Shield tunneling machine attitude prediction method and system

Abstract

The invention relates to the technical field of shield tunneling machine attitude prediction, and provides a shield tunneling machine attitude prediction method and system, and the method comprises the following steps: carrying out the correlation analysis and feature extraction of obtained input data; performing feature intersection and feature selection on the extracted features, and merging related parameters based on correlation analysis to obtain data after parameter merging; performing feature fusion on the features after parameter combination to obtain a comprehensive feature matrix; inputting the comprehensive feature matrix into a trained convolutional neural network model to obtain a shield tunneling machine attitude prediction result; and combining the historical attitude data of the shield tunneling machine with the predicted attitude of the shield tunneling machine, carrying out anomaly detection, and correcting the attitude prediction result of the shield tunneling machine. Through data fusion and integration of the multi-dimensional data, the features input into the attitude prediction model are unified, the data diversity is improved through data integration, and the attitude prediction accuracy is improved.

Description

Technical Field

[0001] The present disclosure relates to the technical field related to shield machine posture prediction, and specifically, to a shield machine posture prediction method and system. Background Art

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

[0003] With the continuous expansion of urban underground transportation networks, shield construction technology has attracted much attention as an efficient and safe method for underground tunnel construction. During the shield construction process, the attitude control of the shield machine is crucial to ensure the accuracy and safety of tunnel construction. However, traditional attitude control methods often rely on the subjective judgment of the operator, which has problems such as operation lag and low decision-making accuracy.

[0004] With the development of artificial intelligence technology, deep learning algorithms have shown excellent performance in many fields. In shield construction, the development of sensor technology has enabled a large amount of posture data to be collected and recorded in real time. Compared with traditional machine learning methods, the posture prediction model based on deep learning can make better use of these data to predict the posture of the shield machine. However, the current deep learning model construction process still requires a lot of professional knowledge and manpower investment, and lacks an automated end-to-end learning process, which limits its application in shield construction. And when processing the multidimensional data involved in the shield machine posture prediction process, the accuracy of data feature recognition of different dimensions and dimensions is a major challenge. When processing these multidimensional complex data, the existing deep learning network has the problem of low recognition accuracy, which directly affects the accuracy of posture prediction. Summary of the invention

[0005] In order to solve the above problems, the present invention proposes a shield machine posture prediction method and system, which unifies the features input into the posture prediction model through data fusion and integration of multi-dimensional data, and improves the diversity of data through data integration, thereby improving the accuracy of posture prediction.

[0006] In order to achieve the above objectives, the present disclosure adopts the following technical solutions:

[0007] One or more embodiments provide a shield machine posture prediction method, comprising the following steps:

[0008] Perform correlation analysis and feature extraction on the acquired input data;

[0009] Perform feature crossover and feature selection on the extracted features, merge related parameters based on correlation analysis, and obtain the data after parameter merging;

[0010] The features after parameter merging are subjected to feature fusion to obtain a comprehensive feature matrix;

[0011] The comprehensive feature matrix is ​​input into the trained convolutional neural network model to obtain the shield machine posture prediction result;

[0012] The historical posture data of the shield machine is combined with the predicted posture of the shield machine to form the time series data of the shield machine posture changes over time. Anomaly detection is performed based on the deep learning model to correct the prediction results of the shield machine posture.

[0013] One or more embodiments provide a shield machine posture prediction system, including:

[0014] Data feature extraction module: configured to perform correlation analysis and feature extraction on the acquired input data;

[0015] Data enhancement module: configured to perform feature crossover and feature selection on the extracted features, merge related parameters based on correlation analysis, and obtain parameter-merged data;

[0016] Data integration module: configured to perform feature fusion on the features after parameter merging to obtain a comprehensive feature matrix;

[0017] Prediction module: configured to input the comprehensive feature matrix into the trained convolutional neural network model to obtain the TBM posture prediction result;

[0018] Correction module: It is configured to merge the historical posture data of the shield machine with the predicted posture of the shield machine to form time series data of the change of the shield machine posture over time, perform anomaly detection based on the deep learning model, and correct the prediction results of the shield machine posture.

[0019] An electronic device comprises a memory and a processor, and computer instructions stored in the memory and running on the processor. When the computer instructions are run by the processor, the steps in the above-mentioned shield machine posture prediction method are completed.

[0020] A computer-readable storage medium is used to store computer instructions. When the computer instructions are executed by a processor, the steps in the above-mentioned shield machine posture prediction method are completed.

[0021] Compared with the prior art, the present invention has the following beneficial effects:

[0022] The present invention utilizes the correlation analysis and feature extraction technology of input data to extract the most representative features from complex and diverse data; through the steps of feature crossover, selection and merging, the relevant parameter data is further optimized and fused, reducing redundancy and improving the expressiveness of the data. Then, through the feature fusion technology, a comprehensive feature matrix is ​​formed, which is used as input for the trained convolutional neural network model to finally obtain the posture prediction result of the shield machine. In addition, by combining the historical posture data of the shield machine with the predicted data to form time series data, and then performing anomaly detection through the deep learning model, the preliminary prediction results are corrected to improve the accuracy and robustness of the posture prediction.

[0023] The present invention unifies the features input into the posture prediction model through data fusion and integration of multi-dimensional data, and improves the diversity of data through data integration, thereby enhancing the learning and generalization capabilities of the model. Ultimately, the accuracy of shield machine posture prediction has been significantly improved, especially in complex tunneling environments, and can effectively reduce prediction deviations caused by data redundancy or anomalies.

[0024] The advantages of the present disclosure and the advantages of additional aspects will be described in detail in the following specific embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] The accompanying drawings constituting a part of the present disclosure are used to provide a further understanding of the present disclosure. The exemplary embodiments of the present disclosure and the description thereof are used to explain the present disclosure but do not constitute a limitation of the present disclosure.

[0026] Figure 1 is a flowchart of a method for predicting the posture of a shield machine according to Embodiment 1 of the present disclosure;

[0027] Figure 2 is a structural diagram of a convolutional neural network model of Embodiment 1 of the present disclosure;

[0028] Figure 3 It is a schematic diagram of the convolutional neural network model training process of Example 1 of the present disclosure. DETAILED DESCRIPTION

[0029] The present disclosure is further described below in conjunction with the accompanying drawings and embodiments.

[0030] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present disclosure. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present disclosure belongs.

[0031] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit exemplary embodiments according to the present disclosure. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof. It should be noted that, in the absence of conflict, the various embodiments in the present disclosure and the features in the embodiments can be combined with each other. The embodiments will be described in detail below in conjunction with the accompanying drawings.

[0032] Example 1

[0033] In the technical solutions disclosed in one or more embodiments, Figures 1 to 3 As shown, a shield machine posture prediction method includes the following steps:

[0034] Step 1: Perform correlation analysis and feature extraction on the acquired input data;

[0035] Step 2: Perform feature crossover and feature selection on the extracted features, merge the related parameters based on correlation analysis, and obtain the data after parameter merging;

[0036] Step 3: Fuse the features after parameter merging to obtain a comprehensive feature matrix;

[0037] Step 4: Input the comprehensive feature matrix into the trained convolutional neural network model to obtain the shield machine posture prediction result;

[0038] Step 5: Merge the historical posture data of the shield machine with the predicted posture of the shield machine to form the time series data of the change of the shield machine posture over time, perform anomaly detection based on the deep learning model, and correct the prediction result of the shield machine posture.

[0039] In this embodiment, the most representative features are extracted from complex and diverse data using correlation analysis and feature extraction techniques of input data. Through feature crossover, selection and merging steps, the relevant parameter data is further optimized and fused, reducing redundancy and improving the expressiveness of the data. Then, through feature fusion technology, a comprehensive feature matrix is ​​formed, which is used as input for the trained convolutional neural network model to finally obtain the posture prediction result of the shield machine. In addition, by combining the historical posture data of the shield machine with the predicted data to form time series data, and then performing anomaly detection through a deep learning model, the preliminary prediction results are corrected to improve the accuracy and robustness of the posture prediction.

[0040] This implementation method unifies the features input into the posture prediction model through data fusion and integration of multi-dimensional data, and improves the diversity of data through data integration, enhancing the learning ability and generalization ability of the model. Ultimately, the accuracy of shield machine posture prediction has been significantly improved, especially in complex tunneling environments, and can effectively reduce prediction deviations caused by data redundancy or anomalies.

[0041] In step 1, the input data obtained include: internal data of the shield machine and external perception of the shield machine; wherein the internal data of the shield machine includes shield machine operation data and internal sensing data, and the external sensing data of the shield machine includes geological data;

[0042] Specifically, the shield machine operation data includes the current posture angle, thrust, torque, cutterhead status data, grouting data, guidance and positioning data, shield machine displacement (X-axis, Y-axis, Z-axis), shield machine forward acceleration and pressure, vibration monitoring data, shield machine structure stress and deformation monitoring data, etc.

[0043] Geological data includes soil type, rock layer distribution, water content and other data, which can be connected to the geographic information system GIS collection and combined with the shield machine operation data to jointly characterize the influencing factors of the shield machine posture;

[0044] In this embodiment, the shield machine operation data and the geological data collected based on GIS are integrated to form a unified data set, which can be used for multi-modal prediction of the shield machine posture;

[0045] In step 1, correlation analysis and feature extraction are performed on the acquired input data, including the following steps:

[0046] Step 1: Preprocess the acquired input data, including data cleaning, data standardization and missing value processing;

[0047] Data cleaning: remove invalid data and noise, such as detecting and removing extreme values ​​and noise data caused by sensor failure or other reasons to ensure the validity of the data;

[0048] Data standardization: Standardize feature data of different scales to the same range (such as between 0 and 1) to eliminate dimensional differences and ensure that the model processes different features consistently.

[0049] Missing value processing: fill or remove missing values, use interpolation, mean filling or other appropriate methods to fill missing values ​​in time series, or remove severely missing data rows to ensure data integrity and continuity;

[0050] In step 1, the preprocessed data is subjected to correlation analysis according to the data type, and the correlation between the input data is calculated;

[0051] 11) Perform linear correlation analysis on the internal data of the shield machine (such as displacement, acceleration, pressure, etc.), calculate the Pearson correlation coefficient between each feature, set the first coefficient threshold, and identify the data exceeding the set first coefficient threshold as highly correlated feature data;

[0052] 12) Perform nonlinear correlation analysis on the shield machine external sensing data, use mutual information and distance correlation to capture the nonlinear relationship between features, and identify data whose correlation exceeds the set threshold;

[0053] Specifically, methods such as mutual information can be used to analyze the nonlinear relationship between the external perception data of the shield machine (such as soil type and moisture content in geological data) and the operating parameters of the shield machine, and capture complex associations that cannot be identified by traditional linear methods;

[0054] In step 1, feature extraction is performed. Specifically, principal component analysis (PCA) is performed on the high-dimensional data set obtained after linear correlation analysis and nonlinear correlation analysis to extract the main feature components and retain the most important information in the data. This will reduce the dimension of the data and reduce the computational complexity while maintaining the key features of the data;

[0055] In the above scheme, the input data is fully preprocessed and optimized before entering the posture prediction model. The data cleaning and standardization steps improve the quality of the data, the correlation analysis identifies the key features in the data, and the principal component analysis reduces the data dimension, making the model more efficient for training and recognition. In addition, the combination of linear correlation analysis and nonlinear correlation analysis ensures that the correlation of different types of data can be accurately identified, thereby further improving the prediction ability and accuracy of the model.

[0056] In step 2, the extracted features are subjected to feature crossover and feature selection, and the related parameters are merged based on correlation analysis. The steps are as follows:

[0057] Step 21, feature crossover: perform crossover operation on the extracted features to generate new combined features; different types of data can be combined into one feature data, such as performing feature crossover on displacement, acceleration, and pressure to combine the three data into one feature data;

[0058] Step 22, feature selection: Based on the importance scoring method, select the features generated after feature crossing; filter out the features that contribute most to posture prediction, eliminate irrelevant or redundant features, and optimize model performance.

[0059] Specifically, the importance scoring method can adopt Lasso regression, random forest and other methods;

[0060] Step 23: Based on the correlation analysis results, for the data after feature selection, merge the data whose correlation exceeds the set threshold;

[0061] In the above scheme, the input dimension of the shield machine posture prediction model can be simplified by combining highly correlated parameters through correlation analysis;

[0062] In this embodiment, the extracted data of different types are first combined through feature crossover operation to generate new features to improve the data's expressiveness and information density. Next, the importance scoring method is used to screen these cross-generated features, retaining the most important features for posture prediction while eliminating irrelevant or redundant features. Finally, based on the correlation analysis results, the screened feature data is merged again to optimize the structure of the feature set, reduce the complexity of the model, and improve the training efficiency of the model.

[0063] Through feature crossover operations, new combined features are creatively generated, enriching the expressiveness of the data and helping the model capture more complex patterns. Through feature selection, it is ensured that the model only uses the most relevant features, reducing computational overhead and the risk of overfitting. In addition, by further merging relevant features, the data structure is simplified, making the model more efficient and accurate in posture prediction.

[0064] In step 3, the features after the parameters are merged are fused to obtain a comprehensive feature matrix. Specifically, the features after the parameters are merged are normalized first, and then the feature data after the parameters are merged are connected, fused and spliced ​​using a full connection operation;

[0065] Among them, the merged feature data includes sensor data, geological data and shield machine operation parameter data; the processed sensor data, geological data and shield machine operation parameters are feature fused to form a comprehensive feature matrix.

[0066] Through the above steps, effective processing and integration of input data can be achieved, and the predictive ability and stability of the model can be enhanced.

[0067] In some embodiments, the shield machine posture prediction model can adopt a convolutional neural network model, including multiple convolutional layers and pooling layers, for extracting high-level feature representations from raw data.

[0068] Furthermore, this embodiment introduces a spatial attention mechanism into the convolution operation of the convolutional neural network model. The structure of the improved convolutional neural network model is as follows: Figure 2As shown, it includes: an input layer, a convolution layer, an attention layer, a pooling layer, a fully connected layer and an output layer connected in sequence; wherein the convolution layer, the attention layer and the pooling layer connected in sequence are used as a group of convolution blocks based on the attention mechanism, and one or more convolution blocks based on the attention mechanism are set in the model;

[0069] Input Layer: used to receive the comprehensive feature matrix obtained in step 3;

[0070] Convolutional layer: used to perform convolution operations on the input feature data, capture local spatial features, and form feature maps;

[0071] In this embodiment, there are two convolutional layers in the figure, namely Conv Layer I and Conv Layer II, which are located in different parts, indicating that the network extracts features twice.

[0072] Attention Layer: Based on the attention mechanism, it focuses on specific parts of the input data and dynamically adjusts the weights of certain features, thereby improving the performance of the model.

[0073] Each convolutional layer in the convolutional neural network of this embodiment is followed by an attention layer, which means that after feature extraction, the network re-weights important features;

[0074] Pooling Layer: It is used to reduce the dimension of feature maps. Maximum pooling or average pooling can be used. This process can reduce computational complexity and retain important information.

[0075] The convolutional neural network model in this embodiment has two pooling layers, namely Pooling Layer I and Pooling Layer II, which follow the attention layer and can further process the feature map adjusted by the attention mechanism;

[0076] Fully connected layer (FC Layer): The fully connected layer flattens the feature map of the previous layer into a one-dimensional vector and connects it to the output layer. The function of the fully connected layer is to map the abstract features to the final classification or regression results. Figure 2 There are two fully connected layers, FC Layer I and FC Layer II, which are used to increase the expressive power of the network;

[0077] Output Layer: The output layer is used to generate the final prediction result. It can use softmax function or sigmoid function, depending on the category of the task, which is classification or regression.

[0078] In this embodiment, a spatial attention mechanism is introduced into the convolution operation, so that the model can focus on specific areas in the multi-sensor data, such as specific parts of the shield machine or areas with changing geological conditions. At the same time, a modular network structure is designed to facilitate real-time adjustment and optimization according to different shield parameters and geological conditions;

[0079] Furthermore, in the convolutional neural network model training step, the stochastic gradient descent optimization algorithm used may include batch gradient descent, mini-batch gradient descent, or stochastic gradient descent methods.

[0080] As a feasible solution, in the convolutional neural network model evaluation step, the prediction accuracy evaluation includes calculating indicators such as the mean square error and root mean square error between the predicted results and the actual results.

[0081] Optionally, during the training of the convolutional neural network model, a transfer learning method is used to apply a pre-trained model obtained from a similar prediction task on the shield machine to the posture prediction task, and during the tunneling process of the shield machine, the data of the shield machine is updated to the training data set for the shield machine posture prediction, and end-to-end automated training is used to continuously update the parameters of the convolutional neural network model;

[0082] Furthermore, during the training process, the hyperparameters such as learning rate, batch size, number of network layers and number of neurons are optimized through Bayesian optimization and genetic algorithm of automatic deep learning. Finally, a feedback loop mechanism is established, that is, a feedback loop from the posture prediction of the shield machine to the construction adjustment, and the feedback of construction results is used to further optimize the model and prediction process.

[0083] The above training scheme can expand the amount of data and speed up the model fitting. At the same time, the incremental learning method is also adopted. As the tunneling progress of the shield machine progresses, the model parameters are continuously updated, so that the model can more accurately predict the current posture information.

[0084] During the training of the convolutional neural network model, Bayesian optimization and genetic algorithm in the automatic deep learning framework are used to automatically optimize hyperparameters. The specific parameter optimizations are as follows:

[0085] 1) Learning rate adjustment: Dynamically adjust the learning rate through Bayesian optimization to find the optimal learning rate to accelerate the convergence of the model;

[0086] 2) Optimization of the number of network layers and neurons: Genetic algorithms are used to automatically adjust the number of convolutional layers, pooling layers, and fully connected layers and the number of neurons in each layer to ensure the optimal model complexity:

[0087] 3) Batch size adjustment: Dynamically adjust the batch size during training to adapt to different data scales, avoid memory overload and improve training efficiency:

[0088] 4) Hyperparameter search and adjustment: Automatically search for the best hyperparameter combination, such as activation function, regularization coefficient, etc., through grid search or random search to improve the generalization ability of the model.

[0089] Step 5 is the process of correcting the prediction results, including the following steps:

[0090] Step 51, preprocessing the time series data of the change of the shield machine posture over time, and performing time series analysis to obtain the shield machine posture change on different time scales;

[0091] Optionally, the time series data can be denoised and normalized;

[0092] Specifically, wavelet transform is used to decompose the time series data and extract features of different scales;

[0093] Step 52: Divide the time series data after time series analysis according to the set sliding window, and use each window data as an analysis unit;

[0094] Step 53: extract statistical time domain features from the window data, including mean, variance, kurtosis, skewness, etc.; for multidimensional time series, it is also necessary to extract relevant features between dimensions;

[0095] Step 54, frequency domain analysis: Use Fourier transform to convert time domain data to frequency domain and extract frequency domain features, such as frequency distribution, energy spectrum, etc.

[0096] Step 55: Input the extracted time domain features and frequency domain features into the trained LSTM autoencoder deep learning model for Xining recognition, identify abnormal time window data based on reconstruction errors, and correct the shield machine posture prediction results based on the abnormal recognition results.

[0097] Specifically, the anomaly detection process of the LSTM autoencoder is as follows:

[0098] Step 5-1: Model training: Use normal time series data to train the LSTM autoencoder model to minimize the reconstruction error;

[0099] Step 5-2: Input the time series data into the trained LSTM autoencoder to generate reconstructed data;

[0100] Step 5-3, calculating the reconstruction error of each time window, such as calculating the mean square error (MSE) between the reconstructed data and the original data;

[0101] Step 5-4, detect anomaly: According to the set threshold, when the reconstruction error is greater than the set threshold, the corresponding time window data is judged to be abnormal data.

[0102] This implementation greatly improves the accuracy of anomaly detection by combining feature extraction and analysis in the time domain and frequency domain, so that the shield machine posture prediction results can be adjusted in time to avoid prediction errors caused by data anomalies. Using LSTM autoencoders to process time series data can effectively capture long-term dependencies in the data, further improve the ability to process complex time series data, and enhance the robustness and accuracy of posture prediction.

[0103] Example 2

[0104] Based on Example 1, this embodiment provides a shield machine posture prediction system, including:

[0105] Data feature extraction module: configured to perform correlation analysis and feature extraction on the acquired input data;

[0106] Data enhancement module: configured to perform feature crossover and feature selection on the extracted features, merge related parameters based on correlation analysis, and obtain parameter-merged data;

[0107] Data integration module: configured to perform feature fusion on the features after parameter merging to obtain a comprehensive feature matrix;

[0108] Prediction module: configured to input the comprehensive feature matrix into the trained convolutional neural network model to obtain the TBM posture prediction result;

[0109] Correction module: It is configured to merge the historical posture data of the shield machine with the predicted posture of the shield machine to form time series data of the change of the shield machine posture over time, perform anomaly detection based on the deep learning model, and correct the prediction results of the shield machine posture.

[0110] It should be noted here that the various modules in this embodiment correspond one-to-one to the various steps in Example 1, and the specific implementation process is the same, which will not be repeated here.

[0111] Example 3

[0112] This embodiment provides an electronic device, including a memory and a processor, and computer instructions stored in the memory and running on the processor. When the computer instructions are run by the processor, the steps of a shield machine posture prediction method in embodiment 1 are completed.

[0113] Example 4

[0114] This embodiment provides a computer-readable storage medium for storing computer instructions. When the computer instructions are executed by a processor, the steps in a shield machine posture prediction method in Embodiment 1 are completed.

[0115] The above description is only a preferred embodiment of the present disclosure and is not intended to limit the present disclosure. For those skilled in the art, the present disclosure may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present disclosure shall be included in the protection scope of the present disclosure.

[0116] Although the above describes the specific implementation methods of the present disclosure in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present disclosure. Technical personnel in the relevant field should understand that on the basis of the technical solution of the present disclosure, various modifications or variations that can be made by those skilled in the art without creative work are still within the scope of protection of the present disclosure.

Claims

1. A shield machine posture prediction method, characterized in that: The steps include: Perform correlation analysis and feature extraction on the acquired input data; Perform feature crossover and feature selection on the extracted features, merge related parameters based on correlation analysis, and obtain the data after parameter merging; The features after parameter merging are subjected to feature fusion to obtain a comprehensive feature matrix; The comprehensive feature matrix is ​​input into the trained convolutional neural network model to obtain the shield machine posture prediction result; The historical posture data of the shield machine is combined with the predicted posture of the shield machine to form the time series data of the shield machine posture changes over time. Anomaly detection is performed based on the deep learning model to correct the prediction results of the shield machine posture.

2. A shield machine posture prediction method as claimed in claim 1, characterized in that: The obtained input data is subjected to correlation analysis and feature extraction, including the following steps: Preprocess the acquired input data, including data cleaning, data standardization, and missing value processing; Perform correlation analysis on the preprocessed data according to data type and calculate the correlation between input data; Conduct linear correlation analysis on the internal data of the shield machine body, calculate the Pearson correlation coefficient between each feature, set a first coefficient threshold, and identify data exceeding the set first coefficient threshold as highly correlated feature data; Conduct nonlinear correlation analysis on shield machine external sensing data, use mutual information and distance correlation to capture nonlinear relationships between features, and identify data with correlation exceeding the set threshold; Or, perform feature extraction, specifically: Principal component analysis is performed on the high-dimensional data set obtained after linear and nonlinear correlation analysis to extract the main characteristic components as the extracted features.

3. A shield machine posture prediction method as claimed in claim 1, characterized in that: The steps of performing feature crossover and feature selection on the extracted features and merging related parameters based on correlation analysis are as follows: The extracted features are cross-operated to generate new combined features, so that different types of data can be combined into one piece of feature data; Based on the importance scoring method, the features generated after feature crossover are selected; the features that contribute to posture prediction are screened out, and irrelevant or redundant features are eliminated; Based on the results of correlation analysis, for the data after feature selection, the data whose correlation exceeds the set threshold is merged.

4. A shield machine posture prediction method as claimed in claim 1, characterized in that: The spatial attention mechanism is introduced in the convolution operation of the convolutional neural network model. The improved convolutional neural network model structure includes: an input layer connected in sequence, one or more convolution blocks based on the attention mechanism, a fully connected layer and an output layer; the convolution block based on the attention mechanism includes a convolution layer, an attention layer and a pooling layer connected in sequence.

5. A shield machine posture prediction method according to claim 1, characterized in that: The stochastic gradient descent optimization algorithm used in the convolutional neural network model training step; Or / and, during the training process of the convolutional neural network model, the pre-trained model obtained from similar prediction tasks about the shield machine is applied to the posture prediction task by using the transfer learning method, and during the excavation process of the shield machine, the data of the shield machine is updated to the training data set of the shield machine posture prediction, and end-to-end automatic training is used to continuously update the parameters of the convolutional neural network model.

6. A shield machine posture prediction method according to claim 1, characterized in that: The process of correcting the prediction results includes the following steps: Preprocess the time series data of the change of shield machine posture over time, and perform time series analysis to obtain the change of shield machine posture on different time scales; According to the set sliding window, the time series data after time series analysis is divided, and each window data is used as an analysis unit; Extract statistical time domain features from window data; Use Fourier transform to convert time domain data into frequency domain and extract frequency domain features; The extracted time domain features and frequency domain features are input into the trained LSTM autoencoder deep learning model for recognition. Based on the reconstruction error, the abnormal time window data is identified, and the shield machine posture prediction results are corrected based on the abnormal recognition results.

7. A shield machine posture prediction method according to claim 1, characterized in that: The anomaly detection process of LSTM autoencoder is as follows: Use normal time series data to train the LSTM autoencoder model to minimize the reconstruction error; Input the time series data into the trained LSTM autoencoder to generate reconstructed data; Calculate the reconstruction error for each time window; According to the set threshold, when the reconstruction error is greater than the set threshold, the corresponding time window data is judged to be abnormal data.

8. A shield machine posture prediction system, characterized in that: include: Data feature extraction module: configured to perform correlation analysis and feature extraction on the acquired input data; Data enhancement module: configured to perform feature crossover and feature selection on the extracted features, merge related parameters based on correlation analysis, and obtain parameter-merged data; Data integration module: configured to perform feature fusion on the features after parameter merging to obtain a comprehensive feature matrix; Prediction module: configured to input the comprehensive feature matrix into the trained convolutional neural network model to obtain the TBM posture prediction result; Correction module: It is configured to merge the historical posture data of the shield machine with the predicted posture of the shield machine to form time series data of the change of the shield machine posture over time, perform anomaly detection based on the deep learning model, and correct the prediction results of the shield machine posture.

9. An electronic device, characterized in that: The method comprises a memory and a processor, and computer instructions stored in the memory and executed on the processor. When the computer instructions are executed by the processor, the steps in the method for predicting the posture of a shield machine described in any one of claims 1 to 7 are completed.

10. A computer-readable storage medium, characterized in that: Used to store computer instructions, which, when executed by a processor, complete the steps of a shield machine posture prediction method as described in any one of claims 1-7.

Images