Railway kilometer marker prediction method, device and medium based on geospatial data

Abstract

The application discloses a railway kilometer marker prediction method and device based on geospatial data and a medium, relates to the technical field of rail transit, and comprises the following steps: processing railway surveying and mapping data of a train operation line to obtain geospatial data; for each data point in the geospatial data, calculating distance characteristics of the data point based on the longitude and latitude of the data point, and calculating elevation change characteristics of the data point based on the elevation of the data point; taking the position characteristics, distance characteristics and elevation change characteristics of the data point as input and taking the kilometer marker of the data point as a label; training a first initial model and a second initial model respectively to obtain a first trained model and a second trained model; integrating the first trained model and the second trained model to obtain an integrated model; and predicting the kilometer marker of any position point in the train operation line by using the integrated model. The application can improve the accuracy and reliability of railway kilometer marker prediction.

Description

Technical Field

[0001] This application relates to the field of rail transit technology, and in particular to a method, device and medium for predicting railway kilometer markers based on geospatial data. Background Technology

[0002] In fields such as transportation engineering and geographic surveying, it is often necessary to determine the corresponding kilometer marker values ​​based on geographic coordinate information (i.e., longitude and latitude). For example, in the construction and maintenance of railways and highways, accurately obtaining kilometer markers is crucial for tasks such as route planning and facility positioning.

[0003] Traditional railway kilometer marker positioning mainly relies on manual measurement or single sensor data, which suffers from low efficiency, high cost, and poor environmental adaptability. Related technologies often use geographic coordinate information as input and predict railway kilometer markers based on a single machine learning model (such as a neural network model). However, this method, relying solely on geographic coordinate information, fails to adequately utilize multi-source geospatial information (i.e., longitude, latitude, and elevation), resulting in limited prediction accuracy. Furthermore, this method, based on only a single machine learning model, is prone to overfitting or insufficient generalization ability when data is missing or terrain changes abruptly, further limiting prediction accuracy and reliability. Summary of the Invention

[0004] The purpose of this application is to provide a method, device, and medium for predicting railway kilometer markers based on geospatial data, which can improve the accuracy and reliability of railway kilometer marker prediction.

[0005] To achieve the above objectives, this application provides the following solution.

[0006] Firstly, this application provides a method for predicting railway kilometer markers based on geospatial data, the method comprising:

[0007] Obtain railway surveying data of the train operation line; the railway surveying data includes the longitude, latitude, elevation and kilometer marker of the starting point and the longitude, latitude, elevation and kilometer marker of the ending point of each segment of the train operation line, as well as the longitude, latitude and elevation of multiple surveying points in the train operation line, and all the segments are connected in sequence to form the train operation line;

[0008] The railway survey data is processed to calculate the kilometer marker for each survey point, thus obtaining geospatial data.

[0009] For each data point in the geographic space data, based on the longitude and latitude of the data point, a great circle spherical distance of the data point to a center point of a train operation line is calculated to obtain a distance feature of the data point, and based on the elevation of the data point, an elevation change feature of the data point is calculated; the data point includes a starting point, an ending point and a surveying point in the geographic space data;

[0010] The first initial model and the second initial model are trained respectively by taking the position feature, the distance feature and the elevation change feature of the data point as input and taking the kilometer mark of the data point as label, to obtain a first trained model and a second trained model, and the first trained model and the second trained model are integrated to obtain an integrated model; the position feature includes longitude, latitude and elevation;

[0011] The integrated model is used to predict the kilometer mark of any position point in the train operation line.

[0012] In a second aspect, the present application provides a computer device, comprising a memory, a processor and a computer program stored in the memory and capable of running on the processor, and the processor executes the computer program to realize the above-mentioned railway kilometer mark prediction method based on geographic space data.

[0013] In a third aspect, the present application provides a computer readable storage medium having a computer program stored thereon, and the computer program is executed by a processor to realize the above-mentioned railway kilometer mark prediction method based on geographic space data.

[0014] According to the specific embodiments provided by the present application, the present application has the following technical effects.

[0015] The application provides a railway kilometer marker prediction method based on geospatial data, equipment and a medium. Geospatial data is obtained by processing railway surveying and mapping data of a train operation line. For each data point in the geospatial data, a great circle spherical distance from the data point to a center point of the train operation line is calculated based on the longitude and latitude of the data point, to obtain a distance feature of the data point. An elevation change feature of the data point is calculated based on the elevation of the data point. The position feature, distance feature and elevation change feature of the data point are taken as inputs, and the kilometer marker of the data point is taken as a label. A first initial model and a second initial model are trained respectively to obtain a first trained model and a second trained model. The first trained model and the second trained model are integrated to obtain an integrated model. The integrated model is used to predict the kilometer marker of any position point in the train operation line. The longitude, latitude and elevation are fused to design the position feature, distance feature and elevation change feature. The position feature, distance feature and elevation change feature are subsequently taken as inputs to predict the railway kilometer marker. Compared with a method in which only geographic coordinate information is taken as input, the method can solve the problem of insufficient fusion and utilization of multi-source geospatial information, improve accuracy, and take the position feature, distance feature and elevation change feature as input. The integrated model of two machine learning models is used to predict the railway kilometer marker. Compared with a method in which only a single machine learning model is used for prediction, the method can solve the problem of overfitting or insufficient generalization ability when data is missing or the terrain is suddenly changed, improve accuracy and reliability, and finally fuse multi-source geospatial information and two machine learning models to predict the railway kilometer marker, improve the accuracy and reliability of railway kilometer marker prediction. BRIEF DESCRIPTION OF DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiments will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0017] Figure 1 An application environment diagram of a railway kilometer marker prediction method based on geospatial data provided by Embodiment 1 of the present application.

[0018] Figure 2 A flowchart of a railway kilometer marker prediction method based on geospatial data provided by Embodiment 1 of the present application.

[0019] Figure 3 A technical route diagram of a railway kilometer marker prediction method based on geospatial data provided by Embodiment 1 of the present application.

[0020] Figure 4 The schematic diagram of the calculation process of the kilometer marker of the survey point provided for Embodiment 1 of the present application.

[0021] Figure 5 The schematic diagram of the process of constructing the feature matrix of the training set, the verification set and the test set provided for Embodiment 1 of the present application.

[0022] Figure 6 The schematic diagram of the process of constructing the double-branch prediction model of the feedforward neural network model and the wide learning system model provided for Embodiment 1 of the present application.

[0023] Figure 7 The schematic diagram of the process of integrating the two prediction models provided for Embodiment 1 of the present application.

[0024] Figure 8 The schematic diagram of the comparison of the prediction results of different models provided for Embodiment 1 of the present application.

[0025] Figure 9 The schematic diagram of the prediction error curve of different models provided for Embodiment 1 of the present application.

[0026] Figure 10 The error distribution histogram of different models provided for Embodiment 1 of the present application.

[0027] Figure 11 The error and elevation (i.e. height) relationship diagram of different models provided for Embodiment 1 of the present application.

[0028] Figure 12 The structural schematic diagram of a computer device provided for Embodiment 2 of the present application. DETAILED DESCRIPTION

[0029] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all the other embodiments obtained by those of ordinary skill in the art without any creative work, fall within the scope of protection of the present application.

[0030] Embodiment 1.

[0031] The railway kilometer marker prediction method based on geographic spatial data provided by the embodiments of the present application can be applied to, for example, Figure 1The application environment shown. Among them, the terminal communicates with the server through the network. The data storage system can store the data required by the server to process. The data storage system can be separately arranged, or integrated on the server, or placed on the cloud or other servers. The terminal can send a prediction request to be processed to the server, and the server receives the prediction request to be processed. For the prediction request to be processed, the server obtains the railway surveying and mapping data of the train operation line; the railway surveying and mapping data is processed to calculate the kilometer mark of each surveying and mapping point to obtain geographic spatial data; for each data point in the geographic spatial data, based on the longitude and latitude of the data point, the great circle spherical distance from the data point to the center point of the train operation line is calculated to obtain the distance feature of the data point, and based on the elevation of the data point, the elevation change feature of the data point is calculated; the position feature, distance feature and elevation change feature of the data point are taken as input, and the kilometer mark of the data point is taken as label, and the first initial model and the second initial model are trained respectively to obtain the first trained model and the second trained model. The first trained model and the second trained model are integrated to obtain an integrated model; the integrated model is used to predict the kilometer mark of any position point in the train operation line. The server can feed back the prediction result of the kilometer mark of any position point in the train operation line to the terminal.

[0032] In addition, in some embodiments, the railway kilometer mark prediction method based on geographic spatial data can also be implemented by the server or the terminal alone, such as being directly processed by the terminal for the prediction request to be processed, or being processed by the server for the prediction request to be processed from the data storage system.

[0033] Among them, the terminal can be, but not limited to, various desktop computers, notebook computers, smart phones, tablet computers, Internet of Things devices and portable wearable devices, the Internet of Things devices can be smart speakers, smart televisions, smart air conditioners, smart vehicle devices, etc., and the portable wearable devices can be smart watches, smart bracelets, head-mounted devices, etc. The server can be implemented by an independent server or a server cluster composed of multiple servers, and can also be a cloud server.

[0034] In an exemplary embodiment, as Figure 2 shown, a railway kilometer mark prediction method based on geographic spatial data is provided, which is executed by a computer device, specifically by a terminal or a server, or by both the terminal and the server, and in the embodiments of the present application, the method is applied to the server in Figure 1 , including the following steps.

[0035] Step S1, obtaining railway surveying and mapping data of a train operation line; the railway surveying and mapping data includes longitude, latitude and elevation of a starting point and a terminal point of each line segment in the train operation line, and longitude, latitude and elevation of a plurality of surveying and mapping points in the train operation line, and all the line segments are sequentially connected to form the train operation line.

[0036] Step S2, processing the railway surveying and mapping data to calculate the kilometer mark of each surveying and mapping point to obtain geographic spatial data.

[0037] Step S3, for each data point in the geographic spatial data, based on the longitude and latitude of the data point, calculating the great circle spherical distance from the data point to the center point of the train operation line to obtain the distance feature of the data point, and based on the elevation of the data point, calculating the elevation change feature of the data point; the data points include the starting point, the terminal point and the surveying and mapping points in the geographic spatial data.

[0038] Step S4, taking the position feature, distance feature and elevation change feature of the data point as input, taking the kilometer mark of the data point as label, training the first initial model and the second initial model respectively to obtain the first trained model and the second trained model, integrating the first trained model and the second trained model to obtain an integrated model; the position feature includes longitude, latitude and elevation.

[0039] Step S5, using the integrated model to predict the kilometer mark of any position point in the train operation line.

[0040] Implementing the above steps S1 to S5, the embodiment fuses longitude, latitude and elevation to design position feature, distance feature and elevation change feature, takes the position feature, distance feature and elevation change feature as input, and uses an integrated model integrating two machine learning models to predict the railway kilometer mark, which can solve the problem of limited prediction accuracy caused by insufficient utilization of multi-source geographic spatial information fusion, and the problem of limited prediction accuracy and insufficient reliability caused by overfitting or insufficient generalization ability of a single machine learning model, and improve the prediction accuracy and reliability.

[0041] Hereinafter, the method for predicting the railway kilometer mark based on geographic spatial data will be described in detail. Figure 3 The method for predicting the railway kilometer mark based on geographic spatial data used in the embodiment will be described in detail, which includes the following steps.

[0042] (I) Data acquisition.

[0043] The embodiment obtains railway surveying and mapping data of a train operation line. The railway surveying and mapping data refers to geographical spatial information related to a railway obtained by various surveying and mapping means in the process of railway construction, operation and maintenance. The railway surveying and mapping data includes longitude, latitude and elevation of a starting point and a terminal point of each line segment in the train operation line, and longitude, latitude and elevation of a plurality of surveying and mapping points in the train operation line, but does not determine the affiliation of the surveying and mapping points to the line segments, that is, does not determine which line segment the surveying and mapping points belong to. All the line segments are sequentially connected to form the train operation line, which is equivalent to randomly dividing the train operation line into a plurality of line segments.

[0044] (II) Data preprocessing.

[0045] The embodiment processes the railway surveying and mapping data to calculate the kilometer mark of each surveying and mapping point to obtain geographical spatial data.

[0046] Processing the railway surveying and mapping data includes deleting duplicate data points and calculating the kilometer mark of the surveying and mapping points to obtain geographical spatial data. At this time, as shown in FIG. 1, the railway surveying and mapping data is processed to calculate the kilometer mark of each surveying and mapping point to obtain geographical spatial data, and specifically includes the following steps. Figure 4

[0047] (1) Delete duplicate initial data points in the railway surveying and mapping data to obtain initial geographical spatial data. The initial data points include starting points, terminal points and surveying and mapping points.

[0048] Duplicate initial data points refer to initial data points with the same longitude, latitude and elevation. For duplicate initial data points, only one is retained. For example, if the longitude, latitude and elevation of two initial data points are the same, one initial data point is deleted.

[0049] (2) For each surveying and mapping point in the initial geographical spatial data, calculate the projection parameters of the surveying and mapping point to each line segment. For each line segment, calculate the three-dimensional coordinate vector of the projection point of the surveying and mapping point projected onto the line segment based on the projection parameters, and calculate the shortest arc length distance of the surveying and mapping point to the line segment based on the three-dimensional coordinate vector of the projection point. Select the line segment corresponding to the minimum value of the shortest arc length distance as the nearest line segment, and determine the kilometer mark of the surveying and mapping point based on the three-dimensional coordinate vector of the projection point of the surveying and mapping point projected onto the nearest line segment to calculate the kilometer mark of each surveying and mapping point to obtain geographical spatial data.

[0050] ​For each line segment in the train operation line, longitude and latitude of the starting point and longitude and latitude of the ending point of the line segment are extracted, and the projection parameter and the shortest arc length distance of the surveying point to each line segment are calculated based on longitude and latitude of the surveying point, longitude and latitude of the starting point of each line segment, and longitude and latitude of the ending point of each line segment, so as to obtain the shortest arc length distance corresponding to each line segment. Finally, the minimum value of the shortest arc length distances corresponding to all line segments is taken as the minimum distance, and the kilometer mark of the surveying point is taken as the nearest kilometer mark. At this time, the calculation method of the kilometer mark of the surveying point is as follows: the line segment corresponding to the minimum value of the shortest arc length distance is taken as the nearest line segment, and the kilometer mark of the surveying point is obtained by interpolation calculation between the kilometer mark of the starting point of the nearest line segment and the kilometer mark of the ending point of the nearest line segment based on the three-dimensional coordinate vector of the projection point of the surveying point on the nearest line segment, the three-dimensional coordinate vector of the starting point of the nearest line segment, and the three-dimensional coordinate vector of the ending point of the nearest line segment.

[0051] In the calculation of the projection parameter of the surveying point to each line segment, the calculation specifically includes: for each line segment, the three-dimensional coordinate vector of the surveying point, the three-dimensional coordinate vector of the starting point of the line segment, and the three-dimensional coordinate vector of the ending point of the line segment are calculated based on longitude and latitude of the surveying point, longitude and latitude of the starting point of the line segment, and longitude and latitude of the ending point of the line segment, the first vector difference is obtained by calculating the difference between the three-dimensional coordinate vector of the surveying point and the three-dimensional coordinate vector of the starting point of the line segment, the second vector difference is obtained by calculating the difference between the three-dimensional coordinate vector of the ending point of the line segment and the three-dimensional coordinate vector of the starting point of the line segment, and the projection parameter of the surveying point to the line segment is calculated based on the first vector difference and the second vector difference.

[0052] In the conversion of longitude and latitude of the surveying point and the end points (including the starting point and the ending point) of the line segment into the three-dimensional coordinate vector, for the point (x, y, z) , the three-dimensional coordinate vector thereof is:

[0053] ;

[0054] wherein, x is the x coordinate of the point; r is the radius of the earth; is the latitude of the point; is the longitude of the point; y is the y coordinate of the point; z is the z coordinate of the point; is the three-dimensional coordinate vector of the point.

[0055] The calculation formula of the projection parameter is:

[0056] ;

[0057] wherein, ​is a projection parameter; is a first vector difference, i.e., a vector difference between the surveying point and the start point of the line segment in the three-dimensional space, equal to a difference between a three-dimensional coordinate vector of the surveying point and a three-dimensional coordinate vector of the start point of the line segment, specifically, subtracting the three-dimensional coordinate vector of the start point of the line segment from the three-dimensional coordinate vector of the surveying point; is a second vector difference, i.e., a vector from the start point of the line segment to the end point of the line segment, i.e., a vector difference between the end point of the line segment and the start point of the line segment in the three-dimensional space, equal to a difference between a three-dimensional coordinate vector of the end point of the line segment and a three-dimensional coordinate vector of the start point of the line segment, specifically, subtracting the three-dimensional coordinate vector of the start point of the line segment from the three-dimensional coordinate vector of the end point of the line segment.

[0058] wherein, after the projection parameter is calculated, a three-dimensional coordinate vector of a projection point of the surveying point on the line segment is calculated based on the projection parameter, specifically including: calculating a product of the projection parameter and the second vector difference, and calculating a sum of the three-dimensional coordinate vector of the start point of the line segment and the product, to obtain the three-dimensional coordinate vector of the projection point of the surveying point on the line segment.

[0059] The calculation formula of the three-dimensional coordinate vector of the projection point is:

[0060] ;

[0061] wherein, is the three-dimensional coordinate vector of the projection point; is the three-dimensional coordinate vector of the start point of the line segment; is the projection parameter; is the second vector difference.

[0062] wherein, after the three-dimensional coordinate vector of the projection point is calculated, a shortest arc length distance of the surveying point to the line segment is calculated based on the three-dimensional coordinate vector of the projection point, specifically including: based on the three-dimensional coordinate vector of the surveying point and the three-dimensional coordinate vector of the projection point, calculating an included angle between the surveying point and the projection point, calculating a product of the earth radius and the included angle, to obtain the shortest arc length distance of the surveying point to the line segment.

[0063] The calculation formula of the included angle between the surveying point and the projection point is:

[0064] ;

[0065] wherein, is the included angle between the surveying point and the projection point; is the three-dimensional coordinate vector of the surveying point; is the three-dimensional coordinate vector of the projection point; represents a calculation of a length.

[0066] The calculation formula of the shortest arc length distance is:

[0067] ;

[0068] wherein, is the shortest arc length distance; is the radius of the earth; is the included angle between the surveying point and the projection point.

[0069] (Three) feature calculation.

[0070] The embodiment can first perform outlier filtering, elevation mutation detection and coordinate deduplication on the geographic spatial data. The outlier filtering is indirectly filtering outliers through data verification and missing value processing. The elevation mutation detection is achieved by using the adjacent elevation difference threshold method to detect the elevation mutation. The core is to quickly locate and eliminate abnormal points by comparing the elevation change amplitude of adjacent points with a preset threshold. The coordinate deduplication is to perform row-level deduplication on the longitude, latitude and elevation columns in validData by using the unique function to obtain standardized geographic spatial data. Subsequent feature calculation is based on the standardized geographic spatial data, that is, the standardized geographic spatial data is taken as new geographic spatial data to execute subsequent steps.

[0071] For each data point in the geographic spatial data, the embodiment calculates the great circle spherical distance from the data point to the center point of the train operation line based on the longitude and latitude of the data point to obtain the distance feature of the data point. The elevation change feature of the data point is calculated based on the elevation of the data point. The data point includes the start point, the end point and the surveying point in the geographic spatial data.

[0072] The embodiment can divide the geographic spatial data, construct a mapping relationship between the three-dimensional geographic space and the kilometer mark by using the natural neighborhood interpolation algorithm, and fuse the great circle spherical distance and the elevation gradient change, that is, calculate the great circle spherical distance of each data point to the center point of the train operation line, and combine the elevation gradient change between the continuous data points to generate an enhanced spatial feature matrix (i.e., a spatial enhanced feature matrix). The enhanced spatial feature matrix includes the position feature, the distance feature and the elevation change feature of each data point.

[0073] For each data point in the geographic spatial data, the embodiment calculates the great circle spherical distance from the data point to the center point of the train operation line based on the longitude and latitude of the data point to obtain the distance feature of the data point. The elevation change feature of the data point is calculated based on the elevation of the data point. The data point includes the start point, the end point and the surveying point in the geographic spatial data.

[0074] (1) The natural neighborhood interpolation algorithm is used to interpolate the geographic spatial data to obtain a plurality of interpolation points, and the longitude, latitude, elevation and kilometer mark of each interpolation point are determined.

[0075] First, the longitude and latitude of each interpolation point are determined, and then the elevation and kilometer mark of each interpolation point are calculated. When calculating the elevation and kilometer mark of the interpolation point, the calculation formula is:

[0076] ;

[0077] wherein, is the elevation or kilometer mark of the interpolation point; is the number of natural neighborhood points; is the spatial weight of the first natural neighborhood point based on the Voronoi diagram; is the elevation or kilometer mark of the first natural neighborhood point, when is the elevation, then is the elevation, when is the kilometer mark, then is the kilometer mark.

[0078] (2) For each interpolated data point in the geographic spatial data, based on the longitude and latitude of the interpolated data point, the great circle spherical distance from the interpolated data point to the center point of the train operation line is calculated to obtain the distance feature of the interpolated data point, and based on the elevation of the interpolated data point, the elevation change feature of the interpolated data point is calculated. The interpolated data point includes the start point, the end point, the surveying point and the interpolation point in the geographic spatial data.

[0079] In this embodiment, the interpolated data point is taken as a new data point, and the subsequent steps are executed.

[0080] wherein, based on the longitude and latitude of the data point, the great circle spherical distance from the data point to the center point of the train operation line is calculated to obtain the distance feature of the data point, and based on the elevation of the data point, the elevation change feature of the data point is calculated, which specifically includes the following steps.

[0081] (1) Taking the longitude and latitude of the data point and the longitude and latitude of the center point of the train operation line as inputs, the great circle spherical distance from the data point to the center point of the train operation line is calculated by using the spherical triangle formula to obtain the distance feature of the data point.

[0082] When calculating the great circle spherical distance from the data point to the center point of the train operation line, the radius of the earth is set as , the longitude and latitude of the data point are , the longitude and latitude of the center point of the train operation line are , and according to the spherical triangle formula, the great circle spherical distance from the data point to the center point of the train operation line is calculated as:

[0083] ;

[0084] in, This is the distance between the surfaces of the great spheres. The radius of the Earth; For data points Latitude; The center point of the train line Latitude; For data points Longitude; The center point of the train line Longitude.

[0085] After calculating the distance to the great sphere, the great sphere distance is directly used as the distance feature of the data points.

[0086] Another method for calculating the distance across a great sphere is:

[0087] Using Earth's radius =6371000m, data points calculated based on the Haversine formula ( , ) and center point ( , The distance from the great sphere:

[0088] ;

[0089] in, Dimensions of data points Latitude of the center point The difference; Longitude of data points Longitude relative to the center point The difference.

[0090] (2) If the data point is the first data point, then set the elevation change feature of the data point to be equal to the elevation of the data point, and obtain the elevation change feature of the data point; otherwise, set the elevation change feature of the data point to be equal to the difference between the elevation of the data point and the elevation of the previous data point, and obtain the elevation change feature of the data point.

[0091] When extracting the elevation change characteristics of data points, the data points are arranged sequentially along the direction from the starting point to the ending point of the train route. Let the elevation sequence of each data point be denoted as... , Let be the number of data points, then the th The formula for calculating the elevation variation characteristics of each data point is:

[0092] ;

[0093] wherein, is an elevation change feature of the th data point; is an elevation of the th data point; is an elevation of the th data point.

[0094] After obtaining the distance feature and the elevation change feature of each data point (which can be replaced by the data point after interpolation), the embodiment can subsequently divide all data points into a training set and a validation set according to a certain proportion, train an initial model using the training set to obtain a trained model, verify the trained model using the validation set to obtain a prediction error of the trained model, integrate the trained model based on the prediction error to obtain an integrated model, or subsequently divide all data points into a training set, a validation set and a test set according to a certain proportion, train an initial model using the training set to obtain a trained model, verify the trained model using the validation set to obtain a prediction error of the trained model, integrate the trained model based on the prediction error to obtain an integrated model, and test the performance of the integrated model using the test set.

[0095] As another implementation, the embodiment can first divide to obtain a training set, a validation set and a test set, and then determine the distance feature and the elevation change feature of each data point in the training set, the validation set and the test set, respectively, to obtain a training set feature matrix, a validation set feature matrix and a test set feature matrix, as shown in Figure 5 For each data point in the geographic spatial data, based on the longitude and the latitude of the data point, the great circle spherical distance of the data point to the center point of the train operation line is calculated to obtain the distance feature of the data point, and based on the elevation of the data point, the elevation change feature of the data point is calculated, which specifically includes the following steps.

[0096] (1) The data points in the geographic spatial data are divided into an initial training set, an initial validation set and an initial test set.

[0097] The data points in the geographic spatial data are divided into an initial training set, an initial validation set and an initial test set according to a certain proportion.

[0098] (2) The natural neighborhood interpolation algorithm is used to interpolate the initial training set, the initial validation set and the initial test set respectively to obtain a training set, a validation set and a test set.

[0099] The natural neighbor interpolation algorithm is used to interpolate the initial training set, the initial validation set and the initial test set respectively, a plurality of interpolation points are generated in the initial training set, the initial validation set and the initial test set, and the longitude, latitude, elevation and kilometer mark of each interpolation point are determined, so as to obtain the training set, the validation set and the test set. The training set is used for training the model, the validation set is used for verifying the performance of the model, facilitating integration to obtain an integrated model, and the test set is used for testing the performance of the integrated model, objectively evaluating the generalization ability of the integrated model and avoiding misjudgment of the performance of the integrated model due to overfitting.

[0100] Taking the training set as an example, the data points (start point, end point, surveying point and interpolation point) of the training set are subjected to Voronoi division, each data point corresponds to a Voronoi cell, and the cell boundary is formed by the perpendicular bisector of the data point and the adjacent data point. For any interpolation point, its natural neighborhood is found, that is, all natural neighborhood points intersecting with the Voronoi cell of the interpolation point. These natural neighborhood points are the most adjacent points of the interpolation point in space. According to the area / volume intrusion proportion of the natural neighborhood points after the interpolation point is inserted, the weight of each natural neighborhood point is calculated, and finally the elevation and kilometer mark of the interpolation point are obtained by weighted average calculation.

[0101] (3) The distance feature and the elevation change feature of each data point (start point, end point, surveying point and interpolation point) in the training set, the validation set and the test set are calculated respectively.

[0102] For each data point in the training set, the validation set and the test set, based on the longitude and latitude of the data point, the great circle spherical distance of the data point to the center point of the train operation line is calculated to obtain the distance feature of the data point, and based on the elevation of the data point, the elevation change feature of the data point is calculated.

[0103] Wherein, based on the longitude and latitude of the data point, the great circle spherical distance of the data point to the center point of the train operation line is calculated to obtain the distance feature of the data point, and based on the elevation of the data point, the elevation change feature of the data point is calculated, which specifically includes the following steps.

[0104] 1) Taking the longitude and latitude of the data point and the longitude and latitude of the center point of the train operation line as inputs, the spherical triangle formula is used to calculate the great circle spherical distance of the data point to the center point of the train operation line to obtain the distance feature of the data point.

[0105] 2) If the data point is the first data point, the elevation change feature of the data point is set to be equal to the elevation of the data point to obtain the elevation change feature of the data point; otherwise, the elevation change feature of the data point is set to be equal to the difference between the elevation of the data point and the elevation of the previous data point of the data point to obtain the elevation change feature of the data point.

[0106] All data points are first globally and continuously numbered in the physical order from the start to the end, and then the elevation change characteristics of the data points are calculated according to the global and continuous numbering.

[0107] For the training set, let the longitude and latitude of the i-th data point be , and the longitude and latitude of the center point of the train operation line be , then the distance feature of the i-th data point is .

[0108] ;

[0109] wherein, is the distance feature of the i-th data point in the training set; is the radius of the earth; is the latitude of the i-th data point in the training set; is the latitude of the center point of the train operation line; is the longitude of the i-th data point in the training set; is the longitude of the center point of the train operation line. Similarly, the distance feature of the i-th data point in the validation set and the distance feature of the i-th data point in the test set are also calculated in the above manner.

[0110] For the training set, the validation set and the test set, respectively, is calculated, and the elevation change feature of the i-th data point in the training set , the elevation change feature of the i-th data point in the validation set and the elevation change feature of the i-th data point in the test set are obtained.

[0111] .

[0112] (4) Construct the feature matrix of the training set, the validation set and the test set.

[0113] Let the position feature vector of the training set (including the longitude, latitude and elevation of each data point in the training set) be , the distance feature vector (including the distance feature of each data point in the training set) be , and the elevation change feature vector (including the elevation change feature of each data point in the training set) be , then the construction formula of the training set feature matrix is:​​​​​​​​​

[0114] ;

[0115] Similarly, the construction formula of the validation set feature matrix and the test set feature matrix is respectively:

[0116] ;

[0117] ;

[0118] wherein, is the position feature vector of the validation set (including the longitude, latitude and elevation of each data point in the validation set); is the distance feature vector of the validation set (including the distance feature of each data point in the validation set); is the elevation change feature vector of the validation set (including the elevation change feature of each data point in the validation set); is the position feature vector of the test set (including the longitude, latitude and elevation of each data point in the test set); is the distance feature vector of the test set (including the distance feature of each data point in the test set); is the elevation change feature vector of the test set (including the elevation change feature of each data point in the test set).

[0119] Subsequently, the initial model is trained using the training set to obtain a trained model, the trained model is verified using the validation set to obtain the prediction error of the trained model, the trained model is integrated based on the prediction error to obtain an integrated model, and the performance of the integrated model is tested using the test set.

[0120] (4) Model construction.

[0121] The embodiment constructs a double-branch prediction model of the feedforward neural network model and the width learning system model, wherein the width learning system model comprises an orthogonalization weight matrix and a multi-type enhancement layer, and specifically adopts an orthogonalization enhancement layer and a radial basis function dynamic enhancement layer to improve the nonlinear characteristic expression capability.

[0122] As shown in Figure 6 , a double-branch prediction model of the feedforward neural network model and the width learning system model is constructed, specifically including the following steps.

[0123] (1) Construct a feedforward neural network model with 3 hidden layers, and then train it using the trainbr (Bayesian Regularization) algorithm.

[0124] The feedforward neural network model consists of one input layer, three hidden layers, and one output layer connected in sequence.

[0125] (2) Construct a width learning system model.

[0126] The width learning system model comprises an input layer, a feature node layer, an enhancement module, and an output layer connected in sequence. The enhancement module includes a first enhancement layer, a second enhancement layer, and a third enhancement layer. The inputs of the first, second, and third enhancement layers are all connected to the outputs of the feature node layer. The outputs of the feature node layer, the first, second, and third enhancement layers are all connected to the inputs of the output layer. The first enhancement layer uses a linear transformation and the ReLU activation function, the second enhancement layer uses a linear transformation and the Sigmoid activation function, and the third enhancement layer uses a linear transformation and the exponential activation function. The first enhancement layer is an orthogonal enhancement layer, and the third enhancement layer is a radial basis function dynamic enhancement layer.

[0127] The following is an introduction to each part.

[0128] 1) Feature node layer.

[0129] The feature node layer is used to calculate the unbiased feature node output and the bias of the feature nodes, and further calculates the final output of the feature nodes.

[0130] First, calculate the output of the unbiased feature nodes:

[0131] ;

[0132] in, Outputs unbiased feature nodes; The first feature node output without bias Line 1 The column has a total of element values. OK List, , , The number of feature nodes. The number of samples (i.e., data points); The transpose of the principal component loading matrix; This is the normalized feature matrix of the training set.

[0133] Then calculate the bias of the feature nodes:

[0134] ;

[0135] wherein, is the bias of the feature node; is the element value of the i-th row of the bias of the feature node, and there are N rows in total, ; is the average function used to calculate the average value.

[0136] Finally, the result of the bias-free feature node output added to the bias of the feature node is nonlinearly transformed by the Exponential Linear Unit (ELU) activation function to obtain the final output of the feature node:

[0137] ;

[0138] wherein, is the final output of the feature node; is the element value of the i-th row and the j-th column of the final output of the feature node, and there are N rows and M columns in total, , ; is the ELU activation function.

[0139] The calculation formula of is:

[0140] ;

[0141] wherein, is the negative input coefficient of the ELU activation function, controlling the exponential decay amplitude when the input is negative, , balancing the nonlinear expression and gradient stability, and keeping the dimension of the final output of the feature node consistent with the dimension of the bias-free feature node output.

[0142] 2) Enhancement module.

[0143] Determine the number of nodes in each enhancement layer in the enhancement module, construct multiple enhancement layers, and the total number of enhancement layers is three, i.e., the enhancement module includes three enhancement layers, and calculate the output of each enhancement layer.

[0144] The first enhancement layer adopts the ReLU (Rectified Linear Unit) activation function and the linear enhancement node of the orthogonalization weight matrix, and the output of the first enhancement layer is calculated through linear transformation and ReLU activation function:

[0145] ​​​​​​ ;

[0146] in, This is the output of the first enhancement layer; For the first enhancement layer output, the first Line number The column has a total of element values. OK List, , , This represents the number of nodes in the first enhancement layer. It is the ReLU activation function; This is the weight matrix of the first enhancement layer; This is the bias vector for the first enhancement layer.

[0147] The ReLU activation function is defined as:

[0148] ;

[0149] in, This is the maximum value function, used to calculate the maximum value; The weight matrix of the first enhancement layer is the first... Line number The element values ​​of the column; For the final output of the feature node Line number The element values ​​of the column; The first enhancement layer's bias vector is the first... The element value of the row.

[0150] The second enhancement layer uses non-linear enhancement nodes with the Sigmoid activation function. The output of the second enhancement layer is calculated through linear transformation and the Sigmoid activation function.

[0151] ;

[0152] in, Output for the second enhancement layer; For the output of the second enhancement layer Line number The column has a total of element values. OK List, , , This represents the number of nodes in the second enhancement layer. Use the Sigmoid activation function; This is the weight matrix for the second enhancement layer; This is the bias vector for the second enhancement layer.

[0153] The Sigmoid activation function is defined as:

[0154] ;

[0155] in, The weight matrix of the second enhancement layer is the first... Line 1 The element values ​​of the column; The bias vector of the second enhancement layer The element value of the row.

[0156] The third enhancement layer employs dynamic enhancement nodes with Gaussian functions, and its output is calculated through linear transformation and exponential activation functions.

[0157] ;

[0158] in, Output for the third enhancement layer; For the output of the third enhancement layer Line 1 The column has a total of element values. OK List, , , This represents the number of nodes in the third enhancement layer. It is an exponential activation function; This is the weight matrix for the third enhancement layer; This is the bias vector for the third enhancement layer.

[0159] The exponential activation function is defined as:

[0160] ;

[0161] in, The weight matrix of the third enhancement layer is the first... Line 1 The element values ​​of the column; The bias vector of the third enhancement layer The element value of the row.

[0162] 3) Output layer.

[0163] The output layer is used to combine the outputs of the feature node layer and each enhancement layer, calculate the output weight matrix, and then complete the prediction based on the output weight matrix.

[0164] The final output of the feature nodes and three-layer enhancement layer output Combined into a feature matrix i.e. " " represents splicing.

[0165] Recalculating the output weight matrix :

[0166] ;

[0167] wherein, is a regularization coefficient; is a unit matrix; is the normalized training set label, i.e. the kilometer mark of each data point in the training set.

[0168] (Five) Model integration.

[0169] In this embodiment, the position feature, distance feature and elevation change feature of the data point are taken as the input, and the kilometer mark of the data point is taken as the label. The first initial model and the second initial model are trained respectively to obtain the first trained model and the second trained model. The first trained model and the second trained model are integrated to obtain an integrated model. The position feature includes longitude, latitude and elevation.

[0170] The first initial model is a feedforward neural network model, and the second initial model is a width learning system model. The width learning system model includes an input layer, a feature node layer, an enhancement module and an output layer connected in sequence. The enhancement module includes a first enhancement layer, a second enhancement layer and a third enhancement layer. The input end of the first enhancement layer, the input end of the second enhancement layer and the input end of the third enhancement layer are connected to the output end of the feature node layer. The output end of the feature node layer, the output end of the first enhancement layer, the output end of the second enhancement layer and the output end of the third enhancement layer are connected to the input end of the output layer. The first enhancement layer is an enhancement layer using linear transformation and ReLU activation function. The second enhancement layer is an enhancement layer using linear transformation and Sigmoid activation function. The third enhancement layer is an enhancement layer using linear transformation and exponential activation function.

[0171] In this embodiment, the model weight is dynamically allocated based on the validation set error to realize meter-level high-precision integrated prediction, as shown in Figure 7 The model weight is dynamically allocated based on the validation set error, which specifically includes: calculating the mean absolute error of the feedforward neural network model and the width learning system model on the validation set, calculating the respective weights of the feedforward neural network model and the width learning system model based on the mean absolute error on the validation set, and integrating the feedforward neural network model and the width learning system model based on the weights. The advantages of the two prediction models are integrated. Specifically, the weights are allocated according to the mean absolute error of the two prediction models on the validation set. The prediction model with smaller mean absolute error is more accurate, and is given greater weight. The integrated prediction result is output based on the integrated model subsequently.

[0172] At this time, the position feature, distance feature and elevation change feature of the data points are taken as inputs, and the kilometer mark of the data points is taken as a label. The first initial model and the second initial model are trained respectively to obtain a first trained model and a second trained model. The first trained model and the second trained model are integrated to obtain an integrated model, and the specific steps include the following.

[0173] (1) All data points are divided into a training set and a validation set.

[0174] (2) The position feature, distance feature and elevation change feature of the data points in the training set are taken as inputs, and the kilometer mark of the data points in the training set is taken as a label. The first initial model and the second initial model are trained respectively to obtain a first trained model and a second trained model.

[0175] (3) The position feature, distance feature and elevation change feature of the data points in the validation set are taken as inputs, and the first prediction kilometer mark of the data points in the validation set is determined by using the first trained model. Based on the first prediction kilometer mark and the kilometer mark of the data points in the validation set, the mean absolute error is calculated to obtain the first prediction error of the first trained model.

[0176] The average absolute error of the feedforward neural network model on the validation set is calculated as follows:

[0177]

[0178] wherein, is the number of data points in the validation set; is the prediction value of the kilometer mark of the data points in the validation set by the feedforward neural network model, i.e. the first prediction kilometer mark; is the true value of the kilometer mark of the data points in the validation set, i.e. the kilometer mark. After the average absolute error of the feedforward neural network model on the validation set is calculated, the average absolute error of the feedforward neural network model on the validation set is directly taken as the first prediction error.

[0179] (4) The position feature, distance feature and elevation change feature of the data points in the validation set are taken as inputs, and the second prediction kilometer mark of the data points in the validation set is determined by using the second trained model. Based on the second prediction kilometer mark and the kilometer mark of the data points in the validation set, the mean absolute error is calculated to obtain the second prediction error of the second trained model.

[0180]

[0181] The average absolute error of the width learning system model on the validation set is calculated as follows: ​​​​​

[0182] ;

[0183] wherein, is the predicted value of the width learning system model for predicting the kilometer marker of the i-th data point in the validation set, i.e., the second predicted kilometer marker.

[0184] After calculating the mean absolute error of the width learning system model on the validation set, the mean absolute error of the width learning system model on the validation set is directly taken as the second prediction error.

[0185] (5) Based on the first prediction error and the second prediction error, the first weight of the first trained model and the second weight of the second trained model are calculated.

[0186] According to the validation set error, the weights of the feedforward neural network model and the width learning system model are calculated, and the calculation formula of the first weight is:

[0187] ;

[0188] wherein, is the first weight; is the first prediction error; is the second prediction error.

[0189] The calculation formula of the second weight is:

[0190] ;

[0191] wherein, is the second weight.

[0192] (6) Based on the first weight and the second weight, the first trained model and the second trained model are integrated to obtain an integrated model.

[0193] The present embodiment adopts a variety of feature engineering methods, provides a railway kilometer marker prediction method based on geographic spatial data and machine learning model fusion, comprehensively considers factors such as spatial relationship of geographic coordinates, interpolation information and elevation change, so that the model can better capture the characteristics of the data, by combining two different types of prediction models of feedforward neural network and width learning system, and performing model integration, the advantages of the two prediction models are fully utilized, compared with a single model, the generalization ability is better, the accuracy of kilometer marker prediction is improved, and finally the accuracy and reliability of railway kilometer marker prediction are improved.

[0194] (Six) Model application.

[0195] The present embodiment predicts the kilometer marker of any position point in the train operation line by using the integrated model. ​

[0196] The integrated model is used to predict the kilometer mark of any position point in the train operation line, specifically comprising: based on the longitude and latitude of any position point in the train operation line, calculating the great circle spherical distance from the position point to the center point of the train operation line to obtain the distance feature of the position point, based on the elevation of the position point, calculating the elevation change feature of the position point, taking the position feature, distance feature and elevation change feature of the position point as input, and using the integrated model to determine the predicted kilometer mark of the position point.

[0197] In the calculation of the elevation change feature, the position point and the data point are again globally sequentially numbered in the order of the starting point to the ending point, at this time the previous position point of the position point is determined, the difference between the elevation of the position point and the elevation of the previous position point is calculated to obtain the elevation change feature of the position point.

[0198] Wherein, taking the position feature, distance feature and elevation change feature of the position point as input, using the integrated model to determine the predicted kilometer mark of the position point, specifically comprising: taking the position feature, distance feature and elevation change feature of the position point as input, using the first trained model in the integrated model to determine the first kilometer mark prediction value of the position point, taking the position feature, distance feature and elevation change feature of the position point as input, using the second trained model in the integrated model to determine the second kilometer mark prediction value of the position point, and based on the first weight and the second weight, the first kilometer mark prediction value and the second kilometer mark prediction value are weighted and summed to obtain the predicted kilometer mark of the position point.

[0199] The calculation formula of the integrated prediction result of the predicted kilometer mark of the position point is:

[0200] ;

[0201] Wherein, is the predicted kilometer mark of the position point; is the first weight; is the first kilometer mark prediction value; is the second weight; is the second kilometer mark prediction value.

[0202] The embodiment can also set a system module to execute the above method, specifically comprising: data preprocessing, used to perform coordinate boundary verification (longitude , latitude ) and elevation mutation detection (threshold ), coordinate boundary check ensures that the coordinates of the input model conform to geographical common sense, avoids abnormal values interfering with training / prediction, and elevation mutation detection is used to eliminate abnormal points with dramatic changes in terrain (such as data collection errors and tunnel entrance and exit mutations) to ensure that the model learns stable terrain features; the real-time positioning engine deploys a lightweight width learning system model (by sparsification and reducing the number of nodes to reduce model computation and meet real-time requirements), and the response time is 30 kilometers, and through online learning (incremental weight W update), the model adapts to newly collected line data, such as changes in terrain of extended sections; the offline training platform provides a Bayesian regularized neural network training interface, supports historical data backfilling and model version management; and the visualization terminal generates an abnormal section traceability report.

[0203] In the real-time positioning engine, a sparse width learning system model is used, and model parameter calibration is triggered once every 100 meters; the number of nodes in the enhanced layer is dynamically adjusted to meet the node number , where is the current number of valid samples. The purpose of dynamically adjusting the number of nodes in the enhanced layer is to reduce the number of nodes to avoid overfitting when the sample is small (such as the initial stage of line extension), and to increase the number of nodes to improve accuracy when the sample is large (mature section of the line).

[0204] As shown in Figures 8-11 , the zero error reference in Figure 9 is the actual value, as shown in Figure 8 , it can be clearly seen that the integrated model of the two model integration is better than the two separate machine learning models in fitting accuracy and prediction stability, and the prediction result is closer to the true value, as shown in Figure 9 , through the whole cycle comparison of error dynamic change, it is fully proved that the integrated model has comprehensive superiority in error amplitude control, fluctuation suppression, direction balance and extreme correction, as shown in Figure 10 , through the quantitative comparison of error distribution, it can be seen that the integrated model is significantly better than the single machine learning model in overall accuracy, stability and extreme error control, as shown in Figure 11 , it can be seen that the integrated model is significantly better than the single machine learning model in error control ability, extreme terrain adaptability and error distribution uniformity in the whole altitude range. By comparing the prediction effects of different models, it is confirmed that the accuracy and reliability of the method of the embodiment are significantly improved.

[0205] The embodiment discloses a railway kilometer marker prediction method based on geographic space data and machine learning model fusion, relates to the technical field of rail transit, and comprises the following steps: acquiring railway surveying and mapping data of a train operation line, calculating unknown kilometer markers, deleting repeated point data, and obtaining geographic space data; dividing the geographic space data, constructing a mapping relationship between three-dimensional geographic space and kilometer markers through a natural neighborhood interpolation algorithm, fusing great circle spherical surface distance and elevation gradient change characteristics, and generating an enhanced space feature matrix; constructing a double-branch prediction model of a feedforward neural network model and a width learning system model, wherein the width learning system model adopts an orthogonalization enhancement layer and a radial basis function dynamic enhancement layer to improve the nonlinear feature expression capability; dynamically assigning model weights based on verification set errors, integrating the advantages of the two prediction models, and realizing high-precision integrated prediction of meter-level kilometer markers. The embodiment improves the accuracy and reliability of railway kilometer marker prediction.

[0206] The application also provides an application scenario of the railway kilometer marker prediction method based on geographic space data. Specifically, the railway kilometer marker prediction method based on geographic space data can be applied in a railway kilometer marker prediction scenario. The railway kilometer marker prediction scenario comprises a prediction link and a display link. The prediction link is used for predicting the kilometer marker of any position point of the train operation line, and the display link is used for displaying the predicted kilometer marker to a user. The railway kilometer marker prediction method based on geographic space data belongs to the prediction link.

[0207] Embodiment 2.

[0208] In an exemplary embodiment, a computer device, which can be a server or a terminal, is provided, and an internal structure diagram of the computer device can be as shown in Figure 12 The computer device comprises a processor, a memory, an input / output interface (I / O) and a communication interface. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device comprises a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium to run. The database of the computer device is used to store data. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with external terminals through network connection. The computer program is executed by the processor to implement a railway kilometer marker prediction method based on geographic space data.

[0209] Those skilled in the art can understand that,Figure 12 The structure shown in the figure is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. The specific computer device can include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0210] In an exemplary embodiment, a computer device is provided, comprising a memory and a processor, the memory storing a computer program, and the processor implementing the railway kilometer marker prediction method based on geospatial data in embodiment 1 when executing the computer program.

[0211] Embodiment 3.

[0212] In an exemplary embodiment, a computer readable storage medium is provided, storing a computer program, and the computer program implements the railway kilometer marker prediction method based on geospatial data in embodiment 1 when executed by a processor.

[0213] The technical features of the above embodiments can be combined in any manner. To make the description concise, not all possible combinations of the technical features in the above embodiments are described, but as long as the combinations of the technical features do not contradict, they should be considered within the scope of the present application.

[0214] The principles and implementation modes of the present application are described by using specific examples herein, and the above descriptions of the embodiments are only used to help understand the method of the present application and its core idea. For those skilled in the art, according to the idea of the present application, the specific implementation mode and application range can be changed. In summary, the content of the present application should not be understood as a limitation.

Claims

1. A method for predicting railway kilometer markers based on geospatial data, characterized in that, The method for predicting railway kilometer markers based on geospatial data includes: Obtain railway surveying data of the train operation line; the railway surveying data includes the longitude, latitude, elevation and kilometer marker of the starting point and the longitude, latitude, elevation and kilometer marker of the ending point of each segment of the train operation line, as well as the longitude, latitude and elevation of multiple surveying points in the train operation line, and all the segments are connected in sequence to form the train operation line; The railway survey data is processed to calculate the kilometer marker for each survey point, thus obtaining geospatial data. For each data point in the geospatial data, the great spherical distance from the data point to the center point of the train running line is calculated based on the longitude and latitude of the data point to obtain the distance characteristics of the data point. Based on the elevation of the data point, the elevation change characteristics of the data point are calculated. The data points include the start point, end point and mapping point in the geospatial data. Using the location features, distance features, and elevation change features of the data points as input, and the kilometer markers of the data points as labels, the first initial model and the second initial model are trained respectively to obtain the first trained model and the second trained model. The first trained model and the second trained model are then integrated to obtain the integrated model; the location features include longitude, latitude, and elevation. The integrated model is used to predict the kilometer marker at any point along the train route; Specifically, the location features, distance features, and elevation change features of the data points are used as inputs, and the kilometer markers of the data points are used as labels. A first initial model and a second initial model are trained respectively to obtain a first trained model and a second trained model. The first trained model and the second trained model are then integrated to obtain an integrated model, which specifically includes: All the data points are divided into a training set and a validation set; Using the location features, distance features, and elevation change features of the data points in the training set as input, and the kilometer markers of the data points in the training set as labels, the first initial model and the second initial model are trained respectively to obtain the first trained model and the second trained model. Using the location features, distance features, and elevation change features of the data points in the validation set as input, the first predicted kilometer markers of the data points in the validation set are determined using the first trained model. Based on the first predicted kilometer markers and the kilometer markers of the data points in the validation set, the mean absolute error is calculated to obtain the first prediction error of the first trained model. Using the location features, distance features, and elevation change features of the data points in the validation set as input, the second predicted kilometer markers of the data points in the validation set are determined using the second trained model. Based on the second predicted kilometer markers and the kilometer markers of the data points in the validation set, the mean absolute error is calculated to obtain the second prediction error of the second trained model. Based on the first prediction error and the second prediction error, the first weight of the first trained model and the second weight of the second trained model are calculated. Based on the first weight and the second weight, the first trained model and the second trained model are integrated to obtain an integrated model; The formula for calculating the first weight is: ； in, It is the first weight; This is the first prediction error; This is the second prediction error; The formula for calculating the second weight is: ； in, It is the second weight.

2. The railway kilometer marker prediction method based on geospatial data according to claim 1, characterized in that, The railway surveying data is processed to calculate the kilometer markers for each surveying point, resulting in geospatial data, specifically including: Duplicate initial data points are deleted from the railway surveying data to obtain initial geospatial data; the initial data points include the starting point, the ending point, and the surveying point. For each survey point in the initial geospatial data, the projection parameters from the survey point to each line segment are calculated. For each line segment, the three-dimensional coordinate vector of the projection point of the survey point onto the line segment is calculated based on the projection parameters. The shortest arc distance from the survey point to the line segment is calculated based on the three-dimensional coordinate vector of the projection point. The line segment corresponding to the minimum value of the shortest arc distance is selected as the nearest line segment. Based on the three-dimensional coordinate vector of the projection point of the survey point onto the nearest line segment, the kilometer marker of the survey point is determined to calculate the kilometer marker of each survey point, thus obtaining geospatial data.

3. The railway kilometer marker prediction method based on geospatial data according to claim 2, characterized in that, Calculating the projection parameters from the survey point to each line segment specifically includes: for each line segment, based on the longitude and latitude of the survey point, the longitude and latitude of the starting point of the line segment, and the longitude and latitude of the ending point of the line segment, calculating the three-dimensional coordinate vector of the survey point, the three-dimensional coordinate vector of the starting point of the line segment, and the three-dimensional coordinate vector of the ending point of the line segment; calculating the difference between the three-dimensional coordinate vector of the survey point and the three-dimensional coordinate vector of the starting point of the line segment to obtain a first vector difference; calculating the difference between the three-dimensional coordinate vector of the ending point of the line segment and the three-dimensional coordinate vector of the starting point of the line segment to obtain a second vector difference; and calculating the projection parameters from the survey point to the line segment based on the first vector difference and the second vector difference. The calculation of the three-dimensional coordinate vector of the projection point of the survey point onto the line segment based on the projection parameters specifically includes: calculating the product of the projection parameters and the difference of the second vector, and calculating the sum of the three-dimensional coordinate vector of the starting point of the line segment and the product, to obtain the three-dimensional coordinate vector of the projection point of the survey point onto the line segment. The shortest arc distance from the surveying point to the line segment is calculated based on the three-dimensional coordinate vector of the projection point. Specifically, this includes: calculating the angle between the surveying point and the projection point based on the three-dimensional coordinate vector of the surveying point and the three-dimensional coordinate vector of the projection point; and calculating the product of the Earth's radius and the angle to obtain the shortest arc distance from the surveying point to the line segment.

4. The railway kilometer marker prediction method based on geospatial data according to claim 1, characterized in that, For each data point in the geospatial data, based on the longitude and latitude of the data point, the great spherical distance from the data point to the center point of the train running line is calculated to obtain the distance characteristics of the data point. Based on the elevation of the data point, the elevation change characteristics of the data point are calculated, specifically including: The geospatial data is interpolated using a natural neighborhood interpolation algorithm to obtain multiple interpolation points, and the longitude, latitude, elevation, and kilometer marker of each interpolation point are determined. For each interpolated data point in the geospatial data, the great spherical distance from the interpolated data point to the center point of the train running line is calculated based on the longitude and latitude of the interpolated data point to obtain the distance characteristics of the interpolated data point. Based on the elevation of the interpolated data point, the elevation change characteristics of the interpolated data point are calculated. The interpolated data points include the start point, end point, survey point, and interpolation point in the geospatial data.

5. The railway kilometer marker prediction method based on geospatial data according to claim 1, characterized in that, Based on the longitude and latitude of the data points, the great spherical distance from the data points to the center point of the train running line is calculated to obtain the distance characteristics of the data points. Based on the elevation of the data points, the elevation variation characteristics of the data points are calculated, specifically including: Using the longitude and latitude of the data points and the longitude and latitude of the center point of the train line as input, the great sphere distance from the data points to the center point of the train line is calculated using the spherical trigonometry formula to obtain the distance characteristics of the data points. If the data point is the first data point, then the elevation change characteristic of the data point is set to be equal to the elevation of the data point, and the elevation change characteristic of the data point is obtained; otherwise, the elevation change characteristic of the data point is set to be equal to the difference between the elevation of the data point and the elevation of the previous data point, and the elevation change characteristic of the data point is obtained.

6. The railway kilometer marker prediction method based on geospatial data according to claim 1, characterized in that, The first initial model is a feedforward neural network model, and the second initial model is a wide learning system model. The wide learning system model includes an input layer, a feature node layer, an enhancement module, and an output layer connected in sequence. The enhancement module includes a first enhancement layer, a second enhancement layer, and a third enhancement layer. The input terminals of the first enhancement layer, the second enhancement layer, and the third enhancement layer are all connected to the output terminal of the feature node layer. The output terminals of the feature node layer, the first enhancement layer, the second enhancement layer, and the third enhancement layer are all connected to the input terminal of the output layer. The first enhancement layer is an enhancement layer using a linear transformation and the ReLU activation function, the second enhancement layer is an enhancement layer using a linear transformation and the Sigmoid activation function, and the third enhancement layer is an enhancement layer using a linear transformation and the exponential activation function.

7. The railway kilometer marker prediction method based on geospatial data according to claim 1, characterized in that, Predicting the kilometer marker at any point along the train route using the ensemble model specifically includes: Based on the longitude and latitude of any point on the train line, the great spherical distance from the point to the center point of the train line is calculated to obtain the distance characteristics of the point. Based on the elevation of the point, the elevation change characteristics of the point are calculated. Using the location features, distance features, and elevation change features of the location point as input, the ensemble model is used to determine the predicted kilometer marker of the location point.

8. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and capable of running on the processor, characterized in that the processor executes the computer program to implement the railway kilometer marker prediction method based on geospatial data as described in any one of claims 1-7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the railway kilometer marker prediction method based on geospatial data as described in any one of claims 1-7.

Images