High-speed rail line infrastructure state evaluation method based on air-overhead-vehicle-ground integration

Abstract

The invention discloses a high-speed rail line infrastructure state evaluation method based on air-overhead-vehicle-ground integration, and belongs to the technical field of high-speed rail line infrastructure state evaluation. The method comprises the following steps: acquiring a multi-source data set; converting data in the multi-source data set into a plurality of structured evaluation indexes to obtain an evaluation index set; performing time sequence alignment, mileage alignment and space alignment on the plurality of structured evaluation indexes in the evaluation index set to obtain a plurality of evaluation vectors; processing the plurality of evaluation vectors by using a state evaluation model to obtain a prediction result of the road, bridge, tunnel and rail, the prediction result including a degradation state prediction result and a residual life prediction result; determining a real-time degradation state and a real-time remaining life based on the real-time monitoring data; and updating parameters of the state evaluation model based on differences among the real-time degradation state, the real-time residual life and the prediction result. The method can accurately evaluate the state of the road, bridge and tunnel rails.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the technical field of high-speed rail line infrastructure state evaluation, in particular to a high-speed rail line infrastructure state evaluation method based on space-air-ground integration. BACKGROUND

[0002] With the continuous expansion of China's high-speed rail network, the long-term service performance evaluation of infrastructure is facing major challenges.

[0003] The current high-speed rail line includes tracks, bridges, tunnels and other structures, and its state is affected by multiple factors such as geological environment and load action. Traditional manual inspection and single-point detection cannot comprehensively consider the influence of multiple factors on high-speed rail line infrastructure, resulting in inaccurate evaluation results of high-speed rail line infrastructure state. SUMMARY

[0004] The present disclosure provides a high-speed rail line infrastructure state evaluation method based on space-air-ground integration, which can accurately evaluate the state of tracks, bridges, tunnels and tracks. The technical solution at least includes the following solutions: In a first aspect, a high-speed rail line infrastructure state evaluation method based on space-air-ground integration is provided, comprising: obtaining a multi-source data set, the multi-source data set including data collected by multi-source sensors for monitoring tracks, bridges, tunnels and tracks; converting the data in the multi-source data set into a plurality of structured evaluation indicators to obtain an evaluation indicator set, the structured evaluation indicators being used for state evaluation of tracks, bridges, tunnels and tracks; after time alignment, mileage alignment and spatial alignment of the plurality of structured evaluation indicators in the evaluation indicator set, a plurality of evaluation vectors are obtained; using a state evaluation model to process the plurality of evaluation vectors to obtain a prediction result of tracks, bridges, tunnels and tracks, the prediction result including a degradation state prediction result and a remaining life prediction result; determining a real-time degradation state and a real-time remaining life based on real-time monitoring data; updating the parameters of the state evaluation model based on the difference between the real-time degradation state, the real-time remaining life and the prediction result.

[0005] Optionally, the state evaluation model comprises a convolutional neural network (CNN) connected in sequence, a long short-term memory (LSTM) network, a cross-modal feature interaction layer, and a multi-task output layer, the CNN is used to extract spatial features, the LSTM is used to extract time sequence features, the cross-modal feature interaction layer is used to fuse the spatial features and the time sequence features, and the multi-task output layer is used to output the prediction result of the road, bridge, tunnel and rail, and the parameters of the state evaluation model are updated based on the difference between the real-time degradation state, the real-time remaining life and the prediction result, comprising: in the case that the difference between the real-time degradation state, the real-time remaining life and the prediction result of the roadbed satisfies a roadbed difference threshold, updating the forgetting gate weight matrix parameters of the LSTM; in the case that the difference between the real-time degradation state, the real-time remaining life and the prediction result of the bridge satisfies a bridge difference threshold, updating the convolution kernel parameters of the CNN; in the case that the difference between the real-time degradation state, the real-time remaining life and the prediction result of the tunnel satisfies a tunnel difference threshold, updating the multi-head attention weight of the cross-modal feature interaction layer; and in the case that the difference between the real-time degradation state, the real-time remaining life and the prediction result of the rail satisfies a rail difference threshold, updating the fully connected layer weight of the multi-task output layer.

[0006] Optionally, the updating of the forgetting gate weight matrix parameters of the LSTM comprises re-weighting the forgetting gate weight matrix parameters of the LSTM based on a time sequence attention mechanism, the updating of the convolution kernel parameters of the CNN comprises updating the convolution kernel parameters of the CNN based on a gradient clipping update manner, the updating of the multi-head attention weight of the cross-modal feature interaction layer comprises updating the multi-head attention weight of the cross-modal feature interaction layer using a federated learning collaborative fine-tuning algorithm, and the updating of the fully connected layer weight of the multi-task output layer comprises updating the fully connected layer weight of the multi-task output layer using an adaptive momentum optimization algorithm.

[0007] Optionally, the time series alignment of the plurality of structured evaluation indexes in the evaluation index set comprises: achieving the time series alignment of the plurality of structured evaluation indexes in the evaluation index set by using a dynamic time warping (DTW) algorithm to obtain a set of alignment paths, the set of alignment paths comprising a plurality of time alignment paths; and achieving data synchronization by using linear interpolation for any index pair in a first time alignment path, the first time alignment path being any time alignment path in the set of alignment paths. The mileage alignment of the plurality of structured evaluation indexes in the evaluation index set comprises: setting a plurality of mileage points on a route, and achieving the mileage alignment by using a piecewise cubic spline function for adjacent mileage points. The spatial alignment of the plurality of structured evaluation indexes in the evaluation index set comprises: converting coordinates of the plurality of structured evaluation indexes to a same spatial coordinate system by using a seven-parameter Helmert transformation.

[0008] Optionally, the first time point is the first mileage point.

[0009] wherein, is the evaluation vector corresponding to the first time point, is the evaluation vector corresponding to the first mileage point, is a coordinate of the first mileage point after spatial alignment, is an evaluation index of the first mileage point at the first time point, is the evaluation index of the first mileage point at the first time point, are positive integers, has a value range of 1 to , is a total number of mileage points, has a value range of 1 to , is a time series length, has a value range of 1 to , is a total number of evaluation indexes.

[0010] Optionally, the CNN is a three-dimensional CNN, and the CNN extracts the spatial features of the evaluation vector in the following manner:

[0011] wherein, is a spatial feature of . is an activation function, , is a time neighborhood offset, is a space neighborhood offset, denotes a case where the time neighborhood offset is , and the space neighborhood offset is , is a convolution kernel weight of an th dimension of is a convolution kernel weight matrix of the CNN, , is an element-wise multiplication, is a bias of the CNN.

[0012] The second aspect also provides a space-air-ground integrated high-speed rail line infrastructure state evaluation device, including: an acquisition module configured to acquire a multi-source data set, the multi-source data set including data collected by multi-source sensors for monitoring a road-bridge-tunnel-rail; a conversion module configured to convert data in the multi-source data set into a plurality of structured evaluation indexes to obtain an evaluation index set, the structured evaluation indexes being used for state evaluation of the road-bridge-tunnel-rail; an alignment module configured to obtain a plurality of evaluation vectors after time sequence alignment, mileage alignment, and spatial alignment of the plurality of structured evaluation indexes in the evaluation index set; a prediction module configured to process the plurality of evaluation vectors by using a state evaluation model to obtain a prediction result of the road-bridge-tunnel-rail, the prediction result including a degradation state prediction result and a remaining life prediction result; a real-time monitoring module configured to determine a real-time degradation state and a real-time remaining life based on real-time monitoring data; and an update module configured to update parameters of the state evaluation model based on a difference between the real-time degradation state, the real-time remaining life, and the prediction result.

[0013] Optionally, the state evaluation model comprises a convolutional neural network (CNN) connected in sequence, a long short-term memory (LSTM) network, a cross-modal feature interaction layer, and a multi-task output layer, the CNN is configured to extract spatial features, the LSTM is configured to extract time sequence features, the cross-modal feature interaction layer is configured to fuse the spatial features and the time sequence features, and the multi-task output layer is configured to output the prediction result of the road, bridge, tunnel, and rail; the update module is further configured to update a forgetting gate weight matrix parameter of the LSTM when a difference between the real-time degradation state, the real-time remaining life of the embankment, and the prediction result meets an embankment difference threshold; update a convolution kernel parameter of the CNN when a difference between the real-time degradation state, the real-time remaining life of the bridge, and the prediction result meets a bridge difference threshold; update a multi-head attention weight of the cross-modal feature interaction layer when a difference between the real-time degradation state, the real-time remaining life of the tunnel, and the prediction result meets a tunnel difference threshold; and update a fully connected layer weight of the multi-task output layer when a difference between the real-time degradation state, the real-time remaining life of the rail, and the prediction result meets a rail difference threshold.

[0014] Optionally, the update module is further configured to reweight the forgetting gate weight matrix parameter of the LSTM based on a time sequence attention mechanism; update the convolution kernel parameter of the CNN based on a gradient clipping update manner; update the multi-head attention weight of the cross-modal feature interaction layer using a federated learning collaborative fine-tuning algorithm; and update the fully connected layer weight of the multi-task output layer using an adaptive momentum optimization algorithm.

[0015] Optionally, the alignment module is further configured to perform time sequence alignment on a plurality of structured evaluation indexes in the evaluation index set using a dynamic time warping (DTW) algorithm to obtain an aligned path set, the aligned path set comprising a plurality of time-aligned paths; perform data synchronization on any index pair in a first time-aligned path using linear interpolation, the first time-aligned path being any time-aligned path in the aligned path set; set a plurality of mileage points on the road, and perform mileage alignment on adjacent mileage points using a piecewise cubic spline function; and convert coordinates of the plurality of structured evaluation indexes to the same spatial coordinate system using a seven-parameter Helmert transformation.

[0016] In a third aspect, a computer device is also provided, comprising a memory and a processor, at least one computer program is stored in the memory, the at least one computer program is loaded and executed by the processor, thereby executing the above-mentioned space-air-ground integrated high-speed rail line infrastructure state evaluation method.

[0017] In a fourth aspect, a computer-readable storage medium is provided, and the computer-readable storage medium stores at least one computer program. The at least one computer program is loaded and executed by a processor, so as to implement the space-air-ground integrated high-speed rail line infrastructure state evaluation method described in the above embodiments.

[0018] In a fifth aspect, a computer program product is provided, and the computer program product includes computer programs / instructions. When the computer programs / instructions are executed by a processor, the method of the first aspect is implemented.

[0019] The technical solutions provided by the embodiments of the present disclosure have at least the following beneficial effects: In the embodiments of the present disclosure, by obtaining a multi-source data set, converting data in the multi-source data set into a plurality of structured evaluation indexes to obtain an evaluation index set, after time sequence alignment, mileage alignment and spatial alignment of the plurality of structured evaluation indexes in the evaluation index set, a plurality of evaluation vectors are obtained, the plurality of evaluation vectors are processed by using a state evaluation model to obtain a prediction result of a road-bridge-tunnel-rail, the prediction result includes a degradation state prediction result and a residual life prediction result, a real-time degradation state and a real-time residual life are determined based on real-time monitoring data, and parameters of the state evaluation model are updated based on differences between the real-time degradation state, the real-time residual life and the prediction result, thereby realizing a state dynamic evaluation and model closed-loop optimization mechanism of high-speed rail infrastructure. The state evaluation model realizes feature extraction and fusion analysis of multi-source data, and establishes an end-to-end mapping relationship from original data to degradation state and residual life. Based on the differences between the real-time degradation state, the real-time residual life and the prediction result, the parameters of the state evaluation model are updated, which is equivalent to using the differences between the real-time degradation state, the real-time residual life and the prediction result to perform closed-loop feedback on the parameters of the state evaluation model, thereby realizing continuous optimization of model parameters based on real-time monitoring data, and effectively improving the accuracy of the prediction result output by the state evaluation model. BRIEF DESCRIPTION OF DRAWINGS

[0020] In order to more clearly illustrate the technical solutions in the embodiments of the present disclosure, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present disclosure, and other drawings can be obtained by those skilled in the art without creative labor.

[0021] Figure 1 A flowchart of a space-air-ground integrated high-speed rail line infrastructure state evaluation method provided by an example embodiment of the present disclosure is shown; Figure 2A flow chart of a high-speed rail line infrastructure state evaluation method based on space-air-vehicle-ground integration is shown according to an example embodiment of the present disclosure. Figure 3 A structural schematic diagram of a high-speed rail line infrastructure state evaluation device based on space-air-vehicle-ground integration is shown according to an example embodiment of the present disclosure. Figure 4 A structural schematic diagram of a computer device according to an example embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0022] Unless otherwise defined, technical terms or scientific terms used herein should be interpreted as is normally understood by one of ordinary skill in the art to which the present disclosure pertains. The use of terms such as "first", "second", "third" and the like in the present patent application and claims does not imply any order, quantity, or importance, but is merely used to distinguish different components. Similarly, the use of terms such as "one" or "a" or "an" does not limit the quantity of referenced elements to one. The use of terms such as "including", "containing", or "comprising" and the like does not exclude the presence of other elements or items. The use of terms such as "connected" or "coupled" or the like does not necessarily denote a direct or indirect physical or mechanical connection or linkage, but can also include an electrical connection or linkage.

[0023] In order to make the purpose, technical solutions and advantages of the present disclosure clearer, the following will further describe the embodiments of the present disclosure in conjunction with the drawings.

[0024] Figure 1 A flow chart of a high-speed rail line infrastructure state evaluation method based on space-air-vehicle-ground integration is shown according to an example embodiment of the present disclosure, which can be executed by a computer device. Referring to FIG. 1, the method comprises the following steps. Figure 1 The method comprises the following steps. In step 101, a multi-source data set is acquired.

[0025] The multi-source data set comprises data collected by multi-source sensors for monitoring road, bridge, tunnel and track.

[0026] The road, bridge, tunnel and track include track, bridge, tunnel and subgrade of high-speed rail.

[0027] The subgrade is a foundation structure of the high-speed rail line, including embankment, cutting, slope and other soil structures. The bridge is an overhead structure that crosses obstacles, including piers, beam bodies, supports and the like. The tunnel is a structure that passes through underground or mountains, including lining, inverted arches and drainage systems. The track is a driving load-bearing system composed of steel rails, fasteners and ballast beds. The road, bridge, tunnel and track mentioned herein are all related to high-speed rail.

[0028] Rails are a subset of tracks and are the core load-bearing components of tracks; bridge piers are a subset of bridges and are the vertical support structures of bridges. Track monitoring includes but is not limited to rails (it also includes fasteners, track bed, etc.), while bridge monitoring includes but is not limited to bridge piers (it also includes beams, supports, etc.).

[0029] According to the classification of roads, bridges, tunnels, and tracks, the multi-source data set can include four types of data: roadbed inspection data, track inspection data, bridge inspection data, and tunnel inspection data.

[0030] Roadbed inspection data can be acquired using spaceborne multispectral imaging, distributed optical fiber, and other equipment. For multispectral images acquired by spaceborne multispectral imaging, the slope moisture content of the roadbed can be retrieved using NDVI (Normalized Difference Vegetation Index) (spatial resolution 2m). For distributed optical fiber, the roadbed strain distribution can be monitored based on OTDR (Optical Time-Domain Reflectometer) (spatial positioning accuracy 0.5m). Preprocessing of the roadbed inspection data includes atmospheric correction of multispectral images and fiber optic strain moving average.

[0031] Bridge inspection data can be acquired using equipment such as spaceborne InSAR (Interferometric Synthetic Aperture Radar), airborne LiDAR (Light Detection and Ranging), and GNSS (Global Navigation Satellite System) monitoring stations. For point cloud data acquired by spaceborne InSAR, differential interferometry can be used to extract pier settlement sequence data (accuracy ±2mm). For point cloud data acquired by airborne LiDAR, the Time-of-Flight (ToF) method can be used to construct a 3D point cloud of the bridge surface, automatically marking the location and width of cracks on the bridge surface. For satellite signals acquired by GNSS monitoring stations, carrier phase calculation can be used to achieve dynamic displacement monitoring of the bridge (horizontal accuracy 1mm, vertical accuracy 2mm). Preprocessing of bridge inspection data includes InSAR atmospheric delay correction and LiDAR point cloud radius filtering.

[0032] The tunnel detection data can be collected by a UAV infrared thermal imager, a vehicle-mounted laser scanner and the like. For the infrared image data collected by the UAV infrared thermal imager, the lining void area of the tunnel can be identified based on the radiation thermometry principle (temperature difference threshold 2℃); for the point cloud data collected by the vehicle-mounted laser scanner, the tunnel boundary deformation can be detected by the laser triangulation method (axial resolution 5mm). The preprocessing of the tunnel detection data includes infrared thermal image non-uniformity correction, laser radar intensity normalization and the like.

[0033] The track detection data can be collected by a vehicle-mounted rail ultrasonic flaw detector, a track geometry detection system, a trackside optical fiber sensor and the like. For the ultrasonic data collected by the vehicle-mounted rail ultrasonic flaw detector, the internal defects of the rail head / rail waist can be detected by the pulse reflection method to generate the damage level (0-4 levels) and position coordinates. For the data collected by the track geometry detection system, the deformation parameters (millimeter level precision) such as track gauge, levelness, height and the like can be output based on the inertial reference method and optical distance measurement; for the sensor data collected by the trackside optical fiber sensor, the rail longitudinal stress change curve (sampling frequency 1kHz) can be analyzed by using the Brillouin scattering principle. The preprocessing of the track detection data includes wavelet packet denoising for rail ultrasonic waves, Kalman filtering for track geometry data and the like.

[0034] After the preprocessing of the above four types of data, the signal-to-noise ratio and the spatial resolution of the above four types of data can be unified (such as signal-to-noise ratio , spatial resolution 0.05m), which lays a data foundation for subsequent fusion analysis. The multi-source sensors involved in collecting the above four types of data include satellites, UAVs, trains and the ground, realizing the integrated multi-source data collection of high-speed rail line infrastructure in space, air, train and ground, so that the multi-source data set can be obtained.

[0035] The multi-source data set in the embodiment of the present disclosure is the historical multi-source data of road, bridge, tunnel and track, which can be used to train the subsequent state evaluation model.

[0036] In step 102, the data in the multi-source data set is converted into a plurality of structured evaluation indexes to obtain an evaluation index set.

[0037] The structured evaluation index is used for state evaluation of road, bridge, tunnel and track.

[0038] In the embodiment of the present disclosure, the classification is performed according to road, bridge, tunnel and track, so that the structured evaluation index includes four types of structured evaluation indexes corresponding to roadbed, bridge, tunnel and track. The following will illustrate examples of the four types of structured evaluation indexes.

[0039] (1) The structured evaluation indexes corresponding to the roadbed include: settlement acceleration, slope stability coefficient.

[0040] (2) The structured evaluation indexes corresponding to the bridge include: support displacement early warning value, crack propagation rate.

[0041] (3) The structured evaluation indexes corresponding to the tunnel include: lining defect density, convergence deformation amount.

[0042] (4) The structured evaluation indexes corresponding to the track include: rail damage index, track quality index (TQI).

[0043] After the multi-source data set is obtained, any manner in the related art can be used to convert the multi-source data set into the structured evaluation indexes described above.

[0044] In step 103, after the plurality of structured evaluation indexes in the evaluation index set are time-sequentially aligned, milepost-aligned, and space-aligned, a plurality of evaluation vectors are obtained.

[0045] Optionally, time-sequentially aligning the plurality of structured evaluation indexes in the evaluation index set includes the following two steps.

[0046] First, a dynamic time warping algorithm is used to time-sequentially align the plurality of structured evaluation indexes in the evaluation index set, to obtain a set of alignment paths.

[0047] The set of alignment paths includes a plurality of time alignment paths.

[0048] The dynamic time warping (DTW) algorithm aligns time series by calculating an optimal matching path between the two time series, thereby obtaining an aligned time path. Data from different sensors or monitoring devices belong to different time series in the time dimension. After the multi-source data is converted into evaluation indexes, the data at each time point can be converted into evaluation indexes, that is, the evaluation indexes converted from the multi-source data from different sensors also belong to different time series in the time dimension. Therefore, the plurality of structured evaluation indexes in the multi-source evaluation index set can be time-sequentially aligned based on the DTW algorithm.

[0049] After time-sequentially aligning based on the DTW, a plurality of alignment paths are obtained, each alignment path including a plurality of index pairs, and an index pair including data from two different time series. An index pair is actually a time point obtained by aligning data from two different time series.

[0050] Second, linear interpolation is used to synchronize the data in any index pair in the first time alignment path.

[0051] The first time alignment path is any time alignment path in the alignment path set.

[0052] Exemplarily, the linear interpolation is implemented by using formula (1).

[0053] (1) In formula (1), is an index pair in the first time alignment path, , is data from different time series, , is an interpolation weight coefficient, , is an interpolated index pair, that is, time-aligned data.

[0054] The time path obtained by the DTW algorithm is actually composed of multiple time points, and there is a certain interval between the time points. Thus, multiple time points do not obtain a continuous time curve. Linear interpolation is performed on the time points in the time path, so that the time path becomes a continuous and smooth time curve, thereby realizing time series alignment.

[0055] Optionally, the multiple structured evaluation indexes in the evaluation index set are subjected to mile alignment, including: multiple mile points are set on the line, and piecewise cubic spline function is used to realize mile alignment on adjacent mile points.

[0056] Exemplarily, a mile interval value can be preset, and multiple mile points are set on the line based on the mile interval value. Here, since the embodiments of the present disclosure are for state evaluation of high-speed rail line infrastructure, the line here is the high-speed rail line to be studied. Exemplarily, the mile interval value can be 500 meters, for example, a mile point can be set every 500m along the line In addition, the GNSS coordinates of each mile point, the mile value of each mile point on the line (for example, the mile distance from the starting point of the line to the mile point) and the multiple evaluation indexes included in the mile point need to be recorded.

[0057] In implementation, a linear reference system (Linear Referencing System, LRS) can be used to realize data coding and management based on mile points. Each mile point can be coded as: line number+mile point serial number+structure type code.

[0058] The structure type code is used to indicate which structure in the road-bridge-tunnel-track the mile point belongs to. For example, the structure type code is: 01 (road), 02 (bridge), 03 (tunnel), and 04 (track).

[0059] Exemplarily, the bridge at the 500th mileage point of the K123 line of the Beijing-Shanghai high-speed railway can be coded as JHG123-500-02, where JHG123 represents the K123 line of the Beijing-Shanghai high-speed railway, 500 represents the 500th mileage point, and 02 represents the bridge.

[0060] The mileage point can be abstractly understood as a spatial position, which can include any structure of a road, a bridge, a tunnel, and a track, or multiple structures (for example, a certain mileage point includes a tunnel and a track in the tunnel). In this case, when coding the mileage point, the value of the structure type coding can be increased. For example, the 500th mileage point of the K123 line of the Beijing-Shanghai high-speed railway includes a tunnel and a track, and the spatial position at this point can be coded as JHG123-500-03-04.

[0061] After multiple mileage points are set, the data collected by each sensor or monitoring device can be inserted into the mileage point, and the data collected by the sensors or monitoring devices at the same mileage point is equivalent to achieving mileage alignment; further, the evaluation indexes calculated from these mileage-aligned data can also achieve mileage alignment.

[0062] However, only the mileage alignment at the mileage point is not enough, and such mileage alignment is equivalent to intermittent and discontinuous alignment. Therefore, interpolation processing needs to be performed between adjacent two mileage points, multiple interpolation mileage points are inserted between the adjacent two mileage points, and continuous mileage alignment is achieved through the interpolation mileage points. In the embodiments of the present disclosure, a piecewise cubic spline function is used to insert multiple interpolation mileage points between adjacent two mileage points to achieve continuous mileage alignment. This process can be represented by formula (2).

[0063] (2) In formula (2), is a piecewise cubic spline function used for interpolation in the interval between adjacent two mileage points , and is a spline coefficient, which can be solved by a boundary condition.

[0064] Exemplarily, the boundary condition is represented by formula (3).

[0065] (3) In formula (3), is the mileage point corresponding to multiple evaluation indexes, is the mileage point corresponding to the evaluation index, is a piecewise cubic spline function used for interpolation in the interval between adjacent two mileage points Piecewise cubic spline function for interpolation within an interval This demonstrates that the first derivative of the piecewise cubic spline function is continuous. This demonstrates the continuity of the second derivative of the piecewise cubic spline function. The meanings of the other parameters in formula (3) are the same as those in formula (2), and are omitted here for further explanation.

[0066] Optionally, multiple structured evaluation indicators in the evaluation indicator set can be spatially aligned, including: using a seven-parameter Helmert transform to convert the coordinates of multiple structured evaluation indicators to the same spatial coordinate system, such as the Earth-Centered Earth-Fixed (ECEF) coordinate system.

[0067] The coordinates of the evaluation index are the coordinates of the sensor or monitoring device that calculates the evaluation index. Different sensors or monitoring devices may output coordinate formats that differ. Through the seven-parameter Helmholtz transformation, the coordinates of multiple structured evaluation indices can be transformed to the same spatial coordinate system.

[0068] There are many implementation methods for the seven-parameter Helmert transform, which will not be detailed here.

[0069] After time-series alignment, mileage alignment, and spatial alignment, evaluation metrics for the same mileage point at the same time belong to the same evaluation vector. In this way, multiple evaluation vectors can be obtained.

[0070] For example, the first Time of the first The evaluation vector corresponding to the mileage point can be represented by formula (4).

[0071] (4) in, For the first Time of the first The evaluation vector corresponding to the mileage point For spatial alignment, the first Time of the first The coordinates of the mileage point For the first Time of the first Mileage point One evaluation indicator, , , All are positive integers. The value range is 1 to , The total number of mileage points. The value range is 1 to , The length of the time series. the value range of the evaluation index is 1 to , is the total number of evaluation indexes.

[0072] All evaluation vectors can constitute an input tensor of the state evaluation model, which is expressed by formula (5).

[0073] (5) In formula (5), is the input tensor of the state evaluation model, and the meanings of other parameters in formula (5) are the same as those of the parameters in formula (4), and details are omitted here.

[0074] In the embodiments of the present disclosure, a three-dimensional monitoring system of space-air-ground-vehicle integration is constructed, and through the collaborative configuration of multi-platform and multi-scale sensors, the state perception of all elements and all-weather of high-speed rail infrastructure is realized. Through the establishment of a unified data preprocessing process, the standardized conversion and quality improvement of multi-source heterogeneous data are realized, providing a reliable data foundation for subsequent analysis.

[0075] In step 104, the state evaluation model is used to process the plurality of evaluation vectors to obtain the prediction result of the road-bridge-tunnel-railway.

[0076] The prediction result includes a degradation state prediction result and a residual life prediction result.

[0077] Optionally, the state evaluation model includes a convolutional neural network (CNN) connected in sequence, a long short-term memory network (LSTM), a cross-modal feature interaction layer, and a multi-task output layer. The CNN is used to extract spatial features, the LSTM is used to extract time sequence features, the cross-modal feature interaction layer is used to fuse spatial features and time sequence features, and the multi-task output layer is used to output the prediction result of the road-bridge-tunnel-railway.

[0078] Optionally, the CNN is a three-dimensional CNN. In this case, the CNN uses the following formula (6) to extract the spatial features of the evaluation vector.

[0079] (6) In formula (6), is the spatial feature of , is an activation function, , is a time neighborhood offset, is a spatial neighborhood offset, indicates that the time neighborhood offset is , the spatial neighborhood offset is , the spatial neighborhood offset is , the first dimensional convolution kernel weight, is a convolution kernel weight matrix of the CNN, , is an element-wise multiplication, is a bias of the CNN.

[0080] Based on the LSTM, the time sequence feature extraction and the degradation process modeling can be realized.

[0081] In the LSTM, the forgetting gate controls the memory strength of the historical state, thereby realizing the retention of long-term degradation trends (such as steel rail fatigue accumulation); the input gate adjusts the weight of the current feature, effectively capturing short-term mutations (such as sudden track geometric deformation); and the output gate determines the output proportion of the hidden state, realizing dynamic focusing of the feature.

[0082] The memory unit of the LSTM fuses historical and current information through the gating mechanism to form a time sequence degradation track of the infrastructure, so as to realize the degradation process modeling. For example: progressive change of bridge support displacement, expansion rate of tunnel lining crack, etc.

[0083] The cross-modal feature interaction layer realizes the fusion of spatial features and time sequence features through a multi-head attention mechanism to obtain fusion features, which are input into the multi-task output layer to obtain the prediction results of the road, bridge, tunnel and track, wherein the prediction results include the degradation state prediction result and the remaining life prediction result.

[0084] Exemplarily, the degradation state includes 5 levels (more levels can also be set according to requirements), and the multi-task output layer can output the probability distribution of the road, bridge, tunnel and track in each degradation state to realize the output of the degradation state prediction result. In addition, the degradation state prediction result contains the degradation state of the road, bridge, tunnel and track at each time in the future period of time.

[0085] Similarly, the remaining life prediction result contains the remaining life of the road, bridge, tunnel and track at each time in the future period of time.

[0086] In the prediction result output by the state prediction model, each mileage point outputs a four-element prediction result of the road, bridge, tunnel and track. For the structure contained in the mileage point, the corresponding part of the four-element prediction result outputs the corresponding prediction result, and for the structure not contained in the mileage point, the corresponding part of the four-element prediction result outputs 0. For example, if a mileage point includes a tunnel and a track, but does not include a roadbed and a bridge, the roadbed and bridge part of the road, bridge, tunnel and track four-element prediction result of the mileage point is 0.

[0087] For different degradation states and degradation life of road, bridge, tunnel and rail, some maintenance suggestions can be preset, which can be output at the same time as the prediction result. For example, when the state prediction model predicts that the degradation state of the roadbed is at a certain level, the maintenance suggestion corresponding to the level can be output.

[0088] In step 105, the real-time degradation state and the real-time remaining life are determined based on the real-time monitoring data.

[0089] Here, the real-time monitoring data is obtained in the same way as the multi-source data set, the difference is that the data in the multi-source data set is historical data, and the real-time monitoring data is real-time collected data. Similarly, the real-time monitoring data also needs to be processed in the way of steps 101 to 103 before it can be used to calculate the real-time degradation state and the real-time remaining life.

[0090] After obtaining a plurality of evaluation vectors based on real-time monitoring data, the real-time degradation state and the real-time remaining life of the road, bridge, tunnel and rail can be determined by a preset threshold. However, this method can only determine the degradation state and the remaining life corresponding to the current data or the historical data, and cannot predict the degradation state and the remaining life of the road, bridge, tunnel and rail in the future, for example, the degradation state and the remaining life of the road, bridge, tunnel and rail in the future. Therefore, in the embodiment of the present disclosure, the prediction of the degradation state and the remaining life of the road, bridge, tunnel and rail in the future is realized by steps 101 to 104.

[0091] In step 106, the parameters of the state evaluation model are updated based on the difference between the real-time degradation state, the real-time remaining life and the prediction result.

[0092] In the embodiment of the present disclosure, through steps 101 to 106, the state dynamic evaluation and model closed-loop optimization mechanism of high-speed rail infrastructure are realized. Among them, the feature extraction and fusion analysis of multi-source data are realized by the state evaluation model, and the end-to-end mapping relationship from the original data to the degradation state and the remaining life is established. Based on the difference between the real-time degradation state, the real-time remaining life and the prediction result, the parameters of the state evaluation model are updated, which is equivalent to using the difference between the real-time degradation state, the real-time remaining life and the prediction result to close-loop feedback the parameters of the state evaluation model, realizing continuous optimization of model parameters based on real-time monitoring data, and effectively improving the accuracy of the prediction result output by the state evaluation model.

[0093] Figure 2 A flowchart of a high-speed rail line infrastructure state evaluation method based on space-air-ground integration provided by another example embodiment of the present disclosure is shown, which can be executed by a computer device. Referring to Figure 1 , the method comprises: In step 201, a multi-source data set is obtained.

[0094] The multi-source data set comprises data collected by multi-source sensors for monitoring the road, bridge, tunnel and rail; In step 202, the data in the multi-source data set is converted into a plurality of structured evaluation indexes to obtain an evaluation index set.

[0095] The structured evaluation indexes are used for state evaluation of the road, bridge, tunnel and rail.

[0096] In step 203, after the plurality of structured evaluation indexes in the evaluation index set are time-aligned, mile-aligned and space-aligned, a plurality of evaluation vectors are obtained.

[0097] In step 204, the plurality of evaluation vectors are processed by using a state evaluation model to obtain a prediction result of the road, bridge, tunnel and rail.

[0098] The prediction result comprises a degradation state prediction result and a residual life prediction result.

[0099] In step 205, real-time degradation state and real-time residual life are determined based on real-time monitoring data.

[0100] The related content of steps 201 to 205 is described in the foregoing steps 101 to 105, and detailed description is omitted here.

[0101] In step 206, parameters of the state evaluation model are updated based on the difference between the real-time degradation state, the real-time residual life and the prediction result.

[0102] Optionally, step 206 comprises the following four cases.

[0103] (1) In the case that the difference between the real-time degradation state, the real-time residual life and the prediction result of the subgrade satisfies a subgrade difference threshold, the forgetting gate weight matrix parameter of the LSTM is updated.

[0104] In implementation, the forgetting gate weight matrix parameter of the LSTM can be reweighted based on a time sequence attention mechanism to update the forgetting gate weight matrix parameter of the LSTM.

[0105] The degradation or life of the subgrade is greatly affected by long-term deformation. In the case that the difference between the real-time degradation state, the real-time residual life and the prediction result of the subgrade satisfies a subgrade difference threshold, it is indicated that the model parameters related to the long-term degradation trend in the state evaluation model need to be adjusted, that is, the LSTM forgetting gate weight matrix parameter needs to be adjusted. By updating the forgetting gate weight matrix parameter of the LSTM, the sensitivity of slow degradation trends such as settlement and slope slip can be enhanced.

[0106] (2) In the case that the difference between the real-time degradation state, the real-time residual life and the prediction result of the bridge satisfies a bridge difference threshold, the convolution kernel parameter of the CNN is updated.

[0107] In implementation, the convolution kernel parameters of the CNN can be updated based on the gradient clipping update manner.

[0108] The degradation or service life of the bridge is greatly affected by local damage of the bridge, which can be reflected in spatial features. If the difference between the real-time degradation state and the real-time remaining life of the bridge and the prediction result satisfies the bridge difference threshold, it indicates that the related parameters of the spatial features in the state evaluation model need to be adjusted. By updating the convolution kernel parameters of the CNN, the extraction capability of the state evaluation model for spatial features can be adjusted, and the identification accuracy of local damage such as cracks and support displacement can be improved.

[0109] (3) In the case where the difference between the real-time degradation state and the real-time remaining life of the tunnel and the prediction result satisfies the tunnel difference threshold, the multi-head attention weight of the cross-modal feature interaction layer is updated.

[0110] In implementation, the multi-head attention weight of the cross-modal feature interaction layer can be updated by using a federated learning collaborative fine-tuning algorithm (such as the FedProx algorithm).

[0111] The degradation or service life of the tunnel needs to consider multiple parameters such as water seepage and deformation data, that is, the degradation or service life of the tunnel is directly related to the coupling of multiple parameters. In the case where the difference between the real-time degradation state and the real-time remaining life of the tunnel and the prediction result satisfies the tunnel difference threshold, it indicates that the model parameters related to the coupling of multiple parameters in the state evaluation model need to be adjusted, that is, the multi-head attention weight of the cross-modal feature interaction layer needs to be adjusted. By updating the multi-head attention weight of the cross-modal feature interaction layer, the correlation analysis capability of water seepage-deformation-convergence can be optimized.

[0112] (4) In the case where the difference between the real-time degradation state and the real-time remaining life of the track and the prediction result satisfies the track difference threshold, the full connection layer weight of the multi-task output layer is updated.

[0113] In implementation, the full connection layer weight of the multi-task output layer can be updated by using an adaptive momentum optimization algorithm (such as the AdamW algorithm).

[0114] The track involves multiple components (such as rails, ballast beds, fasteners, etc.), and different weights are usually set for the multiple components in the full connection layer of the multi-task output layer to balance the influence of the multiple components on the track. In the case where the difference between the real-time degradation state and the real-time remaining life of the track and the prediction result satisfies the track difference threshold, it indicates that the weight setting of the multiple components of the track in the full connection layer of the multi-task output layer is unreasonable, and the full connection layer weight of the multi-task output layer needs to be updated to calibrate the comprehensive weight of the components such as rails, ballast beds, and fasteners in the track.

[0115] According to whether the output of the state evaluation model satisfies the above four conditions, a corresponding update strategy can be selected. For example, if the output of the state evaluation model satisfies conditions (1) and (2), the relevant update strategies in conditions (1) and (2) (updating the forgetting gate weight matrix parameters of the LSTM and updating the convolution kernel parameters of the CNN) are used to update the parameters of the state evaluation model.

[0116] In the embodiments of the present disclosure, through steps 201 to 206, the state dynamic evaluation and model closed-loop optimization mechanism of the high-speed rail infrastructure are realized. Through the state evaluation model, the feature extraction and fusion analysis of multi-source data are realized, and an end-to-end mapping relationship from the original data to the degradation state and the remaining life is established. Among them, based on the difference between the real-time degradation state, the real-time remaining life and the prediction result, the parameters of the state evaluation model are updated, which is equivalent to the difference between the real-time degradation state, the real-time remaining life and the prediction result for the closed-loop feedback of the parameters of the state evaluation model, realizing the continuous optimization of the model parameters by the real-time monitoring data, and maintaining the accuracy of the evaluation result.

[0117] The following is a device embodiment of the present application. For details not described in detail in the device embodiment, please refer to the above method embodiments.

[0118] Figure 3 The structure schematic diagram of the high-speed rail line infrastructure state evaluation device based on space-air-ground integration provided by an example embodiment of the present disclosure is shown. Referring to Figure 3 The high-speed rail line infrastructure state evaluation device based on space-air-ground integration 300 includes an acquisition module 301, a conversion module 302, an alignment module 303, a prediction module 304, a real-time monitoring module 305 and an update module 306.

[0119] The acquisition module 301 is configured to acquire a multi-source data set, wherein the multi-source data set includes data collected by a plurality of sensors for monitoring roads, bridges, tunnels and rails.

[0120] The conversion module 302 is configured to convert data in the multi-source data set into a plurality of structured evaluation indicators to obtain an evaluation indicator set, wherein the structured evaluation indicators are used for state evaluation of roads, bridges, tunnels and rails.

[0121] The alignment module 303 is configured to perform time alignment, mileage alignment and spatial alignment on a plurality of structured evaluation indicators in the evaluation indicator set to obtain a plurality of evaluation vectors.

[0122] The prediction module 304 is configured to process the plurality of evaluation vectors by using a state evaluation model to obtain a prediction result of roads, bridges, tunnels and rails, wherein the prediction result includes a degradation state prediction result and a remaining life prediction result.

[0123] The real-time monitoring module 305 is configured to determine a real-time degradation state and a real-time residual life based on real-time monitoring data.

[0124] The updating module 306 is configured to update parameters of the state assessment model based on a difference between the real-time degradation state, the real-time residual life and the prediction result.

[0125] Optionally, the state assessment model comprises a convolutional neural network (CNN) connected in sequence, a long short-term memory (LSTM) network, a cross-modal feature interaction layer and a multi-task output layer, the CNN is configured to extract spatial features, the LSTM is configured to extract time sequence features, the cross-modal feature interaction layer is configured to fuse the spatial features and the time sequence features, and the multi-task output layer is configured to output the prediction result of the road, bridge, tunnel and track, and the updating module 306 is further configured to update a forgetting gate weight matrix parameter of the LSTM when a difference between the real-time degradation state, the real-time residual life and the prediction result of the roadbed satisfies a roadbed difference threshold, update a convolution kernel parameter of the CNN when a difference between the real-time degradation state, the real-time residual life and the prediction result of the bridge satisfies a bridge difference threshold, update a multi-head attention weight of the cross-modal feature interaction layer when a difference between the real-time degradation state, the real-time residual life and the prediction result of the tunnel satisfies a tunnel difference threshold, and update a fully connected layer weight of the multi-task output layer when a difference between the real-time degradation state, the real-time residual life and the prediction result of the track satisfies a track difference threshold.

[0126] Optionally, the updating module 306 is further configured to reweight the forgetting gate weight matrix parameter of the LSTM based on a time sequence attention mechanism, update the convolution kernel parameter of the CNN based on a gradient clipping update manner, update the multi-head attention weight of the cross-modal feature interaction layer by using a federated learning collaborative fine-tuning algorithm, and update the fully connected layer weight of the multi-task output layer by using an adaptive momentum optimization algorithm.

[0127] Optionally, the alignment module 303 is further configured to implement time sequence alignment of a plurality of structured evaluation indexes in the evaluation index set by using a dynamic time warping (DTW) algorithm to obtain an aligned path set, the aligned path set comprises a plurality of time-aligned paths, implement data synchronization by using linear interpolation for any index pair in a first time-aligned path, the first time-aligned path is any time-aligned path in the aligned path set, implement mileage alignment by using a piecewise cubic spline function for adjacent mileage points, and convert coordinates of the plurality of structured evaluation indexes to a same spatial coordinate system by using a seven-parameter Helmert transformation.

[0128] It should be noted that: the above embodiment provides the high-speed rail line infrastructure state evaluation device based on the space-air-ground integration to evaluate the high-speed rail line infrastructure state, and only the above-mentioned division of each functional module is used as an example for illustration. In actual application, the above-mentioned functions can be distributed to be completed by different functional modules according to needs, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. In addition, the high-speed rail line infrastructure state evaluation device based on the space-air-ground integration provided in the above embodiment and the high-speed rail line infrastructure state evaluation method based on the space-air-ground integration belong to the same concept, and the specific implementation process is described in the method embodiment, which will not be repeated here.

[0129] The division of the modules in the embodiments of the present disclosure is illustrative, and is only a logical functional division. In actual implementation, there can be another division manner. In addition, each functional module in each embodiment of the present disclosure can be integrated in one processor, or can be a separate physical existence, or two or more modules can be integrated into one module. The integrated module can be realized in the form of hardware or in the form of a software functional module.

[0130] When the integrated module is realized in the form of a software functional module and sold or used as an independent product, it can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present disclosure essentially or the part that contributes to the prior art or the whole or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium, and includes a plurality of instructions for causing an end device (which can be a personal computer, a mobile phone, or a communication device, etc.) or a processor to execute all or part of the steps of the method of each embodiment of the present disclosure. The foregoing storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various program code storage media.

[0131] Figure 4 is a structural schematic diagram of the computer device provided by the embodiments of the present disclosure. As shown in Figure 4 the computer device 400 includes a processor 401 and a memory 402.

[0132] The processor 401 can include one or more processing cores, such as a 4-core processor, an 8-core processor, and the like. The processor 401 can be implemented in at least one of a hardware form of a DSP (Digital Signal Processing), an FPGA (Field-Programmable Gate Array), a PLA (Programmable Logic Array). The processor 401 can also include a main processor and a coprocessor, the main processor being a processor for processing data in an awake state, also referred to as a CPU (Central Processing Unit), and the coprocessor being a low-power processor for processing data in a standby state. In some embodiments, the processor 401 can be integrated with a GPU (Graphics Processing Unit) that is responsible for rendering and drawing of content to be displayed by the display screen. In some embodiments, the processor 401 can further include an AI (Artificial Intelligence) processor for processing computing operations related to machine learning.

[0133] The memory 402 can include one or more computer-readable storage media that can be non-transitory. The memory 402 can also include a high-speed random access memory, and a nonvolatile memory such as one or more disk storage devices, flash storage devices. In some embodiments, the non-transitory computer-readable storage medium in the memory 402 is used to store at least one instruction for being executed by the processor 401 to implement the space-air-ground integrated high-speed rail line infrastructure state evaluation method provided in the embodiments of the present disclosure.

[0134] Those skilled in the art can understand that, Figure 4 The structure shown in the figure does not constitute a limitation on the computer device 400, and can include more or fewer components than those shown, or combine certain components, or adopt a different arrangement of components.

[0135] The embodiments of the present disclosure also provide a non-transitory computer-readable storage medium, when instructions in the storage medium are executed by a processor of a computer device, the computer device is enabled to perform the space-air-ground integrated high-speed rail line infrastructure state evaluation method provided in the embodiments of the present disclosure.

[0136] The embodiment of the present disclosure further provides a computer program product comprising computer programs / instructions which, when executed by a processor, implement the space-air-ground integrated high-speed rail line infrastructure state evaluation method provided in the embodiment of the present disclosure.

[0137] The above merely describes optional embodiments of the present disclosure, and is not intended to limit the present disclosure. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present disclosure shall be included in the protection scope of the present disclosure.

Claims

1. A method for assessing the condition of high-speed railway infrastructure based on an integrated air-space-vehicle-ground system, characterized in that, The method includes: Acquire a multi-source data set, which includes data collected by multi-source sensors for monitoring roads, bridges, tunnels and tracks; The data in the multi-source data set is converted into multiple structured evaluation indicators to obtain an evaluation indicator set, which is used to evaluate the condition of roads, bridges, tunnels and railways. After performing time-series alignment, mileage alignment, and spatial alignment on multiple structured evaluation indicators in the set of evaluation indicators, multiple evaluation vectors are obtained. The multiple evaluation vectors are processed using a state assessment model to obtain prediction results for roads, bridges, tunnels, and railways. The prediction results include degradation state prediction results and remaining life prediction results. Real-time degradation status and real-time remaining lifetime are determined based on real-time monitoring data; The parameters of the state assessment model are updated based on the difference between the real-time degradation state, the real-time remaining lifetime, and the prediction result.

2. The method according to claim 1, characterized in that, The state evaluation model comprises a convolutional neural network (CNN), a long short-term memory network (LSTM), a cross-modal feature interaction layer, and a multi-task output layer, connected in sequence. The CNN is used to extract spatial features, the LSTM is used to extract temporal features, the cross-modal feature interaction layer is used to fuse the spatial and temporal features, and the multi-task output layer is used to output the prediction results for roads, bridges, tunnels, and railways. The step of updating the parameters of the state assessment model based on the difference between the real-time degradation state, the real-time remaining lifetime, and the prediction result includes: If the difference between the real-time degradation state of the roadbed, the real-time remaining lifetime, and the prediction result meets the roadbed difference threshold, the forget gate weight matrix parameters of the LSTM are updated. If the difference between the bridge's real-time degradation state, the real-time remaining lifetime, and the prediction result meets the bridge difference threshold, the convolution kernel parameters of the CNN are updated. If the difference between the real-time degradation state of the tunnel, the real-time remaining lifetime, and the prediction result meets the tunnel difference threshold, update the multi-head attention weights of the cross-modal feature interaction layer. If the difference between the real-time degradation state of the orbit, the real-time remaining lifetime, and the prediction result meets the orbit difference threshold, the weights of the fully connected layer of the multi-task output layer are updated.

3. The method according to claim 2, characterized in that, The step of updating the forget gate weight matrix parameters of the LSTM includes: reweighting the forget gate weight matrix parameters of the LSTM based on a temporal attention mechanism; The updating of the convolution kernel parameters of the CNN includes: updating the convolution kernel parameters of the CNN based on gradient clipping. The step of updating the multi-head attention weights of the cross-modal feature interaction layer includes: updating the multi-head attention weights of the cross-modal feature interaction layer using a federated learning collaborative fine-tuning algorithm; The step of updating the fully connected layer weights of the multi-task output layer includes: updating the fully connected layer weights of the multi-task output layer using an adaptive momentum optimization algorithm.

4. The method according to claim 2 or 3, characterized in that, The step of aligning multiple structured evaluation indicators in the evaluation indicator set according to time sequence includes: The Dynamic Time Warping (DTW) algorithm is used to perform time-series alignment of multiple structured evaluation indicators in the evaluation indicator set to obtain an alignment path set, which includes multiple time alignment paths. For any index pair in the first time alignment path, linear interpolation is used to achieve data synchronization, where the first time alignment path is any time alignment path in the set of alignment paths. The step of aligning multiple structured evaluation metrics in the evaluation metric set with mileage includes: Multiple mileage points are set on the line, and mileage alignment is achieved by using a piecewise cubic spline function for adjacent mileage points; The step of spatially aligning multiple structured evaluation indicators in the set of evaluation indicators includes: The coordinates of the multiple structured evaluation indicators are transformed to the same spatial coordinate system using a seven-parameter Helmert transform.

5. The method according to claim 4, characterized in that, No. Time of the first The evaluation vector corresponding to the mileage point is represented in the following way: in, For the first Time of the first The evaluation vector corresponding to the mileage point For the space alignment of the first Time of the first The coordinates of the mileage point For the first Time of the first Mileage point One evaluation indicator, , , All are positive integers. The value range is 1 to , The total number of mileage points. The value range is 1 to , The length of the time series. The value range is 1 to , This represents the total number of evaluation indicators.

6. The method according to claim 5, wherein the CNN is a three-dimensional CNN, and the CNN extracts the spatial features of the evaluation vector in the following manner: in, for Spatial characteristics, For activation function, , This is the time neighborhood offset. This is the spatial neighborhood offset. Indicates the time neighborhood offset as Spatial neighborhood offset is In this case, The Convolutional kernel weights in each dimension, Here is the convolutional kernel weight matrix of the CNN. , For element-wise multiplication, This is the bias of the CNN.

7. A high-speed railway line infrastructure condition assessment device based on integrated space-air-ground system, characterized in that, The device includes: The acquisition module is used to acquire a multi-source data set, which includes data collected by multi-source sensors for monitoring roads, bridges, tunnels and tracks; The conversion module is used to convert the data in the multi-source data set into multiple structured evaluation indicators to obtain an evaluation indicator set, which is used to evaluate the status of roads, bridges, tunnels and railways. The alignment module is used to perform time-series alignment, mileage alignment, and spatial alignment on multiple structured evaluation indicators in the evaluation indicator set to obtain multiple evaluation vectors. The prediction module is used to process the multiple evaluation vectors using a state assessment model to obtain prediction results for roads, bridges, tunnels and railways. The prediction results include degradation state prediction results and remaining life prediction results. The real-time monitoring module is used to determine the real-time degradation status and real-time remaining lifetime based on real-time monitoring data; An update module is used to update the parameters of the state assessment model based on the difference between the real-time degradation state, the real-time remaining lifetime, and the prediction result.

8. A computer device, characterized in that, The computer device includes a memory and a processor, wherein the memory stores at least one computer program, which is loaded and executed by the processor to implement the method according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores at least one computer program, which is loaded and executed by a processor to implement the method according to any one of claims 1 to 6.

10. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instructions are executed by the processor, they implement the method described in any one of claims 1 to 6.

Images