Lithology real-time prediction method in tunneling process of tunnel boring machine

Abstract

The invention relates to the technical field of tunneling machine intelligent control, in particular to a lithology real-time prediction method in the tunneling process of a tunneling machine, which comprises the following steps: detecting effective data in the tunneling process of the tunneling machine; inputting the effective data into the lithology prediction model, and obtaining an output lithology prediction result; during the period, dividing the valid data into at least one feature block according to the size of the feature block; obtaining a time domain attention matrix, a frequency domain attention matrix, a frequency domain feature and a time domain feature corresponding to the feature block according to the set LSTM network and the mixed attention mechanism; based on the time domain attention matrix, the frequency domain attention matrix, the frequency domain features and the time domain features, obtaining a lithology prediction result; and judging the working state of the next tunneling circulating tunnel boring machine according to the lithological prediction result. The method has the beneficial effects that local features in address data are analyzed more meticulously while real-time lithology prediction is realized in the TBM construction process, so that the sensitivity to lithology change is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent control of tunnel boring machines, and in particular to a real-time prediction method for lithology during the tunneling process of a tunnel boring machine. Background Art

[0002] As a highly efficient underground engineering machinery and equipment, Tunnel Boring Machine (TBM) occupies an important position in modern underground engineering construction and is widely used in water conservancy tunnels, mine tunnels, subway projects and other large-scale infrastructure projects. The advantage of TBM is that it can approach complex geological conditions with relatively high efficiency, while ensuring the safety of construction through automated control systems. Compared with traditional drilling and blasting methods, TBM automated operation significantly reduces dependence on manpower and also reduces the impact on the environment, such as less noise pollution and dust pollution. However, TBM still has certain limitations in geological adaptability, and its performance is easily affected by different geological conditions. Therefore, the widespread application of TBM has certain limitations.

[0003] Under complex geological conditions, the performance of TBMs varies significantly. For example, in hard rock environments, the wear rate of TBM cutters will increase significantly, while in soft rock, the tunneling resistance will increase. Therefore, rock properties have become a key factor affecting TBM construction efficiency, cutter wear, and construction safety. Traditional rock property prediction methods rely on geological surveys, advance drilling, and geophysical exploration, which cannot achieve real-time rock property prediction, and the above methods require a lot of manpower and material resources. In order to improve the adaptability of TBMs under different geological conditions, real-time rock property prediction has become an important technical means. Summary of the invention

[0004] 1. Technical issues to be resolved In view of the above-mentioned shortcomings and deficiencies of the prior art, the present invention provides a real-time prediction method for rock properties during the excavation process of a tunnel boring machine, which solves the technical problems that traditional rock property prediction methods rely on geological surveys, advance drilling, geophysical exploration and other methods, cannot achieve real-time prediction of rock properties, and the above methods require a lot of manpower and material resources.

[0005] (II) Technical solution In order to achieve the above object, the main technical solutions adopted by the present invention include: In a first aspect, an embodiment of the present invention provides a method for real-time prediction of rock properties during tunneling of a tunnel boring machine, comprising: S11, real-time detection of valid data corresponding to each excavation cycle during the excavation process of the tunnel boring machine; S12, inputting the obtained valid data into a pre-trained lithology prediction model to obtain a lithology prediction result output by the lithology prediction model; During this period, characteristic parameters corresponding to the effective data are obtained according to the effective data corresponding to each excavation cycle; According to the obtained characteristic parameters corresponding to the valid data and the preset characteristic block size, the characteristic parameters corresponding to the valid data are divided into at least one characteristic block, and the time domain signal and the frequency domain signal corresponding to each characteristic block are obtained; The frequency domain signal and time domain signal corresponding to each feature block are respectively input into the pre-set LSTM network to obtain the frequency domain features and time domain features corresponding to the feature block; According to the frequency domain features and time domain features corresponding to each feature block and the pre-set hybrid attention mechanism, the time domain attention matrix and frequency domain attention matrix corresponding to each feature block are obtained; Based on the time domain attention matrix, frequency domain attention matrix, frequency domain features and time domain features corresponding to each feature block, the lithology prediction results corresponding to the effective data are obtained; S13, judging the working state of the tunnel boring machine in the next excavation cycle according to the lithology prediction result corresponding to the valid data.

[0006] Optionally, each excavation cycle includes: a rising section and a stabilization section; The valid data corresponding to each excavation cycle includes: the state parameters of each unit time in the first preset time period in the ascending section corresponding to the excavation cycle and the state parameters of each unit time in the second preset time period in the stable section; Then the S11 includes: Real-time detection of the state parameters corresponding to each unit time in each excavation cycle during the tunnel boring machine excavation process, and screening of all state parameters in each excavation cycle to obtain valid data corresponding to each excavation cycle; Each status parameter includes: cutter head speed, excavation speed, total thrust, cutter head torque, shield pressure, gripper pressure, gripper pump pressure, main engine belt conveyor pump pressure and main engine belt conveyor speed.

[0007] Optionally, the feature parameters include batch size, sequence length and feature dimension; Then the S12, according to the characteristic parameters corresponding to the obtained valid data and the preset characteristic block size, divides the valid data into at least one characteristic block, and obtains the time domain signal and the frequency domain signal corresponding to each characteristic block, including: According to the sequence length and feature dimension corresponding to the obtained valid data, the preset feature block size and the preset formula 1, the valid data is divided into at least one feature block, each feature block has its corresponding sequence length and feature dimension; the formula 1 is: ; Wherein, P is the number of feature blocks corresponding to valid data, T is the sequence length corresponding to valid data, and L is the preset feature block size; According to the sequence length and feature dimension corresponding to each feature block and the preset formula 2, each feature block is reshaped to obtain the time domain signal corresponding to each feature block; the formula 2 is: X time =X[:,:P×L,:].view(B,P,L,D); Among them, X time is the time domain signal corresponding to the feature block, B is the batch size corresponding to the valid data, D is the feature dimension corresponding to the feature block, view is the reshaping function, and X[:,:P×L,:] is the slicing operation on the valid data; According to the time domain signal corresponding to each feature block and the preset formula 3, the frequency domain signal corresponding to each feature block is obtained; the formula 3 is: X freq =Re(FFT(X time ,dim=2)); Among them, X freq is the frequency domain signal corresponding to the feature block, Re means taking the real part of the complex number, FFT is the fast Fourier transform, dim=2 means operating in the second dimension, X time is the time domain signal corresponding to the feature block.

[0008] Optionally, the LSTM network includes a time domain LSTM network and a frequency domain LSTM network; Then, the S12, respectively inputting the frequency domain signal and the time domain signal corresponding to each feature block into a preset LSTM network to obtain the frequency domain features and the time domain features corresponding to the feature block, includes: The frequency domain signal corresponding to the feature block is input into the preset frequency domain LSTM network to capture the frequency domain features corresponding to the feature block; wherein the processing process of the frequency domain LSTM network is as follows: L freq ,(h freq ,c freq )=LSTM freq (X freq ); Among them, L freq is the frequency domain feature corresponding to the feature block, LSTM freq is the frequency domain LSTM network, X freq is the frequency domain signal corresponding to the feature block, h freq is the short-term memory parameter in the frequency domain LSTM network, c freqis the long-term memory parameter in the frequency domain LSTM network; The time domain signal corresponding to the feature block is input into a preset time domain LSTM network to capture the time domain features corresponding to the feature block; wherein the processing process of the time domain LSTM network is as follows: L time ,(h time ,c time )=LSTM time (X time ); Among them, L time is the time domain feature corresponding to the feature block, LSTM time is the time domain LSTM network, X time is the time domain signal corresponding to the feature block, h time is the short-term memory parameter in the time domain LSTM network, c time is the long-term memory parameter in the time domain LSTM network.

[0009] Optionally, the hybrid attention mechanism includes a multi-head attention mechanism and a multi-head convolutional projection attention mechanism; Then, the S12, according to the frequency domain features and time domain features corresponding to each feature block, and the pre-set hybrid attention mechanism, obtains the time domain attention matrix and the frequency domain attention matrix corresponding to each feature block, including: Based on the time domain features and frequency domain features corresponding to all feature blocks, the time domain features and frequency domain features corresponding to each feature block are normalized respectively; According to the time domain features corresponding to each feature block, the preset hybrid attention mechanism and the preset formula 4, the time domain attention matrix corresponding to the feature block is obtained; The formula 4 is: O time =α 1 Attention time (Q r ,K r ,V r )+β 1 Attention time (Q c ,K c ,V c ); Among them, O time is the time domain attention matrix corresponding to the time domain features of the feature block, Attention time (Q r ,K r ,V r ) is the output matrix of the multi-head convolution projection attention mechanism corresponding to the time domain features of the feature block; Attentiontime (Q c ,K c ,V c ) is the output matrix of the multi-head attention mechanism corresponding to the temporal features of the feature block, α 1 is the pre-set multi-head convolution projection attention weight of the time domain feature, β 1 The multi-head attention weights for the pre-set temporal features; At the same time, according to the frequency domain features corresponding to each feature block, the preset hybrid attention mechanism and the preset formula 5, the frequency domain attention matrix corresponding to the feature block is obtained; The formula five is: O freq =α 2 Attention freq (Q r ,K r ,V r )+β 2 Attention freq (Q c ,K c, V c ); Among them, O freq is the frequency domain attention matrix corresponding to the frequency domain features of the feature block, Attention freq (Q r ,K r ,V r ) is the output matrix of the multi-head convolution projection attention mechanism corresponding to the frequency domain features of the feature block; Attention freq (Q c ,K c, V c ) is the output matrix of the multi-head attention mechanism corresponding to the frequency domain features of the feature block, α 2 is the pre-set multi-head convolution projection attention weight of the frequency domain feature, β 2 is the multi-head attention weights of the pre-set frequency domain features.

[0010] Optionally, the S12, based on the time domain attention matrix, frequency domain attention matrix, frequency domain features and time domain features corresponding to each feature block, obtains the lithology prediction result corresponding to the valid data, including: According to the time domain attention matrix, frequency domain attention matrix, time domain features and frequency domain features corresponding to each feature block, the fusion features corresponding to each feature block are obtained; According to the fusion features corresponding to each feature block and the pre-set weighted attention mechanism, the global information of all feature blocks is obtained, that is, the global context vector corresponding to the valid data; According to the global context vector corresponding to the valid data, the lithology prediction result corresponding to the valid data is obtained.

[0011] Optionally, the S12, obtaining a fusion feature corresponding to each feature block according to a time domain attention matrix, a frequency domain attention matrix, a time domain feature, and a frequency domain feature corresponding to each feature block, includes: The time domain attention matrix and the frequency domain attention matrix corresponding to each feature block are respectively input into the preset first fully connected layer, and feature transformation is performed based on the preset formula 6; ; Among them, A time is the transformation result of the time domain attention matrix corresponding to the feature block, W time is the preset time domain weight matrix, O time is the temporal attention matrix corresponding to the temporal features of the feature block; A freq is the transformation result of the frequency domain attention matrix corresponding to the feature block, W freq is the preset time domain weight matrix, O freq is the frequency domain attention matrix corresponding to the frequency domain features of the feature block; Then, based on the preset formula 7, the transformation result of the time domain attention matrix corresponding to each feature block is residually connected with the time domain feature corresponding to the feature block, and the transformation result of the frequency domain attention matrix corresponding to each feature block is residually connected with the frequency domain feature corresponding to the feature block; the formula 7 is: ; Among them, R time is the residual connection result of the feature block in the time domain, L time is the time domain feature corresponding to the feature block, R freq is the residual connection result of the feature block in the frequency domain, L freq is the time domain feature corresponding to the feature block; Finally, based on the preset formula eight, the residual connection results of each feature block in the frequency domain and the residual connection results in the time domain are feature spliced ​​to obtain the fusion features corresponding to each feature block; the formula eight is: C = concat(R time ,R freq ,dim=2); Where C is the fusion feature corresponding to the feature block, concat is the concatenation operation, and dim=2 is the operation on the second feature dimension.

[0012] Optionally, the S12, according to the fusion feature corresponding to each feature block and a preset weighted attention mechanism, obtains the global information of all feature blocks, including: According to the fusion features corresponding to each feature block and the preset attention score weight matrix, and the preset formula nine, the attention score matrix corresponding to each feature block is obtained; the formula nine is: S=CW score ; Among them, S is the attention score matrix corresponding to the feature block, C is the fusion feature corresponding to the feature block, and W score is the pre-set attention score weight matrix; Then, the attention score matrix corresponding to each feature block is normalized by the preset softmax function to generate the attention weight corresponding to each feature block; Finally, according to the attention weight and fusion features corresponding to each feature block and the preset formula 10, the global context vector corresponding to the valid data is obtained; the formula 10 is: ; Among them, A is the attention weight corresponding to the feature block, c is the global context vector corresponding to the valid data, C is the fusion feature corresponding to the feature block, i is the index of the feature block, and P is the number of feature blocks corresponding to the valid data.

[0013] Optionally, the S12, obtaining the lithology prediction result corresponding to the valid data according to the global context vector corresponding to the valid data, includes: The fusion feature corresponding to the last feature block in all feature blocks of valid data is concatenated with the global context vector corresponding to the valid data to obtain the connection vector corresponding to the valid data; the specific formula is as follows: p=concat(c,H t ); Among them, p is the connection vector corresponding to the valid data, concat is the concatenation operation, c is the global context vector corresponding to the valid data, and H t is the fusion feature corresponding to the last feature block; Subsequently, the fusion features corresponding to the valid data are input into the preset second fully connected layer, and the lithology prediction results corresponding to the valid data are obtained based on the preset formula 11; the formula 11 is: ; Among them, W f and W p are all pre-set weights of the second fully connected layer, v is the intermediate feature generated by nonlinear transformation, p is the connection vector corresponding to the valid data, and O is the lithology prediction result corresponding to the valid data.

[0014] Optionally, before S11, the step further includes: S10, training the lithology prediction model based on a preset model training set to obtain a trained lithology prediction model; During this period, the hyperparameters of the lithology prediction model were optimized in real time based on the improved EWOA-GWO-MRFO and the pre-set model training strategy; The model training strategy includes: When in the exploration stage, the optimal hyperparameter is found based on the preset formula 12; the formula 12 is: ; Among them, X t+1 and X t are the positions of the gray wolf individuals at the t+1th iteration and the tth iteration, respectively, α is the individual with the best fitness, A is the step size dynamic adjustment factor, a is the dynamic control factor, r 1 is a random number in the range [0,1], sin(2πl) is a nonlinear perturbation term, X -☆ is the global optimal solution of the current population; When in the predation phase, the optimal hyperparameters are found based on the pre-set formula 13; the formula 13 is: X t+1 =X β -A·|2·r 2 ·X β -X t |; Among them, X β is the individual with the second best fitness, r 2 is a random number in the range [0,1]; When in the spiral predation stage, the optimal hyperparameter is found based on the pre-set formula 14; the formula 14 is: X t+1= |X ☆ -X t |·e b·l cos(2πl)+η·(X rand -X t )+γ·(X a -X t ); Among them, l is a random factor with a value range of [0,1], η and γ are dynamic control factors of jump and collaboration respectively, b is a constant, and X rand Any individual in a population.

[0015] (III) Beneficial effects The beneficial effects of the present invention are as follows: a real-time lithology prediction method during tunnel boring machine excavation of the present invention adopts a lithology prediction model to predict the current lithology in real time during TBM operation, and decomposes the data into small segments for analysis. Compared with the prior art, it can realize real-time lithology prediction during TBM construction and more carefully analyze the local characteristics in geological data, thereby improving the sensitivity to lithology changes. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 The present invention is a flow chart of a method for real-time prediction of rock properties during tunnel boring machine excavation. DETAILED DESCRIPTION

[0017] In order to better explain the present invention and facilitate understanding, the present invention is described in detail below through specific implementation modes in conjunction with the accompanying drawings.

[0018] A method for real-time prediction of lithology during tunnel boring machine excavation proposed in an embodiment of the present invention adopts a novel lithology prediction model, including a patch mechanism, an LSTM network with time domain and frequency domain decomposition, a hybrid attention mechanism, residual connection and normalization, and a global attention mechanism. By decomposing the data into small fragments, the Patch mechanism allows the model to analyze the local features in the geological data more carefully, thereby improving the sensitivity to lithology changes. The time domain data is converted into frequency domain data using Fourier transform to capture the periodic characteristics and frequency information in the geological data. Subsequently, an LSTM network with time domain and frequency domain decomposition is adopted. The time domain LSTM network can effectively memorize and utilize long-term time dependency information, and the frequency domain LSTM network can identify periodic and oscillatory features in the data. The hybrid attention mechanism combines dynamic convolution projection and multi-head attention mechanism to achieve comprehensive capture of data features. That is, through the local connection characteristics of the convolution layer, the dynamic convolution projection module can effectively extract the spatial and frequency features in the data. Multi-head attention allows the model to focus on multiple feature dimensions at the same time, enhancing the ability to analyze complex data. The hybrid attention mechanism can dynamically adjust the focus of attention to cope with different geological features and changes. In order to further improve the stability and training efficiency of the model, residual connections and normalization techniques are added to the model. Residual connections can alleviate the problems of gradient disappearance and gradient explosion. Batch Normalization ensures the stability of data distribution and reduces internal covariance shift. The global attention mechanism allows the model to consider global context information when processing data, not just local features.

[0019] In order to better understand the above technical solution, exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided to enable a clearer and more thorough understanding of the present invention and to fully convey the scope of the present invention to those skilled in the art.

[0020] Example 1

[0021] This embodiment proposes a real-time prediction method for rock properties during tunnel boring by a tunnel boring machine. Figure 1 As shown, including: S11, real-time detection of valid data corresponding to each excavation cycle during the excavation process of the tunnel boring machine; S12, inputting the obtained valid data into a pre-trained lithology prediction model to obtain a lithology prediction result output by the lithology prediction model; During this period, characteristic parameters corresponding to the effective data are obtained according to the effective data corresponding to each excavation cycle; According to the obtained characteristic parameters corresponding to the valid data and the preset characteristic block size, the characteristic parameters corresponding to the valid data are divided into at least one characteristic block, and the time domain signal and the frequency domain signal corresponding to each characteristic block are obtained; The frequency domain signal and time domain signal corresponding to each feature block are respectively input into the pre-set LSTM network to obtain the frequency domain features and time domain features corresponding to the feature block; According to the frequency domain features and time domain features corresponding to each feature block and the pre-set hybrid attention mechanism, the time domain attention matrix and frequency domain attention matrix corresponding to each feature block are obtained; Based on the time domain attention matrix, frequency domain attention matrix, frequency domain features and time domain features corresponding to each feature block, the lithology prediction results corresponding to the valid data are obtained; the lithology prediction results include limestone, granite, tuffaceous sandstone, diorite, etc.

[0022] S13, judging the working state of the tunnel boring machine in the next excavation cycle according to the lithology prediction result corresponding to the valid data.

[0023] This embodiment provides a method for real-time prediction of lithology during tunnel boring machine excavation. It uses a lithology prediction model to predict the current lithology in real time during TBM operation, and decomposes the data into small segments for analysis. Compared with the existing technology, it can realize real-time lithology prediction during TBM construction, and analyze the local characteristics of geological data in more detail, thereby improving the sensitivity to lithology changes.

[0024] Example 2

[0025] This embodiment provides a method for real-time prediction of rock properties during tunneling by a tunnel boring machine, comprising: S11, real-time detection of valid data corresponding to each excavation cycle during the excavation process of the tunnel boring machine; S12, inputting the obtained valid data into a pre-trained lithology prediction model to obtain a lithology prediction result output by the lithology prediction model; S13. The lithology prediction result corresponding to the valid data is used to determine the working state of the tunnel boring machine in the next excavation cycle.

[0026] Among them, the excavation cycle of a tunnel boring machine usually includes an ascending section and a stable section; the ascending section refers to the stage that the tunnel boring machine goes through when it just starts to excavate, also known as the "starting section" or "advancement section"; the stable section refers to the stage when the tunnel boring machine is close to reaching the target position during the excavation process; in the ascending section, the tunnel boring machine gradually accelerates from a stationary state and begins to excavate a tunnel in the underground rock and soil. In the stable section, the tunnel boring machine is close to reaching the target position and needs to excavate steadily, adapt to the underground rock and soil conditions and make necessary adjustments.

[0027] The valid data corresponding to each excavation cycle include: the state parameters of each unit time in the first preset time period in the ascending segment corresponding to the excavation cycle and the state parameters of each unit time in the second preset time period in the stable segment; usually the state parameters of the first 10 seconds of the ascending segment and the first 15 seconds of the stable segment are selected, that is, including these 25 seconds, the state parameters corresponding to each second; the state parameters per second include: cutter head speed, excavation speed, total thrust, cutter head torque, shield pressure, support shoe pressure, support shoe pump pressure, main engine belt conveyor pump pressure and main engine belt conveyor speed, etc.

[0028] The above S11 specifically includes: real-time detection of the state parameters corresponding to each unit time (usually 1s) in each excavation cycle during the tunnel boring machine excavation process, and screening all the state parameters in each excavation cycle to obtain valid data corresponding to each excavation cycle.

[0029] In the above S12, the processing process of the lithology prediction model specifically includes: According to the effective data corresponding to each excavation cycle, characteristic parameters corresponding to the effective data are obtained; the characteristic parameters include batch size, sequence length and characteristic dimension; According to the sequence length and feature dimension corresponding to the obtained valid data, the preset feature block size and the preset formula 1, the valid data is divided into at least one feature block, each feature block has its corresponding sequence length and feature dimension; the formula 1 is: ; Wherein, P is the number of feature blocks corresponding to valid data, T is the sequence length corresponding to valid data, and L is the preset feature block size; According to the sequence length and feature dimension corresponding to each feature block and the preset formula 2, each feature block is reshaped to obtain the time domain signal corresponding to each feature block; the formula 2 is: X time =X[:,:P×L,:].view(B,P,L,D); Among them, X time is the time domain signal corresponding to the feature block, B is the batch size corresponding to the valid data, D is the feature dimension corresponding to the feature block, view is the reshaping function, and X[:,:P×L,:] is the slicing operation on the valid data; According to the time domain signal corresponding to each feature block obtained, the time domain signal is converted into a frequency domain signal using fast Fourier transform, so that we can capture the periodicity and trend information in the data, that is, the frequency domain signal corresponding to each feature block is obtained through formula three; the formula three is: X freq =Re(FFT(X time ,dim=2)); Among them, X freq is the frequency domain signal corresponding to the feature block, Re means taking the real part of the complex number, FFT is the fast Fourier transform, dim=2 means operating in the second dimension, X time is the time domain signal corresponding to the feature block.

[0030] The frequency domain signal and time domain signal corresponding to each feature block are input into the pre-set frequency domain LSTM network and time domain LSTM network respectively; That is, the frequency domain signal corresponding to the feature block is input into a preset frequency domain LSTM network to capture the frequency domain features corresponding to the feature block; wherein the processing process of the frequency domain LSTM network is as follows: L freq ,(h freq ,c freq )=LSTM freq (X freq ); Among them, L freq is the frequency domain feature corresponding to the feature block, LSTM freq is the frequency domain LSTM network, X freq is the frequency domain signal corresponding to the feature block, h freq is the short-term memory parameter in the frequency domain LSTM network, c freq is the long-term memory parameter in the frequency domain LSTM network; The time domain signal corresponding to the feature block is input into a preset time domain LSTM network to capture the time domain features corresponding to the feature block; wherein the processing process of the time domain LSTM network is as follows: L time ,(h time ,c time )=LSTM time (X time ); Among them, L time is the time domain feature corresponding to the feature block, LSTM time is the time domain LSTM network, X time is the time domain signal corresponding to the feature block, h time is the short-term memory parameter in the time domain LSTM network, c time is the long-term memory parameter in the time domain LSTM network.

[0031] Based on the time domain features and frequency domain features corresponding to all feature blocks, the time domain features and frequency domain features corresponding to each feature block are normalized respectively; Based on the preset hybrid attention mechanism, the time domain features and frequency domain features corresponding to each feature block are processed respectively to obtain the time domain attention matrix and frequency domain attention matrix corresponding to the feature block; For the time domain features, the processing is performed by formula 4, which is: O time =α 1 Attention time (Q r ,K r ,V r )+β 1 Attention time (Q c ,K c ,V c ); Among them, O time is the time domain attention matrix corresponding to the time domain features of the feature block, Attention time (Q r ,K r ,V r ) is the output matrix of the multi-head convolution projection attention mechanism corresponding to the time domain features of the feature block, Attention time (Q c ,K c ,V c ) is the output matrix of the multi-head attention mechanism corresponding to the temporal features of the feature block, α 1 is the pre-set multi-head convolution projection attention weight of the time domain feature, β 1The multi-head attention weights for the pre-set temporal features; The frequency domain features are processed by Formula 5, which is: O freq =α 2 Attention freq (Q r ,K r ,V r )+β 2 Attention freq (Q c ,K c, V c ); Among them, O freq is the frequency domain attention matrix corresponding to the frequency domain features of the feature block, Attention freq (Q r ,K r ,V r ) is the output matrix of the multi-head convolution projection attention mechanism corresponding to the frequency domain features of the feature block; Attention freq (Q c ,K c, V c ) is the output matrix of the multi-head attention mechanism corresponding to the frequency domain features of the feature block, α 2 is the pre-set multi-head convolution projection attention weight of the frequency domain feature, β 2 is the multi-head attention weights of the pre-set frequency domain features.

[0032] The time domain attention matrix and the frequency domain attention matrix corresponding to each feature block are respectively input into the preset first fully connected layer, and feature transformation is performed based on the preset formula 6; ; Among them, A time is the transformation result of the time domain attention matrix corresponding to the feature block, W time is the preset time domain weight matrix, O time is the temporal attention matrix corresponding to the temporal features of the feature block; A freq is the transformation result of the frequency domain attention matrix corresponding to the feature block, W freq is the preset time domain weight matrix, O freq is the frequency domain attention matrix corresponding to the frequency domain features of the feature block; Then, based on the preset formula 7, the transformation result of the time domain attention matrix corresponding to each feature block is residually connected with the time domain feature corresponding to the feature block, and the transformation result of the frequency domain attention matrix corresponding to each feature block is residually connected with the frequency domain feature corresponding to the feature block; the formula 7 is: ; Among them, R time is the residual connection result of the feature block in the time domain, L time is the time domain feature corresponding to the feature block, R freq is the residual connection result of the feature block in the frequency domain, L freq is the time domain feature corresponding to the feature block; Finally, based on the preset formula eight, the residual connection results of each feature block in the frequency domain and the residual connection results in the time domain are feature spliced ​​to obtain the fusion features corresponding to each feature block; the formula eight is: C = concat(R time ,R freq ,dim=2); Where C is the fusion feature corresponding to the feature block, concat is the concatenation operation, and dim=2 is the operation on the second feature dimension.

[0033] According to the fusion features corresponding to each feature block and the preset attention score weight matrix, and the preset formula nine, the attention score matrix corresponding to each feature block is obtained; the formula nine is: S=CW score ; Among them, S is the attention score matrix corresponding to the feature block, C is the fusion feature corresponding to the feature block, and W score is the pre-set attention score weight matrix; Then, the attention score matrix corresponding to each feature block is normalized by the preset softmax function to generate the attention weight corresponding to each feature block; Finally, according to the attention weight and fusion features corresponding to each feature block and the preset formula 10, the global context vector corresponding to the valid data is obtained; the formula 10 is: ; Among them, A is the attention weight corresponding to the feature block, c is the global context vector corresponding to the valid data, C is the fusion feature corresponding to the feature block, i is the index of the feature block, and P is the number of feature blocks corresponding to the valid data.

[0034] The fusion feature corresponding to the last feature block in all feature blocks of valid data is concatenated with the global context vector corresponding to the valid data to obtain the connection vector corresponding to the valid data; the specific formula is as follows: p=concat(c,H t ); Among them, p is the connection vector corresponding to the valid data, concat is the concatenation operation, c is the global context vector corresponding to the valid data, and H tis the fusion feature corresponding to the last feature block; Subsequently, the fusion features corresponding to the valid data are input into the preset second fully connected layer, and the lithology prediction results corresponding to the valid data are obtained based on the preset formula 11; the formula 11 is: ; Among them, W f and W p are all pre-set weights of the second fully connected layer, v is the intermediate feature generated by nonlinear transformation, p is the connection vector corresponding to the valid data, O is the lithology prediction result corresponding to the valid data, tanh is the hyperbolic tangent function, and dropout is regularization.

[0035] Traditional CNNs are limited to focusing on local information and have difficulty capturing long-distance dependencies in data, especially when dealing with complex time series data. Although 1D-CNN can be applied to time series data, it still cannot fully capture global time dependencies. LSTM networks give the same weight to inputs at different time steps, making it difficult to flexibly adjust the importance of each time step. Transformers are not good at capturing local details because they tend to evenly distribute attention globally.

[0036] The lithology prediction model provided in this embodiment includes a patch mechanism, an LSTM network with time domain and frequency domain decomposition, a hybrid attention mechanism, residual connection and normalization, and a global attention mechanism. By decomposing the data into small fragments, the Patch mechanism allows the model to analyze the local features in the geological data more carefully, thereby improving the sensitivity to lithology changes. The time domain data is converted into frequency domain data using Fourier transform to capture the periodic characteristics and frequency information in the geological data. Subsequently, an LSTM network with time domain and frequency domain decomposition was used. The time domain LSTM network can effectively memorize and utilize long-term time dependency information, and the frequency domain LSTM network can identify periodic and oscillatory features in the data. The hybrid attention mechanism combines dynamic convolution projection and multi-head attention mechanism to achieve comprehensive capture of data features, that is, through the local connection characteristics of the convolution layer, the dynamic convolution projection module can effectively extract the spatial and frequency features in the data, and the multi-head attention allows the model to focus on multiple feature dimensions at the same time, enhancing the ability to analyze complex data. The hybrid attention mechanism can dynamically adjust the focus of attention to cope with different geological features and changes. In order to further improve the stability and training efficiency of the model, residual connections and normalization techniques are added to the model. Residual connections can alleviate the problems of gradient vanishing and gradient exploding, and Batch Normalization ensures the stability of data distribution and reduces internal covariance shift. The global attention mechanism allows the model to consider global context information when processing data, rather than just local features.

[0037] Example 3

[0038] This embodiment provides a lithology prediction model training method, including: Acquire a model training set for training a lithology prediction model, wherein the model training includes state parameters of a tunnel boring machine under various lithology conditions; Based on the model training set, the lithology prediction model is trained to obtain a trained lithology prediction model; During model training, the hyperparameters of the lithology prediction model are optimized based on the improved EWOA-GWO-MRFO and the pre-set model training strategy; The model training strategy includes: when in the exploration stage, finding the optimal hyperparameter based on the preset formula 12; the formula 12 is: ; Among them, X t+1 and X t are the positions of the gray wolf individuals at the t+1th iteration and the tth iteration, respectively, a is the individual with the best fitness, A is the step size dynamic adjustment factor, a is the dynamic control factor, r 1 is a random number in the range [0,1], sin(2πl) is a nonlinear perturbation term, X ☆ is the global optimal solution of the current population; When in the predation phase, the optimal hyperparameters are found based on the pre-set formula 13; the formula 13 is: X t+1 =X β -A·|2·r 2 ·X β -X t |; Among them, X β is the individual with the second best fitness, r 2 is a random number in the range [0,1]; When in the spiral predation stage, the optimal hyperparameter is found based on the pre-set formula 14; the formula 14 is: X t+1= |X ☆ -X t |·e b·l cos(2πl)+η·(X rand -X t )+γ·(X a -X t ); Among them, l is a random factor with a value range of [0,1], η and γ are dynamic control factors of jump and collaboration respectively, b is a constant, and Xrand Any individual in a population.

[0039] The model proposed in this embodiment has many hyperparameters, such as learning rate, number of neurons, number of network layers, etc. In order to achieve appropriate performance of the model, the improved EWOA-GWO-MRFO is used to optimize the hyperparameters of the model. In order to improve the performance of the whale optimization algorithm (EWOA), the α wolf and β wolf guidance mechanisms of the gray wolf optimization algorithm (GWO) are introduced into the exploration phase and the surrounding predation phase, and the vortex predation and chain predation behaviors of the manta ray optimization algorithm (MRFO) are introduced into the spiral predation phase, and the whale spiral search formula is retained. This improved scheme not only enhances the global search capability and local search accuracy, but also adds jump and mutation mechanisms to avoid falling into local optimality.

[0040] In the exploration phase, the whale optimization algorithm performs global search by simulating random perturbations. To improve the search efficiency, the guidance behavior of the leader wolf (α wolf) in GWO is combined. Through the guidance of α wolf, individual whales can efficiently approach the global optimal solution.

[0041] In the hunting phase, the whales gradually approached the prey, and then introduced the sub-leader wolf (β wolf) behavior of the gray wolf algorithm, so that the individual whales can more accurately approach the optimal solution through collaboration during local search. The guidance behavior of the sub-leader wolf (β wolf) enhances the whale's search ability in the local area and avoids falling into the local optimum.

[0042] In the spiral predation stage, the core formula of the whale optimization algorithm is retained, and the vortex predation and chain predation of MFO are combined to enhance the flexibility and jumping ability of local search. The classic path formula of the whale algorithm is retained to simulate the spiral movement of whales around prey; vortex predation: through the random jumping mechanism, the search is made more flexible and avoids falling into the local optimum; chain predation: through the cooperative behavior between individuals, the accuracy of local search is enhanced.

[0043] Example 4 This embodiment provides a method for real-time prediction of rock properties during tunneling by a tunnel boring machine, comprising: In TBM construction, real-time monitoring and analysis of state parameters are crucial. These data are usually time series data, including cutter head speed, excavation speed, total thrust, cutter head torque, shield pressure, gripper pressure, gripper pump pressure, main machine belt pump pressure and main machine belt speed during tunnel excavation. In order to better capture the local patterns and periodic features in these time series data, we split the input data (i.e., valid data) into multiple feature blocks and extract frequency domain features. This method helps the model capture the changing patterns of data at different time scales and enhances the understanding and analysis of TBM construction time series data. The selected input data is the first 15 seconds of continuous data in the ascending segment and the first 10 seconds of continuous data in the stable segment, a total of 25 seconds of data.

[0044] First, the input valid data X has dimensions (B, T, D), where B represents the batch size / batch, T represents the sequence length, and D represents the input feature dimension. In order to process these data, the input valid data is divided into multiple feature blocks based on the pre-set feature block size, that is, the following formula: ; Wherein, P is the number of feature blocks corresponding to valid data, T is the sequence length corresponding to valid data, and L is the preset feature block size; Then, the segmented feature blocks are reshaped accordingly through the view function, namely: X time =X[:,:P×L,:].view(B,P,L,D); Among them, X time is the time domain signal corresponding to the feature block, B is the batch size corresponding to the valid data, D is the feature dimension corresponding to the feature block, view is the reshaping function, and X[:,:P×L,:] is the slicing operation on the valid data.

[0045] In order to more accurately capture the periodic characteristics in the TBM construction time series data, Fourier transform is performed on each feature block to extract frequency domain features. Fourier transform converts the time domain signal into a frequency domain signal, so that the periodicity and trend information in the data can be captured. The formula is as follows: ; Among them, x(n) represents the time domain signal, X(f) represents the complex representation of the frequency domain signal, N is the sequence length of the feature block, f is the frequency, and j is the imaginary unit of the complex imaginary part.

[0046] Here, the complex part of the Fourier transform result only retains the real part, that is: X freq =Re(FFT(X time ,dim=2)); Among them, Xfreq is the frequency domain signal corresponding to the feature block, Re means taking the real part of the complex number, FFT is the fast Fourier transform, dim=2 means operating in the second dimension, X time is the time domain signal corresponding to the feature block.

[0047] Through the above data preprocessing method, the local characteristics and periodic characteristics of the TBM construction process can be better captured, and the model's ability to recognize and understand these characteristics can be enhanced.

[0048] Furthermore, the frequency domain features and time domain features of these feature blocks are processed separately through the LSTM network. Among them, when processing the TBM construction time series data, the LSTM network captures long-term and short-term dependencies through the gating mechanism. In order to process these time series data, the LSTM network is used to process the time domain and frequency domain signals to capture information from different domains.

[0049] The specific process of LSTM is as follows. The input gate controls the number of inputs to the cell state: i t =σ(W i x t +U i h t-1 +b i ); where t usually represents the current time step in the sequence, and i t is the output of the input gate, b i is the bias vector of the input gate, h t-1 is the hidden state at the previous moment, W i is the weight matrix of the input gate, U i is the weight matrix in the input gate related to the hidden state at the previous moment, and σ is the Sigmoid activation function; The forget gate controls how much of the cell state from the previous time step is retained: f t =σ(W f x t +U f h t-1 +b f ); Among them, f t is the output of the forget gate, W f is the weight matrix of the forget gate, U f is the weight matrix related to the hidden state at the previous moment in the forget gate, b f is the bias vector of the forget gate.

[0050] o t =σ(W o x t +U o ht-1 +b o ); Among them, t is the output of the output gate, W o is the weight matrix of the output gate, U o is the weight matrix in the output gate related to the hidden state at the previous moment, b o is the bias vector of the output gate.

[0051] Calculation of new candidate cell states: ; in, is the memory state at the current moment, W c is the weight matrix of the state candidate, U c is the weight matrix related to the hidden state at the previous moment in the state candidate, b c is the bias vector of the state candidate.

[0052] Cell status update: ; c t is the unit status at the current moment.

[0053] Hide status updates: h t =o t ⊙tanh(c t ); Among them, ⊙ is element-wise multiplication.

[0054] Specifically, Time domain LSTM network processing: L time ,(h time ,c time )=LSTM time (X time ); Among them, L time is the time domain feature corresponding to the feature block, LSTM time is the time domain LSTM network, X time is the time domain signal corresponding to the feature block, h time is the short-term memory parameter in the time domain LSTM network, c time is the long-term memory parameter in the time domain LSTM network.

[0055] Frequency domain LSTM network processing: L freq ,(h freq ,c freq )=LSTM freq (X freq ); Among them, L freq is the frequency domain feature corresponding to the feature block, LSTM freq is the frequency domain LSTM network, X freq is the frequency domain signal corresponding to the feature block, h freq is the short-term memory parameter in the frequency domain LSTM network, c freq is the long-term memory parameter in the frequency domain LSTM network; The time domain signal corresponding to the feature block is input into the preset time domain LSTM network to capture the time domain features corresponding to the feature block; wherein the processing process of the time domain LSTM network is as follows: When processing TBM construction time series data, the difference in data distribution between different batches will affect the training effect of the model. In order to improve the stability of training and the convergence speed of the model, we batch normalized the features within each batch to avoid excessive differences in data distribution between different batches, thereby improving the generalization ability of the model.

[0056] Batch normalization normalizes the features within each batch as shown below: ; Among them, μ batch and are the mean and variance of the batch, ε is a small constant set for numerical stability, and L x For L time or L freq .

[0057] In order to preserve the original feature information and improve training stability when processing TBM construction time series data, we add the attention outputs in the time domain and frequency domain to the features processed by LSTM to form residual connections. The residual connection effectively alleviates the gradient vanishing problem, improves training performance, and preserves the original feature information.

[0058] First, the attention outputs of time domain features and frequency domain are calculated, that is, the attention matrices of frequency domain features and time domain features are calculated through the hybrid attention mechanism.

[0059] In TBM construction time series data, the extraction of local features is crucial because these local features may contain key information about changes in construction parameters. In order to improve the local feature extraction in the multi-head attention mechanism and enable the model to better capture local patterns, we use convolution operations to generate queries (Q c ), key (K c ) and the value matrix (V c ). The convolution operation extracts local features by learning the convolution kernel weights, as shown in the following formula: Q c =Conv1D(X),Kc =Conv1D(X),V c =Conv1D(X); The key is to use local perception to extract features from adjacent time steps, Q c , K c and V c The calculation process is similar to that of Q c Take this as an example to show the calculation process: ; Among them, K is the size of the convolution kernel, W convk is the weight of the kth convolution kernel, x i+k is the i+kth time step of the input sequence.

[0060] Next, we use convolutional projection to calculate the attention score: ; T is the transpose, d k The dimension of the key.

[0061] In order to calculate the dependencies between different positions in the global scope, a multi-head attention mechanism is used to generate query, key and value matrices and calculate attention weights to capture long-range dependencies in the input sequence. The multi-head attention mechanism processes information from different subspaces in parallel through multiple attention heads, thereby improving the expressiveness of the model.

[0062] First, a linear transformation generates query, key, and value matrices as shown in the following equations: ; Among them, W Q , W K and W V are the weight parameters for the query, key, and value matrices respectively.

[0063] Next, the dot product of the query and the key is calculated and scaled by the square root of the key dimension to obtain the scaled attention matrix as follows: ; T is the transpose, d k is the dimension of the key, Q r K r T is the dot product of the query and key matrices.

[0064] To combine local and global features, we adopt a dynamic weighted fusion of the results of convolutional projection and traditional attention, which enables the model to adaptively balance the contribution of local and global information. This hybrid projection attention mechanism can simultaneously exploit detailed information within the local window and long-range dependencies in the global sequence, thereby enhancing the comprehensive understanding of the construction process.

[0065] First, we calculate the attention results of convolutional projection and traditional attention mechanism, and then get the final attention output through weighted fusion.

[0066] For the time domain, it is calculated by the following formula: O time =α 1 Attention time (Q r ,K r ,V r )+β 1 Attention time (Q c ,K c ,V c ); Among them, O time is the time domain attention matrix corresponding to the time domain features of the feature block, Attention time (Q r ,K r ,V r ) is the output matrix of the multi-head convolution projection attention mechanism corresponding to the time domain features of the feature block, Attention time (Q c ,K c ,V c ) is the output matrix of the multi-head attention mechanism corresponding to the temporal features of the feature block, α 1 is the pre-set multi-head convolution projection attention weight of the time domain feature, β 1 The multi-head attention weights for the pre-set temporal features; For the frequency domain: O freq =α 2 Attention freq (Q r ,K r ,V r )+β 2 Attention freq (Q c ,K c, V c ); Among them, O freq is the frequency domain attention matrix corresponding to the frequency domain features of the feature block, Attention freq (Q r ,K r ,V r ) is the output matrix of the multi-head convolution projection attention mechanism corresponding to the frequency domain features of the feature block, Attention freq (Qc ,K c, V c ) is the output matrix of the multi-head attention mechanism corresponding to the frequency domain features of the feature block, α 2 is the pre-set multi-head convolution projection attention weight of the frequency domain feature, β 2 is the multi-head attention weights of the pre-set frequency domain features.

[0067] Specifically, the following examples are given: First, the convolutional projection query, key, and value matrices are calculated. Then, the query, key, and value matrices are split into multiple heads: ; Where Q c , K c and V c It is split into h-head; Next, we calculate the scaled dot product attention for each head: ; The scaled dot product attention of each head is represented by Q ch , K ch 、V ch Represented by h, where h represents the number of attention heads. Then, the outputs of all heads are concatenated: ; Next, a similar treatment is performed on the traditional multi-head attention mechanism; ; Here, Q r , K r , V r Decomposed into h heads.

[0068] Next, we calculate the scaled dot product attention for each head: ; Results by Q rh , K rh , V rh Represented by h, where h represents the number of attention heads. Then, the outputs of all heads are concatenated: O r =concat(Attention(Q r1 ,K r1 ,V r1 ),...,Attention(Q rh ,K rh ,V rh )); Finally, the convolutional projection and traditional attention results are weighted fused: Of =αO r +βO c .

[0069] Through the first fully connected layer, the time domain attention matrix and frequency domain attention matrix corresponding to the obtained feature block are transformed, that is: ; Among them, A time is the transformation result of the time domain attention matrix corresponding to the feature block, W time is the preset time domain weight matrix, O time is the temporal attention matrix corresponding to the temporal features of the feature block; A freq is the transformation result of the frequency domain attention matrix corresponding to the feature block, W freq is the preset time domain weight matrix, O freq is the frequency domain attention matrix corresponding to the frequency domain features of the feature block; Next, a residual connection is performed by adding the transformed attention output to the features processed by the LSTM; that is: ; Among them, R time is the residual connection result of the feature block in the time domain, L time is the time domain feature corresponding to the feature block, R freq is the residual connection result of the feature block in the frequency domain, L freq is the time domain feature corresponding to the feature block; Finally, when generating the context vector, the residual connection results in the time domain and frequency domain are combined to form a comprehensive feature, which improves the representation ability of the model and can better capture the diversity of data, that is: C = concat (R time ,R freq ,dim=2); Where C is the fusion feature corresponding to the feature block, concat is the concatenation operation, and dim=2 is the operation on the second feature dimension.

[0070] Then, calculate the attention score: S=CW score ; Among them, S is the attention score matrix corresponding to the feature block, C is the fusion feature corresponding to the feature block, and W score is the pre-set attention score weight matrix; Generate context vector: ; Among them, A is the attention weight corresponding to the feature block, c is the global context vector corresponding to the valid data, C is the fusion feature corresponding to the feature block, i is the index of the feature block, and P is the number of feature blocks corresponding to the valid data.

[0071] Finally, to generate the final output, the context vector is concatenated with the hidden state from the previous time step and passed to a fully connected layer and a dropout layer: The fusion feature corresponding to the last feature block in all feature blocks of valid data is concatenated with the global context vector corresponding to the valid data to obtain the connection vector corresponding to the valid data; the specific formula is as follows: p=concat(c,H t ); Among them, p is the connection vector corresponding to the valid data, concat is the concatenation operation, c is the global context vector corresponding to the valid data, and H t is the fusion feature corresponding to the last feature block; Subsequently, the fusion features corresponding to the valid data are input into the preset second fully connected layer, and the lithology prediction results corresponding to the valid data are obtained based on the preset formula 11; the formula 11 is: ; Among them, W f and W p are all pre-set weights of the second fully connected layer, v is the intermediate feature generated by nonlinear transformation, p is the connection vector corresponding to the valid data, O is the lithology prediction result corresponding to the valid data, tanh is the hyperbolic tangent function, and dropout is regularization.

[0072] This embodiment proposes a real-time prediction method for rock properties during tunneling by a tunnel boring machine. The method predicts the current rock properties in real time based on the state parameters during the TBM construction process, so that the parameters and construction plans during the tunneling process can be dynamically adjusted according to the prediction results, thereby effectively reducing the risk of equipment damage and sudden accidents.

[0073] In the description of the present invention, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the features. In the description of the present invention, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.

[0074] In the present invention, unless otherwise clearly specified and limited, the terms "installed", "connected", "connected", "fixed" and the like should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium; it can be the internal connection of two elements or the interaction relationship between two elements. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0075] In the present invention, unless otherwise clearly specified and limited, when a first feature is “on” or “below” a second feature, it may be that the first and second features are in direct contact, or that the first and second features are in indirect contact through an intermediate medium. Moreover, when a first feature is “above”, “above” or “above” a second feature, it may be that the first feature is directly above or obliquely above the second feature, or it may simply mean that the first feature is higher in level than the second feature. When a first feature is “below”, “below” or “below” a second feature, it may be that the first feature is directly below or obliquely below the second feature, or it may simply mean that the first feature is lower in level than the second feature.

[0076] In the description of this specification, the description of the terms "one embodiment", "some embodiments", "embodiment", "example", "specific example" or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, unless they are contradictory.

[0077] Although the embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and are not to be construed as limitations of the present invention. A person skilled in the art may alter, modify, replace and modify the above embodiments within the scope of the present invention.

Claims

1. A method for real-time prediction of rock properties during tunnel boring machine excavation, characterized in that: include: S11, real-time detection of valid data corresponding to each excavation cycle during the excavation process of the tunnel boring machine; S12, inputting the obtained valid data into a pre-trained lithology prediction model to obtain a lithology prediction result output by the lithology prediction model; During this period, characteristic parameters corresponding to the effective data are obtained according to the effective data corresponding to each excavation cycle; According to the obtained characteristic parameters corresponding to the valid data and the preset characteristic block size, the characteristic parameters corresponding to the valid data are divided into at least one characteristic block, and the time domain signal and the frequency domain signal corresponding to each characteristic block are obtained; The frequency domain signal and time domain signal corresponding to each feature block are respectively input into the pre-set LSTM network to obtain the frequency domain features and time domain features corresponding to the feature block; According to the frequency domain features and time domain features corresponding to each feature block and the pre-set hybrid attention mechanism, the time domain attention matrix and frequency domain attention matrix corresponding to each feature block are obtained; Based on the time domain attention matrix, frequency domain attention matrix, frequency domain features and time domain features corresponding to each feature block, the lithology prediction results corresponding to the effective data are obtained; S13, judging the working state of the tunnel boring machine in the next excavation cycle according to the lithology prediction result corresponding to the valid data.

2. The method for real-time prediction of rock properties during tunnel boring by a tunnel boring machine according to claim 1, characterized in that: Each excavation cycle includes: a rising section and a stabilization section; The valid data corresponding to each excavation cycle includes: the state parameters of each unit time in the first preset time period in the ascending section corresponding to the excavation cycle and the state parameters of each unit time in the second preset time period in the stable section; Then the S11 includes: Real-time detection of the state parameters corresponding to each unit time in each excavation cycle during the tunnel boring machine excavation process, and screening of all state parameters in each excavation cycle to obtain valid data corresponding to each excavation cycle; Each status parameter includes: cutter head speed, excavation speed, total thrust, cutter head torque, shield pressure, gripper pressure, gripper pump pressure, main engine belt conveyor pump pressure and main engine belt conveyor speed.

3. The method for real-time prediction of rock properties during tunnel boring by a tunnel boring machine according to claim 1, characterized in that: The feature parameters include batch size, sequence length and feature dimension; Then the S12, according to the characteristic parameters corresponding to the obtained valid data and the preset characteristic block size, divides the valid data into at least one characteristic block, and obtains the time domain signal and the frequency domain signal corresponding to each characteristic block, including: According to the sequence length and feature dimension corresponding to the obtained valid data, the preset feature block size and the preset formula 1, the valid data is divided into at least one feature block, each feature block has its corresponding sequence length and feature dimension; the formula 1 is: ； Wherein, P is the number of feature blocks corresponding to valid data, T is the sequence length corresponding to valid data, and L is the preset feature block size; According to the sequence length and feature dimension corresponding to each feature block and the preset formula 2, each feature block is reshaped to obtain the time domain signal corresponding to each feature block; the formula 2 is: X time =X[:,:P×L,:].view(B,P,L,D); Among them, X time is the time domain signal corresponding to the feature block, B is the batch size corresponding to the valid data, D is the feature dimension corresponding to the feature block, view is the reshaping function, and X[:,:P×L,:] is the slicing operation on the valid data; According to the time domain signal corresponding to each feature block and the preset formula 3, the frequency domain signal corresponding to each feature block is obtained; the formula 3 is: X freq =Re(FFT(X time ,dim=2)); Among them, X freq is the frequency domain signal corresponding to the feature block, Re means taking the real part of the complex number, FFT is the fast Fourier transform, dim=2 means operating in the second dimension, X time is the time domain signal corresponding to the feature block.

4. The method for real-time prediction of rock properties during tunnel boring by a tunnel boring machine according to claim 1, characterized in that: The LSTM network includes a time domain LSTM network and a frequency domain LSTM network; Then, the S12, respectively inputting the frequency domain signal and the time domain signal corresponding to each feature block into a preset LSTM network to obtain the frequency domain features and the time domain features corresponding to the feature block, includes: The frequency domain signal corresponding to the feature block is input into the preset frequency domain LSTM network to capture the frequency domain features corresponding to the feature block; wherein the processing process of the frequency domain LSTM network is as follows: L freq ,(h freq ,c freq )=LSTM freq (X freq ); Among them, L freq is the frequency domain feature corresponding to the feature block, LSTM freq is the frequency domain LSTM network, X freq is the frequency domain signal corresponding to the feature block, h freq is the short-term memory parameter in the frequency domain LSTM network, c freq is the long-term memory parameter in the frequency domain LSTM network; The time domain signal corresponding to the feature block is input into a preset time domain LSTM network to capture the time domain features corresponding to the feature block; wherein the processing process of the time domain LSTM network is as follows: L time ,(h time ,c time )=LSTM time (X time ); Among them, L time is the time domain feature corresponding to the feature block, LSTM time is the time domain LSTM network, X time is the time domain signal corresponding to the feature block, h time is the short-term memory parameter in the time domain LSTM network, c time is the long-term memory parameter in the time domain LSTM network.

5. The method for real-time prediction of rock properties during tunnel boring by a tunnel boring machine according to claim 1, characterized in that: The hybrid attention mechanism includes a multi-head attention mechanism and a multi-head convolutional projection attention mechanism; Then, the S12, according to the frequency domain features and time domain features corresponding to each feature block, and the pre-set hybrid attention mechanism, obtains the time domain attention matrix and the frequency domain attention matrix corresponding to each feature block, including: Based on the time domain features and frequency domain features corresponding to all feature blocks, the time domain features and frequency domain features corresponding to each feature block are normalized respectively; According to the time domain features corresponding to each feature block, the preset hybrid attention mechanism and the preset formula 4, the time domain attention matrix corresponding to the feature block is obtained; The formula 4 is: O time =α1·Attention time (Q r ,K r ,V r )+β1·Attention time (Q c ,K c ,V c ); Among them, O time is the time domain attention matrix corresponding to the time domain features of the feature block, Attention time (Q r ,K r ,V r ) is the output matrix of the multi-head convolution projection attention mechanism corresponding to the time domain features of the feature block, Attention time (Q c ,K c ,V c ) is the multi-head attention mechanism output matrix corresponding to the time domain features of the feature block, α1 is the pre-set multi-head convolution projection attention weight of the time domain features, and β1 is the pre-set multi-head attention weight of the time domain features; At the same time, according to the frequency domain features corresponding to each feature block, the preset hybrid attention mechanism and the preset formula 5, the frequency domain attention matrix corresponding to the feature block is obtained; The formula five is: O freq =α2·Attention freq (Q r ,K r ,V r )+β2·Attention freq (Q c ,K c, V c ); Among them, O freq is the frequency domain attention matrix corresponding to the frequency domain features of the feature block, Attention freq (Q r ,K r ,V r ) is the output matrix of the multi-head convolution projection attention mechanism corresponding to the frequency domain features of the feature block; Attention freq (Q c ,K c, V c ) is the multi-head attention mechanism output matrix corresponding to the frequency domain features of the feature block, α2 is the pre-set multi-head convolution projection attention weight of the frequency domain features, and β2 is the pre-set multi-head attention weight of the frequency domain features.

6. The method for real-time prediction of rock properties during tunnel boring by a tunnel boring machine according to claim 1, characterized in that: The S12, based on the time domain attention matrix, frequency domain attention matrix, frequency domain features and time domain features corresponding to each feature block, obtains the lithology prediction result corresponding to the valid data, including: According to the time domain attention matrix, frequency domain attention matrix, time domain features and frequency domain features corresponding to each feature block, the fusion features corresponding to each feature block are obtained; According to the fusion features corresponding to each feature block and the pre-set weighted attention mechanism, the global information of all feature blocks is obtained, that is, the global context vector corresponding to the valid data; According to the global context vector corresponding to the valid data, the lithology prediction result corresponding to the valid data is obtained.

7. The method for real-time prediction of rock properties during tunneling by a tunnel boring machine according to claim 6, characterized in that: The S12, according to the time domain attention matrix, frequency domain attention matrix, time domain features and frequency domain features corresponding to each feature block, obtains the fusion features corresponding to each feature block, including: The time domain attention matrix and the frequency domain attention matrix corresponding to each feature block are respectively input into the preset first fully connected layer, and feature transformation is performed based on the preset formula 6; ； Among them, A time is the transformation result of the time domain attention matrix corresponding to the feature block, W time is the preset time domain weight matrix, O time is the temporal attention matrix corresponding to the temporal features of the feature block; A freq is the transformation result of the frequency domain attention matrix corresponding to the feature block, W freq is the preset time domain weight matrix, O freq is the frequency domain attention matrix corresponding to the frequency domain features of the feature block; Then, based on the preset formula 7, the transformation result of the time domain attention matrix corresponding to each feature block is residually connected with the time domain feature corresponding to the feature block, and the transformation result of the frequency domain attention matrix corresponding to each feature block is residually connected with the frequency domain feature corresponding to the feature block; the formula 7 is: ； Among them, R time is the residual connection result of the feature block in the time domain, L time is the time domain feature corresponding to the feature block, R freq is the residual connection result of the feature block in the frequency domain, L freq is the time domain feature corresponding to the feature block; Finally, based on the preset formula eight, the residual connection results of each feature block in the frequency domain and the residual connection results in the time domain are feature spliced ​​to obtain the fusion features corresponding to each feature block; the formula eight is: C=concat(R time ,R freq ,dim=2); Where C is the fusion feature corresponding to the feature block, concat is the concatenation operation, and dim=2 is the operation on the second feature dimension.

8. The method for real-time prediction of rock properties during tunneling by a tunnel boring machine according to claim 6, characterized in that: The S12, according to the fusion features corresponding to each feature block and the preset weighted attention mechanism, obtains the global information of all feature blocks, including: According to the fusion features corresponding to each feature block and the preset attention score weight matrix, and the preset formula nine, the attention score matrix corresponding to each feature block is obtained; the formula nine is: S=CW score ; Among them, S is the attention score matrix corresponding to the feature block, C is the fusion feature corresponding to the feature block, and W score is the pre-set attention score weight matrix; Then, the attention score matrix corresponding to each feature block is normalized by the preset softmax function to generate the attention weight corresponding to each feature block; Finally, according to the attention weight and fusion features corresponding to each feature block and the preset formula 10, the global context vector corresponding to the valid data is obtained; the formula 10 is: ； Among them, A is the attention weight corresponding to the feature block, c is the global context vector corresponding to the valid data, C is the fusion feature corresponding to the feature block, i is the index of the feature block, and P is the number of feature blocks corresponding to the valid data.

9. The method for real-time prediction of rock properties during tunnel boring by a tunnel boring machine according to claim 6, characterized in that: The step S12, obtaining the lithology prediction result corresponding to the valid data according to the global context vector corresponding to the valid data, includes: The fusion feature corresponding to the last feature block in all feature blocks of valid data is concatenated with the global context vector corresponding to the valid data to obtain the connection vector corresponding to the valid data; the specific formula is as follows: p=concat(c,H t ); Among them, p is the connection vector corresponding to the valid data, concat is the concatenation operation, c is the global context vector corresponding to the valid data, and H t is the fusion feature corresponding to the last feature block; Subsequently, the fusion features corresponding to the valid data are input into the preset second fully connected layer, and the lithology prediction results corresponding to the valid data are obtained based on the preset formula 11; the formula 11 is: ； Among them, W f and W p are all pre-set weights of the second fully connected layer, v is the intermediate feature generated by nonlinear transformation, p is the connection vector corresponding to the valid data, O is the lithology prediction result corresponding to the valid data, tanh is the hyperbolic tangent function, and dropout is regularization.

10. The method for real-time prediction of rock properties during tunneling by a tunnel boring machine according to claim 1, characterized in that: The S11 and before also include: S10, training the lithology prediction model based on a preset model training set to obtain a trained lithology prediction model; During this period, the hyperparameters of the lithology prediction model were optimized in real time based on the improved EWOA-GWO-MRFO and the pre-set model training strategy; The model training strategy includes: When in the exploration stage, the optimal hyperparameter is found based on the preset formula 12; the formula 12 is: ； Among them, X t+1 and X t are the positions of the gray wolf individuals at the t+1th iteration and the tth iteration, respectively, α is the individual with the best fitness, A is the step size dynamic adjustment factor, a is the dynamic control factor, r1 is a random number in the range of [0,1], sin(2πl) is the nonlinear perturbation term, X ☆ is the global optimal solution of the current population; When in the predation phase, the optimal hyperparameters are found based on the pre-set formula 13; the formula 13 is: X t+1 =X β -A·|2·r2·X β -X t |; Among them, X β is the individual with the second best fitness, r2 is a random number in the range of [0,1]; When in the spiral predation stage, the optimal hyperparameter is found based on the pre-set formula 14; the formula 14 is: X t+1= |X ☆ -X t |·e b·l cos(2πl)+η·(X rand -X t )+γ·(X α -X t ); Among them, l is a random factor with a value range of [0,1], η and γ are dynamic control factors of jump and collaboration respectively, b is a constant, and X rand Any individual in a population.

Images