Slurry balance shield attitude prediction method

Abstract

The invention provides a slurry balance shield attitude prediction method. The method comprises the steps of data preprocessing, model construction, model evaluation index establishment and the like. According to the method, a Transform model and an LSTM (Long Short Term Memory) model are combined, and the shield attitude is predicted through shield tunneling characteristic parameters. Compared with a traditional theory, the technology overcomes the limitation of a single theory, has the advantages of being lower in cost, higher in generalization ability and calculation efficiency, higher in prediction precision and the like, and has better performance in predicting the tunneling posture of the slurry balance shield.

Description

Technical Field

[0001] The present invention relates to the field of shield tunneling posture prediction, and in particular to a slurry balance shield posture prediction method. Background Art

[0002] With the rapid development of subway and road tunnel construction in the era of urbanization, shield machines (TBMs) are widely used in tunnel construction due to their high construction speed, minimal impact on the surrounding environment, and high safety. However, with increasing demands for high-quality construction and safety assurance, precise control of the TBM's posture has become a critical issue.

[0003] Inadequate shield machine attitude control can lead to a range of engineering problems, such as tunnel deviation from the designed route, segment assembly deviations, and tunnel misalignment. Currently, shield machine attitude control relies primarily on the engineering experience of the machine operator. There is currently no precise and feasible method for verifying the machine's tunneling attitude, making deviations from the designed route unavoidable during construction. Therefore, research is needed to accurately predict and control shield machine attitude. Summary of the Invention

[0004] The purpose of the present invention is to provide a slurry shield posture prediction method to solve the problems existing in the prior art.

[0005] The technical solution adopted to achieve the purpose of the present invention is as follows: a slurry shield posture prediction method, comprising the following steps:

[0006] 1) Collect the working condition data and shield posture data generated during the construction process. The working condition data includes propulsion speed, pitch angle arc, propulsion F group pressure, mud water tank pressure, cutterhead speed, cutterhead power, cutterhead extrusion pressure and propulsion pump pressure.

[0007] 2) Preprocess the data and divide it into training set and test set.

[0008] 3) Construct a slurry shield posture prediction model. This model uses a Transformer as a feature extractor, extracting global features from the input sequence through a self-attention mechanism. It also uses an LSTM as a decoder to process the extracted features and capture both short-term and long-term temporal dependencies. The input variables of the slurry shield posture prediction model are the shield machine's operating condition data from the past, and the output variables are the shield machine's posture parameters for the future.

[0009] 4) The slurry shield attitude prediction model was trained based on the training set and tested on the test set to select the slurry shield attitude prediction model with the best performance. The prediction performance of the model was evaluated using mean absolute error, root mean square error, and coefficient of determination.

[0010] 5) Input the shield working condition data before the time period to be predicted into the trained slurry balance shield posture prediction model to obtain the shield posture data prediction result for the time period to be predicted.

[0011] Furthermore, in step 1), the data is normalized to convert data of different dimensions and ranges to the same scale to eliminate the dimensional differences between features. Correlation analysis and hierarchical clustering are used to remove redundant feature parameters in the shield posture prediction process.

[0012] Furthermore, the time step or spatial position information is embedded into the data through position encoding. Position encoding is usually generated by sine and cosine functions, and the calculation formula is as follows:

[0013]

[0014] Where PE represents position encoding. pos represents data position. model Represents the data embedding dimension.

[0015] Furthermore, in step S1), the method for removing redundant feature parameters is:

[0016] A. By calculating the correlation matrix between features and target variables, we can select features with an absolute value of correlation coefficient greater than 0.2 to filter out factors with weak correlation with the target variable.

[0017] B. Use the “average linkage method” in hierarchical clustering to group the filtered features, and set the cluster distance threshold to 0.8 to reveal the similarity structure between features.

[0018] C. Select the feature with the highest correlation with the target variable in each cluster to form the final feature set.

[0019] Furthermore, the Transformer model framework includes an encoder and a decoder. The encoder and decoder include a multi-head attention layer, a feedforward neural layer, and a summation and regularization layer.

[0020] Furthermore, the multi-head attention layer maps data to output data using a query matrix, a key matrix, and a value matrix. The feedforward neural layer applies nonlinear transformations to the features at each position, further extracting deep features. The summation and regularization layers ensure that the model does not suffer from the vanishing gradient problem during data processing.

[0021] Furthermore, the LSTM model framework includes a forget gate, input gate, output gate, and cell state. The forget gate combines the cell state from the previous time step with the current input data to determine the data's retention state. The input gate determines whether information enters the storage cell by controlling the importance of the input data. The output gate controls the output of the neural network at the current time step by combining the current input, the previous hidden state, and the cell state. The cell state is a long-term storage of information. Through the control of the forget gate and input gate, the cell state can selectively retain or introduce new information.

[0022] Furthermore, step 3) specifically includes the following sub-steps:

[0023] 3.1) Preprocess the input data to ensure data quality and consistency. The processed data is mapped to a high-dimensional feature space through an embedding layer, and position encoding is introduced to preserve the position information of the sequence in the Transformer.

[0024] 3.2) The processed output feature data from the Transformer encoder is passed to the LSTM decoder, and the flow of information is dynamically adjusted through the LSTM gating mechanism.

[0025] 3.3) Through the fully connected layer, the output feature data is converted into the required prediction results.

[0026] The present invention also discloses a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the steps of the above method are implemented.

[0027] The present invention also discloses a slurry shield posture prediction device, comprising a memory, a processor, and executable instructions stored in the memory and executable in the processor. When the processor executes the executable instructions, the above method is implemented.

[0028] The technical benefits of this invention are undeniable: a big data-based machine learning approach, combined with the Transformer and LSTM models, predicts shield tunneling posture using characteristic parameters of shield tunneling. Compared to traditional theories, this technology overcomes the limitations of single theories and offers advantages such as lower cost, greater generalization, improved computational efficiency, and higher prediction accuracy. It also demonstrates superior performance in predicting slurry shield tunneling posture, providing valuable theoretical guidance for engineering design. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 It is a flow chart;

[0030] Figure 2 Schematic diagram of the Transformer model framework;

[0031] Figure 3 This is a schematic diagram of the LSTM recurrent unit framework;

[0032] Figure 4 Schematic diagram of the Transformer-LSTM model framework. DETAILED DESCRIPTION

[0033] The present invention will be further described below with reference to the following examples, but it should not be understood that the scope of the present invention is limited to the following examples. Without departing from the above technical ideas of the present invention, various substitutions and modifications can be made according to common technical knowledge and customary means in the art, and all should be included in the scope of protection of the present invention.

[0034] Example 1:

[0035] With the development of computer science, artificial intelligence has gradually been applied to the field of shield machine posture prediction. Big data-based machine learning methods can mine data and establish connections between construction parameters, ground parameters, and shield machine posture. However, traditional machine learning algorithms still have difficulty capturing high-dimensional data and suffer from gradient vanishing and gradient exploding issues.

[0036] See also Figure 1 This embodiment provides a method for predicting the posture of a slurry shield, comprising the following steps:

[0037] 1) Collect the working condition data and shield posture data generated during the construction process. The working condition data includes propulsion speed, pitch angle radian, roll angle radian, propulsion F group pressure, mud water tank pressure, cutterhead speed, cutterhead power, cutterhead extrusion pressure, as well as propulsion pump pressure, horizontal deviation and vertical deviation.

[0038] 2) Preprocess the data and divide it into training set and test set.

[0039] 3) Construct a slurry shield posture prediction model. This model uses a Transformer as a feature extractor, extracting global features from the input sequence through a self-attention mechanism. It also uses an LSTM as a decoder to process the extracted features and capture both short-term and long-term temporal dependencies. The input variables of the slurry shield posture prediction model are the shield machine's operating condition data from the past, and the output variables are the shield machine's posture parameters for the future.

[0040] 4) The slurry shield attitude prediction model is trained based on the training set, and the trained slurry shield attitude prediction model is tested based on the test set to select the slurry shield attitude prediction model with the best performance level. The mean absolute error, root mean square error and determination coefficient are used to evaluate the prediction performance of the model. MAE is used to evaluate the average deviation between the model prediction results and the actual data, reflecting the robustness of the model to the overall error; RMSE is used to quantify the degree of variance between the model's prediction results and the actual results, reflecting the overall error between the predicted value and the true value; R 2 Indicators used to quantify the goodness of fit of the regression model and evaluate the model's ability to explain the data.

[0041] 5) Input the shield working condition data at k1+1 moments before the time period to be predicted into the trained slurry balance shield posture prediction model to obtain the shield posture data prediction result for the time period to be predicted.

[0042] Example 2:

[0043] The main contents of this embodiment are the same as those of embodiment 1, wherein the data is normalized to convert data of different dimensions and ranges to the same scale to eliminate the dimensional differences between features;

[0044] Correlation analysis and hierarchical clustering methods are used to remove redundant characteristic parameters in the shield posture prediction process.

[0045] Specifically, during the data preprocessing process, the Transformer feature extractor cannot effectively process sequence data, and it is necessary to embed the time step or spatial position information into the data through position encoding. This position encoding is usually generated by sine and cosine functions, and the calculation formula is as follows:

[0046]

[0047] Where, PE represents position encoding; pos represents data position; d model Represents the data embedding dimension.

[0048] Specifically, the method for removing redundant feature parameters is:

[0049] First, by calculating the correlation matrix between features and the target variable (horizontal and vertical deviations of the shield tunneling posture), features with absolute correlation coefficients greater than 0.2 were selected to filter out factors with weak correlations with the target variable. Subsequently, the "average linkage method" in hierarchical clustering was used to group the selected features, with a cluster distance threshold of 0.8 to reveal the similarity structure between features. Finally, within each cluster, the feature with the highest correlation with the target variable was selected to form the final feature set.

[0050] Example 3:

[0051] The main content of this embodiment is the same as that of Embodiment 1 or 2, wherein Transformer serves as a feature extractor and extracts global features of the input sequence through a self-attention mechanism; LSTM serves as a decoder and processes the extracted features to capture short-term and long-term dependencies in the time series.

[0052] Specifically, the Transformer-LSTM model is composed of a Transformer model and an LSTM model.

[0053] Among them, the Transformer model framework is mainly composed of an encoder and a decoder. Figure 2 As shown in Figure 2, the encoder and decoder consist of a multi-head attention layer, a feedforward neural layer, and a summation and regularization layer. The multi-head attention layer maps data to output data using a query matrix, a key matrix, and a value matrix. The feedforward neural layer performs a nonlinear transformation on the features at each position to further extract deep features. The summation and regularization layer ensures that the model does not suffer from the vanishing gradient problem during data processing.

[0054] Among them, the LSTM model framework consists of a forget gate, an input gate, an output gate, and a cell state, such as Figure 3 As shown in the figure, the forget gate combines the cell state of the previous time step with the current input data to determine the data retention state; the input gate determines whether information enters the storage unit by controlling the importance of the input data; the output gate controls the output of the neural network at the current time step by combining the current input, the previous hidden state, and the cell state; the cell state is the long-term storage of information. Through the control of the forget gate and the input gate, the cell state can selectively retain or introduce new information.

[0055] Example 4:

[0056] The main contents of this embodiment are the same as any one of Embodiments 1 to 3, wherein step 3) specifically includes the following sub-steps:

[0057] 3.1) Preprocess the input data to ensure data quality and consistency. The processed data is mapped to a high-dimensional feature space through an embedding layer, and position encoding is introduced to preserve the position information of the sequence in the Transformer.

[0058] 3.2) The processed output feature data from the Transformer encoder is passed to the LSTM decoder, and the flow of information is dynamically adjusted through the LSTM gating mechanism.

[0059] 3.3) Through the fully connected layer, the output feature data is converted into the required prediction results, such as Figure 4 shown.

[0060] Example 5:

[0061] This embodiment provides a computer-readable storage medium having a computer program stored thereon. When the program is executed by a processor, the steps of the method described in any one of Embodiments 1 to 4 are implemented.

[0062] Example 6:

[0063] This embodiment provides a slurry shield posture prediction device, comprising a memory, a processor, and executable instructions stored in the memory and executable in the processor. When the processor executes the executable instructions, the method described in any one of embodiments 1 to 4 is implemented.

Claims

1. A method for predicting the posture of a slurry shield, characterized in that: The following steps are involved: 1) Collecting the working condition data and shield attitude data generated during the construction process of the shield machine; the working condition data includes propulsion speed, pitch angle radian, roll angle radian, propulsion F group pressure, mud water tank pressure, cutterhead speed, cutterhead power, cutterhead extrusion pressure, propulsion pump pressure, horizontal deviation and vertical deviation; 2) Preprocess the data and divide it into training set and test set; 3) Constructing a slurry shield posture prediction model; the slurry shield posture prediction model uses Transformer as a feature extractor and performs global feature extraction on the input sequence through a self-attention mechanism; the slurry shield posture prediction model uses LSTM as a decoder to process the extracted features and capture short-term and long-term dependencies in the time series; the input variables of the slurry shield posture prediction model are the shield machine's operating condition data at the past time, and the output variables are the shield machine's posture parameters at the future time; 4) The slurry shield attitude prediction model is trained based on the training set, and the trained slurry shield attitude prediction model is tested based on the test set to select the slurry shield attitude prediction model with the best performance level; the prediction performance of the model is evaluated using the mean absolute error, root mean square error, and coefficient of determination; 5) Input the shield working condition data before the time period to be predicted into the trained slurry balance shield posture prediction model to obtain the shield posture data prediction result for the time period to be predicted.

2. The method for predicting the posture of a slurry shield according to claim 1, characterized in that: In step 1), the data is normalized and data of different dimensions and ranges are converted to the same scale to eliminate the dimensional differences between features; correlation analysis and hierarchical clustering method are used to remove redundant feature parameters in the shield posture prediction process.

3. The method for predicting the posture of a slurry shield according to claim 1, characterized in that: The time step or spatial position information is embedded into the data through position encoding; the position encoding is usually generated by sine and cosine functions, and the calculation formula is as follows: Where, PE represents position code; pos represents data position; d model Represents the data embedding dimension.

4. A slurry shield posture prediction method according to claim 1, characterized in that: In step S1), the method for removing redundant feature parameters is: A. By calculating the correlation matrix between features and target variables, we can select features with a correlation coefficient greater than 0.2 to filter out factors with weak correlation with the target variable. B. Use the "average linkage method" in hierarchical clustering to group the filtered features and set the cluster distance threshold to 0.8 to reveal the similarity structure between features; C. Select the feature with the highest correlation with the target variable in each cluster to form the final feature set.

5. The method for predicting the posture of a slurry shield according to claim 1, characterized in that: The Transformer model framework includes an encoder and a decoder; the encoder and decoder include a multi-head attention layer, a feedforward neural layer, and a summation and regularization layer.

6. A slurry shield posture prediction method according to claim 4, characterized in that: The multi-head attention layer maps data to data output through the query matrix, key matrix, and value matrix. The feedforward neural layer performs nonlinear transformation on the features of each position to further extract deep features. The summation and regularization layers ensure that the model does not suffer from the vanishing gradient problem during data processing.

7. The method for predicting the posture of a slurry shield according to claim 1, characterized in that: The LSTM model framework includes a forget gate, an input gate, an output gate, and a cell state. The forget gate combines the cell state of the previous time step with the current input data to determine the data retention state. The input gate determines whether information enters the storage unit by controlling the importance of the input data. The output gate controls the output of the neural network at the current time step by combining the input at the current moment, the previous hidden state, and the cell state; The cell state is a long-term storage of information. Through the control of the forget gate and the input gate, the cell state can selectively retain or introduce new information.

8. The method for predicting the posture of a slurry shield according to claim 1, characterized in that: Step 3) specifically includes the following sub-steps: 3.1) Preprocess the input data to ensure data quality and consistency; map the processed data into a high-dimensional feature space through an embedding layer, and introduce position encoding to preserve the position information of the sequence in the Transformer; 3.2) The processed output feature data from the Transformer encoder is passed to the LSTM decoder, and the LSTM gating mechanism dynamically regulates the flow of information; 3.3) Through the fully connected layer, the output feature data is converted into the required prediction results.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

10. A slurry shield posture prediction device, characterized by: The method comprises a memory, a processor, and executable instructions stored in the memory and executable in the processor; when the processor executes the executable instructions, the method according to any one of claims 1 to 8 is implemented.

Images