A TBM excavation speed prediction method and system

Abstract

The present invention relates to the technical field of TBM, and in particular to a method and system for predicting a TBM excavation speed. The method comprises: S1, obtaining TBM parameter data of a first time period before a current time point; S2, segmenting the TBM parameter data of the first time period before the current time point to obtain N parameter data sequence sets; S3, inputting the N parameter data sequence sets into a trained iWOA-TSM model in sequence to obtain a prediction result; the prediction result comprises: a predicted value of the TBM excavation speed per second in a second time period after the current time point; pre-training a preset iWOA-TSM model with a training data set to obtain the trained iWOA-TSM model; the training data set comprises: a plurality of parameter data sequence sets segmented from the TBM parameter data of a historical time period and the true values ​​of the excavation speeds respectively corresponding to the parameter data sequence sets segmented from the TBM parameter data of the historical time period.

Description

Technical Field

[0001] The present invention relates to the technical field of TBM, and in particular to a TBM excavation speed prediction method and system. Background Art

[0002] As an efficient tunnel construction equipment, Tunnel Boring Machine (TBM) is widely used in tunnel construction projects such as subways, railways, highways, and water conservancy projects. TBM can efficiently and safely carry out tunneling construction under different geological conditions, and its tunneling speed is one of the key indicators to measure construction efficiency and progress. However, due to the influence of many complex factors during the TBM tunneling process, such as changes in geological conditions, fluctuations in equipment status, and operator control strategies, the tunneling speed often shows nonlinearity, time-varying, and uncertainty. Therefore, accurately predicting the TBM tunneling speed is of great significance for construction progress control and construction safety assurance.

[0003] However, the existing TBM excavation speed prediction methods have the following major problems: Most of the existing prediction models only focus on the modeling of a single feature dimension, ignoring the integration of the time dimension and the feature dimension, resulting in the model's insufficient use of multi-dimensional feature information. During the TBM excavation process, the speed change is not only related to the parameters at the current time point, but also affected by the parameter changes in the historical time period. It is difficult for the existing model to effectively capture this long-term dependency, which affects the accuracy of the prediction. In practical applications, the existing prediction models are often difficult to adapt to the complex and changeable geological environment and construction conditions, and the prediction effect is unstable and cannot meet the actual needs of the project. Summary of the invention

[0004] In view of the above-mentioned shortcomings and deficiencies of the prior art, the present invention provides a TBM excavation speed prediction method and system.

[0005] In order to achieve the above object, the main technical solutions adopted by the present invention include:

[0006] The embodiment of the present invention provides a TBM excavation speed prediction method, comprising:

[0007] S1, obtaining TBM parameter data every second in the first time period before the current time point;

[0008] S2, segmenting the TBM parameter data in the first time period before the current time point to obtain N parameter data sequence sets;

[0009] S3, inputting N parameter data sequence sets into the trained iWOA-TSM model in sequence to obtain prediction results; the prediction results include: the predicted value of the TBM excavation speed per second in the second time period after the current time point;

[0010] Wherein, a preset iWOA-TSM model is trained in advance using a training data set to obtain a trained iWOA-TSM model;

[0011] The training data set includes: a plurality of parameter data sequence sets obtained by segmenting the TBM parameter data of the historical time period and the real values ​​of the tunneling speeds corresponding to the parameter data sequence sets obtained by segmenting the TBM parameter data of the historical time period;

[0012] The iWOA-TSM model includes a hybrid cross-dimensional module and a time projection module in sequence;

[0013] The hybrid cross-dimensional module is used to process the parameter data sequence set input to the iWOA-TSM model to obtain fused feature data; the fused feature data is obtained by fusing the time dimension features and feature dimension features of the parameter data sequence set input to the iWOA-TSM model;

[0014] The time projection module is used to perform prediction based on the fused feature data to obtain a prediction result.

[0015] Preferably,

[0016] The TBM parameter data in the first time period before the current time point are segmented according to data, and the nth parameter data sequence set among the N parameter data sequence sets is X n , the n+ith parameter data sequence set is X n+i ;

[0017] Among them, X n ={x （a） 、x （a+1） , ..., x （a+L－1）};

[0018] Among them, x （a） represents the TBM parameter data at the ath second in the first time period;

[0019] x （a+L－1） , represents the TBM parameter data at the (a+L-1)th second in the first time period;

[0020] L is the time dimension of the parameter data series set;

[0021] X n+i ={x （a+i） 、x （a+i+1） , ..., x （a+i+L－1）}.

[0022] Preferably,

[0023] The first time period is the time period from the current time point to the first time point;

[0024] The first time point is 30-128 seconds before the current time point;

[0025] Each TBM parameter data includes: D parameter characteristics;

[0026] The second time period is a time period from the current time point to the second time point;

[0027] The second time point is between 128-500 seconds after the current time point.

[0028] Preferably,

[0029] The hybrid cross-dimensional module includes:

[0030] The first batch normalization layer is used to perform batch normalization processing on the parameter data sequence set in the input mixed cross-dimensional module to obtain the parameter data sequence set after batch normalization processing;

[0031] A first dimension transformation unit, used to perform a first dimension transformation on the parameter data sequence set after batch normalization processing to obtain a parameter data sequence set after the first dimension transformation;

[0032] A time matrix unit, used to calculate the parameter data sequence set after the first dimension transformation in the time dimension using a feedforward neural network to obtain a time matrix calculation result;

[0033] A second dimension transformation unit, used for performing a second dimension transformation on the time matrix calculation result to obtain data after the second dimension transformation;

[0034] A second batch normalization layer, used to perform batch normalization processing on the first data to obtain batch-normalized first data;

[0035] The first data is the residual between the parameter data sequence set input into the hybrid cross-dimensional module and the data after the second dimension transformation;

[0036] A feature dimension calculation unit, used to calculate the first data after batch normalization on the feature dimension by using a feedforward neural network to obtain a feature dimension calculation result;

[0037] A third dimension transformation unit, used for performing a third dimension transformation on the second data to obtain data after the third dimension transformation;

[0038] The second data is the residual between the first data and the characteristic dimension calculation result;

[0039] A time and feature dimension matrix calculation unit, used to calculate the data transformed by the third dimension in the time dimension and the channel dimension using a feedforward neural network to obtain a comprehensive calculation result;

[0040] The FEA unit is used to process the third data to obtain a processing result, and obtain output data of the hybrid cross-dimensional module according to the processing result and the third data;

[0041] The third data is the residual between the comprehensive calculation result and the data after the second dimension transformation;

[0042] Among them, the output data of the mixed cross-dimensional module is the residual between the processing result and the third data, and the output data of the mixed cross-dimensional module is used as the fused feature data.

[0043] Preferably,

[0044] The dimension of the parameter data sequence set after batch normalization is L×D;

[0045] D is the characteristic dimension of each TBM parameter data in the parameter data series set;

[0046] The dimension of the parameter data sequence set after the first dimension transformation is D×L;

[0047] The dimension of the data after the second dimension transformation is L×D;

[0048] The data after the third dimension transformation is represented by a L×D vector.

[0049] Preferably,

[0050] The FEA unit processes the third data to obtain a processing result, which specifically includes:

[0051] The FEA unit performs dimension-up mapping on the third data in the feature dimension to obtain first-dimensional data E; the feature dimension of the first-dimensional data E is S, where S≥5D;

[0052] Normalizing the first dimensional data E in the time dimension using a softmax function to obtain normalized first dimensional data E;

[0053] Perform a dimensionality reduction operation on the normalized first dimension data E along the feature dimension to restore the feature dimension that is the same as the third data, and obtain output data of the time dimension;

[0054] The FEA unit transposes the third data in the time dimension and performs dimensionality-up mapping in the feature dimension to obtain second-dimensional data B; the feature dimension of the second-dimensional data B is S;

[0055] The second dimension data B is normalized using the softmax function on the feature dimension to obtain the normalized second dimension data B;

[0056] Performing a dimensionality reduction operation on the normalized second dimensional data B along the feature dimension to restore the feature dimension that is the same as the third data, and converting it back to the order of the time dimension of the third data to obtain output data of the feature dimension;

[0057] The output data of the time dimension and the output data of the feature dimension are concatenated in the last dimension to obtain fused data;

[0058] A dimension reduction operation is performed on the fused data in terms of feature dimension to make it consistent with the feature dimension of the third data, thereby obtaining a processing result.

[0059] Preferably,

[0060] The trained iWOA-TSM model refers to an optimized model obtained by tuning the improved whale optimization algorithm in the process of training the preset iWOA-TSM model using the training data.

[0061] Preferably,

[0062] Wherein, in the improved whale optimization algorithm, the whale individuals are represented as hyperparameters in the iWOA-TSM model, and the positions of the whale individuals are updated to determine the optimal positions of the whale individuals;

[0063] The control factor A in the improved whale optimization algorithm fluctuates randomly in the interval [-2, 2], r 1 is a random number uniformly distributed in the interval [0, 1];

[0064] Among them, in the improved whale optimization algorithm, when r 1 When <0.5 and |A|≤1, the individual whale position is updated using formula (1);

[0065] Formula (1) is:

[0066] ;

[0067] ;

[0068] Indicates the next step position of the individual whale from its current position;

[0069] Indicates the current position of the individual whale;

[0070] t represents the current iteration number;

[0071] T is the maximum number of iterations;

[0072] r is a random number in the range of [0, 1];

[0073] e is the base of natural logarithms;

[0074] in, is the current best individual whale position;

[0075] In the improved whale optimization algorithm, when r 1 When <0.5 and |A|>1, the individual position of the whale is updated using formula (2);

[0076] Formula (2) is:

[0077] ;

[0078] is the position of a randomly selected individual whale, indicating a global search;

[0079] In the improved whale optimization algorithm, when r 1 When ≥0.5, the individual whale position is updated using formula (3);

[0080] Formula (3) is:

[0081] ;

[0082] ;

[0083] ;

[0084] l is a random number that changes randomly in the range [0, 1];

[0085] and are the initial maximum and minimum random ranges respectively.

[0086] Preferably,

[0087] The characteristics of each TBM parameter data in the parameter data sequence set include: TBM excavation speed, cutter head power, cutter head torque, thrust force, gripper shoe pressure, gripper shoe pump pressure, and main machine belt pump pressure.

[0088] This embodiment also provides a TBM excavation speed prediction system, including:

[0089] at least one processor; and,

[0090] At least one memory is communicatively connected to the processor, wherein the memory stores program instructions executable by the processor, and the processor calls the program instructions to execute the TBM excavation speed prediction method as described above.

[0091] The beneficial effects of the present invention are as follows: A TBM excavation speed prediction method of the present invention predicts the TBM excavation speed through the iWOA-TSM model. iWOA is an improved whale optimization algorithm that can automatically optimize parameters during model training, avoiding errors and instability caused by manual parameter adjustment in traditional methods. A hybrid cross-dimensional module is used to perform deep feature extraction on the input data, effectively capturing the changing trends of the time dimension and feature dimension, and reducing the deviation and error of the model. The time projection module models the fused feature data, making full use of the temporal dependency information in the time series data, and improving the accuracy of the prediction results.

[0092] In addition, the fused feature data in the hybrid cross-dimensional module combines the time dimension features and feature dimension features of the input data, and is fused through the residual structure, which can simultaneously extract and utilize feature information of different dimensions. This design of fused feature data enables the model to more comprehensively capture the key factors affecting the tunneling speed when dealing with complex and nonlinear TBM parameters, thereby enhancing the robustness and stability of the model.

[0093] The time projection module uses the long-term dependency modeling capability to conduct in-depth analysis of data in historical time periods based on the fusion of feature data, capturing long-term trends and cyclical changes in time series. This modeling capability is particularly suitable for long-term span data during TBM excavation, and can maintain high prediction accuracy under different geological conditions and construction environments.

[0094] The iWOA optimization algorithm introduces a dynamic adjustment strategy, which enables the model to conduct extensive global searches in the initial stage to avoid falling into local optimality; in the later stage, it pays more attention to local searches, which improves the accuracy of parameter optimization. Through iWOA parameter optimization, the model can reach convergence in fewer iterations, reducing training time and computing resource consumption, and improving the performance stability of the model on different data sets. BRIEF DESCRIPTION OF THE DRAWINGS

[0095] Figure 1 A flow chart of a TBM excavation speed prediction method of the present invention;

[0096] Figure 2 A schematic diagram of a prediction result obtained by inputting a parameter data sequence set into a trained iWOA-TSM model in an embodiment of the present invention;

[0097] Figure 3 Schematic diagram of the hybrid cross-dimensional module processing process in the iWOA-TSM model in an embodiment of the present invention;

[0098] Figure 4It is a schematic diagram of a process in which the FEA unit processes the third data to obtain a processing result in an embodiment of the present invention.

[0099] Description of Reference Numerals

[0100] 1: The first batch of normalization layers;

[0101] 2: first dimension transformation unit;

[0102] 3: Time matrix unit;

[0103] 4: second dimension transformation unit;

[0104] 5: The second batch of normalization layers;

[0105] 6: Feature dimension calculation unit;

[0106] 7: Time and feature dimension matrix calculation unit;

[0107] 8: FEA unit;

[0108] 9: third data;

[0109] E: normalized first dimension data;

[0110] B: normalized second dimension data;

[0111] 10: Fusion data;

[0112] 11: Processing results. DETAILED DESCRIPTION

[0113] 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.

[0114] 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.

[0115] See also Figure 1 This embodiment provides a TBM excavation speed prediction method, including:

[0116] S1, obtaining TBM parameter data every second in the first time period before the current time point;

[0117] S2, segmenting the TBM parameter data in the first time period before the current time point to obtain N parameter data sequence sets;

[0118] In this embodiment, the TBM parameter data in the first time period before the current time point is segmented according to data, and the nth parameter data sequence set among the N parameter data sequence sets is X n , the n+ith parameter data sequence set is X n+i ;

[0119] Among them, X n ={x （a） 、x （a+1） , ..., x （a+L－1）};

[0120] Among them, x （a） represents the TBM parameter data at the ath second in the first time period;

[0121] x （a+L－1） , represents the TBM parameter data at the (a+L-1)th second in the first time period;

[0122] L is the time dimension of the parameter data series set;

[0123] X n+i ={x （a+i） 、x （a+i+1） , ..., x （a+i+L－1）}.

[0124] The first time period is the time period from the current time point to the first time point;

[0125] The first time point is 30-128 seconds before the current time point;

[0126] Each TBM parameter data includes: D parameter characteristics;

[0127] S3, inputting N parameter data sequence sets into the trained iWOA-TSM model in sequence to obtain prediction results; the prediction results include: the predicted value of the TBM excavation speed per second in the second time period after the current time point;

[0128] The second time period is a time period from the current time point to the second time point;

[0129] The second time point is between 128-500 seconds after the current time point.

[0130] Among them, see Figure 2 , pre-training a preset iWOA-TSM model using a training data set to obtain a trained iWOA-TSM model;

[0131] The training data set includes: a plurality of parameter data sequence sets obtained by segmenting the TBM parameter data of the historical time period and the real values ​​of the tunneling speeds corresponding to the parameter data sequence sets obtained by segmenting the TBM parameter data of the historical time period;

[0132] The iWOA-TSM model includes a hybrid cross-dimensional module and a time projection module in sequence;

[0133] The hybrid cross-dimensional module is used to process the parameter data sequence set input to the iWOA-TSM model to obtain fused feature data; the fused feature data is obtained by fusing the time dimension features and feature dimension features of the parameter data sequence set input to the iWOA-TSM model;

[0134] The time projection module is used to perform prediction based on the fused feature data to obtain a prediction result.

[0135] For example, suppose a TBM is excavating a tunnel and the excavation speed in the future is predicted in order to carry out effective construction planning and management.

[0136] Assuming the current time point is t, the data of the first time period from the current time point t is taken, and the length of this time period is between 30-128 seconds. Assuming the length of the first time period is 60 seconds.

[0137] The parameter data of the TBM are recorded once every second. These data include D characteristics, such as: TBM excavation speed, cutterhead power, cutterhead torque, thrust, gripper pressure, gripper pump pressure, and main engine belt pump pressure.

[0138] For example, from t-60 seconds to t, 60 seconds of data are collected, and each second contains D = 7 features. Then the 60 seconds of data are segmented, assuming that the length of each data sequence set is L = 10 seconds, and the sliding window method is used, each sliding 1 second, so that multiple data sequence sets can be obtained:

[0139] X1={x(t-60),x(t-59),…,x(t-51)};

[0140] X2={x(t-59),x(t-58),…,x(t-50)};

[0141] X3={x(t-58),x(t-57),…,x(t-49)}; ...

[0142] XN={x(t-11),x(t-10),…,x(t)}.

[0143] Each Xn (parameter data sequence set) is a 10-second data sequence containing D features, that is, an L×D matrix.

[0144] The obtained N parameter data sequence sets are input into the trained iWOA-TSM model in sequence. The hybrid cross-dimensional module in the trained iWOA-TSM model processes each parameter data sequence set, extracts the trend information of the time series and the correlation between the features, and generates fused feature data. For example, there may be a certain correlation between the cutter head speed and the propulsion speed, and the hybrid cross-dimensional module can capture this correlation and make adjustments to extract the purified trend signal.

[0145] The time projection module in the trained iWOA-TSM model will predict the TBM excavation speed in the second time period in the future based on the fused feature data and using the time series prediction method.

[0146] In the practical application of this embodiment, Figure 3 As shown, the hybrid cross-dimensional module includes:

[0147] The first batch normalization layer 1 is used to perform batch normalization processing on the parameter data sequence set in the input mixed cross-dimensional module to obtain the parameter data sequence set after batch normalization processing;

[0148] The dimension of the parameter data sequence set after batch normalization is L×D; D is the feature dimension of each TBM parameter data in the parameter data sequence set;

[0149] In this embodiment, the first batch of normalization layers performs batch normalization processing, the purpose of which is to eliminate the dimensional differences between different features and improve the convergence speed and stability of the model.

[0150] A first dimension transformation unit 2, used for performing a first dimension transformation on the parameter data sequence set after batch normalization processing to obtain a parameter data sequence set after the first dimension transformation;

[0151] The dimension of the parameter data sequence set after the first dimension transformation is D×L;

[0152] A time matrix unit 3 is used to calculate the parameter data sequence set after the first dimension transformation in the time dimension using a feedforward neural network to obtain a time matrix calculation result;

[0153] The feedforward neural network in this embodiment has a powerful nonlinear mapping capability and can effectively model high-dimensional features, thereby improving prediction accuracy.

[0154] A second dimension transformation unit 4 is used to perform a second dimension transformation on the time matrix calculation result to obtain data after the second dimension transformation; the dimension of the data after the second dimension transformation is L×D;

[0155] A second batch normalization layer 5, used for performing batch normalization processing on the first data to obtain batch-normalized first data;

[0156] The first data is the residual between the parameter data sequence set input into the hybrid cross-dimensional module and the data after the second dimension transformation;

[0157] A feature dimension calculation unit 6, configured to use a feedforward neural network to calculate the first data after batch normalization on the feature dimension to obtain a feature dimension calculation result;

[0158] The third dimension transformation unit is used to perform a third dimension transformation on the second data to obtain data after the third dimension transformation; the data after the third dimension transformation is represented by a vector of L×D.

[0159] The second data is the residual between the first data and the characteristic dimension calculation result;

[0160] A time and feature dimension matrix calculation unit 7, used to calculate the data after the third dimension transformation in the time dimension and the channel dimension using a feedforward neural network to obtain a comprehensive calculation result;

[0161] The FEA unit 8 is used to process the third data 9 to obtain a processing result, and obtain output data of the hybrid cross-dimensional module according to the processing result and the third data;

[0162] The third data 9 is the residual between the comprehensive calculation result and the data after the second dimension transformation;

[0163] Among them, the output data of the mixed cross-dimensional module is the residual between the processing result and the third data, and the output data of the mixed cross-dimensional module is used as the fused feature data.

[0164] In this embodiment, through multiple dimensional transformations and residual calculations, the hybrid cross-dimensional module can capture the time dimension characteristics of the input data, the nonlinear relationship between the feature dimension characteristics, and extract the fused comprehensive features. Residual calculation can effectively remove redundant information and retain only the effective change features in the input data, thereby improving the accuracy of feature extraction. Using a residual structure can avoid overfitting, because the hybrid cross-dimensional module compares the input data with the intermediate calculation results and only retains the effective information in the input data. This structure can better adapt to noise and outliers in the data, and improve the prediction performance of the model in different scenarios and different data distributions.

[0165] Among them, see Figure 4 In this embodiment, the FEA unit 8 processes the third data 9 to obtain a processing result, which specifically includes:

[0166] The FEA unit 8 performs dimensional mapping on the third data 9 in the feature dimension to obtain the first dimensional data E; the feature dimension of the first dimensional data E is S, where S≥5D;

[0167] The process of performing dimension-up mapping on the third data 9 by the FEA unit 8 uses the matrix M fmk , the matrix M fmk The shape of is [D, S];

[0168] Normalizing the first dimensional data E in the time dimension using a softmax function to obtain normalized first dimensional data E;

[0169] Perform a dimensionality reduction operation on the normalized first dimension data E along the feature dimension to restore the feature dimension that is the same as the third data, and obtain output data of the time dimension;

[0170] The process of FEA unit 8 reducing the dimension of the normalized first dimension data E uses the matrix M fmv , the matrix M fmv The shape of is [S, D];

[0171] The FEA unit transposes the third data in the time dimension and performs dimensionality-up mapping in the feature dimension to obtain second-dimensional data B; the feature dimension of the second-dimensional data B is S;

[0172] The FEA unit uses the matrix M to perform dimensional mapping of the third data in the time dimension. tmk , the matrix M tmk The shape of is [D, S];

[0173] The second dimension data B is normalized using the softmax function on the feature dimension to obtain the normalized second dimension data B;

[0174] Performing a dimensionality reduction operation on the normalized second dimensional data B along the feature dimension to restore the feature dimension that is the same as the third data, and converting it back to the order of the time dimension of the third data to obtain output data of the feature dimension;

[0175] The FEA unit uses the matrix M to reduce the dimension of the normalized second-dimensional data B. tmv , the matrix M tmv The shape of is [S, D];

[0176] The output data of the time dimension and the output data of the feature dimension are concatenated in the last dimension to obtain fused data 10;

[0177] A dimension reduction operation is performed on the fused data 10 in terms of feature dimension to make it consistent with the feature dimension of the third data, thereby obtaining a processing result 11.

[0178] The matrix M is used in the dimension reduction operation on the feature dimension of the fused data. fusion , the matrix M fusion The shape of is [2D, D].

[0179] For example, assume that the third data is a matrix of size [L, D].

[0180] Assume: time series length L = 100 (i.e. 100 time points), feature dimension D = 10 (each time point has 10 eigenvalues). The third data is processed by FEA unit, and the specific process is as follows:

[0181] The third data [100, 10] is mapped to the first dimension data E, whose feature dimension is S. Assuming S = 50, the size of the first dimension data E is [100, 50].

[0182] Then, the first dimension data E is normalized by softmax in the time dimension (100 time points) to obtain the normalized E. This operation is equivalent to normalizing each eigenvalue (i.e., each column). The size of the normalized data is still [100, 50].

[0183] The normalized E is subjected to dimensionality reduction, restoring the feature dimension from 50 to the original 10.

[0184] The data size after dimensionality reduction is [100, 10], that is, it is restored to the same feature dimension as the third data.

[0185] The third data [100, 10] is transposed to [10, 100] in the time dimension, and then mapped to the second dimensional data B. Assuming that the feature dimension after dimensionality upgrade is still S = 50, then: the size of the second dimensional data B is [10, 50].

[0186] The second dimension data B is normalized by softmax on the feature dimension (i.e., 50 features) to obtain the normalized B, whose size is still [10, 50].

[0187] The normalized B is reduced to 10 along the feature dimension to obtain a matrix of size [10, 10].

[0188] Revert to the original time dimension order and obtain feature dimension output data of size [100, 10].

[0189] The output data [100, 10] of the time dimension and the output data [100, 10] of the feature dimension are concatenated in the last dimension to obtain fused data of size [100, 20]. The fused data is reduced in dimension to restore it to the same feature dimension [100, 10] as the third data. Finally, the processing result of the FEA unit is obtained, that is, a matrix of size [100, 10].

[0190] In this embodiment, through dimensional mapping, the FEA unit increases the representation ability of data in the feature dimension and time dimension, so that the model can capture more features and time correlation. The dimension S after feature dimensionality upgrading is much larger than the original dimension D (S≥5D), which means that the model can extract features from higher dimensions during the dimensionality upgrading process, which helps to capture complex patterns and features. Using softmax normalization operations in the time dimension and feature dimension respectively can effectively balance the influence of different features and time points and reduce the influence of biased data. Softmax normalization makes the distribution of data more uniform, avoids the situation where some eigenvalues ​​are too large or too small, thereby improving the stability and robustness of the model. The FEA unit processes the information of the time dimension and feature dimension separately and then splices and fuses them. This strategy can retain the dynamic changes in the time series while extracting the correlation between features. This two-way processing and fusion of the time dimension and the feature dimension can capture more fine-grained information, so that the output data has stronger representation and prediction capabilities.

[0191] The time projection module is used to project and map data in the time dimension.

[0192] The trained iWOA-TSM model refers to an optimized model obtained by tuning the improved whale optimization algorithm in the process of training the preset iWOA-TSM model using the training data.

[0193] Specifically, the time series model (TSM) is used as the basic model. The model structure includes a hybrid cross-dimensional module, a time projection module, and a feature fusion module to capture long-term and short-term dependency features in the time series. Before optimization, the model's hyperparameters (such as learning rate, number of layers, number of nodes, etc.) and network weights are initially set and have not been tuned. Through iWOA's intelligent search and optimization of hyperparameters, the model's structure is more adapted to specific time series data, and the trained model can better capture dynamic changes and time series patterns in the data.

[0194] In the conventional whale optimization algorithm, there are three key processes: 1. Encircling Prey; 2. Bubble-net Attack; 3. Random search for prey.

[0195] In the process of surrounding prey, in nature, humpback whales will gradually surround their prey after discovering it. In the whale optimization algorithm, the current optimal solution position (considered to be the prey position) is used to guide other individuals to approach this position, thereby surrounding the prey.

[0196] Bubble Net Attack Process,When hunting, humpback whales form a unique spiral bubble net around their prey, a behavior known as,bubble net attack.,In conventional whale optimization algorithms, this process is simulated by two strategies:,spiral update position and surround update position.

[0197] In order to avoid falling into the local optimum, whales will randomly search for prey for a part of the time, that is, perform a global search. At this time, whales no longer follow the current optimal position, but choose a random position to search.

[0198] In this embodiment, in the improved whale optimization algorithm, the whale individuals are represented as hyperparameters in the iWOA-TSM model, and the positions of the whale individuals are updated to determine the optimal positions of the whale individuals;

[0199] The control factor A in the improved whale optimization algorithm fluctuates randomly in the interval [-2, 2], r 1 is a random number uniformly distributed in the interval [0, 1]. In this embodiment, the control factor A is in the range of [-2, 2], which can increase the diversity of the search and enable the whale optimization algorithm to dynamically adjust between global search and local development. When A is large, the individual whale is more inclined to conduct global search; when A is small, the individual will focus on local development, which can balance the exploration and development capabilities of the whale optimization algorithm.

[0200] Among them, in the improved whale optimization algorithm, when r 1 When <0.5 and |A|≤1, the individual whale position is updated using formula (1);

[0201] Formula (1) is:

[0202] ;

[0203] ;

[0204] Indicates the next step position of the individual whale from its current position;

[0205] represents the current position of the individual whale; t represents the current number of iterations; T represents the maximum number of iterations; r represents a random number in the range [0, 1]; e represents the base of the natural logarithm; where, is the current best individual whale position; The decay function in , the search intensity can be gradually weakened during the iteration process to promote convergence.

[0206] In the improved whale optimization algorithm, when r 1 When <0.5 and |A|>1, the individual position of the whale is updated using formula (2);

[0207] Formula (2) is:

[0208] ;

[0209] is the position of a randomly selected whale individual, indicating a global search; in this embodiment, by selecting a random individual position , the algorithm can randomly explore in the search space to avoid falling into the local optimum. When |A|>1, the individual whale is far away from the current optimal position, which promotes it to jump out of the local search area and conduct new global exploration.

[0210] In the improved whale optimization algorithm, when r 1 When ≥0.5, the individual whale position is updated using formula (3);

[0211] Formula (3) is:

[0212] ;

[0213] ;

[0214] ;

[0215] is the spiral parameter;

[0216] l is a random number that changes randomly in the range [0, 1];

[0217] and are the initial maximum and minimum random ranges respectively.

[0218] In this embodiment, formula (3) simulates the position of the whale approaching the prey along the spiral path, effectively combining the characteristics of linear encirclement and nonlinear encirclement. The introduction of the spiral parameter can prevent the algorithm from falling into encirclement in a single direction, increase the diversity of the search path, and help to better approach the global optimal solution.

[0219] This embodiment also provides a TBM excavation speed prediction system, including:

[0220] At least one processor; and at least one memory in communication with the processor, wherein the memory stores program instructions executable by the processor, and the processor calls the program instructions to execute the TBM excavation speed prediction method as described above.

[0221] A TBM excavation speed prediction method in this embodiment predicts the TBM excavation speed through the iWOA-TSM model. iWOA is an improved whale optimization algorithm that can automatically optimize parameters during model training, avoiding errors and instability caused by manual parameter adjustment in traditional methods. A hybrid cross-dimensional module is used to perform deep feature extraction on the input data, effectively capturing the changing trends of the time dimension and feature dimension, and reducing the deviation and error of the model. The time projection module models the fused feature data, making full use of the temporal dependency information in the time series data, and improving the accuracy of the prediction results.

[0222] 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.

[0223] 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.

[0224] 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.

[0225] 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.

[0226] 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 TBM excavation speed prediction method, characterized in that: include: S1, obtaining TBM parameter data every second in the first time period before the current time point; S2, segmenting the TBM parameter data in the first time period before the current time point to obtain N parameter data sequence sets; S3, inputting N parameter data sequence sets into the trained iWOA-TSM model in sequence to obtain prediction results; the prediction results include: the predicted value of the TBM excavation speed per second in the second time period after the current time point; Wherein, a preset iWOA-TSM model is trained in advance using a training data set to obtain a trained iWOA-TSM model; The training data set includes: a plurality of parameter data sequence sets obtained by segmenting the TBM parameter data of the historical time period and the real values ​​of the tunneling speeds corresponding to the parameter data sequence sets obtained by segmenting the TBM parameter data of the historical time period; The iWOA-TSM model includes a hybrid cross-dimensional module and a time projection module in sequence; The hybrid cross-dimensional module is used to process the parameter data sequence set input to the iWOA-TSM model to obtain fused feature data; the fused feature data is obtained by fusing the time dimension features and feature dimension features of the parameter data sequence set input to the iWOA-TSM model; The time projection module is used to perform prediction based on the fused feature data to obtain a prediction result.

2. The TBM excavation speed prediction method according to claim 1, characterized in that: The TBM parameter data in the first time period before the current time point are segmented according to data, and the nth parameter data sequence set among the N parameter data sequence sets is X n , the n+ith parameter data sequence set is X n+i ; Among them, X n ={x （a） 、x （a+1） , ..., x （a+L－1） }; Among them, x （a） represents the TBM parameter data at the ath second in the first time period; x （a+L－1） , represents the TBM parameter data at the (a+L-1)th second in the first time period; L is the time dimension of the parameter data series set; X n+i ={x （a+i） 、x （a+i+1） , ..., x （a+i+L－1） }.

3. The TBM excavation speed prediction method according to claim 2, characterized in that: The first time period is the time period from the current time point to the first time point; The first time point is 30-128 seconds before the current time point; Each TBM parameter data includes: D parameter characteristics; The second time period is a time period from the current time point to the second time point; The second time point is between 128-500 seconds after the current time point.

4. The TBM excavation speed prediction method according to claim 3 is characterized in that: The hybrid cross-dimensional module includes: The first batch normalization layer is used to perform batch normalization processing on the parameter data sequence set in the input mixed cross-dimensional module to obtain the parameter data sequence set after batch normalization processing; A first dimension transformation unit, used to perform a first dimension transformation on the parameter data sequence set after batch normalization processing to obtain a parameter data sequence set after the first dimension transformation; A time matrix unit, used to calculate the parameter data sequence set after the first dimension transformation in the time dimension using a feedforward neural network to obtain a time matrix calculation result; A second dimension transformation unit, used for performing a second dimension transformation on the time matrix calculation result to obtain data after the second dimension transformation; A second batch normalization layer, used to perform batch normalization processing on the first data to obtain batch-normalized first data; The first data is the residual between the parameter data sequence set input into the hybrid cross-dimensional module and the data after the second dimension transformation; A feature dimension calculation unit, used to calculate the first data after batch normalization on the feature dimension by using a feedforward neural network to obtain a feature dimension calculation result; A third dimension transformation unit, used for performing a third dimension transformation on the second data to obtain data after the third dimension transformation; The second data is the residual between the first data and the characteristic dimension calculation result; A time and feature dimension matrix calculation unit, used to calculate the data transformed by the third dimension in the time dimension and the channel dimension using a feedforward neural network to obtain a comprehensive calculation result; The FEA unit is used to process the third data to obtain a processing result, and obtain output data of the hybrid cross-dimensional module according to the processing result and the third data; The third data is the residual between the comprehensive calculation result and the data after the second dimension transformation; Among them, the output data of the mixed cross-dimensional module is the residual between the processing result and the third data, and the output data of the mixed cross-dimensional module is used as the fused feature data.

5. The TBM excavation speed prediction method according to claim 4, characterized in that: The dimension of the parameter data sequence set after batch normalization is L×D; D is the characteristic dimension of each TBM parameter data in the parameter data series set; The dimension of the parameter data sequence set after the first dimension transformation is D×L; The dimension of the data after the second dimension transformation is L×D; The data after the third dimension transformation is represented by a L×D vector.

6. The TBM excavation speed prediction method according to claim 5, It is characterized in that The FEA unit processes the third data to obtain a processing result, which specifically includes: The FEA unit performs dimension-up mapping on the third data in the feature dimension to obtain first-dimensional data E; the feature dimension of the first-dimensional data E is S, where S≥5D; Normalizing the first dimensional data E in the time dimension using a softmax function to obtain normalized first dimensional data E; Perform a dimensionality reduction operation on the normalized first dimension data E along the feature dimension to restore the feature dimension that is the same as the third data, and obtain output data of the time dimension; The FEA unit transposes the third data in the time dimension and performs dimensionality-up mapping in the feature dimension to obtain second-dimensional data B; the feature dimension of the second-dimensional data B is S; The second dimension data B is normalized using the softmax function on the feature dimension to obtain the normalized second dimension data B; Performing a dimensionality reduction operation on the normalized second dimensional data B along the feature dimension to restore the feature dimension that is the same as the third data, and converting it back to the order of the time dimension of the third data to obtain output data of the feature dimension; The output data of the time dimension and the output data of the feature dimension are concatenated in the last dimension to obtain fused data; A dimension reduction operation is performed on the fused data in terms of feature dimension to make it consistent with the feature dimension of the third data, thereby obtaining a processing result.

7. The TBM excavation speed prediction method according to claim 6, characterized in that: The trained iWOA-TSM model refers to an optimized model obtained by tuning the improved whale optimization algorithm in the process of training the preset iWOA-TSM model using the training data.

8. The TBM excavation speed prediction method according to claim 7, It is characterized in that Wherein, in the improved whale optimization algorithm, the whale individuals are represented as hyperparameters in the iWOA-TSM model, and the positions of the whale individuals are updated to determine the optimal positions of the whale individuals; The control factor A in the improved whale optimization algorithm fluctuates randomly in the interval [-2, 2], and r1 is a random number uniformly distributed in the interval [0, 1]; Wherein, in the improved whale optimization algorithm, when r1<0.5 and |A|≤1, the individual position of the whale is updated using formula (1); Formula (1) is: ； ； Indicates the next step position of the individual whale from its current position; Indicates the current position of the individual whale; t represents the current iteration number; T is the maximum number of iterations; r is a random number in the range of [0, 1]; e is the base of natural logarithms; in, is the current best individual whale position; In the improved whale optimization algorithm, when r1<0.5 and |A|>1, the individual position of the whale is updated using formula (2); Formula (2) is: ； is the position of a randomly selected individual whale, indicating a global search; In the improved whale optimization algorithm, when r1≥0.5, the individual whale positions are updated using formula (3); Formula (3) is: ； ； ； l is a random number that changes randomly in the range [0, 1]; and are the initial maximum and minimum random ranges respectively.

9. The TBM excavation speed prediction method according to claim 8, characterized in that: The characteristics of each TBM parameter data in the parameter data sequence set include: TBM excavation speed, cutter head power, cutter head torque, thrust force, gripper shoe pressure, gripper shoe pump pressure, and main machine belt pump pressure.

10. A TBM excavation speed prediction system, characterized in that: include: at least one processor; as well as, At least one memory is communicatively connected to the processor, wherein the memory stores program instructions executable by the processor, and the processor calls the program instructions to execute the TBM excavation speed prediction method as described in any one of claims 1-9.

Images