A shield machine posture multi-target control system and method based on a space-time transformer

Abstract

The application discloses a shield machine posture multi-target control system and method based on a space-time Transformer, and belongs to the technical field of equipment control. The system comprises the following steps: acquiring real-time posture data and environment data of a shield machine, obtaining current multi-source monitoring data, constructing a dynamic virtual geological model, dividing a plurality of pre-control intervals in combination with historical posture data, identifying control contradictions to generate an interval contradiction set, generating a strategy sequence for each pre-control interval, coordinating and processing the strategy sequence by using a dynamic control chart, generating a coordinated control instruction, adjusting parameters of a PID controller to control the action of a shield machine actuator, completing posture control, calculating the deviation between current posture data and a target posture trajectory to obtain deviation data, and updating historical posture data. Through the combination of geological feedforward perception, multi-target contradiction analysis and collaborative control, the foreseeability, accuracy and self-adaptive ability to complex geology of shield machine posture control can be improved.

Description

Technical Field

[0001] This invention relates to the field of equipment control technology, and in particular to a multi-objective control system and method for the attitude of a tunnel boring machine based on spatiotemporal Transformer. Background Technology

[0002] As a large-scale tunnel boring machine, the core function of a tunnel boring machine (TBM) is to excavate tunnels underground according to a predetermined design axis and simultaneously complete the assembly and support of tunnel segments. During the excavation process, precisely controlling the TBM's attitude—its spatial position, pitch angle, yaw angle, and roll angle—is crucial to ensuring the quality, safety, and efficiency of tunnel construction. TBM attitude control is a complex, multivariate, strongly coupled, nonlinear, and time-varying control process, influenced by a combination of factors including geological conditions, equipment status, and operational behavior.

[0003] In existing technologies, the attitude control of tunnel boring machines (TBMs) mainly relies on experienced operators for manual or semi-automatic operation, supplemented by automatic control systems based on proportional-integral-derivative (PID) control algorithms. These systems typically generate corrective torque by adjusting the pressure or stroke of the propulsion cylinders based on real-time attitude deviations measured by sensors such as laser guidance systems or gyroscopes, thereby causing the TBM to return to its design axis. This control mode is essentially a passive control based on deviation feedback, meaning that corrective measures are only taken after an attitude deviation is detected.

[0004] However, the aforementioned existing technical solutions have significant limitations. First, the control method based on deviation feedback has an inherent lag. When faced with sudden changes in geological conditions, such as encountering soft upper layers and hard lower layers, or isolated boulders, the system often responds slowly, easily leading to large attitude deviations or even over-excavation or under-excavation. Second, traditional PID controller parameters are usually fixed values ​​or can only be adjusted to a limited extent, making it difficult to adapt to drastic changes in geological parameters during tunneling and lacking adaptability in complex and variable strata. Furthermore, multiple objectives of attitude control often conflict, and existing methods struggle to intelligently balance and coordinate these contradictory objectives under different working conditions, frequently resulting in oscillations or unintended consequences in the control process. Summary of the Invention

[0005] To address the aforementioned issues, this invention provides a multi-objective control system and method for tunnel boring machine (TBM) attitude based on spatiotemporal Transformer. By combining geological feedforward sensing, multi-objective conflict analysis, and collaborative control, the predictability, accuracy, and adaptability to complex geological conditions of the TBM attitude control can be improved.

[0006] The above objectives can be achieved through the following approach: A multi-objective attitude control method for a tunnel boring machine (TBM) based on a spatiotemporal Transformer includes: acquiring real-time attitude data and environmental data of the TBM to obtain current multi-source monitoring data; constructing a dynamic virtual geological model for predicting the geological conditions ahead based on the current multi-source monitoring data; acquiring historical attitude data; dividing the TBM into multiple pre-control intervals according to the dynamic virtual geological model and the historical attitude data, identifying control contradictions within each pre-control interval, and generating an interval contradiction set; generating a strategy sequence containing a target attitude trajectory and control weight strategy for each pre-control interval for the interval contradiction set; coordinating the strategy sequence using a preset dynamic control diagram for coordinating different control strategies to generate a coordinated control command; adjusting the parameters of the PID controller to control the actions of the TBM actuator according to the coordinated control command to complete attitude control; acquiring the current attitude data of the TBM after attitude control, calculating the deviation between the current attitude data and the target attitude trajectory to obtain deviation data, and updating the historical attitude data using the deviation data.

[0007] Optionally, obtaining real-time attitude data and environmental data of the tunnel boring machine to obtain current multi-source monitoring data includes: acquiring position, angle, and speed data through attitude sensors on the tunnel boring machine to form attitude data; acquiring earth pressure and torque data through environmental sensors on the tunnel boring machine to form environmental data; and performing time alignment and data fusion on the attitude data and the environmental data to generate current multi-source monitoring data.

[0008] Optionally, constructing a dynamic virtual geological model for predicting the geological conditions ahead based on the current multi-source monitoring data includes: filtering and normalizing the current multi-source monitoring data to obtain preprocessed data; using a preset neural network based on a self-attention mechanism to extract spatiotemporal features from the preprocessed data to generate a current spatiotemporal feature vector; inputting the current spatiotemporal feature vector into a preset geological data prediction model to output predicted data on the soil stiffness distribution and pressure trend in the unexcavated area ahead of the tunnel boring machine; and using the predicted data to construct a dynamic virtual geological model.

[0009] Optionally, the construction of the geological data prediction model includes: acquiring historical multi-source monitoring data and performing filtering and normalization processing to obtain historical processed data; using the neural network based on the self-attention mechanism to extract spatiotemporal features from the historical processed data to generate historical spatiotemporal feature vectors; acquiring historical soil stiffness distribution and historical pressure trends to obtain historical real-time data; and training a regression prediction network with the historical spatiotemporal feature vectors as input and the historical real-time data as output to obtain the geological data prediction model.

[0010] Optionally, generating the interval conflict set includes: acquiring historical attitude data and setting a gradient threshold based on the historical attitude data; calculating the spatial gradient of soil stiffness along the tunneling direction in the dynamic virtual geological model, and identifying locations where the spatial gradient exceeds the gradient threshold as geological boundary points; dividing the tunneling path into multiple pre-control intervals based on the geological boundary points; for each pre-control interval, analyzing the conflict degree between attitude control targets, identifying control conflicts, and summarizing all identified control conflicts into an interval conflict set.

[0011] Optionally, generating a strategy sequence containing a target attitude trajectory and a control weight strategy for each pre-regulation interval for the set of interval contradictions includes: determining the primary and secondary control objectives for each pre-regulation interval based on the control contradictions; calculating the target attitude trajectory using a multi-objective optimization algorithm based on the primary and secondary control objectives; assigning weights to the attitude control objectives according to the control contradictions to form a control weight strategy; and combining the target attitude trajectory with the control weight strategy to form a strategy sequence.

[0012] Optionally, generating coordinated control instructions includes: extracting attitude parameters at the boundary of adjacent pre-adjustment intervals from the policy sequence; calculating the discontinuity of the attitude parameters at the boundary using a preset arbitration rule in the dynamic control graph for resolving policy conflicts, and generating a conflict resolution solution; and adjusting the policies of adjacent pre-adjustment intervals in the policy sequence based on the conflict resolution solution to generate coordinated control instructions.

[0013] Optionally, the step of adjusting the parameters of the PID controller to control the action of the tunnel boring machine's actuator according to the coordinated control command to complete the attitude control includes: extracting a control weight strategy from the coordinated control command, adjusting the proportional, integral, and derivative parameters of the tunnel boring machine's PID controller to obtain adaptive control parameters; using the deviation between the target attitude trajectory and the actual attitude data as input, calculating the real-time control quantity using the PID controller containing the adaptive control parameters; converting the real-time control quantity into a drive signal and sending it to the tunnel boring machine's actuator to complete the attitude control.

[0014] Optionally, updating the historical attitude data using the deviation data includes: after attitude control, acquiring current attitude data of the tunnel boring machine at multiple time points to form a current attitude sequence; calculating the difference between the current attitude sequence and the target attitude trajectory to generate deviation data; and updating the historical attitude data using the deviation data.

[0015] Based on the same inventive concept, this invention also provides a multi-objective control system for tunnel boring machine (TBM) attitude based on a spatiotemporal Transformer. The system includes: a data acquisition module for acquiring real-time attitude data and environmental data of the TBM to obtain current multi-source monitoring data; a geological feedforward sensing module for constructing a dynamic virtual geological model for predicting the geological conditions ahead based on the current multi-source monitoring data; a spatiotemporal situation analysis module for acquiring historical attitude data, dividing the data into multiple pre-control intervals according to the dynamic virtual geological model and the historical attitude data, identifying control contradictions within each pre-control interval, and generating an interval contradiction set; and a strategy generation module. For the set of interval contradictions, a strategy sequence containing the target attitude trajectory and control weight strategy is generated for each pre-controlled interval; a control coordination module is used to coordinate the strategy sequence using a preset dynamic control diagram for coordinating different control strategies, and generate coordinated control instructions; an instruction execution module is used to adjust the parameters of the adaptive controller to control the action of the tunnel boring machine's actuator according to the coordinated control instructions, and complete the attitude control; a feedback update module is used to acquire the current attitude data of the tunnel boring machine after attitude control, calculate the deviation between the current attitude data and the target attitude trajectory to obtain deviation data, and use the deviation data to update the historical attitude data.

[0016] Compared with the prior art, the present invention has the following advantages: 1. This invention enables effective prediction of the geological conditions of the unexcavated area in front of the tunnel boring machine by constructing a dynamic virtual geological model. This feedforward perception capability based on spatiotemporal Transformer allows the control system to shift from traditional delayed response to proactive prediction, formulating response strategies in advance for upcoming geological changes, thereby enhancing the foresight of attitude control and avoiding the risk of attitude instability caused by sudden geological changes.

[0017] 2. This invention divides the tunneling path into multiple pre-controlled intervals and identifies the main control contradictions within each interval, achieving the decomposition and refined management of complex control tasks. Customized solutions, including target attitude trajectories and control weight strategies, are generated for specific contradictions in different intervals. This resolves the inherent conflicts between multiple objectives such as pitch, yaw, and roll in attitude control, making control decisions more targeted and improving attitude control accuracy under complex conditions such as uneven geological formations.

[0018] 3. This invention utilizes dynamic control charts to coordinate the segmented generated strategy sequences, ensuring control continuity and smoothness during transitions between different pre-control intervals. This coordination mechanism eliminates sudden command changes that may occur due to strategy switching, avoids impacts on the tunnel boring machine's actuators and overshoot and oscillations during the control process, guarantees the stability and safety of the entire tunneling process, and improves control quality.

[0019] 4. This invention establishes a complete closed-loop learning and optimization mechanism, from execution deviation feedback to perception models, control strategies, and coordination rules. By continuously calculating the deviation between the actual trajectory and the target trajectory, the system can continuously update historical data, iteratively optimize geological prediction models and high-level decision-making rules, endowing the entire control system with the ability to self-evolve, enabling it to continuously adapt to new geological environments and working conditions, and achieve continuous improvement in control performance.

[0020] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures pointed out in the description, claims and drawings. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the 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.

[0022] Figure 1 This is a flowchart illustrating a multi-objective control method for tunnel boring machine attitude based on spatiotemporal Transformer according to an embodiment of the present invention.

[0023] Figure 2 This is a schematic diagram of the strategy sequence generation and coordination process according to an embodiment of the present invention.

[0024] Figure 3 This is a schematic diagram of the structure of a multi-objective control system for tunnel boring machine attitude based on spatiotemporal Transformer according to an embodiment of the present invention. Detailed Implementation

[0025] 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.

[0026] Reference Figure 1One embodiment of the present invention proposes a multi-objective control method for the attitude of a tunnel boring machine based on spatiotemporal Transformer. By combining geological feedforward perception, multi-objective contradiction analysis and collaborative control, the predictability, accuracy and adaptability of the attitude control of the tunnel boring machine to complex geology can be improved.

[0027] The method described in this embodiment specifically includes: Acquire real-time attitude data and environmental data of the tunnel boring machine to obtain current multi-source monitoring data; Based on the current multi-source monitoring data, a dynamic virtual geological model is constructed to predict the geological conditions ahead. Historical attitude data is acquired, and based on the dynamic virtual geological model and the historical attitude data, multiple pre-control intervals are divided, and control contradictions within each pre-control interval are identified to generate an interval contradiction set. For the set of interval contradictions, a strategy sequence containing the target attitude trajectory and control weight strategy is generated for each pre-adjustment interval; Using a preset dynamic control diagram for coordinating different control strategies, the strategy sequence is coordinated to generate coordinated control instructions; According to the coordinated control command, the parameters of the PID controller are adjusted to control the action of the tunnel boring machine's actuator, thereby completing attitude control; After attitude control, the current attitude data of the tunnel boring machine is acquired, and the deviation between the current attitude data and the target attitude trajectory is calculated to obtain deviation data. The historical attitude data is then updated using the deviation data.

[0028] This invention first constructs a dynamic virtual geological model capable of sensing geological changes ahead by integrating multi-source data from real-time monitoring, enabling prediction of the future tunneling environment. Based on this prediction, and combined with historical experience data, the complex tunneling path is decomposed into several pre-control intervals with clear control challenges, and the inherent contradictions between attitude control objectives within each interval are identified. Subsequently, for the specific contradictions in each interval, a locally optimal target attitude trajectory and control weight strategy are generated. To ensure the smoothness of global control, this method uses a dynamic control chart to coordinate these segmented strategies, generating a continuous and smooth coordinated control command. This command is used to adjust the parameters of the adaptive controller in real time to execute precise attitude control actions. Finally, by comparing the deviation between the actual attitude and the target trajectory, feedback data is generated. This data is not only used to update the historical database but also enables continuous optimization of the entire control system, forming a complete self-learning and adaptive closed loop.

[0029] This invention enhances the intelligence level of tunnel boring machine (TBM) attitude control and its adaptability to complex geological conditions. By introducing a geological prediction mechanism, the method transforms passive, delayed responses into proactive, pre-emptive adjustments, mitigating the risk of attitude instability caused by sudden geological changes and enhancing the forward-looking nature of the control. Through segmented analysis and conflict identification of the tunneling path, refined management of multi-objective control conflicts is achieved, making the control strategy more targeted and effective, and improving the accuracy of attitude control. The strategy coordination mechanism ensures a smooth transition during the switching of different working conditions, avoiding abrupt changes in control commands and guaranteeing the overall stability of the tunneling process. Finally, the feedback update closed loop endows the system with the ability to self-learn and continuously optimize, enabling it to continuously accumulate experience and improve control performance, thereby achieving more reliable, efficient, and high-quality TBM tunneling attitude control in variable and uncertain environments.

[0030] Optionally, obtaining the real-time attitude data and environmental data of the tunnel boring machine to obtain the current multi-source monitoring data includes: Position, angle, and speed data are acquired through attitude sensors on the tunnel boring machine to form attitude data; Environmental data is generated by acquiring earth pressure and torque data through environmental sensors on the tunnel boring machine. The attitude data and environmental data are time-aligned and fused to generate current multi-source monitoring data.

[0031] Specifically, attitude sensors installed on the tunnel boring machine (TBM), such as laser guidance systems, gyroscopes, and inclinometers, acquire real-time data describing the TBM's spatial state. This data constitutes attitude data, primarily including the TBM's position in a three-dimensional coordinate system, its angles around each axis, and its excavation speed. Simultaneously, environmental sensors installed in the TBM's cutterhead drive system and soil chamber, such as torque and earth pressure sensors, acquire real-time data describing the interaction between the TBM and the geological environment. This data constitutes environmental data, primarily including the torque experienced by the cutterhead during rotation and the soil pressure at the excavation face. Because different sensors have varying sampling frequencies and data transmission delays, direct use can lead to timestamp discrepancies, failing to accurately reflect the TBM's overall state at any given moment. Therefore, time alignment processing is required for the acquired attitude and environmental data. This processing uses techniques such as interpolation or resampling to unify all data to the same time reference. After time alignment, data fusion is performed, integrating data from different sources into a structured dataset—the current multi-source monitoring data. This fusion process can be represented as concatenating different data vectors into a comprehensive feature vector at each time step t: , in, It represents the current multi-source monitoring data generated at time t, and it is a comprehensive state vector. This represents the position data obtained from the attitude sensor. This represents the angle data obtained from the attitude sensor. This represents the velocity data obtained from the attitude sensor. This represents earth pressure data obtained from environmental sensors. This represents torque data acquired from environmental sensors. This vector comprehensively describes the tunnel boring machine's motion and attitude at a specific moment and its interaction with the surrounding geological environment.

[0032] By temporally aligning and fusing attitude and environmental data, this method generates current multi-source monitoring data. This high-dimensional fused data provides rich and reliable input for the subsequent construction of a dynamic virtual geological model, enabling the model to more accurately capture the complex spatiotemporal correlation between attitude changes and environmental parameters. This not only improves the accuracy of predicting the geological conditions ahead but also lays a solid foundation for the generation and coordination of subsequent control strategies, thereby enhancing the predictability and adaptability of the entire attitude multi-objective control system and ultimately achieving smoother and more precise control of the tunnel boring machine's attitude.

[0033] Optionally, constructing a dynamic virtual geological model for predicting future geological conditions based on the current multi-source monitoring data includes: The current multi-source monitoring data is filtered and normalized to obtain preprocessed data; Using a pre-defined neural network based on a self-attention mechanism, spatiotemporal features are extracted from the preprocessed data to generate a current spatiotemporal feature vector; The current spatiotemporal feature vector is input into a preset geological data prediction model, and the predicted data of soil stiffness distribution and pressure trend in the unexcavated area in front of the tunnel boring machine are output. The predicted data are used to construct a dynamic virtual geological model.

[0034] 1. Signal Denoising: Different denoising algorithms are employed for data with different physical characteristics. For attitude data (position, angle, velocity), a kinematic Kalman filter algorithm is used to provide optimal state estimation while filtering out noise. For environmental data (earth pressure, torque), a wavelet thresholding denoising algorithm is used to smooth the data while retaining abrupt changes that may indicate geological changes. 2. Feature Extraction: A hybrid deep learning model is used to automatically extract the spatiotemporal features of the denoised data. This model consists of a one-dimensional convolutional neural network (1D-CNN) and a Transformer encoder connected in series. First, the 1D-CNN layer processes the time series data of each sensor to extract local temporal patterns. Then, the Transformer encoder, through its self-attention mechanism, fuses the CNN output features from all sensors and calculates global temporal dependencies, ultimately generating a high-dimensional vector that comprehensively summarizes the dynamic characteristics of the tunnel boring machine's excavation process—the current spatiotemporal feature vector.

[0035] This current spatiotemporal feature vector, rich in spatiotemporal information, is input into a pre-defined geological data prediction model. This model is a regression prediction network trained on historical data. It receives the current spatiotemporal feature vector as input and outputs quantitative predictions of geological parameters in the unexcavated area ahead of the tunnel boring machine (TBM), specifically including predicted data such as soil stiffness distribution along the tunneling path and excavation face pressure trends. Finally, these predicted data are organized and structured to construct a digital geological environment map that updates in real time as the TBM advances—a dynamic virtual geological model.

[0036] Through the steps described above, this method transforms real-time monitoring data into predictions of geological conditions ahead. Utilizing a neural network based on a self-attention mechanism, it can deeply mine the hidden spatiotemporal correlations within multi-source data, obtaining more accurate geological feedforward information than traditional methods. The constructed dynamic virtual geological model provides predictability for subsequent control decisions, enabling the control system to shift from passive response to proactive prediction, preparing in advance for impending geological changes. This enhances the intelligence and foresight of the tunnel boring machine's attitude control, providing crucial technical support for achieving smooth and efficient tunneling.

[0037] Optionally, the construction of geological data prediction models includes: Historical multi-source monitoring data is acquired and filtered and normalized to obtain historical processed data; Using the neural network based on the self-attention mechanism, spatiotemporal features are extracted from the historical processed data to generate historical spatiotemporal feature vectors; Historical soil stiffness distribution and historical pressure trends are obtained to acquire historical real-time data; Using the historical spatiotemporal feature vector as input and the historical real-time data as output, a regression prediction network is trained to obtain a geological data prediction model.

[0038] Specifically, firstly, a large amount of historical multi-source monitoring data needs to be collected. This data comes from previous tunnel boring machine (TBM) projects or completed sections of the same project, containing complete records of the TBM's operation under different geological conditions. These historical multi-source monitoring data undergo the same filtering and normalization operations as real-time processing to eliminate noise and standardize the data scale, thereby obtaining historically processed data that can be used for model training.

[0039] Subsequently, the serialized historical processing data is input into a pre-defined neural network based on a self-attention mechanism. This network, through its inherent attention weight calculation, captures and encodes key dynamic changes in the historical data sequence, thereby generating a historical spatiotemporal feature vector containing spatiotemporal information for each historical data mining process. This step transforms the raw, high-dimensional time-series data into a compact and highly representative feature representation.

[0040] Meanwhile, to construct the supervisory signal for training, it is necessary to obtain the actual geological conditions corresponding to historical multi-source monitoring data in time and space, namely, the historical soil stiffness distribution and historical pressure trends, which together constitute historical real-time data. This historical real-time data usually comes from geological survey reports in the early stages of the project, borehole sampling analysis, and actual inversion analysis of the strata after tunneling. It constitutes the learning objective of the model, which is the result we hope the model can predict.

[0041] Finally, the model training phase begins. A regression prediction network is trained using a series of generated historical spatiotemporal feature vectors as the input dataset and corresponding historical real-time data as the output label set. The training objective is to minimize the difference between the geological parameters predicted by the network and the actual historical real-time data. This difference is quantified using a loss function: , In this optimization objective Representational regression prediction network The set of trainable parameters, including weights and biases. It is a historical spatiotemporal feature vector extracted through a self-attention network. This corresponds to historical real-time data, i.e., the actual geological labels. The training process utilizes optimization algorithms such as gradient descent to repeatedly adjust the parameters. This continues until the loss function converges to its minimum value. After training, the result has the optimal parameters. Regression prediction network This is the final geological data prediction model.

[0042] Through supervised training based on historical data and real geological conditions, the geological data prediction model learns a deep mapping relationship between the tunnel boring machine's operating status and the geological conditions ahead. This gives the model powerful generalization capabilities, enabling it to accurately predict parameters of unknown geological areas based on real-time monitoring data. The construction of this model, by embedding engineering experience and geological knowledge into the neural network, provides reliable feedforward perception capabilities for the entire control system, and is a key link in achieving predictive attitude control and mitigating geological risks.

[0043] Optionally, the generation of the interval contradiction set includes: Acquire historical attitude data and set a gradient threshold based on the historical attitude data; Calculate the spatial gradient of soil stiffness along the tunneling direction in the dynamic virtual geological model, and identify the locations where the spatial gradient exceeds the gradient threshold as geological boundary points; Based on the geological boundary points, the tunneling path is divided into multiple pre-controlled intervals; For each pre-adjustment interval, the degree of conflict between attitude control targets is analyzed, control contradictions are identified, and all identified control contradictions are summarized into an interval contradiction set.

[0044] Specifically, the process first retrieves and analyzes historical attitude data, particularly data from sections where the tunnel boring machine's attitude changed significantly as it traversed different geological strata. By statistically analyzing the rate of change of attitude parameters with tunneling distance in this historical data, a reasonable gradient threshold can be set. This gradient threshold represents the degree of geological change historically significant enough to cause difficulties in attitude control.

[0045] Next, using the dynamic virtual geological model constructed in the previous step, the predicted soil stiffness distribution data along the planned tunneling path is extracted. Soil stiffness is a key geological parameter affecting the stress balance and attitude changes of the tunnel boring machine. To quantify the degree of its variation, it is necessary to calculate the spatial gradient of soil stiffness along the tunneling direction, which reflects the gradual or abrupt changes in geological conditions. The calculation can be expressed as: , in, It is the spatial gradient of soil stiffness along the excavation direction. The soil stiffness is predicted by a dynamic virtual geological model and is the tunneling distance. The function. This represents the distance along the tunnel boring machine's path. Subsequently, the spatial gradient will be calculated. The absolute value of the spatial gradient is compared with a pre-set gradient threshold. When the absolute value of the spatial gradient at a certain location exceeds the threshold, it means that there is a significant change in geological conditions at that point, which may lead to a sudden change in the stress on the tunnel boring machine. These locations are identified and marked as geological boundary points.

[0046] Based on all identified geological boundary points, the entire forward tunneling path is naturally divided into multiple continuous segments, each of which is a pre-control interval. Each pre-control interval has relatively consistent or smoothly changing geological characteristics. For each pre-control interval, it is necessary to deeply analyze the inherent contradictions in the attitude control of the tunnel boring machine (TBM). Attitude control typically involves multiple objectives, such as maintaining the correct vertical slope, keeping the accurate horizontal axis, and preventing excessive roll of the machine. When traversing heterogeneous strata, the control actions required to achieve these objectives may conflict with each other; this is the control contradiction. For example, increasing the subduction force to counteract the upward tendency caused by soft upper and hard lower strata may exacerbate the yaw caused by horizontal geological heterogeneity. By analyzing the geological characteristics within each pre-control interval and combining it with the TBM dynamics model, the conflicts between the control objectives most likely to occur under specific geological conditions are identified, and the control contradictions identified in all pre-control intervals are summarized to form the final set of interval contradictions.

[0047] This method proactively identifies geological boundaries and delineates pre-control intervals, decomposing a continuous, complex, long-distance control problem into a series of discrete sub-problems with clearly defined control challenges. Identifying and forming a set of interval contradictions allows the control system to anticipate the major difficulties and trade-offs that may be encountered at different tunneling stages. This situational analysis-based strategy ensures that subsequent control decisions are no longer blind and generic, but rather tailored to the specific contradictions of each section, improving the accuracy and effectiveness of the control strategy and providing a decision-making basis for achieving smooth and stable tunneling under complex geological conditions.

[0048] Optionally, generating a strategy sequence containing the target attitude trajectory and control weight strategy for each pre-adjustment interval for the set of interval contradictions includes: Based on the control contradictions, determine the primary and secondary control objectives for each pre-regulation interval; Based on the primary control objective and the secondary control objective, the target attitude trajectory is calculated using a multi-objective optimization algorithm; Based on the control contradiction, weights are assigned to the attitude control target to form a control weight strategy, and the target attitude trajectory is combined with the control weight strategy to form a strategy sequence.

[0049] Specifically, the control objectives are first prioritized based on the identified control conflicts within each pre-control interval. According to the primary and secondary relationships of the influence of geological conditions on the tunnel boring machine's attitude, the main and secondary control objectives within that interval are clearly defined. For example, when a typical heterogeneous stratum with a soft upper layer and a hard lower layer is predicted, the tunnel boring machine is highly prone to head-up or head-down movements. In this case, vertical attitude control, i.e., precise maintenance of the pitch angle, becomes the main control objective, while horizontal alignment may be considered a secondary control objective.

[0050] After determining the priorities of the control objectives, an optimal target attitude trajectory is calculated and generated using a multi-objective optimization algorithm. This trajectory is not a simple geometric design axis, but a dynamic path that is physically smooth and can be achieved, taking into account anticipated geological disturbances and control priorities. The optimization process aims to minimize a weighted objective function that integrates the deviations of primary and secondary control objectives from the ideal tunneling axis. For the objective function, we have: , in, It is the comprehensive cost function to be minimized. It is the target attitude trajectory to be solved, which is the tunneling distance. The function. This represents the length of the current pre-regulation range. and These represent the cost functions for the deviations of the primary and secondary control objectives relative to the design axis, respectively; for example, they could be the squares of the attitude deviations. and These are dimensionless weighting coefficients, which are set based on the analysis of control contradictions, and Much larger This is to reflect the absolute priority of the main control objectives. At its minimum, The value of is the target attitude trajectory. By solving this optimization problem, the obtained target attitude trajectory satisfies the main control objective while also taking into account secondary objectives, achieving optimal balance under specific geological constraints.

[0051] Simultaneously, based on the same control contradiction analysis, specific control weights are assigned to various attitude control targets within this pre-control interval, forming a control weight strategy. This set of weights exists in vector form, quantifying the controller's response intensity to different dimensions such as pitch, yaw, and roll deviations in subsequent real-time control stages. For example, in the interval where vertical attitude is the primary contradiction, the weight of pitch control is significantly increased, while the weights of other dimensions are correspondingly decreased. Finally, the calculated target attitude trajectory is combined with this control weight strategy to form a complete strategy sequence, used to guide the tunnel boring machine's attitude control within this pre-control interval.

[0052] By generating customized strategy sequences for each pre-control interval, this method closely integrates macro-level path planning with micro-level control focus. It no longer strives for perfect control of all objectives with equal priority under all conditions, but rather proactively makes trade-offs and compromises based on anticipated difficulties, generating the most targeted and executable control blueprint. This divide-and-conquer, focus-oriented strategy generation approach resolves the inherent conflicts in multi-objective control, avoids oscillations or neglecting certain aspects of the control system under complex operating conditions, and thus provides core decision support for achieving stable, accurate, and proactively adaptable intelligent tunneling.

[0053] Optionally, the generation of coordination control instructions includes: Extract attitude parameters at the boundary between adjacent pre-adjustment intervals from the strategy sequence; Using the arbitration rules in the preset dynamic control diagram for resolving policy conflicts, the discontinuity of the attitude parameters at the boundary is calculated, and a conflict resolution solution is generated. Based on the conflict resolution, the strategies of adjacent pre-regulation intervals in the strategy sequence are adjusted to generate coordinated control instructions.

[0054] Specifically, such as Figure 2 As shown, the process of generating coordinated control commands in this method aims to solve the problem of strategy discontinuity that may arise from segmented optimization, ensuring the smoothness of the tunnel boring machine's movements when traversing different pre-control intervals. Since the strategy sequence for each pre-control interval is generated independently based on its internal control contradictions, there may be a jump in the target attitude defined by the two strategy sequences at the boundary between adjacent intervals. To eliminate this discontinuity, it is first necessary to extract the attitude parameters at the boundary between any two adjacent pre-control intervals from the generated strategy sequences. Specifically, the target attitude parameters at the end of the strategy sequence of the previous pre-control interval and the target attitude parameters at the beginning of the strategy sequence of the next pre-control interval are extracted. Subsequently, the discontinuity of these two attitude parameters at the boundary is calculated. This discontinuity quantifies the magnitude and direction of the instantaneous attitude change required during strategy switching, representing a potential control conflict. This discontinuity can be represented as a difference vector.

[0055] , in, It is the vector of discontinuities in attitude parameters at the boundary. This represents the target attitude parameter vector at the end of the i-th pre-adjustment interval, which is extracted from the i-th policy sequence. The target attitude parameter vector representing the starting point of the (i+1)th pre-adjustment interval is extracted from the (i+1)th strategy sequence. Both vectors contain multi-dimensional information describing the tunnel boring machine's attitude, such as position and angle.

[0056] Next, a pre-defined dynamic control chart is used to resolve this conflict. This dynamic control chart embeds a series of arbitration rules for resolving strategy conflicts. These rules are based on an expert knowledge base established using tunnel boring machine dynamics constraints and engineering experience, and can be adjusted according to discontinuity vectors. The system automatically generates an optimal conflict resolution solution based on the magnitude and direction of the angle. For example, arbitration rules can stipulate that when there is a large discontinuity in the attitude angle, a smooth transition trajectory should be generated within a certain distance before and after the boundary, such as by using cubic spline interpolation or polynomial curves to connect the two discontinuities, thereby generating specific transition path parameters as a conflict resolution solution.

[0057] Finally, based on this conflict resolution, the original policy sequence is adjusted. Specifically, the generated transition path parameters are used to correct the target attitude trajectory near the boundary of adjacent pre-control intervals, replacing the original jump points with a smooth connecting curve. The adjusted policy sequences are then recombined to form a globally continuous and smooth final control command, i.e., the coordinated control command.

[0058] Through this coordination process, this method integrates a series of segmented, locally optimal control strategies into a globally coordinated, physically feasible, continuous control sequence. It effectively avoids equipment shocks and attitude control overshoot or oscillations caused by sudden command changes at strategy switching points, ensuring the smoothness and safety of the tunnel boring machine's attitude transition under complex and variable geological conditions, and improving the overall control quality and stability of the tunneling process.

[0059] Optionally, adjusting the parameters of the PID controller to control the action of the tunnel boring machine's actuator according to the coordinated control command to complete the attitude control includes: The control weight strategy is extracted from the coordinated control command, and the proportional, integral, and derivative parameters of the shield machine's PID controller are adjusted to obtain adaptive control parameters. The deviation between the target attitude trajectory and the actual attitude data is used as input, and a PID controller containing the adaptive control parameters is used to calculate the real-time control quantity. The real-time control quantity is converted into a drive signal and sent to the actuator of the tunnel boring machine to complete the attitude control.

[0060] Specifically, in this method, adjusting the parameters of the PID controller according to the coordinated control commands to control the actions of the tunnel boring machine's actuators is the final execution step for achieving precise attitude control. First, the control weight strategy corresponding to the current tunneling position is extracted in real time from the globally coordinated control commands. This strategy, in the form of a weight vector, clarifies the emphasis on different attitude control dimensions such as pitch and yaw under the current working conditions. These weights are used to dynamically adjust the proportional, integral, and derivative parameters of the core of the tunnel boring machine's attitude control—the PID controller—making it an adaptive controller capable of real-time changes, thereby obtaining adaptive control parameters.

[0061] Subsequently, the target attitude trajectory contained in the coordinated control command is compared with the actual attitude data acquired in real time by sensors, and the deviation between the two in each control dimension is calculated. This deviation serves as the input to the adaptive PID controller. The controller uses a PID algorithm incorporating adaptive control parameters to calculate the real-time control quantity that needs to be applied. The calculation process can be expressed as follows: , in, It is the real-time control quantity calculated at time t, representing the correction force that needs to be applied. It is the attitude deviation at the current moment, obtained by subtracting the target attitude trajectory from the actual attitude data. , and These are based on the control weight strategy Dynamically adjusted proportional, integral, and derivative adaptive control parameters. It is extracted from the coordinated control commands, enabling the controller to use a larger control gain for a stronger response when facing primary challenges, such as pitch control, while using a smaller gain for secondary challenges to achieve smoothness. The calculated real-time control quantity... It is an abstract control command. Finally, the command is converted into executable drive signals. For example, a corrective torque is generated by adjusting the pressure difference of the propulsion cylinder group, and these signals are sent to the actuators of the tunnel boring machine, such as the hydraulic propulsion system, to complete the precise closed-loop control of the tunnel boring machine's attitude.

[0062] This method allows the controller's performance to be intelligently adjusted based on a pre-planned control weighting strategy, rather than being static. When traversing critical sections with drastically changing geological conditions, the tunnel boring machine (TBM) can overcome external disturbances with greater force and faster response, ensuring its attitude remains within the primary control objective. In sections with stable geological conditions, it can be controlled more gently, avoiding unnecessary overshoot and oscillations. This adaptive execution mechanism improves the accuracy, stability, and adaptability of attitude control to complex working conditions.

[0063] Optionally, updating the historical attitude data using the deviation data includes: After attitude control, the current attitude data of the tunnel boring machine at multiple time points are acquired to form the current attitude sequence; Calculate the difference between the current attitude sequence and the target attitude trajectory to generate deviation data; The historical attitude data is updated using the deviation data.

[0064] Specifically, the process of updating using deviation data is a core element in forming a closed-loop learning and self-optimization mechanism, enabling the entire control system to continuously improve. After the tunnel boring machine (TBM) actuator completes a stage of attitude control action, the system continuously acquires the TBM's current attitude data at multiple time points. This data, sorted by time or tunneling mileage, constitutes an actual tunneling trajectory, i.e., the current attitude sequence. Subsequently, the system compares this actual current attitude sequence point by point with the target attitude trajectory tracked by the controller, calculating the difference between the two to generate deviation data. This deviation data is the most direct quantitative indicator reflecting the control effect. , in, Represents the distance of tunneling The deviation data at that location is a multi-dimensional vector that includes deviations in position and angle, among other things. It is the current attitude sequence obtained through sensors. This is the target attitude trajectory previously determined by coordinated control commands. This deviation data will trigger a multi-level update and optimization cycle.

[0065] First, the latest current attitude sequence, along with the calculated deviation data, is added to the historical attitude data, continuously enriching and updating the system's historical experience base and providing data samples that are closer to the current working conditions for subsequent iterative learning. Second, the deviation data is used to update the geological data prediction model. Large control deviations often indicate discrepancies between previous geological predictions and actual conditions. This deviation can serve as an error feedback signal, allowing for fine-tuning of the geological data prediction model's internal parameters through online learning algorithms, or triggering offline retraining of the model after a certain amount of data accumulation, thereby improving the model's accuracy in predicting subsequent geological conditions. Finally, the system extracts control performance indicators from the deviation data to quantify control quality, such as the integral absolute error or maximum overshoot over the entire interval. These indicators are used to evaluate the effectiveness of arbitration rules in the dynamic control chart under different geological conditions. If a certain type of arbitration rule consistently leads to poor control performance indicators in practice, the system will automatically adjust the rule's parameters or logic, such as changing the rate of policy smooth transition, thereby optimizing higher-level policy coordination mechanisms.

[0066] Based on the same inventive concept, such as Figure 3 As shown, the present invention also provides a multi-objective control system for the attitude of a tunnel boring machine based on a spatiotemporal Transformer, the system comprising: The data acquisition module is used to acquire real-time attitude data and environmental data of the tunnel boring machine to obtain current multi-source monitoring data; The geological feedforward sensing module is used to construct a dynamic virtual geological model for predicting the geological conditions ahead based on the current multi-source monitoring data. The spatiotemporal situation analysis module is used to acquire historical attitude data, divide multiple pre-control intervals based on the dynamic virtual geological model and the historical attitude data, identify control contradictions within each pre-control interval, and generate an interval contradiction set. The strategy generation module is used to generate a strategy sequence containing the target attitude trajectory and control weight strategy for each pre-adjustment interval for the set of interval contradictions. The control and coordination module is used to coordinate the strategy sequence using a preset dynamic control diagram for coordinating different control strategies, and generate coordinated control instructions. The instruction execution module is used to adjust the parameters of the adaptive controller according to the coordinated control instructions to control the action of the tunnel boring machine's actuator and complete the attitude control. The feedback update module is used to acquire the current attitude data of the tunnel boring machine after attitude control, calculate the deviation between the current attitude data and the target attitude trajectory to obtain deviation data, and use the deviation data to update the historical attitude data.

[0067] To verify the feasibility of this invention in practice, it was applied to a tunnel boring machine (TBM) project for a subway line in a certain city. The tunnel in this project needs to traverse complex geological strata with varied and challenging geological conditions, including soft upper layers and hard lower layers, and alternating layers of sand, gravel, and clay. This places extremely high demands on the attitude control of the TBM. Traditional manual or semi-automatic control methods are ill-suited to handle attitude disturbances caused by sudden geological changes, often resulting in excessive deviations in the tunneling axis, segment misalignment, and drastic fluctuations in equipment load. This project employs the spatiotemporal Transformer-based multi-objective control system and method for TBM attitude, as described in this invention, to achieve intelligent, precise, and proactive control of the TBM's attitude.

[0068] In this embodiment, the project team utilizes the data acquisition module of this invention to acquire multi-source monitoring data such as the position, angle, speed, earth pressure, and torque of the tunnel boring machine (TBM) in real time through the laser guidance system, gyroscope, earth pressure sensor, and torque sensor on the TBM. Based on this data, the geological feedforward sensing module constructs a dynamic virtual geological model to predict the soil stiffness distribution in the unexcavated area ahead. The spatiotemporal situation analysis module divides the pre-control intervals according to the predicted geological changes and identifies control conflicts. Subsequently, the strategy generation module generates a strategy sequence for each interval, containing the target attitude trajectory and control weight strategy. The control coordination module performs smooth transition processing on these strategies to form coordinated control commands. The command execution module adjusts the PID control parameters according to the commands, driving the propulsion cylinders and other actuators to move. Finally, the feedback update module performs closed-loop optimization of the historical database, geological model, and control rules based on the deviation between the actual tunneling attitude and the target.

[0069] In terms of geological feedforward sensing, when the tunnel boring machine reached position K15+230, the data acquisition module collected data on uneven earth pressure distribution and abnormal fluctuations in cutterhead torque. Based on this current multi-source monitoring data, the geological feedforward sensing module used its self-attention mechanism-based neural network to analyze spatiotemporal characteristics and successfully predicted that at position 15 meters ahead, i.e., K15+245, it would enter a typical "soft on top, hard on the bottom" stratum, and predicted that the spatial gradient of soil stiffness in the vertical direction would exceed a preset threshold.

[0070] Based on this prediction, the spatiotemporal situation analysis module immediately marked K15+245 as the geological boundary point and divided the area from K15+245 to K15+300 into a critical pre-control interval. Within this interval, the system identified the main control contradiction as follows: the subduction moment applied to resist the upward trend may lead to an increase in the tunnel boring machine's roll angle. Therefore, the core challenge of this interval was clearly identified in the set of contradictions within the interval.

[0071] Subsequently, the strategy generation module addressed this contradiction by determining "pitch angle control" as the primary control objective for this range, and "roll angle stability" as the secondary control objective. Through a multi-objective optimization algorithm, the system generated a target attitude trajectory slightly offset downwards within the K15+245 to K15+300 range to allow for correction space. Simultaneously, the generated control weight strategy significantly improved the response weights of the pitch attitude PID control. , ), and appropriately reduced the weight of roll control.

[0072] During the strategy switch, the control coordination module detected a slight discontinuity in attitude parameters between the old and new strategy sequences at the K15+245 boundary. The module immediately invoked the arbitration rule in the dynamic control graph to generate a smooth transition scheme, fine-tuning the strategy within 5 meters before and after the boundary, forming a globally continuous coordinated control command and avoiding abrupt control changes.

[0073] When the tunnel boring machine (TBM) entered the "soft on top, hard on the bottom" geological stratum, the command execution module, based on the high-weight strategy in the coordinated control commands, drove the adaptive controller to respond strongly and rapidly to deviations in pitch attitude, effectively suppressing the upward tendency. Ultimately, the TBM smoothly passed through this challenging geological section. Current attitude data obtained by the feedback update module showed that the maximum vertical deviation in the experimental section was only 12mm, while the maximum deviation in the control section in similar strata reached 45mm. This deviation data was also used to update the historical attitude database and fine-tune the geological data prediction model, improving the accuracy of subsequent predictions.

[0074] This invention improves the accuracy and stability of shield tunneling attitude control under complex geological conditions through feedforward prediction and adaptive control. The entire process is highly automated, reducing reliance on operator experience and ensuring construction safety and efficiency.

[0075] It should be noted that the electrical connections between the various units described above do not necessarily represent direct or indirect connections. Any indirect connection method can be applied to the embodiments of the present invention as long as it achieves the purpose of the present invention. The above descriptions are merely exemplary embodiments of the present invention and should not be construed as limiting the scope of the present invention.

[0076] All equivalent changes and modifications made in accordance with the teachings of this invention are still within the scope of this invention. Those skilled in the art will readily conceive of other embodiments of this invention upon considering the specification and the disclosure of practical truth. This application is intended to cover any variations, uses, or adaptations of this invention that follow the general principles of this invention and include common knowledge or conventional techniques in the art not described herein.

Claims

1. A multi-objective control method for the attitude of a tunnel boring machine based on spatiotemporal Transformer, characterized in that: Acquire real-time attitude data and environmental data of the tunnel boring machine to obtain current multi-source monitoring data; Based on the current multi-source monitoring data, a dynamic virtual geological model is constructed to predict the geological conditions ahead. Historical attitude data is acquired, and based on the dynamic virtual geological model and the historical attitude data, multiple pre-control intervals are divided, and control contradictions within each pre-control interval are identified to generate an interval contradiction set. For the set of interval contradictions, a strategy sequence containing the target attitude trajectory and control weight strategy is generated for each pre-adjustment interval; Using a preset dynamic control diagram for coordinating different control strategies, the strategy sequence is coordinated to generate coordinated control instructions; According to the coordinated control command, the parameters of the PID controller are adjusted to control the action of the tunnel boring machine's actuator, thereby completing attitude control; After attitude control, the current attitude data of the tunnel boring machine is acquired, and the deviation between the current attitude data and the target attitude trajectory is calculated to obtain deviation data. The historical attitude data is then updated using the deviation data.

2. The method for multi-objective attitude control of a tunnel boring machine based on spatiotemporal Transformer according to claim 1, characterized in that, The acquisition of real-time attitude data and environmental data of the tunnel boring machine to obtain current multi-source monitoring data includes: Position, angle, and speed data are acquired through attitude sensors on the tunnel boring machine to form attitude data; Environmental data is generated by acquiring earth pressure and torque data through environmental sensors on the tunnel boring machine. The attitude data and environmental data are time-aligned and fused to generate current multi-source monitoring data.

3. The method for multi-objective attitude control of a tunnel boring machine based on spatiotemporal Transformer according to claim 2, characterized in that, The construction of a dynamic virtual geological model for predicting future geological conditions based on the current multi-source monitoring data includes: The current multi-source monitoring data is filtered and normalized to obtain preprocessed data; Using a pre-defined neural network based on a self-attention mechanism, spatiotemporal features are extracted from the preprocessed data to generate a current spatiotemporal feature vector; The current spatiotemporal feature vector is input into a preset geological data prediction model, and the predicted data of soil stiffness distribution and pressure trend in the unexcavated area in front of the tunnel boring machine are output. The predicted data are used to construct a dynamic virtual geological model.

4. The method for multi-objective attitude control of a tunnel boring machine based on spatiotemporal Transformer according to claim 3, characterized in that, The construction of geological data prediction models includes: Historical multi-source monitoring data is acquired and filtered and normalized to obtain historical processed data; Using the neural network based on the self-attention mechanism, spatiotemporal features are extracted from the historical processed data to generate historical spatiotemporal feature vectors; Historical soil stiffness distribution and historical pressure trends are obtained to acquire historical real-time data; Using the historical spatiotemporal feature vector as input and the historical real-time data as output, a regression prediction network is trained to obtain a geological data prediction model.

5. The method for multi-objective attitude control of a tunnel boring machine based on spatiotemporal Transformer according to claim 3, characterized in that, The set of contradictions generated in intervals includes: Acquire historical attitude data and set a gradient threshold based on the historical attitude data; Calculate the spatial gradient of soil stiffness along the tunneling direction in the dynamic virtual geological model, and identify the locations where the spatial gradient exceeds the gradient threshold as geological boundary points; Based on the geological boundary points, the tunneling path is divided into multiple pre-controlled intervals; For each pre-adjustment interval, the degree of conflict between attitude control targets is analyzed, control contradictions are identified, and all identified control contradictions are summarized into an interval contradiction set.

6. The method for multi-objective attitude control of a tunnel boring machine based on spatiotemporal Transformer according to claim 5, characterized in that, The step of generating a strategy sequence containing the target attitude trajectory and control weight strategy for each pre-adjustment interval for the set of interval contradictions includes: Based on the control contradictions, determine the primary and secondary control objectives for each pre-regulation interval; Based on the primary control objective and the secondary control objective, the target attitude trajectory is calculated using a multi-objective optimization algorithm; Based on the control contradiction, weights are assigned to the attitude control target to form a control weight strategy, and the target attitude trajectory is combined with the control weight strategy to form a strategy sequence.

7. The method for multi-objective attitude control of a tunnel boring machine based on spatiotemporal Transformer according to claim 6, characterized in that, The generated coordination control instructions include: Extract attitude parameters at the boundary between adjacent pre-adjustment intervals from the strategy sequence; Using the arbitration rules in the preset dynamic control diagram for resolving policy conflicts, the discontinuity of the attitude parameters at the boundary is calculated, and a conflict resolution solution is generated. Based on the conflict resolution, the strategies of adjacent pre-regulation intervals in the strategy sequence are adjusted to generate coordinated control instructions.

8. The method for multi-objective attitude control of a tunnel boring machine based on spatiotemporal Transformer according to claim 7, characterized in that, The step of adjusting the parameters of the PID controller to control the action of the tunnel boring machine's actuator according to the coordinated control command, and completing the attitude control, includes: The control weight strategy is extracted from the coordinated control command, and the proportional, integral, and derivative parameters of the shield machine's PID controller are adjusted to obtain adaptive control parameters. The deviation between the target attitude trajectory and the actual attitude data is used as input, and a PID controller containing the adaptive control parameters is used to calculate the real-time control quantity. The real-time control quantity is converted into a drive signal and sent to the actuator of the tunnel boring machine to complete the attitude control.

9. A method for multi-objective attitude control of a tunnel boring machine based on spatiotemporal Transformer according to claim 8, characterized in that, The step of updating the historical attitude data using the deviation data includes: After attitude control, the current attitude data of the tunnel boring machine at multiple time points are acquired to form the current attitude sequence; Calculate the difference between the current attitude sequence and the target attitude trajectory to generate deviation data; The historical attitude data is updated using the deviation data.

10. A multi-objective control system for the attitude of a tunnel boring machine based on a spatiotemporal Transformer, characterized in that, The system includes: The data acquisition module is used to acquire real-time attitude data and environmental data of the tunnel boring machine to obtain current multi-source monitoring data; The geological feedforward sensing module is used to construct a dynamic virtual geological model for predicting the geological conditions ahead based on the current multi-source monitoring data. The spatiotemporal situation analysis module is used to acquire historical attitude data, divide multiple pre-control intervals based on the dynamic virtual geological model and the historical attitude data, identify control contradictions within each pre-control interval, and generate an interval contradiction set. The strategy generation module is used to generate a strategy sequence containing the target attitude trajectory and control weight strategy for each pre-adjustment interval for the set of interval contradictions. The control and coordination module is used to coordinate the strategy sequence using a preset dynamic control diagram for coordinating different control strategies, and generate coordinated control instructions. The instruction execution module is used to adjust the parameters of the adaptive controller according to the coordinated control instructions to control the action of the tunnel boring machine's actuator and complete the attitude control. The feedback update module is used to acquire the current attitude data of the tunnel boring machine after attitude control, calculate the deviation between the current attitude data and the target attitude trajectory to obtain deviation data, and use the deviation data to update the historical attitude data.

Images