Intelligent tunnel bias voltage identification method based on artificial intelligence algorithm

Through the intelligent tunnel bias determination method based on artificial intelligence algorithm, the problem of time-consuming and labor-intensive bias determination in tunnel survey and design is solved, automatic, fast and accurate bias determination is achieved, and design efficiency is improved and intelligence.

CN118940024APending Publication Date: 2024-11-12CHINA RAILWAY DESIGN GRP CO LTD +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410814753.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-06-24
Publication Date
2024-11-12

AI Technical Summary

Technical Problem

During the tunnel survey and design process, the identification of biased tunnels is time-consuming and labor-intensive and subjective, which leads to an increase in the possibility of misjudgment and affects the design efficiency and degree of intelligence.

Method used

Using the intelligent tunnel bias determination method based on artificial intelligence algorithm, the tunnel bias situation is marked, data definition, combined and merge bias position and degree, divide training sets and test sets, use machine learning algorithms, convolutional neural networks and Point cloud network conducts intelligent judgment to achieve automatic, fast and accurate bias judgment.

Benefits of technology

It improves design efficiency, reduces the possibility of misjudgment of subjective judgments by designers, and improves the degree of intelligence in the railway tunnel field, which is in line with the current intelligent trend.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118940024A_ABST
    Figure CN118940024A_ABST
Patent Text Reader

Abstract

The invention discloses a tunnel bias voltage intelligent identification method based on an artificial intelligence algorithm, and the method comprises the following steps: marking a tunnel bias voltage condition based on line data, and recording tunnel bias voltage data; performing datamation definition on the tunnel bias voltage data in each paragraph: combining and merging bias voltage positions and bias voltage degrees in the tunnel bias voltage data to obtain bias voltage attributes of the paragraph; obtaining data of all paragraphs of the tunnel, and dividing into a training set and a test set; structured data bias voltage intelligent identification based on a machine learning algorithm is carried out; performing tunnel cross section black and white image bias intelligent identification based on the convolutional neural network; tunnel plane RGB image bias voltage intelligent identification based on the convolutional neural network is carried out; and tunnel 3D point cloud bias voltage intelligent identification based on the point cloud network is carried out. The method can effectively improve the design efficiency, reduces the misjudgment possibility when designers subjectively judge whether the bias voltage exists or not, and accords with the overall trend of intelligence in the current railway field.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention belongs to the technical field of tunnel investigation and design, and in particular relates to an intelligent identification method for tunnel bias voltage based on an artificial intelligence algorithm. Background Art

[0002] During the tunnel survey and design process, affected by the terrain and line, unfavorable terrain conditions such as eccentric pressure will inevitably appear in the tunnel portal section. Due to the thin overlying soil layer and complex stress on the surrounding rock and lining structure, eccentric pressure tunnels often experience disasters such as overall slope sliding and surface collapse during construction.

[0003] On the one hand, the railway tunnel design specifications stipulate that "the vertical distance t from the outer spandrel of the tunnel to the ground surface is used to determine whether the tunnel is designed as a bias tunnel". The judgment requires distance measurement and formula calculation. It is time-consuming and laborious to judge the bias of multiple design sections of a design section. On the other hand, different designers have different experiences and backgrounds, and the judgment of bias is subjective. For the same terrain, one designer may judge that there is no bias, while another designer may judge that there is a bias, which leads to misjudgment of whether there is a bias.

[0004] Therefore, it is necessary to propose a tunnel bias pressure judgment method that is convenient, fast, accurate, intelligent, and has low human subjectivity.

[0005] With the rapid rise of artificial intelligence technology, it has penetrated into all aspects of society, and has also set off a wave of intelligence in the railway field. Intelligentization in the tunnel field is the development direction of the railway field and an inevitable trend in the development of railway engineering construction under the new situation. Therefore, the artificial intelligence algorithm is applied to the identification of tunnel bias. Designers only need to transfer terrain data to the tunnel bias intelligent identification system to automatically, quickly and accurately determine its bias properties, improve design efficiency, reduce the possible errors in the subjective judgment of different designers on whether there is bias, and at the same time improve the current level of intelligence in the field of railway tunnels, providing conditions for realizing intelligent design in the tunnel field. Summary of the invention

[0006] The present invention is proposed to solve the problems existing in the prior art, and its purpose is to provide a tunnel bias voltage intelligent identification method based on artificial intelligence algorithm.

[0007] The technical solution of the present invention is: a tunnel bias voltage intelligent identification method based on an artificial intelligence algorithm, comprising the following steps:

[0008] A. Based on the line data, mark the tunnel bias situation and record the tunnel bias data;

[0009] B. Define the tunnel bias data in each section:

[0010] C. Combining and merging the bias position and bias degree in the tunnel bias data to obtain the bias attribute of the section;

[0011] D. Obtain data from all sections of the tunnel and divide them into training sets and test sets;

[0012] E. Conduct intelligent identification of structured data bias based on machine learning algorithms;

[0013] F. Conduct intelligent identification of bias voltage of black and white images of tunnel cross sections based on convolutional neural networks;

[0014] G. Conduct intelligent bias identification of tunnel plane RGB images based on convolutional neural network;

[0015] H. Perform intelligent identification of tunnel 3D point cloud bias pressure based on point cloud network.

[0016] Furthermore, step A marks the tunnel bias condition based on the line data and records the tunnel bias data. The specific process is as follows:

[0017] First, in the 3D GIS-based road, bridge and tunnel 3D scheme design software, import the line data into the GIS terrain model;

[0018] Then, a tunnel outline of fixed size is selected and the 3D model of the tunnel structure is placed at each mileage position of the line;

[0019] Then, by observing the relationship between the tunnel structure and the terrain, mark the tunnel bias situation, record the tunnel bias data, and export it into an XML format file.

[0020] Furthermore, the tunnel bias voltage data is used as original sample data for intelligent identification of tunnel bias voltage.

[0021] Furthermore, step B defines the tunnel bias data in each section in a digital manner, and the specific process is as follows:

[0022] First, the exported XML file contains multiple <learndata>Data, each <learndata>The data contains the data information in the same paragraph, and the data information is defined in a digital way;

[0023] Then, <biasside>is the bias position, which has three categories: no bias, left bias, and right bias;

[0024] After that, <biasdegree>The bias level has three categories: no bias, great bias, and small bias;

[0025] After that, <tunnelwidth>is the tunnel design width, in m;

[0026] After that, <tunnelheight>is the tunnel design height, in m;

[0027] After that, <pointspace>The horizontal spacing between two adjacent elevation points is 1m by default;

[0028] at last, <pointmatrix>This is the elevation data of the section. Each section contains elevation data of multiple cross sections. <string>, each <string>Contains the elevation data of different positions in sequence from the left to the right of the cross section, in meters, and the horizontal spacing of each data point is 1 meter.

[0029] Furthermore, step C combines the bias position and bias degree in the tunnel bias data to obtain the bias attribute of the section. The specific process is as follows:

[0030] First, change the <biasside>Bias position and <biasdegree>The two attributes of bias degree are combined and merged to obtain the bias attribute of the paragraph;

[0031] Then, the obtained bias attribute is used as the output label. There are five categories of bias attributes: no bias, large left bias, small left bias, large right bias, and small right bias. They are represented by numerical values, corresponding to 0, 1, 2, 3, and 4 respectively.

[0032] Furthermore, step D obtains the data of all sections of the tunnel and divides them into training sets and test sets. The specific process is as follows:

[0033] First, the data of all sections of the tunnel are obtained based on the above software;

[0034] Then, the original sample data is divided into training set and test set in a ratio of 0.9:0.1.

[0035] Furthermore, step E performs intelligent identification of structured data bias based on machine learning algorithms. The specific process is as follows:

[0036] e1. Establish a data set based on machine learning algorithms;

[0037] e2. Establish a structured data bias intelligent identification algorithm model based on machine learning algorithms. The algorithm model uses five machine learning algorithms: linear discriminant analysis (LDA), K-nearest neighbor method (KNN), decision tree (CART), naive Bayes (NB) and support vector machine (SVM);

[0038] e3. Use the dataset to train and test different machine learning algorithms;

[0039] e4. The decision tree (CART) has the best effect, the fastest training speed, and the test set accuracy can reach more than 99%.

[0040] Furthermore, the data set includes a single sample input, and the single sample input is obtained as follows:

[0041] First, the elevation data values ​​of x cross sections in each paragraph and y points in each cross section in the XML file are merged and sorted in order into x*y dimensional elevation data values;

[0042] Then, the x*y dimensional elevation data value is used as input for a single sample.

[0043] Furthermore, the data set includes output labels, and the output labels are obtained as follows:

[0044] The bias attribute value corresponding to each paragraph is used as the output label.

[0045] Furthermore, the data set is written into an Excel file, where each row represents a paragraph and a sample. Each sample includes x*y+1 data, the first x*y data are input, and the last data is output, forming structured elevation data.

[0046] The beneficial effects of the present invention are as follows:

[0047] From the perspectives of different data forms and different artificial intelligence algorithms, the present invention proposes four terrain data processing methods and their corresponding bias intelligent identification methods, including: a structured data bias intelligent identification method based on a machine learning algorithm, a tunnel cross-section black and white image bias intelligent identification method based on a convolutional neural network, a tunnel plane RGB image bias intelligent identification method based on a convolutional neural network, and a tunnel 3D point cloud bias intelligent identification method based on a point cloud network.

[0048] The four terrain data processing methods and the corresponding bias pressure intelligent identification methods provided by the present invention can effectively improve the design efficiency, reduce the possibility of misjudgment by designers when making subjective judgments on whether there is bias pressure, and improve the level of intelligence in the current railway tunnel field, which is in line with the overall trend of intelligence in the current railway field. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 is a system architecture diagram of the present invention;

[0050] Figure 2 It is a schematic diagram of the XML file in the present invention;

[0051] Figure 3 is a schematic diagram of a structured data set in the present invention;

[0052] Figure 4 It is a black and white picture of the tunnel cross section in the present invention;

[0053] Figure 5 It is the improved VGG16 model diagram in the present invention;

[0054] Figure 6 is the tunnel plane RGB map in the present invention;

[0055] Figure 7 is the three-dimensional coordinates of the tunnel section point cloud in the present invention;

[0056] Figure 8 It is the 3D point cloud visualization diagram in the present invention;

[0057] Fig. 9 It is the pointnet model diagram in the present invention. DETAILED DESCRIPTION

[0058] Hereinafter, the present invention will be described in detail with reference to the accompanying drawings and embodiments:

[0059] like Figures 1 to 9 As shown, the tunnel bias voltage intelligent identification method based on artificial intelligence algorithm includes the following steps:

[0060] A. Based on the line data, mark the tunnel bias situation and record the tunnel bias data;

[0061] B. Define the tunnel bias data in each section:

[0062] C. Combining and merging the bias position and bias degree in the tunnel bias data to obtain the bias attribute of the section;

[0063] D. Obtain data from all sections of the tunnel and divide them into training sets and test sets;

[0064] E. Conduct intelligent identification of structured data bias based on machine learning algorithms;

[0065] F. Conduct intelligent identification of bias voltage of black and white images of tunnel cross sections based on convolutional neural networks;

[0066] G. Conduct intelligent bias identification of tunnel plane RGB images based on convolutional neural network;

[0067] H. Perform intelligent identification of tunnel 3D point cloud bias pressure based on point cloud network.

[0068] Step A marks the tunnel bias condition based on the line data and records the tunnel bias data. The specific process is as follows:

[0069] First, in the 3D GIS-based road, bridge and tunnel 3D scheme design software, import the line data into the GIS terrain model;

[0070] Then, a tunnel outline of fixed size is selected and the 3D model of the tunnel structure is placed at each mileage position of the line;

[0071] Then, by observing the relationship between the tunnel structure and the terrain, mark the tunnel bias situation, record the tunnel bias data, and export it into an XML format file.

[0072] The tunnel bias voltage data is used as original sample data for intelligent identification of tunnel bias voltage.

[0073] Specifically, the method for extracting the biased terrain feature data of the tunnel in step A is to select the mileage range of the tunnel entrance, set the sampling interval of the terrain feature data (such as 1m), obtain the terrain cross-section according to the sampling interval along the direction of the railway line, and the left and right side range of the cross-section is not less than 10m. The elevation of the ground point is obtained along the cross-section according to the sampling interval, and the difference between the ground point elevation and the line elevation is calculated. All the differences are arranged in order to form the biased terrain feature data of the tunnel, including but not limited to storage in the xml file format.

[0074] Step B defines the tunnel bias data in each section in a digital way. The specific process is as follows:

[0075] First, the exported XML file contains multiple <learndata>Data, each <learndata>The data contains the data information in the same paragraph, and the data information is defined in a digital way;

[0076] Then, <biasside>is the bias position, which has three categories: no bias, left bias, and right bias;

[0077] After that, <biasdegree>The bias level has three categories: no bias, great bias, and small bias;

[0078] After that, <tunnelwidth>is the tunnel design width, in m;

[0079] After that, <tunnelheight>is the tunnel design height, in m;

[0080] After that, <pointspace>The horizontal spacing between two adjacent elevation points is 1m by default;

[0081] at last, <pointmatrix>This is the elevation data of the section. Each section contains elevation data of multiple cross sections. <string>, each <string>Contains the elevation data of different positions in sequence from the left to the right of the cross section, in meters, and the horizontal spacing of each data point is 1 meter.

[0082] Step C combines the bias position and bias degree in the tunnel bias data to obtain the bias attribute of the section. The specific process is as follows:

[0083] First, change the <biasside>Bias position and <biasdegree>The two attributes of bias degree are combined and merged to obtain the bias attribute of the paragraph;

[0084] Then, the obtained bias attribute is used as the output label. There are five categories of bias attributes: no bias, large left bias, small left bias, large right bias, and small right bias. They are represented by numerical values, corresponding to 0, 1, 2, 3, and 4 respectively.

[0085] Step D obtains the data of all sections of the tunnel and divides them into training sets and test sets. The specific process is as follows:

[0086] First, the data of all sections of the tunnel are obtained based on the above software;

[0087] Then, the original sample data is divided into training set and test set in a ratio of 0.9:0.1.

[0088] Step E performs intelligent identification of structured data bias based on machine learning algorithms. The specific process is as follows:

[0089] e1. Establish a data set based on machine learning algorithms;

[0090] e2. Establish a structured data bias intelligent identification algorithm model based on machine learning algorithms. The algorithm model uses five machine learning algorithms: linear discriminant analysis (LDA), K-nearest neighbor method (KNN), decision tree (CART), naive Bayes (NB) and support vector machine (SVM);

[0091] e3. Use the dataset to train and test different machine learning algorithms;

[0092] e4. The decision tree (CART) has the best effect, the fastest training speed, and the test set accuracy can reach more than 99%.

[0093] The data set includes a single sample input, and the single sample input is obtained as follows:

[0094] First, the elevation data values ​​of x cross sections in each paragraph and y points in each cross section in the XML file are merged and sorted in order into x*y dimensional elevation data values;

[0095] Then, the x*y dimensional elevation data value is used as input for a single sample.

[0096] The data set includes output labels, and the output labels are obtained as follows:

[0097] The bias attribute value corresponding to each paragraph is used as the output label.

[0098] Write the data set into an Excel file. Each row represents a paragraph and a sample. Each sample includes x*y+1 data. The first x*y data are input and the last data is output, forming structured elevation data.

[0099] Specifically, step F performs intelligent bias identification of black and white images of tunnel cross sections based on convolutional neural networks, including generating black and white images of tunnel cross sections. Generating black and white images of tunnel cross sections includes: drawing the elevation values ​​of each cross section into a black and white line graph, and exporting a black and white image .png file. The image represents the terrain information of a single cross section, and together with the bias attributes of the section where the cross section is located, constitutes the input and output of a sample.

[0100] Specifically, step F performs intelligent identification of bias voltage of black and white images of tunnel cross sections based on convolutional neural networks, and the specific process is as follows:

[0101] f1. Establish a data set based on convolutional neural network.

[0102] Draw the y elevation data points of each cross section in the xml file into a line graph, with the spacing between adjacent points being 1m. The middle position elevation is the ground elevation of the vertex of the left track of the line. Export it as a .png file. The image is the input of the sample, and the bias attribute of the section where the cross section is located is used as the output.

[0103] Each black and white image represents a cross section and a sample, resulting in a total sample.

[0104] f2. Establish an intelligent bias voltage identification algorithm model for black and white images of tunnel cross sections based on convolutional neural networks.

[0105] The common convolutional neural network VGG16 model is improved. The input size of the cross-sectional grayscale image is 256*512*3. After multiple convolutional layers, pooling layers, and fully connected layers, an improved VGG network model is formed. The output layer size is 5*1, which represents the prediction of five bias attributes.

[0106] After the training is completed, the test set is tested and its bias identification accuracy is about 90%.

[0107] Specifically, step G performs intelligent bias identification of tunnel plane RGB image based on convolutional neural network, including generating tunnel plane RGB image, wherein the generating tunnel plane RGB image includes: drawing the elevation data of multiple cross sections of each section into a plane RGB image of the section, the plane coordinates of the image pixels represent different cross section position orders and position orders of different elevation points of the cross section, and the different colors of the pixels represent different elevation values, so the RGB image contains the plane position information and elevation information, which together with the bias attribute of the section constitute the input and output of a sample.

[0108] Specifically, step G performs intelligent bias recognition of the tunnel plane RGB image based on a convolutional neural network, and the specific process is as follows:

[0109] g1. Establish a data set based on convolutional neural network.

[0110] First, each paragraph in the XML file contains x cross sections, each cross section contains y elevation data, and is drawn into a planar RGB image;

[0111] Then, the position of each pixel block in the image is determined by the position order of different cross-sections and the position order of different points in the cross-section;

[0112] Then, the first row of pixel blocks in the figure represents the first cross section in the paragraph, and the third pixel block in the first row represents the third position point on the left side of the first cross section. Different elevation values ​​are assigned different colors;

[0113] Finally, each RGB image represents a paragraph, a sample input, and the biased attributes of the paragraph are output.

[0114] g2. Establish a tunnel plane RGB image bias intelligent identification algorithm model based on convolutional neural network, using an improved VGG model that is the same as the model in step 62, and its bias identification accuracy can reach 99%.

[0115] Specifically, step H performs intelligent identification of tunnel 3D point cloud bias based on point cloud network, including generating tunnel 3D point cloud, wherein generating tunnel 3D point cloud includes: writing multiple cross-sectional elevation data and plane coordinates contained in a paragraph into a txt file, the file contains three columns of data coordinates, namely plane coordinates and elevation, and each txt file and the corresponding bias attribute together constitute the input and output of a sample.

[0116] Specifically, step H performs intelligent identification of the bias pressure of the tunnel 3D point cloud based on the point cloud network, and the specific process is as follows:

[0117] h1. Establish a dataset based on point cloud network.

[0118] First, write the elevation data of x cross sections contained in each paragraph of the XML file and their pixel square plane coordinates into the txt file;

[0119] Then, the first column in the file indicates the order of the location points starting from the left in the same cross section, the second column indicates the order of the cross sections in the paragraph, and the third column indicates the elevation of the corresponding point;

[0120] After that, each txt file represents the input of a paragraph and a sample, and the bias attribute of the paragraph is output;

[0121] Finally, import the point cloud txt file into Cloud Compare software for point cloud visualization.

[0122] h2. Establish a tunnel plane RGB image bias intelligent identification algorithm model based on convolutional neural network.

[0123] First, the point cloud network pointnet model is used for point cloud classification. The model input dimension is xy*3, which represents the three coordinate values ​​of the xy points in the paragraph. After passing through multiple neural networks, the output size is 5*1, which represents the prediction of five bias attributes.

[0124] Then, after training, the test set was tested and its bias identification accuracy was about 75%.

[0125] Embodiment 1

[0126] The intelligent identification method of tunnel bias voltage based on artificial intelligence algorithm includes the following steps:

[0127] Step A. In the three-dimensional road, bridge and tunnel design software based on three-dimensional GIS software, import the line data into the GIS terrain model, select the tunnel outline of a fixed size, and place the three-dimensional model of the tunnel structure at each mileage position of the line. By observing the relationship between the tunnel structure and the terrain, mark the tunnel bias situation, and record the tunnel bias data and export it to an XML format file as the original sample data for intelligent identification of tunnel bias.

[0128] Step B. The exported XML file content is as follows Figure 2 As shown, the file contains multiple <learndata>Data, each <learndata>The data includes the location, elevation and bias attributes within the same paragraph, which are:

[0129] b1. <biasside>For bias position, there are three categories: no bias, left bias, and right bias;

[0130] b2. <biasdegree>For the degree of bias, there are three categories: no bias, very large bias, and small bias;

[0131] b3. <tunnelwidth>is the tunnel design width, in m;

[0132] b4. <tunnelheight>is the tunnel design height, in m;

[0133] b5. <pointspace>The horizontal spacing between two adjacent elevation points is 1m by default;

[0134] b6. <pointmatrix>This is the elevation data of the section. Each section contains elevation data of multiple cross sections. <string>, each <string>Contains the elevation data of different positions in sequence from the left to the right of the cross section, in meters, and the horizontal spacing of each data point is 1 meter.

[0135] Step C. <biasside>Bias position and <biasdegree>The two attributes of bias degree are combined and merged to obtain the bias attribute of the paragraph, which is the output label of the present invention. There are five categories: no bias, large left bias, small left bias, large right bias, and small right bias. They are expressed as numerical values ​​corresponding to 0, 1, 2, 3, and 4 respectively.

[0136] Step D. Based on the above design software, a total of about 13,000 tunnel sections are marked, each section contains elevation data of 11 cross sections, and each cross section contains 101 data points with a total length of 100m from left to right. The above sample data are divided into training set and test set according to the ratio of 0.9:0.1.

[0137] Step E. Intelligent identification method of structured data bias based on machine learning algorithm.

[0138] e1. Create a data set, merge the elevation data values ​​of 11 cross sections in each paragraph and 101 points in each cross section in the XML file, and sort them in order into 11*101=1111-dimensional elevation data values ​​as the input of a single sample, and use the bias attribute value corresponding to each paragraph as the output label, totaling about 13,000 samples.

[0139] Then write these data into an Excel file, each row represents a paragraph, a sample, each sample includes 1112 data, the first 1111 data are input, and the last data is output, forming structured elevation data, such as Figure 3 shown.

[0140] e2. Algorithm model, using five machine learning algorithms, including linear discriminant analysis (LDA), K-nearest neighbor method (KNN), decision tree (CART), naive Bayes (NB) and support vector machine (SVM), for training and testing respectively. Among them, decision tree (CART) has the best effect and the fastest training speed, and the accuracy of the test set can reach more than 99%.

[0141] Step F. Intelligent bias identification method of black and white images of tunnel cross-section based on convolutional neural network.

[0142] f1. Create a data set and plot the 101 elevation data points of each cross section in the xml file into a line graph. The distance between adjacent points is 1m. The 51st elevation is the ground elevation of the vertex of the left track of the line. Export it to a .png file, such as Figure 4 As shown, the image is the input of the sample, and the bias attribute of the section where the cross section is located is the output. Each black and white image represents a cross section and a sample, totaling about 13000*11=143000 samples.

[0143] f2. Algorithm model, improve the common convolutional neural network VGG16 model, the cross-sectional grayscale image input size is 256*512*3, after multiple convolution layers, pooling layers, and fully connected layers, to form an improved VGG network model, such as Figure 5 As shown in the figure, the output layer size is 5*1, which indicates the prediction of five bias attributes. After training, the bias identification accuracy is about 90% when tested on the test set.

[0144] Step G. Intelligent bias identification method of tunnel plane RGB image based on convolutional neural network.

[0145] g1. Create a data set and draw the 11 cross sections contained in each paragraph of the XML file into a planar RGB image, where each cross section contains 101 elevation data, such as Figure 6 As shown, the position of each pixel block in the figure is determined by the position order of different cross-sections and the position order of different points in the cross-section.

[0146] like Figure 6 The first row of pixels in the image represents the first cross section in the paragraph, and the third pixel in the first row represents the third position point on the left side of the first cross section. Different elevation values ​​are assigned different colors. Each RGB image represents the input of a paragraph and a sample, and the "bias attribute" of the paragraph is output, totaling about 13,000 samples.

[0147] g2. Algorithm model, using the improved VGG model which is the same as the model in step 62, its bias recognition accuracy can reach 99%.

[0148] Step H. This method for intelligently identifying bias pressure of 3D point cloud of tunnel based on point cloud network.

[0149] h1. Create a data set and write the elevation data of the 11 cross sections contained in each paragraph of the XML file and their pixel block plane coordinates into a txt file, such as Figure 7 As shown in the figure, the first column in the file indicates the order of the position points starting from the left in the same cross section, the second column indicates the order of the cross sections in the paragraph, and the third column indicates the elevation of the corresponding point. Each txt file represents the input of a paragraph and a sample, and the "bias attribute" of the paragraph is used as the output, totaling about 13,000 samples. Import the point cloud txt file into the Cloud Compare software for point cloud visualization, as shown in the figure below. Figure 8 shown.

[0150] h2. Algorithm model, using the point cloud network pointnet model for point cloud classification. The network architecture of this model is as follows Fig. 9 As shown in the figure, the model input dimension is 1111*3, which means the three coordinate values ​​of the 1111 points in the paragraph. After passing through multiple neural networks, the output dimension is 5*1, which means the prediction of five bias attributes. After training, the model is tested on the test set, and the bias identification accuracy is about 75%.

[0151] Step i. After training and testing the four algorithms, the accuracy is shown in Table 1.

[0152] Table 1 Test accuracy of four bias intelligent identification methods

[0153]

[0154] i1. As can be seen from the table: the first and third methods have the highest accuracy, reaching more than 99%, indicating that these two methods can automatically and accurately determine the bias properties of tunnel sections, and the training time of the first method is much less than that of the third method. Therefore, the first method "structured data bias intelligent identification method based on machine learning algorithm" can be used for actual survey and design tasks.

[0155] i2. As can be seen from the table: the second method is based on cross-section intelligent bias identification, and its accuracy is average, about 85%. This may be because the bias properties of the entire section are marked, rather than the bias properties of each cross section, resulting in the annotation of each cross section may not be accurate enough. Therefore, the accuracy of the method based on the cross-section black and white image may be limited, but it can achieve cross-section-level, more precise tunnel bias identification, so it also has certain practicality in actual survey and design.

[0156] i3. As can be seen from the table: the fourth method has poor accuracy and cannot accurately identify the bias properties. This may be due to problems such as sparse point clouds. Therefore, it is not recommended for use in survey and design.

[0157] The four intelligent identification methods for tunnel bias voltage based on artificial intelligence algorithms proposed in the present invention each have their own advantages and can effectively perform intelligent identification of tunnel bias voltage.

[0158] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.< / biasdegree> < / biasside> < / string> < / string> < / pointmatrix> < / pointspace> < / tunnelheight> < / tunnelwidth> < / biasdegree> < / biasside> < / learndata> < / learndata> < / biasdegree> < / biasside> < / string> < / string> < / pointmatrix> < / pointspace> < / tunnelheight> < / tunnelwidth> < / biasdegree> < / biasside> < / learndata> < / learndata> < / biasdegree> < / biasside> < / string> < / string> < / pointmatrix> < / pointspace> < / tunnelheight> < / tunnelwidth> < / biasdegree> < / biasside> < / learndata> < / learndata>

Claims

1. Intelligent identification method of tunnel bias pressure based on artificial intelligence algorithm, Features: The following steps are involved: A. Based on the line data, mark the tunnel bias situation and record the tunnel bias data; B. Define the tunnel bias data in each section: C. Combining and merging the bias position and bias degree in the tunnel bias data to obtain the bias attribute of the section; D. Obtain data from all sections of the tunnel and divide them into training sets and test sets; E. Conduct intelligent identification of structured data bias based on machine learning algorithms; F. Conduct intelligent identification of bias voltage of black and white images of tunnel cross sections based on convolutional neural networks; G. Conduct intelligent bias identification of tunnel plane RGB images based on convolutional neural network; H. Perform intelligent identification of tunnel 3D point cloud bias pressure based on point cloud network.

2. According to claim 1, the tunnel bias voltage intelligent identification method based on artificial intelligence algorithm, Features: Step A marks the tunnel bias condition based on the line data and records the tunnel bias data. The specific process is as follows: First, in the 3D GIS-based road, bridge and tunnel 3D scheme design software, import the line data into the GIS terrain model; Then, a tunnel outline of fixed size is selected and the 3D model of the tunnel structure is placed at each mileage position of the line; Then, by observing the relationship between the tunnel structure and the terrain, mark the tunnel bias situation, record the tunnel bias data, and export it into an XML format file.

3. According to claim 2, the tunnel bias voltage intelligent identification method based on artificial intelligence algorithm, Features: The tunnel bias voltage data is used as original sample data for intelligent identification of tunnel bias voltage.

4. According to claim 1, the tunnel bias voltage intelligent identification method based on artificial intelligence algorithm, Features: Step B defines the tunnel bias data in each section in a digital way. The specific process is as follows: First, the exported XML file contains multiple <learndata>Data, each <learndata> The data contains the data information within the same paragraph, and defines the data information in a digitized manner.< / learndata> < / learndata> Then, <biasside> is the bias position, which has three categories: no bias, left bias, and right bias;< / biasside> After that, <biasdegree> The bias level has three categories: no bias, great bias, and small bias;< / biasdegree> After that, <tunnelwidth> is the tunnel design width, in m;< / tunnelwidth> After that, <tunnelheight> is the tunnel design height, in m;< / tunnelheight> After that, <pointspace> The horizontal spacing between two adjacent elevation points is 1m by default;< / pointspace> at last, <pointmatrix>This is the elevation data of the section. Each section contains elevation data of multiple cross sections. <string>, each <string> Contains the elevation data of different positions in sequence from the left to the right of the cross section, in meters, and the horizontal spacing of each data point is 1 meter.< / string> < / string> < / pointmatrix> 5. According to claim 1, the tunnel bias voltage intelligent identification method based on artificial intelligence algorithm, Features: Step C combines the bias position and bias degree in the tunnel bias data to obtain the bias attribute of the section. The specific process is as follows: First, change the <biasside>Bias position and <biasdegree> The two attributes of bias degree are combined and merged to obtain the bias attribute of the paragraph;< / biasdegree> < / biasside> Then, the obtained bias attribute is used as the output label. There are five categories of bias attributes: no bias, large left bias, small left bias, large right bias, and small right bias. They are represented by numerical values, corresponding to 0, 1, 2, 3, and 4 respectively.

6. The tunnel bias voltage intelligent identification method based on artificial intelligence algorithm according to claim 1, Features: Step D obtains the data of all sections of the tunnel and divides them into training sets and test sets. The specific process is as follows: First, the data of all sections of the tunnel are obtained based on the above software; Then, the original sample data is divided into training set and test set in a ratio of 0.9:0.

1.

7. The tunnel bias voltage intelligent identification method based on artificial intelligence algorithm according to claim 1, Features: Step E performs intelligent identification of structured data bias based on machine learning algorithms. The specific process is as follows: e1. Establish a data set based on machine learning algorithms; e2. Establish a structured data bias intelligent identification algorithm model based on machine learning algorithms. The algorithm model uses five machine learning algorithms: linear discriminant analysis (LDA), K-nearest neighbor method (KNN), decision tree (CART), naive Bayes (NB) and support vector machine (SVM); e3. Use the dataset to train and test different machine learning algorithms; e4. The decision tree (CART) has the best effect, the fastest training speed, and the test set accuracy can reach more than 99%.

8. The method for intelligently identifying tunnel bias voltage based on artificial intelligence algorithm according to claim 7, Features: The data set includes a single sample input, and the single sample input is obtained as follows: First, the elevation data values ​​of x cross sections in each paragraph and y points in each cross section in the XML file are merged and sorted in order into x*y dimensional elevation data values; Then, the x*y dimensional elevation data value is used as input for a single sample.

9. The method for intelligently identifying tunnel bias voltage based on artificial intelligence algorithm according to claim 8, Features: The data set includes output labels, and the output labels are obtained as follows: The bias attribute value corresponding to each paragraph is used as the output label.

10. The method for intelligently identifying tunnel bias voltage based on artificial intelligence algorithm according to claim 9, Features: Write the data set into an Excel file. Each row represents a paragraph and a sample. Each sample includes x*y+1 data. The first x*y data are input and the last data is output, forming structured elevation data.