characteristics of the prediction model with strong correlation and multiple inputs and single outputs can be used.
1. Prediction parameters
The invention takes the first 26 groups of data of 31 groups of sample data as a training set and the last 5 groups as a test set, and the length of an input history series and the prediction step length are both 1. The input of the CNN-LSTM-Attention model is a two-dimensional structure of 31 multiplied by 5, so that Conv2D two-dimensional convolution layers are constructed, the number of layers is two, the number of convolution kernels of the first Conv2D layer is 32, and the size is 3 multiplied by 3; the number of Conv2D convolution kernels of the second layer is 64, and the size is 3 multiplied by 3. The two convolution layer activation functions are relu functions, and the pooling mode is two-dimensional global average pooling. The number of the LSTM hidden layer neurons is 32, the weight and bias of the LSTM network are updated by adopting an Adam gradient descent method, the maximum iteration number is 300, the initial learning rate is 0.001, the descent factor is 0.25, and finally, partial neurons are randomly deactivated for preventing overfitting, so that the Dropout layer is arranged to improve the model generalization capability, and the deactivation rate is 0.25.
In order to measure the prediction effect of the CNN-LSTM-attribute multi-feature prediction model, the root mean square error RMSE (root mean square error), the average absolute percentage error MAPE (Mean Absolute Percentage Error) and the average absolute error MAE (mean absolute error) are adopted as indexes for evaluating the prediction performance of the model.
2. Experimental results and analysis
In order to evaluate the prediction accuracy and the fitting generalization capability of the model, decision coefficients are introducedAnd the two evaluation indexes of the RMSE judge whether the model is over-fitted. The result of evaluating the goodness of the regression model is shown in fig. 7, wherein (a) is a training set and (b) is a test set, and the two indexes of the training set and the test set have no significant difference, so that the model can be considered to have better generalization ability prediction performance.
The invention compares the prediction result of CNN-LSTM-attribute with CNN-LSTM model, LSTM model and CNN model respectively, and the result is shown in figure 8.
As shown in table 2, the prediction result evaluation index pair of the model and other models is generally smaller than the RMSE, MAPE and MAE of the CNN-LSTM-Attention model than the CNN-LSTM, LSTM and CNN models, and compared with the combination model CNN-LSTM, the prediction accuracy of the combination model CNN-LSTM is better than that of two single models LSTM and CNN models.
Table 2 evaluation index of each predicted result
3. Simulation test and analysis
In order to remove the influence of inconsistent time intervals and data discreteness on model prediction results, the prediction performance of the model used by the method is further verified, and limited data are enabled to generate more equivalent data by utilizing a data enhancement (Data Augmentation) means to artificially expand a training data set, so that the network learning capacity is improved, and the training data distribution situation is fitted conveniently. And converting 31-period data with unequal intervals into 301-period monitoring data of 1 time/3 d by adopting a cubic spline interpolation method, taking the first 260 period as a training set and the later 51 period as a testing set, and carrying out simulation experiment prediction. The interpolated data are input into the CNN-LSTM-Attention model provided by the invention, and the comparison result and evaluation index of the prediction result and other models are shown in the table 3 and the figure 9.
TABLE 3 evaluation index of each prediction result
The deformation conditions of the P point and each influence characteristic point simulated by the cubic spline function can be seen, the interpolated data is better in convergence and smoothness, and the influence of unequal time intervals on the time sequence prediction accuracy is avoided. From the model prediction results and the evaluation index results, the following points can be seen:
(1) The RMSE, MAPE and MAE of the CNN-LSTM model (CNN-LSTM-Attention model) that introduced the Attention mechanism to assign weights according to probability were reduced by 0.16%, 0.17% and 0.137% respectively, compared to the CNN-LSTM model that did not introduce the Attention mechanism but randomly assigned weights.
(2) The combination model CNN-LSTM is reduced by 0.117 percent and 0.356 percent compared with the RMSE of two single models LSTM and CNN respectively; MAPE is reduced by 0.23% and 1.52% respectively; MAE was reduced by 0.029% and 0.257%, respectively.
(3) CNNs possess local weight sharing and translational invariance to better extract features, but because of lack of memory mechanisms, they are less suitable for processing sequence data, and in the case of smaller data volume, CNN prediction accuracy is better than LSTM, and as the data volume increases, CNN prediction accuracy is lower than that of LSTM with long-term memory function.
The invention has the following conclusion:
(1) According to the invention, the PS-InSAR technology is utilized to obtain the deformation results of images of different strips of the same rail on the surface of the 3 subway lines along the line of the Shijia, and the deformation results of the two groups of data are spliced by calculating the integral deformation deviation of the two groups of data and unifying the coordinate reference standard of the two strips of data.
(2) According to PS-InSAR interpretation results, several obvious deformation areas are calibrated, accumulated deformation values of the central line of the subway after the Kriging interpolation are extracted for deformation gradient analysis, the fact that the central line of the No. 1 line is stable in deformation along the line is found, the larger point of the linear deformation gradient of the No. 3 line appears in a village station in the city, and the whole line of the No. 2 line finds a plurality of points with larger gradient values, which are all places needing key monitoring and maintenance in the daily operation process of the subway.
(3) The present invention selects the settlement point of Xiaohe Avenue Station with the largest settlement as the prediction target. Considering that the settlement influencing factors are not only human and natural factors, but also the deformation correlation between adjacent points, five strongly correlated points around the point are selected as input for the influence feature. A CNN-LSTM Attention multi-dimensional prediction model is established, and the prediction results are compared with CNN-LSTM, LSTM, and CNN to obtain the highest accuracy of the model.
(4) In order to further verify the prediction accuracy and superiority of the model of the invention, the original data is interpolated into high-frequency equidistant time sequence data, and the prediction errors of the four models are reduced compared with those before interpolation, wherein the accuracy of the CNN-LSTM-Attention model is highest.
The above embodiments are merely preferred embodiments of the present invention, the protection scope of the present invention is not limited thereto, and any simple changes or equivalent substitutions of technical solutions that can be obviously obtained by those skilled in the art within the technical scope of the present invention disclosed in the present invention belong to the protection scope of the present invention.
Claims (6)
1. A prediction method for geological deformation of a subway along a line is characterized by comprising the following steps:
collecting subway Sentinel-1A image data of different strips of a plurality of scenes on the same track;
processing the subway Sentinel-1A image data by using a time sequence PS-InSAR technology, and inverting to obtain surface subsidence time sequence data of a plurality of subway lines in a period;
analyzing the stability of the subway along the line through the uneven surface deformation in the surface subsidence time sequence data, wherein the specific expression is as follows:
Wherein grad i is the i-point deformation gradient; d i is the deformation value at the point i; d i+1 and D i-1 are deformation values of a point before and after the point i respectively; s i+1,i-1 is the distance between the front point and the rear point of i;
constructing a CNN-LSTM-attribute prediction model considering the deformation relevance of adjacent points;
Inputting the earth surface subsidence time sequence data into a CNN-LSTM-Attention prediction model, extracting high-dimensional features by using CNN, excavating time sequence rules of the high-dimensional features by using LSTM, processing feature importance differences by using Attention mechanism Attention, and finally extracting line deformation gradient distribution features along the railway;
in the CNN-LSTM-Attention prediction model, other monitoring points around the site P are searched by taking the site P as a center, and the correlation r between the adjacent point and the site P is measured by using the Pearson correlation coefficient, wherein the specific expression is as follows:
wherein X i、Yi is two variables, And/>N is the number of samples, which is the average value;
obtaining a prediction result of deformation of the subway along the line through the line deformation gradient distribution characteristics;
The processing of the subway Sentinel-1A image data by using the time sequence PS-InSAR technology comprises the following steps:
Registering a public main image in the subway Sentinel-1A image data with the rest images by using a public main image algorithm to obtain N interference pairs;
processing the N interference pairs by using a differential interference processing method to obtain an interference pattern after flattening, and obtaining an oblique distance DEM and a synthetic phase;
performing optimization selection of PS points on the interference pattern after the flattening to obtain amplitude dispersion index distribution, and performing PS networking modeling;
the inversion method comprises the following steps of:
performing a first inversion operation on the model obtained by the PS network construction to obtain a residual terrain phase and a deformation rate;
Performing a second inversion operation on the model obtained by the PS network construction based on the residual topography phase and the deformation rate to obtain a final deformation rate, a coherence coefficient and a final height, and obtaining the separation of the atmospheric delay phases by using an atmospheric correction algorithm;
And performing geocoding on the final deformation rate, the coherence coefficient and the height to obtain final sedimentation time sequence data.
2. The method for predicting geological deformation of a subway line according to claim 1, wherein the optimization selection of PS points is performed on the interference pattern after the flattening, and the method comprises the following steps:
setting a threshold value by using an amplitude dispersion method to obtain a PS point;
And optimizing and selecting PS points by a phase stability analysis least square estimation method.
3. The method for predicting geological deformation along a subway line according to claim 1, wherein when the subway Sentinel-1A image data is processed by using a time sequence PS-InSAR technology, relevant parameters are required to be set for processing, and the method comprises the following steps:
Setting a time base line, a multi-vision ratio in an interference treatment process, an amplitude deviation index threshold value, a linear rate and an elevation residual error, and setting a space base line to be 45% of a critical base line;
acquiring Sentinel-1A image settlement information within a certain time range by using a time sequence PS-InSAR technology with related parameters;
And carrying out splicing treatment on sedimentation results of different strip data of the research area, wherein the specific expression is as follows:
Wherein N i is the number of PS points in the common overlapping area of the two strips; d j and d' j are cumulative deformation values of the same-name PS points of the two strips; Δd i is the overall deviation of the two bands.
4. The method for predicting geological deformation of a subway line according to claim 1, wherein the constructing of the CNN-LSTM-Attention prediction model considering the deformation relevance of adjacent points comprises the following steps:
The convolutional neural network CNN is used as an input end of a prediction model, so that input network parameters are reduced, and the complexity of data reconstruction in the feature extraction process is reduced;
the LSTM network with memory cells is combined with the convolutional neural network CNN, and the cell state is enhanced through a gate structure;
The outputs of the CNN and LSTM networks are processed in a probability distribution weight way by using an attention mechanism.
5. The method for predicting geological deformation of subway line according to claim 4, wherein said using LSTM network with memory cells combined with convolutional neural network CNN to enhance cell state through gate structure comprises the steps of:
Constructing an LSTM network through a forget gate, an input gate and an output gate;
The calculation formula of the forgetting gate is as follows:
ft＝σ(Wf·[ht-1,xt]+bf)
the input gate calculation formula is as follows:
it＝σ(Wi·[ht-1,xt]+bi)
the output gate calculation formula is as follows:
ot＝σ(Wo·[ht-1,xt]+bo)
ht＝ot*Tanh(Ct)
the cell state is obtained through the outputs of the forgetting gate, the input gate and the output gate, and the specific expression is as follows:
Wherein h is a hidden state, which represents short-term memory, h t is a hidden state at the current time, and h t-1 represents a hidden state at the previous time; c is the cell state, which is the long-term memory, C t is the memory cell state at the current time, and C t-1 is the cell state at the previous time; f t represents a forgetting door; x represents input, x t represents input value at current time; i t represents an input gate, o t represents an output gate; sigma is a Sigmoid activation function; w f、Wi、Wo and W c refer to the weight matrix of the forgetting gate, the input gate, the output gate and the cell state, respectively; b f、bi、bo are respectively forgetting gate, input gate and output gate offset, and Tanh is an activation function.
6. The method for predicting the geological deformation of the subway line according to claim 1, wherein after a prediction result of the deformation of the subway line is obtained through the line deformation gradient distribution characteristics, a root mean square error RMSE, a mean absolute percentage error MAPE and a mean absolute error MAE are adopted as indexes for evaluating the prediction performance of a model, and the specific expression is as follows:
where y i is the true value, Is the predicted value and n is the number of samples.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311489535.5A CN117236518B (en) | 2023-11-10 | 2023-11-10 | Prediction method for geological deformation of subway along line |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311489535.5A CN117236518B (en) | 2023-11-10 | 2023-11-10 | Prediction method for geological deformation of subway along line |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117236518A CN117236518A (en) | 2023-12-15 |
| CN117236518B true CN117236518B (en) | 2024-05-24 |
Family
ID=89086332
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311489535.5A Active CN117236518B (en) | 2023-11-10 | 2023-11-10 | Prediction method for geological deformation of subway along line |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117236518B (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108663017A (en) * | 2018-08-13 | 2018-10-16 | 伟志股份公司 | A kind of method of ground settlement along monitoring city underground |
| CN111680343A (en) * | 2020-04-27 | 2020-09-18 | 中煤科工集团武汉设计研究院有限公司 | Deep foundation pit support structure deformation prediction method based on deep learning |
| CN113204824A (en) * | 2021-05-21 | 2021-08-03 | 上海大学 | Multi-model fusion shield construction settlement prediction method and system |
| CN115077474A (en) * | 2022-06-14 | 2022-09-20 | 天津华北地质勘查总院 | Ground settlement trend prediction method and system based on machine learning |
| CN115494501A (en) * | 2022-09-29 | 2022-12-20 | 南宁市勘测设计院集团有限公司 | Urban infrastructure deformation monitoring method based on high-resolution PS-InSAR |
-
2023
- 2023-11-10 CN CN202311489535.5A patent/CN117236518B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108663017A (en) * | 2018-08-13 | 2018-10-16 | 伟志股份公司 | A kind of method of ground settlement along monitoring city underground |
| CN111680343A (en) * | 2020-04-27 | 2020-09-18 | 中煤科工集团武汉设计研究院有限公司 | Deep foundation pit support structure deformation prediction method based on deep learning |
| CN113204824A (en) * | 2021-05-21 | 2021-08-03 | 上海大学 | Multi-model fusion shield construction settlement prediction method and system |
| CN115077474A (en) * | 2022-06-14 | 2022-09-20 | 天津华北地质勘查总院 | Ground settlement trend prediction method and system based on machine learning |
| CN115494501A (en) * | 2022-09-29 | 2022-12-20 | 南宁市勘测设计院集团有限公司 | Urban infrastructure deformation monitoring method based on high-resolution PS-InSAR |
Non-Patent Citations (3)
| Title |
|---|
| 京津冀地区高速铁路沿线区域地表沉降监测及时序演化态势分析;杨斌 等;中国铁道科学(第01期);全文 * |
| 佛山地铁沿线时序InSAR形变时空特征分析;刘琦 等;武汉大学学报(信息科学版)(第07期);全文 * |
| 基于LSTM-GM神经网络模型的深基坑沉降变形预测;袁志明 等;江西理工大学学报(第01期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117236518A (en) | 2023-12-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110986747B (en) | Landslide displacement combined prediction method and system | |
| CN110232471B (en) | Rainfall sensor network node layout optimization method and device | |
| KR100982447B1 (en) | Landslide occurrence prediction system and predicting method using the same | |
| Ding et al. | Monitoring, analyzing and predicting urban surface subsidence: A case study of Wuhan City, China | |
| CN116596321B (en) | Building risk monitoring and evaluating method and system based on neural network | |
| CN107798425A (en) | A kind of space-time based on big data obscures degrees of exposure assessment system and method | |
| CN109615118A (en) | Based on big data hazards control Informatization Service integrated control system and method | |
| CN112882032B (en) | Method and device for dynamically monitoring geological disaster SAR in key area of gas pipeline | |
| CN113251914A (en) | Surface deformation prediction method combining InSAR technology and long-term memory neural network | |
| CN113487105B (en) | Landslide geological disaster space prediction method and device and computer equipment | |
| CN111709609A (en) | Geological disaster susceptibility evaluation method | |
| CN113536659B (en) | Method, system and storage medium for rapidly predicting post-earthquake road disaster area | |
| CN116822185B (en) | Daily precipitation data space simulation method and system based on HASM | |
| CN112097733A (en) | Surface deformation monitoring method combining InSAR technology and geographic detector | |
| Feng et al. | Automatic selection of permanent scatterers-based GCPs for refinement and reflattening in InSAR DEM generation | |
| CN114170524A (en) | Single-day tidal flat digital terrain construction method based on high-time-space satellite data fusion | |
| An et al. | Ground subsidence monitoring in based on UAV-LiDAR technology: a case study of a mine in the Ordos, China | |
| CN113126093A (en) | Geological early warning method | |
| CN117236518B (en) | Prediction method for geological deformation of subway along line | |
| CN117113032A (en) | SBAS-InSAR ground subsidence prediction method considering ground load | |
| CN116307857A (en) | Built-up area environment evaluation method based on remote sensing image scene understanding | |
| Liu et al. | Road density analysis based on skeleton partitioning for road generalization | |
| CN114239379A (en) | Transmission line geological disaster analysis method and system based on deformation detection | |
| Li et al. | Detecting, Monitoring, and Analyzing the Surface Subsidence in the Yellow River Delta (China) Combined with CenterNet Network and SBAS‐InSAR | |
| Prakash et al. | Assessment of Urban Built-Up Volume Using Geospatial Methods: A Case Study of Bangalore |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |