A shield machine intelligent direction adjustment system and control method based on model predictive control

Abstract

The present invention discloses a shield machine intelligent steering system and control method based on model predictive control. The system's shield posture measurement module collects shield posture data and outputs shield posture deviation values in combination with a preset tunnel design axis; the data acquisition module collects and stores shield parameters; the model predictive controller receives shield parameters and outputs a propulsion cylinder pressure signal; the oil pressure posture module receives the pressure signal and adjusts the shield machine posture. The method includes: a simulation system designs a model predictive controller to construct an intelligent steering system; the system collects shield data and outputs a pressure signal to control the shield machine's propulsion cylinder to execute an action. The present invention solves the problem of manually operating the shield machine steering. The model predictive control algorithm can handle the coupling relationship in the shield machine steering process, better adapt to system changes and external interference, greatly improve the intelligence level of the shield, and effectively solves the disadvantage that it is difficult to accurately establish or impossible to establish a complex shield-environment interaction model using a mathematical model.

Description

Technical Field

[0001] The present invention relates to an intelligent shield direction adjustment system, and in particular to an intelligent shield direction adjustment system and a control method based on model predictive control. Background Art

[0002] The shield machine is a highly efficient, environmentally friendly, and safe large-scale tunnel excavation equipment. During construction, the shield machine must excavate along the tunnel's designed axis. Excessive deviation between the actual excavation axis and the designed axis can directly cause the tunnel's shape and dimensions to deviate from design requirements, reducing tunnel construction quality and increasing safety risks and construction costs. Therefore, adjusting the shield machine's excavation direction according to the designed axis to ensure that the actual excavation axis matches the designed axis as closely as possible is key to ensuring tunnel construction quality.

[0003] Traditionally, shield machine (TBM) tunneling direction adjustment relies on manual control by the shield driver. Based on the deviation of the shield machine from the designed axis, as measured by the shield guidance system, the driver adjusts the pressure in the thrust cylinders of different sections, thereby changing the shield machine's tunneling direction. This process constitutes a human-in-the-loop (HIL) closed-loop control system, in which the shield driver primarily analyzes deviations, makes decisions, and outputs control signals. Consequently, traditional TBM attitude control relies heavily on the driver's experience and intuition. However, due to the complex human-machine-environment interaction, the shield machine's significant inertia, and the numerous parameters the driver must monitor and manipulate, manual attitude control suffers from low accuracy, poor real-time performance, and lags in control. This leads to significant serpentine tunneling in actual projects. Furthermore, the long and intense control work in a noisy construction environment can easily fatigue the shield driver and cause him to lose focus, further complicating the difficulty of manual direction adjustment. The training period for shield drivers is long and the cost is high, and the shortage of qualified shield drivers is a major challenge facing the industry. In response to the above-mentioned defects of manual direction adjustment of traditional shield machines, the development of automated and intelligent shield direction adjustment technology has become the development trend and research hotspot of the industry, and a large amount of research and development work has been carried out at home and abroad.

[0004] At present, the technologies for automated and intelligent shield direction adjustment are mainly divided into three categories. Among them, the first category is the method based on logical rules. This method establishes logical rules based on manual experience, and converts the shield posture deviation information into partitioned cylinder control instructions through the established logical rules, which is equivalent to automating manual control by using the method of logical programming. Patent CN1800583 uses a total station equipped with a space vector method calculation program module to automatically obtain the shield posture. According to the selected correction mode, the shield posture automatic correction software is used to calculate the group position of the shield cut center under the shield posture. Finally, the calculation results and control instructions are transmitted to the control module of the shield propulsion system through the PLC (Programmable Logic Controller) module for corresponding adjustment, thereby achieving the purpose of continuous automatic correction of the shield posture. This method can reduce the number of operators and the fluctuation of the direction adjustment deviation is lower than that of manual correction. Patent CN102518446A uses a laser total station to measure the real-time position information of the front, hinged, and rear sections of the shield machine. Based on this information, it calculates the vertical and horizontal deviations of the front, hinged, and rear sections relative to the tunnel design axis, as well as the relative deviations and relative deviation changes of the shield machine as a whole in the vertical and horizontal planes. It also calculates the relative vertical and horizontal deflections of the front, rear, and overall sections of the shield machine, sets multiple preset thresholds, and determines whether these deviations exceed the preset thresholds. When the thresholds are exceeded, the controller uses the thrust of the jacks of each section of the shield machine measured by the pressure sensor or the main thrust of the shield machine to determine the correction method and the position of the jacks performing the correction. The pressure value of the jacks performing the correction is determined according to an empirical formula, and their expansion and contraction are controlled to adjust the deflection posture of the front and rear sections of the shield machine, ultimately achieving control of the shield machine's excavation route. Patent CN102606165A uses an industrial computer to receive real-time shield machine posture data transmitted from a laser automatic posture measurement system, and compares it with the designed shield posture data to determine whether to send a correction control signal to the controller. When the correction control signal is sent, the controller sends a drive signal to the drive mechanism based on the signal, thereby controlling the thrust cylinder and the folding hydraulic cylinder to make corresponding movements, adjusting the shield machine posture and achieving automatic correction. The second category is based on machine learning methods. This type of method uses a neural network training model to obtain a mapping from the shield machine posture deviation to the direction adjustment instruction. This is equivalent to using the information available during the operation of the shield machine through the neural network model to obtain the shield machine posture deviation or control instruction in advance, thereby achieving the control of the shield machine.Patent CN109779649A proposes a real-time shield tunneling axis deviation correction system based on big data. This system performs pre-processing operations such as outlier processing and data fusion on data collected from the construction site, including the shield machine cutterhead torque, cutterhead speed, propulsion speed, zoned jack stroke, total thrust, zoned hydraulic valve opening, zoned output oil pressure, horizontal / vertical deviation of the cutterhead / shield tail, shield slope angle, excavation face soil pressure, change in tunnel design axis coordinates, shield tail clearance, and excavation face soil information. The system then uses a long short-term memory (LSTM) self-learning model to train a trajectory strategy module. The trajectory strategy module determines the target position to which the shield machine will be adjusted for the next construction unit. The hydraulic attitude module outputs the required zoned hydraulic cylinder pressure values based on the target position to control the shield machine's directional adjustment. Furthermore, this method can optimize the trajectory strategy module based on new data generated at the work site, in order to achieve real-time directional adjustment of the shield axis based on big data. Patent CN108868807A proposes an intelligent control method for shield tunneling correction. This method first divides the shield's thrust cylinders into multiple zones and, based on the segment burial depth and historical shield construction data, determines the pressure distribution for each zone's thrust cylinders. Next, the controller extracts features from the real-time data of the shield's attitude and shield tail clearance, obtained by an automatic measurement system. This data is combined with a custom correction curve equation and correction distance formula to calculate a correction curve and shield advance distance. Based on the shield advance distance, combined with the burial depth and historical construction data, a random forest algorithm is used to determine the oil pressure adjustment for each zone's thrust cylinders. The pressure is then output to adjust the shield's attitude during the advance. After the shield machine advances a certain distance, its attitude extends to the next adjustment distance. The difference between the shield's attitude and the correction curve is calculated to obtain a correction assessment value, which is then incorporated into the shield's attitude adjustment. Furthermore, the output oil pressure and attitude changes during the shield's attitude adjustment process are stored as historical data, and this data is used to train a random forest prediction model every 10 rings. Patent CN112922609A uses collected shield tunneling and guidance sample data to train a gated recurrent unit (GRU) model, resulting in a shield machine posture prediction model. This posture prediction model processes the current tunneling and guidance data in real time, deriving the desired zone pressure and average propulsion speed output by the shield machine under that state. This model then controls the machine's propulsion, thereby enabling intelligent tunneling.Patent CN110195592A established a hybrid deep learning model WCNN-LSTM (Wide Convolutional Neural Networks-Long Short Term Memory) for predicting shield machine posture. The model takes the shield machine excavation parameters corresponding to the t+j moment or the t+1 to t+j time period as input, and the shield machine posture signal at the t+j moment or the t+1 to t+j time period as output. Therefore, it can predict the shield machine posture prediction value in the t+1 to t+j time period under the current input value, and can compare it with the design value to obtain the deviation. If the deviation exceeds the allowable range, the shield machine driver can adjust the input value applied to the shield machine at the t+j moment or the t+1 to t+j time period in advance, so as to complete the correction operation in advance before the shield machine excavation becomes inaccurate. A big data-driven shield machine attitude control method has also been developed. This method has two sub-models: one for setting the shield machine attitude control target, and the other for reflecting the relationship between shield propulsion cylinder pressure and shield machine posture deviation. These two sub-models work together to generate a shield machine attitude adjustment strategy, providing a reference for the shield machine driver's decision-making when performing directional maneuvers. All of the above-mentioned technologies are based on supervised learning. Furthermore, patent CN114019795A uses a reinforcement learning framework to construct a shield machine simulation correction environment that maps shield machine posture deviations to directional maneuvering instructions. It also establishes a shield machine correction decision model. Using an agent-based evaluation method and a value function network structure, the shield machine correction decision model is trained multiple times in the simulation environment to obtain the final model. The final model outputs the propulsion cylinder pressure values and cutterhead steering decisions for each shield machine section, providing the shield machine driver with reference and assisting him in directional maneuvers. A third category of methods is based on fuzzy control. These methods establish fuzzy rules based on human experience. These rules convert shield machine posture deviation information into directional cylinder control instructions for each section, achieving automatic shield machine attitude control. The current research on shield machine attitude control based on the fuzzy PID (Proportion Integration Differentiation) method uses a dual closed-loop feedback fuzzy PID control strategy to adjust the shield propulsion hydraulic cylinder speed to achieve accurate control of the shield tunneling trajectory. Patent CN113931648A proposes a control system for automatically adjusting the attitude of a shield machine. The system consists of a host computer, a slave computer, a shield machine attitude measurement system, a shield machine propulsion system, and feedback sensors. The shield machine attitude measurement system transmits the current attitude deviation value of the shield machine to the slave computer in real time. The host computer reads this data and key shield machine excavation parameters from the slave computer and calculates recommended pressure values for the shield machine's four zone propulsion cylinders based on the fuzzy control algorithm. The slave computer controls the shield machine propulsion system to perform corresponding actions based on these four recommended values, thereby achieving automatic directional adjustment of the shield machine.Patent CN106522973A proposes an automatic deviation correction system for a shield machine. This system uses a guide measurement device to collect the current operating status of the shield machine and transmits it to a computing device. The control device, programmed with IPC (Inter-Process Communication) software, converts the parameter values of the shield machine's current operating status stored in the computing device into control instructions based on fuzzy rules, thereby achieving automatic deviation correction for the shield machine. However, since the fuzzy rules of fuzzy control require manual formulation, this method is essentially based on human experience. Furthermore, patent CN113586075A proposes a system and method for automatically correcting the attitude of the shield machine axis relative to the tunnel axis. The system includes a deviation angle calculation unit and an angle correction control unit. The deviation angle calculation unit can collect and determine the horizontal and vertical deviation angles of the shield machine axis relative to the tunnel axis in real time. Based on these deviation angles, the system's adaptive module generates shield machine force adjustment strategies for adjusting the horizontal and vertical deviation angles. The angle correction control unit then controls the horizontal and vertical deflection of the shield machine. The system also includes a thrust correction unit and a propulsion control unit. The thrust correction unit monitors the actual total thrust of the shield machine and compares the actual horizontal and vertical components of the actual total thrust with the target components generated by the adaptive module to obtain the corresponding corrections. Finally, the propulsion control unit corrects the current actual total thrust based on the current corrections to control the shield machine's propulsion. However, this method fails to account for the impact of different geological conditions.

[0005] Among existing shield machine automatic steering technologies, logic-based rule-based approaches rely solely on shield machine posture deviation information to calculate shield propulsion cylinder pressure for shield steering / correction operations. These approaches fail to fully utilize other valuable information generated during shield machine operation, such as geological information. Furthermore, due to the complexity of geological conditions and the actual control process, logic-based rule-based shield steering technologies suffer from unstable control and difficulty adapting to complex and changing geology. Furthermore, the formulation of logic rules still relies on human experience, requiring manual intervention. Fuzzy control-based approaches also require manual formulation of fuzzy rules and are unable to adapt to varying geological conditions, resulting in limited intelligence. Machine learning-based approaches, on the other hand, suffer from the drawbacks of relying on extensive historical data training and poor generalization. Alternatively, they can only provide shield machine operators with decision-making assistance for shield steering, failing to completely replace manual steering operations. Consequently, existing shield machine automatic steering technologies are unable to overcome the numerous problems associated with manual operation. Summary of the Invention

[0006] To address the problems in the background technology, the present invention provides a shield machine intelligent steering system and control method based on model predictive control. The present invention comprises two parts: a shield machine intelligent steering system based on model predictive control and a control method. The shield machine intelligent steering system based on model predictive control mainly consists of four parts: a shield machine posture measurement module, a data acquisition module, an intelligent steering controller, and a hydraulic posture module. It can realize five functions: acquisition of geological information and shield machine geometric parameters, real-time acquisition of shield machine excavation parameters, real-time acquisition of shield machine posture data, control signal output, and shield machine posture change. Before the shield machine is started, geological information of the excavation area is obtained through geological exploration and stored in the shield machine's data acquisition module. At the same time, the shield machine's data acquisition module also stores shield machine geometric parameters and is responsible for the real-time acquisition of shield machine excavation parameters. The shield machine's posture data is mainly measured by the shield machine's shield posture measurement module. The intelligent steering controller is a controller that adopts a model predictive control algorithm and has high control accuracy, strong stability, and good robustness. In order to optimize the shield machine intelligent steering controller based on model predictive control, the present invention also builds a shield machine intelligent steering simulation system, which mainly consists of three parts: the shield machine intelligent steering controller based on model predictive control, the tunneling parameter prediction module and the simulation module. The model predictive controller includes three parts: a prediction model, a loss function and constraints, and an optimization algorithm. The prediction model is built using machine learning or deep neural networks, which can better express the complex, nonlinear, and multivariable shield-environment dynamic interaction process. The model predictive controller combines the prediction model and constraints to perform optimization solutions to obtain the optimal control sequence at the current moment. The tunneling parameter prediction module is used to receive parameters output from the model predictive controller and output corresponding predicted values to the simulation module, mainly providing some input parameters for the simulation module. The simulation module is built using machine learning or deep neural networks, mainly providing input and high-precision training simulation environment for the model predictive controller during the design phase, optimizing the design parameters of the model predictive controller and verifying its performance. The intelligent shield machine steering method based on model predictive control is as follows: at a set time point, the shield machine intelligent steering controller receives the shield machine's current excavation parameters, shield geometry parameters, current geological parameters, and the shield machine's current posture deviation measured by the shield machine posture measurement module from the shield machine data acquisition module. Based on these parameters, it generates a control input sequence that meets the constraints. Then, it uses the predictive model to perform a multi-step prediction of the shield machine's posture changes under the control input sequence. The prediction results are compared with the control target. Through continuous rolling optimization, the optimal control sequence is finally generated and output to the shield machine's hydraulic posture module. The hydraulic posture module adjusts the pressure of the corresponding partitioned propulsion cylinder based on the control signal, thereby controlling the movement of the corresponding partitioned propulsion cylinder to change the shield machine's posture.When the shield machine's position changes, the intelligent shield direction controller receives updated data from the shield machine's position measurement module and data acquisition module at the next set time. It then outputs the optimal pressure control sequence for each shield section's thrust cylinders to the shield's thrust hydraulic system. This continuous cycle enables intelligent adjustment of the shield machine's tunneling direction without human intervention.

[0007] The technical solution adopted in the present invention is:

[0008] 1. A shield machine intelligent direction adjustment system based on model predictive control, comprising:

[0009] The shield posture measurement module is used to collect the shield posture data of the shield machine during tunnel excavation and then output the current shield posture deviation value based on the shield posture data and the preset tunnel design axis.

[0010] The data acquisition module is used to collect and store the shield machine's current shield posture deviation value, shield excavation parameters, geological parameters and shield geometric parameters.

[0011] The model prediction controller is used to receive the current shield posture deviation value, shield tunneling parameters, geological parameters and shield geometric parameters stored in the data acquisition module and then output the pressure signal of the propulsion cylinder of each partition.

[0012] The hydraulic posture module is used to receive the pressure signals of the thrust cylinders of each partition and control the thrust cylinders of each partition of the shield machine to perform actions to continue tunnel excavation.

[0013] The shield posture deviation includes the horizontal deviation of the shield head, the horizontal deviation of the shield tail, the vertical deviation of the shield head, the vertical deviation of the shield tail and the pitch angle.

[0014] The number of shield propulsion cylinder partitions of the shield machine is 4, that is, the shield propulsion cylinders are divided into four groups: A, B, C, and D. The propulsion cylinders in group A are the propulsion cylinders for the right partition of the shield, the propulsion cylinders in group B are the propulsion cylinders for the lower partition of the shield, the propulsion cylinders in group C are the propulsion cylinders for the left partition of the shield, and the propulsion cylinders in group D are the propulsion cylinders for the upper partition of the shield.

[0015] The model prediction controller includes a prediction model, an optimization algorithm, a loss function, and constraints. Under the constraints, the optimization algorithm calculates the thrust cylinder pressure value for each subarea and outputs it to the prediction model. Under the constraints, the prediction model receives the current shield posture deviation value, current shield tunneling parameters (excluding the thrust cylinder pressure values for each subarea), current geological parameters, and shield geometric parameters as input, processes them, and outputs the predicted shield posture deviation value to the optimization algorithm. The loss function is simultaneously updated, and the cycle continues until the loss function is minimized. Ultimately, the optimal thrust cylinder pressure value for each subarea at the current moment and under the current state is output. The constraints constrain the output value of the optimization algorithm.

[0016] The prediction model is specifically a fully connected neural network (FCN) model, a convolutional neural network (CNN) model, a recurrent neural network (RNN) model or a long short-term memory network (LSTM) model.

[0017] The optimization algorithm is specifically Sequential Least Squares Quadratic Programming (SLSQP), trust-constrained optimization (trust-constr), Newton-Conjugate Gradient (Newton-CG), Ant Colony Optimization (ACO) or Genetic Algorithm (GA), etc., to speed up the speed of finding the optimal sequence of propulsion cylinder pressure values in each partition, that is, to accelerate convergence.

[0018] The optimization algorithm is a key component of the shield machine's intelligent steering controller based on model predictive control. Its purpose is to find the optimal sequence of thrust cylinder pressure values for each partition to minimize the loss function. The goal of the optimization algorithm is to select the thrust cylinder pressure values for each partition at each moment so that the shield machine's attitude deviation, as output by the prediction model, is minimized over a series of future time steps, thereby minimizing the loss function.

[0019] The loss function is as follows:

[0020] Loss function = a × shield head horizontal deviation 2 +b×vertical deviation of shield head 2 +c×horizontal deviation of shield tail 2 +d×

[0021] Shield tail vertical deviation 2 +e×pitch angle 2

[0022] Where a, b, c, d and e are the first, second, third, fourth and fifth loss coefficients respectively.

[0023] The constraints mentioned above include the actual maximum and minimum allowable pressures of the propulsion cylinders of each partition, thereby constraining the range of propulsion cylinder pressure values of each partition output by the optimization algorithm. It should be noted that the minimum pressure value of the propulsion cylinder of the partition below the shield is generally much higher than the minimum value of other partitions to prevent the shield from crashing. At the same time, the constraints also include the maximum allowable change rate of the shield propulsion cylinder pressure, thereby constraining the change rate of the propulsion cylinder pressure values of each partition output by the optimization algorithm. That is, at the current moment, the difference between the shield propulsion cylinder pressure value of each partition output by the optimization algorithm and the corresponding shield propulsion cylinder pressure value of each partition transmitted by the shield data acquisition module cannot be greater than the actual shield propulsion cylinder pressure change value allowed.

[0024] Compared to other controllers, the MPC-based intelligent shield steering controller can account for the various constraints of actual shield operation. It can perform multi-step predictions using a trained predictive model that reflects the dynamic interaction between the shield and the environment. It can also compare the prediction results with the control objectives to generate an optimal sequence of thrust cylinder pressure values for each shield section. This allows it to better resist the effects of external interference on the shield, significantly enhancing the shield's intelligence. Furthermore, this MPC uses data-based models such as machine learning or deep neural network models as its prediction model, effectively addressing the difficulty of establishing an accurate shield rock-machine interaction model using traditional physical modeling methods and increasing the accuracy of the prediction model.

[0025] 2. A control method for a shield machine intelligent steering system based on model predictive control, comprising:

[0026] Step 1) Use the shield tunneling intelligent steering simulation system to design a model predictive controller, and build an intelligent steering system based on the model predictive controller.

[0027] Step 2) When the shield machine is performing tunnel excavation, the intelligent direction adjustment system collects the shield machine's shield posture data, shield excavation parameters, geological parameters and shield geometric parameters in real time, and outputs the pressure signals of each section propulsion cylinder of the shield machine in real time based on the preset tunnel design axis. The pressure signals of each section propulsion cylinder are used to control the execution of the action of each section propulsion cylinder of the shield machine to continue tunnel excavation.

[0028] In the step 1), the shield machine intelligent direction adjustment simulation system includes a tunneling parameter prediction module and a simulation module. The shield machine intelligent direction adjustment simulation system is used to design a model prediction controller, specifically as follows:

[0029] The initial model predictive controller is constructed using the prediction model, optimization algorithm, loss function and constraint conditions. The current simulated shield posture deviation value, shield tunneling parameters, geological parameters and shield geometric parameters are input into the model predictive controller. The model predictive controller outputs the predicted shield tunneling cylinder pressure values of each partition to the tunneling parameter prediction module and simulation module respectively. At the same time, the shield tunneling parameters and geological parameters other than the shield tunneling cylinder pressure values of each partition are input into the tunneling parameter prediction module. The tunneling parameter prediction module outputs the predicted shield total thrust, thrust speed, cutterhead torque and cutterhead speed and outputs them to the simulation module. At the same time, the current simulated shield posture deviation value, shield tunneling parameters other than the shield tunneling cylinder pressure values of each partition, total thrust, thrust speed, cutterhead torque and cutterhead speed, geological parameters and shield geometric parameters are input into the simulation module. The simulation module outputs the current simulated shield posture deviation value to itself and the model predictive controller respectively to complete the closed loop, realize the training and optimization of the model predictive controller, and finally design the final model predictive controller.

[0030] The shield tunneling parameters include the thrust cylinder pressure value of each shield section, the total thrust force of the shield, the thrust speed, the cutter head rotation speed, the cutter head torque, the grouting pressure and the articulation pressure, etc.

[0031] Geological parameters include elevation, natural density, soil particle density, internal friction angle, cohesion, natural compressive strength, saturated compressive strength, permeability, Poisson's ratio, elastic model and shear modulus.

[0032] The geometric parameters of the shield machine include cutterhead diameter, opening ratio, total length of the main machine, front shield diameter, middle shield diameter, tail shield diameter, propulsion cylinder diameter, propulsion stroke, articulation hydraulic cylinder diameter and articulation stroke, etc.

[0033] The shield machine intelligent direction adjustment system and control method based on model predictive control described in the present invention can realize unmanned autonomous direction adjustment of the shield machine, greatly improving the intelligence level of the shield machine. The shield machine intelligent direction adjustment system based on model predictive control and its design method have the following technical advantages and innovations: First, the model predictive intelligent direction adjustment controller is used to replace the shield machine driver to perform decision-making tasks and output control instructions during the direction adjustment of the shield machine, without the need for shield machine driver operation, effectively avoiding a series of drawbacks caused by manual operation of the shield machine direction adjustment, such as misoperation, excessive adjustment and untimely adjustment; second, the model predictive control algorithm takes into account multiple constraints and can be adjusted according to changes in the shield system. It can use the system model to perform multi-step predictions and compare the prediction results with the control target to generate the optimal control strategy, which can effectively handle multi-input and multi-output shield systems and solve the problems of shield machines. The coupling relationship between the thrust cylinder pressures in each section of the thrust cylinder and the shield machine's posture deviations during the maneuvering process enables efficient control of complex, nonlinear shield dynamic systems. This allows for better adaptation to system changes and external disturbances, making it suitable for controlling complex shield systems and significantly enhancing the shield machine's intelligence. Thirdly, the prediction model in the model predictive controller is constructed using machine learning or deep neural networks. Because machine learning or deep neural networks can express arbitrarily complex nonlinear mapping relationships, they can better capture the nonlinear dynamic behavior of the shield system. Furthermore, machine learning or deep neural network models can easily handle multivariable systems, overcoming the modeling difficulties and low precision of traditional physical modeling methods. Furthermore, machine learning or deep neural network models can be learned and updated online in real time, effectively improving the accuracy and applicability of the prediction model. Fourthly, the prediction model, simulation module, and tunneling parameter prediction module in the model predictive controller are trained using historical tunneling parameters from multiple shield machines, improving the accuracy, applicability, and precision of each module. Fifthly, the shield machine intelligent maneuvering simulation system, built based on the simulation module, can optimize not only the model predictive controller but also other controllers, expanding the scope of application of the shield machine intelligent maneuvering controller design method. In summary, the present invention has great application potential in the shield intelligent direction adjustment system and its design method.

[0034] The beneficial effects of the present invention are:

[0035] The intelligent shield machine turning system proposed in the present invention utilizes a model predictive controller to replace the shield machine driver in shield machine turning operations, effectively solving a series of problems that occur when manually operating the shield machine turning, such as misoperation, excessive adjustment, and untimely adjustment. Compared with other control algorithms, the model predictive control algorithm can consider multiple constraints and select the optimal control strategy through future predictions. It can effectively handle the coupling relationship between various control variables and various state variables during the shield machine turning process, and better adapt to system changes and external interference, greatly improving the intelligence level of the shield machine. At the same time, machine learning or deep neural network methods are used to build the simulation module and the prediction model in the model predictive controller, and a large amount of historical tunneling data from different shield machines is used to train the model, effectively solving the disadvantage that it is difficult or impossible to accurately establish or establish complex shield systems using mathematical models. In addition, the proposed intelligent shield machine turning simulation system can be used to design and optimize a variety of different controllers, greatly expanding the scope of application. In summary, the present invention has great application potential in intelligent shield machine turning systems and control methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 This is the structure diagram of the shield machine intelligent steering controller based on model predictive control;

[0037] Figure 2 This is the schematic diagram of the shield machine intelligent direction adjustment simulation system;

[0038] Figure 3 Design flow chart for simulation module;

[0039] Figure 4 This is a schematic diagram of the control method for the shield machine intelligent steering system based on model predictive control. DETAILED DESCRIPTION

[0040] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0041] like Figure 4As shown, the present invention's intelligent shield adjustment system based on model predictive control includes: a shield posture measurement module for collecting shield posture data of the shield machine during tunnel excavation and then outputting the shield machine's current posture deviation value based on the shield posture data and the preset tunnel design axis; a data acquisition module for collecting and storing the shield machine's current posture deviation value, shield excavation parameters, geological parameters, and shield geometric parameters; a model predictive controller for receiving the shield machine's current posture deviation value, shield excavation parameters, geological parameters, and shield geometric parameters stored in the data acquisition module and then outputting pressure signals for each section propulsion cylinder; and an oil pressure posture module for receiving the pressure signals for each section propulsion cylinder and controlling each section propulsion cylinder of the shield machine to execute actions to continue tunnel excavation. Shield posture deviations include shield head horizontal deviation, shield tail horizontal deviation, shield head vertical deviation, shield tail vertical deviation, and pitch angle.

[0042] The number of shield thrust cylinder partitions of the shield machine is 4, that is, the shield thrust cylinders are divided into four groups: A, B, C, and D. The thrust cylinders in group A are the thrust cylinders in the right partition of the shield, the thrust cylinders in group B are the thrust cylinders in the lower partition of the shield, the thrust cylinders in group C are the thrust cylinders in the left partition of the shield, and the thrust cylinders in group D are the thrust cylinders in the upper partition of the shield.

[0043] like Figure 1 As shown, the model predictive controller includes a prediction model, an optimization algorithm, a loss function, and constraints. Under the constraints, the optimization algorithm calculates the thrust cylinder pressure value for each partition and outputs it to the prediction model. The prediction model receives the current shield posture deviation value, the current shield tunneling parameters (excluding the thrust cylinder pressure value for each partition), the current geological parameters, and the shield geometry parameters, processes them, and outputs the predicted shield posture deviation value to the optimization algorithm. Simultaneously, the loss function is updated, and the cycle continues until the loss function is minimized. Ultimately, the optimal thrust cylinder pressure value for each partition is output for the current moment and state. The constraints constrain the output value of the optimization algorithm.

[0044] The prediction model is specifically a fully connected neural network model, a convolutional neural network model, a recurrent neural network model or a long short-term memory network model.

[0045] Optimization algorithms, such as sequential least squares quadratic programming, trust region constrained optimization, Newton-conjugate gradient method, ant colony algorithm, or genetic algorithm, are used to accelerate the search for the optimal sequence of thrust cylinder pressure values for each sub-area, i.e., to accelerate convergence. The optimization algorithm is a key component of the shield machine intelligent steering controller based on model predictive control. Its function is to find the optimal sequence of thrust cylinder pressure values for each sub-area to minimize the loss function. The goal of the optimization algorithm is to select the thrust cylinder pressure value for each sub-area at each moment so that the shield machine attitude deviation value output by the prediction model is minimized over a series of future time steps, thereby minimizing the loss function.

[0046] The loss function is as follows:

[0047] Loss function = a × shield head horizontal deviation 2 +b×vertical deviation of shield head 2 +c×horizontal deviation of shield tail 2 +d×

[0048] Shield tail vertical deviation 2 +e×pitch angle 2

[0049] Where a, b, c, d and e are the first, second, third, fourth and fifth loss coefficients respectively.

[0050] The constraints include the actual maximum and minimum allowable pressures of the thrust cylinders in each partition, thereby constraining the range of thrust cylinder pressure values output by the optimization algorithm. It should be noted that the minimum thrust cylinder pressure of the partition below the shield is generally much higher than the minimum values of other partitions to prevent the shield from crashing. At the same time, the constraints also include the maximum allowable rate of change of the thrust cylinder pressure in the shield, thereby constraining the rate of change of the thrust cylinder pressure values output by the optimization algorithm in each partition. That is, at the current moment, the difference between the thrust cylinder pressure values output by the optimization algorithm for each partition of the shield and the corresponding thrust cylinder pressure values for each partition transmitted by the shield data acquisition module cannot be greater than the actual allowable thrust cylinder pressure change value of the shield.

[0051] Compared to other controllers, the MPC-based intelligent shield steering controller can account for the various constraints of actual shield operation. It can perform multi-step predictions using a trained predictive model that reflects the dynamic interaction between the shield and the environment. It can also compare the prediction results with the control objectives to generate an optimal sequence of thrust cylinder pressure values for each shield section. This allows it to better resist the effects of external interference on the shield, significantly enhancing the shield's intelligence. Furthermore, this MPC uses data-based models such as machine learning or deep neural network models as its prediction model, effectively addressing the difficulty of establishing an accurate shield rock-machine interaction model using traditional physical modeling methods and increasing the accuracy of the prediction model.

[0052] The control method of the shield intelligent direction adjustment system based on model predictive control of the present invention includes:

[0053] Step 1) Use the shield tunneling intelligent steering simulation system to design a model predictive controller, and build an intelligent steering system based on the model predictive controller.

[0054] like Figure 2 As shown in step 1), the shield machine intelligent direction adjustment simulation system includes a tunneling parameter prediction module and a simulation module. The model prediction controller is designed using the shield machine intelligent direction adjustment simulation system, as follows:

[0055] The initial model predictive controller is constructed using the prediction model, optimization algorithm, loss function and constraint conditions. The current simulated shield posture deviation value, shield tunneling parameters, geological parameters and shield geometric parameters are input into the model predictive controller. The model predictive controller outputs the predicted shield tunneling cylinder pressure values of each partition to the tunneling parameter prediction module and simulation module respectively. At the same time, the shield tunneling parameters and geological parameters other than the shield tunneling cylinder pressure values of each partition are input into the tunneling parameter prediction module. The tunneling parameter prediction module outputs the predicted shield total thrust, thrust speed, cutterhead torque and cutterhead speed and outputs them to the simulation module. At the same time, the current simulated shield posture deviation value, shield tunneling parameters other than the shield tunneling cylinder pressure values of each partition, total thrust, thrust speed, cutterhead torque and cutterhead speed, geological parameters and shield geometric parameters are input into the simulation module. The simulation module outputs the current simulated shield posture deviation value to itself and the model predictive controller respectively to complete the closed loop, realize the training and optimization of the model predictive controller, and finally design the final model predictive controller.

[0056] Shield tunneling parameters include the thrust cylinder pressure values of each shield section, total shield thrust force, thrust speed, cutterhead speed, cutterhead torque, grouting pressure and articulation pressure, etc.

[0057] Geological parameters include elevation, natural density, soil particle density, internal friction angle, cohesion, natural compressive strength, saturated compressive strength, permeability, Poisson's ratio, elastic model and shear modulus.

[0058] The geometric parameters of the shield machine include cutterhead diameter, opening ratio, total length of the main machine, front shield diameter, middle shield diameter, tail shield diameter, propulsion cylinder diameter, propulsion stroke, articulation hydraulic cylinder diameter and articulation stroke, etc.

[0059] Step 2) When the shield machine is performing tunnel excavation, the intelligent direction adjustment system collects the shield machine's shield posture data, shield excavation parameters, geological parameters and shield geometric parameters in real time, and outputs the pressure signals of each section propulsion cylinder of the shield machine in real time based on the preset tunnel design axis. The pressure signals of each section propulsion cylinder are used to control the execution of the action of each section propulsion cylinder of the shield machine to continue tunnel excavation.

[0060] The system of the present invention can realize five functions: acquisition of geological information and shield geometric parameters, real-time acquisition of shield excavation parameters, real-time acquisition of shield posture data, control signal output, and shield machine posture change. Among them, the real-time acquisition of shield posture data is realized by the shield posture measurement module; the storage of geological information and shield geometric parameters, real-time acquisition of shield excavation parameters, and acquisition of shield posture deviation in the shield posture measurement module are realized by the data acquisition module; the output of control signals to the hydraulic posture module is realized by the intelligent direction adjustment controller based on model predictive control; and the change of shield machine posture is realized by the hydraulic posture module controlling the thrust cylinders of each shield section to perform corresponding actions. With respect to the traditional method of manually adjusting the shield machine's excavation direction by the shield driver, the shield driver first needs to judge the shield machine's offset direction based on the shield posture deviation on the shield machine control panel at regular intervals, and then the shield driver relies on experience to adjust the pressure of the thrust cylinders of the corresponding shield section, thereby completing the adjustment of the shield machine's excavation direction. The intelligent shield direction adjustment system based on model predictive control proposed in the present invention receives the current shield excavation parameters, current shield posture deviation, shield geometric parameters and current geological information transmitted from the shield data acquisition module at a set time point through the intelligent direction adjustment controller, and outputs corresponding control signals to the oil pressure posture module to adjust the pressure of the thrust cylinder of the corresponding partition, that is, the intelligent direction adjustment controller is used instead of the shield driver to complete the adjustment of the shield machine's excavation direction.

[0061] Real-time shield machine posture data collection is accomplished through the shield machine's posture measurement module. This module compares the measured shield posture data with the preset tunnel design axis and, after calculation, transmits the shield posture deviation to the shield machine's data acquisition module. The shield posture deviations measured by the posture measurement module and converted to the shield data acquisition module primarily include horizontal deviation, vertical deviation, horizontal orientation, vertical orientation, and pitch angle. Real-time collection of shield machine excavation parameters is accomplished through the shield machine's data acquisition module. Furthermore, the shield machine's data acquisition module stores shield geometry parameters and geological information about the excavation area obtained through geological surveys before the shield machine begins. The shield machine's data acquisition module primarily transmits the shield machine's current excavation parameters, posture deviation values, geological parameters of the current excavation area, and shield geometry parameters to the intelligent direction control controller. After receiving the current tunneling parameters of the shield machine, the current posture deviation of the shield machine, the shield machine's geometric parameters, and the current geological parameters, the intelligent direction adjustment controller calculates and outputs control signals to the hydraulic control valves of each partition of the shield machine's oil pressure posture module, adjusting the pressure of the propulsion cylinders in each partition to adjust the tunneling direction of the shield machine. The entire control process is a closed-loop control, which effectively improves the accuracy of the shield machine's direction adjustment and suppresses the disturbance of the geological load on the tunneling direction. In addition, this intelligent shield direction adjustment method still retains a manual adjustment channel to facilitate the shield machine driver's operation in emergency situations. In addition, the model prediction intelligent controller used in this intelligent shield direction adjustment method can also be replaced with an intelligent controller equipped with other algorithms.

[0062] In order to optimize the shield machine intelligent steering controller based on model predictive control, the present invention also builds a shield machine intelligent steering simulation system, such as Figure 2 As shown in Figure 1, it mainly consists of three parts: a shield intelligent direction adjustment controller based on model predictive control, a tunneling parameter prediction module, and a simulation module. The simulation module in the shield intelligent direction adjustment simulation system is mainly used to reflect the dynamic interaction process between the shield and the environment. Its design flow chart is shown in Figure 1. Figure 3 As shown in the figure. The first step is to obtain the total construction data of multiple shield machines, and then preprocess these data, including filtering out shield shutdown data, processing abnormal data, interpolating missing values, and separating the data of the initial stage and the stable stage of the shield operation process. Then, the input and output features of the simulation module are constructed and the parameters not used in the model are deleted. The input features of the simulation module include the tunneling parameters of the shield machine, the geometric parameters of the shield machine, the geological parameters and the shield posture deviation. The output feature of the simulation module is the posture deviation of the shield machine. Then, a machine learning model or a neural network model is used to build a simulation module, and the processed data is used to train the model. The model parameters are continuously modified until the simulation module's prediction accuracy of the shield posture deviation meets the specified requirements.

[0063] A tunneling parameter prediction module is built using a machine learning model or a deep neural network model. The input data for this module consists of geological parameters and shield tunneling parameters. Its output features include the shield's total thrust, thrust speed, cutterhead torque, and cutterhead speed. Using the same process as the design simulation module, the model is trained using processed historical tunneling data from multiple shield machines. Model parameters are continuously modified until the required accuracy is achieved. The output data from the tunneling parameter prediction module and the output data from the model prediction controller serve as part of the simulation module's input.

[0064] The intelligent shield direction adjustment controller based on model predictive control mainly includes three parts: prediction model, loss function and constraint conditions, and optimization algorithm. Unlike conventional model predictive controllers, the prediction model in the present invention is a rock-machine interaction model built using a machine learning model or a deep neural network model, which avoids the problem of difficulty in establishing an accurate shield rock-machine interaction model using traditional physical modeling methods. Among them, the input features of the prediction model include shield machine excavation parameters, shield machine geometric parameters, shield posture deviation and geological parameters, and the output feature of the prediction model is shield posture deviation. The prediction model is trained and optimized using the processed historical excavation data of multiple shield machines. Unlike the simulation module, the prediction model has higher requirements for accuracy. In the shield intelligent direction adjustment simulation system, the input of the model predictive controller is the shield posture deviation output by the simulation module and the geological parameters, shield excavation parameters, and shield geometric parameters in the historical shield excavation data, and the output is the thrust cylinder pressure of each partition of the shield machine. It is important to note that the shield cylinder pressures in each section of the shield machine, as input features to the prediction model, are not the actual pressures input to the model predictive controller. Instead, they are generated by the optimization algorithm. Within the model predictive controller, the prediction model uses the current input features (excluding the shield cylinder pressures) and the future cylinder pressures calculated by the optimization algorithm to predict the shield machine's posture deviation within a set timeframe, thereby performing optimal control. The optimization algorithm generates the cylinder pressures for each section of the shield machine, which is related to the loss function and constraints. The constraints of the model predictive controller are the minimum and maximum allowable cylinder pressures and the range of pressure change allowed within a set timeframe, i.e., the pressure change rate. The current cylinder pressures input to the model predictive controller are primarily used to determine whether the cylinder pressures generated by the optimization algorithm meet the pressure change rate constraints. Furthermore, the minimum cylinder pressure in the shield section below the shield machine differs from that in other sections. It has a pressure significantly greater than 0 bar and not less than a set limit to prevent the shield machine from crashing. The constraint function of the model predictive controller is defined as the square of the deviation values for each shield machine's posture, as predicted by the prediction model, multiplied by the sum of the corresponding coefficients. At each moment, the model predictive controller determines the optimal shield propulsion cylinder pressure for that given moment, based on criteria such as minimizing the shield machine's posture deviation and ensuring stable pressure changes in each section of the shield propulsion cylinder. The model predictive controller's optimization algorithm combines the prediction model, constraints, and loss function to achieve an optimal solution, ultimately determining the optimal control value for the shield propulsion cylinder pressure at that moment.

[0065] Taking a shield machine with a four-zone shield thrust cylinder as an example, the working principle of the shield intelligent steering simulation system optimizing the shield intelligent steering controller based on model predictive control is explained. The process of optimizing the shield intelligent steering controller based on model predictive control by the shield intelligent steering simulation system includes three steps. The first step is to prepare a set of shield machine geometric parameters, geological parameters, shield tunneling parameters, and shield posture deviations from historical tunneling data of multiple consecutive tunneling sections that have not participated in the training prediction model, tunneling parameter prediction module, and simulation module as the initial input values of the model predictive controller. The shield machine geometric parameters, geological parameters, shield tunneling parameters, and shield tunneling parameters other than the thrust cylinder pressure, total thrust force, thrust speed, cutterhead torque, and cutterhead speed of each zone (hereinafter referred to as other tunneling parameters 2) are then used as partial initial inputs to the simulation module. Furthermore, the geological parameters and shield tunneling parameters other than the thrust cylinder pressure of each zone (hereinafter referred to as other tunneling parameters 1) are also required as partial initial inputs to the tunneling parameter prediction module. To improve the model predictive controller's interference tolerance, noise is often added to the simulation module's input features. In the second step, the model predictive controller receives inputs consisting of shield posture deviations, shield excavation parameters, shield geometry, and geological parameters from historical tunneling data. These shield posture deviations, along with all other parameters except for the thrust cylinder pressures in each shield section, shield geometry, and geological parameters, are then passed to the prediction model as partial inputs. The thrust cylinder pressures in each shield section are also passed to constraints, which are used to limit the range of thrust cylinder pressures generated by the optimization algorithm. The optimization algorithm generates thrust cylinder pressures for each shield section based on the shield posture deviations, a loss function, and the constraints. These pressures are then passed to the prediction model. The prediction model then generates shield posture deviations, which are used to update the loss function. Within a set prediction step, the optimization algorithm and prediction model in the model predictive controller continuously update the predicted control variables and predicted state variables. Combining constraints and loss functions, they determine the optimal control output—the current optimal thrust cylinder pressure for each shield machine section—and transmit this output to the tunneling parameter prediction module and simulation module. The tunneling parameter prediction module, after receiving geological parameters and other tunneling parameters (1) from the shield machine's historical tunneling data, and the thrust cylinder pressures for each shield section output by the model predictive controller, outputs the predicted total thrust force, thrust speed, cutterhead torque, and cutterhead speed to the simulation module. The simulation module, after receiving the thrust cylinder pressures for each shield section output by the model predictive controller, the total thrust force, thrust speed, cutterhead torque, and cutterhead speed output by the tunneling parameter prediction module, and shield machine geometric parameters, geological parameters, and other tunneling parameters (2) from the shield machine's historical tunneling data, outputs the simulated shield head horizontal deviation, shield head vertical deviation, shield tail horizontal deviation, shield tail vertical deviation, and pitch angle to the model predictive controller. The output values are then fed back to the simulation module as partial input for the next simulation run.Based on the updated input, the model predictive controller optimizes the optimal thrust cylinder pressure for each section of the shield machine at the current moment, and then outputs it to the tunneling parameter prediction module and simulation module. This constitutes a closed-loop process, completing the establishment of the model predictive controller training environment. The entire closed-loop process will continue to run until the prepared data is used up or all posture deviations of the shield machine tend to 0. The third step is to continuously optimize the proportion of each indicator in the model predictive controller loss function and the model predictive controller adjustment frequency according to the actual operation rules of the shield machine and the time from the controller outputting the control signal to the shield machine thrust cylinder to perform the corresponding action, until all posture deviations of the shield machine tend to 0 and the shield machine thrust cylinder pressure change and adjustment frequency meet the requirements, while avoiding serpentine motion of the shield machine.

[0066] In addition, in order to improve the accuracy of the prediction model, a tunneling parameter prediction module can also be added inside the model prediction controller to provide the prediction model with corresponding parameters such as the total shield propulsion force, propulsion speed, cutterhead torque and cutterhead speed when the predicted shield propulsion cylinder pressure changes.

[0067] When a shield machine intelligent direction adjustment controller that meets the conditions is deployed on the actual shield machine, it can realize intelligent direction adjustment of the shield machine by cooperating with the shield machine posture measurement module, data acquisition module and oil pressure posture module.

[0068] The principle diagram of the shield machine intelligent direction adjustment system control method based on model predictive control is as follows: Figure 4 As shown in Figure 2, the model-predictive control-based intelligent shield steering controller receives the current shield excavation parameters, current shield posture deviation, shield geometry, and current geological information from the shield data acquisition module at a set time. Based on these parameters, it generates a pressure sequence for each shield section's thrust cylinder that meets the constraints. The prediction model then performs a multi-step prediction of the shield machine's posture changes under the influence of each section's thrust cylinder pressure sequence. The prediction results are compared with the control target. Through continuous rolling optimization, the optimal shield section's thrust cylinder pressure sequence is ultimately generated and output to the shield machine's hydraulic posture module. The hydraulic posture module adjusts the pressure of the corresponding section's thrust cylinder based on the control signal, thereby controlling the movement of the corresponding section's thrust cylinder to change the shield machine's posture. When the shield machine's posture changes, the intelligent steering controller receives updated data from the shield posture measurement module and data acquisition module at the next set time. It then outputs the optimal shield section's thrust cylinder pressure control sequence to the shield machine's hydraulic posture module. This continuous cycle enables the intelligent adjustment of the shield machine's excavation direction without human intervention.

[0069] In addition, the input features of the prediction model of the shield machine intelligent steering controller based on model predictive control are all provided by the shield machine data acquisition module, enabling the prediction model to predict the shield machine posture deviation more accurately.

[0070] This method takes a shield machine with a shield propulsion cylinder divided into four zones as an example, but it is not only applicable to shield machines with a shield propulsion cylinder divided into four zones. It is also applicable to shield machines with other partition types, such as six zones. It only requires changing the input structure of the model predictive controller prediction model, the tunneling parameter prediction module, and the simulation module.

[0071] In addition, the controller design method of this shield intelligent steering system is not limited to designing controllers equipped with model predictive control algorithms, but is also applicable to controllers equipped with other intelligent algorithms. It is only necessary to change the corresponding input characteristics of the controller according to the properties of the installed control algorithm.

Claims

1. A shield machine intelligent direction adjustment system based on model predictive control, characterized in that: include: The shield posture measurement module is used to collect the shield posture data of the shield machine during tunnel excavation and then output the current shield posture deviation value based on the shield posture data and the preset tunnel design axis; The data acquisition module is used to collect and store the shield machine's current posture deviation value, shield tunneling parameters, geological parameters and shield geometric parameters; The model prediction controller is used to receive the shield's current posture deviation value, shield tunneling parameters, geological parameters and shield geometric parameters stored in the data acquisition module and then output the pressure signal of each partition propulsion cylinder; The hydraulic posture module is used to receive the pressure signals of the thrust cylinders of each section and control the thrust cylinders of each section of the shield machine to execute actions to continue tunnel excavation; The model prediction controller includes a prediction model, an optimization algorithm, a loss function and constraint conditions. The optimization algorithm calculates the pressure value of the propulsion cylinder of each partition under the constraint conditions and outputs it to the prediction model. The prediction model receives the input of the current shield posture deviation value, the current shield tunneling parameters except the pressure value of the propulsion cylinder of each partition of the shield, the current geological parameters and the shield geometric parameters, and outputs the predicted shield posture deviation value to the optimization algorithm after processing. At the same time, the loss function is updated, and the cycle is continuously repeated until the loss function reaches the minimum, and finally the optimal pressure value of the propulsion cylinder of each partition at the current moment and current state is output; The loss function is as follows: Loss function = a × shield head horizontal deviation 2 +b×vertical deviation of shield head 2 +c×horizontal deviation of shield tail 2 +d×vertical deviation of shield tail 2 +e×pitch angle 2 Where a, b, c, d and e are the first, second, third, fourth and fifth loss coefficients respectively; The constraints include the actual maximum and minimum allowable pressures of each partitioned propulsion cylinder, thereby constraining the range of pressure values of each partitioned propulsion cylinder output by the optimization algorithm; at the same time, the constraints also include the maximum allowable change rate of the shield propulsion cylinder pressure, thereby constraining the change rate of the pressure values of each partitioned propulsion cylinder output by the optimization algorithm.

2. The shield machine intelligent direction adjustment system based on model predictive control according to claim 1 is characterized by: The shield posture deviation includes the horizontal deviation of the shield head, the horizontal deviation of the shield tail, the vertical deviation of the shield head, the vertical deviation of the shield tail and the pitch angle; The number of shield propulsion cylinder partitions of the shield machine is 4, that is, the shield propulsion cylinders are divided into four groups: A, B, C, and D. The propulsion cylinders in group A are the propulsion cylinders for the right partition of the shield, the propulsion cylinders in group B are the propulsion cylinders for the lower partition of the shield, the propulsion cylinders in group C are the propulsion cylinders for the left partition of the shield, and the propulsion cylinders in group D are the propulsion cylinders for the upper partition of the shield.

3. The shield machine intelligent direction adjustment system based on model predictive control according to claim 1 is characterized by: The prediction model is specifically a fully connected neural network model, a convolutional neural network model, a recurrent neural network model or a long short-term memory network model.

4. The shield machine intelligent direction adjustment system based on model predictive control according to claim 1 is characterized by: The optimization algorithm is specifically sequential least squares quadratic programming, trust region constrained optimization, Newton-conjugate gradient method, ant colony algorithm or genetic algorithm.

5. The control method of the shield intelligent direction adjustment system based on model predictive control according to any one of claims 1 to 4, characterized in that: include: Step 1) Design a model predictive controller using the shield tunneling intelligent steering simulation system, and build an intelligent steering system based on the model predictive controller; Step 2) While the shield machine is tunneling, the intelligent steering system collects real-time shield posture data, tunneling parameters, geological parameters, and shield geometry parameters. Combined with the preset tunnel design axis, it outputs real-time pressure signals for each section of the shield machine's thrust cylinders. These pressure signals are used to control the actions of each section of the shield machine's thrust cylinders to continue tunneling.

6. The control method of the shield machine intelligent direction adjustment system based on model predictive control according to claim 5 is characterized by: In the step 1), the shield machine intelligent direction adjustment simulation system includes a tunneling parameter prediction module and a simulation module. The shield machine intelligent direction adjustment simulation system is used to design a model prediction controller, specifically as follows: The initial model predictive controller is constructed using the prediction model, optimization algorithm, loss function and constraint conditions. The current simulated shield posture deviation value, shield tunneling parameters, geological parameters and shield geometric parameters are input into the model predictive controller. The model predictive controller outputs the predicted shield tunneling cylinder pressure values of each partition to the tunneling parameter prediction module and simulation module respectively. At the same time, the shield tunneling parameters and geological parameters other than the shield tunneling cylinder pressure values of each partition are input into the tunneling parameter prediction module. The tunneling parameter prediction module outputs the predicted shield total thrust, thrust speed, cutterhead torque and cutterhead speed and outputs them to the simulation module. At the same time, the current simulated shield posture deviation value, shield tunneling parameters other than the shield tunneling cylinder pressure values of each partition, total thrust, thrust speed, cutterhead torque and cutterhead speed, geological parameters and shield geometric parameters are input into the simulation module. The simulation module outputs the current simulated shield posture deviation value to itself and the model predictive controller respectively to complete the closed loop, realize the training and optimization of the model predictive controller, and finally design the final model predictive controller.

7. The control method of the shield machine intelligent direction adjustment system based on model predictive control according to claim 5 is characterized by: The shield tunneling parameters include the thrust cylinder pressure value of each shield section, the total thrust force of the shield, the thrust speed, the cutterhead speed, the cutterhead torque, the grouting pressure and the articulation pressure; Geological parameters include elevation, natural density, soil particle density, internal friction angle, cohesion, natural compressive strength, saturated compressive strength, permeability, Poisson's ratio, elastic model and shear modulus; The geometric parameters of the shield machine include cutterhead diameter, opening ratio, total length of the main machine, front shield diameter, middle shield diameter, tail shield diameter, propulsion cylinder diameter, propulsion stroke, articulation hydraulic cylinder diameter and articulation stroke.

Images