A spatio-temporal analysis method for land subsidence monitoring data

Abstract

The application discloses a kind of space-time analysis methods for ground subsidence monitoring data, multiple monitoring points are laid out on the surface of target area, the spatial coordinates of each monitoring point are determined, the time series data filtered at each monitoring point is used as sample, and a deep learning prediction model is established and trained to predict the deformation condition of future time;Combined with the spatial coordinate data of each monitoring point, the filtered real monitoring data and the prediction data of the deep learning prediction model as sample, a spatial interpolation model for each time period is established;The spatial interpolation surface of each monitoring time period is combined, a three-dimensional reconstruction model of the cumulative deformation of the surface in the time domain in the area is established by three-dimensional reconstruction, and the implicit relationship between the time domain and the spatial domain is established by means of three-dimensional sliding average method;By slicing, extracting the isosurface of the established deformation three-dimensional space-time model, the deformation law and influence range of the target area in the construction process are analyzed, which is convenient for timely adjusting construction method and reinforcement measures.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of test detection, and relates to a kind of space-time analysis method for ground subsidence monitoring data, and the present technology is mainly applied in the process of subway, highway and other engineering construction, and the analysis and processing work of ground subsidence monitoring data. BACKGROUND

[0002] At present, China is vigorously promoting the construction of subway, highway and other projects. Deformation monitoring is to continuously monitor the deformation of surrounding environment and self structure in the construction influence area during the construction process, analyze its deformation form and predict the development law. For engineering construction, deformation monitoring can monitor the deformation of surrounding environment and buildings, and take measures in time to abnormal conditions to ensure the safety of construction and people's property. Due to the restriction of terrain and site conditions, the monitoring data is often discontinuous in time and space, so the deformation data of limited monitoring points is interpolated in time and space to study and analyze the deformation field of each stage in the construction influence area, which is of great significance to construction decision.

[0003] At present, the analysis, prediction analysis and spatial interpolation analysis of time series data have been widely studied, and many analysis methods have been proposed, but each has its own limitations for on-site production, and cannot meet the intuitive, comprehensive, accurate and reliable requirements. Therefore, the present application combines Kalman filter model, LSTM deep learning algorithm, Kriging spatial interpolation algorithm and three-dimensional reconstruction technology to propose a space-time analysis method for ground subsidence monitoring data. SUMMARY

[0004] The purpose of the present application is to provide a space-time analysis method for ground subsidence monitoring data. Compared with conventional measurement, deformation monitoring usually belongs to precision measurement, which has higher precision requirement, and filtering and denoising can effectively reduce measurement error in measurement process, which is of great significance to analysis of monitoring data. At the same time, due to the restriction of terrain and site conditions, monitoring data is often discontinuous in time and space, so the deformation data of limited monitoring points is interpolated in time and space to study and analyze the deformation field of each stage in the construction influence area, which is of great significance to construction decision.

[0005] The present application is realized by the following technical solutions:

[0006] A space-time analysis method for ground subsidence monitoring data, comprising the following steps:

[0007] S1: multiple monitoring points are laid out on the ground surface of the target area, spatial coordinates of the monitoring points are determined, time series data of cumulative deformation of each monitoring point in the monitoring process are counted, and a data set sample is constructed;

[0008] S2: the cumulative deformation time series data of each monitoring point in step S1 are preprocessed by removing abnormal data and filling missing data;

[0009] S3: the cumulative deformation time series data of each monitoring point obtained in step S2 after preprocessing are filtered;

[0010] S4: a deep learning prediction model is established and trained by taking the filtered time series data of each monitoring point as a sample, and future deformation of each monitoring point in a period of time is predicted;

[0011] S5: a Kriging spatial interpolation model of each time period is established by taking spatial coordinate data of each monitoring point, filtered real monitoring data and prediction data of the deep learning prediction model as samples;

[0012] S6: spatial interpolation surfaces of each monitoring time period are combined, a three-dimensional reconstruction model of cumulative deformation of the ground surface in the area in the time domain is established by three-dimensional reconstruction, and an implicit relationship between the time domain and the spatial domain is established by means of three-dimensional sliding average method, so that deformation of an arbitrary coordinate point in the target area in the time period at an arbitrary time point can be predicted;

[0013] S7: the deformation rule and influence range of the target area in the construction process are analyzed by slicing and extracting the isosurface of the established deformation three-dimensional space-time model.

[0014] In the above technical solution, in step S2, the abnormal data is removed according to the 3σ principle.

[0015] In the above technical solution, in step S2, the missing value is filled by using the window sliding average method.

[0016] In the above technical solution, in step S3, the Klaman model is used for filtering.

[0017] In the above technical solution, in step S4, the LSTM deep learning prediction model is used to predict the deformation of each monitoring point in a period of time in the future.

[0018] In the above technical solution, in step S4, the first 90% of the data sample is taken as a training set, and the last 10% is taken as a test set.

[0019] The advantages and beneficial effects of the present application are:

[0020] (1) can solve the individual data anomaly and missing problem of monitoring process under the restriction of monitored condition; (2) can effectively reduce the noise interference problem of monitoring data; (3) can effectively realize the prediction of monitoring data in time domain and space domain, and the visual analysis of the settlement law of the surface of the analysis area; (4) the present application has the advantages of intuitive visualization, comprehensive function, accurate prediction result, and high reliability. BRIEF DESCRIPTION OF DRAWINGS

[0021] Figure 1 is a flow chart of the spatio-temporal analysis method for ground subsidence monitoring data of the present application.

[0022] Figure 2 is a curve graph of the original data and the preprocessed data given by the embodiment.

[0023] Figure 3 is a data curve graph before and after filtering given by the embodiment.

[0024] Figure 4 is a prediction result graph of the LSTM model in the embodiment.

[0025] Figure 5 is a Kriging spatial interpolation model graph established by the embodiment.

[0026] Figure 6 is a three-dimensional reconstruction model graph of the cumulative deformation of the target area ground in the time domain established by the embodiment.

[0027] Figure 7 is a graph obtained by slicing the established deformation three-dimensional spatio-temporal model.

[0028] Figure 8 is a graph obtained by extracting the isosurface of the established deformation three-dimensional spatio-temporal model.

[0029] For ordinary skilled persons in the art, other related drawings can be obtained according to the above drawings without creative labor. DETAILED DESCRIPTION

[0030] In order to enable the personnel in the technical field to better understand the present application scheme, the technical scheme of the present application will be further described below in combination with specific embodiments.

[0031] A spatio-temporal analysis method for ground subsidence monitoring data, referring to the accompanying Figure 1 , comprising the following steps:

[0032] S1: multiple monitoring points are arranged on the surface of the target area, the spatial coordinates of each monitoring point are determined, and the time series data of the cumulative deformation monitored by each monitoring point in the monitoring process is counted to construct a data set sample.

[0033] S2: Preprocess the time series data of cumulative deformation at each monitoring point in step S1, including removing outlier data and filling in missing data.

[0034] In this embodiment, outlier data is removed according to the 3σ principle (i.e., outliers are defined as values ​​in a set of measurements that deviate from the mean by more than 3 times the standard deviation).

[0035] Missing values ​​are filled using a window moving average method:

[0036]

[0037] Where: μ is the result calculated after moving average; N is the number of non-null values ​​within the moving window; A i Let i be the i-th non-empty value within the moving window; i is the index of the value within the moving window; for example... Figure 2 The figure shown is a graph of the original data and the preprocessed data given in this embodiment.

[0038] S3: Due to measurement errors during the monitoring process, in order to ensure the reliability of subsequent analysis, it is necessary to perform noise reduction filtering on the preprocessed cumulative deformation time series data of each monitoring point obtained in step S2. In this embodiment, the Klaman model is used for noise reduction filtering. The noise reduction filtering process of the Klaman model is as follows:

[0039] State equation: x t =A t,t-1 B t u t +ε t

[0040] Monitoring equation: z t =H t x t +δ t

[0041] State prediction equation:

[0042] State covariance prediction:

[0043] Status update based on new monitoring values:

[0044] Update state covariance matrix prediction:

[0045] Filter gain matrix:

[0046] Where, x t Let A be the state vector at time t. t,t-1is the state transition matrix of the system from time t-1 to time t, z t is the monitoring value vector at time t, H is the monitoring matrix at time t, and ε t is the system noise matrix at time t, δ t is the monitoring noise matrix at time t, P t is the covariance at time t, u t is the input vector at time t, z t is the monitoring matrix at time t, R t is the measurement covariance matrix (reflecting the accuracy of sensor measurement), Q t is the covariance matrix (reflecting the size of the influence of the state on the outside world); when the system is monitored at time t, z t , the predicted value can be corrected using the monitoring value to obtain the state estimation x t of the system at time t, and the recursive prediction and filtering are repeatedly performed. As shown in Figure 3 , the data curve graph before and after filtering given in the embodiment is shown.

[0047] S4: Taking the time series data of each monitoring point after noise reduction filtering as a sample, a deep learning prediction model is established and trained to predict the deformation of each monitoring point in the future period of time. Specifically, in the embodiment, the LSTM deep learning prediction model is used to predict the deformation of each monitoring point in the future period of time, and the LSTM deep learning prediction model is expressed as follows:

[0048] i t =τ(U i x t +W i h t-1 +b i )

[0049] g t =τ(U g x t +W g h t-1 +b f )

[0050] c t =g t c t-1 +i t tanh(U c x t +W c h t-1 +b c )

[0051] y Nt =τ(U0x t +W0h t-1 +b0)

[0052] h t = y Nt tanh(c t )

[0053] where i t , g t , y Nt are the output variables of the input gate, output gate, and forget gate, respectively, U i , U g , U o are the weight matrices of the input variables in the above three gates, W i , W g , W o are the weight matrices of the hidden state variables in the above three gates, b i , b g , b o are the bias terms in the above three gates, τ(·) is an activation function (to nonlinearize the neural network), x t and h t-1 represent the input and candidate state variables at time t and t-1, respectively, c t is the cell state variable, and h t is the candidate memory state variable.

[0054] In this embodiment, the first 90% of the data samples are used as the training set, and the last 10% are used as the test set. During the training process, the number of hidden layer units of the LSTM network model is set to 100, and the initial learning rate is specified as 0.005. After training, the root mean square error (RMSE) and the loss reach a stable minimum after about 50 iterations. The maximum error between the predicted value and the actual value after training is 0.32 mm, and the root mean square error (RMSE) is 0.22766, indicating that the LSTM network model has a small prediction error and good prediction performance. The prediction results are shown in Figure 4 .

[0055] S5: Combine the spatial coordinate data of each monitoring point, the filtered real monitoring data, and the prediction data of the deep learning prediction model as samples to establish a Kriging spatial interpolation model for each time period. The specific process is as follows:

[0056] Assuming that the spatial process Z(s i ), i-1,…,n is known, the optimal linear unbiased prediction value of the s0 point is where

[0057] If s0 is an unmonitored point (s0≠s i ), then λ1,…,λ n are calculated by the following formula:

[0058] Where: λ0≡(λ1,…,λ) n ,m)',γ0≡(γ(s0-s1),…,γ(s0-s n ),1)',

[0059]

[0060] Mutation function

[0061] The prediction variance matrix is:

[0062] If s0 is a monitoring point (s0≠s) i ), then λ1,…,λ n for:

[0063] Where: Z(s) i ) represents the measurement value at the i-th position, and λ i Let be the weight coefficient of the i-th point, s0 be the predicted location, n be the number of measurements, Λ0 be the angular symmetric matrix, and γ(h) be the variogram. The k-th component of γ0 is measured using the monitoring noise variance. To obtain by substitution.

[0064] See appendix Figure 5 This embodiment divides the target area into 100×100 interpolation grids based on the spatial coordinate range of each monitoring point, and establishes a Kriging spatial interpolation model based on the actual cumulative deformation values ​​of surface monitoring points at the same time (e.g., the 26th day).

[0065] S6: Combine the spatial interpolation surfaces of each monitoring time period, establish a three-dimensional reconstruction model of the cumulative surface deformation in the time domain of the area through three-dimensional reconstruction, and establish the implicit connection between the time domain and the spatial domain by using the three-dimensional moving average method, so that the deformation of any coordinate point in the target area at any time point in the current time period can be predicted.

[0066] In this embodiment, see Appendix Figure 6 The cumulative displacement values ​​of the 100×100 interpolation grid over 68 days (the first 62 days are filtered real monitoring data, and the last 6 days are LSTM predicted data) are used to construct a 100×100×68 three-dimensional matrix (X and Y axes represent the spatial coordinates of each point within the region, and the Z axis represents the monitoring time). Figure 6 The three-dimensional reconstruction model of the cumulative surface deformation of the target area in the time domain is shown. By using the three-dimensional moving average method, the implicit connection between the time domain and the spatial domain is established, so that the deformation of any coordinate point in the region at any time point in the current time period can be predicted.

[0067] S7: see attached Figure 7 and 8 By slicing the established three-dimensional space-time model of deformation, extracting the isosurface and other operations, the deformation law and influence range of the target area in the construction process are analyzed, so that the construction method and reinforcement measures can be adjusted in time.

[0068] The above has made an exemplary description of the application, it should be explained that, without departing from the core of the application, any simple deformation, modification or other equivalent replacement of the person skilled in the art without creative labor falls within the protection scope of the application.

Claims

1. A spatio-temporal analysis method for ground subsidence monitoring data, characterized in that, The method comprises the following steps: S1: arranging a plurality of monitoring points on the ground surface of the target area, determining the spatial coordinates of each monitoring point, and counting the time series data of the cumulative deformation of each monitoring point during the monitoring process to construct a data set sample; S2: performing pre-processing of the cumulative deformation time series data of each monitoring point in step S1, including abnormal data elimination and missing data filling; S3: performing filtering processing on the pre-processed cumulative deformation time series data of each monitoring point obtained in step S2; S4: taking the filtered time series data of each monitoring point as a sample, establishing and training a deep learning prediction model to predict the deformation of each monitoring point in a future period of time; S5: combining the spatial coordinate data of each monitoring point, the filtered real monitoring data, and the prediction data of the deep learning prediction model as samples to establish a Kriging spatial interpolation model for each time period; S6: combining the spatial interpolation surfaces of each monitoring time period to establish a three-dimensional reconstruction model of the cumulative deformation of the ground surface in the time domain in the area, and establishing an implicit relationship between the time domain and the spatial domain by means of the three-dimensional sliding average method, so that the deformation of any coordinate point in the target area in the time period at any time point can be predicted; S7: analyzing the deformation law and influence range of the target area in the construction process by slicing and extracting the isosurface of the established deformation three-dimensional space-time model.

2. The method for spatio-temporal analysis of monitoring data for ground subsidence according to claim 1, characterized in that: In step S2, the abnormal data is eliminated according to the 3σ principle.

3. The method for spatio-temporal analysis of monitoring data for ground subsidence according to claim 1, characterized in that: In step S2, the missing values are filled by using the window sliding average method.

4. The method for spatio-temporal analysis of monitoring data for ground subsidence according to claim 1, characterized in that: In step S3, the Klaman model is used for filtering.

5. The method for spatio-temporal analysis of monitoring data for ground subsidence according to claim 1, characterized in that: In step S4, the LSTM deep learning prediction model is used to predict the deformation of each monitoring point in a future period of time.

6. The method for spatio-temporal analysis of monitoring data for ground subsidence according to claim 1, characterized in that: In step S4, the first 90% of the data sample is taken as the training set, and the last 10% is taken as the test set.

Images