Tunnel surrounding rock ground temperature field prediction method and system based on mining learning spatial features

Abstract

The invention discloses a tunnel surrounding rock ground temperature field prediction method and system based on mining learning spatial features, and relates to the technical field of tunnel surrounding rock ground temperature field prediction. The method comprises the following steps: acquiring temperature data to be processed, and normalizing the temperature data; constructing a three-dimensional temperature model, processing the temperature data by using the three-dimensional temperature model, designing a three-dimensional convolution kernel and three-dimensional pooling operation, mining spatial features in the temperature data, and learning to obtain a three-dimensional heat source distribution prediction result; and carrying out data point expansion and visualization operation on the three-dimensional heat source distribution prediction result. According to the invention, the three-dimensional space position of the high-temperature heat source can be rapidly and accurately positioned, the intensity distribution condition of the heat source can be clearly and visually presented, and comprehensive and visual thermal environment information is provided.

Description

Technical Field

[0001] The present invention relates to the technical field of tunnel surrounding rock geothermal field prediction, and in particular to a tunnel surrounding rock geothermal field prediction method and system based on mining and learning spatial features. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] In recent years, as the focus of transportation line construction in various countries around the world has gradually shifted to mountainous areas, a large number of deep buried tunnels have been built to overcome complex terrain conditions. However, with the increase in tunnel excavation depth, a series of severe problems have arisen. Among them, the continuous increase in surrounding rock temperature has become a key problem. A large number of studies have shown that there is a significant linear relationship between tunnel burial depth and rock temperature. Generally, with each increase in burial depth, the surrounding rock temperature will rise accordingly. At the same time, due to plate movement, geological tectonic activities, and underground heat flow, high-temperature areas are easily generated in the surrounding rock during tunnel construction. This high-temperature environment will not only cause changes in the physical and mechanical properties of the tunnel surrounding rock, such as reduced rock strength and increased deformation, which will in turn affect the stability of the tunnel; it will also pose a serious threat to the health of construction workers, reduce construction efficiency, and increase construction costs. For example, in a high-temperature environment, construction workers are prone to symptoms such as heat stroke and fatigue, which will increase the rate of work errors and hinder construction progress. In addition, high temperatures may also cause mechanical equipment failures in the tunnel, further affecting the smooth progress of construction.

[0004] In this context, it is particularly important to accurately grasp the detailed information of high-temperature heat sources in tunnels. However, traditional methods can no longer meet the needs in this regard. Due to the complex terrain of the tunnel and the irregular distribution of the temperature field, the spatial characteristics of the temperature field are difficult to be mined. The existing methods for predicting the geothermal field of the tunnel surrounding rock do not fully grasp the thermal environment information and cannot accurately locate the high-temperature heat source. Summary of the invention

[0005] In view of the shortcomings of the prior art, the purpose of the present invention is to provide a method and system for predicting the geothermal field of tunnel surrounding rock based on mining and learning spatial features, which can quickly and accurately locate the three-dimensional spatial position of high-temperature heat sources, and can also clearly and intuitively present the intensity distribution of heat sources, providing comprehensive and intuitive thermal environment information.

[0006] In order to achieve the above object, the present invention is implemented through the following technical solutions: The first aspect of the present invention provides a method for predicting the geothermal field of surrounding rock of a tunnel based on mining and learning spatial features, comprising the following steps: Obtain the temperature data to be processed and perform normalization on the temperature data; Construct a three-dimensional temperature model, process the temperature data using the three-dimensional temperature model, design a three-dimensional convolutional kernel and three-dimensional pooling operations to mine and learn the spatial features in the temperature data, and obtain the prediction result of the three-dimensional heat source distribution; Perform data point expansion and visualization operations on the prediction result of the three-dimensional heat source distribution.

[0007] Further, the specific steps for normalizing the temperature data are as follows: Reorder according to the sorting rules of the x, y, and z coordinates of the temperature data. Specifically, consider the temperature data with the same x value as the data on the same plane, the temperature data with the same x and y values as the data on the same line, and arrange them in ascending order of x, y, and z. Extract the corresponding temperature values and place them into a three-dimensional grid.

[0008] Further, the specific steps for constructing the three-dimensional temperature model are as follows: Adopt a 3D ConvNet network as the three-dimensional temperature model and determine the structural parameters of the three-dimensional temperature model; Obtain a training set, train the three-dimensional temperature model using the training set, and optimize the three-dimensional temperature model using cross-entropy error.

[0009] Even further, the training set is composed of temperature data of the simulated on-site test acquisition points with labels.

[0010] Even further, the three-dimensional temperature model includes a convolutional layer, a pooling layer, a normalization layer, and a fully connected layer. Among them, the convolutional layer adopts a deep structure containing 4 convolutional blocks.

[0011] Further, the specific steps for data point expansion of the prediction result of the three-dimensional heat source distribution are as follows: Use a cubic spline interpolation function to fit and obtain the interpolation function between the original coordinate data and the temperature data; Then substitute the newly generated coordinate data into the corresponding interpolation function to obtain the corresponding temperature value.

[0012] Even further, the specific steps for visualizing the prediction result of the three-dimensional heat source distribution are as follows: After the interpolation of the temperature data is completed, plot the interpolated temperature data, and assign different colors to different regions according to different temperature values to obtain a continuous and smooth three-dimensional visualization effect diagram.

[0013] The second aspect of the present invention provides a tunnel surrounding rock geothermal field prediction system based on mining and learning spatial features, including: A data acquisition module, configured to acquire the temperature data to be processed and normalize the temperature data; The temperature field prediction module is configured to construct a three-dimensional temperature model, process temperature data using the three-dimensional temperature model, design a three-dimensional convolution kernel and three-dimensional pooling operations to mine and learn the spatial features in the temperature data, and obtain a three-dimensional heat source distribution prediction result; The expansion and visualization module is configured to perform data point expansion and visualization operations on the three-dimensional heat source distribution prediction result.

[0014] In the third aspect of the present invention, a medium is provided, on which a program is stored, and when the program is executed by a processor, the steps in the tunnel surrounding rock geothermal field prediction method based on mining and learning spatial features as described in the first aspect of the present invention are implemented.

[0015] In the fourth aspect of the present invention, a device is provided, including a memory, a processor, and a program stored on the memory and executable on the processor. When the processor executes the program, the steps in the tunnel surrounding rock geothermal field prediction method based on mining and learning spatial features as described in the first aspect of the present invention are implemented.

[0016] The above one or more technical solutions have the following beneficial effects: The present invention discloses a tunnel surrounding rock geothermal field prediction method and system based on mining and learning spatial features. The high-temperature heat source prediction process covers multiple key steps. After data normalization processing, it is input into the 3D ConvNet network for training, giving full play to the advantages of the 3D ConvNet network, mining and learning spatial features, realizing accurate prediction and imaging of the three-dimensional heat source distribution, providing strong technical support for mastering the heat source situation in the tunnel, and helping to formulate effective protection measures.

[0017] The present invention uses the cubic spline interpolation method to process the inversion result data, interpolates and fills the data gaps on the specified yz plane, makes the visualization graph continuous and smooth, improves the resolution, and clearly shows the details of the temperature distribution. Drawing a three-dimensional graph of the interpolated data realizes an intuitive display, converts the abstract data into graphical information, helps to comprehensively grasp the spatial structure characteristics of the temperature field, provides an intuitive basis for tunnel engineering decision-making, and effectively improves the engineering quality and safety.

[0018] The advantages of the additional aspects of the present invention will be partially given in the following description, partially become obvious from the following description, or be understood through the practice of the present invention. Description of the Drawings

[0019] The specification drawings constituting a part of the present invention are used to provide a further understanding of the present invention. The schematic embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation to the present invention.

[0020] Figure 1 It is a flow chart of the tunnel surrounding rock geothermal field prediction method based on mining and learning spatial features in the first embodiment of the present invention; Figure 2 This is the visualization effect diagram in the first embodiment of the present invention; Figure 3 This is the three-dimensional solid effect diagram in the first embodiment of the present invention. Detailed implementation manners

[0021] It should be noted that the following detailed description is exemplary and is intended to provide further illustration of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the technical field to which the present invention belongs.

[0022] It should be noted that the terms used herein are only for describing specific implementation manners and are not intended to limit the exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular forms are also intended to include the plural forms. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof; Embodiment 1: The first embodiment of the present invention provides a method for predicting the geothermal field of tunnel surrounding rock based on mining the spatial features of the learning space. First, data normalization processing is aimed at making it meet the input requirements of the 3D ConvNet network, that is, arranging it in order into three-dimensional solid grid data; then inputting it into the 3D ConvNet network for training, using three-dimensional convolution kernels and three-dimensional pooling operations to deeply mine and learn the spatial features of the data, effectively extracting the key information of the data, reducing the redundancy and noise of the data, improving the recognition and prediction ability of the model, and saving the optimal tunnel surrounding rock temperature model; in order to obtain a continuous and smooth three-dimensional solid visualization effect diagram, the cubic spline interpolation algorithm is used to realize the expansion of data points in each direction, and finally the accurate prediction and imaging of the three-dimensional heat source distribution are realized.

[0023] As Figure 1 shown, it specifically includes the following steps: Step 1: Obtain the temperature data to be processed and perform normalization processing on the temperature data.

[0024] Step 1.1: Obtain the temperature data to be processed.

[0025] In a specific implementation manner, the temperature data to be processed includes in-hole temperature data and surface temperature data. Among them, the in-hole temperature data is collected by an optical fiber rod, and the surface temperature data is collected by an infrared camera.

[0026] Step 1.2: Perform normalization processing on the temperature data.

[0027] In a specific embodiment, re - sorting is performed according to the sorting rules of the temperature data in the x, y, and z coordinates. Specifically, the temperature data with the same x - value is regarded as the data on the same plane, the temperature data with the same x - value and y - value is regarded as the data on the same line, and they are arranged in ascending order of x, y, and z. Then, the corresponding temperature values are extracted and placed into a three - dimensional grid. The above process can be implemented by writing a data normalization processing module in Python. The processed data not only meets the data input requirements of the 3D ConvNet network but also further improves the spatial consistency and coherence of the data.

[0028] In this embodiment, Python software can be selected for data processing. Of course, in other embodiments, other software can be selected according to the situation.

[0029] Step 2: Construct a three - dimensional temperature model, use the three - dimensional temperature model to process the temperature data, design three - dimensional convolutional kernels and three - dimensional pooling operations to mine and learn the spatial features in the temperature data, and obtain the three - dimensional heat source distribution prediction result, as Figure 2 shown.

[0030] In a specific embodiment, this embodiment uses a 3D ConvNet network as the three - dimensional temperature model and determines the structure parameters of the three - dimensional temperature model; using a 3DCNN network for training has better spatial information modeling capabilities and retains the spatial information of the input signal. The three - dimensional temperature model includes a convolutional layer, a pooling layer, a normalization layer, and a fully - connected layer. Among them, the convolutional layer adopts a deep structure containing 4 convolutional blocks.

[0031] Specifically, select the network architecture: the normalized data is input into the 3D ConvNet network for training. The 3D ConvNet network sets the optimizer for updating the model parameters to stochastic gradient descent (SGD), the momentum to 0.9, and the epoch to 10000; the 3D ConvNet network uses three - dimensional convolutional kernels and three - dimensional pooling operations to deeply mine and learn the spatial features of the data, effectively extracts the key information of the data, reduces the redundancy and noise of the data, and improves the recognition and prediction capabilities of the model; in order to ensure that the constructed 3DCNN network architecture is both reasonable and robust, a normalization layer is introduced in all network structures to accelerate the training process and improve the generalization ability of the model; at the same time, a pooling layer is also equipped to reduce the dimension of the feature map, reduce the amount of calculation, and retain important features; finally, the features are mapped to the final result through the fully - connected layer to achieve accurate prediction.

[0032] Reasons for choosing 3DCNN in this embodiment: 3D ConvNet is a three-dimensional convolutional neural network well-suited for spatial feature learning. Compared with 2D ConvNet, due to 3D convolution and 3D pooling operations, 3D ConvNet has better spatial information modeling capabilities. In 3D ConvNets, convolution and pooling operations are performed spatially, while in 2D ConvNets, they are only done on a plane. 2D convolution applied to a single or multiple images will output an image. 2D ConvNets lose the spatial information of the input signal after each convolution operation. 3D convolution preserves the spatial information of the input signal and generates an output volume. In 3D ConvNet, the depth of the filter is smaller than the depth of the input layer (kernel size < channel size), so the 3D filter can move along all three directions (height, width, and the channels of the image). Since the filter slides through 3D space, the output values are also presented in the form of 3D space, and thus a 3D data is finally output. In addition, similar to the ability of 2D ConvNet to decode target spatial relationships in two-dimensional space, 3D ConvNet further expands this ability, enabling it to deeply explore and describe the complex spatial relationships between targets in three-dimensional space.

[0033] Determine the kernel size: A deep structure consisting of 4 convolutional blocks is adopted, with a kernel size of 3×3×3, and the stride of the kernel is uniformly set to 1. The 3DCNN network architecture is characterized by a deep structure with 4 convolutional blocks and a kernel size of 3×3×3. It is found through experiments that the loss value of the deep network is significantly reduced compared to the shallow structure. Especially when using a 3×3×3 kernel size, the network achieves optimal performance, being able to balance the performance of the model and the requirements of computing resources and realizing efficient and accurate feature extraction. The reason for choosing a deep structure with 4 convolutional blocks and a convolutional kernel size of 3×3×3: Two basic but representative 3D CNN network architectures were constructed for experiments. The 3D CNN network with a shallow structure has a relatively simple design and only contains two convolutional blocks; the 3D CNN network with a deep structure is more complex and contains four convolutional blocks. Inside each convolutional block, convolutional kernels of various sizes are configured, namely (1×1×1), (3×3×3), and (5×5×5). By comparing the loss values of 6 experimental schemes with different network structures and different convolutional kernel sizes, it is found that regardless of how the convolutional kernel size is adjusted, the 3D CNN network with a shallow structure is difficult to achieve effective convergence when dealing with complex datasets. This phenomenon may be attributed to the relatively small number of convolutional layers and pooling layers in the shallow network structure, resulting in limitations in feature extraction and processing of non-linear transformations. In contrast, through optimization strategies such as backpropagation algorithm and gradient descent, the deep network can continuously adjust its parameters to gradually reduce the loss function. In the experiment, it is found that the loss value of the deep network is significantly lower than that of the shallow structure. Especially when using a convolutional kernel of size 3×3×3, the network achieves the best performance. Although when the convolutional kernel size increases to 5×5×5, the training effect of the network still remains good, comparable to the case of the 3×3×3 convolutional kernel, but as the convolutional kernel size further increases, the number of model parameters and computational complexity increase sharply, and the demand for computing resources also rises significantly, resulting in a significant extension of the training time. Therefore, the deep network structure is finally selected, and 3×3×3 is determined as the optimal convolutional kernel size. This choice aims to balance the performance of the model and the demand for computing resources to achieve efficient and accurate feature extraction.

[0034] The reason for setting the stride of the convolutional kernel to 1: The strides of these convolutional kernels are all uniformly set to 1, aiming to not change the spatial dimension of the input data, thereby maintaining the integrity of information, and at the same time being able to extract feature information more deeply, ensuring the stability of the data size during the convolution process and avoiding unnecessary information loss.

[0035] Step 2.2: Obtain the training set, use the training set to train the three-dimensional temperature model, and optimize the three-dimensional temperature model using cross-entropy error.

[0036] Step 2.2.1: The training set consists of temperature data of the simulated on-site test collection points with labels.

[0037] In this embodiment, the use of data labels in the training session of the 3D ConvNet network is different from other projects. Although simulated temperature data is also used as data labels, these labels do not directly correspond to the positions of the input data; the input data comes from the temperature data of the simulated on-site test collection points, and the data labels are set based on the simulated data.

[0038] Different from the prior art, usually during the training process, for tagging, known data is tagged according to classification categories, and the tagged data is allocated as a training set and a test set according to a certain ratio, with a part for training and a part for testing. The tags are generally field names or custom numbers, and the output result after training the network is the tag. In this embodiment, the temperature data is tagged and divided according to a ratio of 7:3, with 7 parts as input data and 3 parts as output data, and these data are all used in the training stage. A 3DCNN network is used to learn the relationship between the input data and the output data, and finally a three-dimensional temperature model is obtained through multiple rounds of learning.

[0039] Specifically, a physical model with different heat source sizes, different heat source temperatures, and different heat source positions is established by using Comsol numerical simulation. The simulated temperature data at the specified position can be obtained by inserting probes at the specified positions. The temperature data is divided into two parts, one part as input data and the other part tagged as output data, and then input into the 3DCNN for training to learn the relationship between the input and the output. After optimization, a three-dimensional temperature model is obtained. The characteristic of the training set is that the temperature data is discontinuous, which are individual temperature surfaces in the drilling direction. As shown in the appendix Figure 2 shown, subsequent interpolation and data encryption are carried out to obtain a continuous and smooth predicted temperature field, as shown in the appendix Figure 3 shown.

[0040] Step 2.2.2: Optimize the three-dimensional temperature model by using cross-entropy error.

[0041] In a specific implementation manner, the cross-entropy between the predicted probability distribution and the true label is calculated voxel by voxel, and the formula is: ).

[0042] Among them, is the cross-entropy, N is the total number of voxels, C is the number of categories, the one-hot encoding of the true label, the predicted probability. Forward propagation, loss calculation, backpropagation, and parameter update are repeatedly executed in multiple epochs until the model converges, that is, the optimization is completed.

[0043] Step 2.3: Process the temperature data by using the three-dimensional temperature model.

[0044] Collect temperature data from on-site tests, including in-hole temperature data and surface temperature data. Input these data into the 3DCNN, and the temperature data at the output position corresponding to the tag can be obtained. Then, cubic spline interpolation is used to expand and encrypt the data points in each direction at unknown temperature positions. Finally, continuous and smooth visualization images are drawn through the matplotlib function in the built-in library of Python.

[0045] Step 3: Perform data point expansion and visualization on the three-dimensional heat source distribution prediction results.

[0046] It can be seen from Figure 2 that due to the sparse distribution of data points, there are a large number of visual gaps inside the three-dimensional structure, making it difficult to form a continuous and smooth temperature gradient display. The intervals between data points prevent the accurate and complete depiction of the temperature change trend of the tunnel surrounding rock. Therefore, expansion is needed to increase the data points.

[0047] Step 3.1: Expand the data points of the three-dimensional heat source distribution prediction results.

[0048] Step 3.1.1: Use a cubic spline interpolation function to fit and obtain the interpolation function between the original coordinate data and the temperature data.

[0049] To obtain a continuous and smooth three-dimensional visualization effect diagram, in this embodiment, the cubic spline interpolation algorithm is used to expand the data points in each direction. First, read the prediction data output by the 3DCNN network in the order of the x direction, y direction, z direction, and predicted temperature; divide it into four arrays; since interpolation expansion needs to be performed on multiple surfaces and multiple lines, it can be assumed that x = i (i is any x value output by the 3dcnn network). Taking the yz plane as an example, limit the range to the specified yz plane; process the y direction and z direction separately; use the cubic spline interpolation function; fit and obtain the interpolation function between the original coordinate data and the temperature data output by the three-dimensional temperature model. Specifically: when the y-axis direction is the same, 100 points are evenly inserted between every two points in the z-axis direction, and the corresponding coordinate values are saved; in this embodiment, the built-in spicy library of python is used to call the cubic spline interpolation function to fit and obtain the interpolation function between the original z-axis coordinate data and the temperature data.

[0050] Step 3.1.2: Then substitute the newly generated coordinate data into the corresponding interpolation function to obtain the corresponding temperature value.

[0051] Substitute the newly generated coordinate data into the corresponding interpolation function to obtain the corresponding temperature value, and at the same time save this value into the python container. The operations for the remaining lines and surfaces are the same as above, and a for loop function can be used to iterate through each step. When the loop exits, it represents the end of the code execution. At this time, the data in each direction has been expanded.

[0052] In this embodiment, python software can be selected for data processing. Of course, in other embodiments, other software can be selected according to the situation.

[0053] Step 3.2: Visualize the three-dimensional heat source distribution prediction results.

[0054] Using this display method of three-dimensional graphics, it is possible to visually observe the spatial distribution of temperature data; for a more intuitive display effect, the coordinates of the tunnel contour are obtained through simulation software, and the shape of the tunnel is dug out on the three-dimensional temperature map, and finally a three-dimensional temperature map with the tunnel shape is obtained, as Figure 3 shown. This three-dimensional visualization method not only makes the data more intuitive and easy to understand, but also provides strong support for subsequent analysis and decision-making.

[0055] In a specific embodiment, after the temperature data interpolation is completed, the interpolated temperature data is plotted. In this embodiment, the matplotlib library of Python code is used to plot the interpolated temperature data on a three-dimensional graph. During the plotting process, different colors are assigned to different regions according to different temperature values, and a continuous and smooth three-dimensional visualization effect diagram is obtained, so as to realize the visualization of temperature data. This display method of three-dimensional graphics makes it possible to visually observe the spatial distribution of temperature data.

[0056] The coordinates of the tunnel contour are obtained through the comsol simulation software, and the shape of the tunnel is dug out on the three-dimensional temperature map. Specifically, first, semi-circular coordinates of the tunnel contour size are generated according to the tunnel structure characteristics, combined with the temperature zone coordinates known in the previous steps, the data points inside the tunnel contour are deleted, and the data points outside the contour are retained, so that the shape of the tunnel can be dug out on the three-dimensional temperature map. Finally, a three-dimensional temperature map with the tunnel shape is obtained, as Figure 3 shown. This three-dimensional visualization method not only makes the data more intuitive and easy to understand, but also provides strong support for subsequent analysis and decision-making.

[0057] In this embodiment, the comsol simulation software can be selected for data processing. Of course, in other embodiments, other software can be selected according to the situation.

[0058] The present invention adopts an intuitive data visualization technical process for processing tunnel temperature data. Combining the advantages of the 3DConvNet network for heat source inversion to achieve accurate imaging, using the cubic spline interpolation method to process the inversion results and draw a three-dimensional graph to make the data visualization intuitive and effective, providing comprehensive and reliable technical support for the temperature field evaluation, heat source control, decision-making, etc. of tunnel engineering, and effectively ensuring the project quality and safety.

[0059] Embodiment 2: Embodiment 2 of the present invention provides a tunnel surrounding rock geothermal field prediction system based on mining learning spatial features, including: A data acquisition module, configured to acquire the temperature data to be processed and perform normalization processing on the temperature data; A temperature field prediction module, configured to construct a three-dimensional temperature model, process temperature data using the three-dimensional temperature model, design a three-dimensional convolution kernel and three-dimensional pooling operations to mine and learn spatial features in the temperature data, and obtain a three-dimensional heat source distribution prediction result; An expansion and visualization module, configured to perform data point expansion and visualization operations on the three-dimensional heat source distribution prediction result.

[0060] Embodiment III: Embodiment III of the present invention provides a medium, on which a program is stored, and when the program is executed by a processor, it implements the steps in the power demand response method based on the data matching algorithm as described in Embodiment I of the present invention. The steps are as follows: The detailed steps are the same as those of the power demand response method based on the data matching algorithm provided in Embodiment I, and will not be elaborated here.

[0061] Embodiment IV: Embodiment IV of the present invention provides a device, including a memory, a processor, and a program stored on the memory and executable on the processor. When the processor executes the program, it implements the steps in the power demand response method based on the data matching algorithm as described in Embodiment I of the present invention. The steps are as follows: The detailed steps are the same as those of the power demand response method based on the data matching algorithm provided in Embodiment I, and will not be elaborated here.

[0062] The steps involved in the above Embodiments II, III, and IV correspond to those of Method Embodiment I. For specific implementation manners, reference may be made to the relevant description part of Embodiment I.

[0063] Those skilled in the art should understand that the above-mentioned modules or steps of the present invention can be implemented by a general-purpose computer device. Optionally, they can be implemented by program codes executable by a computing device, so that they can be stored in a storage device and executed by the computing device, or they can be separately fabricated into individual integrated circuit modules, or multiple modules or steps among them can be fabricated into a single integrated circuit module to be implemented. The present invention is not limited to any specific combination of hardware and software.

[0064] Although the specific implementation manners of the present invention are described above in conjunction with the accompanying drawings, it is not a limitation to the protection scope of the present invention. Those skilled in the art should understand that based on the technical solutions of the present invention, various modifications or deformations that can be made without creative efforts by those skilled in the art are still within the protection scope of the present invention.

Claims

1. A method for predicting the geothermal field of tunnel surrounding rock based on mining and learning spatial features, characterized in that: The following steps are involved: Obtain the temperature data to be processed and perform normalization on the temperature data; Construct a three-dimensional temperature model, use the three-dimensional temperature model to process the temperature data, design three-dimensional convolution kernels and three-dimensional pooling operations to mine and learn the spatial features in the temperature data, and obtain the three-dimensional heat source distribution prediction results; Perform data point expansion and visualization operations on the three-dimensional heat source distribution prediction results.

2. The method for predicting the geothermal field of surrounding rock of a tunnel based on mining and learning spatial features according to claim 1, characterized in that: The specific steps for normalizing temperature data are as follows: Rearrange the temperature data according to the sorting rules of the x, y, and z coordinates. Specifically, treat the temperature data with the same x value as data on the same surface, and the temperature data with the same x and y values ​​as data on the same line. Arrange x, y, and z in ascending order, extract the corresponding temperature values, and place them in the three-dimensional grid.

3. The method for predicting the geothermal field of surrounding rock of a tunnel based on mining and learning spatial features according to claim 1, characterized in that: The specific steps to construct a three-dimensional temperature model are: A 3D ConvNet network is used as a three-dimensional temperature model, and the structural parameters of the three-dimensional temperature model are determined; A training set is obtained, the three-dimensional temperature model is trained using the training set, and the three-dimensional temperature model is optimized using the cross entropy error.

4. The method for predicting the geothermal field of surrounding rock of a tunnel based on mining and learning spatial features according to claim 3, characterized in that: The training set consists of labeled temperature data from simulated field test collection points.

5. The method for predicting the geothermal field of surrounding rock of a tunnel based on mining and learning spatial features according to claim 3, characterized in that: The three-dimensional temperature model includes a convolution layer, a pooling layer, a normalization layer and a fully connected layer, wherein the convolution layer adopts a deep structure containing 4 convolution blocks.

6. The method for predicting the geothermal field of surrounding rock of a tunnel based on mining and learning spatial features according to claim 1, characterized in that: The specific steps for expanding the data points of the three-dimensional heat source distribution prediction results are as follows: The interpolation function between the original coordinate data and the temperature data is obtained by fitting the cubic spline interpolation function; Then substitute the newly generated coordinate data into the corresponding interpolation function to obtain the corresponding temperature value.

7. The method for predicting the geothermal field of surrounding rock of a tunnel based on mining and learning spatial features according to claim 6, characterized in that: The specific steps for visualizing the three-dimensional heat source distribution prediction results are: After the temperature data interpolation is completed, the interpolated temperature data is plotted, and different areas are assigned different colors according to different temperature values ​​to obtain a continuous and smooth three-dimensional visualization effect map.

8. A tunnel surrounding rock geothermal field prediction system based on mining and learning spatial features, characterized in that: include: A data acquisition module is configured to acquire temperature data to be processed and perform normalization processing on the temperature data; The temperature field prediction module is configured to construct a three-dimensional temperature model, use the three-dimensional temperature model to process the temperature data, design three-dimensional convolution kernels and three-dimensional pooling operations to mine and learn the spatial features in the temperature data, and obtain the three-dimensional heat source distribution prediction results; The expansion and visualization module is configured to perform data point expansion and visualization operations on the three-dimensional heat source distribution prediction results.

9. A computer-readable storage medium, characterized in that: A plurality of instructions are stored therein, and the instructions are suitable for being loaded by a processor of a terminal device and executing the method for predicting the geothermal field of surrounding rock of a tunnel based on mining and learning spatial features as described in any one of claims 1-7.

10. A terminal device, characterized in that: It includes a processor and a computer-readable storage medium, the processor is used to implement each instruction; the computer-readable storage medium is used to store multiple instructions, and the instructions are suitable for being loaded by the processor and executing the tunnel surrounding rock geothermal field prediction method based on mining and learning spatial features as described in any one of claims 1-7.

Images