Data-driven model modeling method and control system suitable for EPB shield tunneling machine soil pressure system

Abstract

The invention discloses a data driving model modeling method and control system suitable for an EPB shield tunneling machine soil pressure system, and relates to the technical field of tunnel construction equipment control, and the method comprises the steps: collecting and preprocessing the historical operation data of the shield tunneling machine soil pressure system, and defining the system state of the shield tunneling machine soil pressure system; establishing an approximate Koopman model in a finite-dimensional observation space based on the observation function, and establishing a linear predictor; introducing an integrator and an augmented state, constructing an augmented state matrix, and calculating a multi-step state and output prediction based on a prediction time domain; defining constraint conditions for operation data, setting a system prediction objective function, calculating a closed-loop system stability inequality, and combining the output to generate a linear matrix inequality; and performing rolling iterative optimization on the parameters of the earth pressure system of the shield tunneling machine by adopting an interior point method. When an earth pressure system is subjected to external bounded disturbance, the earth pressure can still converge to an expected value, the possible risk of collapse or upheaval of the tunnel face is reduced, and economic losses are reduced.

Description

Technical Field

[0001] This invention relates to the field of tunnel construction equipment control technology, and in particular to a data-driven modeling method and control system applicable to the earth pressure system of EPB shield tunneling machines. Background Technology

[0002] A tunnel boring machine (TBM) is a specialized piece of equipment for tunnel construction that integrates mechanical, electrical, hydraulic, sensor, and information technology. Modern TBMs cut into the soil using a closed cutterhead, stabilizing the excavation face using slurry balance or earth pressure balance technology, and simultaneously completing muck transport, segment assembly, and tunnel lining, achieving efficient full-face tunneling. EPB stands for Earth pressure balance.

[0003] In existing technologies, the core problem of tunnel boring machines (TBMs) lies in maintaining a dynamic match between the soil pressure in the tunnel chamber and the water and soil pressure at the excavation face: too low pressure can easily lead to excessive surface subsidence or even collapse; too high pressure can cause surface heave or strata perforation. The main challenges stem from the time-varying nature of soil parameters, the lag in soil removal, and the strong coupling of the propulsion system. Traditional PID control relies on experience to adjust the propulsion speed and screw conveyor rotation speed, making it difficult to respond to geological disturbances in real time, resulting in large pressure fluctuations, low control accuracy, and ultimately threatening construction safety.

[0004] Existing technology proposes a model predictive control (MPC) scheme based on nonlinear least squares support vector machine (LS-SVM) and particle swarm optimization (PSO). The core of this scheme is to use LS-SVM to establish a dynamic nonlinear relationship between earth pressure and shield control parameters (engine speed and screw conveyor speed), achieving high-precision prediction of earth pressure. With the objective of minimizing the deviation between the predicted earth pressure and the setpoint, a control parameter optimization model is constructed. The optimal engine speed and screw conveyor speed are then solved in real-time using PSO, ultimately dynamically maintaining the earth chamber pressure balance. However, this scheme has optimization limitations, specifically as follows:

[0005] (1) It fails to explicitly handle unknown bounded disturbances in the external environment, such as geological changes and equipment wear. It only relies on the error between the model prediction value and the actual value for feedback adjustment, which may increase the earth pressure control error in some strong disturbance scenarios.

[0006] (2) The global search characteristics of the ant colony algorithm result in a long time for a single optimization, which cannot meet the requirement of the 1-second control cycle of the tunnel boring machine.

[0007] (3) The above scheme only considers the speed of the screw conveyor and the propulsion speed as the main factors affecting the pressure of the earth chamber, and ignores the key influence of the total propulsion force on the pressure of the earth chamber.

[0008] (4) No conditions are given to guarantee the stability of the system. Summary of the Invention

[0009] To address the problems existing in the prior art, this invention provides a data-driven modeling method and control system suitable for the earth pressure system of EPB tunnel boring machines. When the system is subjected to unknown bounded disturbances from the outside, a robust control strategy is used to achieve unbiased tracking of earth pressure in the sealed chamber, improving the system's adaptability to geological uncertainties. Ultimately, this achieves stable earth pressure control under complex geological conditions.

[0010] The technical solution of this invention is implemented as follows:

[0011] A data-driven modeling method for the earth pressure system of EPB tunnel boring machines includes the following steps:

[0012] S1. Collect and preprocess historical operating data of the earth pressure system of the tunnel boring machine; the historical operating data refers to past operating data; the operating data includes input information and output information.

[0013] S2. Based on the historical operating data, define the system state of the tunnel boring machine's earth pressure system;

[0014] S3. Based on the observation function, establish an approximate Koopman model in a finite-dimensional observation space; based on the approximate Koopman model, establish a linear predictor;

[0015] S4. Introduce an integrator and augmented state to construct an augmented state matrix, and calculate multi-step state and output predictions based on the prediction time domain;

[0016] S5. Define the constraints on the running data and set the system prediction objective function; calculate the closed-loop system stability inequality based on Lyapunov theory; combine the state and output prediction, the constraints, the system prediction objective function, and the closed-loop system stability inequality, and generate a linear matrix inequality using Schur's complement and S-lemma;

[0017] Schur complement is a core tool in block matrix theory, used to simplify matrix operations and analyze system properties. S-lemmas are a core tool for handling the logical implications of quadratic inequalities, especially for verifying constraint compatibility in robust optimization and control systems.

[0018] S6. Perform rolling iterative optimization using the interior point method, including:

[0019] S6-1. Obtain the current running data and update the augmentation state;

[0020] S6-2. Solve the linear matrix inequality to obtain the optimal control input increment;

[0021] S6-3. Update the input information of the tunnel boring machine earth pressure system using the optimal control input increment; S6-4. Repeat S6-1 to S6-3 to continuously update the prediction time domain.

[0022] As a further optimization of the above scheme, in step S1, the historical operation data is collected in 1-second intervals;

[0023] The input information includes SCRS, AS, and TFF; the output information includes EP; SCRS represents the screw conveyor speed, AS represents the propulsion speed, TTF represents the cutterhead torque; EP represents the soil chamber pressure.

[0024] Because tunnel boring machines (TBMs) operate intermittently, meaning that the continuous working time of a single advance is relatively short, the amount of sampling data for a single continuous working segment is insufficient. Therefore, data from multiple continuous working segments are merged according to time windows and preprocessed to remove outliers.

[0025] As a further optimization of the above scheme, in S2, the system state is represented by formula (1), that is:

[0026] x(k)=[y(k) T …y(kn a ) T u(k-1) T …u(kn b ) T ] T (1); Formula (1) is constructed using the NARX model. NARX stands for Nonlinear AutoRegressive with eXogenous inputs. It is a nonlinear dynamic system modeling tool that combines autoregressive terms with external inputs and is widely used in time series forecasting and complex system identification.

[0027] Where k represents time; y(·) represents the output information at the specified time; u(·) represents the input information at the specified time; x(·) represents the system state at the specified time; n a This indicates the length of time the input information is stored; n b This indicates the length of time the output is stored in memory;

[0028] Converting formula (1) into state-space form, we get formula (2), which is:

[0029]

[0030] Where ω(·) represents the unmeasurable bounded disturbance signal, i.e., process noise; f ω (·) represents an unknown mapping function, that is, a nonlinear state transition function with uncertainty;

[0031] p represents the dimension of the output information, specifically the dimension of EP, I p×p This represents the p-order identity matrix.

[0032] The above model uses the NARX model, which stands for Nonlinear AutoRegressive with eXogenous inputs. It is a nonlinear dynamic system modeling tool that combines autoregressive terms with external inputs and is widely used in time series forecasting and complex system identification.

[0033] As a further optimization of the above scheme, a Gaussian kernel function is used as the basis function to construct the observation function; alternatively, polynomial functions, neural networks, etc., can also be used as basis functions. However, for this method, experiments have shown that the Gaussian kernel function has the best application effect. The Gaussian kernel function is expressed as: Where i∈{1,2,…,n φ}, The center point of the Gaussian kernel function is represented by exp(·); exp(·) represents the exponential decay function, and ||·|| represents the Euclidean norm; x represents the system state; n φ It is the subscript of the center point of the Gaussian kernel function, indicating the number of center points selected.

[0034] The observation function is expressed as:

[0035] Introducing the finite-dimensional Koopman operator The approximate Koopman model is established as Equation (3), namely:

[0036]

[0037] Where, n g =n+m, where m represents the dimension of the input information, and n is... The dimension; Represent a real number, and C is a matrix used to reconstruct the original output information;

[0038] The Koopman operator is a theoretical tool proposed by mathematician Bernard Koopman in 1931. It is used to transform nonlinear dynamical systems into infinite-dimensional linear systems, thereby simplifying the analysis and control of complex systems.

[0039] The solutions for A, B, and C are:

[0040]

[0041] in, k represents the square of the L2 norm of a specified parameter. n =max{n a ,n b}, N m This indicates the number of historical operational data collected;

[0042] Assume data is collected when ω = 0. and The former represents output information, and the latter represents input information, and transforms the data into... and The analysis yields:

[0043]

[0044] in, ω represents the inverse of a matrix; the pseudo-inverse, also known as the generalized inverse, is a "quasi-inverse matrix" operation defined for non-invertible matrices (such as non-square matrices and singular matrices), used to solve approximate solutions to systems of linear equations or least squares problems. ω = 0 indicates that the unmeasurable bounded perturbation signal is 0.

[0045] The linear predictor in the observation space is constructed as shown in formula (4):

[0046]

[0047] Where F is a known scaling matrix used to project ω(k) from x(k) to the output space associated with z(k); d(k) = ξ(k).

[0048] By order A linear predictor in the observation space can be constructed, taking into account modeling errors and all unknown uncertainties. Since ω(k) is bounded and ξ(k) = [ω(k)]... T ,0,…,0] T Therefore, d(k) is bounded, which means there exists a compact set. Make

[0049] As a further optimization of the above scheme, in order to transform the earth pressure setpoint tracking problem of the EPB tunnel boring machine into a stabilization problem, an integrator is introduced. The integrator is represented as:

[0050]

[0051] Among them, y r This is the preset earth pressure reference value;

[0052] Then, formula (4) is transformed into velocity form, which is expressed as formula (5), that is:

[0053]

[0054] The augmented state is introduced into formula (5). And matrix M A M B M D Then we get formula (6):

[0055] in,

[0056] Let C e =[0 p×n I p×p Reconstructing y(k) yields:

[0057] Since d(k) is bounded, and so It is bounded. Without loss of generality, we assume... but This represents the XOR operation.

[0058] By introducing an integrator with augmented states, the earth pressure tracking error is transformed into a state variable, and an augmented state space model including disturbance compensation is constructed.

[0059] Within the prediction time domain N, the state sequence is defined as follows:

[0060]

[0061] Where y(k+j|k) is the output information predicted at time k+j; For the unknown perturbation at time k+j; u(k+j|k), These represent the input information at time k+j, namely the control input, the control input increment, and the augmented state; col{·} represents the column vector;

[0062] U, Z e Y and D are the control input increment sequence, control input sequence, augmented state sequence, prediction output sequence, and unknown disturbance sequence in the prediction time domain N, respectively.

[0063] Based on the above formula (6) and the augmented state at the current time k The state and output prediction for j steps forward can be obtained;

[0064] When j∈[1,N), the state and output prediction are expressed as Equation (7):

[0065]

[0066] in,

[0067]

[0068] Y r =col{y r ,…,y r} N-1 That is, Y r It consists of N-1 y r The column vectors formed are similar in the following text; M represents A N-2, Similarly;

[0069] When j = N, the state and output prediction are expressed as formula (8):

[0070]

[0071] in,

[0072] As a further optimization of the above scheme, the constraint condition is expressed as formula (9):

[0073]

[0074] Where u represents the input information; y represents the output information; (·) max and(·) min These represent the maximum and minimum values ​​corresponding to a parameter, respectively.

[0075] Let A y =col{I p×p ,-I p×p}, b y =col{y max ,-y min}, A u =col{I m×m ,-I m×m}, b u =col{u max ,-u min}, then according to the formula (9), a set is generated, which is represented by formula (10):

[0076]

[0077] As a further optimization of the above scheme, the system prediction objective function is expressed as formula (11):

[0078]

[0079] in,

[0080]

[0081] W1 and W2 are positive definite matrices; when j≥N y hour, Where K is the state feedback gain matrix.

[0082] A positive definite matrix is ​​a real symmetric matrix in which the result of substituting any non-zero vector into its quadratic form function is always positive.

[0083] To obtain a robust controller, a quadratic cost function, i.e., the objective function, is defined in the infinite time domain:

[0084]

[0085] Considering that all future control input increments in the objective function All control methods are determined by state feedback control laws, which may lead to an overly conservative control approach. Therefore, this proposal provides a multi-stage scheduling control algorithm. The core idea of ​​this algorithm is to divide the infinite time-domain objective function into two parts, namely:

[0086]

[0087] When j≥N y hour, Where K is the state feedback gain matrix.

[0088] Ultimately, the robust control problem (objective function) of the system is transformed into a minimization problem (Equation (11)). As a further optimization of the above scheme, the stability inequality of the closed-loop system is expressed as Equation (12):

[0089]

[0090] Where, V(j,k)=z e (k+j|k)Pz e (k+j|k) and j≥N y P is a positive definite matrix.

[0091] By combining Lyapunov theory with its application, the asymptotic stability of the closed-loop system can be guaranteed.

[0092] As a further optimization of the above scheme, the linear matrix inequality is expressed as formula (13):

[0093]

[0094] in,

[0095]

[0096]

[0097] Where γ is The upper bound is given by Π = kG; H(α)=diag{α j ,j=1,…,pN}, diag represents a diagonal matrix; α=col{α j ,j=1,…,pN};α j M is a positive real number; M = P -1 Δ1 and Δ2 are dimension-matching matrices;

[0098]

[0099] "Dimensional matching" refers to the two matrices involved in matrix calculations being compatible in dimensions, such as "Δ11". pN×1 In the text, the column dimension of "Δ1" and "1" are... pN×1 The row dimensions are the same.

[0100] The present invention also provides a control system for a tunnel boring machine (TBM) for controlling the TBM, and applies the above-mentioned data-driven modeling method for the earth pressure system of an EPB TBM.

[0101] Compared with the prior art, the present invention achieves the following beneficial effects:

[0102] (1) This invention introduces an integrator through an augmented state to transform the earth pressure tracking error into a state variable, constructing an augmented state space model that includes disturbance compensation. The integrator eliminates steady-state errors, and combined with the disturbance modeling capability of the Koopman operator, it ensures that when an external unknown bounded disturbance tends to a steady state, the earth pressure system can achieve target tracking with complete error elimination. This control method significantly suppresses the influence of external bounded disturbances on the control system, effectively ensuring that the earth pressure converges precisely to the set value, thereby significantly reducing the risk of face collapse or surface uplift caused by inaccurate pressure control, reducing economic losses, and ensuring the safety and stability of tunnel boring.

[0103] (2) This invention takes into account the rotational speed of the screw conveyor, the propulsion speed and the total propulsion force, and realizes multi-parameter coordinated control to improve control accuracy and response speed.

[0104] (3) The present invention provides a rigorous theoretical guarantee through the Lyapunov method, ensuring that the earth pressure control system error converges to zero in the infinite time domain, thus solving the divergence problem that may exist in the long-term operation of the existing technology.

[0105] (4) This invention utilizes Schur's complement and S-lemma to transform the nonlinear inequality optimization problem into a linear matrix inequality convex optimization problem, which can be solved using existing convex optimization methods, such as the interior-point method. The convex optimization transformation significantly reduces computational complexity. It solves the problems of existing technologies (such as those using heuristic or global search algorithms) where single-optimization takes a long time and cannot meet the requirement of 1-second-level control implementation.

[0106] (5) Linear matrix inequality constraints can ensure that the control input is within a safe range, avoid parameter out-of-bounds problems caused by manual adjustment in actual engineering, avoid equipment overload damage, and reduce operation and maintenance costs. Attached Figure Description

[0107] Figure 1 This is a flowchart illustrating a data-driven modeling method for the earth pressure system of an EPB tunnel boring machine, provided by an embodiment of the present invention. Detailed Implementation

[0108] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions in the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0109] like Figure 1 As shown, this embodiment provides a control system for a tunnel boring machine (TBM), which applies a data-driven modeling method suitable for the earth pressure system of an EPB TBM, including the following steps:

[0110] S1. Collect historical operating data of the earth pressure system of the tunnel boring machine in 1-second intervals and preprocess the historical operating data. The historical operating data is the past operating data. The operating data includes input information and output information. The input information includes SCRS, AS and TFF. The output information includes EP. SCRS represents the speed of the screw conveyor, AS represents the propulsion speed, TTF represents the cutterhead torque, and EP represents the earth pressure chamber.

[0111] Because tunnel boring machines (TBMs) operate intermittently, meaning that the continuous working time of a single advance is relatively short, the amount of sampling data for a single continuous working segment is insufficient. Therefore, data from multiple continuous working segments are merged according to time windows and preprocessed to remove outliers.

[0112] S2. Define the system state of the earth pressure system of the tunnel boring machine based on historical operating data.

[0113] In this embodiment, the system state is represented by formula (1), that is:

[0114] x(k)=[y(k) T …y(kn a ) T u(k-1) T …u(kn b ) T ] T (1); Formula (1) is constructed using the NARX model. NARX stands for Nonlinear AutoRegressive with eXogenous inputs. It is a nonlinear dynamic system modeling tool that combines autoregressive terms with external inputs and is widely used in time series forecasting and complex system identification.

[0115] Where k represents time; y(·) represents the output information at the specified time; u(·) represents the input information at the specified time; x(·) represents the system state at the specified time; n a Indicates the length of time the input information is stored; n b Indicates the length of time the output is stored in memory;

[0116] Converting formula (1) into state-space form, we get formula (2), which is:

[0117]

[0118] Where ω(·) represents the unmeasurable bounded disturbance signal, i.e., process noise; f ω (·) represents an unknown mapping function, that is, a nonlinear state transition function with uncertainty;

[0119] p represents the dimension of the output information, specifically the dimension of EP, I p×p This represents the p-order identity matrix.

[0120] The above model uses the NARX model, which stands for Nonlinear AutoRegressive with eXogenous inputs. It is a nonlinear dynamic system modeling tool that combines autoregressive terms with external inputs and is widely used in time series forecasting and complex system identification.

[0121] S3. Establish an approximate Koopman model in the finite-dimensional observation space based on the observation function. Specifically, the Gaussian kernel function is used as the basis function to construct the observation function; the Gaussian kernel function is expressed as: Where i∈{1,2,…,n φ}, The center point of the Gaussian kernel function is represented by exp(·); exp(·) represents the exponential decay function; ‖·‖ represents the Euclidean norm; x represents the system state; n φ It is the subscript of the center point of the Gaussian kernel function, indicating the number of center points selected.

[0122] The observation function is expressed as:

[0123] Introducing the finite-dimensional Koopman operator An approximate Koopman model is established, expressed as formula (3), namely:

[0124]

[0125] Where, n g =n+m, where m represents the dimension of the input information, and n is... The dimension; Represent a real number, and C is a matrix used to restore the original output information;

[0126] The Koopman operator is a theoretical tool proposed by mathematician Bernard Koopman in 1931. It is used to transform nonlinear dynamical systems into infinite-dimensional linear systems, thereby simplifying the analysis and control of complex systems.

[0127] The solutions for A, B, and C are:

[0128]

[0129] in, k represents the square of the L2 norm of a specified parameter. n =max{n a ,b b}, N m This indicates the number of historical operational data collected;

[0130] Assume data is collected when ω = 0. and The former represents output information, and the latter represents input information, and transforms the data into... and The analysis yields:

[0131]

[0132] in, ω represents the inverse of a matrix; the pseudo-inverse, also known as the generalized inverse, is a "quasi-inverse matrix" operation defined for non-invertible matrices (such as non-square matrices and singular matrices), used to solve approximate solutions to systems of linear equations or least squares problems. ω = 0 indicates that the unmeasurable bounded perturbation signal is 0.

[0133] Based on the approximate Koopman model, a linear predictor in the observation space is constructed, expressed as Equation (4):

[0134]

[0135] Where F is a known scaling matrix used to project ω(k) from x(k) to the output space associated with z(k); d(k) = ξ(k).

[0136] By order A linear predictor in the observation space can be constructed, taking into account modeling errors and all unknown uncertainties. Since ω(k) is bounded, and

[0137] ξ(k)=[ω(k) T ,0,…,0] T Therefore, d(k) is bounded, which means there exists a compact set. Make

[0138] S4. In this embodiment, to transform the earth pressure setpoint tracking problem of the EPB tunnel boring machine into a stabilization problem, an integrator is introduced. The integrator is represented as:

[0139]

[0140] Among them, y r This is the preset earth pressure reference value;

[0141] Then formula (4) is transformed into velocity form, which is expressed as formula (5), that is:

[0142]

[0143] Introduce augmented states into formula (5) And matrix M A M B M D Then we get formula (6):

[0144] in,

[0145] Let C e =[0p×n I p×p Reconstructing y(k) yields:

[0146] Since d(k) is bounded, and so It is bounded. Without loss of generality, we assume... but This represents the XOR operation.

[0147] By introducing an integrator with augmented states, the earth pressure tracking error is transformed into a state variable, and an augmented state space model including disturbance compensation is constructed.

[0148] Multi-step state and output predictions are calculated based on the prediction time domain; specifically, within the prediction time domain N, the state sequence is defined as follows:

[0149]

[0150] Where y(k+j|k) is the output information of the prediction at time k+j; For the unknown perturbation at time k+j; u(k+j|k), These represent the input information at time k+j, namely the control input, the control input increment, and the augmented state; col{·} represents the column vector;

[0151] U, Z e Y and D are the control input increment sequence, control input sequence, augmented state sequence, prediction output sequence, and unknown disturbance sequence in the prediction time domain N, respectively.

[0152] Based on formula (6) and the augmented state z at the current time k e (k|k) can be used to obtain the state and output prediction for j steps forward;

[0153] When j∈[1,N), the state and output prediction are expressed as Equation (7):

[0154]

[0155] in,

[0156]

[0157] Y r =col{y r ,…,y r} N-1 That is, Y r It consists of N-1 y r The column vectors formed are similar in the following text; M representsA N-2, Similarly;

[0158] When j = N, the state and output prediction are expressed as formula (8):

[0159]

[0160] in,

[0161] S5. Define the constraints on the running data and set the system prediction objective function.

[0162] In this embodiment, the constraint condition is expressed as formula (9):

[0163]

[0164] Where u represents input information; y represents output information; (·) max and(·) min These represent the maximum and minimum values ​​corresponding to a parameter, respectively.

[0165] Let A y =col{I p×p ,-I p×p}, b y =col{y max ,-y min}, A u =col{I m×m ,-I m×m}, b u =col{u max ,-u min}, then according to formula (9), a set is generated, which is expressed as formula (10):

[0166]

[0167] In this embodiment, to obtain a robust controller, a quadratic cost function, i.e., the objective function, is defined in the infinite time domain:

[0168]

[0169] Considering that all future control input increments in the objective function All control methods are determined by state feedback control laws, which may lead to an overly conservative control approach. Therefore, this proposal provides a multi-stage scheduling control algorithm. The core idea of ​​this algorithm is to divide the infinite time-domain objective function into two parts, namely:

[0170]

[0171] When j≥Ny hour, Where K is the state feedback gain matrix.

[0172] Ultimately, the robust control problem (objective function) of the system is transformed into a minimization problem (Equation (11)), namely:

[0173] (11). W1 and W2 are positive definite matrices. A positive definite matrix is ​​a real symmetric matrix in which the result of substituting any non-zero vector into its quadratic form function is always positive.

[0174] The stability inequality of the closed-loop system is calculated based on Lyapunov theory; in this embodiment, the stability inequality of the closed-loop system is expressed as formula (12):

[0175]

[0176] in, And j≥N y P is a positive definite matrix.

[0177] By combining Lyapunov theory with its application, the asymptotic stability of the closed-loop system can be guaranteed.

[0178] Combining state and output prediction, constraints, system prediction objective function, and closed-loop system stability inequality, a linear matrix inequality is generated using Schur's complement and S-lemma, expressed as formula (13):

[0179]

[0180] in,

[0181] Where γ is The upper bound is given by ∏ = kG;

[0182] H(α)=diag{α j ,j=1,…,pN}, diag represents a diagonal matrix; α=col{α j ,j=1,…,pN};pN is a positive real number;M=P -1 Δ1 and Δ2 are dimension-matching matrices;

[0183] "Dimensional matching" refers to the two matrices involved in matrix calculations being compatible in dimensions, such as "Δ11". pN×1 In the text, the column dimension of "Δ1" and "1" are... pN×1 The row dimensions are the same.

[0184] Schur complement is a core tool in block matrix theory, used to simplify matrix operations and analyze system properties. S-lemmas are a core tool for handling the logical implications of quadratic inequalities, especially for verifying constraint compatibility in robust optimization and control systems.

[0185] S6. Perform rolling iterative optimization using the interior point method, including:

[0186] S6-1. Obtain the current running data and update the augmentation state;

[0187] S6-2. Solve the linear matrix inequalities to obtain the optimal control input increment;

[0188] S6-3. Update the input information of the tunnel boring machine's earth pressure system using the optimal control input increment;

[0189] S6-4. Repeat S6-1 to S6-3 to continuously update the prediction time domain.

[0190] Based on the disclosure and teachings of the foregoing specification, those skilled in the art can make changes and modifications to the above embodiments. Therefore, the present invention is not limited to the specific embodiments disclosed and described above, and some modifications and changes to the present invention should also fall within the protection scope of the claims of the present invention. Furthermore, although some specific terms are used in this specification, these terms are only for convenience of explanation and do not constitute any limitation on the present invention.

Claims

1. A data-driven modeling method for the earth pressure system of an EPB tunnel boring machine, characterized in that, Includes the following steps: S1. Collect and preprocess historical operating data of the earth pressure system of the tunnel boring machine; the historical operating data refers to past operating data; the operating data includes input information and output information. S2. Based on the historical operating data, define the system state of the tunnel boring machine's earth pressure system; S3. Based on the observation function, establish an approximate Koopman model in a finite-dimensional observation space; based on the approximate Koopman model, establish a linear predictor; S4. Introduce an integrator and augmented state to construct an augmented state matrix, and calculate multi-step state and output predictions based on the prediction time domain; S5. Define the constraints on the running data and set the system prediction objective function; calculate the closed-loop system stability inequality based on Lyapunov theory; combine the state and output prediction, the constraints, the system prediction objective function, and the closed-loop system stability inequality to generate a linear matrix inequality. S6. Perform rolling iterative optimization using the interior point method, including: S6-1. Obtain the current running data and update the augmentation state; S6-2. Solve the linear matrix inequality to obtain the optimal control input increment; S6-3. Update the input information of the tunnel boring machine earth pressure system using the optimal control input increment; S6-4. Repeat S6-1 to S6-3 to continuously update the prediction time domain.

2. The data-driven modeling method for the earth pressure system of an EPB tunnel boring machine according to claim 1, characterized in that, In step S1, the historical operation data is collected in 1-second intervals. The input information includes SCRS, AS, and TFF; the output information includes EP; SCRS represents the screw conveyor speed, AS represents the propulsion speed, TTF represents the cutterhead torque; EP represents the soil chamber pressure.

3. The data-driven modeling method for the earth pressure system of an EPB tunnel boring machine according to claim 2, characterized in that, In S2, the system state is represented by formula (1), that is: x(k)=[y(k) T …y(k-n a ) T u(k-1) T …u(k-n b ) T ] T (1); Where k represents time; y(·) represents the output information at the specified time; u(·) represents the input information at the specified time; x(·) represents the system state at the specified time; n a This indicates the length of time the input information is stored; n b This indicates the length of time the output is stored in memory; Converting formula (1) into state-space form, we get formula (2), which is: Where ω(·) represents an unmeasurable bounded disturbance signal; f ω (·) represents an unknown mapping function; p represents the dimension of the output information, I p×p This represents the p-order identity matrix.

4. The data-driven modeling method for the earth pressure system of an EPB tunnel boring machine according to claim 3, characterized in that, The Gaussian kernel function is used as the basis function to construct the observation function; the Gaussian kernel function is expressed as: Where i∈{1,2,…,n φ }, The center point of the Gaussian kernel function is represented by exp(·); exp(·) represents the exponential decay function, and ‖·‖ represents the Euclidean norm; x represents the system state. The observation function is expressed as: Introducing the finite-dimensional Koopman operator The approximate Koopman model is established as Equation (3), namely: Where, n g =n+m, where m represents the dimension of the input information, and n is... The dimension; Represent a real number, and C is a matrix used to reconstruct the original output information; The solutions for A, B, and C are: in, k represents the square of the L2 norm of a specified parameter. n =max{n a ,b b }, N m This indicates the number of historical operational data collected; Assume data is collected when ω = 0. and Transform the data and The analysis yields: in, Indicates the inverse of a matrix; The linear predictor in the observation space is constructed as shown in formula (4): Where F is a known scaling matrix used to project ω(k) from x(k) to the output space associated with z(k); d(k) = ξ(k).

5. The data-driven modeling method for the earth pressure system of an EPB tunnel boring machine according to claim 4, characterized in that, The integrator is represented as: Among them, y r This is the preset earth pressure reference value; Then, formula (4) is transformed into velocity form, which is expressed as formula (5), that is: The augmented state is introduced into formula (5). And matrix M A M B M D Then we get formula (6): in, Let C e =[0 p×n I p×p Reconstructing y(k) yields: Within the prediction time domain N, the state sequence is defined as follows: Where y(k+j|k) is the output information predicted at time k+j; For the unknown perturbation at time k+j; u(k+j|k), These represent the input information, control input increment, and augmented state at time k+j, respectively; col{·} represents a column vector; U, Z e Y and D are the control input increment sequence, control input sequence, augmented state sequence, prediction output sequence, and unknown disturbance sequence in the prediction time domain N, respectively. Based on the above formula (6) and the augmented state at the current time k The state and output prediction for j steps forward can be obtained; When j∈[1,N), the state and output prediction are expressed as Equation (7): in, AND r =col{y r ,…,and r } N-1 ; When j = N, the state and output prediction are expressed as formula (8): in, 6. The data-driven modeling method for the earth pressure system of an EPB tunnel boring machine according to claim 5, characterized in that, The constraint condition is expressed as formula (9): Where u represents the input information; y represents the output information; (·) max and(·) min These represent the maximum and minimum values ​​corresponding to a parameter, respectively. Let A y =col{I p×p ,-I p×p }, b y =col{y max ,-y min }, A u =col{I m×m ,-I m×m }, b u =col{u max ,-u min }, then according to the formula (9), a set is generated, which is represented by formula (10):

7. A data-driven modeling method for the earth pressure system of an EPB tunnel boring machine according to claim 6, characterized in that, The system prediction objective function is expressed as formula (11): in, W1 and W2 are positive definite matrices; when j≥N y hour, Where K is the state feedback gain matrix.

8. The data-driven modeling method for the earth pressure system of an EPB tunnel boring machine according to claim 7, characterized in that, The stability inequality of the closed-loop system is expressed as formula (12): in, And j≥N y P is a positive definite matrix.

9. A data-driven modeling method for the earth pressure system of an EPB tunnel boring machine according to claim 8, characterized in that, The linear matrix inequality is expressed as formula (13): in, Where γ is The upper bound is given by Π = kG; H(α)=diag{α j ,j=1,…,pN}, diag represents a diagonal matrix; α=col{α j ,j=1,…,pN};α j M is a positive real number; M = P -1 Δ1 and Δ2 are dimension-matching matrices; 10. A control system for a tunnel boring machine (TBM), used to control the TBM, characterized in that, The data-driven modeling method for the earth pressure system of EPB tunnel boring machines, as described in any one of claims 1 to 9, is applied.

Images