Residue soil water content prediction method and system suitable for shield tunnel construction

Abstract

The invention discloses a muck water content prediction method and system suitable for shield tunnel construction, and the method comprises the steps: obtaining first data corresponding to to-be-measured muck in the shield tunnel construction in real time based on a self-adaptive window; in combination with a triangle cumulative sectional area method and an AE-FPFH-ICP algorithm, performing parallel processing to generate second data and third data corresponding to the to-be-measured muck; constructing a PLS-SVR combined model, and according to the PLS-SVR combined model, respectively combining the first data, the second data and the third data, and generating fourth data corresponding to the to-be-measured muck in real time; according to the method, the system and the platform corresponding to the method, high precision and robustness are shown in muck water content detection, the constructed PLS-SVR combined model is excellent in performance when processing a complex nonlinear relation, the water content of muck can be accurately predicted in real time, adaptability is high, and the high-precision requirement of tunnel construction for water content detection is met.

Description

Technical Field

[0001] The invention belongs to the technical field of shield tunnel construction measurement and processing, and in particular relates to a method and system for predicting soil moisture content suitable for shield tunnel construction. Background Art

[0002] In the process of tunnel construction and maintenance, accurate measurement of the volume and moisture content of slag is crucial to ensure construction quality and efficiency. Traditional methods have problems such as low accuracy and poor efficiency, which are difficult to meet the needs of modern engineering.

[0003] In recent years, the application of laser radar (LiDAR) and near-infrared (NIR) spectrometers combined with machine learning technology in these measurements has become increasingly widespread. Yu et al. (2021) proposed a volume measurement method based on a mobile LiDAR system, which accurately obtains the three-dimensional volume information of tunnel debris by processing point cloud data in real time. LiDAR technology can provide extremely accurate volume measurements due to its high resolution and high precision. However, the main disadvantage of this method is the high equipment cost and huge computing resource requirements, especially when processing a large amount of point cloud data in real time, the computational complexity increases sharply. In addition, in complex tunnel environments, LiDAR data acquisition is easily affected by factors such as ambient light and dust, resulting in reduced measurement accuracy.

[0004] In addition, Li et al. (2019) studied the method of calculating the volume of slag soil using optical measurement and three-dimensional reconstruction technology using drones. UAV optical measurement technology is highly flexible and can cover complex terrain and obtain a large amount of data. However, the measurement accuracy of this method is easily affected when the lighting conditions vary greatly or the environment is complex. In addition, the three-dimensional reconstruction process requires processing a large amount of image data, which is complex and requires high computing resources. Zhang et al. (2020) proposed a volume measurement method based on laser scanning technology and triangular mesh method. This method calculates the volume by generating a triangular mesh of the tunnel section, accurately describes complex geometric shapes, and is suitable for volume calculation of irregular surfaces. However, the disadvantage of this method is that the amount of calculation required to generate and process triangular meshes is huge, the quality requirements for point cloud data are high, and data noise or missing may significantly affect the measurement accuracy. Feng et al. (2022) studied a laser scanning method combined with multi-view stereo measurement technology, generating a three-dimensional model by fusing laser scanning data from multiple perspectives for calculating the volume of slag soil. This method can effectively reduce data blind spots and improve measurement accuracy, especially in environments with complex shapes or measurement blind spots. However, the complexity of data acquisition, fusion and processing, as well as the high demand for computing resources, have become a major challenge for this method. The voxel method proposed by Gong and Huang (2020) estimates the volume by dividing the point cloud data into three-dimensional voxel units and calculating the number of filled voxels. The voxel method is simple to operate and suitable for quick estimation, but when dealing with complex geometric shapes, the accuracy is limited by the size of the voxel, and the amount of calculation increases significantly as the voxel unit is reduced. Wang, Wang Fuli and others proposed a cross-sectional area integration method to calculate the volume of belt conveyor slag, but the belt conveyor speed cannot be obtained in real time. The default belt conveyor speed is uniform or the speed is obtained through the encoder on the belt conveyor. The efficiency and automation level of volume measurement need to be improved.

[0005] That is to say, the existing volume measurement methods mainly face the following limitations in technical applications. First, many methods rely on fine grid generation or voxel division when dealing with complex geometric shapes and irregular surfaces, resulting in huge computational complexity and extremely high data quality requirements, which makes these methods susceptible to noise and data loss in practical applications. Secondly, the ability to process a large amount of point cloud data in real time is insufficient, resulting in high computational complexity and poor real-time performance, especially in dynamic environments, making it difficult to achieve accurate volume measurement. Furthermore, some methods rely on multi-view data fusion or multi-device synchronization, which increases the complexity of data processing and the demand for computing resources, limiting its applicability in resource-constrained environments. In addition, existing volume measurement methods generally lack automated processing capabilities, especially in dynamic environments where it is difficult to automatically calculate the moving speed of objects, which in turn affects the overall accuracy and efficiency of volume measurement. These methodological limitations indicate that the existing technology still has a lot of room for improvement in terms of accuracy, real-time performance, robustness, and automation.

[0006] In addition, in terms of soil moisture detection, some scholars have used images to determine soil moisture. Although the image detection method is convenient in determining soil moisture, it is limited by factors such as accuracy, environmental conditions, material heterogeneity, and quantitative ability. This method has certain limitations. Gorji studied the application of handheld NIR spectrometers in determining the moisture content of basil leaves. This study used multivariate analysis methods, especially partial least squares regression (PLS) models, to analyze spectral data, extract the main components related to moisture content, and develop an accurate prediction model. This method can effectively process high-dimensional spectral data and provide accurate moisture content estimates, but its ability to model nonlinear relationships is limited.

[0007] In addition, Gibertoni designed a NIR system for on-site measurement of alfalfa moisture content. The study used the PLS method for data processing and achieved high-precision detection of alfalfa moisture content by analyzing NIR spectral data collected on-site. The PLS method can simplify the data processing process without losing key information through dimensionality reduction and regression analysis. However, since PLS is mainly a linear regression model, its performance may be limited when faced with complex nonlinear data. Aguilar-Castro used NIR reflectance spectroscopy technology when studying the measurement of ceramic gypsum moisture content. This study demonstrated the potential application of NIR spectroscopy in industrial materials, especially in the rapid online detection of moisture content. The NIR reflectance spectroscopy method used in the study was based on the correlation between spectral reflectance and material moisture content. Through multiple measurements and calibrations, an efficient prediction model was established. However, when dealing with different materials and surface characteristics, the reflectance spectroscopy method may be affected by surface roughness, color and gloss, thereby affecting the measurement accuracy. Pu's research focuses on the pharmaceutical field, using NIR spectroscopy technology to non-destructively detect the moisture content in freeze-dried drugs. The study used a spectral analysis method similar to the above, combined with a calibration model to predict the moisture content in drugs. However, since the detection of moisture content in drugs involves complex matrix effects, NIR spectroscopy may require more complex preprocessing and calibration steps when dealing with these complex interferences. Shi Wenqiang used near-infrared spectroscopy to predict soil moisture content in northern cold regions, using the SVM method. SVM performs well in dealing with nonlinear relationships and high-dimensional data, especially in predicting the moisture content of slag soil. It can achieve high accuracy, but SVM has high requirements for parameter tuning, and the kernel function selection and penalty parameter setting may lead to overfitting or underfitting if not carefully. Secondly, SVM has high computational complexity when processing high-dimensional spectral data, and the training process is time-consuming and resource-intensive. In addition, SVM is sensitive to noise and outliers, and is easily affected by a small number of outliers, which in turn affects the prediction accuracy and stability.

[0008] That is to say, although NIR spectroscopy technology has shown application potential in moisture detection, existing studies generally rely on linear models (such as partial least squares regression, PLS) for data analysis. Although these models perform well in processing high-dimensional spectral data and providing high prediction accuracy, they are insufficient in dealing with complex nonlinear relationships and often fail to capture deep patterns and complex interactions in the data. In addition, the adaptability of existing NIR spectroscopy methods to different materials and environmental conditions also has certain limitations, making it difficult to cope with the changing conditions in practical applications. These limitations highlight the shortcomings of existing methods in dealing with nonlinear and complex environments, and it is necessary to explore more powerful combined models.

[0009] Therefore, the above existing studies generally rely on linear models. Although they perform well in processing high-dimensional spectral data and providing high prediction accuracy, they are insufficient in dealing with complex nonlinear relationships and often fail to capture deep patterns and complex interactions in the data. In addition, the adaptability of existing NIR spectroscopy methods to different materials and environmental conditions also has certain limitations, and it is difficult to cope with technical problems and defects in variable conditions in practical applications. It is urgent to design and develop a method and system for predicting soil moisture content suitable for shield tunnel construction. Summary of the invention

[0010] In order to overcome the shortcomings and difficulties of the above-mentioned prior art, the purpose of the present invention is to provide a method and system for predicting the moisture content of slag suitable for shield tunnel construction, which can quickly provide information on chemical composition, especially outstanding in moisture detection.

[0011] The first purpose of the present invention is to provide a method for predicting the moisture content of slag suitable for shield tunnel construction; the second purpose of the present invention is to provide a system for predicting the moisture content of slag suitable for shield tunnel construction; the third purpose of the present invention is to provide a platform for predicting the moisture content of slag suitable for shield tunnel construction.

[0012] The first object of the present invention is achieved in that the method comprises the following steps:

[0013] Based on the adaptive window, first data corresponding to the slag to be measured in the shield tunnel construction is acquired in real time; wherein the first data is the operation status data of the conveyor belt;

[0014] Combined with the triangle cumulative cross-sectional area method and the AE-FPFH-ICP algorithm, parallel processing is performed to generate the second data and the third data corresponding to the slag to be measured; wherein the second data is the actual cross-sectional area data of the slag; and the third data is the point cloud data corresponding to the slag operation environment and after local area optimization and feature enhancement processing;

[0015] A PLS-SVR combined model is constructed, and fourth data corresponding to the slag to be measured is generated in real time based on the PLS-SVR combined model and in combination with the first data, the second data and the third data respectively; wherein the fourth data is the predicted data of the moisture content of the slag during the shield tunnel construction process.

[0016] Furthermore, the real-time acquisition of first data corresponding to the slag to be measured in the shield tunnel construction based on the adaptive window also includes:

[0017] The fifth data and the sixth data corresponding to the muck to be measured in the shield tunnel construction are respectively obtained, and the local window size in the optical flow calculation is dynamically adjusted according to the fifth data; wherein the fifth data is the gradient intensity data of the local area of ​​the image; the sixth data is the original image data of the muck to be measured; the calculation formula for dynamically adjusting the local window size in the optical flow calculation according to the fifth data is as follows:

[0018]

[0019] Where W0 is the initial window size, α is the adjustment factor, Represents the gradient strength of the local area, and N is the total number of pixels in the window;

[0020] Downsampling the sixth data layer by layer, and constructing at least one image pyramid from high resolution to low resolution;

[0021] Based on the sixth data, optical flow vector data corresponding to each level of image is generated, and the optical flow vector data is fused layer by layer from bottom to top.

[0022] Furthermore, the real-time acquisition of first data corresponding to the slag to be measured in the shield tunnel construction based on the adaptive window also includes:

[0023] In each window, the optical flow vectors of all pixels are weighted and accumulated, and the weights are dynamically adjusted according to the gradient size and consistency of the pixels; the formula for weighted accumulation is:

[0024]

[0025] In the formula, [u i ,v i ] is the optical flow vector of the ith pixel, w i is the weight of the i-th pixel, and the weight calculation formula is:

[0026]

[0027] In the formula, and They represent the gradient of the pixel in the x and y directions respectively, and σ is a parameter used to adjust the weight distribution.

[0028] Furthermore, the method combines the triangle cumulative cross-sectional area method and the AE-FPFH-ICP algorithm to generate the second data and the third data corresponding to the muck to be measured by parallel processing, and further includes:

[0029] Combined with the triangle cumulative cross-sectional area method, the seventh data and the eighth data corresponding to the muck cross section are calculated and generated respectively, and the second data is generated according to the seventh data and the eighth data; wherein the seventh data is the cross-sectional area data of the muck; the eighth data is the cross-sectional area data of the empty belt; the area of ​​each triangle is calculated by the following formula:

[0030]

[0031] In the formula, (x i ,y i ) and (x i+1 ,y i+1 ) are the coordinates of two adjacent points, (x o ,y o ) is the coordinate of the origin, and the areas of all triangles are added up to get the total cross-sectional area:

[0032]

[0033] In the formula, A total Represents the total cross-sectional area, A i represents the area of ​​the ith triangle, and n is the total number of points.

[0034] Furthermore, the method combines the triangle cumulative cross-sectional area method and the AE-FPFH-ICP algorithm to generate the second data and the third data corresponding to the muck to be measured by parallel processing, and further includes:

[0035] Generate and obtain feature data corresponding to the source point cloud and the target point cloud respectively, and fuse them into at least one joint feature descriptor;

[0036] The joint feature vector is normalized and a corresponding relationship between the source point cloud and the target point cloud is created; wherein the normalization formula is:

[0037]

[0038] In the formula, is the standardized joint feature descriptor, μ is the mean of the feature vector, and σ is the standard deviation of the feature vector;

[0039] Generate matching pair geometric information corresponding to the corresponding relationship, and generate a corresponding local transformation matrix based on the matching pair geometric information and in combination with minimizing the error between the source point cloud and the target point cloud; wherein the error function of the local area is:

[0040]

[0041] In the formula, E i is the error of local region i, T iis the transformation matrix of the local area, and are the points in the source point cloud and the target point cloud respectively.

[0042] Furthermore, the constructing of the PLS-SVR combined model, based on the PLS-SVR combined model, respectively combining the first data, the second data and the third data, to generate fourth data corresponding to the muck to be measured in real time, also includes:

[0043] Generate and obtain ninth data corresponding to the muck to be measured; wherein the ninth data is analysis training sample spectral information data corresponding to the muck to be measured;

[0044] According to the ninth data, and in combination with partial least squares regression and support vector regression methods, a corresponding prediction model is established.

[0045] Furthermore, the constructing of the PLS-SVR combined model, based on the PLS-SVR combined model, respectively combining the first data, the second data and the third data, to generate fourth data corresponding to the muck to be measured in real time, also includes:

[0046] The standard normal transformation processes the spectral data corresponding to the muck to be measured and generates the corresponding spectral data matrix;

[0047] Based on the PLS-SVR combined model, the tenth data in the spectral data is extracted, and the correlation between the tenth data and the moisture content of the slag is maximized; wherein the tenth data is low-dimensional feature data;

[0048] In combination with the support vector regression method, the tenth data is processed by regression analysis to generate corresponding eleventh data; wherein the eleventh data is a hyperplane of optimal separation data corresponding to the muck to be measured;

[0049] According to the eleventh data, in combination with the SVR model, the moisture content data corresponding to the muck to be measured is generated; wherein the SVR model needs to meet the following constraints:

[0050]

[0051] Where w is the weight vector, b is the bias, which indicates the offset of the hyperplane, ò is the error tolerance, which defines the maximum prediction error range allowed by the model, and φ(T i ) is the kernel function, which maps the input features to a high-dimensional space. Here, the radial basis function (RBF) is selected as the kernel function, which is defined as follows:

[0052] K(T i ,T j )=exp(-γ‖Ti -T j ‖ 2 ) (35)

[0053] In the formula, γ is the parameter of the kernel function, which controls the influence range of the RBF kernel function;

[0054] Among them, the specific prediction formula for water content data is:

[0055]

[0056] In the formula, is the predicted value, α i is the Lagrange multiplier obtained during the optimization process, T i is the feature matrix of the i-th training sample, represented by the low-dimensional features extracted by PLS, T is the feature matrix of the new sample, K(T i ,T) is the similarity between the new sample and the training sample in the feature space, which is calculated by the RBF kernel function.

[0057] The second object of the present invention is achieved in this way: the system is applied to the method for predicting soil moisture content applicable to shield tunnel construction, and the system comprises:

[0058] A data acquisition unit, used for acquiring first data corresponding to the muck to be measured in shield tunnel construction in real time based on an adaptive window; wherein the first data is the operation status data of the conveyor belt;

[0059] A data processing unit is used to combine the triangle cumulative cross-sectional area method and the AE-FPFH-ICP algorithm to generate second data and third data corresponding to the slag to be measured by parallel processing; wherein the second data is the actual cross-sectional area data of the slag; and the third data is the point cloud data corresponding to the slag operation environment and after local area optimization and feature enhancement processing;

[0060] Construct a generation unit for constructing a PLS-SVR combination model, and generate fourth data corresponding to the slag to be measured in real time based on the PLS-SVR combination model and in combination with the first data, the second data and the third data respectively; wherein the fourth data is the predicted data of the moisture content of the slag during the shield tunnel construction process.

[0061] Furthermore, the data acquisition unit further includes:

[0062] The first acquisition module is used to respectively acquire fifth data and sixth data corresponding to the muck to be measured in the shield tunnel construction, and dynamically adjust the local window size in the optical flow calculation according to the fifth data; wherein the fifth data is the gradient intensity data of the local area of ​​the image; the sixth data is the original image data of the muck to be measured; the calculation formula for dynamically adjusting the local window size in the optical flow calculation according to the fifth data is as follows:

[0063]

[0064] Where W0 is the initial window size, α is the adjustment factor, Represents the gradient strength of the local area, and N is the total number of pixels in the window;

[0065] A processing and construction module, used for downsampling the sixth data layer by layer, and constructing at least one image pyramid from high resolution to low resolution;

[0066] A first generating module is used to generate optical flow vector data corresponding to each level of image based on the sixth data, and fuse and process the optical flow vector data layer by layer from bottom to top;

[0067] The processing and adjustment module is used to perform weighted accumulation processing on the optical flow vectors of all pixels in each window, and dynamically adjust the weights according to the gradient size and consistency of the pixels; the formula for weighted accumulation is:

[0068]

[0069] In the formula, [u i ,v i ] is the optical flow vector of the ith pixel, w i is the weight of the i-th pixel, and the weight calculation formula is:

[0070]

[0071] In the formula, and Represent the gradient of the pixel in the x and y directions respectively, and σ is a parameter used to adjust the weight distribution;

[0072] And / or, the data processing unit further includes:

[0073] The second generation module is used to calculate and generate the seventh data and the eighth data corresponding to the muck cross section respectively by combining the triangle cumulative cross-sectional area method, and generate the second data according to the seventh data and the eighth data; wherein the seventh data is the cross-sectional area data of the muck; the eighth data is the cross-sectional area data of the empty belt; the area of ​​each triangle is calculated by the following formula:

[0074]

[0075] In the formula, (x i ,y i ) and (x i+1 ,y i+1 ) are the coordinates of two adjacent points, (x o ,y o ) is the coordinate of the origin, and the areas of all triangles are added up to get the total cross-sectional area:

[0076]

[0077] In the formula, A total Represents the total cross-sectional area, A i represents the area of ​​the i-th triangle, and n is the total number of points;

[0078] The third generation module is used to generate and obtain feature data corresponding to the source point cloud and the target point cloud respectively, and fuse them into at least one joint feature descriptor;

[0079] The processing and creation module is used to standardize the joint feature vector and create a corresponding relationship between the source point cloud and the target point cloud; wherein the standardization formula is:

[0080]

[0081] In the formula, is the standardized joint feature descriptor, μ is the mean of the feature vector, and σ is the standard deviation of the feature vector;

[0082] The fourth generation module is used to generate matching pair geometric information corresponding to the corresponding relationship, and based on the matching pair geometric information, the corresponding local transformation matrix is ​​generated in combination with minimizing the error between the source point cloud and the target point cloud; wherein the error function of the local area is:

[0083]

[0084] In the formula, E i is the error of local region i, T i is the transformation matrix of the local area, and are points in the source point cloud and the target point cloud and / or respectively, the construction generation unit further includes:

[0085] A fifth generating module is used to generate and obtain ninth data corresponding to the slag to be measured; wherein the ninth data is the analysis training sample spectral information data corresponding to the slag to be measured;

[0086] A model building module, used to build a corresponding prediction model based on the ninth data and in combination with partial least squares regression and support vector regression methods;

[0087] The sixth generation module is used for processing the spectral data corresponding to the muck to be measured by standard normal transformation and generating a corresponding spectral data matrix;

[0088] A first processing module is used to extract the tenth data in the spectral data based on the PLS-SVR combination model, and maximize the correlation between the tenth data and the soil moisture content; wherein the tenth data is low-dimensional feature data;

[0089] The second processing module is used to process the tenth data by regression analysis in combination with a support vector regression method, and generate corresponding eleventh data; wherein the eleventh data is a hyperplane of optimal separation data corresponding to the muck to be measured;

[0090] The seventh generation module is used to generate the moisture content data corresponding to the muck to be measured according to the eleventh data and in combination with the SVR model; wherein the SVR model needs to meet the following constraints:

[0091]

[0092] Where w is the weight vector, b is the bias, which indicates the offset of the hyperplane, ò is the error tolerance, which defines the maximum prediction error range allowed by the model, and φ(T i ) is the kernel function, which maps the input features to a high-dimensional space. Here, the radial basis function (RBF) is selected as the kernel function, which is defined as follows:

[0093] K(T i ,T j )=exp(-γ‖T i -T j ‖ 2 ) (35)

[0094] In the formula, γ is the parameter of the kernel function, which controls the influence range of the RBF kernel function;

[0095] Among them, the specific prediction formula for moisture content data is:

[0096]

[0097] In the formula, is the predicted value, α i is the Lagrange multiplier obtained during the optimization process, T i is the feature matrix of the i-th training sample, represented by the low-dimensional features extracted by PLS, T is the feature matrix of the new sample, K(T i,T) is the similarity between the new sample and the training sample in the feature space, which is calculated by the RBF kernel function.

[0098] The third object of the present invention is achieved as follows: it includes a processor, a memory and a slag moisture content prediction platform control program applicable to shield tunnel construction; wherein the slag moisture content prediction platform control program applicable to shield tunnel construction is executed by the processor, the slag moisture content prediction platform control program applicable to shield tunnel construction is stored in the memory, and the slag moisture content prediction platform control program applicable to shield tunnel construction implements the slag moisture content prediction method applicable to shield tunnel construction.

[0099] The present invention uses a method based on an adaptive window to obtain first data corresponding to the slag to be measured in shield tunnel construction in real time; wherein the first data is the operating status data of the conveyor belt; combined with the triangle cumulative cross-sectional area method and the AE-FPFH-ICP algorithm, parallel processing is performed to generate second data and third data corresponding to the slag to be measured; wherein the second data is the actual cross-sectional area data of the slag; the third data is point cloud data corresponding to the slag operating environment and after local area optimization and feature enhancement processing; a PLS-SVR combined model is constructed, and according to the PLS-SVR combined model, and in combination with the first data, the second data and the third data respectively, fourth data corresponding to the slag to be measured is generated in real time; wherein the fourth data is the predicted data of the moisture content of the slag during the shield tunnel construction process, and the system and platform corresponding to the method can quickly provide information about the chemical composition, especially in terms of moisture detection.

[0100] That is to say, the PLS-SVR combined model proposed in the scheme of the present invention combines the linear dimension reduction capability of PLS ​​and the nonlinear modeling capability of SVR, and shows high accuracy and robustness in the detection of soil moisture content. The model performs well in processing complex nonlinear relationships, can accurately predict the soil moisture content in real time, has strong adaptability, and meets the high-precision requirements of tunnel construction for moisture content detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0101] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0102] Figure 1 This is a schematic diagram of the support structure of a device for predicting soil moisture content in shield tunnel construction according to the present invention;

[0103] Figure 2 It is a schematic diagram of the structure inside the box of a device for predicting soil moisture content in shield tunnel construction according to the present invention;

[0104] Figure 3 It is a schematic diagram of a technical route of an embodiment of a method for predicting soil moisture content applicable to shield tunnel construction of the present invention;

[0105] Figure 4 A schematic diagram of a conveyor belt of an embodiment of a method for predicting soil moisture content applicable to shield tunnel construction according to the present invention;

[0106] Figure 5 This is a schematic diagram of point cloud data of an embodiment of a method for predicting soil moisture content applicable to shield tunnel construction according to the present invention;

[0107] Figure 6 This is a schematic diagram of a process flow of soil moisture prediction for an embodiment of a soil moisture prediction method for shield tunnel construction according to the present invention;

[0108] Figure 7 A schematic flow chart of the steps of a method for predicting soil moisture content applicable to shield tunnel construction according to the present invention;

[0109] Figure 8 This is a schematic diagram of the architecture of a system for predicting soil moisture content in shield tunnel construction according to the present invention;

[0110] Fig. 9 This is a schematic diagram of a platform architecture for predicting soil moisture content in shield tunnel construction according to the present invention;

[0111] 1-top connecting device; 2-rotating mechanism; 3-connecting rod; 4-box; 5-spectrometer; 6-laser radar.

[0112] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings in conjunction with the embodiments. DETAILED DESCRIPTION

[0113] In order to better understand the purpose, technical solutions and advantages of the present invention, the present invention is further described below in conjunction with the accompanying drawings and specific implementation methods. Those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification.

[0114] The present invention may also be implemented or applied through other different specific examples, and the details in this specification may also be modified and changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention.

[0115] It should be noted that if the embodiments of the present invention involve directional indications (such as up, down, left, right, front, back, etc.), the directional indications are only used to explain the relative position relationship, movement status, etc. between the components under a certain specific posture (as shown in the accompanying drawings). If the specific posture changes, the directional indication will also change accordingly.

[0116] In addition, if there are descriptions involving "first", "second", etc. in the embodiments of the present invention, the descriptions of "first", "second", etc. are only used for descriptive purposes and cannot be understood as indicating or suggesting their relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. Secondly, the technical solutions between the various embodiments can be combined with each other, but they must be based on the ability of ordinary technicians in the field to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be deemed that such a combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

[0117] Preferably, the present invention is a method for predicting soil moisture content suitable for shield tunnel construction, which is applied in one or more terminals or servers. The terminal is a device that can automatically perform numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes but is not limited to microprocessors, application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), digital signal processors (DSP), embedded devices, etc.

[0118] The terminal can be a computing device such as a desktop computer, a notebook, a PDA, a cloud server, etc. The terminal can interact with the client through a keyboard, a mouse, a remote control, a touch pad, or a voice control device.

[0119] The present invention is to realize a method, system and platform for predicting the moisture content of slag suitable for shield tunnel construction.

[0120] like Figure 7 , which is a flow chart of a method for predicting soil moisture content applicable to shield tunnel construction provided by an embodiment of the present invention.

[0121] In this embodiment, the method for predicting the moisture content of slag suitable for shield tunnel construction can be applied to a terminal or a fixed terminal with a display function. The terminal is not limited to a personal computer, a smart phone, a tablet computer, a desktop computer or an all-in-one computer equipped with a camera, etc.

[0122] The method for predicting the moisture content of soil slag applicable to shield tunnel construction can also be applied to a hardware environment consisting of a terminal and a server connected to the terminal via a network. The network includes but is not limited to: a wide area network, a metropolitan area network or a local area network. The method for predicting the moisture content of soil slag applicable to shield tunnel construction of the embodiment of the present invention can be executed by a server, can be executed by a terminal, or can be executed by both a server and a terminal.

[0123] For example, for a terminal that needs to predict the moisture content of slag applicable to shield tunnel construction, the function of predicting the moisture content of slag applicable to shield tunnel construction provided by the method of the present invention can be directly integrated on the terminal, or a client for implementing the method of the present invention can be installed. For another example, the method provided by the present invention can also be run on a server or other device in the form of a software development kit (SDK), and an interface for predicting the moisture content of slag applicable to shield tunnel construction is provided in the form of SDK. The terminal or other device can realize the function of predicting the moisture content of slag applicable to shield tunnel construction through the provided interface. The present invention is further described below in conjunction with the accompanying drawings.

[0124] like Figure 3-Figure 7 As shown, the present invention provides a method for predicting soil moisture content applicable to shield tunnel construction, the method comprising the following steps:

[0125] S1. Based on an adaptive window, first data corresponding to the slag to be measured in shield tunnel construction is acquired in real time; wherein the first data is the operation status data of the conveyor belt;

[0126] S2, combining the triangle cumulative cross-sectional area method and the AE-FPFH-ICP algorithm, parallel processing is performed to generate second data and third data corresponding to the slag to be measured; wherein the second data is the actual cross-sectional area data of the slag; and the third data is the point cloud data corresponding to the slag operation environment and after local area optimization and feature enhancement processing;

[0127] S3. Construct a PLS-SVR combined model, and generate in real time fourth data corresponding to the slag to be measured based on the PLS-SVR combined model and respectively combining the first data, the second data and the third data; wherein the fourth data is the predicted data of moisture content of the slag during the shield tunnel construction process.

[0128] The method of acquiring first data corresponding to the slag to be measured in shield tunnel construction in real time based on the adaptive window also includes:

[0129] S11, respectively obtaining fifth data and sixth data corresponding to the muck to be measured in the shield tunnel construction, and dynamically adjusting the local window size in the optical flow calculation according to the fifth data; wherein the fifth data is the gradient intensity data of the local area of ​​the image; the sixth data is the original image data of the muck to be measured; the calculation formula for dynamically adjusting the local window size in the optical flow calculation according to the fifth data is as follows:

[0130]

[0131] Where W0 is the initial window size, α is the adjustment factor, Represents the gradient strength of the local area, and N is the total number of pixels in the window;

[0132] S12, downsampling the sixth data layer by layer, and constructing at least one image pyramid from high resolution to low resolution;

[0133] S13. Generate optical flow vector data corresponding to each level of image based on the sixth data, and fuse the optical flow vector data layer by layer from bottom to top.

[0134] The method of acquiring first data corresponding to the slag to be measured in shield tunnel construction in real time based on the adaptive window also includes:

[0135] S14, in each window, weighted accumulation processing is performed on the optical flow vectors of all pixels, and the weights are dynamically adjusted according to the gradient size and consistency of the pixels; wherein the formula for weighted accumulation is:

[0136]

[0137] In the formula, [u i ,v i ] is the optical flow vector of the ith pixel, w i is the weight of the i-th pixel, and the weight calculation formula is:

[0138]

[0139] In the formula, and They represent the gradient of the pixel in the x and y directions respectively, and σ is a parameter used to adjust the weight distribution.

[0140] The method combines the triangle cumulative cross-sectional area method and the AE-FPFH-ICP algorithm to generate the second data and the third data corresponding to the muck to be measured by parallel processing, and also includes:

[0141] S21. Combined with the triangle cumulative cross-sectional area method, the seventh data and the eighth data corresponding to the cross-sectional area of ​​the slag are calculated and generated respectively, and the second data is generated according to the seventh data and the eighth data; wherein the seventh data is the cross-sectional area data of the belt loaded with slag; the eighth data is the cross-sectional area data of the empty belt; the area of ​​each triangle is calculated by the following formula:

[0142]

[0143] In the formula, (x i ,y i ) and (x i+1 ,y i+1 ) are the coordinates of two adjacent points, (x o ,y o ) is the coordinate of the origin, and the areas of all triangles are added up to get the total cross-sectional area:

[0144]

[0145] In the formula, A total Represents the total cross-sectional area, A i represents the area of ​​the ith triangle, and n is the total number of points.

[0146] The method combines the triangle cumulative cross-sectional area method and the AE-FPFH-ICP algorithm to generate the second data and the third data corresponding to the muck to be measured by parallel processing, and also includes:

[0147] S22, respectively generating and acquiring feature data corresponding to the source point cloud and the target point cloud, and fusing them into at least one joint feature descriptor;

[0148] S23, normalize the joint feature vector and create a corresponding relationship between the source point cloud and the target point cloud; wherein the normalization formula is:

[0149]

[0150] In the formula, is the standardized joint feature descriptor, μ is the mean of the feature vector, and σ is the standard deviation of the feature vector;

[0151] S24, generating matching pair geometric information corresponding to the corresponding relationship, and generating a corresponding local transformation matrix based on the matching pair geometric information and in combination with minimizing the error between the source point cloud and the target point cloud; wherein the error function of the local area is:

[0152]

[0153] In the formula, E i is the error of local region i, Ti is the transformation matrix of the local area, and are the points in the source point cloud and the target point cloud respectively.

[0154] The method further comprises: constructing a PLS-SVR combined model, and generating fourth data corresponding to the slag to be measured in real time according to the PLS-SVR combined model and combining the first data, the second data and the third data respectively; wherein the fourth data is the predicted data of the moisture content of the slag during the shield tunnel construction process; and further comprising:

[0155] S31, generating and acquiring ninth data corresponding to the muck to be measured; wherein the ninth data is analysis training sample spectral information data corresponding to the muck to be measured;

[0156] S32. Establish a corresponding prediction model based on the ninth data and in combination with partial least squares regression and support vector regression methods.

[0157] The constructing of the PLS-SVR combined model, and generating fourth data corresponding to the muck to be measured in real time according to the PLS-SVR combined model and combining the first data, the second data and the third data respectively, further includes:

[0158] S33, processing the spectral data corresponding to the muck to be measured by standard normal transformation, and generating a corresponding spectral data matrix;

[0159] S34, based on the PLS-SVR combined model, extracting the tenth data in the spectral data, and maximizing the correlation between the tenth data and the soil moisture content; wherein the tenth data is low-dimensional feature data;

[0160] S35, combining support vector regression method, regression analysis processing the tenth data, and generating corresponding eleventh data; wherein the eleventh data is a hyperplane of optimal separation data corresponding to the muck to be measured;

[0161] S36. Generate moisture content data corresponding to the soil to be measured based on the eleventh data and in combination with the SVR model; wherein the SVR model needs to meet the following constraints:

[0162]

[0163] Where w is the weight vector, b is the bias, which indicates the offset of the hyperplane, ò is the error tolerance, which defines the maximum prediction error range allowed by the model, and φ(T i ) is the kernel function, which maps the input features to a high-dimensional space. Here, the radial basis function (RBF) is selected as the kernel function, which is defined as follows:

[0164] K(T i ,T j )=exp(-γ‖T i -T j ‖ 2 ) (35)

[0165] In the formula, γ is the parameter of the kernel function, which controls the influence range of the RBF kernel function;

[0166] Among them, the specific prediction formula for water content data is:

[0167]

[0168] In the formula, is the predicted value, α i is the Lagrange multiplier obtained during the optimization process, T i is the feature matrix of the i-th training sample, represented by the low-dimensional features extracted by PLS, T is the feature matrix of the new sample, K(T i ,T) is the similarity between the new sample and the training sample in the feature space, which is calculated by the RBF kernel function.

[0169] Specifically, in an embodiment of the present invention, the method of combining PLS with support vector regression (SVR) proposed in this scheme, by combining the linear dimensionality reduction capability of PLS ​​with the nonlinear modeling capability of SVR, the method of this scheme can more effectively capture complex nonlinear relationships, improve the accuracy and robustness of prediction, and enhance the adaptability of the model in a changing environment, thereby providing a more innovative and practical moisture detection solution.

[0170] The dual laser radar algorithm proposed in this invention uses one laser radar to measure speed and the other to measure the cross-sectional point cloud, and combines the FPFH_ICP inter-frame point cloud matching algorithm to accurately calculate the volume by integrating the speed and the cross-sectional point cloud area. Compared with the existing algorithms, this method has significant advantages.

[0171] At the same time, the Spectral Prediction Fusion Model (SPFM) proposed in this solution uses a combined model of PLS ​​and support vector regression (SVR) for prediction, which has higher accuracy and robustness than existing single algorithms. The SPFM model is not only highly adaptable and able to maintain performance in a changing environment, but also has the advantages of rapid response and real-time measurement, making it very suitable for field application scenarios that require rapid decision-making. By combining these innovative methods, it performs well in accuracy, efficiency, and automation, providing more reliable technical support for volume measurement and moisture content prediction in tunnel construction. The technical roadmap of the solution of the present invention is as follows: Figure 3 shown.

[0172] In order to fully obtain the muck data, the present invention designs a muck condition monitoring system MCMS (Muck condition monitoring system) that integrates a variety of advanced sensing equipment to achieve comprehensive monitoring of the belt conveyor. The hardware includes two parts: a detection bracket and a sensor. The bracket consists of four parts: a rotating shaft, a connecting rod, a top connecting device, and a box. It is made of all-aluminum alloy lathe processing, 3D printing and other advanced technologies. The sensor integration includes two single-line laser radars, a near-infrared spectrometer, two 4K high-definition cameras and a 4K wide-angle camera. The two laser radars are used to collect cross-sectional point cloud data and longitudinal point cloud data, respectively. The two high-definition cameras are used to collect video files in front of and behind the equipment, respectively, and the wide-angle camera is used to collect video files directly below the equipment. The near-infrared spectrometer collects near-infrared band spectral information of the soil below. This multi-sensor fusion solution can provide comprehensive environmental perception capabilities to ensure the accuracy and integrity of the data. Such as Figure 1 is a schematic diagram of the bracket structure. Figure 2 It is a box-body structure. The specific equipment parameters are shown in Table 1 below.

[0173] Table 1 - Sensor parameters

[0174]

[0175] The solution of the present invention realizes two different fixing methods. If there is an installation platform, the device can be installed on the platform. If there is no installation position, the device can be fixed on the pipe segment. The design of two rotating bearings plus four connecting rods allows the height and angle of the bracket to be adjusted to meet the needs of different tunnels and measurements. The sensor can be arranged directly above the conveyor belt, which can improve the acquisition accuracy of the point cloud, improve the accuracy of spectral information acquisition, and improve the clarity of video or box acquisition. It is suitable for complex tunnel environments. In terms of overall design, the bracket is not only powerful, but also has high stability and durability, which can ensure the long-term stable operation of the monitoring equipment in harsh construction environments. In addition, the bracket is also portable, and its modular design allows each measuring device to be quickly disassembled and assembled, which is convenient for movement and deployment at different construction sites.

[0176] Improved optical flow method to determine the motion state of the conveyor belt: In order to realize the automatic power on and off of the equipment, the scheme of the present invention uses the optical flow method to detect the movement of the slag on the conveyor belt, so as to determine whether the conveyor belt is running. The traditional optical flow method is based on the assumption of constant brightness and assumes that the brightness of pixels in the image does not change over time. By calculating the movement of pixels in a local window between consecutive frames, the traditional method constructs the optical flow equation, and uses the gradient information of all pixels in the local window to solve the optical flow vector by the least squares method. This method assumes that the movement within the window is linear and is suitable for processing image movement within a small range. However, the traditional optical flow method is easily affected by noise and complex motion, and the fixed window size also limits its adaptability in multi-scale scenarios.

[0177] The improved optical flow method uses two main techniques, adaptive window selection and multi-scale pyramid method, to enhance its robustness and accuracy in complex environments. This method is particularly suitable for handling complex multi-scale motion scenes such as conveyor belt motion detection in tunnel construction environments. These improvements enable the optical flow method to not only perform well in motions of different scales, but also effectively deal with noise problems and provide more accurate optical flow estimates.

[0178] First, the solution of the present invention dynamically adjusts the local window size in the optical flow calculation through an adaptive window selection method. The traditional method uses a fixed-size window, which is simple and easy to implement, but may cause the accuracy of optical flow estimation to decrease in areas with uneven motion intensity. Therefore, the solution of the present invention introduces an adaptive window selection method to dynamically adjust the window size according to the gradient intensity of the local area of ​​the image. Specifically, the window size W(x0, y0) is determined by the initial window size W0 plus a correction term based on the local gradient intensity. The calculation of this correction term is based on the average value of the pixel gradient in the window. Therefore, in areas with intense motion, the window will automatically increase to cover a larger motion area; in areas with relatively gentle motion, the window will be reduced to reduce noise interference. This improved formula can be expressed as:

[0179]

[0180] Where W0 is the initial window size, α is the adjustment factor, represents the gradient strength of the local area, and N is the total number of pixels in the window. By adaptively adjusting the window size, the optical flow estimation is more sensitive to areas with different motion intensities, reducing the impact of noise and improving the overall accuracy.

[0181] Secondly, the solution of the present invention adopts a multi-scale pyramid method to handle complex motions at different scales. This method constructs an image pyramid from high resolution to low resolution by downsampling the original image layer by layer. On each level of the image, the optical flow method is used to calculate the optical flow vector. The optical flow calculation is based on the classic brightness constant assumption, that is, the brightness of the object remains unchanged during movement. Specifically, the optical flow equation can be expressed as:

[0182] I(x,y,t)=I(x+u,y+v,t+Δt) (2)

[0183] Taylor expansion of the equation and ignoring higher-order terms yields the following optical flow equation:

[0184]

[0185] in, and Represent the gradient of the image in the horizontal and vertical directions respectively, represents the temporal gradient of the image, u l and v l are the optical flow components of the l-th layer pyramid image respectively. The process of layer-by-layer optical flow vector fusion starts from the lowest layer, where the optical flow vectors calculated will be fused upward layer by layer. At each level l, the optical flow vector is enlarged to the resolution of the previous level through interpolation and used as the initial estimate to continue the optical flow calculation:

[0186] (u l-1 ,v l-1 )=upsample(u l ,v l )+Lucas-Kanade(I l-1 ,I l-1+Δt ) (4)

[0187] Among them, upsample(u l ,v l ) is the optical flow vector (u l ,v l ) is upsampled to the resolution of the previous level. This step enables the optical flow estimation of the low-resolution layer to be used to guide the optical flow calculation of the high-resolution layer, thereby improving the optical flow accuracy of the high-resolution layer.

[0188] This layer-by-layer fusion process continues until the highest resolution level l = 1 is reached, and the global optical flow vector (u1, v1) is finally obtained. This bottom-up optical flow fusion method effectively utilizes the large-scale motion information in low-resolution images while refining the optical flow estimate on high-resolution images.

[0189] Through layer-by-layer fusion, the optical flow results of the low-resolution layer provide a good initial estimate for the high-resolution layer, thereby reducing the computational burden of complex motion estimation in the high-resolution layer and ultimately improving the robustness and accuracy of the entire optical flow estimation.

[0190] In addition, in order to further reduce the optical flow error caused by local noise, a weighted accumulation method is introduced. In each window, the optical flow vectors of all pixels are weighted and accumulated, and the weights are dynamically adjusted according to the gradient size and consistency of the pixels. This method ensures that more consistent pixels contribute more to the optical flow estimation, while the influence of noisy pixels is minimized. The formula for weighted accumulation can be expressed as:

[0191]

[0192] Among them, [u i ,v i ] is the optical flow vector of the ith pixel, w i is the weight of the i-th pixel, and the weight calculation formula is:

[0193]

[0194] In this formula, and Represent the gradient of the pixel in the x and y directions respectively, and σ is a parameter used to adjust the weight distribution. Through this weighted accumulation method, the weight of pixels with larger noise in the optical flow estimation is reduced, and the final output optical flow vector is more accurate.

[0195] With these two improvements, the optical flow method can show higher robustness and accuracy when dealing with complex motion scenes. Adaptive window selection allows the optical flow estimation to be flexibly adjusted according to local motion characteristics, the multi-scale pyramid method ensures that motion information at different scales is effectively captured, and the weighted accumulation rule further reduces the impact of noise on optical flow calculation. In summary, these improvements make the improved optical flow method have significant advantages in complex scenes, especially in situations where large-scale displacements and subtle local motions need to be processed.

[0196] Through the above improvements, the optical flow method shows higher accuracy and robustness in complex tunnel construction environments, ensuring accurate judgment of the conveyor belt's operating status, thereby realizing automated management and energy-saving control of the equipment. The adaptive window selection method reduces the impact of noise by dynamically adjusting the window size; the multi-scale pyramid method improves the robustness of optical flow calculation through multi-scale calculation and fusion; the weighted accumulation method minimizes the impact of noise on optical flow calculation through weight adjustment. These improvements enable the optical flow method to more accurately detect the operating status of the conveyor belt in complex environments.

[0197] LiDAR cross-sectional measurement of slag: An innovative "Triangular Accumulation Cross-sectional Area (TACA)" method is proposed. This method decomposes the cross-sectional point cloud data into multiple triangles, calculates the area of ​​each triangle and accumulates them to obtain the total cross-sectional area. This algorithm includes two main steps: first, the cross-sectional area of ​​slag and the cross-sectional area of ​​the empty belt are calculated separately, and then the two are subtracted to obtain the actual cross-sectional area of ​​slag.

[0198] In order to obtain continuous point cloud data, a single-line laser radar is used to scan the muck on the belt at a frequency of 120Hz. These data are divided into two sets: point cloud data loaded with muck and point cloud data of empty belt. In order to improve the data quality, statistical filtering is first applied to the point cloud data, and then further processed in combination with the Savitzky-Golay (SG) filtering method.

[0199] Statistical filtering determines and removes outliers by calculating the distance from each point in the point cloud to the point in its neighborhood, thereby improving the overall quality of the data. Specifically, for each point p in the point cloud, first find all the points N(p) in its neighborhood, and then calculate the Euclidean distance between point p and each point q in its neighborhood

[0200]

[0201] Next, the average distance and standard deviation of each point are calculated, and whether it is an abnormal point is determined based on the preset distance threshold and standard deviation threshold. Usually a standard deviation threshold σ is set th :

[0202] σ th =k×σ (8)

[0203] σ th = k × σ, where k is a constant and σ represents the standard deviation of the data point. Finally, these outliers that exceed the threshold range are removed.

[0204] The advantage of the statistical filter is that it can effectively remove isolated noise points in the point cloud data, thereby improving the smoothness of the data.

[0205] In order to further smooth the data and remove high-frequency noise, the solution of the present invention combines the use of SG filtering. SG filtering performs polynomial fitting on the points in the sliding window and replaces the values ​​of the points in the original window with the fitted polynomial values, thereby achieving data smoothing. The specific steps include: selecting a sliding window containing an odd number of points, usually 2m+1 points, and then performing polynomial fitting on the points in the sliding window, assuming that the fitted polynomial is P(x):

[0206] P(x)=a0+a1x+a2x 2 +…+a n x n (9)

[0207] Finally, the fitted polynomial value is used to replace the value of the original point in the window. By combining statistical filtering and SG filtering, the scheme of the present invention can significantly improve the quality of point cloud data and provide a reliable basis for subsequent cross-sectional area calculation. The innovation of combining these two filtering methods is that statistical filtering can effectively remove isolated noise points, while SG filtering can smooth high-frequency noise, ensuring high precision and smoothness of point cloud data.

[0208] After preprocessing the point cloud data, the cross-sectional area is calculated using the triangle cumulative cross-sectional area method (TACA). The specific method is to select the belt center as the origin O and process each point cloud data set in turn. The first triangle is formed by selecting the origin O and the first and second points in the point cloud, and then the second triangle is formed by the origin, the second point and the third point, and so on. The area of ​​each triangle is calculated using the following formula:

[0209]

[0210] Among them, (x i ,y i ) and (x i+1 ,y i+1 ) are the coordinates of two adjacent points, (x o ,y o ) are the coordinates of the origin. The areas of all triangles are added together to get the total cross-sectional area:

[0211]

[0212] Among them, A total Represents the total cross-sectional area. i represents the area of ​​the i-th triangle. n is the total number of points. In order to calculate the cross-sectional area of ​​the slag, the total cross-sectional area A of the slag is calculated separately. soil and the total cross-sectional area of ​​the empty belt A belt Then subtract the two to get the cross-sectional area of ​​the soil:

[0213] A final =A soil -A belt (12)

[0214] The triangle cumulative cross-sectional area method improves the accuracy and efficiency of point cloud data in tunnel muck measurement by refining the calculation process of cross-sectional area. This algorithm is particularly suitable for application scenarios of single-line laser radar high-frequency data acquisition, ensuring the accuracy of volume measurement during tunnel construction and maintenance. Its high accuracy, efficiency and robustness significantly improve the quality control capability of tunnel engineering.

[0215] LiDAR speed measurement algorithm: The present invention also proposes an innovative algorithm, named AdaptiveEnhanced FPFH-ICP (AE-FPFH-ICP), which is designed for the application of belt conveyor speed measurement in a tunnel environment using a single-line LiDAR. The algorithm combines a high-frequency frame rate of 120Hz with the optimization of the special environment in the tunnel, aiming to improve the matching accuracy, computational efficiency, robustness and adaptability of point cloud data processing. Its core idea is to achieve efficient and accurate point cloud matching and volume calculation through local area optimization and feature enhancement. The dynamic ICP algorithm reduces the computational complexity and improves the matching accuracy by dividing the point cloud data into multiple local areas and independently registering them for two adjacent frames of point clouds. Each area calculates the transformation matrix independently to ensure the local matching accuracy, and then fuses it into a global transformation matrix to adapt to the overall environmental changes. AE feature enhancement extracts and fuses multiple point cloud features, such as FPFH, SHOT and Spin Image features, to form a joint feature descriptor, and normalizes it to eliminate the influence of feature magnitude, and finally improves robustness and accuracy through Euclidean distance matching. FPFH generates feature descriptors by calculating the geometric features of each point and its neighborhood, while dynamic ICP finds the optimal alignment by iteratively minimizing the point cloud error. Combined with high-frequency data acquisition and local optimization, it is suitable for high-frequency matching of adjacent frames, effectively improving the accuracy and efficiency of point cloud matching.

[0216] In the AE-FPFH-ICP algorithm, the source point cloud and the target point cloud are firstly subjected to feature extraction. Specifically, geometric feature descriptors such as FPFH, SHOT, and Spin Image are used to generate feature vectors for each point. The purpose of feature extraction is to provide a descriptor that can quantify the local geometric structure of each point in the point cloud, thereby providing a basis for subsequent matching and registration.

[0217] The extraction formula of the joint feature descriptor is:

[0218]

[0219] in, They are the FPFH, SHOT and Spin Image features of point p. By fusing these features into a joint feature descriptor, the local geometric information of the point cloud can be more comprehensively described, and the robustness of the registration can be improved. In order to eliminate the influence of the feature magnitude, the joint feature vector is normalized. The normalization formula is:

[0220]

[0221] in, is the standardized joint feature descriptor, μ is the mean of the feature vector, and σ is the standard deviation of the feature vector. Standardization ensures that all feature vectors are on the same scale, thereby reducing the impact of feature magnitude on registration accuracy. During the registration process, feature matching is performed using Euclidean distance. Specifically, for the standardized joint feature descriptor, the Euclidean distance between point p and point q is calculated as:

[0222]

[0223] By minimizing the Euclidean distance, the optimal point pair matching is determined, thus establishing a correspondence between the source point cloud and the target point cloud. Once enough matching pairs are found, the algorithm divides the source point cloud into multiple local regions. For each local region, the local transformation matrix T is calculated by minimizing the error between the source point cloud and the target point cloud based on the geometric information of the matching pairs. i The error function of the local area is defined as:

[0224]

[0225] Among them, E i is the error of local region i, T i is the transformation matrix of the local area, and are the points in the source point cloud and the target point cloud respectively. By minimizing this error function, the local transformation matrix T can be optimized i , thus achieving point cloud registration in the local area. Local transformation matrix T i The calculation of is based on the matching point pairs between the source point cloud and the target point cloud. First, the centroid of the point cloud in the local area needs to be calculated. For the local area i, the centroids of the source point cloud and the target point cloud are expressed as:

[0226]

[0227] After calculating the centroid, the coordinates of each point are decentralized to obtain the point coordinates without the influence of the centroid:

[0228]

[0229] The decentralized point coordinates are used to construct the covariance matrix:

[0230]

[0231] The covariance matrix H describes the geometric relationship between the source point cloud and the target point cloud in the local area. Next, the rotation matrix and translation vector are solved by SVD decomposition. First, the covariance matrix H is decomposed by SVD to obtain:

[0232] H=UΣV T (20)

[0233] The rotation matrix R is calculated from i and the translation vector t i :

[0234]

[0235] Finally, the transformation matrix T of local area i i It is expressed as:

[0236]

[0237] This transformation matrix is ​​used to align the source point cloud with the target point cloud in a local area.

[0238] After all local regions are registered, the transformation matrices of each local region are fused to form a global transformation matrix. The calculation formula of the global transformation matrix T is:

[0239]

[0240] Where K is the number of local regions, T i is the transformation matrix of the i-th local area. By fusing the transformation results of the local areas, the obtained global transformation matrix can align the entire source point cloud to the target point cloud to achieve overall point cloud registration. After the global transformation matrix is ​​preliminarily calculated, the AE-FPFH-ICP algorithm further improves the accuracy of the registration through ICP iterative optimization. Through continuous iteration of nearest neighbor search and error minimization calculation, the algorithm updates the global transformation matrix until the error converges or reaches a predetermined number of iterations. The AE-FPFH-ICP algorithm ensures the efficiency and robustness of point cloud registration through local area optimization, global transformation matrix fusion, and multi-feature fusion and standardization. This algorithm is particularly suitable for complex tunnel environments. By gradually optimizing the matching results, it significantly improves the accuracy and efficiency of point cloud registration.

[0241] Volume calculation: After acquiring the point cloud data and completing the registration, the velocity v(t) is calculated by the AE-FPFH-ICP algorithm. The velocity can be determined by calculating the displacement and time difference between two consecutive frames:

[0242]

[0243] where dS is the displacement of the belt during the time interval dt.

[0244] After calculating the speed and cross-sectional area, the volume of the slag is calculated by integration. Assuming that the speed of the belt conveyor is v(t) and the cross-sectional area changes over time as A(t), the volume V can be calculated by the following formula:

[0245]

[0246] In order to perform numerical integration, the Simpson method is selected. The Simpson method has higher accuracy because it approximates the integral through quadratic interpolation, which can better handle the nonlinear changes of cross-sectional area and velocity data and reduce the integral error. The specific Simpson method formula is as follows:

[0247]

[0248] In the above formula: V represents the calculated volume, t0 and t1 are the start and end time of the volume calculation, A(t) is the cross-sectional area at time t, v(t) is the velocity at time t, n is the total number of data points, ti is the i-th time point, A(t i ) is the cross-sectional area at the i-th time point, v(t i ) is the velocity at the ith time point.

[0249] Through the numerical integration method, the AE-FPFH-ICP algorithm can achieve efficient and accurate point cloud data matching and volume calculation, and can adapt to complex tunnel environments, significantly improving the matching accuracy and calculation efficiency, and enhancing the robustness and adaptability of the algorithm.

[0250] Near-infrared spectroscopy moisture content prediction based on PLS-SVR: In the measurement of tunnel slag moisture content, the near-infrared spectrometer can achieve accurate prediction of slag moisture content. In order to achieve this goal, 11 groups of samples were prepared, with moisture contents of 0%, 10%, 20%, 30%, 40%, 50%, 60%, 70%, 80%, 90% and 100%. Each group of samples contains 100 data points, totaling 1100 samples. 80 groups of samples were selected for model training, and the remaining 20 groups were used for verification. By analyzing the spectral information of the training samples, combining PLS (partial least squares regression) and support vector regression (SVR) methods, a prediction model was established.

[0251] A total of 1100 samples. The data was randomly divided into a training set and a validation set. 80 sets of samples in each set (800 data points with sample moisture content ranging from 0% to 100% in steps of 10%. Each set of samples contains 100 data points, total) were used to train the model, and 20 sets of samples (200 data points) were used to verify the model performance. The PLS-SVR model combines the linear dimensionality reduction capability of PLS ​​and the nonlinear regression capability of SVR, and can perform high-precision nonlinear modeling while maintaining the main information of the data. The specific steps are as follows and Figure 6 shown.

[0252] PLS-SVR near-infrared spectroscopy prediction algorithm: In the measurement of tunnel soil moisture content, a combination of PLS ​​(partial least squares regression) and SVR (support vector regression) was used to construct a prediction model. The advantage of this combined model is that PLS can effectively extract the main information in the spectral data and reduce the data dimension, while SVR can handle complex nonlinear relationships and further improve the prediction accuracy.

[0253] Data preprocessing: Standard Normal Variate (SNV): In the experiment, 11 groups of samples were prepared, each with 100 data points, totaling n = 1100n = 1100n = 1100 samples. 800 groups of samples were selected for model training, and the remaining 300 groups of samples were used for model verification. The spectral data of each sample has 121 resolution points, so the spectral data matrix X train It can be expressed as:

[0254]

[0255] Among them, X ij represents the spectral value of the i-th training sample at the j-th resolution point. In order to eliminate the scattering effect and background noise in the spectral data, the standard normal transformation (SNV) is used to preprocess the spectral data. After SNV preprocessing, the spectral data matrix Calculated by the following formula:

[0256]

[0257] Where: μ i is the mean of the i-th sample at 121 resolution points:

[0258]

[0259] σ i is the standard deviation of the ith sample at 121 resolution points:

[0260]

[0261] After SNV preprocessing, the standardized spectral data matrix It has the characteristics of zero mean and one standard deviation, thus reducing the deviation caused by the difference in spectral measurement of different samples.

[0262] PLS extracts low-dimensional features: In the prediction of soil moisture content, the PLS (partial least squares regression) model is used to extract low-dimensional features from spectral data and maximize the correlation between these features and soil moisture content. The main goal of PLS ​​is to extract the principal components most related to the change in moisture content by projecting high-dimensional spectral data into a low-dimensional latent space, thereby reducing the dimensionality while retaining the main information in the spectral data. First, PLS maximizes the spectral data. The weight vector w is calculated by the projected covariance between y and the water content y, and the formula is:

[0263]

[0264] By using this formula, the weight vector w is calculated to extract the principal component direction that is most relevant to the change in moisture content in the spectral data. Among them, w is the weight vector, which represents the direction that is most relevant to the change in moisture content in the spectral data. By calculating the weight vector w, the spectral data matrix can be projected into a low-dimensional space, retaining the main information related to water content. Next, the spectral data matrix is ​​quantified using the weight vector w. Perform projection and obtain the projection matrix T:

[0265]

[0266] Where T is a 800×k matrix, which represents the result of projecting the spectral data into the low-dimensional latent space, and k is the number of principal components extracted. The projection matrix T retains the main characteristic information in the spectral data that can best explain the change in water content and provides input data for subsequent nonlinear regression.

[0267] Nonlinear regression using SVR: After PLS extracts features, SVR (Support Vector Regression) is used to perform regression analysis on the extracted features. The core idea of ​​SVR is to perform nonlinear regression prediction by finding an optimal hyperplane. In order to capture the nonlinear relationship in the data, the radial basis function (RBF) is selected as the kernel function of SVR because the RBF kernel function performs well in dealing with complex nonlinear relationships and has wide applicability.

[0268] The goal of SVR is to find a hyperplane that can optimally separate the data. To do this, the following optimization problem needs to be solved:

[0269]

[0270] Where ‖w‖ is the bi-norm of the weight vector, which indicates the complexity of the model. Minimizing ‖w‖ helps find a smooth hyperplane and avoid overfitting. C is a penalty coefficient used to balance between model complexity and training error. Larger C values ​​will strictly control the error and may result in fewer support vectors, while smaller C values ​​allow more error but produce a smoother model. ξ i and It is a slack variable used to deal with samples that cannot be completely correctly classified, that is, allowing a certain error to improve the generalization ability of the model.

[0271] The SVR model needs to meet the following constraints to ensure that the difference between the predicted value and the true value is within an acceptable range:

[0272]

[0273] Where w is the weight vector. b is the bias, which indicates the offset of the hyperplane. ò Error tolerance defines the maximum prediction error range allowed by the model. φ(T i ) is the kernel function, which maps the input features to a high-dimensional space. Here, the radial basis function (RBF) is selected as the kernel function, which is defined as follows:

[0274] K(T i ,T j )=exp(-γ‖T i -T j ‖ 2 ) (35)

[0275] Among them, γ is the parameter of the kernel function, which controls the influence range of the RBF kernel function.

[0276] By solving the above problem through optimization algorithms (such as Lagrange multiplier method), the optimal weight vector w and bias b can be obtained. This process is actually to find a hyperplane in the high-dimensional feature space that can best describe the input-output relationship. Using the trained SVR model, the water content of new samples can be predicted. The specific prediction formula is:

[0277]

[0278] in, is the predicted value, i.e. the predicted result of the moisture content of the new sample. i is the Lagrange multiplier obtained during the optimization process. T i is the feature matrix of the i-th training sample, represented by the low-dimensional features extracted by PLS. T is the feature matrix of the new sample, that is, the features of the sample to be predicted. K(T i,T) is the similarity between the new sample and the training sample in the feature space, which is calculated by the RBF kernel function.

[0279] Accuracy verification: In order to comprehensively evaluate the actual application effect of the proposed method in tunnel slag state measurement, detailed indoor accuracy verification experiments were designed for slag moisture content prediction, volume measurement, and conveyor belt speed measurement. Through these verification experiments, the accuracy of the PLS-SVR model in moisture content prediction, the accuracy of the AE-FPFH-ICP algorithm combined with the TACA method in slag volume measurement, and the accuracy and robustness of the lidar system in conveyor belt speed measurement were systematically evaluated.

[0280] Verification of moisture content prediction accuracy: In the verification of soil moisture content prediction accuracy, 20 sets of spectral data that did not participate in model training were first prepared as verification sets, which covered soil moisture content at different proportions from 0% to 100%. These verification set data were input into the trained PLS-SVR combined model, and the moisture content prediction value output by the model was compared with the actual moisture content.

[0281] In order to quantify the prediction accuracy of the model, the mean square error (MSE) and the coefficient of determination (R2) were selected as the main evaluation indicators. MSE measures the average square error between the predicted value and the actual value, and the formula is:

[0282]

[0283] Among them, y i represents the actual water content of the i-th sample, represents the predicted water content of the ith sample, and n represents the total number of samples. The lower the value of MSE, the higher the prediction accuracy of the model. 2 It reflects the model's ability to explain the data and is defined as:

[0284]

[0285] in, is the mean of the actual values. The closer the value is to 1, the better the model fits the actual situation.

[0286] The experimental results show that the PLS-SVR model performs well in predicting soil moisture content, with a mean square error (MSE) of 0.02 and a determination coefficient (R 2 It reached 0.90. This shows that the PLS-SVR model can effectively capture the complex relationship between soil moisture content and spectral characteristics, and has high prediction accuracy and reliability.

[0287] Verification of volume measurement accuracy: In the verification of the volume measurement accuracy of slag, a simulated experimental environment was built, and slag samples of different volumes were used as test objects. The dual laser radar system was used for volume measurement. The working principle of the dual laser radar system is to monitor the conveyor belt speed in real time through one laser radar, and collect point cloud data of the slag section through the other laser radar. Based on the high-precision point cloud registration of the AE-FPFH-ICP algorithm, the "Triangle Accumulated Cross-sectional Area Method (TACA)" is used to calculate the cross-sectional area, and the Simpson method is used for numerical integration to obtain the slag volume.

[0288] The volume measurement accuracy is evaluated by the relative error (RE), which is defined as follows:

[0289]

[0290] Among them, V actual is the actual volume of the sample, V predicted is the volume measured by the system. The smaller the relative error, the higher the accuracy of the volume measurement.

[0291] Experiments show that the relative error of the method of the present invention in volume measurement is controlled within 1.5%, which is significantly better than the traditional volume measurement method. This shows that the method has high reliability and accuracy in measuring the volume of slag and can accurately reflect the actual situation of the construction site.

[0292] Speed ​​measurement accuracy verification: In order to verify the accuracy of the laser radar system in conveyor belt speed measurement, the present invention designs a conveyor belt simulation system and uses a high-precision encoder to measure the actual speed of the conveyor belt as a comparison standard. The laser radar collects continuous point cloud data of the conveyor belt in real time, uses the AE-FPFH-ICP algorithm to calculate the displacement between two consecutive frames of point cloud, and then calculates the running speed of the conveyor belt.

[0293] The speed measurement accuracy is evaluated by the mean error (ME) and relative error. The mean error is calculated as:

[0294]

[0295] Among them, v actual,i and v predicted,i Respectively represent the actual speed and predicted speed of the i-th sample. The relative error is defined as:

[0296]

[0297] The experimental results show that the average speed measurement error of the lidar system is within 1%, and the relative error is within 1.2%, which fully demonstrates the high accuracy and stability of the system in conveyor belt speed measurement.

[0298] Through the above accuracy verification experiments, the slag state measurement method proposed in the scheme of the present invention has performed well in the three key links of slag moisture content prediction, volume measurement and conveyor belt speed measurement. The low MSE and high R2 of the PLS-SVR model in moisture content prediction verify that it can still maintain high prediction accuracy in complex environments. The accuracy of the AE-FPFH-ICP algorithm combined with the TACA method and the Simpson integral method in volume measurement has been strongly demonstrated, and the measurement error is controlled within 1.5%. In the speed measurement accuracy verification, the measurement results of the lidar system are highly consistent with the actual situation, indicating that it can work stably and efficiently in a complex tunnel environment. Overall, the method of the scheme of the present invention significantly surpasses the existing technology in terms of accuracy, efficiency, robustness and degree of automation, and provides a more reliable and efficient solution for slag volume measurement and moisture content prediction in tunnel construction.

[0299] The invention proposes an innovative slag status monitoring system MCMS, which combines a single-line laser radar, a near-infrared spectrometer and a 4K camera to achieve comprehensive monitoring of the slag volume and moisture content during shield tunnel construction.

[0300] The MCMS system designed by the scheme of the present invention realizes all-round monitoring of the volume, moisture content and transportation status of the slag in the shield tunnel construction by integrating a single-line laser radar, a near-infrared spectrometer and a 4K camera. The system ensures the comprehensiveness and accuracy of data collection in a complex tunnel environment through multi-sensor fusion. In addition, the scheme of the present invention improves the traditional Lucas-Kanade optical flow method, enhances the robustness and accuracy of the system in conveyor belt motion detection through adaptive window selection and multi-scale pyramid method, and effectively realizes the automatic management of the conveyor belt operation status. The scheme of the present invention proposes the "Triangle Accumulated Cross-sectional Area Method (TACA)", which simplifies the calculation process of the slag volume by decomposing the cross-sectional point cloud data into multiple triangles and accumulating their areas, and significantly improves the measurement accuracy. The improved AE-FPFH-ICP algorithm performs well in point cloud matching, improves efficiency and robustness through local optimization and feature enhancement, and realizes the automatic calculation of the conveyor belt speed, greatly improving the real-time and automation level of volume measurement. The PLS-SVR combined model proposed in the scheme of the present invention combines the linear dimension reduction capability of PLS ​​and the nonlinear modeling capability of SVR, and shows high accuracy and robustness in the detection of soil moisture content. The model performs well in dealing with complex nonlinear relationships, and can accurately predict the soil moisture content in real time. It has strong adaptability and meets the high-precision requirements of tunnel construction for moisture content detection. The multi-sensor fusion and algorithm improvement strategy proposed in the scheme of the present invention has shown significant advantages in accuracy, efficiency, degree of automation and adaptability, and provides reliable technical support for soil volume measurement and moisture content detection in tunnel construction. Future research can further optimize the real-time and robustness of the system, explore the application of this method under the transportation conditions of slag trucks, and expand its application potential in more tunnel construction scenarios.

[0301] To achieve the above object, the present invention also provides a soil moisture content prediction system applicable to shield tunnel construction, wherein the system is applied to the soil moisture content prediction method applicable to shield tunnel construction, such as Figure 8 As shown, the system specifically includes:

[0302] A data acquisition unit, used for acquiring first data corresponding to the muck to be measured in shield tunnel construction in real time based on an adaptive window; wherein the first data is the operation status data of the conveyor belt;

[0303] A data processing unit is used to combine the triangle cumulative cross-sectional area method and the AE-FPFH-ICP algorithm to generate second data and third data corresponding to the slag to be measured by parallel processing; wherein the second data is the actual cross-sectional area data of the slag; and the third data is the point cloud data corresponding to the slag operation environment and after local area optimization and feature enhancement processing;

[0304] Construct a generation unit for constructing a PLS-SVR combination model, and generate fourth data corresponding to the slag to be measured in real time based on the PLS-SVR combination model and in combination with the first data, the second data and the third data respectively; wherein the fourth data is the predicted data of the moisture content of the slag during the shield tunnel construction process.

[0305] The data acquisition unit further includes: a first acquisition module, which is used to respectively acquire fifth data and sixth data corresponding to the muck to be measured in the shield tunnel construction, and dynamically adjust the local window size in the optical flow calculation according to the fifth data; wherein the fifth data is the gradient intensity data of the local area of ​​the image; the sixth data is the original image data of the muck to be measured; the calculation formula for dynamically adjusting the local window size in the optical flow calculation according to the fifth data is as follows:

[0306]

[0307] Where W0 is the initial window size, α is the adjustment factor, Represents the gradient strength of the local area, and N is the total number of pixels in the window;

[0308] A processing and construction module, used for downsampling the sixth data layer by layer, and constructing at least one image pyramid from high resolution to low resolution;

[0309] A first generating module is used to generate optical flow vector data corresponding to each level of image based on the sixth data, and fuse and process the optical flow vector data layer by layer from bottom to top;

[0310] The processing and adjustment module is used to perform weighted accumulation processing on the optical flow vectors of all pixels in each window, and dynamically adjust the weights according to the gradient size and consistency of the pixels; the formula for weighted accumulation is:

[0311]

[0312] In the formula, [u i ,v i ] is the optical flow vector of the ith pixel, w i is the weight of the i-th pixel, and the weight calculation formula is:

[0313]

[0314] In the formula, and Represent the gradient of the pixel in the x and y directions respectively, and σ is a parameter used to adjust the weight distribution;

[0315] And / or, the data processing unit further includes:

[0316] The second generation module is used to calculate and generate the seventh data and the eighth data corresponding to the muck cross section respectively by combining the triangle cumulative cross-sectional area method, and generate the second data according to the seventh data and the eighth data; wherein the seventh data is the cross-sectional area data of the muck; the eighth data is the cross-sectional area data of the empty belt; the area of ​​each triangle is calculated by the following formula:

[0317]

[0318] In the formula, (x i ,y i ) and (x i+1 ,y i+1 ) are the coordinates of two adjacent points, (x o ,y o ) is the coordinate of the origin, and the areas of all triangles are added up to get the total cross-sectional area:

[0319]

[0320] In the formula, A total Represents the total cross-sectional area, A i represents the area of ​​the i-th triangle, and n is the total number of points;

[0321] The third generation module is used to generate and obtain feature data corresponding to the source point cloud and the target point cloud respectively, and fuse them into at least one joint feature descriptor;

[0322] The processing and creation module is used to standardize the joint feature vector and create a corresponding relationship between the source point cloud and the target point cloud; wherein the standardization formula is:

[0323]

[0324] In the formula, is the standardized joint feature descriptor, μ is the mean of the feature vector, and σ is the standard deviation of the feature vector;

[0325] The fourth generation module is used to generate matching pair geometric information corresponding to the corresponding relationship, and based on the matching pair geometric information, the corresponding local transformation matrix is ​​generated in combination with minimizing the error between the source point cloud and the target point cloud; wherein the error function of the local area is:

[0326]

[0327] In the formula, E i is the error of local region i, T i is the transformation matrix of the local area, and are points in the source point cloud and the target point cloud and / or respectively, the construction generation unit further includes:

[0328] A fifth generating module is used to generate and obtain ninth data corresponding to the slag to be measured; wherein the ninth data is the analysis training sample spectral information data corresponding to the slag to be measured;

[0329] A model building module, used to build a corresponding prediction model based on the ninth data and in combination with partial least squares regression and support vector regression methods;

[0330] The sixth generation module is used for processing the spectral data corresponding to the muck to be measured by standard normal transformation and generating a corresponding spectral data matrix;

[0331] A first processing module is used to extract the tenth data in the spectral data based on the PLS-SVR combination model, and maximize the correlation between the tenth data and the soil moisture content; wherein the tenth data is low-dimensional feature data;

[0332] The second processing module is used to process the tenth data by regression analysis in combination with a support vector regression method, and generate corresponding eleventh data; wherein the eleventh data is a hyperplane of optimal separation data corresponding to the muck to be measured;

[0333] The seventh generation module is used to generate the moisture content data corresponding to the muck to be measured according to the eleventh data and in combination with the SVR model; wherein the SVR model needs to meet the following constraints:

[0334]

[0335] Where w is the weight vector, b is the bias, which indicates the offset of the hyperplane, ò is the error tolerance, which defines the maximum prediction error range allowed by the model, and φ(T i ) is the kernel function, which maps the input features to a high-dimensional space. Here, the radial basis function (RBF) is selected as the kernel function, which is defined as follows:

[0336] K(T i ,T j )=exp(-γ‖T i -T j ‖ 2 ) (35)

[0337] In the formula, γ is the parameter of the kernel function, which controls the influence range of the RBF kernel function;

[0338] Among them, the specific prediction formula for water content data is:

[0339]

[0340] In the formula, is the predicted value, α i is the Lagrange multiplier obtained during the optimization process, T i is the feature matrix of the i-th training sample, represented by the low-dimensional features extracted by PLS, T is the feature matrix of the new sample, K(T i ,T) is the similarity between the new sample and the training sample in the feature space, which is calculated by the RBF kernel function.

[0341] In the system solution embodiment of the present invention, the method steps involved in the prediction of soil moisture content suitable for shield tunnel construction have been described above in detail, that is, the functional modules in the system are used to implement the steps or sub-steps in the above method embodiment, which will not be repeated here.

[0342] To achieve the above object, the present invention also provides a device for predicting soil moisture content suitable for shield tunnel construction, such as Figure 1-Figure 2 As shown, the device is applied to the method for predicting soil moisture content applicable to shield tunnel construction, and the device includes a detection bracket assembly and a sensor assembly;

[0343] The detection bracket assembly consists of four detachably connected parts: a rotating shaft, a connecting rod, a top connecting device and a box body; the sensor assembly includes at least two single-line laser radars for collecting cross-sectional point cloud and longitudinal point cloud data, at least one near-infrared spectrometer, at least two 4K high-definition cameras for collecting video files in front of and behind the equipment, and at least one 4K wide-angle camera for collecting video files directly below the equipment; and a near-infrared spectrometer for collecting near-infrared band spectral information of the soil below.

[0344] To achieve the above object, the present invention also provides a soil moisture content prediction platform suitable for shield tunnel construction, such as Fig. 9 As shown, it includes a processor, a memory, and a control program for a platform for predicting the moisture content of slag in shield tunnel construction; wherein the control program for predicting the moisture content of slag in shield tunnel construction is executed by the processor, the control program for predicting the moisture content of slag in shield tunnel construction is stored in the memory, and the control program for predicting the moisture content of slag in shield tunnel construction implements the steps of the method for predicting the moisture content of slag in shield tunnel construction. For example:

[0345] S1. Based on an adaptive window, first data corresponding to the slag to be measured in shield tunnel construction is acquired in real time; wherein the first data is the operation status data of the conveyor belt;

[0346] S2, combining the triangle cumulative cross-sectional area method and the AE-FPFH-ICP algorithm, parallel processing is performed to generate second data and third data corresponding to the slag to be measured; wherein the second data is the actual cross-sectional area data of the slag; and the third data is the point cloud data corresponding to the slag operation environment and after local area optimization and feature enhancement processing;

[0347] S3. Construct a PLS-SVR combined model, and generate in real time fourth data corresponding to the slag to be measured based on the PLS-SVR combined model and respectively combining the first data, the second data and the third data; wherein the fourth data is the predicted data of moisture content of the slag during the shield tunnel construction process.

[0348] The specific details of the steps have been explained above and will not be repeated here.

[0349] In the embodiment of the present invention, the built-in processor of the soil moisture content prediction platform applicable to shield tunnel construction can be composed of integrated circuits, for example, a single packaged integrated circuit, or a plurality of integrated circuits with the same or different functions, including one or more central processing units (CPU), microprocessors, digital processing chips, graphics processors, and combinations of various control chips. The processor uses various interfaces and lines to connect various components, and executes or executes programs or units stored in the memory, and calls data stored in the memory to perform various functions and process data for soil moisture content prediction applicable to shield tunnel construction;

[0350] The memory is used to store program codes and various data. It is installed in the soil moisture content prediction platform suitable for shield tunnel construction, and realizes high-speed and automatic access to programs or data during operation.

[0351] The memory includes read-only memory (ROM), random access memory (RAM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), one-time programmable read-only memory (OTPROM), electronically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, magnetic disk storage, magnetic tape storage, or any other computer-readable medium that can be used to carry or store data.

[0352] The present invention uses a method based on an adaptive window to obtain first data corresponding to the slag to be measured in shield tunnel construction in real time; wherein the first data is the operating status data of the conveyor belt; combined with the triangle cumulative cross-sectional area method and the AE-FPFH-ICP algorithm, parallel processing is performed to generate second data and third data corresponding to the slag to be measured; wherein the second data is the actual cross-sectional area data of the slag; the third data is point cloud data corresponding to the slag operating environment and after local area optimization and feature enhancement processing; a PLS-SVR combined model is constructed, and according to the PLS-SVR combined model, and in combination with the first data, the second data and the third data respectively, fourth data corresponding to the slag to be measured is generated in real time; wherein the fourth data is the predicted data of the moisture content of the slag during the shield tunnel construction process, and the system and platform corresponding to the method can quickly provide information about the chemical composition, especially in terms of moisture detection.

[0353] That is to say, the PLS-SVR combined model proposed in the scheme of the present invention combines the linear dimension reduction capability of PLS ​​and the nonlinear modeling capability of SVR, and shows high accuracy and robustness in the detection of soil moisture content. The model performs well in processing complex nonlinear relationships, can accurately predict the soil moisture content in real time, has strong adaptability, and meets the high-precision requirements of tunnel construction for moisture content detection.

[0354] In other words, the scheme of the present invention introduces an innovative muck condition monitoring system MCMS (Muck Condition Monitoring System), which integrates a single-line laser radar, a near-infrared spectrometer and a 4K camera to achieve comprehensive monitoring of the volume and moisture content of muck during tunnel construction. This scheme improves the traditional optical flow method, and realizes accurate detection of the running state of the conveyor belt through adaptive window selection, multi-scale pyramid method and weighted accumulation method, thereby realizing automatic power on and off of the equipment. In addition, the proposed "triangle cumulative cross-sectional area method (TACA)" improves the cross-sectional measurement accuracy by decomposing point cloud data and accumulating triangle areas. The developed AE-FPFH-ICP algorithm achieves efficient point cloud data matching and calculates the conveyor belt running speed through local optimization and feature enhancement, and combines TACA for volume calculation. The volume calculation uses the Simpson method for numerical integration to ensure accuracy. The PLS-SVR near-infrared spectroscopy prediction model is proposed to predict the moisture content of muck, and verification shows that the model has high accuracy and robustness. The system and algorithm can monitor the volume and moisture content of the belt conveyor slag system, providing reliable data support for tunnel construction.

[0355] The above-mentioned embodiments only express several implementation methods of the present invention, and the description thereof is relatively specific and detailed, but it cannot be understood as limiting the scope of the patent of the present invention. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

Claims

1. A method for predicting soil moisture content suitable for shield tunnel construction, characterized in that: The method comprises the steps of: Based on the adaptive window, first data corresponding to the slag to be measured in the shield tunnel construction is acquired in real time; wherein the first data is the operation status data of the conveyor belt; Combined with the triangle cumulative cross-sectional area method and the AE-FPFH-ICP algorithm, parallel processing is performed to generate the second data and the third data corresponding to the slag to be measured; wherein the second data is the actual cross-sectional area data of the slag; and the third data is the point cloud data corresponding to the slag operation environment and after local area optimization and feature enhancement processing; A PLS-SVR combined model is constructed, and fourth data corresponding to the slag to be measured is generated in real time based on the PLS-SVR combined model and in combination with the first data, the second data and the third data respectively; wherein the fourth data is the predicted data of the moisture content of the slag during the shield tunnel construction process.

2. A method for predicting soil moisture content in shield tunnel construction according to claim 1, characterized in that: The method of acquiring first data corresponding to the slag to be measured in shield tunnel construction in real time based on the adaptive window also includes: The fifth data and the sixth data corresponding to the muck to be measured in the shield tunnel construction are respectively obtained, and the local window size in the optical flow calculation is dynamically adjusted according to the fifth data; wherein the fifth data is the gradient intensity data of the local area of ​​the image; the sixth data is the original image data of the muck to be measured; the calculation formula for dynamically adjusting the local window size in the optical flow calculation according to the fifth data is as follows: Where W0 is the initial window size, α is the adjustment factor, Represents the gradient strength of the local area, and N is the total number of pixels in the window; Downsampling the sixth data layer by layer, and constructing at least one image pyramid from high resolution to low resolution; Based on the sixth data, optical flow vector data corresponding to each level of image is generated, and the optical flow vector data is fused layer by layer from bottom to top.

3. A method for predicting soil moisture content in shield tunnel construction according to claim 1 or 2, characterized in that: The method of acquiring first data corresponding to the slag to be measured in shield tunnel construction in real time based on the adaptive window also includes: In each window, the optical flow vectors of all pixels are weighted and accumulated, and the weights are dynamically adjusted according to the gradient size and consistency of the pixels; the formula for weighted accumulation is: In the formula, [u i ,v i ] is the optical flow vector of the ith pixel, w i is the weight of the i-th pixel, and the weight calculation formula is: In the formula, and They represent the gradient of the pixel in the x and y directions respectively, and σ is a parameter used to adjust the weight distribution.

4. The method for predicting soil moisture content in shield tunnel construction according to claim 1 is characterized in that: The method combines the triangle cumulative cross-sectional area method and the AE-FPFH-ICP algorithm to generate the second data and the third data corresponding to the muck to be measured by parallel processing, and also includes: Combined with the triangle cumulative cross-sectional area method, the seventh data and the eighth data corresponding to the muck cross section are calculated and generated respectively, and the second data is generated according to the seventh data and the eighth data; wherein the seventh data is the cross-sectional area data of the muck; the eighth data is the cross-sectional area data of the empty belt; the area of ​​each triangle is calculated by the following formula: In the formula, (x i ,y i ) and (x i+1 ,y i+1 ) are the coordinates of two adjacent points, (x o ,y o ) is the coordinate of the origin, and the areas of all triangles are added up to get the total cross-sectional area: In the formula, A total Represents the total cross-sectional area, A i represents the area of ​​the ith triangle, and n is the total number of points.

5. A method for predicting soil moisture content in shield tunnel construction according to claim 1 or 4, characterized in that: The method combines the triangle cumulative cross-sectional area method and the AE-FPFH-ICP algorithm to generate the second data and the third data corresponding to the muck to be measured by parallel processing, and also includes: Generate and obtain feature data corresponding to the source point cloud and the target point cloud respectively, and fuse them into at least one joint feature descriptor; The joint feature vector is normalized and a corresponding relationship between the source point cloud and the target point cloud is created; wherein the normalization formula is: In the formula, is the standardized joint feature descriptor, μ is the mean of the feature vector, and σ is the standard deviation of the feature vector; Generate matching pair geometric information corresponding to the corresponding relationship, and generate a corresponding local transformation matrix based on the matching pair geometric information and in combination with minimizing the error between the source point cloud and the target point cloud; wherein the error function of the local area is: In the formula, E i is the error of local region i, T i is the transformation matrix of the local area, and are the points in the source point cloud and the target point cloud respectively.

6. The method for predicting soil moisture content in shield tunnel construction according to claim 1, characterized in that: The constructing of the PLS-SVR combined model, and generating fourth data corresponding to the muck to be measured in real time according to the PLS-SVR combined model and combining the first data, the second data and the third data respectively, further includes: Generate and obtain ninth data corresponding to the muck to be measured; wherein the ninth data is analysis training sample spectral information data corresponding to the muck to be measured; According to the ninth data, and in combination with partial least squares regression and support vector regression methods, a corresponding prediction model is established.

7. A method for predicting soil moisture content in shield tunnel construction according to claim 1 or 6, characterized in that: The constructing of the PLS-SVR combined model, and generating fourth data corresponding to the muck to be measured in real time according to the PLS-SVR combined model and combining the first data, the second data and the third data respectively, further includes: The standard normal transformation processes the spectral data corresponding to the muck to be measured and generates the corresponding spectral data matrix; Based on the PLS-SVR combined model, the tenth data in the spectral data is extracted, and the correlation between the tenth data and the moisture content of the slag is maximized; wherein the tenth data is low-dimensional feature data; In combination with the support vector regression method, the tenth data is processed by regression analysis to generate corresponding eleventh data; wherein the eleventh data is a hyperplane of optimal separation data corresponding to the muck to be measured; According to the eleventh data, in combination with the SVR model, the moisture content data corresponding to the muck to be measured is generated; wherein the SVR model needs to meet the following constraints: Where w is the weight vector, b is the bias, which indicates the offset of the hyperplane, ò is the error tolerance, which defines the maximum prediction error range allowed by the model, and φ(T i ) is the kernel function, which maps the input features to a high-dimensional space. Here, the radial basis function (RBF) is selected as the kernel function, which is defined as follows: K(T i ,T j )=exp(-γ‖T i -T j ‖ 2 )(35) In the formula, γ is the parameter of the kernel function, which controls the influence range of the RBF kernel function; Among them, the specific prediction formula for moisture content data is: In the formula, is the predicted value, α i is the Lagrange multiplier obtained during the optimization process, T i is the feature matrix of the i-th training sample, represented by the low-dimensional features extracted by PLS, T is the feature matrix of the new sample, K(T i ,T) is the similarity between the new sample and the training sample in the feature space, which is calculated by the RBF kernel function.

8. A soil moisture content prediction system suitable for shield tunnel construction, characterized in that: The system is applied to the method for predicting soil moisture content applicable to shield tunnel construction as claimed in any one of claims 1 to 7, and the system comprises: A data acquisition unit, used for acquiring first data corresponding to the muck to be measured in shield tunnel construction in real time based on an adaptive window; wherein the first data is the operation status data of the conveyor belt; A data processing unit is used to combine the triangle cumulative cross-sectional area method and the AE-FPFH-ICP algorithm to generate second data and third data corresponding to the slag to be measured by parallel processing; wherein the second data is the actual cross-sectional area data of the slag; and the third data is the point cloud data corresponding to the slag operation environment and after local area optimization and feature enhancement processing; Construct a generation unit for constructing a PLS-SVR combination model, and generate fourth data corresponding to the slag to be measured in real time based on the PLS-SVR combination model and in combination with the first data, the second data and the third data respectively; wherein the fourth data is the predicted data of the moisture content of the slag during the shield tunnel construction process.

9. The soil moisture content prediction system suitable for shield tunnel construction according to claim 7, characterized in that: The data acquisition unit further includes: The first acquisition module is used to respectively acquire fifth data and sixth data corresponding to the muck to be measured in the shield tunnel construction, and dynamically adjust the local window size in the optical flow calculation according to the fifth data; wherein the fifth data is the gradient intensity data of the local area of ​​the image; the sixth data is the original image data of the muck to be measured; the calculation formula for dynamically adjusting the local window size in the optical flow calculation according to the fifth data is as follows: Where W0 is the initial window size, α is the adjustment factor, Represents the gradient strength of the local area, and N is the total number of pixels in the window; A processing and construction module, used for downsampling the sixth data layer by layer, and constructing at least one image pyramid from high resolution to low resolution; A first generating module is used to generate optical flow vector data corresponding to each level of image based on the sixth data, and fuse and process the optical flow vector data layer by layer from bottom to top; The processing and adjustment module is used to perform weighted accumulation processing on the optical flow vectors of all pixels in each window, and dynamically adjust the weights according to the gradient size and consistency of the pixels; the formula for weighted accumulation is: In the formula, [u i ,v i ] is the optical flow vector of the ith pixel, w i is the weight of the i-th pixel, and the weight calculation formula is: In the formula, and Represent the gradient of the pixel in the x and y directions respectively, and σ is a parameter used to adjust the weight distribution; And / or, the data processing unit further includes: The second generation module is used to calculate and generate the seventh data and the eighth data corresponding to the muck cross section respectively by combining the triangle cumulative cross-sectional area method, and generate the second data according to the seventh data and the eighth data; wherein the seventh data is the cross-sectional area data of the muck; the eighth data is the cross-sectional area data of the empty belt; the area of ​​each triangle is calculated by the following formula: In the formula, (x i ,y i ) and (x i+1 ,y i+1 ) are the coordinates of two adjacent points, (x o ,y o ) is the coordinate of the origin, and the areas of all triangles are added up to get the total cross-sectional area: In the formula, A total Represents the total cross-sectional area, A i represents the area of ​​the i-th triangle, and n is the total number of points; The third generation module is used to generate and obtain feature data corresponding to the source point cloud and the target point cloud respectively, and fuse them into at least one joint feature descriptor; The processing and creation module is used to standardize the joint feature vector and create a corresponding relationship between the source point cloud and the target point cloud; wherein the standardization formula is: In the formula, is the standardized joint feature descriptor, μ is the mean of the feature vector, and σ is the standard deviation of the feature vector; The fourth generation module is used to generate matching pair geometric information corresponding to the corresponding relationship, and based on the matching pair geometric information, the corresponding local transformation matrix is ​​generated in combination with minimizing the error between the source point cloud and the target point cloud; wherein the error function of the local area is: In the formula, E i is the error of local region i, T i is the transformation matrix of the local area, and are the points in the source point cloud and the target point cloud respectively. And / or, the construction generation unit further includes: A fifth generating module is used to generate and obtain ninth data corresponding to the slag to be measured; wherein the ninth data is the analysis training sample spectral information data corresponding to the slag to be measured; A model building module, used to build a corresponding prediction model based on the ninth data and in combination with partial least squares regression and support vector regression methods; The sixth generation module is used for processing the spectral data corresponding to the muck to be measured by standard normal transformation and generating a corresponding spectral data matrix; A first processing module is used to extract the tenth data in the spectral data based on the PLS-SVR combination model, and maximize the correlation between the tenth data and the soil moisture content; wherein the tenth data is low-dimensional feature data; The second processing module is used to process the tenth data by regression analysis in combination with a support vector regression method, and generate corresponding eleventh data; wherein the eleventh data is a hyperplane of optimal separation data corresponding to the muck to be measured; The seventh generation module is used to generate the moisture content data corresponding to the muck to be measured according to the eleventh data and in combination with the SVR model; wherein the SVR model needs to meet the following constraints: Where w is the weight vector, b is the bias, which indicates the offset of the hyperplane, ò is the error tolerance, which defines the maximum prediction error range allowed by the model, and φ(T i ) is the kernel function, which maps the input features to a high-dimensional space. Here, the radial basis function (RBF) is selected as the kernel function, which is defined as follows: K(T i ,T j )=exp(-γ‖T i -T j ‖ 2 )(35) In the formula, γ is the parameter of the kernel function, which controls the influence range of the RBF kernel function; Among them, the specific prediction formula for moisture content data is: In the formula, is the predicted value, α i is the Lagrange multiplier obtained during the optimization process, T i is the feature matrix of the i-th training sample, represented by the low-dimensional features extracted by PLS, T is the feature matrix of the new sample, K(T i ,T) is the similarity between the new sample and the training sample in the feature space, which is calculated by the RBF kernel function.

10. A soil moisture content prediction platform suitable for shield tunnel construction, characterized in that: It includes a processor, a memory, and a platform control program for predicting the moisture content of slag suitable for shield tunnel construction; wherein the platform control program for predicting the moisture content of slag suitable for shield tunnel construction is executed by the processor, the platform control program for predicting the moisture content of slag suitable for shield tunnel construction is stored in the memory, and the platform control program for predicting the moisture content of slag suitable for shield tunnel construction implements the method for predicting the moisture content of slag suitable for shield tunnel construction as described in any one of claims 1 to 7.

Images