Space-time prediction method and device for settlement along railway

Abstract

The invention provides a time-space prediction method and device for settlement along a railway. The prediction method comprises the following steps: identifying an unstable target region based on monitoring data obtained by monitoring a railway line; obtaining time sequence settlement data of each monitoring point in the unstable target area, and performing noise reduction processing by using the optimal noise reduction model to obtain the time sequence settlement data of each monitoring point after noise reduction processing; obtaining an adjacent matrix and a degree matrix based on a preset distance threshold and the distance between every two monitoring points in the unstable target area; and based on the time sequence settlement data, the adjacency matrix and the degree matrix of each monitoring point after noise reduction processing, utilizing a GCN-LSTM prediction model, and adopting a rolling prediction method to carry out advanced prediction on the time sequence settlement data of each monitoring point to obtain an advanced prediction set of each monitoring point. According to the method, the time sequence settlement data of the monitoring points are predicted in advance by constructing the GCN-LSTM prediction model with the space-time sequence coupling characteristics, and the method has the advantage of being high in prediction precision.

Description

Technical Field

[0001] The present invention belongs to the technical field of geological disaster prediction, and in particular relates to a method and device for spatiotemporal prediction of subsidence along a railway. Background Art

[0002] As a vital national transportation infrastructure, the safe operation of railways is directly linked to the national economy and the safety of people's lives and property. In recent years, frequent extreme weather events, geological environmental changes, and increased human engineering activities have led to frequent disasters along railway lines. In particular, geological hazards such as subsidence, karst collapse, and slope instability pose a serious threat to the smoothness of railway lines and driving safety. Traditional monitoring methods, which primarily rely on manual inspections and localized monitoring, suffer from limited coverage, insufficient real-time performance, and delayed hazard identification, making it difficult to promptly identify potential risk areas along long, linear railway lines.

[0003] Furthermore, as a core indicator reflecting the evolution of hazard bodies, surface subsidence trend forecasting has dual value for railway safety and operation: on the one hand, it can assess overall stability through historical and current subsidence measurements, and on the other hand, it can predict potential future risks based on advance subsidence prediction. Current mainstream prediction methods focus on feature mining of single-point time-series subsidence data. While these methods can capture the temporal evolution of monitoring points, they overlook the inherent spatial correlations in InSAR subsidence fields. The lack of spatiotemporal heterogeneity can cause predictions to deviate from the actual distribution pattern of the subsidence field. Summary of the Invention

[0004] The purpose of the present invention is to capture the spatial neighborhood association of the settlement field by introducing a graph convolutional network (GCN), couple the optimal time series prediction model to extract the settlement time series characteristics, and construct a GCN-LSTM prediction model with "space-time series" coupling characteristics to make advance predictions on the time series settlement data of the monitoring points, which has the advantage of high prediction accuracy.

[0005] In a first aspect, the present invention provides a spatiotemporal prediction method for settlement along a railway, the prediction method comprising the following steps: step S10, identifying an unstable target area based on monitoring data obtained by monitoring the railway line; step S20, obtaining the time-series settlement data of each monitoring point in the unstable target area, and performing noise reduction processing using an optimal noise reduction model to obtain the time-series settlement data of each monitoring point after noise reduction processing; step S30, obtaining an adjacency matrix based on a preset distance threshold and the distance between every two monitoring points in the unstable target area, and constructing a degree matrix based on the adjacency relationship of the adjacency matrix; step S40, based on the time-series settlement data of each monitoring point after noise reduction processing, the adjacency matrix and the degree matrix, using a GCN-LSTM prediction model, adopting a rolling prediction method to perform advance prediction on the time-series settlement data of each monitoring point, and obtaining an advance prediction set for each monitoring point, the GCN-LSTM prediction model having a corresponding relationship with the preset distance threshold and being a trained model, wherein the advance prediction set includes a plurality of predicted settlement data having a corresponding relationship with the prediction time.

[0006] In a specific embodiment, the monitoring data includes TQI values ​​collected based on vehicle platforms, settlement data obtained based on aerospace platforms, image data collected by drones based on aviation platforms, inspection data obtained by manual inspections based on ground-based platforms, and geophysical detection data.

[0007] In a specific embodiment, step S10 includes: using the track inspection vehicle of the vehicle platform to obtain the TQI value along the railway, and determining the over-limit section based on the preset TQI value, the TQI value of the over-limit section is greater than or equal to the preset TQI value; based on the aerospace platform, using SBAS-InSAR technology to obtain the time-series settlement data and the average annual settlement rate along the railway, and dividing the railway into multiple settlement target areas and multiple non-sedimentation target areas based on the time-series settlement data and the average annual settlement rate; combining the over-limit section determined by the vehicle platform and the settlement target area and non-sedimentation target area determined by the aviation platform, identifying unstable target areas and target areas to be verified, and if the over-limit section is located in the settlement target area, the settlement target area is identified as an unstable target area; the over-limit section in the non-sedimentation target area and the settlement target area that does not include the over-limit section are both identified as areas to be verified.

[0008] In a specific embodiment, the step S10 also includes: obtaining image data of the area to be verified based on an aerial platform and obtaining inspection data of the area to be verified based on manual inspections on a ground-based platform, and further identifying the area to be verified based on the image data and the inspection data; when the area to be verified has both adjacent construction disturbances and collapse points, the area to be verified is identified as an unstable target area; when the area to be verified has adjacent construction disturbances and no collapse points or there are collapse points and no adjacent construction disturbances, the area to be verified is identified as a further verification area; and the further verification area is verified based on geophysical means.

[0009] In a specific embodiment, the optimal noise reduction model is a wavelet transform model, and the method for obtaining the optimal noise reduction model in step S20 includes: obtaining time-series settlement data of multiple target monitoring points, wherein the multiple target monitoring points are all located in an unstable target area; performing noise reduction processing on the time-series settlement data of each target monitoring point using a wavelet transform model and a Gaussian filter model, respectively, to obtain a first noise reduction result corresponding to the wavelet transform model and a second noise reduction result corresponding to the Gaussian filter model, wherein the first noise reduction result and the second noise reduction result both include the noise-reduced time-series settlement data of each target monitoring point; calculating a noise reduction accuracy evaluation index based on the time-series settlement data of each target monitoring point, the first noise reduction result, and the second noise reduction result, to obtain a first noise reduction accuracy evaluation index corresponding to the wavelet transform model and a second noise reduction accuracy evaluation index corresponding to the Gaussian filter model, wherein the noise reduction accuracy evaluation index includes an average signal-to-noise ratio and an average root mean square error of the multiple target monitoring points; comparing the first noise reduction accuracy evaluation index with the second noise reduction accuracy evaluation index to obtain a first comparison result, and determining that the wavelet transform model is the optimal noise reduction model based on the first comparison result.

[0010] In a specific embodiment, the preset distance threshold is a predetermined optimal distance threshold, and the prediction method further includes a step of obtaining the optimal distance threshold and the GCN-LSTM prediction model, which is located after step S20 and before step S30, including: step (1), based on the time series sedimentation data after noise reduction processing of each monitoring point, using the sliding window method to construct the input features and output features of the time series prediction, and obtaining a time feature data set consisting of multiple groups of data pairs, each group of data pairs including input features and output features with a preset time series length; step (2), splitting the time feature data set into a first training set and a first test set; step (3), first training three time series prediction models based on the first training set, and then evaluating the three trained time series prediction models based on the first test set to obtain the optimal time series prediction model, the three time series prediction models including a Bayesian ridge regression model, a support vector regression model and a long short-term memory network model, and the optimal time series prediction model is a long short-term memory network model; step (4), fusing the GCN model based on the long short-term memory network model to construct a GCN-LSTM prediction model. The method comprises the following steps: setting a plurality of distance thresholds, and constructing a plurality of spatial feature data sets corresponding to the plurality of distance thresholds according to each distance threshold and the distance between each two monitoring points in the unstable target area, wherein each spatial feature data set includes an adjacency matrix and a degree matrix corresponding to the distance threshold; constructing a plurality of second training sets and a plurality of second test sets corresponding to the plurality of distance thresholds based on the time feature data set and the plurality of spatial feature data sets, wherein each second training set includes a first training set and a spatial feature data set, and the spatial feature data sets in different second training sets are different, and each second test set includes a first test set and a spatial feature data set, and the spatial feature data sets in different second test sets are different, and the second training set and the second test set having the same spatial feature data set have a corresponding relationship; and firstly using the plurality of second training sets to train the GCN-LSTM prediction model respectively, and then using the plurality of second test sets to evaluate the trained GCN-LSTM prediction model respectively, and obtaining the optimal distance threshold and the GCN-LSTM prediction model corresponding to the optimal distance threshold.

[0011] In a specific embodiment, the step (3) includes: inputting the first training set into the Bayesian Ridge regression model, the support vector regression model and the long short-term memory network model for training, respectively, to obtain the trained Bayesian Ridge regression model, the support vector regression model and the long short-term memory network model; inputting the first test set into the trained Bayesian Ridge regression model, the support vector regression model and the long short-term memory network model for prediction, respectively, to obtain three groups of first prediction results corresponding to the three time series prediction models, each group of first prediction results includes multiple first prediction settlement data corresponding to multiple input features in the first test set; based on the multiple output features in the first test set and the three groups of first prediction results, respectively calculating the prediction accuracy evaluation index of each time series prediction model, and comparing them to obtain a second comparison result, wherein the prediction accuracy evaluation index includes the average value of the mean absolute error, the average value of the mean square error and the average value of the goodness of fit of multiple monitoring points; based on the second comparison result, determining that the optimal time series prediction model is the long short-term memory network model.

[0012] In a specific embodiment, the step (7) includes: inputting the plurality of second training sets into the GCN-LSTM prediction model for training, obtaining a plurality of trained GCN-LSTM prediction models, wherein the plurality of trained GCN-LSTM prediction models have a one-to-one correspondence with the plurality of distance thresholds; inputting the plurality of second test sets into the trained GCN-LSTM prediction model with the corresponding relationship for prediction, obtaining a plurality of groups of second prediction results corresponding one-to-one to the plurality of distance thresholds, wherein each group of second prediction results includes a plurality of second predicted settlement data corresponding one-to-one to the plurality of input features in the first test set; based on the plurality of output features in the first test set and the plurality of groups of second prediction results, respectively calculating the prediction accuracy evaluation index of each trained GCN-LSTM prediction model, and comparing them to obtain a third comparison result, wherein the prediction accuracy evaluation index includes the average value of the mean absolute error, the average value of the mean square error and the average value of the goodness of fit of the plurality of monitoring points; based on the third comparison result, obtaining the optimal distance threshold and the GCN-LSTM prediction model corresponding to the optimal distance threshold.

[0013] In a specific embodiment, the preset time series length is N, and the time series length of the rolling prediction method is set according to the preset time series length. The steps of using the rolling prediction method to make any advance prediction of any one of the multiple monitoring points include: step (a), obtaining target time series settlement data for the Mth settlement data prediction, the target time series settlement data is composed of the latest N settlement data, and the N settlement data included in the target time series settlement data are measured settlement data and / or predicted settlement data; step (b), using the target time series settlement data, the adjacency matrix and the degree matrix as input, and using the GCN-LSTM prediction model to obtain the predicted settlement data for the Mth prediction.

[0014] In a second aspect, the present invention provides a spatiotemporal prediction device for settlement along a railway, the prediction device comprising: an identification module for identifying an unstable target area based on monitoring data obtained by monitoring the railway line; a noise reduction processing module for obtaining the time-series settlement data of each monitoring point in the unstable target area, and performing noise reduction processing using an optimal noise reduction model to obtain the time-series settlement data of each monitoring point after noise reduction processing; a spatial feature construction module for constructing an adjacency matrix based on a preset distance threshold and the distance between every two monitoring points in the unstable target area, and constructing a degree matrix based on the adjacency relationship of the adjacency matrix; a prediction module for performing advance prediction on the time-series settlement data of each monitoring point based on the time-series settlement data after noise reduction processing of each monitoring point, the adjacency matrix and the degree matrix, using a GCN-LSTM prediction model and a rolling prediction method to obtain an advance prediction set for each monitoring point, the GCN-LSTM prediction model having a corresponding relationship with the preset distance threshold and being a trained model, wherein the advance prediction set includes a plurality of predicted settlement data having a corresponding relationship with the prediction time.

[0015] The beneficial effects of the present invention include at least:

[0016] 1. The spatiotemporal prediction method for settlement along the railway provided by the present invention first identifies the unstable target area, and then based on the time-series settlement data of the monitoring points in the unstable target area and the spatial feature data between multiple monitoring points constructed based on a preset distance threshold, uses the GCN-LSTM prediction model and adopts a rolling prediction method to make advance predictions on the time-series settlement data of each monitoring point. In this way, the present invention introduces a graph convolutional network (GCN) to capture the spatial neighborhood association of the settlement field, couples the optimal time-series prediction model LSTM to extract the settlement time-series features, and constructs a GCN-LSTM prediction model with "space-time" coupling features to make advance predictions on the time-series settlement data of the monitoring points, which has the advantage of high prediction accuracy. At the same time, the present invention mainly makes advance predictions on the settlement data of the monitoring points in the unstable target area. On the one hand, it solves the technical problem of difficult monitoring of unstable areas along long linear railways, and on the other hand, it makes advance predictions in key areas, with high monitoring efficiency and low monitoring cost.

[0017] 2. The present invention constructs a multi-dimensional, three-dimensional system for early identification of unstable target areas along railways. By integrating track inspection vehicle detection, large-scale aerospace monitoring and fine ground exploration technology, a hierarchical identification principle is established, thus breaking through the limitations of traditional methods and improving the timeliness and accuracy of identifying unstable target areas along long and large railway lines.

[0018] In addition to the above-described objects, features and advantages, the present invention has other objects, features and advantages. The present invention will be further described in detail below with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 A schematic flow chart of the steps of a spatiotemporal prediction method for railway settlement along a railway provided in one embodiment of the present invention;

[0020] Figure 2 for Figure 1 The schematic diagram of the identification process of unstable target areas along the railway corresponding to step S10 is shown;

[0021] Figure 3 This is a flowchart of a method for spatiotemporal prediction of railway settlement corresponding to Example 1 of the present invention;

[0022] Figure 4 This is a diagram of monitoring results based on a vehicle platform and a space platform in Example 1 of the present invention, wherein: Figure 4 (a) is a graph showing the relationship between the sedimentation rate obtained from the aerial platform and the monitoring areas along the railway. Figure 4 (b) is a graph showing the relationship between the settlement data obtained from an aerial platform and the location along the railway. Figure 4 (c) in the figure is the TQI value along the railway obtained based on the vehicle platform;

[0023] Figure 5 The image data obtained based on the aerial platform in Example 1 of the present invention;

[0024] Figure 6 This is an adjacency graph of 9 monitoring points constructed according to different distance thresholds in Example 1 of the present invention, wherein: Figure 6 The distance threshold corresponding to (a) is 600m. Figure 6 The distance threshold corresponding to (b) is 800m. Figure 6 The distance threshold corresponding to (c) in the figure is 1000m. Figure 6 The distance threshold corresponding to (d) in is 1200m. Figure 6 The distance threshold corresponding to (e) in is 1400m. Figure 6 The distance threshold corresponding to (f) in the figure is 1600m;

[0025] Figure 7 This is the mean absolute error of the nine monitoring points obtained when the GCN-LSTM prediction model corresponding to different distance thresholds (600m, 800m, 1000m, 1200m, and 1400m) is used for prediction in Example 1 of the present invention;

[0026] Figure 8 This is the mean square error of the nine monitoring points obtained when the GCN-LSTM prediction model corresponding to different distance thresholds (600m, 800m, 1000m, 1200m, and 1400m) is used for prediction in Example 1 of the present invention;

[0027] Figure 9 The goodness-of-fit results of the nine monitoring points obtained when the GCN-LSTM prediction model corresponding to different distance thresholds (600m, 800m, 1000m, 1200m, and 1400m) is used for prediction in Example 1 of the present invention;

[0028] Figure 10 The comparison result of the mean absolute error of the 9 monitoring points corresponding to the GCN-LSTM prediction model in Example 1 of the present invention and the mean absolute error of the 9 monitoring points corresponding to the LSTM time series prediction model;

[0029] Figure 11 The comparison result of the mean square error of the 9 monitoring points corresponding to the GCN-LSTM prediction model in Example 1 of the present invention and the mean square error of the 9 monitoring points corresponding to the LSTM time series prediction model;

[0030] Figure 12 The comparison result of the goodness of fit of the 9 monitoring points corresponding to the GCN-LSTM prediction model in Example 1 of the present invention and the goodness of fit of the 9 monitoring points corresponding to the LSTM time series prediction model;

[0031] Figure 13 This is a time series settlement data diagram drawn based on the measured data set and the advance prediction set of 9 monitoring points in Example 1 of the present invention.

[0032] Figure 14 This is a module diagram of a device for spatiotemporal prediction of settlement along a railway provided in another embodiment of the present invention. DETAILED DESCRIPTION

[0033] The embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0034] In view of the defect that the existing prediction methods mostly focus on the feature mining of single-point time series settlement data and ignore the inherent spatial correlation characteristics of InSAR settlement fields, the present invention introduces a graph convolutional network (GCN) to capture the spatial neighborhood correlation of the settlement field, couples the optimal time series prediction model to extract the settlement time series features, and constructs a GCN-LSTM prediction model with "space-time series" coupling features to make advance predictions on the time series settlement data of the monitoring points, which has the advantage of high prediction accuracy; at the same time, it aims at the limited coverage, lack of real-time performance and To solve the problem of delayed hidden danger identification, the present invention constructs a multi-dimensional, three-dimensional early identification system for unstable target areas along the railway. By integrating track inspection vehicle detection, large-scale aerospace monitoring and fine foundation exploration technology, a hierarchical identification principle is established, thereby breaking through the limitations of traditional methods and improving the timeliness and accuracy of unstable target area identification on long and large railway lines. On the one hand, it can reduce monitoring costs, and on the other hand, it can also realize early prediction of settlement data of each monitoring point in key areas (unstable target areas). In this way, early warning and prevention of settlement risks in unstable target areas can be achieved based on the predicted settlement data.

[0035] It should be noted that the time-series settlement data in the present invention is composed of several settlement data corresponding to the time characteristics in sequence (in chronological order), and the settlement data is the cumulative settlement amount, such as the settlement data A1 of 2023.01.01, the settlement data A2 of 2023.03.01, the settlement data A3 of 2023.05.01, etc.

[0036] See also Figure 1 According to a first aspect of the present invention, the present invention provides a spatiotemporal prediction method for settlement along a railway, the prediction method comprising the following steps:

[0037] Step S10: Identify unstable target areas based on monitoring data obtained from monitoring along the railway line.

[0038] In the present invention, a multi-faceted approach is used to monitor the railway line, and unstable target areas are identified based on the monitoring data obtained by the multi-faceted approach.

[0039] In an optional embodiment, the monitoring data includes TQI values ​​collected based on a vehicle platform, time-series settlement results obtained based on a space platform, image data collected by a drone based on an aviation platform, inspection data obtained by manual inspection based on a ground-based platform, and geophysical detection results.

[0040] Specifically, the vehicle platform collects TQI values ​​through track inspection vehicles; the aerospace platform uses SBAS-InSAR technology to obtain time-series settlement results along the railway, mainly based on Sentinel-1A images, 30m spatial resolution SRTM DEM data released by NASA, and precise orbit determination (POD) data (Precise Orbit Ephemerides) of the Sentinel-1A satellite provided by the European Space Agency; the aviation platform collects image data through drones; the ground-based platform performs manual inspections; and geophysical exploration methods include multi-source frequency domain seismic wave method and micro-motion detection method.

[0041] Please refer to Figure 2 , the step S10 includes:

[0042] Step (1): using the track inspection vehicle on the vehicle platform to obtain the TQI value along the railway, and determining the over-limit section based on the preset TQI value, wherein the TQI value of the over-limit section is greater than or equal to the preset TQI value.

[0043] In the present invention, the preset TQI value is 15, which is determined based on the existing grading standard.

[0044] Step (2): Based on the aerospace platform, SBAS-InSAR technology is used to obtain time-series settlement data and annual average settlement rate along the railway, and based on the time-series settlement data and annual average settlement rate, the railway is divided into multiple settlement target areas and multiple non-sedimentation target areas.

[0045] In the present invention, the sedimentation target area has an obvious sedimentation funnel compared with other areas. It can be understood that the maximum absolute value of the sedimentation data in the sedimentation target area is significantly larger than the absolute value of the sedimentation data in other areas. Distinguishing between sedimentation target areas and non-sedimentation target areas belongs to the existing technology, and technical personnel in this field make judgments based on experience or changes in sedimentation data.

[0046] In the present invention, the core calculation formula of SBAS-InSAR technology is:

[0047]

[0048] In the formula is the interference phase of the pixel (x, r) in the j-th interference pattern; t B , t AThe phase at time (x, r); is the deformation phase of the radar line of sight direction; They are the terrain phase, orbit error, atmospheric delay phase and noise in the interferometer respectively.

[0049] Step (3): combining the over-limit section determined by the vehicle platform and the settlement target area and non-settlement target area determined by the aviation platform, identifying the unstable target area and the target area to be verified. If the over-limit section is located in the settlement target area, the settlement target area is identified as the unstable target area; the over-limit section in the non-settlement target area and the settlement target area not including the over-limit section are both identified as areas to be verified.

[0050] In the present invention, the railway line segments along the railway that do not belong to the unstable target area and the area to be verified are all safe areas.

[0051] In step (3), when identifying unstable target areas, both TQI value and settlement characteristics are considered. If there is obvious settlement characteristics and the TQI value is greater than the preset TQI value (that is, both limiting conditions are met), it is considered to be an unstable target area that needs to be monitored. It is necessary to predict the future time-series settlement data of the unstable target area so as to provide early warning and prevention. If only the TQI value is greater than the preset TQI value or the settlement characteristics meet the preset conditions, it is considered to be an area to be verified and further other monitoring data need to be obtained for further judgment.

[0052] For easier understanding, please refer to Figure 4 ,from Figure 4 It can be seen that the TQI value exceeding limit section K2261+200 is located in the settlement target area B, so the settlement target area is an unstable target area, while the settlement target areas A, C and mileages K2250.6 and K2250.8 do not meet the two limiting conditions at the same time, and are considered to be areas to be verified.

[0053] Step (4): obtaining image data of the area to be verified based on an aerial platform and obtaining inspection data of the area to be verified based on manual inspection on a ground-based platform, and further identifying the area to be verified based on the image data and the inspection data; when the area to be verified has both adjacent construction disturbances and collapse points, the area to be verified is identified as an unstable target area; when the area to be verified has adjacent construction disturbances and no collapse points, or has collapse points and no adjacent construction disturbances, the area to be verified is identified as a further verification area.

[0054] In the present invention, image data of the area to be verified is taken by an unmanned aerial vehicle to determine whether there is any adjacent construction disturbance in the area to be verified, and manual inspection is used to determine whether there is a collapse point in the area to be verified. If both adjacent construction disturbance and collapse point exist, the area to be verified is identified as an unstable target area. If the judgment result is that both adjacent construction disturbance and collapse point do not exist, the area to be verified is identified as a safe area. If the judgment result is that there is an adjacent construction disturbance and no collapse point or there is a collapse point and no adjacent construction disturbance, the area to be verified is identified as a further verification area.

[0055] The present invention combines drone aerial photography with manual inspection to determine whether there is any adjacent construction disturbance and collapse points in the area to be verified. The collected data is highly reliable and can further improve the accuracy of unstable target area identification.

[0056] Step (5) verifies the further verification area based on geophysical means.

[0057] In the present invention, the further verification area is detected by geophysical means, and whether the area is stable is determined based on the detection results. If it is unstable, it is an unstable target area.

[0058] In the present invention, the unstable target area is determined by multiple means, and then the time series settlement data of the monitoring points in the unstable target area are predicted in advance. On the one hand, the monitoring cost can be reduced, and on the other hand, the unstable target area can be monitored in a focused manner to provide timely warnings.

[0059] Step S20: acquiring the time series sedimentation data of each monitoring point in the unstable target area, and performing noise reduction processing using the optimal noise reduction model to obtain the time series sedimentation data of each monitoring point after noise reduction processing.

[0060] In the present invention, the SBAS-InSAR technology is used to obtain the time-series settlement data of each monitoring point in the unstable target area.

[0061] In the present invention, the optimal noise reduction model is a wavelet transform model.

[0062] In an optional embodiment, the method for obtaining the optimal noise reduction model in step S20 includes:

[0063] Step (a): obtaining time-series sedimentation data of a plurality of target monitoring points, wherein the plurality of target monitoring points are all located in an unstable target area.

[0064] When selecting the optimal noise reduction model, the time series settlement data of some or all monitoring points in the unstable target area can be selected as the data to be processed.

[0065] For ease of understanding, for example, assuming that the number of unstable target areas determined in step S10 is 3, and each unstable target area includes 5 monitoring points, in step (a), 3 monitoring points or other number of monitoring points can be selected as target monitoring points in one unstable target area, or 1 or more monitoring points can be selected as target monitoring points in each unstable target area.

[0066] Step (b) performs denoising on the time series settlement data of each target monitoring point using a wavelet transform model and a Gaussian filter model, respectively, to obtain a first denoising result corresponding to the wavelet transform model and a second denoising result corresponding to the Gaussian filter model, wherein the first denoising result and the second denoising result both include the denoised time series settlement data of each target monitoring point.

[0067] In the present invention, the wavelet transform model is referred to as the WT (Wavelet Transform) model; the Gaussian filtering model is referred to as the GF (Gaussian Filtering) model.

[0068] Step (c) calculates a noise reduction accuracy evaluation index based on the time-series sedimentation data of each target monitoring point, the first noise reduction result and the second noise reduction result, and obtains a first noise reduction accuracy evaluation index corresponding to the wavelet transform model and a second noise reduction accuracy evaluation index corresponding to the Gaussian filter model. The noise reduction accuracy evaluation index includes the average signal-to-noise ratio and the average root mean square error of multiple target monitoring points.

[0069] In the present invention, the calculation of the signal-to-noise ratio (SNR) and the root mean square error (RMSE) are both existing technologies and will not be described in detail here.

[0070] It should be noted that when there are multiple target monitoring points, each target monitoring point will obtain a corresponding signal-to-noise ratio and root mean square error when calculating the noise reduction accuracy evaluation index. When performing model evaluation, the arithmetic mean of the signal-to-noise ratio and root mean square error of all target monitoring points is used.

[0071] For ease of understanding, an example is given, assuming that the number of target monitoring points is 5, the average signal-to-noise ratio is the arithmetic mean of the signal-to-noise ratios of the 5 monitoring points, and the average root mean square error is the arithmetic mean of the root mean square errors of the 5 monitoring points.

[0072] Step (d): comparing the first denoising accuracy evaluation index and the second model evaluation model index to obtain a first comparison result, and determining the wavelet transform model as the optimal denoising model based on the first comparison result.

[0073] In the present invention, the first comparison result includes that the signal-to-noise ratio of the wavelet transform model is greater than the signal-to-noise ratio of the Gaussian filter model, and the mean square error of the Gaussian filter model is greater than the mean square error of the wavelet transform model.

[0074] Step S30: obtaining an adjacency matrix based on a preset distance threshold and the distance between every two monitoring points in the unstable target area, and constructing a degree matrix based on the adjacency relationship of the adjacency matrix.

[0075] In the present invention, the predetermined optimal distance threshold is 800m.

[0076] In the present invention, constructing the adjacency matrix and the degree matrix are both existing technologies. The innovation of the present invention is that the spatial feature data is taken into consideration when predicting the settlement data of the monitoring points.

[0077] Construct the adjacency matrix A according to the distance between monitoring points, where i, j, k are monitoring points, e i，j The weight of the edge between monitoring point i and monitoring point j is calculated based on the geographical location (coordinates) of the monitoring points. The specific calculation formula is as follows:

[0078]

[0079] Where, d i,j is the distance between two monitoring points; H is the distance threshold.

[0080] As can be seen from the weight calculation formula above, when the distance between two monitoring points exceeds the distance threshold, the weight value is 0, which means that in practice, only the weight of the edge between two monitoring points with a distance less than the distance threshold H is calculated. In this way, based on different distance thresholds, the adjacency matrix and degree matrix constructed are different, so the adjacency matrix and degree matrix in the spatial feature data have a corresponding relationship with the distance threshold.

[0081] Step S40: Based on the time series settlement data after noise reduction of each monitoring point and the spatial feature data, the GCN-LSTM prediction model is used to adopt a rolling prediction method to perform advance prediction on the time series settlement data of each monitoring point, and an advance prediction set for each monitoring point is obtained. The GCN-LSTM prediction model has a corresponding relationship with a preset distance threshold and is a trained model, wherein the advance prediction set includes multiple predicted settlement data that have a corresponding relationship with the prediction time.

[0082] In the present invention, the rolling prediction method is used to construct the input features and output features of time series prediction. When the time series length is 5, in the first set of data pairs, the input feature is L t-4 , L t-3 , L t-2 , L t-1 , L t , the output feature is L t+1 ; For the a-th group of data pairs, the input feature is L t+a-5 , L t+a-4 , Lt+a-3 , L t+a-2 , L t+a-1 , the output feature is L t+a , a total of a group of settlement data are output.

[0083] It should be noted that the rolling prediction method used in this invention is an existing technology and will not be described in detail here. For details, please refer to the "Deformation Monitoring and Prediction Method along Railways in Karst Areas Based on Time Series InSAR and Machine Learning" published in the Journal of Central South University in March 2025.

[0084] It can be understood that the time interval between two adjacent predicted settlement data in the advance prediction set is the same as the time interval between two adjacent settlement data in the time series settlement data.

[0085] For ease of understanding, let's take an example. Assume that the interval between two adjacent settlement data in the time series settlement data is 12 days, that is, settlement data is obtained every 12 days; the advance prediction time is 600 days, then the advance prediction set includes 50 predicted settlement data, and the interval between two adjacent predicted settlement data is 12 days.

[0086] In an optional implementation, the preset distance threshold is a predetermined optimal distance threshold.

[0087] In an optional embodiment, the prediction method further includes a step of obtaining an optimal distance threshold and the GCN-LSTM prediction model, which is located after step S20 and before step S30, including:

[0088] Step (1): Based on the time series sedimentation data after noise reduction at each monitoring point, the sliding window method is used to construct the input features and output features of the time series prediction, and a time feature data set consisting of multiple groups of data pairs is obtained, each group of data pairs includes input features and output features with a preset time series length.

[0089] In this invention, the sliding window method is used to construct the input features and output features of time series prediction. When the time series length is 5, in the first group of data pairs, the input features are L1, L2, L3, L4, L5, and the output feature is L6; for the t-5 group of data pairs, the input feature is L t-5 , L t-4 , L t-3 , L t-2 , L t-1 , the output feature is L t , to complete the construction of the time feature dataset.

[0090] It should be noted that the sliding window method for constructing input and output features for time series prediction is an existing technology and will not be elaborated on here. For details, please refer to the "Deformation Monitoring and Prediction Method along Railways in Karst Regions Based on Time Series InSAR and Machine Learning" published in the Journal of Central South University in March 2025.

[0091] Step (2): split the time feature data set into a first training set and a first test set.

[0092] In the present invention, the time feature dataset is split into a first training set and a first test set in a ratio of 8:2.

[0093] Step (3): first train the three time series prediction models based on the first training set, and then evaluate the three trained time series prediction models based on the first test set to obtain the optimal time series prediction model. The three time series prediction models include a Bayesian ridge regression model, a support vector regression model, and a long short-term memory network model. The optimal time series prediction model is a long short-term memory network model.

[0094] In the present invention, the Bayesian Ridge Regression model is referred to as the Bayesian Ridge (Bayesian Ridge Regression) model; the Support Vector Regression model is referred to as the SVR (Support Vector Regression) model; and the Long Short-Term Memory Network model is referred to as the LSTM (Long Short-Term Memory) model.

[0095] This step includes:

[0096] Step 3.1: Input the first training set into the Bayesian Ridge regression model, the support vector regression model and the long short-term memory network model for training respectively to obtain the trained Bayesian Ridge regression model, the support vector regression model and the long short-term memory network model.

[0097] Step 3.2: Input the first test set into the trained Bayesian Ridge regression model, support vector regression model and long short-term memory network model for prediction, and obtain three groups of first prediction results corresponding to the three time series prediction models. Each group of first prediction results includes multiple first prediction settlement data corresponding to multiple input features in the first test set.

[0098] In the present invention, the first predicted settlement data obtained by taking the input feature as input has a corresponding relationship between the two.

[0099] For ease of understanding, for example, assuming that there are 200 input features, each first prediction result includes 200 first predicted settlement data predicted based on the 200 input features.

[0100] Step 3.3: Based on the multiple output features in the first test set and the three groups of first prediction results, the prediction accuracy evaluation index of each time series prediction model is calculated respectively, and compared to obtain a second comparison result, wherein the evaluation index includes the average value of the mean absolute error of multiple monitoring points, the average value of the mean square error, and the average value of the goodness of fit.

[0101] In this embodiment, the multiple monitoring points are all monitoring points. In other embodiments, the multiple monitoring points may also be some monitoring points, for example, excluding the monitoring points corresponding to the minimum mean absolute error and the maximum mean absolute error.

[0102] The steps for obtaining the prediction accuracy evaluation index of any of the three time series prediction models include:

[0103] Step (a): obtaining a first prediction result corresponding to the time series prediction model.

[0104] Step (b), based on the first prediction result corresponding to the time series prediction model and multiple output features in the first test set, obtain the mean absolute error, mean square error and goodness of fit corresponding to each monitoring point according to the calculation formula of the mean absolute error, the calculation formula of the mean square error and the calculation formula of the goodness of fit.

[0105] The calculation method of the corresponding mean absolute error, mean square error and goodness of fit of any monitoring point is: based on the first prediction result corresponding to the time series prediction model, multiple first predicted settlement data having a corresponding relationship with each input feature of the monitoring point are obtained, the output feature corresponding to the input feature in the first test set is used as the true value, and the first predicted settlement data corresponding to the input feature is used as the predicted value, and the mean absolute error, mean square error and goodness of fit of the monitoring point are obtained according to the calculation formula of the mean absolute error, the calculation formula of the mean square error and the calculation formula of the goodness of fit.

[0106] Step (c): Based on the calculation formula of the arithmetic mean, the average value of the mean absolute error, the average value of the mean square error and the average value of the goodness of fit of the multiple monitoring points are obtained.

[0107] Step (d): The average value of the mean absolute error, the average value of the mean square error, and the average value of the goodness of fit calculated in step (c) are used as prediction accuracy evaluation indicators of the time series prediction model.

[0108] Step 3.4: Based on the second comparison result, determine that the optimal time series prediction model is a long short-term memory network model.

[0109] In the present invention, the second comparison results include that the mean absolute errors and mean square errors of the long short-term memory network model, the support vector regression model and the Bayesian ridge regression model increase in sequence and the goodness of fit decreases in sequence.

[0110] Step (4) is to fuse the GCN model with the long short-term memory network model to construct a GCN-LSTM prediction model.

[0111] GCN (Graph Convolutional Network) is a deep learning model based on graph structure, which realizes graph data feature learning by aggregating information of neighboring nodes.

[0112] Step (5) sets multiple distance thresholds, and constructs multiple spatial feature data sets corresponding to the multiple distance thresholds based on each distance threshold and the distance between each two monitoring points in the unstable target area, each spatial feature data set includes an adjacency matrix and a degree matrix corresponding to the distance threshold.

[0113] In the present invention, the setting of the distance threshold is related to the minimum distance and the maximum distance between two monitoring points and the settlement monitoring spacing specified in the Railway Roadbed Design Code.

[0114] In the present invention, multiple distance thresholds are set as 600m, 800m, 1000m, 1200m, 1400m and 1600m, and the optimal distance threshold is determined to be 800m.

[0115] In the present invention, a method for constructing a spatial feature data set corresponding to any distance threshold includes: constructing an adjacency matrix based on the distance threshold and the distance between every two monitoring points in the unstable target area; and constructing a degree matrix based on the adjacency relationship of the adjacency matrix.

[0116] Step (6): Based on the time feature data set and the multiple spatial feature data sets, multiple second training sets and multiple second test sets corresponding to the multiple distance thresholds are constructed, each of the second training sets includes the first training set and the spatial feature data set, and the spatial feature data sets in different second training sets are different; each second test set includes the first test set and the spatial feature data set, and the spatial feature data sets in different second test sets are different; the second training set and the second test set with the same spatial feature data set have a corresponding relationship.

[0117] For ease of understanding, an example is given, assuming that a total of three distance thresholds are set, namely the first distance threshold, the second distance threshold and the third distance threshold. Based on step (5), a first adjacency matrix and a first degree matrix corresponding to the first distance threshold, a second adjacency matrix and a second degree matrix corresponding to the second distance threshold, and a third adjacency matrix and a third degree matrix corresponding to the third distance threshold are constructed.

[0118] Correspondingly, the number of the second training set and the second test set obtained in step (6) is also 3. The second training set 01 corresponding to the first distance threshold consists of the first training set, the first adjacency matrix and the first degree matrix, and the second test set 01 corresponding to the first distance threshold consists of the first test set, the first adjacency matrix and the first degree matrix. The second training set 01 and the second test set 01 have a corresponding relationship; the second training set 02 corresponding to the second distance threshold consists of the first training set, the second adjacency matrix and the second degree matrix, and the second test set 02 corresponding to the second distance threshold consists of the first test set, the second adjacency matrix and the second degree matrix. The second training set 02 and the second test set 02 have a corresponding relationship; the second training set 03 corresponding to the third distance threshold consists of the first training set, the third adjacency matrix and the third degree matrix, and the second test set 03 corresponding to the third distance threshold consists of the first test set, the third adjacency matrix and the third degree matrix. The third training set 03 and the third test set 03 have a corresponding relationship.

[0119] Step (7): First, use multiple second training sets to train the GCN-LSTM prediction model respectively, and then use multiple second test sets to evaluate the trained GCN-LSTM prediction model respectively to obtain the optimal distance threshold and the GCN-LSTM prediction model corresponding to the optimal distance threshold.

[0120] This step includes:

[0121] Step 7.1: Input the plurality of second training sets into the GCN-LSTM prediction model for training, and obtain a plurality of trained GCN-LSTM prediction models. The plurality of trained GCN-LSTM prediction models have a one-to-one correspondence with the plurality of distance thresholds.

[0122] Step 7.2: Input multiple second test sets into the trained GCN-LSTM prediction model with corresponding relationships for prediction, and obtain multiple groups of second prediction results corresponding one-to-one to multiple distance thresholds. Each group of second prediction results includes multiple second predicted settlement data corresponding one-to-one to multiple input features in the first test set.

[0123] In step 7.2, multiple second test sets are input into the corresponding trained GCN-LSTM prediction models for prediction. This can be understood as follows: second test set 01 is input into the GCN-LSTM prediction model trained with second training set 01 for prediction, second test set 02 is input into the GCN-LSTM prediction model trained with second training set 02 for prediction, and second test set 03 is input into the GCN-LSTM prediction model trained with second training set 03 for prediction. That is, when the GCN-LSTM prediction model is trained and evaluated, the spatial feature data set used for training and evaluating the model is the same.

[0124] Step 7.3. Based on the multiple output features and multiple groups of second prediction results in the first test set, calculate the prediction accuracy evaluation index of each trained GCN-LSTM prediction model respectively, and compare them to obtain a third comparison result, wherein the evaluation index includes the average of the mean absolute error, the average of the mean square error, and the average of the goodness of fit of multiple monitoring points.

[0125] In an embodiment of the present invention, the multiple monitoring points are all monitoring points. In other embodiments, the multiple monitoring points may also be some monitoring points, for example, excluding the monitoring points corresponding to the minimum mean absolute error and the maximum mean absolute error.

[0126] The steps for obtaining the prediction accuracy evaluation index of the GCN-LSTM prediction model corresponding to any distance threshold include:

[0127] Step (a): obtaining a second prediction result corresponding to the distance threshold.

[0128] For ease of understanding, for example, assuming that the distance threshold is 800m, this step is to obtain the second prediction result output by the GCN-LSTM prediction model corresponding to the distance threshold 800m.

[0129] Step (b), based on the second prediction result corresponding to the distance threshold and multiple output features in the first test set, according to the calculation formula of the mean absolute error, the calculation formula of the mean square error and the calculation formula of the goodness of fit, obtain the mean absolute error, mean square error and goodness of fit corresponding to each monitoring point.

[0130] The calculation method of the mean absolute error, mean square error and goodness of fit corresponding to any monitoring point is: based on the second prediction result corresponding to the distance threshold, multiple second predicted settlement data having a corresponding relationship with the multiple input features of the monitoring point are obtained, the output feature corresponding to the input feature in the first test set is used as the true value, and the second predicted settlement data corresponding to the input feature is used as the predicted value, and the mean absolute error, mean square error and goodness of fit of the monitoring point are obtained according to the calculation formula of the mean absolute error, the calculation formula of the mean square error and the calculation formula of the goodness of fit.

[0131] Step (c): Based on the calculation formula of the arithmetic mean, the average value of the mean absolute error, the average value of the mean square error and the average value of the goodness of fit of the multiple monitoring points are obtained.

[0132] Step (d): The average value of the mean absolute error, the average value of the mean square error, and the average value of the goodness of fit calculated in step (c) are used as the prediction accuracy evaluation indicators of the GCN-LSTM prediction model corresponding to the distance threshold.

[0133] Step 7.4: Based on the third comparison result, obtain the optimal distance threshold and the GCN-LSTM prediction model corresponding to the optimal distance threshold.

[0134] In an optional embodiment, the step of obtaining the optimal distance threshold and the GCN-LSTM prediction model also includes comparing the prediction accuracy evaluation index of the GCN-LSTM prediction model corresponding to the optimal distance threshold with the prediction accuracy evaluation index of the LSTM prediction model.

[0135] In an optional embodiment, the preset time series length is N, and the time series length of the rolling prediction method is set according to the preset time series length. The step of using the rolling prediction method to make any advance prediction for any one of the multiple monitoring points includes:

[0136] Step (a): obtaining target time series settlement data for M-th settlement data prediction, wherein the target time series settlement data is composed of the latest N settlement data.

[0137] When the preset time series length N is 5, the specific steps are: add the settlement data obtained by the M-1 prediction to the target time series settlement data corresponding to the prediction of the M-1 settlement data, delete the M-6 settlement data, and use the M-1, M-2, M-3, M-4 and M-5 settlement data as the target time series settlement data for predicting the M settlement data.

[0138] In an optional embodiment, the N settlement data included in the target time-series settlement data are measured settlement data and / or predicted settlement data.

[0139] For ease of understanding, let's take an example. Assume that the length of the time series N is 5, the aerospace platform obtains settlement data every 12 days, and the latest settlement data is obtained on June 8, 2025. If we want to predict the settlement data on June 20, 2025, the target time series settlement data includes C corresponding to June 8, 2025. a-1 、2025.05.27 corresponding to C a-2 、2025.05.15 corresponding to C a-3 , C corresponding to 2025.05.03 a-4 , C corresponding to 2025.04.21 a-5 After completing the settlement data prediction for June 20, 2025, if you want to predict the settlement data for July 2, 2025, the target time series settlement data includes the settlement prediction value for June 20, 2025, the C corresponding to June 8, 2025, and the settlement prediction value for July 2, 2025. a-1 、2025.05.27 corresponding to C a-2 、2025.05.15 corresponding to C a-3 , C corresponding to 2025.05.03 a-4 ...and so on, that is, when making the first prediction, the five settlement data are all measured settlement data; when making the second prediction, the latest one of the five settlement data is the predicted settlement data, and the other four are all measured settlement data; when making the third prediction, the two latest ones of the five settlement data are predicted settlement data, and the other three are all measured settlement data; ...when making the sixth and subsequent predictions, the five settlement data are all predicted settlement data.

[0140] Step (b): using the target time series settlement data, the adjacency matrix and the degree matrix as input, and using the GCN-LSTM prediction model to obtain the predicted settlement data for the Mth prediction.

[0141] See also Figure 14According to a second aspect of the present invention, a device 100 for spatiotemporal prediction of settlement along a railway is provided. The device 100 comprises: an identification module 101 for identifying an unstable target area based on monitoring data obtained by monitoring the railway; a noise reduction processing module 102 for obtaining time series settlement data of each monitoring point in the unstable target area, and performing noise reduction processing using an optimal noise reduction model to obtain time series settlement data of each monitoring point after noise reduction processing; a spatial feature construction module 103 for constructing a spatial feature based on a predetermined optimal distance threshold and the distance between each two monitoring points in the unstable target area. distance, construct an adjacency matrix, and construct a degree matrix based on the adjacency relationship of the adjacency matrix; a prediction module 104 is used to use the GCN-LSTM prediction model to make an advance prediction of the time series settlement data of each monitoring point based on the time series settlement data after noise reduction of each monitoring point, the adjacency matrix and the degree matrix, and adopt a rolling prediction method to obtain an advance prediction set for each monitoring point, wherein the GCN-LSTM prediction model has a corresponding relationship with the optimal distance threshold and is a trained model, wherein the advance prediction set includes a plurality of predicted settlement data having a corresponding relationship with the prediction time.

[0142] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the working process of the prediction device described above can refer to the content of the aforementioned method embodiment and will not be repeated here.

[0143] Example 1

[0144] To better demonstrate the prediction method of the present invention, a railway section was selected for the experiment based on a specific conventional railway example. InSAR subsidence monitoring was mainly carried out based on Sentinel-1A images, SRTM DEM data with a spatial resolution of 30 m released by NASA, and Precise Orbit Ephemerides (POD) data of the Sentinel-1A satellite provided by the European Space Agency. The spatial baseline threshold (the maximum spatial position difference between two SAR images) was set to 4%, and the temporal baseline threshold was set to 90 days.

[0145] Please refer to Figure 3 , the implementation steps are as follows:

[0146] 1.1 Unstable target area identification

[0147] First, according to the TQI results obtained by the track inspection vehicle, it can be seen that the TQI value trends of the two inspections are roughly the same. It is found that there are three over-limit sections along the railway, namely K2250.6, K2250.8 and K2261.2. Figure 4(c) At the same time, the "space platform" SBAS-InSAR technology was used to carry out a large-scale "census" along a section of the railway. The settlement rate and cumulative settlement results are shown as follows: Figure 4 (a) and (b) show that there are significant differences in settlement characteristics along the railway. Among them, the settlement of areas A (K2253-K2256+500 / K2256.5), B (K2258-K2261+500 / K2261.5), and C (K2243-K2247) is more obvious. The absolute maximum values ​​of the cumulative settlement of the three areas from January 1, 2019 to October 31, 2023 are ranked from large to small as follows: B (-63.8mm) > A (-49.7mm) > C (-35.6mm), among which the settlement of area B is the most obvious. Through the integrated analysis of the "vehicle platform" and the "space platform", it is clear that area B is an "unstable target area", that is, the TQI value exceeding the limit section K2261+200 is located in the settlement target area B. Areas A, C and mileages K2250.6 and K2250.8 need further verification to determine whether they are "unstable target areas".

[0148] In order to further identify the “unstable target area”, drones based on “aerial platforms” were used to take aerial photos of the railway line to monitor its external environment. Through drone aerial photography along the railway line, it was found that there were many human engineering disturbances close to the existing line, such as Figure 5 (a) to (f). The results show that there are 6 engineering disturbances near the existing line, mainly distributed in areas A and B, while no obvious engineering disturbances were found in the vicinity of area C and the above-mentioned TQI value exceeding section K2250.6 and K2250.8. The railway mileages near the construction sites are shown in Table 1, and they are numbered a to f respectively. Among them, most of the construction disturbances are located near the settlement funnel and the historical disaster sites of the roadbed. According to the results of manual inspection, there are 8 historical collapse points in the railway roadbed, all located in areas A and B, namely Figure 4 The specific mileage information of the collapse points T1 to T8 shown in (a) is shown in Table 2. Therefore, through "verification", area A is also demarcated as an "unstable target area", while area C is a safe area. The results of regional stability judgment based on TQI value and multi-source data fusion are shown in Table 3.

[0149] Table 1 Construction disturbance adjacent mileage information

[0150]

[0151] Table 2 Collapse information

[0152]

[0153] Table 3 Results of the “unstable target area” discrimination process

[0154]

[0155] 1.2 Determination of the optimal noise reduction model

[0156] Time-series settlement data from settlement funnel points S1 to S6 and collapse points T1 to T3 within unstable target area A, acquired through multi-element collaborative monitoring, were input into the WT and GF models to complete noise reduction. Monitoring points T3, T2, S4, S6, S5, S2, S3, S1, and T1 were renumbered as monitoring points 1 to 9, respectively. Table 4 shows the accuracy evaluation results for the nine monitoring points. Analysis shows that the average signal-to-noise ratio (SNR) values, from high to low, are: WT (25.372) > GF (24.372); and the average root mean square error (RMSE) values, from high to low, are: GF (3.815 mm) > WT (1.576 mm), indicating that the WT model is the optimal noise reduction model.

[0157] Table 4 Comparison of prediction performance between WT model and GF model

[0158]

[0159] 1.3 Determination of the optimal time series prediction model

[0160] (1) Dataset construction and division

[0161] A data set was constructed based on the time-series sedimentation data after denoising by the WT model, and then the data set was divided into a training set and a test set in a ratio of 8:2 for subsequent prediction analysis.

[0162] (2) Optimal time series prediction model selection results

[0163] The prediction results were evaluated using three prediction accuracy and error evaluation indicators, as shown in Table 5. The average values ​​of the mean absolute error (MAE) from high to low are: Bayesian Ridge Regression BayesianRidge model (1.0667mm) > Support Vector Regression SVR model (0.8854mm) > Long Short-Term Memory Network LSTM model (0.5845mm); the average values ​​of the mean squared error (MSE) from high to low are: BayesianRidge (2.0409mm) > Support Vector Regression SVR model (0.8854mm) > Long Short-Term Memory Network LSTM model (0.5845mm). 2 )>SVR(1.3678mm 2 )>LSTM(0.6877mm 2 ); goodness of fit R 2 The average values ​​from high to low are: LSTM (0.9667) > SVR (0.9326) > BayesianRidge (0.9009), indicating that the LSTM model is the optimal time series prediction model.

[0164] Table 5 Comparison of prediction performance of time series prediction models

[0165]

[0166] 1.4 Determination of the optimal spatiotemporal prediction model

[0167] (1) Building a spatiotemporal prediction model

[0168] The GCN-LSTM model is constructed based on the LSTM model, which is the optimal time series prediction model, and the GCN model is integrated.

[0169] The input data of the GCN-LSTM model includes time feature data, i.e., time series sedimentation data, and spatial feature data, and the output data is time series sedimentation data.

[0170] (2) Construction of multiple second training sets and multiple second test sets

[0171] The temporal feature data in the second training set is the first training set constructed in step 1.3, and the spatial feature data in the second training set is composed of an adjacency matrix and a degree matrix corresponding to the distance threshold; the temporal feature data in the second test set is the first test set constructed in step 1.3, and the spatial feature data in the second test set is composed of an adjacency matrix and a degree matrix corresponding to the distance threshold. The second training set is used to train the spatiotemporal prediction model, and the second test set is used to test the spatiotemporal prediction model, and the second training set and the second test set with the same spatial feature data have a corresponding relationship.

[0172] In this embodiment, a total of 6 second training sets and 6 second test sets are constructed, corresponding to distance thresholds of 600 m, 800 m, 1000 m, 1200 m, 1400 m, and 1600 m, respectively.

[0173] The acquisition of spatial feature data is as follows:

[0174] Monitoring points 1 to 9 of unstable target area A are selected to construct spatial feature data, and are marked on the map to obtain the distance information between each two monitoring points, so as to construct their adjacency relationship based on the distance threshold. The distance interval between the monitoring points in target area A is [114.9m, 2632.5m]. According to the "Railway Roadbed Design Code" (TB10001-2016), the settlement monitoring interval along the railway should be 50m to 100m, so 600m to 1600m is selected as the distance threshold division interval. Based on this, 6 groups of experiments were designed and compared. The distance thresholds were selected as 600m, 800m, 1000m, 1200m, 1400m, and 1600m respectively. The graph structure is as follows: Figure 6 As shown in (a) to (f).

[0175] Taking the experimental group with a distance threshold of 600m as an example, Figure 6 (a) shows that monitoring points 4 and 5, monitoring points 3 and 6, monitoring points 6 and 7, and monitoring points 8 and 9 all have adjacency relationships. A 9×9 adjacency matrix A is constructed based on the distance between nodes, and then a degree matrix D is constructed based on the adjacency relationship. It is a 9×9 diagonal matrix. The values ​​on the diagonal represent the connection density of each node. The expressions of A and D are as follows:

[0176]

[0177] (3) Training of spatiotemporal models and selection of optimal spatiotemporal prediction models

[0178] First, the GCN-LSTM model is trained using the six second training sets. Then, the six second test sets are used to perform time series settlement prediction on monitoring points 1 to 9 of the trained GCN-LSTM model with corresponding relationships. The evaluation results are calculated based on the predicted settlement data and the measured settlement data. The evaluation results are shown in Figures 7 to 9. Figure 9 The prediction results were evaluated using three prediction accuracy and error evaluation indicators. The overall prediction accuracy evaluation results are shown in Table 6. The results show that the experimental group with a distance threshold of 800m has the best prediction effect. Compared with other experimental groups, its prediction error indicators MAE and MSE are the smallest, which are 0.4982 and 0.5771 respectively, and the prediction accuracy indicator R 2 The maximum value is 0.9994. At the same time, the prediction performance of the spatiotemporal prediction model GCN-LSTM model is better than that of the LSTM model, such as Figures 10 to 12 shown.

[0179] Table 6 Comparison of prediction effects of different distance thresholds

[0180]

[0181] It should be noted that the data of the prediction accuracy evaluation index corresponding to different distance thresholds in Table 6 are the arithmetic mean values ​​of the prediction accuracy evaluation indexes of the 9 monitoring points.

[0182] 1.5 Advance Forecast

[0183] Based on the last data pair in the dataset, the spatial data features corresponding to the optimal distance threshold of 800m are used to predict the time series deformation data of monitoring points 1 to 9 through the trained GCN-LSTM prediction model and the rolling prediction method, and the settlement prediction results for the next two years are obtained, as shown below: Figure 13 As shown, it should be noted that Figure 13 The three figures in the first row, from left to right, are the forecast results of monitoring points 1, 2, and 3 in the next two years. Figure 13The three figures in the second row, from left to right, are the forecast results of monitoring points 4, 5, and 6 in the next two years. Figure 13 The three figures in the second row, from left to right, are the prediction results of monitoring points 7, 8, and 9 in the next two years.

[0184] Figure 13 The results show that when the GCN-LSTM model predicts future settlement values ​​with an advance prediction scale of about one year, the surface settlement characteristics of monitoring points 3 to 8 within the prediction scale range are characterized by gradual and slow sinking and have not reached a stable state; monitoring point 1 shows fluctuating settlement; monitoring point 2 shows continuous and slow ground uplift and tends to be stable; the surface settlement change process of monitoring point 9 is rapid settlement first, then slow settlement, and finally tends to be stable.

[0185] It should be noted that the model training method used in this invention uses existing technology and does not improve upon it. Conditions for terminating model training iterations can include reaching the maximum number of training rounds, the loss decreasing by less than a threshold, and other user-defined conditions. These conditions can be used individually or in combination to ensure sufficient training without overfitting, while also balancing the efficient use of computing resources.

[0186] The above content is a further detailed description of the present invention in conjunction with specific preferred embodiments, and the specific implementation of the present invention should not be considered to be limited to these descriptions. For those skilled in the art of the present invention, several simple deductions and substitutions can be made without departing from the concept of the present invention, and all of these should be considered to fall within the scope of protection of the present invention.

Claims

1. A method for spatiotemporal prediction of settlement along a railway, characterized in that: The prediction method comprises the following steps: Step S10: identifying unstable target areas based on monitoring data acquired along the railway line; Step S20: acquiring time-series sedimentation data of each monitoring point in the unstable target area, and performing noise reduction processing using the optimal noise reduction model to obtain time-series sedimentation data of each monitoring point after noise reduction processing; Step S30: constructing an adjacency matrix based on a preset distance threshold and the distance between every two monitoring points in the unstable target area, and constructing a degree matrix based on the adjacency relationship of the adjacency matrix; Step S40: Based on the time series settlement data after noise reduction processing of each monitoring point, the adjacency matrix and the degree matrix, the GCN-LSTM prediction model is used to adopt a rolling prediction method to perform advance prediction on the time series settlement data of each monitoring point, and an advance prediction set for each monitoring point is obtained. The GCN-LSTM prediction model has a corresponding relationship with the preset distance threshold and is a trained model, wherein the advance prediction set includes multiple predicted settlement data that have a corresponding relationship with the prediction time.

2. The spatiotemporal prediction method for railway settlement according to claim 1, characterized in that: The monitoring data includes TQI values ​​collected based on vehicle platforms, settlement data obtained based on aerospace platforms, image data collected by drones based on aviation platforms, inspection data obtained by manual inspections based on ground-based platforms, and geophysical detection data.

3. The spatiotemporal prediction method for railway settlement according to claim 1, characterized in that: The step S10 includes: Using the track inspection vehicle on the vehicle platform to obtain the TQI value along the railway line, and determining the over-gauge section based on the preset TQI value, wherein the TQI value of the over-gauge section is greater than or equal to the preset TQI value; Based on the space platform, SBAS-InSAR technology is used to obtain time-series settlement data and annual average settlement rates along the railway line. Based on the time-series settlement data and annual average settlement rates, the railway line is divided into multiple settlement target areas and multiple non-sedimentation target areas; Combined with the over-limit section determined by the vehicle platform and the settlement target area and non-settlement target area determined by the aviation platform, the unstable target area and the target area to be verified are identified. If the over-limit section is located in the settlement target area, the settlement target area is identified as an unstable target area; the over-limit section in the non-settlement target area and the settlement target area that does not include the over-limit section are both identified as areas to be verified.

4. The spatiotemporal prediction method for railway settlement according to claim 3, characterized in that: The step S10 further includes: Acquiring image data of the area to be verified based on an aerial platform and obtaining inspection data of the area to be verified based on manual inspections on a ground-based platform, and further identifying the area to be verified based on the image data and the inspection data; when the area to be verified has both adjacent construction disturbances and collapse points, identifying the area to be verified as an unstable target area; and when the area to be verified has adjacent construction disturbances and no collapse points, or has collapse points and no adjacent construction disturbances, identifying the area to be verified as a further verification area; The further verification area is verified based on geophysical means.

5. The spatiotemporal prediction method for railway settlement according to claim 1, characterized in that: The optimal noise reduction model is a wavelet transform model. The method for obtaining the optimal noise reduction model in step S20 includes: Acquiring time-series settlement data of a plurality of target monitoring points, wherein the plurality of target monitoring points are all located within an unstable target area; Performing noise reduction processing on the time series settlement data of each target monitoring point using a wavelet transform model and a Gaussian filter model, respectively, to obtain a first noise reduction result corresponding to the wavelet transform model and a second noise reduction result corresponding to the Gaussian filter model, wherein the first noise reduction result and the second noise reduction result both include the noise-reduced time series settlement data of each target monitoring point; Calculate a noise reduction accuracy evaluation index based on the time-series settlement data of each target monitoring point, the first noise reduction result, and the second noise reduction result, and obtain a first noise reduction accuracy evaluation index corresponding to the wavelet transform model and a second noise reduction accuracy evaluation index corresponding to the Gaussian filter model, wherein the noise reduction accuracy evaluation index includes an average signal-to-noise ratio and an average root mean square error of multiple target monitoring points; The first noise reduction accuracy evaluation index and the second noise reduction accuracy evaluation index are compared to obtain a first comparison result, and the wavelet transform model is determined to be the optimal noise reduction model based on the first comparison result.

6. The spatiotemporal prediction method for railway settlement according to any one of claims 1 to 5, characterized in that: The preset distance threshold is a predetermined optimal distance threshold. The prediction method further includes a step of obtaining the optimal distance threshold and the GCN-LSTM prediction model, which is located after step S20 and before step S30, including: Step (1), based on the time series sedimentation data after noise reduction processing at each monitoring point, the input features and output features of the time series prediction are constructed using the sliding window method, and a time feature data set consisting of multiple groups of data pairs is obtained, each group of data pairs includes input features and output features with a preset time series length; Step (2), splitting the time feature data set into a first training set and a first test set; Step (3): firstly train the three time series prediction models based on the first training set, and then evaluate the three trained time series prediction models based on the first test set to obtain the optimal time series prediction model, wherein the three time series prediction models include a Bayesian Ridge regression model, a support vector regression model, and a long short-term memory network model, and the optimal time series prediction model is a long short-term memory network model; Step (4), based on the long short-term memory network model, the GCN model is integrated to construct a GCN-LSTM prediction model; Step (5), setting multiple distance thresholds, and constructing multiple spatial feature data sets corresponding to the multiple distance thresholds based on each distance threshold and the distance between every two monitoring points in the unstable target area, each spatial feature data set including an adjacency matrix and a degree matrix corresponding to the distance threshold; Step (6): Based on the temporal feature dataset and the plurality of spatial feature datasets, a plurality of second training sets and a plurality of second test sets are constructed, each of the second training sets including the first training set and the spatial feature dataset, and the spatial feature datasets in different second training sets are different; each second test set including the first test set and the spatial feature dataset, and the spatial feature datasets in different second test sets are different; and the second training sets and the second test sets including the same spatial feature dataset have a corresponding relationship; Step (7): First, use multiple second training sets to train the GCN-LSTM prediction model respectively, and then use multiple second test sets to evaluate the trained GCN-LSTM prediction model respectively to obtain the optimal distance threshold and the GCN-LSTM prediction model corresponding to the optimal distance threshold.

7. The spatiotemporal prediction method for railway settlement according to claim 6, characterized in that: The step (3) comprises: Inputting the first training set into a Bayesian Ridge regression model, a support vector regression model, and a long short-term memory network model for training, respectively, to obtain a trained Bayesian Ridge regression model, a support vector regression model, and a long short-term memory network model; Inputting the first test set into the trained Bayesian Ridge regression model, support vector regression model, and long short-term memory network model for prediction, respectively, to obtain three groups of first prediction results corresponding one-to-one to the three time series prediction models, each group of first prediction results including multiple first predicted settlement data corresponding one-to-one to multiple input features in the first test set; Based on the multiple output features in the first test set and the three groups of the first prediction results, respectively calculate the prediction accuracy evaluation index of each time series prediction model, and compare them to obtain a second comparison result, wherein the prediction accuracy evaluation index includes the average of the mean absolute error, the average of the mean square error, and the average of the goodness of fit of the multiple monitoring points; Based on the second comparison result, it is determined that the optimal time series prediction model is a long short-term memory network model.

8. The spatiotemporal prediction method for railway settlement according to claim 6, characterized in that: The step (7) comprises: Inputting the plurality of second training sets into the GCN-LSTM prediction model for training respectively, obtaining a plurality of trained GCN-LSTM prediction models, wherein the plurality of trained GCN-LSTM prediction models have a one-to-one correspondence with the plurality of distance thresholds; Inputting the plurality of second test sets into the trained GCN-LSTM prediction model with corresponding relationships for prediction, obtaining a plurality of groups of second prediction results corresponding one-to-one to the plurality of distance thresholds, each group of second prediction results including a plurality of second predicted settlement data corresponding one-to-one to the plurality of input features in the first test set; Based on the multiple output features in the first test set and the multiple groups of second prediction results, respectively calculating the prediction accuracy evaluation index of each trained GCN-LSTM prediction model, and comparing them to obtain a third comparison result, wherein the prediction accuracy evaluation index includes the average of the mean absolute error, the average of the mean square error, and the average of the goodness of fit of the multiple monitoring points; Based on the third comparison result, an optimal distance threshold and a GCN-LSTM prediction model corresponding to the optimal distance threshold are obtained.

9. The spatiotemporal prediction method for railway settlement according to claim 6, characterized in that: The preset time series length is N, and the time series length of the rolling prediction method is set according to the preset time series length. The steps of using the rolling prediction method to make any advance prediction for any one of the multiple monitoring points include: Step (a), obtaining target time series settlement data for the Mth settlement data prediction, wherein the target time series settlement data is composed of the latest N settlement data, and the N settlement data included in the target time series settlement data are measured settlement data and / or predicted settlement data; Step (b): using the target time series settlement data, the adjacency matrix and the degree matrix as input, and using the GCN-LSTM prediction model to obtain the predicted settlement data for the Mth prediction.

10. A device for temporal and spatial prediction of settlement along a railway, characterized in that: The prediction device comprises: An identification module for identifying unstable target areas based on monitoring data obtained from monitoring along the railway line; a noise reduction processing module, configured to obtain time-series sedimentation data of each monitoring point in the unstable target area, and perform noise reduction processing using an optimal noise reduction model to obtain time-series sedimentation data of each monitoring point after noise reduction processing; a spatial feature construction module, configured to construct an adjacency matrix based on a preset distance threshold and the distance between every two monitoring points in the unstable target area, and to construct a degree matrix based on the adjacency relationship of the adjacency matrix; The prediction module is used to use the GCN-LSTM prediction model to perform advance prediction on the time-series settlement data of each monitoring point based on the time-series settlement data after noise reduction processing of each monitoring point, the adjacency matrix and the degree matrix, and adopt a rolling prediction method to obtain an advance prediction set for each monitoring point. The GCN-LSTM prediction model has a corresponding relationship with a preset distance threshold and is a trained model, wherein the advance prediction set includes multiple predicted settlement data that have a corresponding relationship with the prediction time.

Images