Control method and system of quasi-rectangular shield tunneling machine

Abstract

The invention relates to the field of shield tunneling machine systems, and discloses a control method of a quasi-rectangular shield tunneling machine, which comprises the following steps: acquiring section form data of the shield tunneling machine through a laser scanning device to obtain real-time section form data of the quasi-rectangular shield tunneling machine; performing section form difference analysis on the real-time section form data of the quasi-rectangular shield tunneling machine to obtain section form difference data; carrying out distribution density analysis on section form differences according to the section form difference data to obtain section form difference distribution density data; according to the method, by analyzing the cross section form difference of the shield tunneling machine in real time, the deviation between geological changes and equipment behaviors in the operation process of the shield tunneling machine can be found in time, then rapid adjustment is made, and the working efficiency of the shield tunneling machine is improved. Therefore, unnecessary downtime can be reduced, the propelling process of the shield tunneling machine is optimized, and the construction efficiency is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of shield machine systems, in particular to a control method and system for a quasi-rectangular shield machine. BACKGROUND

[0002] A quasi-rectangular shield machine is a special equipment for underground tunnel construction. Compared with traditional circular shield machines, the shield body shape of a quasi-rectangular shield machine is rectangular or close to rectangular. This design has higher space utilization and construction efficiency in specific tunnel construction environments (such as urban underground transportation, pipelines, etc.).

[0003] The existing equipment has the following disadvantages: the existing method often relies on preset parameters and fixed thrust force control strategies, and lacks the ability of real-time monitoring and dynamic adjustment. When encountering complex or changing geological conditions, it is often unable to respond quickly, resulting in low efficiency of shield machine operation, and even the need to stop for manual adjustment. The traditional method is difficult to adjust in real time according to the actual geological changes, often ignoring the influence of geological conditions on shield machine operation. This may result in the setting of thrust force not meeting the actual needs, and thus causing excessive burden on the equipment or insufficient thrust, affecting the construction progress and quality. The traditional method usually relies on manual operation in thrust force adjustment, lacking automation and intelligent level. Manual intervention not only reduces the precision of construction, but also increases the risk of errors, and may cause decision delay in the case of tedious operation.

[0004] Therefore, the present application proposes a control method and system for a quasi-rectangular shield machine to solve the above problems. SUMMARY

[0005] The purpose of the present application is to provide a control method and system for a quasi-rectangular shield machine to solve the problem that the existing method often relies on preset parameters and fixed thrust force control strategies, and lacks the ability of real-time monitoring and dynamic adjustment. When encountering complex or changing geological conditions, it is often unable to respond quickly, resulting in low efficiency of shield machine operation, and even the need to stop for manual adjustment.

[0006] To achieve the above purpose, the present application provides the following technical solution: a control method for a quasi-rectangular shield machine, comprising:

[0007] Collecting shield machine cross-section shape data by a laser scanning device to obtain real-time cross-section shape data of the quasi-rectangular shield machine; performing cross-section shape difference analysis on the real-time cross-section shape data of the quasi-rectangular shield machine to obtain cross-section shape difference data; and performing distribution density analysis between cross-section shape differences according to the cross-section shape difference data to obtain cross-section shape difference distribution density data.

[0008] According to the cross-section shape difference spatial and temporal trajectory data, shield machine motion resistance estimation is performed to obtain cross-section shape related motion resistance data; according to the cross-section shape related motion resistance data, propulsion force demand benchmark simulation evaluation is performed to obtain propulsion force demand benchmark data;

[0009] Real-time geological parameter data of the shield machine is obtained; the stratum friction influence factor is identified according to the cross-section shape related motion resistance data, to obtain the geological related friction influence factor; the propulsion force geological influence correction data is obtained by correcting the propulsion force demand benchmark data according to the geological related friction influence factor;

[0010] The dynamic propulsion force control model is obtained by constructing a shield machine control model for the propulsion force geological influence correction data through a deep reinforcement learning algorithm; the dynamic propulsion force control model is sent to the shield machine main control system to perform real-time adjustment of the shield machine posture and dynamic control of the propulsion force.

[0011] The cross-section shape data of the shield machine is collected by the laser scanning device to obtain the real-time cross-section shape data of the rectangular-like shield machine; the cross-section shape difference data is obtained by analyzing the real-time cross-section shape data of the rectangular-like shield machine, including:

[0012] The initial cross-section point cloud data is dynamically calibrated to obtain calibrated cross-section point cloud data; the cross-section contour is reconstructed according to the calibrated cross-section point cloud data to obtain the cross-section shape data of the rectangular-like shield machine;

[0013] According to the cross-section shape data of the rectangular-like shield machine, cross-section feature point data is extracted to obtain cross-section feature point data; according to the cross-section feature point data, feature point spatial and temporal correlation analysis is performed to obtain feature point spatial and temporal correlation data; according to the feature point spatial and temporal correlation data, cross-section shape dynamic evolution modeling is performed to obtain cross-section shape dynamic evolution model data;

[0014] Real-time posture data of the shield machine is obtained; the posture coupling correction is performed on the cross-section shape dynamic evolution model data according to the real-time posture data of the shield machine to obtain posture related cross-section evolution data; the cross-section deformation threshold is determined according to the posture related cross-section evolution data to obtain cross-section deformation warning data;

[0015] The cross-section shape difference data is obtained by extracting the cross-section shape difference feature of the cross-section deformation warning data through a deep residual network.

[0016] According to the cross-section shape difference data, the distribution density analysis between the cross-section shape differences is performed to obtain the cross-section shape difference distribution density data, including:

[0017] perform cross-section space grid division according to the cross-section morphology difference data to obtain difference space grid data, perform grid difference feature extraction on the difference space grid data to obtain grid difference feature data, perform initial distribution density calculation according to the grid difference feature data to obtain initial cross-section morphology difference distribution density data;

[0018] obtain shield machine advancing speed data, perform speed correlation weight adjustment on the initial cross-section morphology difference distribution density data according to the shield machine advancing speed data to obtain speed corrected distribution density data, and perform difference hot spot area identification according to the speed corrected distribution density data to obtain cross-section morphology difference hot spot area data;

[0019] perform neighborhood density expansion analysis on the cross-section morphology difference hot spot area data by a neighborhood search algorithm to obtain neighborhood expansion density data, perform density gradient direction calculation according to the neighborhood expansion density data to obtain density gradient direction data, and perform density peak point positioning according to the density gradient direction data to obtain cross-section morphology difference density peak value data;

[0020] introduce a space-time convolutional neural network to perform space-time density mode learning on the cross-section morphology difference density peak value data to obtain cross-section morphology difference distribution density data.

[0021] According to the cross-section morphology difference distribution density data, cross-section morphology difference space-time trajectory labeling is performed to obtain cross-section morphology difference space-time trajectory data, which includes:

[0022] According to the cross-section morphology difference distribution density data, a space-time density tensor is constructed to obtain density space-time tensor data, dynamic density flow field reconstruction is performed on the density space-time tensor data to obtain density flow field data, and initial trajectory point extraction is performed according to the density flow field data to obtain initial trajectory point data.

[0023] introduce a particle filter algorithm to perform multi-target tracking on the initial trajectory point data to obtain candidate trajectory segment data, perform trajectory segment correlation analysis according to the candidate trajectory segment data to obtain correlated trajectory segment data, and perform trajectory integrity verification according to the correlated trajectory segment data to obtain complete space-time trajectory data.

[0024] obtain shield machine motion parameter data, perform motion parameter coupling correction on the complete space-time trajectory data according to the shield machine motion parameter data to obtain motion correlated trajectory data, and perform trajectory abnormal point detection according to the motion correlated trajectory data to obtain trajectory abnormal point data.

[0025] The trajectory context feature data is obtained by learning trajectory context features of the trajectory abnormal point data through a spatio-temporal graph neural network.

[0026] The optimized spatio-temporal trajectory data is sent to a shield machine control system to perform spatio-temporal evolution visualization of the cross-section shape difference and dynamic planning of the shield propulsion path, so as to obtain cross-section shape difference spatio-temporal trajectory data.

[0027] The cross-section shape difference spatio-temporal trajectory data is used to estimate the motion resistance of the shield machine, so as to obtain cross-section shape related motion resistance data, including:

[0028] The trajectory feature decoupling is performed according to the cross-section shape difference spatio-temporal trajectory data, so as to obtain trajectory geometric feature data and trajectory dynamics feature data; the spatial curvature analysis is performed on the trajectory geometric feature data, so as to obtain trajectory curvature distribution data; the speed fluctuation analysis is performed according to the trajectory dynamics feature data, so as to obtain speed fluctuation mode data.

[0029] The fluid mechanics simulation algorithm is introduced to perform dynamic resistance mapping on the trajectory curvature distribution data and the speed fluctuation mode data, so as to obtain initial resistance distribution data; the resistance hotspot area is identified according to the initial resistance distribution data, so as to obtain resistance hotspot area data; the local resistance coefficient is calibrated according to the resistance hotspot area data, so as to obtain local resistance coefficient data.

[0030] The shield machine real-time geological parameter data and cutterhead torque data are obtained; the local resistance coefficient data is coupled and corrected according to the shield machine real-time geological parameter data, so as to obtain geological related resistance coefficient data; the torque resistance correlation analysis is performed according to the geological related resistance coefficient data and the cutterhead torque data, so as to obtain torque related resistance data.

[0031] The multi-modal resistance feature fusion is performed on the torque related resistance data through a deep residual network, so as to obtain fused resistance feature data; the resistance spatio-temporal pattern learning is performed according to the fused resistance feature data, so as to obtain resistance spatio-temporal evolution mode data; the motion resistance dynamic prediction is performed according to the resistance spatio-temporal evolution mode data, so as to obtain cross-section shape related motion resistance prediction data.

[0032] The cross-section shape related motion resistance prediction data is sent to the shield machine control system to perform motion resistance visualization monitoring and shield propulsion force dynamic adjustment, so as to obtain cross-section shape related motion resistance data.

[0033] The cross-section shape related motion resistance data is used to perform propulsion force demand benchmark simulation evaluation, so as to obtain propulsion force demand benchmark data, including:

[0034] construct a digital twin shield model according to the cross-section shape related motion resistance prediction data to obtain virtual shield machine data; perform multi-working condition kinematics simulation on the virtual shield machine data to obtain simulation thrust force data; perform resistance-thrust force mapping analysis according to the simulation thrust force data to obtain initial thrust force demand benchmark data;

[0035] introduce stratum hardness distribution data to perform geological adaptability correction on the initial thrust force demand benchmark data to obtain geological related thrust force demand data; perform thrust force redundancy analysis according to the geological related thrust force demand data to obtain thrust force safety margin data; perform benchmark data dynamic weighting according to the thrust force safety margin data to obtain optimized thrust force demand benchmark data;

[0036] obtain shield machine real-time cutterhead rotating speed data and thrust speed data; perform motion parameter coupling verification on the optimized thrust force demand benchmark data according to the shield machine real-time cutterhead rotating speed data and thrust speed data to obtain parameter related verification data; perform benchmark data confidence evaluation according to the parameter related verification data to obtain thrust force demand benchmark confidence data;

[0037] perform spatiotemporal feature learning on the thrust force demand benchmark confidence data through a long short-term memory network to obtain thrust force spatiotemporal evolution mode data; perform thrust force demand dynamic prediction according to the thrust force spatiotemporal evolution mode data to obtain thrust force demand benchmark prediction data;

[0038] send the thrust force demand benchmark prediction data to a shield machine control system, perform thrust force demand visual display and shield thrust system closed-loop control to obtain thrust force demand benchmark data.

[0039] wherein shield machine real-time geological parameter data is obtained; stratum friction influencing factors are identified according to the shield machine real-time geological parameter data to obtain geological related friction influencing factors, including:

[0040] stratum data in front of the shield is collected in real time through a sensor array to obtain original geological parameter data; noise filtering and feature enhancement processing are performed on the original geological parameter data to obtain preprocessed geological parameter data; stratum type intelligent classification is performed according to the preprocessed geological parameter data to obtain stratum type data;

[0041] data fusion is performed on the stratum type data through a federated learning framework to obtain cross-device geological feature data; stratum friction coefficient dynamic mapping is performed according to the cross-device geological feature data to obtain initial friction coefficient distribution data; friction hot spot area identification is performed according to the initial friction coefficient distribution data to obtain friction hot spot area data;

[0042] acquire real-time cutterhead vibration data and soil chamber pressure data of the shield machine; perform construction parameter coupling correction on the friction hotspot region data according to the real-time cutterhead vibration data and the soil chamber pressure data of the shield machine, to obtain construction-related friction distribution data; perform initial calibration of the friction influence factor according to the construction-related friction distribution data, to obtain an initial geological-related friction influence factor;

[0043] perform data enhancement and adversarial verification on the initial geological-related friction influence factor through a generative adversarial network, to obtain robust friction influence factor data; perform stratum friction correlation rule mining according to the robust friction influence factor data, to obtain friction influence factor correlation rule data;

[0044] send the friction influence factor correlation rule data to a shield machine control system, to perform stratum friction influence visualization monitoring and shield propulsion parameter self-adaptive adjustment.

[0045] According to the geological-related friction influence factor, the propulsion force demand baseline data is corrected for geological influence to obtain propulsion force geological influence correction data, including:

[0046] According to the geological-related friction influence factor, a dynamic correction weight matrix is constructed to obtain friction influence weight data; the propulsion force demand baseline data and the friction influence weight data are weighted and fused to obtain weighted propulsion force demand data; initial geological correction evaluation is performed according to the weighted propulsion force demand data to obtain initial geological correction propulsion force data;

[0047] Real-time motion parameter data of the shield machine is introduced to couple and correct the initial geological correction propulsion force data to obtain motion-related correction propulsion force data; propulsion force redundancy analysis is performed according to the motion-related correction propulsion force data to obtain propulsion force safety margin data; according to the propulsion force safety margin data, the correction data is dynamically adjusted to obtain optimized geological correction propulsion force data;

[0048] acquire historical propulsion force adjustment record data of the shield machine; perform correction effect confidence evaluation on the optimized geological correction propulsion force data and the historical adjustment record data through a random forest algorithm to obtain correction confidence score data; according to the correction confidence score data, the correction data is weighted for credibility to obtain credibility-enhanced correction propulsion force data;

[0049] adopt a long short-term memory network to learn spatiotemporal evolution features of the credibility-enhanced correction propulsion force data to obtain propulsion force spatiotemporal correction mode data; perform propulsion force demand dynamic optimization according to the propulsion force spatiotemporal correction mode data to obtain propulsion force geological influence correction prediction data;

[0050] The propulsion force geological influence correction prediction data is sent to a shield machine control system to perform real-time adjustment of the propulsion force and optimization of geological adaptability construction parameters.

[0051] The propulsion force geological influence correction data is subjected to shield machine control model construction by a deep reinforcement learning algorithm to obtain a dynamic propulsion force control model, including:

[0052] A deep reinforcement learning environment is constructed according to the propulsion force geological influence correction data to obtain correction data-driven shield machine kinematic state space data; action dimension decoupling is performed on the state space data to obtain propulsion force adjustment action space data; and initial control strategy initialization is performed according to the action space data to obtain basic propulsion force control strategy data.

[0053] A multi-objective reward function design mechanism is introduced, combined with propulsion efficiency, energy consumption indicators and geological adaptability scores to obtain comprehensive reward function data; and strategy gradient optimization is performed on the basic propulsion force control strategy data according to the comprehensive reward function data to obtain optimized control strategy data.

[0054] Meta-learning framework is adopted to perform cross-geological scene adaptation training on the optimized control strategy data to obtain generalization control strategy data; and control strategy confidence evaluation is performed according to the generalization control strategy data to obtain strategy credibility score data.

[0055] Virtual environment verification is performed on the generalization control strategy data by a digital twin shield platform to obtain simulation control effect data; and strategy parameter dynamic fine-tuning is performed according to the simulation control effect data to obtain final dynamic propulsion force control model data.

[0056] A control system of a quasi-rectangular shield machine, which is applicable to the control method of the quasi-rectangular shield machine, includes:

[0057] A cross-section morphology analysis unit collects shield machine cross-section morphology data by a laser scanning device to obtain real-time cross-section morphology data of the quasi-rectangular shield machine; cross-section morphology difference analysis is performed on the real-time cross-section morphology data of the quasi-rectangular shield machine to obtain cross-section morphology difference data; and distribution density analysis is performed on the cross-section morphology difference data to obtain cross-section morphology difference distribution density data.

[0058] A motion resistance estimation unit performs cross-section morphology difference space-time trajectory labeling according to the cross-section morphology difference distribution density data to obtain cross-section morphology difference space-time trajectory data; shield machine motion resistance estimation is performed according to the cross-section morphology difference space-time trajectory data to obtain cross-section morphology associated motion resistance data; and propulsion force demand benchmark simulation evaluation is performed according to the cross-section morphology associated motion resistance data to obtain propulsion force demand benchmark data.

[0059] The data correction unit obtains real-time geological parameter data of the shield machine, identifies a stratum friction influence factor of the cross-section shape related movement resistance data according to the real-time geological parameter data of the shield machine, to obtain a geological related friction influence factor, and corrects a propulsion force demand reference data according to the geological related friction influence factor, to obtain propulsion force geological influence correction data.

[0060] The execution unit constructs a shield machine control model of the propulsion force geological influence correction data by a deep reinforcement learning algorithm, to obtain a dynamic propulsion force control model, and sends the dynamic propulsion force control model to a shield machine main control system, to perform real-time adjustment of a shield machine posture and dynamic control of a propulsion force.

[0061] Compared with the prior art, the present application has the following advantages:

[0062] (1) The present application can timely find the deviation of geological changes and equipment behavior in the operation process of the shield machine by real-time analysis of the cross-section shape difference of the shield machine, and then make rapid adjustment. This helps to reduce unnecessary downtime and optimize the propulsion process of the shield machine, improving construction efficiency;

[0063] (2) The present application can realize accurate prediction and dynamic control of the propulsion force demand by analyzing the space-time trajectory of the cross-section shape difference and the movement resistance, and constructing a dynamic propulsion force control model by deep reinforcement learning, avoiding equipment burden or construction quality problems caused by excessive or insufficient propulsion force;

[0064] (3) The present application can fully consider the influence of different geological conditions on the operation of the shield machine by real-time acquisition of the geological parameter data of the shield machine and geological correction of the propulsion force, thereby optimizing the propulsion force and operation stability in complex geological environment. This not only improves the ability of the shield machine to adapt to complex environment, but also reduces the propulsion force fluctuation caused by stratum friction and other factors;

[0065] (4) The present application can prevent overload or overload operation, reduce equipment wear and failure probability, and prolong equipment service life by dynamically controlling the propulsion force of the shield machine. At the same time, it can adjust the posture of the shield machine in real time to avoid tilting or instability in complex geological environment, improving construction safety;

[0066] (5) The present application can greatly improve the intelligentization and automation level of the shield machine, reduce the dependence on manual intervention, and improve the accuracy and reliability of construction by using advanced technologies such as deep reinforcement learning to dynamically control the propulsion force and constructing a control model that can adaptively adjust according to real-time data during construction;

[0067] (6) The present application realizes real-time feedback and optimization. Through monitoring, analyzing and adjusting various data of the shield machine in the actual construction process, the construction parameters can be continuously optimized, the intelligence and autonomy of the operation process are improved, and various complex construction environments can be better coped with. BRIEF DESCRIPTION OF DRAWINGS

[0068] Fig. 1 It is a step flow structure schematic diagram of the overall method in an embodiment of the present application.

[0069] Fig. 2 It is a system architecture structure schematic diagram of the overall system in an embodiment of the present application.

[0070] In the figure: 1, cross-section morphology analysis unit; 2, motion resistance estimation unit; 3, data correction unit; 4, execution unit. DETAILED DESCRIPTION

[0071] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. According to the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0072] Please refer to Figs. 1-2 The present application provides a technical solution: a control method of a quasi-rectangular shield machine, comprising:

[0073] S1, collecting cross-section morphology data of the shield machine through a laser scanning device to obtain real-time cross-section morphology data of the quasi-rectangular shield machine; performing cross-section morphology difference analysis on the real-time cross-section morphology data of the quasi-rectangular shield machine to obtain cross-section morphology difference data; and performing distribution density analysis between the cross-section morphology differences according to the cross-section morphology difference data to obtain cross-section morphology difference distribution density data;

[0074] S2, performing space-time trajectory marking of the cross-section morphology differences according to the cross-section morphology difference distribution density data to obtain cross-section morphology difference space-time trajectory data; performing motion resistance estimation of the shield machine according to the cross-section morphology difference space-time trajectory data to obtain cross-section morphology associated motion resistance data; and performing simulation evaluation of the propulsion force demand reference according to the cross-section morphology associated motion resistance data to obtain propulsion force demand reference data;

[0075] S3, obtaining real-time geological parameter data of the shield machine; identifying a stratum friction influence factor according to the real-time geological parameter data of the shield machine to obtain a geological associated friction influence factor; and performing propulsion force geological influence correction on the propulsion force demand reference data according to the geological associated friction influence factor to obtain propulsion force geological influence correction data;

[0076] S4, constructing a shield machine control model for the propulsion force geology influence correction data through a deep reinforcement learning algorithm to obtain a dynamic propulsion force control model; sending the dynamic propulsion force control model to a shield machine main control system to perform real-time adjustment of a shield machine posture and dynamic control of a propulsion force.

[0077] It is worth noting that during operation, the shield machine is typically used for underground tunnel excavation, cutting through soil with a rotating cutterhead and pushing the soil into the tunnel wall with a propulsion device. The rectangular-like shield machine refers to a design that is close to a rectangle in shape or cross-section. Compared to traditional circular shield machines, the rectangular-like design can better adapt to certain specific geological and tunnel shape requirements. Laser scanning device: A laser scanning device measures the three-dimensional shape of an object or space by emitting a laser beam and receiving its reflected signal. In the application of a shield machine, the laser scanning device is used to collect real-time data on the shape of the working section of the shield machine to analyze its shape changes. Cross-section shape data: When the shield machine is in operation, its front end gradually cuts through the soil layer, forming a certain cross-section. Cross-section shape data refers to the shape and size of each section during the operation of the shield machine, usually presented in the form of point cloud data or geometric data. Cross-section shape difference analysis: By comparing the cross-section shape data collected at different time points, the differences are analyzed. For example, the deviation of the cross-section shape when the shield machine operates at different locations or at different times. This helps to detect deviations or abnormal situations during construction. Cross-section shape difference distribution density analysis: Analyze the distribution density of the shape difference in different areas of the shield machine, which is usually used to understand the distribution characteristics of the shape difference in different parts of the space. This helps to assess possible construction problems or uneven soil cutting conditions. Spatiotemporal trajectory labeling: For cross-section shape difference data, label its trajectory of change in time and space. The purpose of spatiotemporal trajectory labeling is to track the evolution of cross-section changes during the operation of the shield machine, helping to analyze its impact on the demand for propulsion force. Motion resistance estimation: During the propulsion process of the shield machine, the geological resistance and friction encountered will affect the propulsion effect. Motion resistance estimation is based on cross-section shape difference data and soil layer type factors to estimate the resistance encountered by the shield machine during operation. Baseline simulation evaluation of propulsion force demand: Propulsion force demand refers to the required propulsion force of the shield machine during the propulsion process to overcome soil resistance and other loads. Baseline simulation evaluation is based on the motion resistance data of the shield machine to simulate a reference propulsion force demand value, providing a basis for subsequent dynamic adjustment. Real-time geological parameter data: The geological environment of the shield machine operation is different, and the geological parameter data includes soil density, hardness, moisture, etc. These data have an important influence on the propulsion force demand and operation efficiency of the shield machine. Real-time geological parameter data can be collected in real time through sensors or drilling equipment. Stratum friction influence factor: The stratum friction influence factor refers to the influence coefficient of geological conditions (such as soil type, water content, etc.) on the friction and resistance during the propulsion process of the shield machine. By identifying these influence factors, the propulsion force demand evaluation can be more accurately corrected. Propulsion force geological influence correction: This step is to correct the propulsion force demand based on geological parameters (such as soil friction factors) to make the propulsion force demand more matched with the actual geological environment, thereby improving the operation accuracy.Deep Reinforcement Learning Algorithm: Deep reinforcement learning is a technique that combines deep learning and reinforcement learning, which can learn the optimal control policy through interaction with the environment. In this method, deep reinforcement learning is used to build a dynamic thrust control model for the shield machine, which can intelligently adjust under changing geological and working conditions. Dynamic Thrust Control Model: A model built based on deep reinforcement learning, which can adjust the thrust of the shield machine in real time to adapt to changing construction environments. The model can adjust the thrust output according to real-time feedback data (such as cross-sectional morphology, geological conditions, etc.), optimizing construction efficiency and safety. Master Control System: The master control system of the shield machine is the core control unit of the shield machine, responsible for coordinating and managing various control modules such as thrust, attitude adjustment, etc. The dynamic thrust control model will be sent to the master control system to execute the corresponding adjustment command.

[0078] In an embodiment, the shield machine cross-section morphology data is collected by a laser scanning device to obtain real-time cross-section morphology data of the quasi-rectangular shield machine; the cross-section morphology difference data is obtained by analyzing the cross-section morphology difference of the real-time cross-section morphology data of the quasi-rectangular shield machine, including: performing dynamic calibration processing on the initial cross-section point cloud data to obtain calibrated cross-section point cloud data; reconstructing the cross-section contour according to the calibrated cross-section point cloud data to obtain the cross-section morphology data of the quasi-rectangular shield machine; extracting the cross-section feature point data according to the cross-section feature point data; performing spatio-temporal correlation analysis on the feature point data to obtain feature point spatio-temporal correlation data; modeling the dynamic evolution of the cross-section morphology according to the feature point spatio-temporal correlation data to obtain cross-section morphology dynamic evolution model data; obtaining real-time attitude data of the shield machine; correcting the cross-section morphology dynamic evolution model data according to the real-time attitude data of the shield machine to obtain attitude-related cross-section evolution data; determining the cross-section deformation threshold according to the attitude-related cross-section evolution data to obtain cross-section deformation warning data; extracting the cross-section morphology difference feature of the cross-section deformation warning data through the deep residual network to obtain the cross-section morphology difference data.

[0079] The cross-section shape data of the shield machine is first acquired by the laser scanning device, and then dynamic calibration is performed to correct the initial point cloud data collected. After calibration, the cross-section profile of the shield machine is reconstructed to obtain real-time cross-section shape data in a rectangular shape. Then, cross-section feature points are extracted, and time-space correlation analysis is performed on these feature points to construct a cross-section shape dynamic evolution model reflecting the shape changes of the shield machine at different times and spatial positions. Real-time attitude data is used to couple and correct the cross-section shape evolution model to ensure that the model more accurately reflects the influence of the attitude change of the shield machine on the cross-section shape. Based on the corrected data, cross-section deformation threshold judgment is performed to form a warning system to prompt potential deformation problems. Finally, feature extraction is performed on the warning data by a deep residual network to further analyze the cross-section shape difference, providing a basis for subsequent construction adjustment. This method combines laser scanning, dynamic modeling, deep learning and other technologies to improve the intelligence and accuracy of the shield machine operation process.

[0080] In an embodiment, distribution density analysis is performed between cross-section shape difference data to obtain cross-section shape difference distribution density data, including: performing cross-section space grid division according to the cross-section shape difference data to obtain difference space grid data; performing grid difference feature extraction on the difference space grid data to obtain grid difference feature data; performing initial distribution density calculation according to the grid difference feature data to obtain initial cross-section shape difference distribution density data; acquiring shield machine advancing speed data; performing speed correlation weight adjustment on the initial cross-section shape difference distribution density data according to the shield machine advancing speed data to obtain speed corrected distribution density data; identifying difference hot spot areas according to the speed corrected distribution density data to obtain cross-section shape difference hot spot area data; performing neighborhood density expansion analysis on the cross-section shape difference hot spot area data by a neighborhood search algorithm to obtain neighborhood expansion density data; performing density gradient direction calculation according to the neighborhood expansion density data to obtain density gradient direction data; performing density peak point positioning according to the density gradient direction data to obtain cross-section shape difference density peak data; introducing a time-space convolutional neural network to learn the time-space density mode of the cross-section shape difference density peak data to obtain the cross-section shape difference distribution density data.

[0081] In this way, firstly, difference space grid data is obtained, and difference features are extracted in the grid to calculate an initial cross-section morphology difference distribution density. Then, the initial distribution density is adjusted in speed correlation weight combined with the shield machine advancing speed data to obtain the distribution density data after speed correction. On this basis, hot spot areas of the cross-section morphology difference are identified, the density of these areas is expanded using a neighborhood search algorithm, the neighborhood expansion density is calculated and the density gradient direction is determined. By analyzing the gradient direction, the density peak points are located to obtain the density peak value data of the cross-section morphology difference. Finally, the spatio-temporal pattern of the density peak value is learned by using a spatio-temporal convolutional neural network, so that more accurate cross-section morphology difference distribution density data is obtained. This method combines grid division, density analysis, machine learning and other technologies to improve the data analysis accuracy and dynamic response capability in the running process of the shield machine.

[0082] In an embodiment, according to the cross-section morphology difference distribution density data, cross-section morphology difference spatio-temporal trajectory labeling is performed to obtain cross-section morphology difference spatio-temporal trajectory data, including: constructing a spatio-temporal density tensor according to the cross-section morphology difference distribution density data to obtain density spatio-temporal tensor data; performing dynamic density flow field reconstruction on the density spatio-temporal tensor data to obtain density flow field data; extracting initial trajectory points according to the density flow field data to obtain initial trajectory point data; introducing a particle filtering algorithm to perform multi-target tracking on the initial trajectory point data to obtain candidate trajectory segment data; performing trajectory segment correlation analysis according to the candidate trajectory segment data to obtain correlated trajectory segment data; performing trajectory integrity checking according to the correlated trajectory segment data to obtain complete spatio-temporal trajectory data; obtaining shield machine motion parameter data; coupling and correcting the complete spatio-temporal trajectory data according to the shield machine motion parameter data to obtain motion correlated trajectory data; detecting trajectory abnormal points according to the motion correlated trajectory data to obtain trajectory abnormal point data; learning trajectory context features of the trajectory abnormal point data through a spatio-temporal graph neural network to obtain trajectory context feature data; performing trajectory completion and smoothing processing according to the trajectory context feature data to obtain optimized spatio-temporal trajectory data; sending the optimized spatio-temporal trajectory data to a shield machine control system to perform cross-section morphology difference spatio-temporal evolution visualization and shield advancing path dynamic planning to obtain cross-section morphology difference spatio-temporal trajectory data.

[0083] Thus designed, firstly, the spatiotemporal density tensor is constructed according to the cross-section morphology difference distribution density data, and the density flow field data is obtained through dynamic density flow field reconstruction. Then, the initial trajectory points are extracted from the density flow field data, and the particle filtering algorithm is used for multi-target tracking of the trajectory points to obtain candidate trajectory segment data. After correlation analysis of the candidate trajectory segments, the integrity of the trajectory segments is ensured, and finally the complete spatiotemporal trajectory data is obtained. Subsequently, combined with the motion parameters of the shield machine, the complete spatiotemporal trajectory data is coupled and corrected to obtain the motion-related trajectory data, and the trajectory abnormal point detection is performed. The context features of the trajectory are learned through the spatiotemporal graph neural network, and the trajectory is completed and smoothed to optimize the spatiotemporal trajectory data. Finally, the optimized spatiotemporal trajectory data is sent to the shield machine control system to realize the visualization of the spatiotemporal evolution of the cross-section morphology difference and the dynamic planning of the shield machine advancing path. This method combines density flow field, particle filtering, spatiotemporal graph neural network and other technologies to provide accurate spatiotemporal trajectory prediction for the shield machine, and improves the safety and efficiency of the shield machine operation process.

[0084] In an embodiment, the shield machine motion resistance estimation is performed according to the cross-section morphology difference spatiotemporal trajectory data to obtain cross-section morphology related motion resistance data, including: trajectory feature decoupling is performed according to the cross-section morphology difference spatiotemporal trajectory data to obtain trajectory geometric feature data and trajectory dynamics feature data; spatial curvature analysis is performed on the trajectory geometric feature data to obtain trajectory curvature distribution data; speed fluctuation analysis is performed according to the trajectory dynamics feature data to obtain speed fluctuation mode data; a fluid mechanics simulation algorithm is introduced to perform dynamic resistance mapping on the trajectory curvature distribution data and the speed fluctuation mode data to obtain initial resistance distribution data; resistance hot spot area identification is performed according to the initial resistance distribution data to obtain resistance hot spot area data; local resistance coefficient calibration is performed according to the resistance hot spot area data to obtain local resistance coefficient data; real-time geological parameter data and cutter torque data of the shield machine are obtained; the local resistance coefficient data is coupled and corrected according to the real-time geological parameter data of the shield machine to obtain geological related resistance coefficient data; torque related resistance data is obtained through torque related resistance analysis according to the geological related resistance coefficient data and the cutter torque data; multi-modal resistance feature fusion is performed on the torque related resistance data through a deep residual network to obtain fused resistance feature data; resistance spatiotemporal pattern learning is performed according to the fused resistance feature data to obtain resistance spatiotemporal evolution pattern data; motion resistance dynamic prediction is performed according to the resistance spatiotemporal evolution pattern data to obtain cross-section morphology related motion resistance prediction data; the cross-section morphology related motion resistance prediction data is sent to the shield machine control system to perform motion resistance visualization monitoring and shield advancing force dynamic adjustment to obtain the cross-section morphology related motion resistance data.

[0085] Dynamic resistance model formula:

[0086]

[0087] wherein, denotes the initial thrust at position and time moment, denotes the density of the object, usually related to the mass and volume of the substance, denotes the drag coefficient, usually dependent on the shape of the object, the state of motion or the fluid dynamics characteristics, is the relevant parameter, denotes the velocity vector at time moment, usually used to describe the magnitude of the velocity of the object, denotes the effective cross-sectional area, usually related to the geometric shape of the object and the fluid flow characteristics, denotes the constant coefficient, usually related to the speed change or object characteristics, denotes the change in velocity at time moment, reflecting the rate of change of velocity, denotes the average velocity, usually the average value of the velocity of the system or object over a period of time.

[0088] In this way, by decoupling the spatial and temporal trajectory data of the cross-sectional shape differences, the geometric and dynamic characteristics of the trajectory are extracted. The spatial curvature analysis of the geometric characteristics of the trajectory obtains the curvature distribution data, while the dynamic characteristics are analyzed by the speed fluctuation analysis to obtain the speed fluctuation mode. Combined with the fluid mechanics simulation algorithm, the curvature distribution and speed fluctuation mode are mapped to the dynamic resistance to obtain the initial resistance distribution data. Then, the resistance hot spot area is identified and the local resistance coefficient is calibrated to generate local resistance coefficient data. Through the real-time geological parameters and cutter torque data of the shield machine, the local resistance coefficient is corrected according to the geological conditions to obtain the geological associated resistance coefficient, and the torque data is analyzed to obtain the torque associated resistance data. Then, the deep residual network is used to fuse the torque associated resistance data in multiple modes, learn the spatiotemporal evolution pattern of the resistance, and then predict the dynamic resistance. Finally, the prediction data is sent to the shield machine control system for visual monitoring of the motion resistance and dynamic adjustment of the shield thrust force. This method provides precise resistance prediction for shield machine operation, optimizing the operation efficiency and safety.

[0089] In an embodiment, the propulsive force demand benchmark simulation evaluation according to the cross-section morphology associated motion resistance data is performed to obtain the propulsive force demand benchmark data, including: constructing a digital twin shield model according to the cross-section morphology associated motion resistance prediction data to obtain virtual shield machine data; performing multi-working condition kinematics simulation on the virtual shield machine data to obtain simulation propulsive force data; performing resistance-propulsive force mapping analysis according to the simulation propulsive force data to obtain initial propulsive force demand benchmark data; introducing stratum hardness distribution data to perform geological adaptability correction on the initial propulsive force demand benchmark data to obtain geological associated propulsive force demand data; performing propulsive force redundancy analysis according to the geological associated propulsive force demand data to obtain propulsive force safety margin data; performing benchmark data dynamic weighting according to the propulsive force safety margin data to obtain optimized propulsive force demand benchmark data; obtaining shield machine real-time cutterhead rotating speed data and propelling speed data; performing motion parameter coupling verification on the optimized propulsive force demand benchmark data according to the shield machine real-time cutterhead rotating speed data and propelling speed data to obtain parameter associated verification data; performing benchmark data confidence evaluation according to the parameter associated verification data to obtain propulsive force demand benchmark confidence data; performing spatio-temporal feature learning on the propulsive force demand benchmark confidence data through a long short-term memory network to obtain propulsive force spatio-temporal evolution mode data; performing propulsive force demand dynamic prediction according to the propulsive force spatio-temporal evolution mode data to obtain propulsive force demand benchmark prediction data; sending the propulsive force demand benchmark prediction data to a shield machine control system to perform propulsive force demand visual display and shield propelling system closed-loop control to obtain the propulsive force demand benchmark data.

[0090] Motion resistance-propulsive force mapping formula:

[0091]

[0092] wherein, represents the simulation propulsive force, represents the real-time motion resistance of the th cross-section partition, represents the partition contact area, represents the partition friction coefficient, represents a constant, usually representing density, friction coefficient or other physical constant, represents the propelling speed, represents the cross-section morphology resistance coefficient.

[0093] In this way, firstly, a digital twin shield model is constructed by using cross-section shape related motion resistance prediction data to generate virtual shield data, and simulation propulsion force data is obtained through multi-working condition kinematics simulation. The simulation data is analyzed for resistance-propulsion force mapping to obtain preliminary propulsion force demand benchmark data. Then, geological adaptability correction is performed by introducing stratum hardness distribution data to obtain geological related propulsion force demand data, and propulsion force redundancy analysis is performed to ensure the safety margin of the propulsion force. Further, the benchmark data is dynamically weighted according to the safety margin data to optimize the propulsion force demand benchmark data. The optimized benchmark data is verified by coupling the motion parameters by obtaining real-time cutterhead speed and propulsion speed data of the shield machine to generate parameter related verification data, and the confidence of the benchmark data is evaluated. Then, the long short-term memory network is used to learn the spatio-temporal characteristics of the propulsion force demand benchmark confidence data to obtain the spatio-temporal evolution pattern of the propulsion force, and then dynamic prediction is performed. Finally, the prediction data is transmitted to the shield machine control system for visual display of the propulsion force demand and closed-loop control of the shield propulsion system. The method improves the accuracy and adaptability of the shield machine propulsion force prediction.

[0094] In an embodiment, real-time geological parameter data of the shield machine is obtained; stratum friction influence factors are identified from the cross-section shape related motion resistance data according to the real-time geological parameter data of the shield machine to obtain geological related friction influence factors, including: stratum data in front of the shield machine is collected in real time by a sensor array to obtain original geological parameter data; noise filtering and feature enhancement processing are performed on the original geological parameter data to obtain preprocessed geological parameter data; stratum type intelligent classification is performed according to the preprocessed geological parameter data to obtain stratum type data; data fusion is performed on the stratum type data by introducing a federated learning framework to obtain cross-device geological feature data; stratum friction coefficient dynamic mapping is performed according to the cross-device geological feature data to obtain initial friction coefficient distribution data; friction hot spot area identification is performed according to the initial friction coefficient distribution data to obtain friction hot spot area data; real-time cutterhead vibration data and soil chamber pressure data of the shield machine are obtained; construction parameter coupling correction is performed on the friction hot spot area data according to the real-time cutterhead vibration data and soil chamber pressure data of the shield machine to obtain construction related friction distribution data; initial calibration of the friction influence factors is performed according to the construction related friction distribution data to obtain initial geological related friction influence factors; data enhancement and adversarial verification of the initial geological related friction influence factors are performed by a generative adversarial network to obtain robust friction influence factor data; stratum friction related rules are mined according to the robust friction influence factor data to obtain friction influence factor related rule data; the friction influence factor related rule data is sent to the shield machine control system to perform stratum friction influence visualization monitoring and shield propulsion parameter self-adaptive adjustment.

[0095] Stratum type classification formula:

[0096]

[0097] wherein, denotes the prediction function value of the model for the formation type under given prior geological data is the conditional probability, denotes, denotes the prediction function value of the model for the formation type and the parameters under given prior geological data is denotes the prediction function of the formation type and for describing the output of the model under given data, and are the model parameters associated with each formation type.

[0098] Thus designed, by real-time acquisition of stratum data in front of the shield machine and noise filtering and feature enhancement, preprocessed geological parameter data are generated, then intelligent classification of the stratum type is performed to obtain stratum type data. The cross-device geological feature data are fused by using a federated learning framework, and the stratum friction coefficient is dynamically mapped according to the fused data to obtain initial friction coefficient distribution data. Further, the friction hotspot area is identified, and the construction parameter coupling correction is performed in combination with the cutterhead vibration and soil chamber pressure data to generate construction associated friction distribution data, and initial calibration of the friction influencing factor is performed. The robustness of the data is enhanced by a generative adversarial network to obtain a robust friction influencing factor, and the association rule mining of the friction influencing factor is performed, and finally the rule data are transmitted to the shield machine control system to perform visual monitoring of the stratum friction influence and adaptive adjustment of the shield advancing parameters. This method realizes real-time and accurate identification of the stratum friction influence, optimizes the friction control in the shield machine advancing process, and improves the construction efficiency and safety.

[0099] In an embodiment, the propulsion force demand benchmark data is corrected according to a geological correlation friction influence factor to obtain propulsion force geological influence correction data, including: constructing a dynamic correction weight matrix according to the geological correlation friction influence factor to obtain friction influence weight data; performing weighted fusion processing on the propulsion force demand benchmark data and the friction influence weight data to obtain weighted propulsion force demand data; performing initial geological correction evaluation according to the weighted propulsion force demand data to obtain initial geological correction propulsion force data; introducing shield machine real-time motion parameter data to perform motion parameter coupling correction on the initial geological correction propulsion force data to obtain motion correlation correction propulsion force data; performing propulsion force redundancy analysis according to the motion correlation correction propulsion force data to obtain propulsion force safety margin data; performing dynamic adjustment of correction data according to the propulsion force safety margin data to obtain optimized geological correction propulsion force data; obtaining shield machine historical propulsion force adjustment record data; performing correction effect confidence evaluation on the optimized geological correction propulsion force data and the historical adjustment record data by a random forest algorithm to obtain correction confidence score data; performing correction data credibility weighting according to the correction confidence score data to obtain credibility enhanced correction propulsion force data; performing spatiotemporal evolution feature learning on the credibility enhanced correction propulsion force data by using a long short-term memory network to obtain propulsion force spatiotemporal correction mode data; performing propulsion force demand dynamic optimization according to the propulsion force spatiotemporal correction mode data to obtain propulsion force geological influence correction prediction data; and sending the propulsion force geological influence correction prediction data to a shield machine control system to perform real-time adjustment of the propulsion force and optimization of geological adaptability construction parameters.

[0100] In this way, first, a dynamic correction weight matrix is constructed according to a geological correlation friction influence factor, weighted fusion of propulsion force demand benchmark data and friction influence weight data is performed to obtain weighted propulsion force demand data, and initial geological correction evaluation is performed to obtain initial geological correction propulsion force data. Then, motion parameter coupling correction is performed on the initial data by using shield machine real-time motion parameter data to obtain motion correlation correction propulsion force data, and propulsion force redundancy analysis is performed to obtain propulsion force safety margin data. Based on the safety margin, dynamic adjustment of correction data is performed to optimize the geological correction propulsion force data. The optimized data and historical propulsion force adjustment record data are evaluated for correction effect by a random forest algorithm to obtain correction confidence scores, and credibility enhancement correction is performed according to the scores. Subsequently, spatiotemporal evolution feature learning is performed on the enhanced data by using a long short-term memory network to predict a propulsion force spatiotemporal correction mode, thereby realizing dynamic optimization of propulsion force demand. Finally, the optimized propulsion force data is sent to a shield machine control system to perform real-time adjustment and optimization of geological adaptability construction parameters. This process improves the accuracy of propulsion force adjustment and construction adaptability.

[0101] In an embodiment, a shield machine control model is constructed for the propulsion force geological influence correction data by a deep reinforcement learning algorithm to obtain a dynamic propulsion force control model, including: constructing a deep reinforcement learning environment according to the propulsion force geological influence correction data to obtain modified data-driven shield machine kinematics state space data; decoupling the action dimension of the state space data to obtain propulsion force adjustment action space data; initializing an initial control strategy according to the action space data to obtain basic propulsion force control strategy data; introducing a multi-objective reward function design mechanism, combining propulsion efficiency, energy consumption indicators and geological adaptability scores to obtain comprehensive reward function data; performing policy gradient optimization on the basic propulsion force control strategy data according to the comprehensive reward function data to obtain optimized control strategy data; training the optimized control strategy data using a meta-learning framework for cross-geological scene adaptation to obtain generalization control strategy data; performing control strategy confidence evaluation according to the generalization control strategy data to obtain strategy credibility score data; verifying the generalization control strategy data through a digital twin shield platform in a virtual environment to obtain simulation control effect data; and performing dynamic fine-tuning of strategy parameters according to the simulation control effect data to obtain final dynamic propulsion force control model data.

[0102] In this design, first, a learning environment is established according to the propulsion force geological influence correction data to obtain modified data-driven shield machine kinematics state space. By decoupling the action dimension, a propulsion force adjustment action space is generated, and a control strategy is initialized. A multi-objective reward function is designed to combine propulsion efficiency, energy consumption and geological adaptability scores to optimize the basic control strategy. Next, a meta-learning framework is introduced to train the optimized strategy for adaptability in different geological scenarios, thereby obtaining a generalization control strategy. The control strategy is evaluated for credibility, and a digital twin shield platform is used for virtual environment verification to test the effectiveness of the strategy. The strategy parameters are fine-tuned according to the simulation data to finally obtain a dynamic propulsion force control model. This method realizes intelligent control of the shield machine under different geological conditions, optimizes the adjustment of the propulsion force, improves construction efficiency and safety, and can adapt to geological changes in real time during actual construction.

[0103] A control system for a quasi-rectangular shield machine, which is applicable to the control method of the quasi-rectangular shield machine described above, comprising:

[0104] A cross-sectional morphology analysis unit 1 acquires shield machine cross-sectional morphology data through a laser scanning device to obtain real-time cross-sectional morphology data of the quasi-rectangular shield machine; performs cross-sectional morphology difference analysis on the real-time cross-sectional morphology data of the quasi-rectangular shield machine to obtain cross-sectional morphology difference data; and performs distribution density analysis on the cross-sectional morphology difference data to obtain cross-sectional morphology difference distribution density data.

[0105] The motion resistance estimation unit 2 performs cross-section shape difference space-time trajectory marking according to the cross-section shape difference distribution density data to obtain cross-section shape difference space-time trajectory data; performs shield machine motion resistance estimation according to the cross-section shape difference space-time trajectory data to obtain cross-section shape related motion resistance data; and performs advance force demand benchmark simulation evaluation according to the cross-section shape related motion resistance data to obtain advance force demand benchmark data;

[0106] The data correction unit 3 obtains shield machine real-time geological parameter data; identifies a stratum friction influence factor according to the shield machine real-time geological parameter data and the cross-section shape related motion resistance data to obtain a geological related friction influence factor; and performs advance force geological influence correction on the advance force demand benchmark data according to the geological related friction influence factor to obtain advance force geological influence correction data;

[0107] The execution unit 4 constructs a shield machine control model for the advance force geological influence correction data through a deep reinforcement learning algorithm to obtain a dynamic advance force control model; and sends the dynamic advance force control model to a shield machine main control system to perform shield machine posture real-time adjustment and dynamic advance force control.

[0108] The embodiments of the present application are described in detail above in combination with the drawings, but the present application is not limited thereto, and various changes can be made within the knowledge of those skilled in the art without departing from the spirit of the present application.

Claims

1. A control method for a rectangular tunnel boring machine, characterized in that, include: The cross-sectional shape data of the tunnel boring machine is collected by a laser scanning device to obtain real-time cross-sectional shape data of a rectangular tunnel boring machine; cross-sectional shape difference analysis is performed on the real-time cross-sectional shape data of the rectangular tunnel boring machine to obtain cross-sectional shape difference data; and distribution density analysis is performed on the cross-sectional shape difference data to obtain cross-sectional shape difference distribution density data. Spatiotemporal trajectory marking of cross-sectional morphology differences is performed based on the distribution density data of cross-sectional morphology differences to obtain spatiotemporal trajectory data of cross-sectional morphology differences; Based on the spatiotemporal trajectory data of cross-sectional shape differences, the motion resistance of the tunnel boring machine is estimated to obtain motion resistance data related to cross-sectional shape. Based on the cross-sectional shape and associated motion drag data, a baseline simulation assessment of propulsion demand is performed to obtain baseline propulsion demand data; Obtain real-time geological parameter data for the tunnel boring machine; Based on the real-time geological parameter data of the tunnel boring machine, the stratum friction influence factor is identified by analyzing the cross-sectional shape-related motion resistance data to obtain the geological-related friction influence factor. Based on the geologically related frictional influence factor, the thrust demand baseline data is corrected for geological influence to obtain the thrust geological influence correction data; A shield tunneling machine control model is constructed using deep reinforcement learning algorithms to correct geological impact data on propulsion force, thereby obtaining a dynamic propulsion force control model. The dynamic propulsion force control model is then sent to the shield tunneling machine's main control system to execute real-time adjustment of the shield tunneling machine's attitude and dynamic control of propulsion force.

2. The control method for a rectangular shield tunneling machine according to claim 1, characterized in that, The cross-sectional shape data of the tunnel boring machine (TBM) is acquired using a laser scanning device to obtain real-time cross-sectional shape data for a rectangular TBM. Cross-sectional shape difference analysis is then performed on the real-time cross-sectional shape data of the rectangular TBM to obtain cross-sectional shape difference data, including: The initial cross-section point cloud data is dynamically calibrated to obtain calibrated cross-section point cloud data; the cross-section contour is reconstructed based on the calibrated cross-section point cloud data to obtain rectangular shield machine cross-section morphology data. Based on the rectangular shield tunneling machine cross-sectional morphology data, cross-sectional feature points are extracted to obtain cross-sectional feature point data; based on the cross-sectional feature point data, spatiotemporal correlation analysis of feature points is performed to obtain spatiotemporal correlation data of feature points; based on the spatiotemporal correlation data of feature points, dynamic evolution modeling of cross-sectional morphology is performed to obtain dynamic evolution model data of cross-sectional morphology. Acquire real-time attitude data of the tunnel boring machine; perform attitude coupling correction on the dynamic evolution model data of the cross-section shape based on the real-time attitude data of the tunnel boring machine to obtain attitude-related cross-section evolution data; determine the cross-section deformation threshold based on the attitude-related cross-section evolution data to obtain cross-section deformation early warning data. The cross-sectional deformation early warning data is processed by a deep residual network to extract cross-sectional morphological difference features, so as to obtain cross-sectional morphological difference data.

3. The control method for a rectangular shield tunneling machine according to claim 2, characterized in that, Based on the cross-sectional morphology difference data, a distribution density analysis of the cross-sectional morphology differences is performed to obtain the cross-sectional morphology difference distribution density number, including: The cross-sectional spatial grid is divided according to the cross-sectional morphology difference data to obtain difference spatial grid data; the difference features within the grid are extracted from the difference spatial grid data to obtain grid difference feature data; the initial distribution density is calculated according to the grid difference feature data to obtain initial cross-sectional morphology difference distribution density data. Acquire tunnel boring machine (TBM) advance speed data; adjust the initial cross-sectional shape difference distribution density data by speed correlation weight based on the TBM advance speed data to obtain speed-corrected distribution density data; identify difference hotspot areas based on the speed-corrected distribution density data to obtain cross-sectional shape difference hotspot area data. The neighborhood density expansion analysis of the hotspot region data of the cross-sectional morphology difference is performed by a neighborhood search algorithm to obtain the neighborhood expansion density data; the density gradient direction is calculated based on the neighborhood expansion density data to obtain the density gradient direction data; and the density peak point is located based on the density gradient direction data to obtain the density peak data of the cross-sectional morphology difference. A spatiotemporal convolutional neural network is introduced to learn the spatiotemporal density pattern of the peak data of cross-sectional morphological difference density in order to obtain the distribution density data of cross-sectional morphological difference.

4. The control method for a rectangular shield tunneling machine according to claim 3, characterized in that, Based on the distribution density data of cross-sectional morphology differences, spatiotemporal trajectories of cross-sectional morphology differences are marked to obtain spatiotemporal trajectory data of cross-sectional morphology differences, including: A spatiotemporal density tensor is constructed based on the cross-sectional morphology difference distribution density data to obtain density spatiotemporal tensor data; dynamic density flow field reconstruction is performed on the density spatiotemporal tensor data to obtain density flow field data; initial trajectory points are extracted based on the density flow field data to obtain initial trajectory point data. A particle filter algorithm is introduced to perform multi-target tracking on the initial trajectory point data to obtain candidate trajectory segment data; trajectory segment correlation analysis is performed based on the candidate trajectory segment data to obtain associated trajectory segment data; trajectory integrity verification is performed based on the associated trajectory segment data to obtain complete spatiotemporal trajectory data. Acquire tunnel boring machine (TBM) motion parameter data; perform motion parameter coupling correction on complete spatiotemporal trajectory data based on TBM motion parameter data to obtain motion-related trajectory data; detect trajectory anomalies based on motion-related trajectory data to obtain trajectory anomaly data; The trajectory context features of outlier points are learned by using a spatiotemporal graph neural network to obtain trajectory context feature data; trajectory completion and smoothing are then performed based on the trajectory context feature data to obtain optimized spatiotemporal trajectory data. The optimized spatiotemporal trajectory data is sent to the tunnel boring machine control system to perform visualization of the spatiotemporal evolution of cross-sectional morphology differences and dynamic planning of the tunnel boring machine propulsion path, so as to obtain the spatiotemporal trajectory data of cross-sectional morphology differences.

5. The control method for a rectangular shield tunneling machine according to claim 4, characterized in that, Based on the spatiotemporal trajectory data of cross-sectional shape differences, the motion resistance of the tunnel boring machine is estimated to obtain motion resistance data related to cross-sectional shape, including: Based on the spatiotemporal trajectory data with the aforementioned cross-sectional morphology differences, trajectory features are decoupled to obtain trajectory geometric feature data and trajectory dynamic feature data; spatial curvature analysis is performed on the trajectory geometric feature data to obtain trajectory curvature distribution data; velocity fluctuation analysis is performed on the trajectory dynamic feature data to obtain velocity fluctuation pattern data. A fluid dynamics simulation algorithm is introduced to dynamically map the trajectory curvature distribution data and the velocity fluctuation pattern data to obtain initial resistance distribution data; resistance hotspot regions are identified based on the initial resistance distribution data to obtain resistance hotspot region data; and local resistance coefficients are calibrated based on the resistance hotspot region data to obtain local resistance coefficient data. Acquire real-time geological parameter data and cutterhead torque data of the tunnel boring machine; perform ground resistance coupling correction on the local resistance coefficient data based on the real-time geological parameter data of the tunnel boring machine to obtain geologically related resistance coefficient data; perform torque-resistance correlation analysis on the geologically related resistance coefficient data and cutterhead torque data to obtain torque-related resistance data. Multimodal resistance features are fused from torque-related resistance data using a deep residual network to obtain fused resistance feature data; spatiotemporal resistance patterns are learned based on the fused resistance feature data to obtain spatiotemporal resistance evolution pattern data; dynamic prediction of motion resistance is performed based on the spatiotemporal resistance evolution pattern data to obtain cross-sectional shape-related motion resistance prediction data. The predicted motion resistance data associated with the cross-sectional shape is sent to the tunnel boring machine control system to perform visual monitoring of motion resistance and dynamic adjustment of the tunnel boring machine propulsion force, so as to obtain the motion resistance data associated with the cross-sectional shape.

6. The control method for a rectangular shield tunneling machine according to claim 5, characterized in that, Based on the cross-sectional shape and associated motion drag data, a baseline simulation assessment of propulsion demand is performed to obtain baseline propulsion demand data, including: A digital twin shield tunneling model is constructed based on the predicted motion resistance data associated with the cross-sectional shape to obtain virtual shield machine data; multi-condition kinematic simulation is performed on the virtual shield machine data to obtain simulated propulsion force data; resistance-propulsion force mapping analysis is performed based on the simulated propulsion force data to obtain the initial propulsion force requirement baseline data; The initial thrust demand baseline data is modified for geological adaptability by introducing formation hardness distribution data to obtain geologically correlated thrust demand data; thrust redundancy analysis is performed based on the geologically correlated thrust demand data to obtain thrust safety margin data; and the baseline data is dynamically weighted based on the thrust safety margin data to obtain optimized thrust demand baseline data. Acquire real-time cutterhead rotation speed data and propulsion speed data of the tunnel boring machine; perform motion parameter coupling verification on the optimized propulsion force demand benchmark data based on the real-time cutterhead rotation speed data and propulsion speed data of the tunnel boring machine to obtain parameter correlation verification data; perform benchmark data confidence assessment based on parameter correlation verification data to obtain propulsion force demand benchmark confidence data; Spatiotemporal feature learning is performed on the propulsion demand baseline confidence data using a long short-term memory network to obtain propulsion spatiotemporal evolution model data; dynamic prediction of propulsion demand is then performed based on the propulsion spatiotemporal evolution model data to obtain propulsion demand baseline prediction data. The aforementioned propulsion demand baseline prediction data is sent to the tunnel boring machine control system to perform propulsion demand visualization and closed-loop control of the tunnel boring machine propulsion system, thereby obtaining the propulsion demand baseline data.

7. The control method for a rectangular shield tunneling machine according to claim 6, characterized in that, Obtain real-time geological parameter data for the tunnel boring machine; Based on the real-time geological parameter data of the tunnel boring machine, the stratum friction influence factors are identified from the cross-sectional shape-related motion resistance data to obtain the geologically related friction influence factors, including: Real-time geological data in front of the tunnel boring machine is collected by a sensor array to obtain raw geological parameter data; noise filtering and feature enhancement processing are performed on the raw geological parameter data to obtain preprocessed geological parameter data; and intelligent classification of strata types is performed based on the preprocessed geological parameter data to obtain strata type data. A federated learning framework is introduced to fuse the stratigraphic type data to obtain cross-device geological feature data; dynamic mapping of stratigraphic friction coefficients is performed based on the cross-device geological feature data to obtain initial friction coefficient distribution data; friction hotspot areas are identified based on the initial friction coefficient distribution data to obtain friction hotspot area data. Acquire real-time cutterhead vibration data and soil chamber pressure data of the tunnel boring machine; perform construction parameter coupling correction on the friction hotspot area data based on the real-time cutterhead vibration data and soil chamber pressure data of the tunnel boring machine to obtain construction-related friction distribution data; perform initial calibration of friction influence factors based on construction-related friction distribution data to obtain initial geological-related friction influence factors. Generative adversarial networks are used to augment and validate the initial geologically correlated frictional influencing factors to obtain robust frictional influencing factor data; based on the robust frictional influencing factor data, stratigraphic frictional correlation rules are mined to obtain frictional influencing factor correlation rule data. The data on the association rules of friction impact factors are sent to the shield machine control system to perform visual monitoring of the impact of ground friction and adaptive adjustment of shield propulsion parameters.

8. The control method for a rectangular shield tunneling machine according to claim 7, characterized in that, Based on the geologically correlated frictional influence factor, the thrust demand baseline data is corrected for geological impact to obtain thrust geological impact correction data, including: A dynamic correction weight matrix is ​​constructed based on the geologically related friction influence factors to obtain friction influence weight data; the propulsion demand baseline data and the friction influence weight data are weighted and fused to obtain weighted propulsion demand data; an initial geological correction assessment is performed based on the weighted propulsion demand data to obtain initial geologically corrected propulsion data. The initial geologically corrected propulsion force data is coupled and corrected by introducing real-time motion parameter data of the tunnel boring machine to obtain motion-related corrected propulsion force data; propulsion force redundancy analysis is performed based on the motion-related corrected propulsion force data to obtain propulsion force safety margin data; and the correction data is dynamically adjusted based on the propulsion force safety margin data to obtain optimized geologically corrected propulsion force data. Obtain historical thrust adjustment records of the tunnel boring machine; evaluate the confidence level of the optimized geologically corrected thrust data and historical adjustment records using a random forest algorithm to obtain corrected confidence score data; and weight the corrected data based on the corrected confidence score data to obtain confidence-enhanced corrected thrust data. A long short-term memory network is used to learn the spatiotemporal evolution characteristics of the reliability-enhanced propulsion data to obtain the spatiotemporal correction model data of propulsion; based on the spatiotemporal correction model data of propulsion, dynamic optimization of propulsion demand is performed to obtain the geological impact correction prediction data of propulsion. The propulsion force geological impact correction prediction data is sent to the tunnel boring machine control system to perform real-time adjustment of propulsion force and optimization of geological adaptability construction parameters.

9. The control method for a rectangular shield tunneling machine according to claim 8, characterized in that, A shield tunneling machine control model is constructed using deep reinforcement learning algorithms to correct geological impact data on propulsion force, resulting in a dynamic propulsion force control model, including: A deep reinforcement learning environment is constructed based on the geological impact correction data of the propulsion force to obtain the kinematic state space data of the tunnel boring machine driven by the correction data; the motion dimension is decoupled from the state space data to obtain the motion space data of the propulsion force adjustment; the initial control strategy is initialized based on the motion space data to obtain the basic propulsion force control strategy data. A multi-objective reward function design mechanism is introduced, which combines propulsion efficiency, energy consumption index and geological adaptability score to obtain comprehensive reward function data; based on the comprehensive reward function data, the basic propulsion control strategy data is optimized by strategy gradient to obtain optimized control strategy data; A meta-learning framework is used to train the optimized control strategy data across geological scenarios to obtain generalized control strategy data; the confidence of the control strategy is evaluated based on the generalized control strategy data to obtain the strategy confidence score data. The generalized control strategy data is verified in a virtual environment using a digital twin shield tunneling platform to obtain simulation control effect data. Based on the simulation control effect data, the strategy parameters are dynamically fine-tuned to obtain the final dynamic propulsion control model data.

10. A control system for a rectangular tunnel boring machine, applicable to the control method for a rectangular tunnel boring machine as described in any one of claims 1-9, characterized in that, include: The cross-sectional shape analysis unit (1) collects the cross-sectional shape data of the shield machine through a laser scanning device to obtain the real-time cross-sectional shape data of the rectangular shield machine; A cross-sectional shape difference analysis was performed on the real-time cross-sectional shape data of a rectangular shield tunneling machine to obtain cross-sectional shape difference data; based on the cross-sectional shape difference data, a distribution density analysis was performed on the cross-sectional shape difference to obtain cross-sectional shape difference distribution density data. The motion resistance estimation unit (2) marks the spatiotemporal trajectory of cross-sectional shape difference based on the cross-sectional shape difference distribution density data to obtain the spatiotemporal trajectory data of cross-sectional shape difference; Based on the spatiotemporal trajectory data of cross-sectional shape differences, the motion resistance of the tunnel boring machine is estimated to obtain motion resistance data related to cross-sectional shape. Based on the cross-sectional shape and associated motion drag data, a baseline simulation assessment of propulsion demand is performed to obtain baseline propulsion demand data; Data correction unit (3) acquires real-time geological parameter data of the tunnel boring machine; Based on the real-time geological parameter data of the tunnel boring machine, the stratum friction influence factor is identified by analyzing the cross-sectional shape-related motion resistance data to obtain the geological-related friction influence factor. Based on the geologically related frictional influence factor, the thrust demand baseline data is corrected for geological influence to obtain the thrust geological influence correction data; The execution unit (4) constructs a shield machine control model by using a deep reinforcement learning algorithm to correct the geological influence of the propulsion force, so as to obtain a dynamic propulsion force control model; and sends the dynamic propulsion force control model to the shield machine main control system to perform real-time adjustment of the shield machine attitude and dynamic control of the propulsion force.

Images