A shield tunneling posture intelligent prediction and control method and system

Abstract

The application provides a shield tunneling posture intelligent prediction and control method and system, the method comprising: acquiring real-time values of key decision parameters in a decision cycle of a shield tunneling machine; the key decision parameters are obtained from multi-source heterogeneous data based on a trained GANDALF model; establishing a segmented shield posture control principle based on a design axis line type of shield tunneling; inputting the real-time values of the key decision parameters in the decision cycle into the trained BO-GANDALF model, outputting a shield tunneling posture prediction result in a prediction stage, and outputting an intelligent control decision of the shield posture in a control stage in combination with the shield tunneling posture prediction result and the segmented shield posture control principle.

Description

Technical Field

[0001] This invention belongs to the field of tunnel boring machine (TBM) construction, specifically relating to a method and system for intelligent prediction and control of TBM tunneling posture. Background Technology

[0002] When the tunnel boring machine (TBM) is turning, the hydraulic cylinders on both sides remain in a state of large extension or retraction on one side for an extended period. The adjustable range of the hydraulic cylinders used to control the TBM's attitude is even smaller, making it more difficult to control the TBM's tunneling posture. Improper attitude control can easily cause the TBM to move in a "snake-like" pattern, causing its trajectory to deviate from the design axis.

[0003] Traditional methods for predicting tunnel boring machine (TBM) attitude include experimental methods, theoretical formula methods, and numerical analysis methods. However, these methods have limited accuracy in predicting TBM attitude and cannot provide prediction results based on real-time monitoring data from actual engineering projects. Therefore, there are certain limitations to establishing accurate TBM attitude and position prediction models using traditional research methods.

[0004] Currently, shield tunneling attitude control faces two bottlenecks: First, existing attitude control methods heavily rely on the subjective experience and judgment of operators. Limited by the dynamic complexity of the underground engineering environment and the driver's personal experience, attitude control suffers from insufficient real-time performance and accuracy deviations. This easily leads to deviations between the tunneling trajectory and the design axis, reducing tunneling efficiency and increasing construction risk. Second, the value of multi-source heterogeneous data is insufficiently exploited. The attitude parameters, position parameters, and tunneling parameters generated in real-time during shield tunneling, along with low-dimensional geological data and tunnel geometry data, constitute a massive multi-source heterogeneous dataset. This dataset contains a wealth of tunneling information, but existing data analysis methods have failed to construct effective feature extraction and knowledge discovery mechanisms, resulting in a lack of data-driven decision support for parameter optimization. Third, the classification of shield attitude control principles for different linear axes is unclear. Most attitude control methods simply aim to bring the shield attitude parameter values ​​closer to zero, but this is divorced from actual construction conditions.

[0005] Therefore, there is an urgent need for an intelligent prediction and control method for shield tunneling posture for different axial alignments of straight lines and curves. Summary of the Invention

[0006] To overcome the shortcomings of existing shield tunneling attitude control methods, which heavily rely on the subjective experience and judgment of operators and lack sufficient value mining of multi-source heterogeneous data, thus failing to meet the requirements of shield tunneling attitude control, this invention provides a method and system for intelligent prediction and control of shield tunneling attitude. Through the BO-GANDALF deep learning model, it achieves precise real-time control of spatial attitude parameters, improves construction efficiency, reduces tunneling deviation, and realizes intelligent optimization control of decision parameters.

[0007] According to one aspect of the present invention, a method for intelligent prediction and control of shield tunneling posture is provided, comprising:

[0008] The real-time values ​​of key decision parameters within the decision-making cycle of the tunnel boring machine are obtained; these key decision parameters are selected from multi-source heterogeneous data based on a trained GANDALF model.

[0009] The principle of segmented shield attitude control is established based on the design axis alignment of shield tunneling.

[0010] The real-time values ​​of key decision parameters within the decision cycle are input into the trained BO-GANDALF model. In the prediction stage, the shield tunneling attitude prediction results are output. In the control stage, the shield tunneling attitude prediction results and the segmented shield attitude control principle are combined to output the intelligent control decision of the shield attitude.

[0011] The training process of the BO-GANDALF model includes:

[0012] Construct a multi-source heterogeneous database for training based on multi-source heterogeneous data;

[0013] A BO-GANDALF model is constructed and trained using a multi-source heterogeneous database. In the prediction phase, the BO-GANDALF model processes the multi-source heterogeneous data (excluding shield attitude parameters) through a GFLU feature extraction module, extracts optimized feature representations, and outputs a feature parameter contribution matrix. A feedforward neural network then outputs predicted values ​​of the shield attitude parameters based on the optimized feature representations and the feature parameter contribution matrix. In the control phase, the BO global optimization strategy, based on the prediction phase results and the segmented shield attitude control principle, solves for the optimal solution within the adjustable range of key decision parameters, outputting intelligent control decisions for the shield attitude.

[0014] As a further technical solution, the screening steps for key decision parameters include:

[0015] Based on the feature parameter contribution value matrix output by the trained BO-GANDALF model, the Top-K feature parameter types with high contribution values ​​are selected from a multi-source heterogeneous database as key decision parameters for selection.

[0016] As a further technical solution, GFLU feature extraction consists of several gated learning units stacked together. Each gated learning unit introduces a learnable parameter vector and uses the t-softmax function to convert the parameter vector into a feature mask. The feature mask is added to the input features of the layer through element-wise multiplication to obtain the feature representation extracted by the layer. Accordingly, the feature representation extracted by the last gated learning unit is output as the optimized feature representation, and the feature mask of the last gated learning unit is used as the feature parameter contribution matrix.

[0017] As a further technical solution, the segmented shield tunneling attitude control principle divides the straight-line segment and the turning segment according to the design axis alignment, including:

[0018] The principle for controlling the shield attitude in straight sections is to control the sum of the absolute values ​​of the deviations between the shield machine attitude parameters and the design axis within a preset deviation distance range on both sides of the design axis. Based on the principle for controlling the shield attitude in straight sections, the principle for controlling the shield attitude in turning sections introduces an additional direction preference to promote turning tendency and controls the horizontal deviation of the shield head within a preset deviation distance range inside the design axis.

[0019] As a further technical solution, the principle of segmented shield tunneling attitude control is expressed in the form of an objective function, the objective function for the straight segment is as follows:

[0020]

[0021] In the formula, The objective function for the attitude parameters of the straight-line segment; These are the shield attitude parameters that need to be controlled during the straight-line section, representing the shield head horizontal deviation, shield tail horizontal deviation, shield head vertical deviation, and shield tail vertical deviation relative to the design axis, respectively. It is a constant; This indicates the critical value of the preset deviation distance range;

[0022] The objective function for the turning segment introduces a direction preference to promote steering tendency in the objective function for the straight segment, and adds a regularization term to control the horizontal deviation of the shield head within a preset deviation distance range. The mathematical expression is as follows:

[0023]

[0024] In the formula, The objective function for the attitude parameters during the turning segment; ; and It is the penalty coefficient; It is related to the stability control weight of the tunnel boring machine; This indicates the critical value of the preset deviation distance range inside the design axis.

[0025] As a further technical solution, the process of obtaining the intelligent control decision of the tunnel boring machine's attitude by using the trained BO-GANDALF model to solve for the optimal solution within the adjustable range of key decision parameters includes:

[0026] Based on the adjustable range of key decision parameters, a control parameter space is constructed. In each iteration, several control parameter combinations are sampled from the control parameter space as evaluation points. A Gaussian process is used to model the mapping relationship between the shield tunneling posture and the control parameter combination in the BO-GANDALF model to obtain a surrogate model. The surrogate model is used to output the shield tunneling posture prediction results for each evaluation point.

[0027] The segmented shield tunnel attitude control principle is used as the objective function of the surrogate model. The expected improvement of the objective function at each candidate point is calculated, and the evaluation point for the next round is selected by the acquisition function.

[0028] The process is iteratively optimized until the stopping criterion is met. The output is the evaluation point that makes the expected value of the objective function optimal. The combination of control parameters represented by this point is taken as the optimal solution within the adjustable range of the key decision parameters, thus obtaining the intelligent control decision of the shield attitude.

[0029] As a further technical solution, the acquisition function is improved based on expectations, and its mathematical expression is as follows:

[0030]

[0031] In the formula, The x represents the expected improvement value, where x is the evaluation point and represents the combination of control parameters to be evaluated; x best To reach the optimal evaluation point in the historical evaluation; Let be the expected value of the objective function to be optimized. It is the expected value of the current optimal objective function. This represents the expected value.

[0032] According to another aspect of this specification, a shield tunneling attitude intelligent prediction and control system is provided, comprising:

[0033] The input module is used to obtain the real-time values ​​of key decision parameters within the decision-making cycle of the tunnel boring machine; the key decision parameters are obtained from multi-source heterogeneous data based on the trained GANDALF model.

[0034] The control principle construction module is used to establish segmented shield attitude control principles based on the design axis alignment of shield tunneling.

[0035] The shield tunneling attitude prediction and control module is used to input the real-time values ​​of key decision parameters within the decision cycle into the trained BO-GANDALF model. In the prediction stage, it outputs the shield tunneling attitude prediction results. In the control stage, it combines the shield tunneling attitude prediction results and the segmented shield attitude control principle to output the intelligent control decision of the shield attitude.

[0036] The training process of the BO-GANDALF model includes:

[0037] Construct a multi-source heterogeneous database for training based on multi-source heterogeneous data;

[0038] A BO-GANDALF model is constructed and trained using a multi-source heterogeneous database. In the prediction phase, the BO-GANDALF model processes the multi-source heterogeneous data (excluding shield attitude parameters) through a GFLU feature extraction module, extracts optimized feature representations, and outputs a feature parameter contribution matrix. A feedforward neural network then outputs predicted values ​​of the shield attitude parameters based on the optimized feature representations and the feature parameter contribution matrix. In the control phase, the BO global optimization strategy, based on the prediction phase results and the segmented shield attitude control principle, solves for the optimal solution within the adjustable range of key decision parameters, outputting intelligent control decisions for the shield attitude.

[0039] According to another aspect of this specification, an electronic device is provided, including a memory and a processor, the memory storing program instructions executed by the processor, the processor calling the program instructions to execute a method for intelligent prediction and control of tunnel boring machine posture.

[0040] According to another aspect of this specification, a non-transitory computer-readable storage medium is provided, the non-transitory computer-readable storage medium storing computer instructions that cause the computer to execute a method for intelligent prediction and control of tunnel boring machine posture.

[0041] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0042] This invention, based on the BO-GANDALF deep learning model, constructs an integrated predictive-control adaptive system for various axis alignments, enabling precise real-time control of spatial attitude parameters. This reduces the reliance on operators' subjective experience and judgment in current attitude control methods. It avoids "snake-like" tunneling caused by improper operation, improving construction efficiency and reducing tunneling deviations. Furthermore, it achieves intelligent optimization control of decision parameters, improving the stability of the tunnel boring machine's attitude control and ensuring construction safety.

[0043] By extracting feature masks using the GFLU module in the BO-GANDALF model, the contribution of feature parameters to attitude parameters can be obtained. This identifies the feature parameters among the control parameters that have the most significant impact on the overall attitude. It reveals the key factors affecting the tunnel boring machine's (TBM) attitude, screens the decision parameters for TBM attitude, and provides knowledge and data support for TBM attitude control.

[0044] Based on the segmented shield tunneling attitude control principle that distinguishes between axis alignment types, the BO-GANDALF model is used to regulate the total thrust affecting the attitude. When the total thrust is controlled within a certain range, the shield attitude is always maintained within the required range to ensure tunnel construction safety and efficiency.

[0045] This invention is based on multi-source heterogeneous data fusion technology and adopts a dual-driven approach of knowledge and data to improve the depth of data mining and the ability to predict shield attitude, thereby achieving intelligent control of total propulsion. The adaptive fuzzy attitude optimization method ensures the dual guarantee of optimal control of the construction axis and construction efficiency, thereby improving the shield tunnel construction capability and intelligence level. Attached Figure Description

[0046] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0047] Figure 1 A flowchart illustrating an intelligent prediction and control method for shield tunneling posture provided in an embodiment of the present invention;

[0048] Figure 2 This is a schematic diagram of the prediction stage structure of the BO-GANDALG model in an embodiment of the present invention;

[0049] Figure 3 This is a flowchart of the BO-GANDALF model in an embodiment of the present invention;

[0050] Figure 4 This is a schematic diagram of the attitude prediction results of the BO-GANDALF model using the tunneling parameters of Guangzhou Metro as an example in an embodiment of the present invention.

[0051] Figure 5 This is a schematic diagram illustrating the BO-GANDALF attitude parameter optimization results using Guangzhou Metro tunneling parameters as an example in this embodiment of the invention.

[0052] Figure 6 This is a schematic diagram illustrating the optimization results of BO-GANDALF key decision parameters, using Guangzhou Metro tunneling parameters as an example, in an embodiment of the present invention.

[0053] Figure 7 This is a schematic diagram of the structure of an intelligent prediction and control system for shield tunneling posture provided in an embodiment of the present invention;

[0054] Figure 8 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0055] It should be noted that:

[0056] The terms “comprising” and “having”, and any variations thereof, in the specification, claims, and accompanying drawings of this invention are intended to cover a non-exclusive inclusion, such as a process, method, system, product, or apparatus that includes a series of steps or units, not necessarily limited to those explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0057] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices. The flowcharts shown in the accompanying drawings are merely illustrative and do not necessarily include all content and operations / steps, nor do they necessarily have to be performed in the described order. For example, some operations / steps can be decomposed, while others can be combined or partially combined; therefore, the actual execution order may change depending on the specific circumstances.

[0058] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. In addition, the technical features of the various embodiments or individual embodiments provided by the present invention can be arbitrarily combined to form new technical solutions. Such combinations are not bound by the order of steps and / or structural composition patterns, but must be based on the ability of those skilled in the art to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed by the present invention.

[0059] like Figure 1 As shown, a method for intelligent prediction and control of shield tunneling posture includes:

[0060] Step 1: Obtain the real-time values ​​of key decision parameters within the tunnel boring machine's decision-making cycle; the key decision parameters are obtained from multi-source heterogeneous data based on the trained GANDALF model.

[0061] Step 2: Establish segmented shield tunnel attitude control principles based on the design axis alignment of the shield tunneling.

[0062] Step 3: Input the real-time values ​​of key decision parameters within the decision cycle into the trained BO-GANDALF model. In the prediction stage, output the shield tunneling attitude prediction results. In the control stage, combine the shield tunneling attitude prediction results and the segmented shield attitude control principle to output the intelligent control decision of the shield attitude.

[0063] Step 1, the screening steps for key decision parameters include:

[0064] Based on the feature parameter contribution value matrix output by the trained BO-GANDALF model, the Top-K feature parameter types with high contribution values ​​are selected from a multi-source heterogeneous database as key decision parameters for selection.

[0065] In step 2, the principle of segmented shield tunnel attitude control is based on dividing the straight section and turning section according to the design axis alignment, including:

[0066] The preset deviation distance range is used for the shield attitude control principle in straight sections. The principle is to control the sum of the absolute values ​​of the deviations between the shield machine attitude parameters and the design axis within the preset deviation distance range of the design axis circumference. Based on the shield attitude control principle in straight sections, the shield attitude control principle in turning sections introduces an additional direction preference to promote turning tendency and controls the horizontal deviation of the shield head within the preset deviation distance range inside the design axis.

[0067] Furthermore, the principle of segmented shield tunnel attitude control is expressed in the form of an objective function, the objective function for the straight segment is as follows:

[0068]

[0069] In the formula, The objective function for the attitude parameters of the straight-line segment; These are the shield attitude parameters that need to be controlled during the straight-line section, representing the shield head horizontal deviation, shield tail horizontal deviation, shield head vertical deviation, and shield tail vertical deviation relative to the design axis, respectively. It is a constant; This indicates the critical value of the preset deviation distance range;

[0070] in, ;

[0071] The objective function for the turning segment introduces a directional preference to promote turning tendency in the objective function for the straight segment, and adds a regularization term to control the horizontal deviation of the shield head within a preset deviation distance range. The mathematical expression is as follows:

[0072]

[0073] In the formula, The objective function for the attitude parameters during the turning segment; ; and It is the penalty coefficient; It is related to the stability control weight of the tunnel boring machine; This indicates the critical value of the preset deviation distance range inside the design axis.

[0074] Specifically, firstly, during the straight-line section, the principle for controlling the shield machine's attitude is clearly defined as follows: the shield machine's attitude is controlled within a preset distance range on both sides of the design axis. Optionally, this invention controls the sum of the absolute values ​​of all attitude deviations within 30mm during the straight-line section. The objective function is as follows:

[0075]

[0076]

[0077] In the formula, The objective function for the attitude parameters of the straight-line segment; These are the attitude parameters that need to be controlled during the straight-line section, representing the shield head horizontal deviation, shield tail horizontal deviation, shield head vertical deviation, and shield tail vertical deviation relative to the design axis, respectively. It is a constant, set to 8.

[0078] The objective function is designed so that the loss is reduced to zero when the absolute value of all attitude deviations is within 30mm. Once any deviation exceeds this threshold, the excess is converted into a penalty term and incorporated into the optimization objective. This guides the optimizer toward a solution space that maintains attitude stability.

[0079] Secondly, in order to adapt to the geometric constraints of the curved path and ensure the adjustability and controllability of the tunneling posture during the turning section, a direction preference was introduced into the control target.

[0080] Optionally, taking a right-hand bend as an example, when turning right, the horizontal deviation of the shield head ( The directional steering tendency is guided into a positive range to facilitate appropriate steering. This aids in smooth navigation through high-curvature sections. To effectively achieve this control objective, an objective function was developed to quantify the deviation between the current attitude of the tunnel boring machine and the final desired state. Furthermore, control parameters related to steering tendency (here...) are... The magnitude of the variable is added to the regularization term, aiming to control the horizontal deviation of the shield head within a preset distance range inside the axis while managing energy consumption and maintaining system stability. Optionally, in the turning section, the horizontal deviation of the shield head is controlled within 30mm inside the axis. The objective function is defined as follows:

[0081]

[0082] In the formula, The objective function for the attitude parameters during the turning segment; The pose parameters predicted by the BO-GANDALF framework; and It is the penalty coefficient; These are weights related to stability control.

[0083] like Figure 3 As shown, in step 3, the training process of the BO-GANDALF model includes:

[0084] Step 3-1: Construct a training multi-source heterogeneous database based on multi-source heterogeneous data; wherein, the multi-source heterogeneous data includes shield attitude parameters, tunneling parameters, asymmetric parameters, soil and rock parameters and tunnel geometric parameters, with shield attitude parameters serving as label data;

[0085] Step 3-2: Construct the BO-GANDALF model and train it using a multi-source heterogeneous database. In the prediction phase, the BO-GANDALF model processes the multi-source heterogeneous data (excluding shield attitude parameters) through the GFLU feature extraction module, extracts optimized feature representations, and outputs a feature parameter contribution matrix. Then, the feedforward neural network outputs the predicted values ​​of the shield attitude parameters based on the optimized feature representations and the feature parameter contribution matrix. In the control phase, the BO global optimization strategy, based on the prediction phase results and the segmented shield attitude control principle, solves for the optimal solution within the adjustable range of key decision parameters and outputs the intelligent control decision of the shield attitude.

[0086] In step 3, the BO-GANDALF model is based on the PyTorch framework and is constructed using the high-performance, highly interpretable, and computationally efficient deep learning framework GANDALF. It learns the optimal representation of features by stacking gating learning units GFLU, and then learns this representation through a feedforward neural network. The non-shared weights allow each unit to learn different information from different inputs to obtain better performance. Combined with the BO global optimization strategy, it serves as a key method for hyperparameter tuning and attitude control of the model.

[0087] Specifically, in step 3-1, the training multi-source heterogeneous database includes five types of data: shield attitude parameters, tunneling parameters, asymmetric parameters, soil and rock parameters, and tunnel geometric parameters.

[0088] Among them, the shield attitude parameters reflect the current position of the shield machine and its relative relationship with the design axis. Specifically, it includes four parameters as the position parameters of the shield: horizontal displacement and vertical displacement of the shield head and tail. In addition, two parameters, pitch angle and roll angle, are added as the motion state parameters of the shield machine. Since the shield attitude parameters themselves have a time-series nature, the difference in shield attitude parameters between rings is calculated as the prediction target of the subsequent prediction model.

[0089] The tunneling parameters are mechanical, electrical, and hydraulic parameters that are automatically collected by the system during the operation of the tunnel boring machine and can reflect the mechanical operating status. These parameters include more than 140 types such as tunneling speed, total propulsion force, cutterhead speed, and cutterhead torque.

[0090] Asymmetric parameters are calculated using the regional operating parameters of the tunnel boring machine (TBM). Specifically, historical data from the TBM tunneling and guidance systems is acquired and preprocessed. Regional operating parameters that significantly influence the TBM's attitude are calculated in zones, including the stroke of the lateral / longitudinal propulsion cylinders and propulsion force. To enhance the influence of asymmetric parameters on the TBM's attitude, the main parameters are divided into four zones: upper, lower, left, and right, according to attitude control principles. The positive and negative values ​​of the asymmetric parameters are consistent with the TBM's attitude, and the longitudinal asymmetric parameter value is the lower zone parameter value minus the upper zone parameter value.

[0091] The soil and rock parameters are derived from the geological survey report. These parameters exhibit significant low-dimensional sparsity, characterized by limited parameter types and low sampling point density.

[0092] The tunnel geometry parameters are low-frequency parameters such as the curvature of the tunnel design axis and the slope.

[0093] To transform low-dimensional parameters into high-dimensional parameters at the ring-level granularity, a linear interpolation method is used to extend low-frequency data such as soil and rock parameters and tunnel design axes to higher frequencies. Using the distance of each ring's advancement as a connection point, a slice of the strata can be obtained, thus achieving the conversion of geological data from low-frequency to high-frequency. Taking soil and rock parameters as an example:

[0094]

[0095]

[0096] in, It is the average soil cohesion corresponding to each ring; It is the cohesion of the i-th soil layer; It is the thickness of the i-th soil layer on the tunnel interface slice; It is the ratio of the thickness of the i-th soil layer in the tunnel; This is the diameter of the tunnel.

[0097] Furthermore, after acquiring the multi-source heterogeneous datasets, a data fusion algorithm is used to fuse the various multi-source heterogeneous datasets to obtain a multi-source heterogeneous database, including:

[0098] Data preprocessing includes missing value imputation, removal of downtime data, data cleaning, outlier detection and removal, and data standardization.

[0099] Unified data granularity: the recording frequency of tunneling parameters is 1 minute, and the recording frequency of shield posture parameters is 1 ring / time.

[0100] Noise reduction: First, to ensure the granularity of the input feature parameters corresponds to the output position and orientation parameters, the feature parameters recorded every 1 minute are averaged in rings. Second, Outlier data is removed based on criteria and engineering experience. When all outlier values ​​are outside the range... If the data contains outliers, it should be removed. After these two noise reduction steps, a multi-source heterogeneous database was established, and the training and test sets were divided in an 8:2 ratio.

[0101] In step 3-2, such as Figure 2 As shown, in the backbone network of the BO-GANDALF model, the feature extraction GFLU is composed of several gated learning units (GFLUs) stacked together. Each gated learning unit introduces a learnable parameter vector and uses the t-softmax function to convert the parameter vector into a feature mask. The feature mask is added to the input features of the layer through element-wise multiplication to obtain the feature representation extracted by the layer. Accordingly, the extracted feature representation of the last gated learning unit is output as the optimized feature representation, and the feature mask of the last gated learning unit is used as the feature parameter contribution matrix.

[0102] Specifically, the gated learning unit GFLU is a core component of the BO-GANDLF model. Its design is inspired by the gated recurrent unit (GRU) and is specifically optimized for static, non-temporal data. A key feature of GFLU is the introduction of learnable feature masks (…). To dynamically filter important features from the data, each gated learning unit GFLU is treated as a GFLU stage, and each GFLU stage has a learnable parameter vector F. The t-softmax function is used to convert F into a sparse feature mask. .

[0103] The t-softmax function, compared to the traditional softmax function, introduces an additional temperature parameter t, thereby controlling the sparsity of the output. Specifically, given a parameter vector F = (F1, F2, ..., F... n The formula for calculating the t-softmax function is as follows:

[0104]

[0105] In the formula, This is a sparse feature mask with values ​​between 0 and 1; , Let be the i-th and j-th elements in the parameter vector F, respectively; and t be the temperature parameter used to adjust the entropy of the softmax function output. When t takes a smaller value, the amplification effect of the exponential function is enhanced, causing the feature weights corresponding to the maximum value to increase significantly, while the weights of other smaller features approach zero, thus achieving sparsity selection. This characteristic makes t-softmax particularly suitable for feature selection tasks, highlighting a few key features while reducing the interference of redundant information and improving the model's learning efficiency.

[0106] Furthermore, sparse feature masks ( The feature values ​​are added using element-wise multiplication. That is, the model learns a weight value between 0 and 1 for each feature parameter input to the model; these weights together form a mask vector. This mask vector is then multiplied element-wise with the original input feature vector. When the mask value corresponding to a feature is close to 1, the information of that feature is almost completely preserved; when the mask value is close to 0, the information of that feature is significantly suppressed or "turned off." In this way, the model can dynamically and selectively strengthen important features and weaken irrelevant or noisy features at each layer, thereby achieving automated feature selection. Mathematically, this can be represented as:

[0107]

[0108] In the formula, denoted as the filtered features, and x represents the original input features.

[0109] Specifically, GFLU includes a reset gate (r n ) and update gate (z n ), r n This determines how much information is forgotten from the previous hidden state, and z n This determines the weight of the current candidate feature representation in the final representation. The specific calculation method is as follows:

[0110] Reset Door:

[0111]

[0112] In the formula, For the reset door in step n, The sigmoid activation function is typically set between (0,1). To reset the gate's bias vector, This is the hidden state at step n-1. To reset the bias term of the gate.

[0113] Update Gate:

[0114]

[0115] In the formula, For the update gate in step n, Update the gate weight matrix. To update the bias vector of the gate.

[0116] Hidden layer:

[0117]

[0118]

[0119] In the formula, This represents the candidate hidden state at step n. Let be the weight matrix of the candidate hidden states. Let be the bias vector of the candidate hidden state. The Hadamard product (element-wise multiplication) represents the element-wise multiplication of two vectors.

[0120]

[0121] Furthermore, each GFLU layer has a different weight matrix, allowing different layers to focus on different subsets of features during the learning process. This hierarchical structure enables the feature representation to be progressively optimized, resulting in an optimized feature representation and thus improving model performance. GFLU layers are stacked sequentially to form a powerful feature representation learning network.

[0122] Furthermore, such as Figure 2 As shown in step 3-2, in the backbone network of the BO-GANDALF model, after the GFLU feature extraction module, the optimized feature representation is passed to a standard feedforward neural network. The feedforward neural network serves as the final prediction layer to further process the feature representation.

[0123] Specifically, in the GFLU feature extraction module, the optimized feature representation (H) obtained after multiple GFLU stages is input into a standard K-layer feedforward neural network. The feedforward neural network maps the optimized feature representation obtained from the GFLU module to the final output dimension. The feedforward neural network is used to transform the feature representation processed by the feedforward neural network into the final prediction result. The formula for the feedforward neural network is:

[0124]

[0125] in, This is the model's predicted output; This represents a feedforward neural network; H represents the input features; W represents the input features. i and b iThese are the weight matrix and bias vector of the i-th layer of the neural network, respectively, where k is the number of layers in the neural network.

[0126] For regression problems, a bias term needs to be added, and the formula for the feedforward neural network is then:

[0127]

[0128] In the formula, T0 is the bias term in the regression task.

[0129] Specifically, in step 3-2, during the training process of the BO-GANDALF model, the mean squared error (MSE) and the coefficient of determination (R²) were introduced to quantitatively evaluate the model performance. 2 The model employs a method for calculating the difference between predicted and observed values, using MSE as the loss function. MSE is a commonly used metric to measure the difference between predicted and observed values. An MSE close to 0 indicates a very small relative difference between the predicted and observed values, meaning the model's predictions are very close to the actual values. Mathematically, this is expressed as:

[0130]

[0131] In the formula, n is the total number of samples. It is the first The true value of each sample It is the first The predicted value for each sample.

[0132] R² (coefficient of determination) is used to measure the goodness of fit of a regression model, representing the proportion of variance of the target variable explained by the model. 2 A value close to 1 indicates that the model can explain most of the variability in the data, specifically its ability to effectively capture changes and complexity within the data, thus performing well when dealing with complex data patterns. Mathematically, this can be represented as:

[0133]

[0134] In the formula, It is the mean of the values. ; It is the sum of squared residuals, which reflects the variability that the model failed to explain; It is the total sum of squares, which reflects the overall variability of the data.

[0135] The model's optimizer uses Adam.

[0136] Preferably, the training process of the BO-GANDALF model is based on BO maximization for hyperparameter tuning:

[0137] After each round of training, the objective function of the surrogate model is approximated by a Gaussian function (to minimize the deviation between the predicted value and the true value).

[0138] Select a sampling function, quantify and evaluate the performance of the surrogate model based on the prediction results of the Gaussian function, and use the sampling function to determine the evaluation point for the next round of training of the surrogate model;

[0139] Iterative optimization involves updating the proxy model based on the evaluation results after each quantitative evaluation and using a collection function to determine the next evaluation point until the stopping criterion is met.

[0140] Specifically, a Gaussian function is used to approximate the objective function of the surrogate model, representing the output distribution of the objective function across different input values ​​in the validation set. By modeling historical data, it can provide predictions for unknown points and estimate the uncertainty of those predictions.

[0141] like Figure 3 The diagram shows a flowchart of the BO-GANDALF model in an embodiment of the present invention. In step 3-2, the process of solving for the optimal solution within the adjustable range of key decision parameters in the control stage of the BO-GANDALF model to obtain the intelligent control decision of the shield tunnel attitude includes:

[0142] Based on the adjustable range of key decision parameters, a control parameter space is constructed. In each iteration, several control parameter combinations are sampled from the control parameter space as evaluation points. A Gaussian process is used to model the mapping relationship between the shield tunneling posture and the control parameter combination in the BO-GANDALF model to obtain a surrogate model. The surrogate model is used to output the shield tunneling posture prediction results for each evaluation point.

[0143] The segmented shield tunnel attitude control principle (straight segment, turning segment) is used as the objective function of the surrogate model. The expected improvement of the objective function at each candidate point is calculated, and the evaluation point for the next round is selected by the acquisition function.

[0144] The process is iteratively optimized until the stopping criterion is met. The output is the evaluation point that makes the expected value of the objective function optimal. The combination of control parameters represented by this point is taken as the optimal solution within the adjustable range of the key decision parameters, thus obtaining the intelligent control decision of the shield attitude.

[0145] The acquisition function is used to determine the next evaluation point of the current BO-GANDALF model based on the prediction results of the surrogate function.

[0146] As a specific implementation method, the acquisition function in this embodiment of the invention adopts Expected Improvement (EI), which measures the expected improvement of the current predicted optimal solution compared to the historical optimal solution. The mathematical expression of EI is as follows:

[0147]

[0148] In the formula, The value to be improved is represented by x, where x is the evaluation point and represents the combination of control parameters; x best To reach the optimal evaluation point in the historical evaluation; Let be the expected value of the objective function to be optimized. It is the expected value of the current optimal objective function. This represents the expected value.

[0149] In each iteration, several evaluation points are selected. First, based on historical evaluation data, a Gaussian process is used to model the mapping relationship between the tunnel boring machine's attitude and control parameters, serving as a surrogate model. This surrogate model can output the predicted mean and uncertainty (variance) of the tunnel boring machine's attitude for any combination of control parameters.

[0150] Secondly, combining the principle of segmented shield attitude control, the objective function of the surrogate model is defined, and the expected improvement of the objective function at each evaluation point is calculated using the prediction distribution of the Gaussian process.

[0151] The optimization potential of each candidate control parameter combination is quantified by the acquisition function (such as the desired improvement in EI), and the point that maximizes the acquisition function is selected as the evaluation point for the next round.

[0152] Repeat the above process to continuously guide the sampling direction until the preset stopping criteria (such as the maximum number of iterations, convergence threshold, etc.) are met. Finally, the optimal combination of control parameters that performs best in the objective function throughout the entire iteration process is output as the intelligent control decision solution of the key decision parameters within the adjustable range, thereby realizing the optimized control of the shield attitude.

[0153] Preferably, to further evaluate the effectiveness of the optimized shield attitude, the mean absolute deviation (MAD) is used to measure the average variation between adjacent data points in the sequence. This index reflects the local continuity and smoothness of the data: the smaller the value, the smoother the curve. Therefore, MAD can directly represent the fluctuation of attitude parameters and the adjustment frequency of key parameters.

[0154]

[0155] In the formula, n is the total number of data points. Let be the value of the i-th data point.

[0156] The coefficient of variation (CV) measures how dispersed data is relative to its average level. A smaller CV value indicates lower dispersion, better stability, and reflects the consistency of key parameters.

[0157]

[0158] in Standard deviation, This is the average value.

[0159] This invention takes the construction process record of Guangzhou Metro Line 18 as the research object. Table 1 shows the control range of key decision parameters in the measured data of Guangzhou Metro Line 18.

[0160] Table 1. Control Range of Key Decision Parameters Based on Actual Measured Data of Guangzhou Metro Line 18

[0161] Prediction Performance Evaluation: In terms of shield tunnel attitude prediction, the BO-GANDALF model used in this invention exhibits higher prediction accuracy compared to traditional neural network prediction models, with an average accuracy exceeding 97%. This is because the feature extraction method employed in this invention can more deeply mine data features, select key decision parameters that contribute more to the prediction of shield tunnel attitude parameters, and demonstrate superior ability to model complex nonlinear relationships between shield tunnel attitude parameters. Compared to traditional monitoring methods (manual total station measurement, average error ±2 mm), the prediction accuracy and real-time performance of this model provide a more reliable technical basis for dynamic attitude control of tunnels with small curvature. The prediction performance of this model has been verified using measured data from Guangzhou Metro Line 18. Figure 4 The model's prediction results for six shield tunneling attitude parameters are presented, and compared with actual field data. For example... Figure 4 As shown, the six shield attitude parameters ( Figure 4 (a) HDSH, Figure 4 (b) HDST, Figure 4 (c) VDSH, Figure 4 (d) VDST, Figure 4 (e) Pitch Figure 4 (f) The predicted curve of Roll closely matches the trend of the actual field data, with most data points overlapping or having minimal offset. For example, in HDSH prediction, the model's output maintains a consistent deviation range of ±2.5 mm from the actual values ​​of most tunnel rings, with only a slight deviation observed in ring 209. This consistency is also reflected in the error distribution histogram, where the residuals of all six parameters are concentrated within ±3 mm, and over 85% of the errors fall within ±1.5 mm, indicating high prediction accuracy of the model. As mentioned earlier, the maximum prediction deviation of -4 mm at Roll is an outlier. Figure 4 According to on-site construction records, this deviation occurred during segment replacement operations, when temporary adjustments to the tunnel boring machine (TBM) propulsion system introduced sudden parameter fluctuations, which the model trained on stable operating data failed to fully capture. Similarly, the low R-value of ring 195... 2While related to instability, a MAE of only 0.052 mm indicates that even under these conditions, the absolute value of the model's prediction error remains negligible, highlighting its robustness to short-term data anomalies. Among all six parameters, the BO-GANDALF model has a mean MAE of 0.728 mm and a mean R0... 2 The R value is 0.886 (for HDSH, HDST, and VDST). 2 The R² value (above 0.9) indicates its superior ability to simulate the complex nonlinear relationship between multi-source parameters and masked pose. Compared to three other models commonly used for pose prediction tasks, the BO-GANDALF model demonstrates better performance in both prediction accuracy and stability. Other models exhibit specific performance limitations: MLP has lower prediction accuracy for pose offsets, CNN lacks stability across metrics, and CART has lower overall accuracy, making it difficult for them to achieve both accuracy and stability across all evaluation metrics. In contrast, GANDALF demonstrates superior prediction ability and generalization performance across all parameters—consistently maintaining the highest R² value. 2 (HDSH = 0.98, HDST = 0.949), the lowest RMSE (e.g., HDSH = 1.1, HDST = 0.963), and the smallest MAE (e.g., HDSH = 0.88, HDST = 1.084). These results strongly support the effectiveness and advantages of BO-GANDALF in predicting the attitude of shield tunnels with small curvature, and verify its ability to handle multi-source heterogeneous data and capture complex nonlinear relationships in attitude control. Compared with traditional monitoring methods, the prediction accuracy and real-time performance of this model provide a more reliable technical basis for dynamic attitude control of tunnels with small curvature. These results collectively confirm that the BO-GANDALF model has satisfactory predictive performance for shield position and attitude, laying a solid foundation for subsequent attitude optimization strategies.

[0162] Control Effect Evaluation: The method of this invention, through accurate prediction and reasonable control strategies, can better maintain the tunnel boring machine's attitude within the normal range, effectively avoiding safety accidents and improving tunneling efficiency, demonstrating significant advantages in control effect. In practical applications, the optimized attitude curve not only controls the machine within the 30mm threshold but also reduces the parameter adjustment frequency by 35%, lowering the workload of the tunnel boring machine operator. The optimized control effect of this model has been verified in the measured data of Guangzhou Metro Line 18, such as... Figure 5As shown, the colored curves represent the attitude parameters before optimization, and the red curve represents the results after optimization. Specifically, the colored curves before optimization fluctuated frequently: the deviation of HDSH reached ±45 mm, and the maximum deviation of HDST reached ±60 mm, far exceeding the safety threshold of 30 mm. After optimization, the red curves of both parameters stabilized within ±25 mm, and the MAD index decreased, directly reflecting the suppression of horizontal attitude instability. The red curves of VDSH and VDST after optimization maintained a smooth trend, with the deviation consistently controlled within ±20 mm. The significantly reduced MAD after optimization indicates a smoother shield attitude, effectively avoiding vertical "snake-like" behavior. The results show that after optimizing the excavation parameters, the shield attitude fluctuation was effectively suppressed. Most attitude parameters remained within a deviation range of ±30 mm, and the MAD values ​​of the attitude parameters decreased, indicating a good control effect. Figure 6 The image shows the optimization results of the BO-GANDALF master control parameters. The results indicate that the driver tends to adopt a relatively cautious and conservative approach when adjusting the cylinder stroke difference. When the posture deviation is large, the driver does not always correct it in time. In contrast, the BO-GANDALF framework provides optimized values, tending to adjust more quickly to effectively correct the shield posture. After optimization, the CV of HCSD shows the largest decrease, indicating better control of the hydraulic cylinder, and the parameter range demonstrates the stability of the posture optimization control. Regarding total thrust and propulsion speed, the average values ​​before and after optimization are basically the same, but the decrease in MAD indicates that the optimization has a better smoothing effect on these parameters. Although the MAD of HCSD increases slightly, implying more frequent adjustments, from... Figure 6 As clearly seen in (a), the attitude correction efficiency has been improved, which helps to improve the overall steering performance. This is particularly beneficial for practical engineering because it can reduce the inefficiency caused by frequent adjustments to decision parameters, thereby improving attitude adjustment efficiency. In summary, guided by the inverse optimization of key parameters within the BO-GANDALF framework, the optimization of excavation parameters significantly improves the control of the tunnel boring machine's attitude. These results verify the feasibility of achieving precise attitude control through parameter optimization and also confirm that the BO-GANDALF framework can provide practical and operable guidance for tunnel boring machine engineering.

[0163] The implementation of the various embodiments of the present invention is based on programmed processing through a system with processor functionality. Therefore, in practical engineering, the technical solutions and functions of the various embodiments of the present invention are encapsulated into various modules. Based on this reality, and building upon the above embodiments, the embodiments of the present invention provide an intelligent prediction and control system for shield tunneling posture, which is used to execute an intelligent prediction and control method for shield tunneling posture from the above method embodiments.

[0164] See Figure 7The system includes:

[0165] The input module is used to obtain the real-time values ​​of key decision parameters within the decision-making cycle of the tunnel boring machine; the key decision parameters are obtained from multi-source heterogeneous data based on the trained GANDALF model.

[0166] The control principle construction module is used to establish segmented shield attitude control principles based on the design axis alignment of shield tunneling.

[0167] The shield tunneling attitude prediction and control module is used to input the real-time values ​​of key decision parameters within the decision cycle into the trained BO-GANDALF model. In the prediction stage, it outputs the shield tunneling attitude prediction results. In the control stage, it combines the shield tunneling attitude prediction results and the segmented shield attitude control principle to output the intelligent control decision of the shield attitude.

[0168] The training process of the BO-GANDALF model includes:

[0169] Construct a multi-source heterogeneous database for training based on multi-source heterogeneous data;

[0170] A BO-GANDALF model is constructed and trained using a multi-source heterogeneous database. In the prediction phase, the BO-GANDALF model processes the multi-source heterogeneous data (excluding shield attitude parameters) through a GFLU feature extraction module, extracts optimized feature representations, and outputs a feature parameter contribution matrix. A feedforward neural network then outputs predicted values ​​of the shield attitude parameters based on the optimized feature representations and the feature parameter contribution matrix. In the control phase, the BO global optimization strategy, based on the prediction phase results and the segmented shield attitude control principle, solves for the optimal solution within the adjustable range of key decision parameters, outputting intelligent control decisions for the shield attitude.

[0171] It should be noted that the system embodiments provided by the present invention are used not only to implement the methods in the above method embodiments, but also to implement the methods in other method embodiments provided by the present invention. The only difference is that corresponding functional modules are set. The principle is basically the same as that of the above system embodiments provided by the present invention. As long as those skilled in the art can improve the modules in the above system embodiments by referring to the specific technical solutions in other method embodiments and combining technical features to obtain corresponding technical means and technical solutions composed of these technical means, on the basis of the above system embodiments, and on the premise of ensuring the practicality of the technical solutions, they can obtain corresponding system-like embodiments for implementing the methods in other method-like embodiments.

[0172] The method in this embodiment of the invention is implemented using an electronic device; therefore, it is necessary to introduce the relevant electronic device. For this purpose, embodiments of the present invention provide an electronic device, such as... Figure 8 As shown, the electronic device includes: at least one processor, a communication interface, at least one memory, and a communication bus, wherein the at least one processor, the communication interface, and the at least one memory communicate with each other via the communication bus. The at least one processor invokes logical instructions stored in the at least one memory to execute all or part of the steps of the methods provided in the foregoing method embodiments.

[0173] Furthermore, when the logical instructions in at least one of the aforementioned memories are implemented as software functional units and sold or used as independent products, they are stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, is embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (a personal computer, server, or network device) to execute all or part of the steps of the methods described in the various method embodiments of the present invention. The aforementioned storage medium includes: USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks—various media for storing program code.

[0174] The system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, located in one place, or distributed across multiple network units. The purpose of this embodiment is achieved by selecting some or all of the modules according to actual needs. Those skilled in the art will understand and implement this without any inventive effort.

[0175] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0176] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0177] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0178] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0179] Based on the same technical concept as the foregoing embodiments, the present invention provides a non-transitory computer-readable storage medium that stores computer instructions that cause the computer to execute a method for intelligent prediction and control of tunnel boring machine attitude.

[0180] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the technical solutions of the embodiments of the present invention.

Claims

1. A method for intelligent prediction and control of shield tunneling posture, characterized in that, include: Obtain real-time values ​​of key decision parameters within the tunnel boring machine's decision-making cycle; The key decision parameters are obtained from multi-source heterogeneous data based on the trained GANDALF model; The principle of segmented shield attitude control is established based on the design axis alignment of shield tunneling. The real-time values ​​of key decision parameters within the decision cycle are input into the trained BO-GANDALF model. In the prediction stage, the shield tunneling attitude prediction results are output. In the control stage, the shield tunneling attitude prediction results and the segmented shield attitude control principle are combined to output the intelligent control decision of the shield attitude. The training process of the BO-GANDALF model includes: Construct a multi-source heterogeneous database for training based on multi-source heterogeneous data; A BO-GANDALF model is constructed and trained using a multi-source heterogeneous database. In the prediction phase, the BO-GANDALF model processes the multi-source heterogeneous data (excluding shield attitude parameters) through a GFLU feature extraction module, extracts optimized feature representations, and outputs a feature parameter contribution matrix. A feedforward neural network then outputs predicted values ​​of the shield attitude parameters based on the optimized feature representations and the feature parameter contribution matrix. In the control phase, the BO global optimization strategy, based on the prediction phase results and the segmented shield attitude control principle, solves for the optimal solutions within the adjustable range of key decision parameters, outputting intelligent control decisions for the shield attitude. The segmented shield tunneling attitude control principle is based on dividing the design axis alignment into straight sections and turning sections, including: The principle for controlling the shield attitude in straight sections is to control the sum of the absolute values ​​of the deviations between the shield machine attitude parameters and the design axis within a preset deviation distance range on both sides of the design axis. Based on the principle for controlling the shield attitude in straight sections, the principle for controlling the shield attitude in turning sections additionally introduces directional preference to promote turning tendency and controls the horizontal deviation of the shield head within a preset deviation distance range inside the design axis. The principle of segmented shield tunnel attitude control is expressed in the form of an objective function, and the objective function for the straight segment is as follows: ; In the formula, The objective function for the attitude parameters of the straight-line segment; These are the shield attitude parameters that need to be controlled during the straight-line section, representing the shield head horizontal deviation, shield tail horizontal deviation, shield head vertical deviation, and shield tail vertical deviation relative to the design axis, respectively. It is a constant; This indicates the critical value of the preset deviation distance range; The objective function for the turning segment introduces a direction preference to promote steering tendency in the objective function for the straight segment, and adds a regularization term to control the horizontal deviation of the shield head within a preset deviation distance range. The mathematical expression is as follows: ; In the formula, The objective function for the attitude parameters during the turning segment; ; and It is the penalty coefficient; It is related to the stability control weight of the tunnel boring machine; This indicates the critical value of the preset deviation distance range inside the design axis.

2. The intelligent prediction and control method for shield tunneling posture as described in claim 1, characterized in that, The screening steps for the key decision parameters include: Based on the feature parameter contribution value matrix output by the trained BO-GANDALF model, the Top-K feature parameter types with high contribution values ​​are selected from a multi-source heterogeneous database as key decision parameters for selection.

3. The intelligent prediction and control method for shield tunneling posture as described in claim 1, characterized in that, The GFLU feature extraction consists of several gated learning units stacked together. Each gated learning unit introduces a learnable parameter vector and uses the t-softmax function to convert the parameter vector into a feature mask. The feature mask is added to the input features of the layer through element-wise multiplication to obtain the feature representation extracted by the layer. Accordingly, the feature representation extracted by the last gated learning unit is output as the optimized feature representation, and the feature mask of the last gated learning unit is used as the feature parameter contribution matrix.

4. The intelligent prediction and control method for shield tunneling posture as described in claim 1, characterized in that, The process of obtaining the intelligent control decision-making for the shield tunneling attitude by using the trained BO-GANDALF model to solve for the optimal solution within the adjustable range of key decision parameters includes: Based on the adjustable range of key decision parameters, a control parameter space is constructed. In each iteration, several control parameter combinations are sampled from the control parameter space as evaluation points. A Gaussian process is used to model the mapping relationship between the shield tunneling posture and the control parameter combination in the BO-GANDALF model to obtain a surrogate model. The surrogate model is used to output the shield tunneling posture prediction results for each evaluation point. The segmented shield tunnel attitude control principle is used as the objective function of the surrogate model. The expected improvement of the objective function at each candidate point is calculated, and the evaluation point for the next round is selected by the acquisition function. The process is iteratively optimized until the stopping criterion is met. The output is the evaluation point that makes the expected value of the objective function optimal. The combination of control parameters represented by this point is taken as the optimal solution within the adjustable range of the key decision parameters, thus obtaining the intelligent control decision of the shield attitude.

5. The intelligent prediction and control method for shield tunneling posture as described in claim 4, characterized in that, The acquisition function is improved based on expectations, and its mathematical expression is as follows: ; In the formula, The x represents the expected improvement value, where x is the evaluation point and represents the combination of control parameters to be evaluated; x best To reach the optimal evaluation point in the historical evaluation; Let be the expected value of the objective function to be optimized. It is the expected value of the current optimal objective function. This represents the expected value.

6. A shield tunneling posture intelligent prediction and control system, used to execute the shield tunneling posture intelligent prediction and control method according to any one of claims 1 to 5, characterized in that, include: The input module is used to obtain the real-time values ​​of key decision parameters during the decision-making cycle of the tunnel boring machine; The key decision parameters are obtained from multi-source heterogeneous data based on the trained GANDALF model; The control principle construction module is used to establish segmented shield attitude control principles based on the design axis alignment of shield tunneling. The shield tunneling attitude prediction and control module is used to input the real-time values ​​of key decision parameters within the decision cycle into the trained BO-GANDALF model. In the prediction stage, it outputs the shield tunneling attitude prediction results. In the control stage, it combines the shield tunneling attitude prediction results and the segmented shield attitude control principle to output the intelligent control decision of the shield attitude. The training process of the BO-GANDALF model includes: Construct a multi-source heterogeneous database for training based on multi-source heterogeneous data; A BO-GANDALF model is constructed and trained using a multi-source heterogeneous database. In the prediction phase, the BO-GANDALF model processes the multi-source heterogeneous data (excluding shield attitude parameters) through a GFLU feature extraction module, extracts optimized feature representations, and outputs a feature parameter contribution matrix. A feedforward neural network then outputs predicted values ​​of the shield attitude parameters based on the optimized feature representations and the feature parameter contribution matrix. In the control phase, the BO global optimization strategy, based on the prediction phase results and the segmented shield attitude control principle, solves for the optimal solution within the adjustable range of key decision parameters, outputting intelligent control decisions for the shield attitude.

7. An electronic device, characterized in that, The method includes a memory and a processor, the memory storing program instructions that are executed by the processor, the processor invoking the program instructions to perform the method according to any one of claims 1 to 5.

8. A non-transitory computer-readable storage medium, characterized in that, The non-transitory computer-readable storage medium stores computer instructions that cause the computer to perform the method described in any one of claims 1 to 5.

Images