Ground surface deformation prediction method and device based on multiple models, medium and equipment

Abstract

The invention discloses an earth surface deformation prediction method and device based on multiple models, a medium and equipment, and is applied to the technical field of earth surface subsidence prediction.The method comprises the steps that first time sequence data of an earth surface to be predicted are acquired, and feature information is extracted; respectively training a plurality of prediction models by taking the feature information as a sample in a preset parameter range; selecting the prediction model with the minimum prediction error value as an optimal model; training the optimal model again by taking the feature information as a sample so as to obtain an optimal parameter in a global range; inputting second time sequence data of the earth surface to be predicted into the optimal model, and predicting a deformation result; the method comprises the following steps: training a plurality of prediction models by using part of time sequence data of a to-be-predicted earth surface within a preset parameter range to realize rapid training, selecting an optimal prediction model, and carrying out global parameter training on the optimal prediction model again to improve the prediction precision of the optimal prediction model, reduce the training time of the model as much as possible, and improve the prediction efficiency. And the prediction efficiency is improved.

Description

Technical Field

[0001] The present application relates to the technical field of surface subsidence prediction, and in particular to a surface deformation prediction method, device, medium and equipment based on multiple models. Background Art

[0002] As the scale of the mining industry gradually expands, the original stress state of the surface is destroyed during the mining process, resulting in stress redistribution. In this process, the rock mass sinks, deforms, and moves, causing a series of disasters such as surface subsidence and landslides. At the same time, it causes damage to surface buildings, railways, farmland, etc., which has a serious negative impact. Monitoring the settlement of the surface above the mining area and predicting the trend of settlement changes can provide early warning of the degree of damage to ground structures caused by surface settlement, and provide a decision-making basis for the prevention and control of geological disasters in mining areas.

[0003] At present, the probability integral method or the subsidence prediction parameter detection method is mostly used to predict mining subsidence in mining areas. Due to the complexity of mining conditions, it is impossible to accurately and comprehensively express the relationship between the probability integral method parameters and the actual geological mining conditions, resulting in inaccurate prediction of mining subsidence by the probability integral method. The subsidence prediction parameter detection method requires the establishment of observation stations in individual mining areas or regional mining areas to detect subsidence prediction parameters, which also has the problem of inaccurate prediction of mining subsidence in mining areas.

[0004] With the continuous development of machine learning technology, more and more machine learning methods have emerged, and machine learning technology has also begun to be applied to the monitoring of surface subsidence in mining areas. However, due to the different settlement forms and laws of different geological structures, and the characteristics of the settlement characteristic settlement curves are also quite different, if a single machine learning method (prediction model) is used to predict settlement in different mining areas, the prediction accuracy will be low. Summary of the invention

[0005] In order to solve the above technical problems, the present application is proposed. The embodiments of the present application provide a method, device, medium and equipment for predicting surface deformation based on multiple models, which train multiple prediction models using time series data of the surface to be predicted within a preset parameter range to achieve rapid training, and evaluate multiple prediction models for the surface to be predicted to obtain the most suitable prediction model for the surface to be predicted, and perform global parameter training on the optimal prediction model to improve prediction accuracy.

[0006] According to one aspect of the present application, a method for predicting surface deformation based on multiple models is provided, comprising: obtaining first time series data of a surface to be predicted; wherein the first time series data includes state information of multiple prediction points in the surface to be predicted obtained at multiple time points; extracting feature information in the first time series data; wherein the feature information represents the state change amount of a single prediction point at the multiple time points; using the feature information as a sample to train multiple prediction models respectively; wherein the parameters of the multiple prediction models are set within a preset range; calculating prediction error values ​​of the multiple prediction models; wherein the prediction error value represents the prediction accuracy of the corresponding prediction model; selecting the prediction model with the smallest prediction error value as the optimal model; using the feature information as a sample to train the optimal model again to obtain the optimal parameters of the optimal model in the entire domain; obtaining second time series data of the surface to be predicted; wherein the second time series data includes state information of all prediction points in the surface to be predicted obtained at multiple time points; inputting the second time series data into the trained optimal model to predict the deformation result of the surface to be predicted.

[0007] In one embodiment, the using the feature information as a sample to train multiple prediction models separately includes: using the first part of the feature information as a training sample to train the multiple prediction models separately; the calculating the prediction error values ​​of the multiple prediction models includes: using the second part of the feature information as a verification sample to calculate the prediction error values ​​of the multiple prediction models separately; wherein, a single training sample includes feature information obtained by a single prediction point at a first continuous time point of a first preset time length, and a single verification sample includes feature information obtained by the single prediction point at a second continuous time point of a second preset time length, and the first continuous time point and the second continuous time point are different, and both are partial time points of the multiple time points.

[0008] In one embodiment, using the second part of the feature information as a verification sample to respectively calculate the prediction error values ​​of the multiple prediction models includes: using the second part of the feature information as a verification sample to respectively calculate multiple verification errors of each of the prediction models; calculating the absolute average of the multiple verification errors of each of the prediction models to obtain the prediction error value of each of the prediction models.

[0009] In one embodiment, the obtaining of first time series data of the surface to be predicted includes: obtaining first image data of the surface to be predicted; wherein the first image data includes image information of multiple prediction points in the surface to be predicted acquired at multiple time points; based on the first image data, the first time series data is generated using a small baseline set interferometry measurement technique.

[0010] In one embodiment, after obtaining the first time series data of the surface to be predicted, the multi-model based surface deformation prediction method also includes: preprocessing the first time series data to obtain processed first time series data; extracting feature information from the first time series data includes: extracting feature information from the processed first time series data.

[0011] In one embodiment, the preprocessing of the first time series data includes: using a preset coherence coefficient and an error estimation value to filter the first time series data.

[0012] In one embodiment, using the feature information as a sample to re-train the optimal model includes: dividing the feature information into multiple samples to train the optimal model; wherein a single sample includes feature information obtained by a single prediction point at a third continuous time point of a third preset time length, and the third continuous time point is a partial time point of the multiple time points.

[0013] According to another aspect of the present application, a surface deformation prediction device based on multiple models is provided, including: a first acquisition module, used to acquire first time series data of a surface to be predicted; wherein the first time series data includes state information of multiple prediction points on the surface to be predicted acquired at multiple time points; an information extraction module, used to extract feature information from the first time series data; wherein the feature information represents the state change of a single prediction point at the multiple time points; a first training module, used to use the feature information as a sample to train multiple prediction models respectively; wherein the parameters of the multiple prediction models are set within a preset range; an error calculation module, used to calculate the error information of the multiple prediction models. prediction error value; wherein the prediction error value represents the prediction accuracy of the corresponding prediction model; a model selection module, used to select the prediction model with the smallest prediction error value as the optimal model; a second training module, used to train the optimal model again using the feature information as a sample to obtain the optimal parameters of the optimal model in the entire domain; a second acquisition module, used to obtain second time series data of the surface to be predicted; wherein the second time series data includes state information of all prediction points in the surface to be predicted obtained at multiple time points; a deformation prediction module, used to input the second time series data into the trained optimal model to predict the deformation result of the surface to be predicted.

[0014] According to another aspect of the present application, a computer-readable storage medium is provided, wherein the storage medium stores a computer program, and the computer program is used to execute any of the above methods.

[0015] According to another aspect of the present application, an electronic device is provided, comprising: a processor; a memory for storing instructions executable by the processor; and the processor is used to execute any of the above-described methods.

[0016] The present application provides a method, device, medium and equipment for predicting surface deformation based on multiple models, which obtains first time series data of the surface to be predicted; wherein the first time series data includes state information of multiple prediction points in the surface to be predicted obtained at multiple time points; extracts feature information in the first time series data; wherein the feature information represents the state change of a single prediction point at multiple time points; uses the feature information as samples to train multiple prediction models respectively; wherein the parameters of the multiple prediction models are set within a preset range; calculates prediction error values ​​of multiple prediction models; wherein the prediction error value represents the prediction accuracy of the corresponding prediction model; selects the prediction model with the smallest prediction error value as the optimal model; and uses the feature information as a sample to train the optimal model again to obtain the optimal model within the entire domain. optimal parameters; obtaining second time series data of the surface to be predicted; wherein the second time series data includes status information of all prediction points in the surface to be predicted obtained at multiple time points; inputting the second time series data into the trained optimal model to predict the deformation result of the surface to be predicted; that is, obtaining part of the time series data of the surface to be predicted, using the part of the time series data to train multiple prediction models within a preset parameter range to achieve rapid training, and evaluating the multiple prediction models for the surface to be predicted to obtain the most suitable prediction model for the surface to be predicted, and again performing global parameter training on the optimal prediction model to improve the prediction accuracy of the optimal prediction model, thereby improving the settlement prediction accuracy of the surface to be predicted, and at the same time, it can also minimize the training time of the model to improve the prediction efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a flow chart of a method for predicting ground deformation based on multiple models provided in an embodiment of the present application.

[0018] Figure 2 It is a structural block diagram of a surface deformation prediction device based on multiple models provided in an embodiment of the present application.

[0019] Figure 3 It is a structural block diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0020] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0021] Furthermore, in the exemplary embodiments, since the same reference numerals denote the same components having the same structure or the same steps of the same method, if one embodiment is exemplarily described, only structures or methods different from the described embodiment are described in other exemplary embodiments.

[0022] Throughout the specification and claims, when a component is described as being “connected” to another component, the component may be “directly connected” to the other component, or “electrically connected” to the other component through a third component. In addition, unless explicitly described to the contrary, the term “include” and its corresponding terms should be understood to include only the components described, and should not be understood to exclude any other components.

[0023] The method provided in the embodiment of the present application can be executed by an electronic device, which can be a server or a terminal device, wherein the server can be an independent physical server, a server cluster or a distributed system composed of multiple physical servers, or a cloud server providing cloud computing services. The terminal device can be a smart phone, a tablet computer, a desktop computer, etc., but is not limited thereto.

[0024] Figure 1 FIG. 1 is a flow chart of a method for predicting ground deformation based on multiple models provided in an embodiment of the present application. Figure 1 As shown, the surface deformation prediction method based on multiple models includes the following steps: Step 110: Obtain first time series data of the surface to be predicted.

[0025] Among them, the first time series data includes status information of multiple prediction points in the surface to be predicted obtained at multiple time points. Specifically, the present application can periodically obtain the time series data of multiple prediction points in the surface to be predicted, for example, periodically obtain it at a time interval of 12 days, that is, collect the time series data of multiple prediction points in the surface to be predicted every 12 days, and obtain continuous time series data through continuous time, that is, the first time series data. It should be understood that the present application can obtain the time series data of all prediction points on the surface to be predicted at a single time, and then select the time series data of some prediction points as the first time series data, or only obtain the time series data of some required prediction points at a single time as the first time series data.

[0026] Step 120: Extract feature information from the first time series data.

[0027] Among them, the feature information characterizes the state change of a single prediction point at multiple time points. Specifically, the present application can use the TsFeatures algorithm to learn the feature information in the first time series data, wherein the feature information can include any one or more combinations of the following information: time series length, mean, variance, noise entropy, variance of variance, variance of mean, trend strength, peak, peak and valley position, features based on horizontal displacement, autocorrelation, ACF features, PACF features, standard deviation of the first-order derivative of the time series, crossover point, binarized mean, heterogeneity, etc. Among them, the variance of the variance represents the variance of the internal variance of the time series group composed of some time series data in the first time series data and the internal variance of other time series groups, and the variance of the mean represents the variance of the internal mean of the time series group composed of some time series data in the first time series data and the internal mean of other time series groups. For example, the first time series data is 100 time series data collected at 100 time points (taking a single prediction point as an example, numbered 1-100 respectively), 1-50, 2-51, 3-52, ..., 51-100 are divided into 51 time series groups, and the means and variances of each time series data in each time series group are calculated respectively, and then the variances between the means of each time series group and the variances of each time series group are calculated, which are the variance of the mean and the variance of the variance. The present application groups the first time series data to obtain the change trend of each time series group, so that when geological dynamics such as mudslides and landslides that cause changes in surface height or morphology occur, they can be reflected only in the time series data of one or more time series groups. That is, the abnormal changes of the surface to be predicted can be determined based on the characteristic information of one or more time series groups, and the weight of the abnormal time series group can be adjusted, or the abnormal time series group can be directly eliminated to minimize the impact of the abnormal change on the deformation result of the surface to be predicted.

[0028] Step 130: Use the feature information as samples to train multiple prediction models respectively.

[0029] Among them, the parameters of multiple prediction models are set within a preset range. The present application uses the feature information of the extracted first time series data as samples to train multiple prediction models respectively, so as to obtain multiple prediction models that can be used for deformation prediction of the surface to be predicted. Specifically, the prediction models may include an autoregressive differential moving average model (SARIMA), Facebook's prediction model (Prophet), Holt-Winter model (Holt-Winter), a long short-term memory recurrent neural network model (LSTM) and a time series decomposition model (STLF), etc. The present application uses a self-supervised learning model selector (SSL-MS) to train each prediction model. The present application sets the parameter range of the prediction model, that is, sets the parameter search range of each prediction model within a preset range according to experience, so as to speed up the training speed of the prediction model and obtain the optimal value within the local range, that is, the training prediction model quickly searches for the optimal parameters within the preset range, which can not only reduce the number of samples required, but also accelerate the training progress.

[0030] Step 140: Calculate prediction error values ​​of multiple prediction models.

[0031] The prediction error value represents the prediction accuracy of the corresponding prediction model. The present application calculates the prediction error value of each prediction model respectively to obtain the prediction accuracy of each prediction model.

[0032] Step 150: Select the prediction model with the smallest prediction error value as the optimal model.

[0033] After calculating the prediction error value of each prediction model, the prediction model that is most suitable for the surface to be predicted can be obtained by comparing the prediction error values, and the prediction model with the smallest prediction error value is used as the optimal model to predict the surface to be predicted; that is, multiple prediction models are trained by a small amount of sample data of the surface to be predicted, and the accuracy of the prediction model that has been initially trained is verified, and the prediction model with the highest accuracy is selected as the final prediction model to predict the deformation of the surface to be predicted, so that the prediction model that is most suitable for the surface to be predicted can be obtained, thereby improving the prediction accuracy.

[0034] Step 160: Use the feature information as samples to train the optimal model again to obtain the optimal parameters of the optimal model in the entire domain.

[0035] After the optimal model is determined, the optimal model is trained again. Specifically, the characteristic information of the first time series data can be used as samples to train the optimal model again, or samples can be obtained again to train the optimal model again. At this time, no restrictions are imposed on the parameters of the optimal model, that is, the optimal parameters of the optimal model are searched over the entire domain, thereby further improving the fit between the optimal model and the surface to be predicted, and then improving the prediction accuracy.

[0036] Step 170: Obtain second time series data of the surface to be predicted.

[0037] Among them, the second time series data includes the status information of all prediction points in the surface to be predicted obtained at multiple time points. The present application predicts the deformation result of the surface to be predicted based on the second time series data by obtaining the second time series data of the surface to be predicted (that is, the time series data of the global prediction points of the surface to be predicted). The present application can obtain the time series data of all prediction points of the surface to be predicted when obtaining the first time series data, select a part of them as the first time series data to train the prediction model, and after the training is completed, directly use the realization data of all prediction points of the surface to be predicted as the second time series data to reduce the number of data acquisition times.

[0038] Step 180: Input the second time series data into the trained optimal model to predict the deformation result of the surface to be predicted.

[0039] The present application inputs the second time series data into the trained optimal model, and uses the optimal model to quickly predict the deformation result of the surface to be predicted, that is, uses the prediction model that is most suitable for the surface to be predicted and has a higher prediction accuracy to predict the deformation result of the surface to be predicted, so as to obtain the most accurate prediction result of the surface to be predicted as much as possible.

[0040] The present application provides a method for predicting surface deformation based on multiple models, which comprises obtaining first time series data of a surface to be predicted; wherein the first time series data comprises state information of multiple prediction points on the surface to be predicted obtained at multiple time points; extracting feature information from the first time series data; wherein the feature information represents the amount of state change of a single prediction point at multiple time points; using the feature information as samples to train multiple prediction models respectively; wherein the parameters of the multiple prediction models are set within a preset range; calculating prediction error values ​​of the multiple prediction models; wherein the prediction error values ​​represent the prediction accuracy of the corresponding prediction models; selecting the prediction model with the smallest prediction error value as the optimal model; and using the feature information as samples to train the optimal model again to obtain the optimal parameters of the optimal model in the entire domain. number; obtaining second time series data of the surface to be predicted; wherein the second time series data includes status information of all prediction points in the surface to be predicted obtained at multiple time points; inputting the second time series data into the trained optimal model to predict the deformation result of the surface to be predicted; that is, obtaining part of the time series data of the surface to be predicted, using the part of the time series data to train multiple prediction models within a preset parameter range to achieve rapid training, and evaluating the multiple prediction models for the surface to be predicted to obtain the most suitable prediction model for the surface to be predicted, and again performing global parameter training on the optimal prediction model to improve the prediction accuracy of the optimal prediction model, thereby improving the settlement prediction accuracy of the surface to be predicted, and at the same time, it can also minimize the training time of the model to improve the prediction efficiency.

[0041] In one embodiment, the specific implementation method of the above-mentioned step 130 may be: using the first part of the feature information as a training sample to train multiple prediction models respectively; correspondingly, the specific implementation method of step 140 may be: using the second part of the feature information as a verification sample to calculate the prediction error values ​​of multiple prediction models respectively; wherein, a single training sample includes the feature information obtained by a single prediction point at a first continuous time point of a first preset time length, and a single verification sample includes the feature information obtained by a single prediction point at a second continuous time point of a second preset time length, and the first continuous time point and the second continuous time point are different, and both are partial time points of multiple time points.

[0042] In one embodiment, the specific implementation method of the above step 140 can be: using the second part of the feature information as a verification sample, and calculating multiple verification errors of each prediction model respectively; calculating the absolute average of the multiple verification errors of each prediction model to obtain the prediction error value of each prediction model.

[0043] Specifically, the present application divides the feature information into two parts: a first part and a second part. The first part is used as a training sample to train multiple prediction models, and the second part is used as a verification sample to verify the prediction accuracy of the prediction model (i.e., calculate the prediction error value of the prediction model). Specifically, the prediction result of the second part is predicted by the first part, and the prediction error value of the prediction model is determined by comparing the actual result and the prediction result of the second part. Preferably, the prediction result of a single verification sample corresponding to the feature information prediction obtained at the first continuous time point closest in time can be selected. For example, the first time series data includes 100 time series data of a single prediction point collected at 100 time points (numbered 1-100 respectively), the 1-90th time series data are used as training samples, the 91st-100th time series data are used as verification samples, and the 91st time series data are used to predict the 91st time series data, that is, the characteristic information of the 1-90th time series data is input into the prediction model as a known quantity, and the prediction data of the next time point (that is, the 91st time point) is predicted, and the single point error of the prediction model is obtained by comparing the 91st time series data and the prediction data. Value, similarly, the 2nd to 91st time series data are used to predict the 92nd time series data, the 3rd to 92nd time series data are used to predict the 93rd time series data, ..., the 10th to 99th time series data are used to predict the 100th time series data, thereby obtaining 10 single-point error values, and based on the 10 single-point error values, the verification error of the corresponding prediction model at the prediction point (such as the absolute average, weighted average, etc.) is obtained. It should be understood that the present application can comprehensively obtain the prediction error value of the corresponding prediction model (such as the absolute average, weighted average, etc.) by combining the verification errors of each prediction point in the sample. The present application uses different and continuous training samples to predict and obtain prediction data for different verification samples. Since mudslides, landslides, etc. may cause geological dynamics such as changes in surface height or morphology, and these geological dynamics may only be reflected in a few time series data or the time series data of a single prediction point, the present application can determine the abnormal changes in the surface to be predicted based on the characteristic information of multiple continuous time series data, and can minimize the impact of the abnormal changes on the verification results of the prediction model by adjusting the weights of the abnormal continuous time series data or directly eliminating the abnormal continuous time series data.

[0044] In one embodiment, the specific implementation method of the above-mentioned step 110 may be: obtaining first image data of the surface to be predicted; wherein the first image data includes image information of multiple prediction points in the surface to be predicted acquired at multiple time points; based on the first image data, a small baseline set interferometry measurement technology is used to generate first time series data.

[0045] Synthetic Aperture Radar Interferometry (InSAR) is a product of the combination of synthetic aperture radar (SAR) satellite and radio astronomy interferometry. It uses radar to transmit microwaves to the target area, and then receives the echo reflected by the target to obtain a SAR complex image pair of the same target area. If there is a coherent condition between the complex image pairs, the conjugate multiplication of the SAR complex image pairs can obtain an interference pattern. According to the phase value of the interference pattern, the path difference of the microwaves in the two imagings is obtained, so as to calculate the topography, landform and slight changes on the surface of the target area, which can be used for the establishment of digital elevation models, crustal deformation detection, etc. InSAR technology can monitor various types of surface subsidence deformation, including surface subsidence monitoring in mining areas. Small Baseline Set Time Series Radar Interferometry (SBAS-InSAR) technology can use multiple images as the main image, and optimize the combination of as many interferometric image pairs as possible by setting reasonable temporal and spatial baseline thresholds, performing differential interferometry processing and multi-view processing, and applying mathematical analysis methods. Finally, as many high-coherence interferometric phase maps as possible are generated. After obtaining a large number of high-coherence interferometric phase maps, the ground object target points with high coherence are extracted according to the coherence coefficient values ​​in the interferometric map, and a linear model is established for these target points; the interferometric map is differentiated through an external digital elevation model (DEM) to obtain a differential interferometric map, and the singular values ​​after phase unwrapping in multiple differential interferometric maps are decomposed using known target points to obtain the linear deformation rate and cumulative deformation value of these target points.

[0046] The present application utilizes radar to continuously and periodically collect the first image data of the surface to be predicted, and performs small baseline set differential interferometry measurement on the radar data to generate the first time series data. Specifically, the selected SAR image is selected according to the time baseline (for example, 90 days) and the spatial baseline (for example, 200 meters) set in time, and the secondary image is selected and processed with the main image for differential interferometry to form multiple interference pairs, and the coherence of the interference pairs is estimated. The high coherence points are selected by pre-setting the time series coherence threshold and the amplitude deviation threshold, and the time series data of the surface deformation points are obtained through repeated iterative calculations. Among them, the first time series data includes the latitude and longitude coordinates of each prediction point on the surface to be predicted, the cumulative sinking time series value of each time period (for example, 12 days), the coherence coefficient of the prediction point and the error estimate value.

[0047] In one embodiment, after step 110, the above-mentioned multi-model-based surface deformation prediction method may also include: preprocessing the first time series data to obtain processed first time series data; correspondingly, the specific implementation method of step 120 may be: extracting feature information from the processed first time series data.

[0048] Due to the complexity of the external environment, there may be certain interference data in the radar data collection process. After obtaining the first time series data, the present application can preprocess the first time series data to optimize the first time series data, thereby providing a more accurate data basis for subsequent calculations.

[0049] In one embodiment, the specific implementation method of the above preprocessing may be: using a preset coherence coefficient and error estimation value to filter the first time series data.

[0050] Specifically, the present application can set a filter by setting the coherence coefficient and the error estimate value, filter out all prediction points with a coherence coefficient less than 0.3 and an error estimate value greater than 10, and obtain multiple high-precision distributed time series deformation monitoring data (including data of multiple prediction points at multiple time points).

[0051] In one embodiment, the specific implementation method of the above-mentioned step 160 may be: dividing the feature information into multiple samples to train the optimal model; wherein a single sample includes the feature information of a single prediction point obtained at the third continuous time point of the third preset time length, and the third continuous time point is a partial time point of the multiple time points.

[0052] Specifically, the present application uses a self-supervised learning parameter selector (SSL-HPT) to input the time series feature information and the optimal model of each sample, and retrain the optimal model to determine the optimal parameters of the optimal model in the entire domain, so as to obtain the optimal parameters of the optimal model for the surface to be predicted, and then improve the deformation prediction accuracy of the surface to be predicted. The present application uses the above-mentioned feature information or the feature information of the newly acquired time series data of the surface to be predicted as a sample to retrain the optimal model, and uses the feature information of each prediction point at a preset time point as a sample to increase the number of samples, and can make the error data (for example, the data error at a single time point is large, and the difference between the error data and the adjacent prediction points is not very large, that is, it is difficult to directly determine it as abnormal data) only exist in part of the samples formed by a single prediction point, thereby reducing its proportion in the sample data, and then reducing its impact on the optimal model. For example, the first time series data includes 100 time series data of a single prediction point collected at 100 time points (numbered 1-100 respectively), and the 100 time series data are split into 51 time series data obtained by combining 1-50, 2-51, ..., 51-100, that is, 51 samples are derived from 1 time series data to improve the training intensity of the optimal model, thereby improving its accuracy; if the 20th time series data has a large error, the 20th time series data will only appear in 20 samples. For a single prediction point, the original 100% of samples with large errors (with the 1st to 100th time series data as samples) is reduced to 20% of samples with large errors, thereby reducing the interference of large errors on the optimal model.

[0053] Preferably, in order to further improve the training effect of the optimal model and the search speed of the optimal parameters, the present application can search for the optimal parameters on the basis of the previous training (that is, the parameters after the previous training are completed as the initial values) to speed up the search speed of the optimal parameters. At the same time, it can also avoid the optimal parameters obtained by the search being inferior to the parameter values ​​obtained by the previous training (for example, the parameter search enters an infinite loop, the search times are insufficient, etc., resulting in only searching for a local optimal solution).

[0054] Figure 2 is a structural block diagram of a surface deformation prediction device based on multiple models provided in an embodiment of the present application. Figure 2As shown, the surface deformation prediction device 20 based on multiple models includes: a first acquisition module 21, used to acquire first time series data of the surface to be predicted; wherein the first time series data includes state information of multiple prediction points in the surface to be predicted acquired at multiple time points; an information extraction module 22, used to extract feature information in the first time series data; wherein the feature information represents the state change of a single prediction point at multiple time points; a first training module 23, used to train multiple prediction models respectively using the feature information as samples; wherein the parameters of the multiple prediction models are set within a preset range; an error calculation module 24, used to calculate the prediction errors of the multiple prediction models. difference; wherein the prediction error value represents the prediction accuracy of the corresponding prediction model; a model selection module 25 is used to select the prediction model with the smallest prediction error value as the optimal model; a second training module 26 is used to train the optimal model again using the feature information as a sample to obtain the optimal parameters of the optimal model in the entire domain; a second acquisition module 27 is used to obtain the second time series data of the surface to be predicted; wherein the second time series data includes the state information of all prediction points in the surface to be predicted obtained at multiple time points; a deformation prediction module 28 is used to input the second time series data into the trained optimal model to predict the deformation result of the surface to be predicted.

[0055] The present application provides a surface deformation prediction device based on multiple models, which obtains the first time series data of the surface to be predicted through the first acquisition module 21; wherein the first time series data includes the state information of multiple prediction points in the surface to be predicted obtained at multiple time points; the information extraction module 22 extracts the feature information in the first time series data; wherein the feature information represents the state change of a single prediction point at multiple time points; the first training module 23 uses the feature information as a sample to train multiple prediction models respectively; wherein the parameters of the multiple prediction models are set within a preset range; the error calculation module 24 calculates the prediction error values ​​of the multiple prediction models; wherein the prediction error value represents the prediction accuracy of the corresponding prediction model; the model selection module 25 selects the prediction model with the smallest prediction error value as the optimal model; the second training module 26 uses the feature information as a sample to train the optimal model again, so as to obtain the prediction accuracy of the corresponding prediction model; the prediction error value is represented by the prediction error value of the prediction model; the model selection module 25 selects the prediction model with the smallest prediction error value as the optimal model; the second training module 26 uses the feature information as a sample to train the optimal model again. The optimal parameters of the optimal model in the entire domain are obtained; the second acquisition module 27 obtains the second time series data of the surface to be predicted; wherein the second time series data includes the status information of all prediction points in the surface to be predicted obtained at multiple time points; the deformation prediction module 28 inputs the second time series data into the trained optimal model to predict the deformation result of the surface to be predicted; that is, part of the time series data of the surface to be predicted is obtained, and multiple prediction models are trained using the part of the time series data within a preset parameter range to achieve rapid training, and multiple prediction models are evaluated for the surface to be predicted to obtain the most suitable prediction model for the surface to be predicted, and the optimal prediction model is trained again with global parameters to improve the prediction accuracy of the optimal prediction model, thereby improving the settlement prediction accuracy of the surface to be predicted, and at the same time, the training time of the model can be minimized to improve the prediction efficiency.

[0056] In one embodiment, the first training module 23 can be further configured as follows: using the first part of the feature information as a training sample to train multiple prediction models respectively; correspondingly, the error calculation module 24 can be further configured as follows: using the second part of the feature information as a verification sample to calculate the prediction error values ​​of multiple prediction models respectively; wherein a single training sample includes feature information obtained by a single prediction point at a first continuous time point of a first preset time length, and a single verification sample includes feature information obtained by a single prediction point at a second continuous time point of the first preset time length, and the first continuous time point and the second continuous time point are different, and both are partial time points of multiple time points.

[0057] In one embodiment, the error calculation module 24 can be further configured as follows: using the second part of the feature information as a verification sample, and calculating multiple verification errors of each prediction model respectively; calculating the absolute average of the multiple verification errors of each prediction model to obtain the prediction error value of each prediction model.

[0058] In one embodiment, the above-mentioned first acquisition module 21 can be further configured to: acquire first image data of the surface to be predicted; wherein the first image data includes image information of multiple prediction points in the surface to be predicted acquired at multiple time points; based on the first image data, generate first time series data using small baseline set interferometry measurement technology.

[0059] In one embodiment, the above-mentioned multi-model-based surface deformation prediction device 20 can be further configured to: preprocess the first time series data to obtain processed first time series data; correspondingly, the information extraction module 22 can be further configured to: extract feature information from the processed first time series data.

[0060] In one embodiment, the specific implementation method of the above preprocessing may be: using a preset coherence coefficient and error estimation value to filter the first time series data.

[0061] In one embodiment, the above-mentioned second training module 26 can be further configured to: divide the feature information into multiple samples to train the optimal model; wherein a single sample includes the feature information obtained by a single prediction point at the third continuous time point of the second preset time length, and the third continuous time point is a partial time point of multiple time points.

[0062] Below, reference Figure 3 The electronic device according to the embodiment of the present application is described. The electronic device may be any one or both of the first device and the second device, or a stand-alone device independent of them, and the stand-alone device may communicate with the first device and the second device to receive the collected input signal from them.

[0063] like Figure 3 As shown, the electronic device 10 includes one or more processors 11 and a memory 12 .

[0064] The processor 11 may be a central processing unit (CPU) or other forms of processing units having data processing capabilities and / or instruction execution capabilities, and may control other components in the electronic device 10 to perform desired functions.

[0065] The memory 12 may include one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may include, for example, a random access memory (RAM) and / or a cache memory (cache), etc. The non-volatile memory may include, for example, a read-only memory (ROM), a hard disk, a flash memory, etc. One or more computer program instructions may be stored on the computer-readable storage medium, and the processor 11 may run the program instructions to implement the methods of the various embodiments of the present application described above and / or other desired functions. Various contents such as input signals, signal components, noise components, etc. may also be stored in the computer-readable storage medium.

[0066] In one example, the electronic device 10 may further include: an input device 13 and an output device 14 , and these components are interconnected via a bus system and / or other forms of connection mechanisms (not shown).

[0067] When the electronic device is a stand-alone device, the input device 13 may be a communication network connector, which is used to receive the collected input signals from the first device and the second device.

[0068] In addition, the input device 13 may also include, for example, a keyboard, a mouse, and the like.

[0069] The output device 14 can output various information to the outside, including the determined distance information, direction information, etc. The output device 14 can include, for example, a display, a speaker, a printer, a communication network and a remote output device connected thereto, and the like.

[0070] Of course, to simplify, Figure 3 Only some of the components related to the present application in the electronic device 10 are shown, and components such as a bus, an input / output interface, etc. are omitted. In addition, according to specific application situations, the electronic device 10 may also include any other appropriate components.

[0071] The computer program product may be written in any combination of one or more programming languages ​​to write program codes for performing the operations of the embodiments of the present application, including object-oriented programming languages, such as Java, C++, etc., and conventional procedural programming languages, such as "C" language or similar programming languages. The program code may be executed entirely on the user computing device, partially on the user device, as an independent software package, partially on the user computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0072] The computer readable storage medium can adopt any combination of one or more readable media. The readable medium can be a readable signal medium or a readable storage medium. The readable storage medium can include, for example, but is not limited to, a system, device or device of electricity, magnetism, light, electromagnetic, infrared, or semiconductor, or any combination of the above. More specific examples (non-exhaustive list) of readable storage media include: an electrical connection with one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), a pluggable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.

[0073] The above description has been given for the purpose of illustration and description. In addition, this description is not intended to limit the embodiments of the present application to the forms disclosed herein. Although multiple example aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, changes, additions and sub-combinations thereof.

Claims

1. A surface deformation prediction method based on multiple models, characterized in that: include: Acquire first time series data of the surface to be predicted; wherein the first time series data includes state information of multiple prediction points on the surface to be predicted acquired at multiple time points; Extracting feature information from the first time series data; wherein the feature information represents the state change amount of a single prediction point at the multiple time points; Using the feature information as samples to train multiple prediction models respectively; wherein the parameters of the multiple prediction models are set within a preset range; Calculating prediction error values ​​of the multiple prediction models; wherein the prediction error values ​​represent the prediction accuracy of the corresponding prediction models; Selecting the prediction model with the smallest prediction error value as the optimal model; Retraining the optimal model using the feature information as a sample to obtain optimal parameters of the optimal model in the entire domain; Acquire second time series data of the surface to be predicted; wherein the second time series data includes state information of all prediction points in the surface to be predicted acquired at multiple time points; The second time series data is input into the trained optimal model to predict the deformation result of the surface to be predicted.

2. The method for predicting ground deformation based on multiple models according to claim 1, characterized in that: The using the feature information as samples to train multiple prediction models respectively comprises: Using the first part of the feature information as training samples to train the multiple prediction models respectively; The calculating the prediction error values ​​of the multiple prediction models comprises: Using the second part of the feature information as a verification sample, respectively calculating the prediction error values ​​of the multiple prediction models; Among them, a single training sample includes feature information of a single prediction point obtained at a first continuous time point of a first preset time length, and a single verification sample includes feature information of the single prediction point obtained at a second continuous time point of a second preset time length, and the first continuous time point and the second continuous time point are different and are both partial time points of the multiple time points.

3. The method for predicting ground deformation based on multiple models according to claim 2, characterized in that: The using the second part of the feature information as a verification sample and calculating the prediction error values ​​of the multiple prediction models respectively includes: Using the second part of the feature information as a validation sample, respectively calculating a plurality of validation errors of each of the prediction models; The absolute average of the multiple validation errors of each of the prediction models is calculated to obtain the prediction error value of each of the prediction models.

4. The method for predicting ground deformation based on multiple models according to claim 1, characterized in that: The step of obtaining first time series data of the surface to be predicted includes: Acquire first image data of the surface to be predicted; wherein the first image data includes image information of multiple prediction points on the surface to be predicted acquired at multiple time points; Based on the first image data, the first time series data is generated by using a small baseline set interferometry measurement technique.

5. The method for predicting ground deformation based on multiple models according to claim 4, characterized in that: After obtaining the first time series data of the surface to be predicted, the surface deformation prediction method based on multiple models further includes: Preprocessing the first time series data to obtain processed first time series data; The extracting feature information from the first time series data includes: Extract feature information from the processed first time series data.

6. The method for predicting ground deformation based on multiple models according to claim 5, characterized in that: The preprocessing of the first time series data includes: The first time series data is filtered using a preset coherence coefficient and an error estimate.

7. The method for predicting ground deformation based on multiple models according to claim 1, characterized in that: The retraining of the optimal model using the feature information as a sample comprises: The feature information is divided into multiple samples to train the optimal model; wherein a single sample includes feature information of a single prediction point obtained at a third continuous time point of a third preset time length, and the third continuous time point is a partial time point of the multiple time points.

8. A surface deformation prediction device based on multiple models, characterized in that: include: A first acquisition module is used to acquire first time series data of the surface to be predicted; wherein the first time series data includes state information of multiple prediction points on the surface to be predicted acquired at multiple time points; An information extraction module, used to extract feature information from the first time series data; wherein the feature information represents the state change amount of a single prediction point at the multiple time points; A first training module is used to train multiple prediction models respectively using the feature information as samples; wherein the parameters of the multiple prediction models are set within a preset range; An error calculation module, used to calculate the prediction error values ​​of the multiple prediction models; wherein the prediction error values ​​represent the prediction accuracy of the corresponding prediction models; A model selection module is used to select the prediction model with the smallest prediction error value as the optimal model; A second training module is used to train the optimal model again using the feature information as a sample to obtain optimal parameters of the optimal model in the entire domain; A second acquisition module is used to acquire second time series data of the surface to be predicted; wherein the second time series data includes state information of all prediction points in the surface to be predicted acquired at multiple time points; The deformation prediction module is used to input the second time series data into the trained optimal model to predict the deformation result of the surface to be predicted.

9. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, and the computer program is used to execute the method according to any one of claims 1 to 7.

10. An electronic device, characterized in that: include: processor; a memory for storing instructions executable by the processor; The processor is used to execute the method described in any one of claims 1 to 7.

Images