Mine goaf intelligent adaptation earth surface deformation monitoring method
By establishing a body deformation monitoring network and using MP-DS-InSAR technology, combining deep learning models for intelligent adaptation correction, the problem of insufficient aging and accuracy of traditional monitoring methods in mining goaf is solved, and high-precision surface deformation monitoring is achieved.
Patent Information
- Application Number
- CN202510190570.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-02-20
- Publication Date
- 2025-06-20
AI Technical Summary
Traditional surface deformation monitoring methods are difficult to meet the needs in terms of mining goaf aging, coverage and monitoring accuracy. The monitoring methods based on InSAR technology have problems such as insufficient scatterer identification, sparse interference points and sparseness of low interference areas, resulting in insufficient deformation monitoring accuracy.
A solid deformation monitoring network is established by using three-dimensional satellite monitoring, drone aerial survey and GNSS ground monitoring, and multi-dimensional fusion data are generated by combining MP-DS-InSAR monitoring technology, and data fusion and correction are performed through Kalman filtering and least squares adjustment models. A deep learning model is used for intelligent adaptation correction, which removes noise, fills in data loss, and corrects system errors.
High-precision monitoring of surface deformation of mining goaf is achieved, the timeliness and accuracy of monitoring results is improved, and the overall deformation trend can be effectively reflected and the monitoring needs under complex geological conditions.
Smart Images
Figure CN120176584A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of terrain monitoring systems, and in particular to an intelligent adaptive surface deformation monitoring method for mined - out areas of mines. Background Technique
[0002] With the extensive development and utilization of mineral resources, underground mining activities are becoming increasingly frequent, and underground spaces are gradually hollowed out. Under the influence of gravity and surface activities, it ultimately leads to surface settlement and causes surface deformation. This kind of deformation not only poses a potential threat to surface buildings, infrastructure and residents' lives, but also may induce a series of secondary disasters, such as ground collapse, landslide, etc., seriously threatening the country's ecological security and social and economic development. Traditional surface deformation monitoring methods mainly rely on ground observations and underground monitoring. However, with the expansion of the scale and quantity of mined - out areas, and the increasing complexity of geological conditions, traditional monitoring means are difficult to meet the actual needs in terms of timeliness, coverage and monitoring accuracy. At present, although the monitoring method based on InSAR technology is widely used in the surface deformation monitoring of mine subsidence areas, there are also certain limitations, such as: ① insufficient scatterer identification; ② sparse interferometric points, especially in low - signal - to - noise ratio areas, it is difficult to accurately obtain surface deformation information; ③ low density of interferometric points; ⑤ the sparsity of low - interference areas leads to insufficient deformation monitoring accuracy and cannot effectively reflect the overall deformation trend, and there is an urgent need for further optimization and improvement.
[0003] Based on this, the present invention designs an intelligent adaptive surface deformation monitoring method for mined - out areas of mines to solve the above problems. Summary of the Invention
[0004] In view of the above - mentioned shortcomings of the prior art, the present invention provides an intelligent adaptive surface deformation monitoring method for mined - out areas of mines.
[0005] To achieve the above objectives, the present invention is realized through the following technical solutions:
[0006] An intelligent adaptive surface deformation monitoring method for mined - out areas of mines, comprising the following steps:
[0007] Step 1, based on stereo satellite monitoring, UAV aerial survey and GNSS ground monitoring, build a stereo deformation monitoring network with different monitoring accuracies and monitoring densities for mined - out areas of mines. The monitoring data in the stereo deformation monitoring network is generated into MP - DS - InSAR monitoring data based on MP - DS - InSAR monitoring technology;
[0008] Step 2, uniformly process the monitoring data with different accuracies, different spatial scales and different time scales and generate multi - dimensional fusion data; and adopt the following method:
[0009] Based on the use of Kalman Filter to dynamically fuse the monitoring data of mined - out areas at different time intervals;
[0010] Based on the Least Squares Adjustment model, jointly process the high - precision positions of GNSS points and the topographic data obtained from UAV aerial surveys;
[0011] Step three, perform intelligent adaptation correction on the MP - DS - InSAR monitoring results. Based on the deep - learning model, learn from the above - mentioned multi - dimensional fusion data, remove noise, fill in missing data, correct systematic errors, optimize the model, and output the corrected high - precision monitoring data.
[0012] Furthermore, the construction steps A of the MP - DS - InSAR monitoring technology:
[0013] Step A1, obtain monitoring data one from stereo satellite monitoring, monitoring data two from UAV aerial surveys, and monitoring data three from GNSS ground monitoring. Each monitoring data has multiple characteristic variables;
[0014] Calculate the covariance matrix C of monitoring data one, monitoring data two, and monitoring data three;
[0015] Step A2, perform spectral decomposition on the covariance matrix C based on the Principal Component Analysis (PCA) model, and finally construct the MP - DS (Principal Component Analysis Model - Distributed Scatterers, MP - DS) scatterer statistical model;
[0016] Step A3, adjust the weights of each interference pair of the MP - DS scatterer statistical model monitoring data based on the Elastic Net regularization model and the Fisher information optimization model;
[0017] Step A4, extract and analyze the effective information in the interference pairs based on the Dense Connectivity Convolutional Neural Network (Dense Net) and the Attention - mechanism Long Short - Term Memory Network (LSTM);
[0018] Step A5, based on the spatio - temporal filter and the spatio - temporal model of atmospheric interference, construct an adaptive filtering technology and an atmospheric correction algorithm to improve the signal - to - noise ratio of the monitoring data;
[0019] Step A6, construct the MP - DS - InSAR monitoring technology based on the above steps.
[0020] Furthermore, the Kalman Filter mathematical model:
[0021]
[0022] Wherein: is the state estimate at the k-th moment, K k is the Kalman gain, z k is the observed value, H k is the observation matrix.
[0023] Furthermore, the Least Squares Adjustment mathematical model;
[0024]
[0025] Wherein: z i is the actually observed elevation value, f(x i , y i ) is the fitted elevation value based on aerial survey data, (x i , y i ) is the plane coordinate.
[0026] Furthermore, the intelligent adaptation correction step B of the MP-DS-InSAR monitoring result:
[0027] Step B1, removing the noise in the MP-DS-InSAR monitoring data based on the Denoising Autoencoder (DAE);
[0028] Step B2, fusing the above multi-dimensional fusion data with the denoised MP-DS-InSAR monitoring data based on a multi-channel Convolutional Neural Network (CNN);
[0029] Step B3, guiding and correcting the MP-DS-InSAR monitoring data based on the LSTM model;
[0030] Step B4, filling the missing data in the MP-DS-InSAR monitoring data based on the Generative Adversarial Network (GAN) and generating high-quality correction data in areas with severe noise;
[0031] Step B5, training the deep learning model based on the cross-validation technique, evaluating the correction effects of the LSTM model and the Generative Adversarial Network (GAN) using the Mean Squared Error (MSE) or the Adversarial Loss, and adjusting the weights of the LSTM model and the Generative Adversarial Network (GAN) to reduce the prediction error;
[0032] Step B6, optimizing the intelligent adaptation correction model through the Incremental Learning and Online Learning models to adapt to new data patterns.
[0033] Step B7, output the high-precision monitoring data after intelligent adaptive correction; specifically:
[0034]
[0035] where: X Corrected is the finally corrected MP-DS-InSAR data, including the results of denoising, correction, and incompleteness, to meet the high-precision monitoring requirements.
[0036] Furthermore, the DAE removes noise through reconstruction, and its loss function is defined as:
[0037]
[0038] where:
[0039] where: n is the time step; m is the spatial dimension; X MP is the MP-DS-InSAR monitoring data; X Fusion is the multi-dimensional fusion data; N is the missing data mask matrix; is the denoised MP-DS-InSAR data; f DAE is the mapping function of the deep convolutional denoising autoencoder (DAE), responsible for removing noise.
[0040] Furthermore, the multi-channel convolutional neural network (CNN) fuses the multi-dimensional fusion data and the denoised MP-DS-InSAR monitoring data, specifically as follows:
[0041]
[0042] Then, use the LSTM model to modify the time series data, and the specific mathematical model is as follows:
[0043]
[0044] Its loss model is as follows:
[0045]
[0046] where: f CNN is the mapping function of the multi-channel convolutional neural network (CNN) for the denoised MP-DS-InSAR data and other auxiliary data X Fusion ; X CNN is to extract the spatial and temporal features of the data; f LSTM is the correction of the time series data; L LSTM is the loss function of the correction module; X Tne (t) is the real data; Correction error; T is the time series.
[0047] Furthermore, first, use the generative adversarial network (GAN) to fill in the missing data. Its generator model (G) is used for data filling. The specific data model is as follows:
[0048]
[0049] Then, use the discriminator (D) to evaluate the filled data, that is, evaluate the authenticity of the generated data. The specific mathematical model is as follows:
[0050]
[0051] Finally, output the completed data after evaluation (including the filled data). The specific mathematical model is as follows:
[0052]
[0053] In the formula: G is the generator, which is used to generate the filling value of the missing data; D is the discriminant model, which is used to judge whether it is real data or generated data; M is the missing mask; L GAN (D, G) is the adversarial loss function of GNA; P data is the real data distribution; P G is the data distribution generated by the generator; X Final is the complete data matrix after filling the missing data.
[0054] Furthermore, use the cross-validation technique to train the deep learning model and optimize its objective function. The specific mathematical model is as follows:
[0055] L Total = αL DAE + βL LSTM + γL GAN ;
[0056] In the formula: L Total is the total loss function; α, β, γ are weighted parameters, which are used to balance the optimization objectives of different modules.
[0057] Furthermore, through the incremental learning and online learning models, update the model weights to improve the intelligence adaptability of the model. Specifically:
[0058]
[0059] In the formula: Θ is the parameter set of the deep learning model; η is the learning rate, which controls the update step of the model parameters; is the gradient of the total loss function with respect to the model parameters, which is used to guide the direction of model optimization.
[0060] Compared with the prior art, the beneficial effects of the present invention are as follows: high-precision monitoring of the ground deformation in the mined-out area of the mine, and realizing the intelligent adaptation correction of the monitoring results. BRIEF DESCRIPTION OF THE DRAWINGS
[0061] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.
[0062] Figure 1 It is a block diagram of an intelligent adaptation ground deformation monitoring method for a mined-out area of a mine according to the present invention;
[0063] Figure 2 It is a flowchart of an intelligent adaptation ground deformation monitoring method for a mined-out area of a mine according to the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS
[0064] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts fall within the scope of protection of the present invention.
[0065] Embodiment 1: In some embodiments, please refer to Figure 1 - Figure 2 of the specification drawings. An intelligent adaptation ground deformation monitoring method for a mined-out area of a mine. Step 1: Based on stereo satellite monitoring, UAV aerial survey, and GNSS ground monitoring, build a stereo deformation monitoring network with different monitoring precisions and monitoring densities for the mined-out area of the mine. The monitoring data in the stereo deformation monitoring network is generated into MP-DS-InSAR monitoring data based on the MP-DS-InSAR monitoring technology;
[0066] Step 2: Uniformly process the monitoring data with different precisions, different spatial scales, and different time scales and generate multi-dimensional fusion data; and adopt the following methods:
[0067] Based on the application of the Kalman Filter to dynamically fuse the monitoring data of the mined-out area of the mine at different time intervals;
[0068] Kalman Filter mathematical model:
[0069]
[0070] In the formula: is the state estimate at the k-th moment, and K k is the Kalman gain, z k is the observed value, and H k is the observation matrix.
[0071] Based on the least squares adjustment model, the high-precision positions of GNSS points are jointly processed with the topographic data of UAV aerial survey;
[0072] Least Squares Adjustment mathematical model;
[0073]
[0074] In the formula: z i is the actually observed elevation value, and f(x i , y i ) is the fitted elevation value based on aerial survey data, and (x i , y i ) are the plane coordinates;
[0075] Step 3, the intelligent adaptation correction of the MP-DS-InSAR monitoring results. Based on the deep learning model, learn from the above multi-dimensional fusion data, remove noise, fill in missing data, correct systematic errors, optimize the model, and output the corrected high-precision monitoring data.
[0076] Construction steps A of the MP-DS-InSAR monitoring technology:
[0077] Step A1, obtain monitoring data 1 from stereo satellite monitoring, monitoring data 2 from UAV aerial survey, and monitoring data 3 from GNSS ground monitoring. Each monitoring data has multiple characteristic variables;
[0078] Calculate the covariance matrix C of monitoring data 1, monitoring data 2, and monitoring data 3;
[0079] Step A2, perform spectral decomposition on the covariance matrix C based on the principal component analysis (PCA) model, and finally construct the MP-DS (Principal Component Analysis Model-Distributed Scatterers, MP-DS) scatterer statistical model;
[0080] The algorithm of the MP-DS scatterer statistical model is:
[0081]
[0082] where u i is λi The eigenvector, the maximum eigenvalue λ i The corresponding eigenvector u i Is the main scattering mechanism, u2ˉu N Is noise.
[0083] Step A3, adjust the weights of each interference pair of the MP-DS scatterer statistical model monitoring data based on the Elastic Net regularization model and the Fisher information optimization model;
[0084] Step A4, extract and analyze the effective information in the interference pair based on the Dense Connectivity Convolutional Neural Network (Dense Net) and the Attention mechanism Long Short-Term Memory Network (LSTM);
[0085] Step A5, based on the spatio-temporal filter and the spatio-temporal model of atmospheric interference, construct an adaptive filtering technique and an atmospheric correction algorithm to improve the signal-to-noise ratio of the monitoring data;
[0086] The mathematical model of the atmospheric correction algorithm is as follows:
[0087]
[0088] Where: Is the corrected phase, Is the original observed phase, Is the phase error caused by atmospheric delay.
[0089] The mathematical model of the adaptive spatio-temporal filtering technique is as follows:
[0090]
[0091] Where D filtered Is the filtered data, ω ij Is the weight coefficient of the spatio-temporal filter, D(x+i,y+j,t) is the unfiltered original data;
[0092] Step A6, construct the MP-DS-InSAR monitoring technology based on the above steps.
[0093] The intelligent adaptation correction step B of the MP-DS-InSAR monitoring result:
[0094] Step B1, remove the noise in the MP-DS-InSAR monitoring data based on the Deep Convolutional Denoising Autoencoder (DAE);
[0095] The DAE removes noise by reconstruction, and its loss function is defined as:
[0096]
[0097] Wherein:
[0098] In the formula: n is the time step; m is the spatial dimension; X MP is the MP-DS-InSAR monitoring data; X Fusion is the multi-dimensional fusion data; N is the missing data mask matrix; is the MP-DS-InSAR data after denoising; f DAE is the mapping function of the deep convolutional denoising autoencoder (DAE), responsible for removing noise.
[0099] Step B2: Based on the multi-channel convolutional neural network (CNN), fuse the above multi-dimensional fusion data with the MP-DS-InSAR monitoring data after denoising;
[0100] The multi-channel convolutional neural network (CNN) fuses the multi-dimensional fusion data and the MP-DS-InSAR monitoring data after denoising, specifically as follows:
[0101]
[0102] Then, use the LSTM model to modify the time series data, and the specific mathematical model is as follows:
[0103]
[0104] Its loss model is as follows:
[0105]
[0106] In the formula: f CNN is the mapping function of the multi-channel convolutional neural network (CNN) for the MP-DS-InSAR data after denoising and other auxiliary data X Fusion ; X CNN is to extract the spatial and temporal features of the data; f LSTM is the correction of the time series data; L LSTM is the loss function of the correction module; X Tne (t) is the real data; Correction error; T is the time series.
[0107] Step B3: Based on the LSTM model, perform guided correction on the MP-DS-InSAR monitoring data;
[0108] The LSTM model performs guided correction on the MP-DS-InSAR monitoring data, thereby enhancing the local accuracy of the correction;
[0109] Step B4: Fill in the missing data in the MP-DS-InSAR monitoring data based on the Generative Adversarial Network (GAN), and generate high-quality corrected data in areas with severe noise;
[0110] First, use the adversarial network (GAN) to fill in the missing data. Its generator model (G) is used for data filling. The specific data model is as follows:
[0111]
[0112] Then, use the discriminator (D) to evaluate the filled data, that is, evaluate the authenticity of the generated data. The specific mathematical model is as follows:
[0113]
[0114] Finally, output the completed data after evaluation (including the filled data). The specific mathematical model is as follows:
[0115]
[0116] In the formula: G is the generator, which is used to generate the filling value of the missing data; D is the discriminant model, which is used to judge whether it is real data or generated data; M is the missing mask; L GAN (D, G) is the adversarial loss function of GNA; P data is the real data distribution; P G is the data distribution generated by the generator; X Final is the complete data matrix after filling in the missing data;
[0117] Fill in the missing data based on the Generative Adversarial Network (GAN), and generate high-quality corrected data in areas with severe noise;
[0118] Step B5: Train the deep learning model based on the cross-validation technique, and use the Mean Squared Error (MSE) or Adversarial Loss to evaluate the correction effects of the LSTM model and the Generative Adversarial Network (GAN), and adjust the weights of the LSTM model and the Generative Adversarial Network (GAN) to reduce the prediction error;
[0119] Use the cross-validation technique to train the deep learning model and optimize its objective function. The specific mathematical model is as follows:
[0120] L Total = αL DAE + βL LSTM + γL GAN ;
[0121] In the formula: L Total is the total loss function; α, β, γ are weighted parameters, which are used to balance the optimization objectives of different modules; LDAE is the loss function for removing noise from the monitoring data; L LSTM is the data correction guidance loss function after the monitoring data is fused; L DAN is the evaluation loss function for filling in the data;
[0122] Use cross-validation technology to train the deep learning model, and use loss functions such as mean squared error (MSE) or adversarial loss to evaluate the model correction effect. Continuously adjust the weights to reduce the prediction error, so as to achieve high-precision correction.
[0123] Step B6, optimize the adaptive correction model through incremental learning and online learning models to adapt to new data patterns.
[0124] Through incremental learning and online learning models, update the model weights and improve the adaptability of the model. Specifically:
[0125]
[0126] Where: Θ is the parameter set of the deep learning model; η is the learning rate, which controls the update step of the model parameters; is the gradient of the total loss function with respect to the model parameters, which is used to guide the direction of model optimization.
[0127] Through incremental learning and online learning models, the adaptive correction model is enabled to continuously self-optimize over time;
[0128] Step B7, output the high-precision monitoring data after adaptive correction, specifically:
[0129]
[0130] Where: X Corrected is the finally corrected MP-DS-InSAR data, including the results of denoising, correction, and completion, to meet the high-precision monitoring requirements.
[0131] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements will not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A method for intelligent adaptive surface deformation monitoring in a mine goaf, characterized by: The following steps are involved: Step 1: Based on stereo satellite monitoring, UAV aerial survey and GNSS ground monitoring, a stereo deformation monitoring network with different monitoring accuracy and monitoring density is built in the goaf area of the mine. The monitoring data in the stereo deformation monitoring network is used to generate MP-DS-InSAR monitoring data based on MP-DS-InSAR monitoring technology. Step 2: Unify the monitoring data of different precisions, different spatial scales and different time scales and generate multi-dimensional fusion data; And adopt the following method: Based on the use of Kalman Filter, the mining goaf monitoring data at different time intervals are dynamically integrated; Based on the Least Squares Adjustment model, the high-precision position of GNSS points is jointly processed with the terrain data obtained by drone aerial survey; Step three, the intelligent adaptive correction of the MP-DS-InSAR monitoring results is carried out based on the deep learning model to learn based on the above multi-dimensional fusion data, remove noise, fill in data gaps, correct system errors, optimize the model, and output the corrected high-precision monitoring data.
2. The method for intelligent adaptive surface deformation monitoring in mine goaf according to claim 1 is characterized in that: MP-DS-InSAR monitoring technology construction steps A: Step A1, obtaining monitoring data 1 of stereo satellite monitoring, monitoring data 2 of drone aerial survey, and monitoring data 3 of GNSS ground monitoring, each monitoring data having multiple characteristic variables; Calculate the covariance matrix C of monitoring data 1, monitoring data 2, and monitoring data 3; Step A2, spectrally decompose the covariance matrix C based on the principal component analysis (PCA) model, and finally construct the MP-DS (Principal Component Analysis Model-Distributed Scatterers, MP-DS) scatterer statistical model; Step A3, adjusting the weights of each interference pair of the monitoring data of the MP-DS scatterer statistical model based on the Elastic Net regularization model and the Fisher information optimization model; Step A4, extracting and analyzing effective information from the interference pair based on densely connected convolutional neural network (Dense Net) and attention mechanism long short-term memory network (LSTM); Step A5, based on the spatiotemporal filter and the spatiotemporal model of atmospheric interference, construct an adaptive filtering technology and an atmospheric correction algorithm to improve the signal-to-noise ratio of the monitoring data; Step A6, constructing MP-DS-InSAR monitoring technology based on the above steps.
3. The method for intelligent adaptive surface deformation monitoring in mine goaf according to claim 2 is characterized in that: Kalman filter mathematical model: Where: is the state estimate at the kth moment, K k is the Kalman gain, z k is the observed value, H k is the observation matrix.
4. The method for intelligent adaptive surface deformation monitoring in mining goaf according to claim 3 is characterized in that: Least Squares Adjustment mathematical model; Where: z i is the actual observed elevation value, f(x i ,y i ) is the fitted elevation value based on the aerial survey data, (x i ,y i ) are plane coordinates.
5. The method for intelligent adaptive surface deformation monitoring in mining goaf according to claim 4 is characterized in that: Step B of intelligent adaptive correction of MP-DS-InSAR monitoring results: Step B1, removing noise from MP-DS-InSAR monitoring data based on a deep convolutional denoising autoencoder (Denoising Autoencoder, DAE); Step B2, fusing the above multi-dimensional fusion data with the denoised MP-DS-InSAR monitoring data based on a multi-channel convolutional neural network (CNN); Step B3, guiding and correcting the MP-DS-InSAR monitoring data based on the LSTM model; Step B4, fill in the missing data in the MP-DS-InSAR monitoring data based on the generative adversarial network (GAN), and generate high-quality corrected data in the noisy area; Step B5, training the deep learning model based on the cross-validation technology, using mean square error (MSE) or adversarial loss (AdversarialLoss) to evaluate the correction effect of the LSTM model and the generative adversarial network (GAN), and adjusting the weights of the LSTM model and the generative adversarial network (GAN) to reduce the prediction error; Step B6, optimizing the intelligent adaptive correction model through incremental learning and online learning models to adapt to the new data pattern. Step B7, outputting high-precision monitoring data after intelligent adaptive correction; specifically: Where: X Corrected The final corrected MP-DS-InSAR data includes denoising, correction and incomplete results to meet the needs of high-precision monitoring.
6. The method for intelligent adaptive surface deformation monitoring in mining goaf according to claim 5 is characterized in that: DAE removes noise by reconstruction, and its loss function is defined as: in: Where: n is the time step; m is the spatial dimension; X MP MP-DS-InSAR monitoring data; X Fusion is the multi-dimensional fusion data; N is the missing data mask matrix; is the MP-DS-InSAR data after denoising; f DAE It is the mapping function of the deep convolutional denoising autoencoder (DAE), which is responsible for removing noise.
7. The method for intelligent adaptive surface deformation monitoring in mine goaf according to claim 6 is characterized in that: A multi-channel convolutional neural network (CNN) is used to fuse the multi-dimensional fusion data and the denoised MP-DS-InSAR monitoring data as follows: Then, the LSTM model is used to modify the time series data. The specific mathematical model is as follows: The loss model is as follows: Where: f CNN is the mapping function of the multi-channel convolutional neural network (CNN) for the denoised MP-DS-InSAR data and other auxiliary data X Fusion ;X CNN To extract the spatial and temporal characteristics of data; f LSTM is the correction of time series data; L LSTM is the loss function of the correction module; X Tne (t) is the real data; Correct the error; T is the time series.
8. The method for intelligent adaptive surface deformation monitoring in mining goaf according to claim 7 is characterized in that: First, we use a GAN to fill in missing data, and its generator model (G) performs data filling. The specific data model is as follows: Then, the judge (D) is used to evaluate the filled data, that is, to evaluate the authenticity of the generated data. The specific mathematical model is as follows: Finally, the completed data after evaluation (including filled data) is output, and its specific mathematical model is as follows: Where: G is the generator, which is used to generate the filling value of missing data; D is the discriminant model, which is used to judge whether it is real data or generated data; M is the missing mask; L GAN (D, G) is the adversarial loss function of GNA; P data is the real data distribution; P G The data distribution generated by the generator; X Final The complete data matrix after filling in the missing data.
9. The method for intelligent adaptive surface deformation monitoring in mining goaf according to claim 8 is characterized in that: Use cross-validation technology to train the deep learning model and optimize its objective function. The specific mathematical model is as follows: L Total =αL DAE +βL LSTM +γL GAN ; Where: L Total is the total loss function; α, β, γ are weighted parameters used to balance the optimization objectives of different modules.
10. The method for intelligent adaptive surface deformation monitoring in mining goaf according to claim 9 is characterized in that: Through incremental learning and online learning models, the model weights are updated to improve the model's intelligent adaptability. Specifically: Where: Θ is the parameter set of the deep learning model; η is the learning rate, which controls the update step size of the model parameters; It is the gradient of the total loss function to the model parameters, which is used to guide the direction of model optimization.