Rock mass fracture prediction method and device and storage medium

Abstract

The invention discloses a rock mass fracture prediction method and device and a storage medium. A deep learning technology is utilized to accurately predict rock mass fractures. According to the method, data continuity is optimized by constructing a data set and applying an image interpolation algorithm, and then learning and verification are performed by using a space-time prediction network. The network captures the spatial and time dependency relationship between the feature maps through a self-attention mechanism, and realizes accurate capture of the dynamic change of the fracture. The technology not only improves the prediction precision and efficiency, but also significantly improves the resolution.

Description

Technical Field

[0001] The present invention relates to a rock mass fissure prediction method, device and storage medium, and belongs to the technical field of deep learning. Background Art

[0002] In the field of rock engineering, especially in tunnel engineering, the discontinuity of rock mass (such as cracks, joints, etc.) has a decisive influence on the stability and safety of the project. Accurate prediction of fracture distribution in rock mass is crucial for tunnel design, construction and subsequent maintenance. Traditionally, statistical analysis of rock mass fractures relies on geological exploration techniques and indirect methods such as sparse drilling and geophysical exploration. These methods are costly, inefficient, and have limited measurement resolution.

[0003] In view of the above problems, this patent aims to propose a new rock fracture prediction method to solve the problems of high cost, low efficiency and limited measurement resolution of existing technologies. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide a rock mass fracture prediction method, device and storage medium to overcome the deficiencies of the prior art.

[0005] The technical solution of the present invention is:

[0006] A first aspect: A rock mass fracture prediction method is provided, the method comprising the following steps:

[0007] S01, constructing the first rock mass fracture map data set during tunnel excavation;

[0008] S02, interpolating the image data of the rock mass fracture map data set by using an image interpolation algorithm to obtain a second rock mass fracture map data set sorted according to a set time interval;

[0009] S03, learning and verifying the preprocessed second rock mass fracture map data set through a spatiotemporal prediction network;

[0010] S04. Predict the rock mass fracture map at the next moment through the trained spatiotemporal prediction network.

[0011] Furthermore, the learning of the preprocessed second rock mass fracture map data set by the spatiotemporal prediction network includes the following steps:

[0012] S01, gradually extracting a feature map of the second rock mass fracture map data set through multi-layer convolution operations;

[0013] S02, for each layer of extracted feature graphs, the information stored in the memory unit at the previous moment and the feature graph at the current moment are used to update the feature graph at the current moment and the information stored in the memory unit at the current moment;

[0014] S03. The final feature map is subjected to multi-layer convolution operations to obtain the rock mass fracture map at the next moment.

[0015] Furthermore, the method for updating the feature map at this moment in step S02 is:

[0016] S02_1. Capture the spatial dependency between feature maps of each layer through the self-attention mechanism;

[0017] S02_2, the memory units corresponding to the same neuron before the current moment are used to capture the temporal dependency between memory units at different moments through the self-attention mechanism;

[0018] S02_3, calculating the weighted feature map of each layer of feature maps through the spatial dependency relationship between the feature maps of each layer and the feature maps of each layer at the current moment; calculating the weighted feature map of each layer of memory unit information at the previous moment through the temporal dependency relationship between memory units at different moments and the memory unit information at different moments;

[0019] S02_4, fusing the weighted feature map of each layer of feature maps and the weighted feature map of each layer of memory unit information at the previous moment to obtain the first fused feature map of each layer at the current moment;

[0020] S02_5. Perform channel connection based on the first fused feature map and the feature map of each layer at this moment to obtain a second fused feature map;

[0021] S02_6. Update the feature map of each layer at this moment through the second fusion feature map and the information stored in the memory unit at this moment.

[0022] Furthermore, the method for updating the information stored in the memory unit at this moment in step S02 is:

[0023] The memory unit storage information at this moment is updated through the second fusion feature map and the memory unit information of each layer at the previous moment.

[0024] Furthermore, the method for calculating the weighted feature map of each layer of feature maps through the spatial dependency relationship between the feature maps of each layer and the feature map of each layer at the current moment is as follows:

[0025]

[0026] Among them, Z h Represents the weighted feature map of each layer of feature map, a h；i,j The i-th row and j-th column of the spatial dependency matrix between feature maps of each layer, W hv Represents the weight matrix of each layer of feature map, H t；jIt represents the j-th feature map of each layer at the current moment, N represents the total number of feature maps of each layer, and the size of N is different for different layers.

[0027] Furthermore, the method for calculating the weighted feature mapping of each layer of memory unit information at the previous moment through the time dependency relationship between memory units at different moments and the memory unit information at different moments is as follows:

[0028]

[0029] Among them, Z m Represents the weighted feature map of each layer of feature map, a m；i,j The i-th row and j-th column of the time dependency matrix between memory units at different times, W mv Represents the weight matrix of the memory unit, M k represents the storage information of the memory unit at time k, and t-1 represents the previous time.

[0030] Furthermore, the method of fusing the weighted feature map of each layer of feature map and the weighted feature map of each layer of memory unit information at the previous moment to obtain the first fused feature map of each layer at the current moment is: Z h and Z m The feature maps of multiple channels are connected channel by channel.

[0031] Furthermore, the method for updating the storage information of the memory unit at the current moment by using the second fused feature map and the information of each layer of memory units at the previous moment is:

[0032]

[0033] Among them, M t Represents the storage information of the memory unit at time t, t is the current time, M t-1 shows the storage information of the memory unit at time t-1, t-1 is the previous time, symbol represents the Hadamard product;

[0034] i′ t =σ(W m；zi *Z+W m；hi *H t +b m；i )

[0035] g′ t =tanh(W m；zg *Z+W m；hg *H t +b m；g )

[0036] Among them, σ() represents the sigmoid activation function, W m；zirepresents the weight matrix connecting the second fusion feature map Z input to the input gate of the memory unit, W m；hi Indicates the connection feature graph H at this moment t The weight matrix of the input gate to the memory cell, b m；i represents the input gate bias term, tan() represents the hyperbolic tangent activation function, W m；zg Represents the connection of the second fusion feature map Z input to the candidate hidden state g′ t The weight matrix, W m；hg The characteristic diagram H at this moment is shown t Input to candidate hidden state g′ t The weight matrix, b m；g Hidden state g′ t The bias matrix.

[0037] The second aspect: provides a rock fracture prediction device, the device comprising: a processor and a memory, wherein the memory stores computer program instructions suitable for execution by the processor, and when the computer program instructions are executed by the processor, the processor executes the rock fracture prediction method described in the first aspect.

[0038] A third aspect: A storage medium is provided, comprising computer program instructions stored on the storage medium, wherein the computer program instructions are executed by a processor to execute the rock fracture prediction method as described in the first aspect.

[0039] The beneficial effects of the present invention are: compared with the prior art,

[0040] 1) The present invention constructs a rock fracture map data set and uses an image interpolation algorithm. The method can generate a rock fracture map with a continuous time series, which provides high-quality input data for the spatiotemporal prediction network. The spatiotemporal prediction network can comprehensively consider information in time and space, and predict future fracture states by learning past fracture data, thereby improving the accuracy and efficiency of the prediction. The fracture resolution predicted by the present invention is comparable to the image used in the training set, and the resolution is greatly improved compared with traditional geological exploration methods.

[0041] 3) The present invention uses a self-attention mechanism and a spatiotemporal model of deep learning to update and optimize the feature representation of the fracture map. The self-attention mechanism allows the model to capture the complex spatial and temporal dependencies between feature maps, and achieves dynamic integration of multi-layer features by calculating weighted feature maps and performing channel connections. In addition, by updating the memory unit to store information (using a gating mechanism and Hadamard product), the model is able to maintain temporal continuity and improve the accuracy of predictions of future fracture states. This method not only enhances the model's ability to capture dynamic changes in rock fractures, but also improves the precision and efficiency of the prediction, thereby more accurately capturing the changing patterns of rock fractures in space and time, and obtaining a more accurate rock fracture prediction map. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 This is a flow chart of Embodiment 1 of the present invention;

[0043] Figure 2 This is a structural block diagram of Embodiment 2 of the present invention;

[0044] Figure 3 This is a structural block diagram of embodiment 3 of the present invention. DETAILED DESCRIPTION

[0045] In order to better understand the above technical solution, the above technical solution will be described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0046] Implementation Example 1:

[0047] refer to Figure 1 , a rock mass fracture prediction method is provided, the method comprising the following steps:

[0048] S01, constructing the first rock mass fracture map data set during tunnel excavation;

[0049] S02, interpolating the image data of the rock mass fracture map data set by using an image interpolation algorithm to obtain a second rock mass fracture map data set sorted according to a set time interval;

[0050] S03, learning and verifying the preprocessed second rock mass fracture map data set through a spatiotemporal prediction network, where the spatiotemporal prediction network may be a ConvLSTM network;

[0051] S04. Predict the rock mass fracture map at the next moment through the trained spatiotemporal prediction network.

[0052] In order to more accurately obtain the variation pattern of the cracks in space and time, in this embodiment, preferably, the learning of the preprocessed second rock mass crack map data set by the spatiotemporal prediction network includes the following steps:

[0053] S01, gradually extracting a feature map of the second rock mass fracture map data set through multi-layer convolution operations;

[0054] S02, for each layer of extracted feature graphs, the information stored in the memory unit at the previous moment and the feature graph at the current moment are used to update the feature graph at the current moment and the information stored in the memory unit at the current moment;

[0055] S03. The final feature map is subjected to multi-layer convolution operations to obtain the rock mass fracture map at the next moment.

[0056] Working principle: The rock mass fracture prediction method used mainly constructs and processes rock mass fracture map datasets, and uses a spatiotemporal prediction network (such as ConvLSTM) to analyze and predict the spatial and temporal changes of fractures. First, an initial fracture map dataset is constructed and data continuity is optimized through an image interpolation algorithm. Then, the features of the fracture map are learned using a spatiotemporal prediction network, which combines convolutional layers to extract spatial features layer by layer, and uses memory units to maintain temporal continuity, ultimately generating a predicted rock mass fracture map for the next moment. This method can effectively capture the dynamic changes of fractures over time and space, thereby improving the accuracy and reliability of fracture prediction.

[0057] Furthermore, the method for updating the feature map at this moment in step S02 is:

[0058] S02_1. Capture the spatial dependency between feature maps of each layer through the self-attention mechanism;

[0059] S02_2, the memory units corresponding to the same neuron before the current moment are used to capture the temporal dependency between memory units at different moments through the self-attention mechanism;

[0060] S02_3, calculating the weighted feature map of each layer of feature maps through the spatial dependency relationship between the feature maps of each layer and the feature maps of each layer at the current moment; calculating the weighted feature map of each layer of memory unit information at the previous moment through the temporal dependency relationship between memory units at different moments and the memory unit information at different moments;

[0061] S02_4, fusing the weighted feature map of each layer of feature maps and the weighted feature map of each layer of memory unit information at the previous moment to obtain the first fused feature map of each layer at the current moment;

[0062] S02_5. Perform channel connection based on the first fused feature map and the feature map of each layer at this moment to obtain a second fused feature map;

[0063] S02_6. Update the feature map of each layer at this moment through the second fusion feature map and the information stored in the memory unit at this moment.

[0064] Furthermore, the method for updating the information stored in the memory unit at this moment in step S02 is:

[0065] The memory unit storage information at this moment is updated through the second fusion feature map and the memory unit information of each layer at the previous moment.

[0066] Furthermore, the method for calculating the weighted feature map of each layer of feature maps through the spatial dependency relationship between the feature maps of each layer and the feature map of each layer at the current moment is as follows:

[0067]

[0068] Among them, Z h Represents the weighted feature map of each layer of feature map, a h；i,j The i-th row and j-th column of the spatial dependency matrix between feature maps of each layer, W hv Represents the weight matrix of each layer of feature map, H t；j It represents the j-th feature map of each layer at the current moment, N represents the total number of feature maps of each layer, and the size of N is different for different layers.

[0069] Furthermore, the method for calculating the weighted feature mapping of each layer of memory unit information at the previous moment through the time dependency relationship between memory units at different moments and the memory unit information at different moments is as follows:

[0070]

[0071] Among them, Z m Represents the weighted feature map of each layer of feature map, a m；i,j The i-th row and j-th column of the time dependency matrix between memory units at different times, W mv Represents the weight matrix of the memory unit, M k represents the storage information of the memory unit at time k, and t-1 represents the previous time.

[0072] Furthermore, the method of fusing the weighted feature map of each layer of feature map and the weighted feature map of each layer of memory unit information at the previous moment to obtain the first fused feature map of each layer at the current moment is: Z n and Z m The feature maps of multiple channels are connected channel by channel.

[0073] Furthermore, the method for updating the storage information of the memory unit at the current moment by using the second fused feature map and the information of each layer of memory units at the previous moment is:

[0074]

[0075] Among them, M tRepresents the storage information of the memory unit at time t, t is the current time, M t-1 shows the storage information of the memory unit at time t-1, t-1 is the previous time, symbol represents the Hadamard product;

[0076] i′ t =σ(W m；zi *Z+W m；hi *H t +b m；i )

[0077] g′ t =tanh(W m；zg *Z+W m；hg *H t +b m；g )

[0078] Among them, σ() represents the sigmoid activation function, W m；zi represents the weight matrix connecting the second fusion feature map Z input to the input gate of the memory unit, W m；hi Indicates the connection feature graph H at this moment t The weight matrix of the input gate to the memory cell, b m；i represents the input gate bias term, tan() represents the hyperbolic tangent activation function, W m；zg Represents the connection of the second fusion feature map Z input to the candidate hidden state g′ t The weight matrix, W m；hg The characteristic diagram H at this moment is shown t Input to candidate hidden state g′ t The weight matrix, b m；g Hidden state g′ t The bias matrix of .

[0079] The present invention uses a self-attention mechanism and a spatiotemporal model of deep learning to update and optimize the feature representation of the fracture map. The self-attention mechanism allows the model to capture the complex spatial and temporal dependencies between feature maps, and achieves dynamic integration of multi-layer features by calculating weighted feature maps and performing channel connections. In addition, by updating the memory unit to store information (using a gating mechanism and Hadamard product), the model can maintain temporal continuity and improve the accuracy of predicting future fracture states. This method not only enhances the model's ability to capture dynamic changes in rock fractures, but also improves the precision and efficiency of predictions.

[0080] Embodiment 2:

[0081] At least one embodiment of the present disclosure also provides a rock fracture prediction device. Figure 2The present invention is a schematic block diagram of a rock mass fracture prediction device provided by at least one embodiment of the present disclosure. The rock mass fracture prediction device comprises: a processor and a memory, wherein the memory stores computer program instructions suitable for execution by the processor, and when the computer program instructions are executed by the processor, the processor executes any one of the rock mass fracture prediction methods in the first embodiment.

[0082] For example, the processor is a central processing unit (CPU), a graphics processor GPU, a tensor processing unit (TPU), or other forms of processing units with data processing capabilities and / or instruction execution capabilities. For example, the processor can be implemented as a general-purpose processor, and can also be a single-chip microcomputer, a microprocessor, a digital signal processor, a dedicated image processing chip, or a field programmable logic array, etc. For example, the memory may include at least one of a volatile memory and a non-volatile memory, for example, the memory may include a read-only memory (ROM), a hard disk, a flash memory, etc. Accordingly, the memory may be implemented as one or more computer program products, and the computer program product may include various forms of computer-readable storage media, and one or more computer program instructions may be stored on the computer-readable storage medium. The processor may run the program instructions to execute any rock mass fracture prediction method provided by at least one embodiment of the present disclosure. The memory may also store various other applications and various data, for example, various data used and / or generated by the application, etc.

[0083] Embodiment three:

[0084] At least one embodiment of the present disclosure also provides a storage medium (eg, a non-transitory storage medium). Figure 3 is a schematic block diagram of a storage medium provided by at least one embodiment of the present disclosure. Figure 3 As shown, the storage medium includes computer program instructions stored on the storage medium. When the computer program instructions are executed by a processor, the rock mass fracture prediction method provided by at least one embodiment of the present disclosure is executed.

[0085] For example, storage media may have many forms, including tangible storage media, carrier media or physical transmission media. Stable storage media may include: optical or magnetic disks, and other storage systems used in computers or similar devices that can implement the system components described in the figure. Unstable storage media may include dynamic memory, such as the main memory of a computer platform. Tangible transmission media may include coaxial cables, copper cables, and optical fibers, such as the lines that form a bus inside a computer system. Carrier transmission media can transmit electrical signals, electromagnetic signals, acoustic signals, or light signals. These signals can be generated by radio frequency or infrared data communication methods. Common storage media (e.g., computer readable media) include hard disks, floppy disks, magnetic tapes, any other magnetic media; CD-ROMs, DVDs, DVD-ROMs, any other optical media; punch cards, any other physical storage media containing a pattern of small holes; RAM, PROM, EPROM, FLASH-EPROM, any other memory chip or tape; carriers that transmit data or instructions, cables or connecting devices that transmit carriers, and any other data that can be read by computer program instructions (e.g., program code) and / or computers.

[0086] Computer program instructions (e.g., program code) for performing the operations of the present disclosure may be written in one or more programming languages, or a combination thereof, including, but not limited to, object-oriented programming languages, such as Java, Smalltalk, C++, and conventional procedural programming languages, such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., through the Internet using an Internet service provider).

[0087] In some examples, the functions described in at least one embodiment of the present disclosure may also be performed at least in part by one or more hardware logic components. For example, without limitation, exemplary types of hardware logic components that may be used include: field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), and the like.

[0088] The above contents are further detailed descriptions of the present invention in combination with specific preferred embodiments, and it cannot be determined that the specific implementation of the present invention is limited to these descriptions. For ordinary technicians in the technical field to which the present invention belongs, several simple deductions or substitutions can be made without departing from the concept of the present invention, which should be regarded as falling within the protection scope of the present invention.

Claims

1. A rock mass fracture prediction method, characterized in that: The method comprises the following steps: S01, constructing the first rock mass fracture map data set during tunnel excavation; S02, interpolating the image data of the rock mass fracture map data set by using an image interpolation algorithm to obtain a second rock mass fracture map data set sorted according to a set time interval; S03, learning and verifying the preprocessed second rock mass fracture map data set through a spatiotemporal prediction network; S04. Predict the rock mass fracture map at the next moment through the trained spatiotemporal prediction network.

2. The rock mass fracture prediction method according to claim 1, characterized in that: The learning of the preprocessed second rock mass fracture map data set by the spatiotemporal prediction network comprises the following steps: S01, gradually extracting a feature map of the second rock mass fracture map data set through multi-layer convolution operations; S02, for each layer of extracted feature graphs, the information stored in the memory unit at the previous moment and the feature graph at the current moment are used to update the feature graph at the current moment and the information stored in the memory unit at the current moment; S03. The final feature map is subjected to multi-layer convolution operations to obtain the rock mass fracture map at the next moment.

3. The rock mass fracture prediction method according to claim 2, characterized in that: The method for updating the feature map at this moment in step S02 is: S02_1. Capture the spatial dependency between feature maps of each layer through the self-attention mechanism; S02_2, the memory units corresponding to the same neuron before the current moment are used to capture the temporal dependency between memory units at different moments through the self-attention mechanism; S02_3, calculating the weighted feature map of each layer of feature maps through the spatial dependency relationship between the feature maps of each layer and the feature maps of each layer at the current moment; calculating the weighted feature map of each layer of memory unit information at the previous moment through the temporal dependency relationship between memory units at different moments and the memory unit information at different moments; S02_4, fusing the weighted feature map of each layer of feature maps and the weighted feature map of each layer of memory unit information at the previous moment to obtain the first fused feature map of each layer at the current moment; S02_5. Perform channel connection based on the first fused feature map and the feature map of each layer at this moment to obtain a second fused feature map; S02_6. Update the feature map of each layer at this moment through the second fusion feature map and the information stored in the memory unit at this moment.

4. The rock mass fracture prediction method according to claim 3, characterized in that: The method for updating the information stored in the memory unit at this moment in step S02 is: The memory unit storage information at this moment is updated through the second fusion feature map and the memory unit information of each layer at the previous moment.

5. The rock mass fracture prediction method according to claim 3, characterized in that: The method for calculating the weighted feature map of each layer of feature maps through the spatial dependency relationship between the feature maps of each layer and the feature map of each layer at the current moment is as follows: Among them, Z h Represents the weighted feature map of each layer of feature map, a h；i,j The i-th row and j-th column of the spatial dependency matrix between feature maps of each layer, W hv Represents the weight matrix of each layer of feature map, H t；j It represents the j-th feature map of each layer at the current moment, N represents the total number of feature maps of each layer, and the size of N is different for different layers.

6. The rock mass fracture prediction method according to claim 5, characterized in that: The method for calculating the weighted feature mapping of each layer of memory unit information at the previous moment through the time dependency relationship between memory units at different moments and the memory unit information at different moments is as follows: Among them, Z m Represents the weighted feature map of each layer of feature map, a m；i,j The i-th row and j-th column of the time dependency matrix between memory units at different times, W mv Represents the weight matrix of the memory unit, M k represents the storage information of the memory unit at time k, and t-1 represents the previous time.

7. The rock mass fracture prediction method according to claim 6, characterized in that: The method of fusing the weighted feature map of each layer of feature map and the weighted feature map of each layer of memory unit information at the previous moment to obtain the first fused feature map of each layer at the current moment is: Z h and Z m The feature maps of multiple channels are connected channel by channel.

8. The rock mass fracture prediction method according to claim 7, characterized in that: The method for updating the storage information of the memory unit at the current moment by using the second fusion feature map and the information of each layer of memory units at the previous moment is: Among them, M t Represents the storage information of the memory unit at time t, t is the current time, M t-1 shows the storage information of the memory unit at time t-1, t-1 is the previous time, symbol represents the Hadamard product; i′ t =σ(W m；zi *Z+W m；hi *H t +b m；i ) g t =tanh(W m；zg *Z+W m；hg *H t +b m；g ) Among them, σ() represents the sigmoid activation function, W m；zi represents the weight matrix connecting the second fusion feature map Z input to the input gate of the memory unit, W m；hi Indicates the connection feature graph H at this moment t The weight matrix of the input gate to the memory cell, b m ; i represents the input gate bias term, tan() represents the hyperbolic tangent activation function, W m；zg Represents the connection of the second fusion feature map Z input to the candidate hidden state g′ t The weight matrix, W m；hg The characteristic diagram H at this moment is shown t Input to candidate hidden state g′ t The weight matrix, b m；g Hidden state g′ t The bias matrix of .

9. A rock mass fracture prediction device, characterized in that: The device comprises: a processor and a memory, wherein the memory stores computer program instructions suitable for execution by the processor, and when the computer program instructions are executed by the processor, the processor executes the rock fracture prediction method according to any one of claims 1 to 8.

10. A storage medium comprising computer program instructions stored on the storage medium, wherein: When the computer program instructions are executed by a processor, the rock fracture prediction method according to any one of claims 1 to 8 is executed.

Images