Intelligent Regulation Method for Shield Tunnels with Extreme Radii in Narrow Spaces

Abstract

The present invention relates to the technical field of shield tunnel construction, and specifically discloses an intelligent regulation method for shield tunnels with an ultimate radius in a narrow space, including: obtaining the state information of the shield machine; collecting formation pressure, moisture content, and soil deformation data to obtain formation characteristic information; analyzing and obtaining an optimal construction adjustment strategy based on the formation characteristic information; based on the optimal construction adjustment strategy, using deep reinforcement learning combined with fuzzy PID control to calculate the optimal tunneling trajectory of the shield machine, and obtaining an adaptable construction plan for the ultimate radius in a narrow space; the present invention jointly optimizes the trajectory of the shield machine based on deep reinforcement learning combined with fuzzy PID control, realizes dynamic attitude correction, and reduces the deviation of the tunneling trajectory. The multi-sensor data fusion improves the accuracy of attitude estimation, realizes the reduction of the shield attitude angle error, effectively reduces the number of manual interventions, and improves the construction efficiency; the automatic thrust and torque optimization control avoids attitude mutations caused by formation changes and improves the stability of the shield machine.

Description

Technical Field

[0001] The present invention belongs to the technical field of shield tunnel construction, and particularly relates to an intelligent control method for shield tunnels with an ultimate radius in a narrow space. Background Art

[0002] In the construction of shield tunnels with an ultimate radius in a narrow space, many technical challenges are faced. The traditional shield construction method mainly relies on manual experience and on-site monitoring. This method often fails to achieve an ideal control effect in a complex and changeable construction environment. Especially under the conditions of a narrow space and an ultimate radius, it becomes particularly difficult to accurately control construction parameters such as the tunneling attitude of the shield machine, the cutterhead torque, and the synchronous grouting pressure.

[0003] In the prior art, the adjustment of the construction parameters of the shield machine mainly depends on the experience and judgment of the operator. However, due to the complexity and uncertainty of the construction environment, as well as the differences in the experience and skill levels of the operators, the adjustment of the construction parameters is often inaccurate and untimely. This not only affects the tunneling efficiency and construction accuracy of the shield machine, but also may increase the construction risk and cost. In addition, the utilization of multi-source data during the shield construction process is insufficient. A large amount of data is generated during the construction of the shield machine, including attitude, thrust, torque, tunneling speed, cutterhead rotation speed, formation resistance, etc. These data contain rich construction information, but the prior art often does not fully explore and utilize these data, resulting in the lack of a scientific basis and accuracy for the adjustment of construction parameters.

[0004] In view of this, the inventor proposes an intelligent control method for shield tunnels with an ultimate radius in a narrow space to solve the above problems. Summary of the Invention

[0005] The purpose of the present invention is to provide an intelligent control method for shield tunnels with an ultimate radius in a narrow space to solve the problems raised in the above background art.

[0006] To achieve the above purpose, the present invention provides the following technical solutions:

[0007] An intelligent control method for shield tunnels with an ultimate radius in a narrow space, comprising:

[0008] Obtaining multi-source data during the shield construction process, and performing data fusion using a multi-modal data fusion algorithm to obtain the state information of the shield machine;

[0009] Collecting formation pressure, water content, and soil deformation data, and using a formation dynamic evolution neural network model, combined with the state information of the shield machine, to predict the formation response during the shield tunneling process to obtain formation characteristic information;

[0010] Based on the formation characteristic information, optimizing the tunneling parameters using reinforcement learning, and combining digital twin simulation analysis to obtain the optimal construction adjustment strategy;

[0011] Based on the optimal construction adjustment strategy, the optimal tunneling trajectory of the shield machine is calculated by combining deep reinforcement learning with fuzzy PID control, and a construction plan adaptable to the limit radius in a narrow space is obtained.

[0012] Preferably, based on the optimal construction adjustment strategy, optimization for adaptability to special environments is carried out to obtain a construction plan adaptable to the limit radius in a narrow space, including the following steps:

[0013] Construct a dynamic shield attitude adjustment model according to the tunnel curvature radius and the minimum construction space constraint;

[0014] Adopt a non-linear trajectory optimization algorithm to calculate the optimal adjustment strategy of the shield attitude in the limit radius area;

[0015] Combined with the formation deformation feedback, the construction parameters are adjusted in real time to improve the adaptability of shield construction with the limit radius in a narrow space.

[0016] Preferably, the multi-modal data fusion algorithm is Bayesian estimation, and the formula for Bayesian estimation is:

[0017] ;

[0018] P(H∣D) represents the posterior probability of the construction environment information H under the sensor data D;

[0019] P(D∣H) represents the likelihood of observing the sensor data D under the specific construction environment information H, the likelihood function;

[0020] P(H) represents the prior probability of the construction environment information, which is modeled based on historical data;

[0021] P(D) represents the total probability of the sensor data, which is used for normalization.

[0022] Preferably, the formula for the formation dynamic evolution neural network model is:

[0023] ;

[0024] ;

[0025] Where: X t represents the input formation monitoring data at the construction time t;

[0026] h t represents the hidden state at time t, h t-1 represents the hidden state at time t-1, representing the formation evolution information;

[0027] W h ,W x ,Wy Denote network weight parameters;

[0028] b h , b y Denote bias terms;

[0029] Yt represents the predicted formation response at time t;

[0030] σ represents the activation function.

[0031] Preferably, based on the formation characteristic information, a reinforcement learning algorithm is used to optimize the tunneling trajectory to obtain the optimal tunneling path; considering parameters such as the total thrust, torque, cutter head rotation speed, and synchronous grouting pressure, a multi-objective optimization method based on the genetic algorithm is used to optimize the construction strategy to obtain the optimal construction adjustment strategy.

[0032] Preferably, the expression of the multi-objective optimization method based on the genetic algorithm includes:

[0033] Fitness calculation:

[0034] ;

[0035] where: fi(x) represents the i-th optimization objective;

[0036] wi represents the weight of the objective;

[0037] Crossover operation:

[0038] ;

[0039] where: xnew is the new construction parameter combination after crossover;

[0040] x1, x2 are the parent parameter sets;

[0041] Mutation operation:

[0042] ;

[0043] where: Xmut is the new construction parameter combination after the mutation operation;

[0044] N(0, σ 2 ) represents a random mutation term following a Gaussian distribution.

[0045] Preferably, the expression of the deep reinforcement learning is:

[0046] ;

[0047] where Q(st, at) represents the value of executing action at in state st;

[0048] α represents the learning rate, controlling the update speed;

[0049] rt represents the reward at the current step;

[0050] γ represents the discount factor, weighing the impact of future rewards on the current decision;

[0051] represents the value of the optimal action in the next state st+1.

[0052] Preferably, the expression combining fuzzy PID control is:

[0053] ;

[0054] where: e(t) represents the attitude error of the shield machine;

[0055] Kp, Ki, Kd represent the PID control parameters, which are dynamically adjusted by fuzzy logic rules;

[0056] The fuzzy logic adjustment rules are as follows:

[0057] If |e(t)| is large, increase Kp to accelerate the response speed;

[0058] If e(t) remains unchanged for a long time, increase Ki to reduce the steady-state error;

[0059] If e(t) changes too fast, increase Kd to suppress oscillations.

[0060] Preferably, the multi-source data includes the attitude, thrust, torque, tunneling speed, cutterhead rotation speed, formation resistance, and tail gap of the shield machine.

[0061] Compared with the prior art, the beneficial effects of the present invention are:

[0062] (1) The present invention jointly optimizes the shield machine trajectory based on deep reinforcement learning (DRL) combined with fuzzy PID control, realizes dynamic attitude correction, and reduces the tunneling trajectory deviation. The multi-sensor data fusion improves the attitude estimation accuracy, realizes the reduction of the shield attitude angle error, effectively reduces the number of manual interventions, and improves the construction efficiency. The automatic thrust and torque optimization control avoids attitude mutations caused by formation changes, improves the stability of the shield machine, and reduces the risk of construction accidents.

[0063] (2) The present invention optimizes the synchronous grouting based on particle swarm optimization combined with genetic algorithm and finite element analysis, reduces the probability of formation settlement, and effectively prevents formation instability; the real-time osmotic pressure monitoring combined with adaptive PID control automatically adjusts the grouting pressure to avoid tail leakage; the slurry utilization rate is improved, the water-cement ratio is optimized, the slurry waste is reduced, and the construction material cost is reduced.

[0064] (3) The present invention fuses multi-sensor data and combines data-driven modeling to improve the prediction ability of shield construction parameters, realizing the adaptive adjustment of intelligent thrust, torque, and grouting pressure; multi-objective optimization ensures the best construction efficiency and safety, reduces the number of shield machine attitude adjustments, shortens the construction time, and saves construction costs. The tunnel axis deviation is small, ensuring that the construction accuracy reaches a high standard. Description of the Drawings

[0065] Figure 1 It is a flowchart of the intelligent control method for shield tunnels with an ultimate radius in a narrow space according to the present invention. Specific Embodiments

[0066] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention. Embodiment 1

[0067] Please refer to Figure 1 As shown in the figure, the intelligent control method for shield tunnels with an ultimate radius in a narrow space includes:

[0068] Obtain multi-source data during the shield construction process, and perform data fusion using a multi-modal data fusion algorithm to obtain the shield machine state information;

[0069] Collect formation pressure, water content, and soil deformation data, and use a formation dynamic evolution neural network model to predict the formation response during shield tunneling in combination with the shield machine state information to obtain formation characteristic information;

[0070] Based on the formation characteristic information, optimize the tunneling parameters using reinforcement learning, and combine digital twin simulation analysis to obtain the optimal construction adjustment strategy;

[0071] Based on the optimal construction adjustment strategy, use deep reinforcement learning combined with fuzzy PID control to calculate the optimal tunneling trajectory of the shield machine, obtain a construction plan adaptable to the ultimate radius in a narrow space, and adjust the construction parameters such as the shield machine tunneling attitude, cutterhead torque, and synchronous grouting pressure for precise control of the shield machine.

[0072] Specifically, based on the optimal construction adjustment strategy, perform special environment adaptability optimization to obtain a construction plan adaptable to the ultimate radius in a narrow space, including the following steps:

[0073] According to the tunnel curvature radius and the minimum construction space constraint, construct a dynamic shield attitude adjustment model;

[0074] Adopt a non - linear trajectory optimization algorithm to calculate the optimal adjustment strategy of the shield attitude in the extreme radius area;

[0075] Combine the formation deformation feedback to adjust the construction parameters in real - time, and improve the adaptability of shield construction with extreme radius in narrow spaces.

[0076] Specifically, the multi - modal data fusion algorithm is Bayesian estimation, and the formula of Bayesian estimation is:

[0077] ;

[0078] P(H∣D) represents the posterior probability of the construction environment information H under the sensor data D;

[0079] P(D∣H) represents the likelihood of observing the sensor data D under the specific construction environment information H, the likelihood function;

[0080] P(H) represents the prior probability of the construction environment information, which is modeled based on historical data;

[0081] P(D) represents the total probability of the sensor data, which is used for normalization.

[0082] This algorithm is used to integrate the shield machine sensor data (such as thrust, torque, settlement, etc.), provide a more stable and noise - resistant construction environment assessment, can effectively improve the data accuracy, and reduce the influence of single - sensor error on the adjustment of construction parameters.

[0083] Specifically, the formula of the formation dynamic evolution neural network model is:

[0084] ;

[0085] ;

[0086] Among them: X t represents the input formation monitoring data at the construction time t;

[0087] h t represents the hidden state at time t, h t-1 represents the hidden state at time t - 1, representing the formation evolution information;

[0088] W h ,W x ,W y represents the network weight parameters;

[0089] b h ,b y represents the bias term;

[0090] Yt represents the predicted formation response at time t;

[0091] σ represents the activation function.

[0092] This model is used to predict the formation deformation trend, helping construction workers adjust construction parameters in advance, prevent problems such as overexcavation and formation collapse, and improve the safety of shield tunneling construction.

[0093] Specifically, based on the formation characteristic information, a reinforcement learning algorithm is used to optimize the tunneling trajectory to obtain the optimal tunneling path; considering parameters such as thrust, torque, cutter head rotation speed, and synchronous grouting pressure, a multi-objective optimization method based on the genetic algorithm is used to optimize the construction strategy to obtain the optimal construction adjustment strategy.

[0094] Collect grouting-related parameters to obtain earth pressure, seepage pressure, tail gap pressure, slurry flow rate, and slurry density, and obtain grouting monitoring data.

[0095] Process the grouting monitoring data, using the particle swarm optimization (PSO) + genetic algorithm (GA-MOO) to calculate the optimal grouting parameters for optimizing the water-cement ratio (W / C), grouting flow rate, and grouting pressure, and obtain the optimal grouting parameters.

[0096] Process the optimal grouting parameters, combine with the finite element analysis (FEM) to calculate the formation deformation, adjust the grouting strategy to ensure that the formation settlement is controlled within 4 mm, and obtain the formation stable grouting plan.

[0097] Process the formation stable grouting plan, based on the adaptive PID control, to dynamically adjust the grouting pressure, optimize the sealing effect of the shield tunnel, and obtain the precise grouting filling effect.

[0098] Specifically, the expression of the multi-objective optimization method based on the genetic algorithm includes:

[0099] Fitness calculation:

[0100] ;

[0101] where: fi(x) represents the i-th optimization objective;

[0102] wi represents the weight of the objective;

[0103] Crossover operation:

[0104] ;

[0105] where: xnew is the new combination of construction parameters after crossover;

[0106] x1, x2 are the set of parent parameters;

[0107] Mutation operation:

[0108] ;

[0109] Where: Xmut is the new construction parameter combination after mutation operation;

[0110] N(0,σ 2 ) represents a random mutation term following a Gaussian distribution.

[0111] This algorithm is used to optimize construction parameters (such as thrust, torque, cutterhead rotation speed, synchronous grouting pressure) so that the shield machine can maintain an efficient and stable construction state under complex geological conditions.

[0112] Specifically, the expression of the deep reinforcement learning is:

[0113] ;

[0114] Where Q(st,at) represents the value of executing action at in state st;

[0115] α represents the learning rate, controlling the update speed;

[0116] rt represents the reward at the current step;

[0117] γ represents the discount factor, weighing the impact of future rewards on the current decision;

[0118] represents the value of the optimal action in the next state st+1.

[0119] This algorithm enables the shield machine to automatically learn the optimal tunneling trajectory, reduce attitude deviation in narrow spaces and construction environments with extreme radii, and improve construction accuracy and efficiency.

[0120] Specifically, the expression of the combination of fuzzy PID control is:

[0121] ;

[0122] Where: e(t) represents the attitude error of the shield machine;

[0123] Kp, Ki, Kd represent PID control parameters, which are dynamically adjusted by fuzzy logic rules;

[0124] The fuzzy logic adjustment rules are as follows:

[0125] If ∣e(t)∣ is large, increase Kp to accelerate the response speed;

[0126] If e(t) remains unchanged for a long time, increase Ki to reduce the steady-state error;

[0127] If e(t) changes too fast, increase Kd to suppress oscillations.

[0128] This algorithm is used to dynamically adjust the attitude of the shield machine, reduce yaw and attitude deviation during tunneling, and improve the accuracy and stability of the shield machine.

[0129] Specifically, the multi-source data includes the attitude of the shield machine, thrust, torque, tunneling speed, cutterhead rotation speed, formation resistance, and tail gap.

[0130] As can be seen from the above, this method addresses the key issues in the construction of shield tunnels with extremely limited radii in narrow spaces and proposes a complete set of intelligent control solutions, including key aspects such as intelligent tunneling trajectory optimization, synchronous grouting intelligent optimization, and adaptive adjustment of construction parameters.

[0131] The present invention combines deep reinforcement learning (DRL) with fuzzy PID control to jointly optimize the trajectory of the shield machine, achieve dynamic attitude correction, and reduce the deviation of the tunneling trajectory. Multi-sensor data fusion improves the accuracy of attitude estimation, reduces the error of the shield attitude angle, effectively reduces the number of manual interventions, and improves construction efficiency. Automatic thrust and torque optimization control avoids sudden attitude changes caused by formation variations, improves the stability of the shield machine, and reduces the risk of construction accidents;

[0132] The present invention performs synchronous grouting optimization based on particle swarm optimization (PSO) combined with genetic algorithm (GA-MOO) and finite element analysis (FEM), reduces the probability of ground settlement, and effectively prevents ground instability; real-time seepage pressure monitoring combined with adaptive PID control automatically adjusts the grouting pressure to avoid tail leakage; improves the utilization rate of grout, optimizes the water-cement ratio, reduces grout waste, and reduces construction material costs;

[0133] The present invention combines multi-sensor fusion (IMU, laser ranging, pressure sensors) with data-driven modeling to improve the prediction ability of shield construction parameters and achieve adaptive adjustment of intelligent thrust, torque, and grouting pressure. Multi-objective optimization (MOO) ensures the optimal balance between construction efficiency and safety, reduces the number of shield machine attitude adjustments, shortens the construction time, and saves construction costs. The tunnel axis deviation is small, ensuring that the construction accuracy meets high standards.

[0134] Example 2

[0135] This method is not only applicable to shield tunnels with extremely limited radii in narrow spaces but also has broad application value in projects such as urban subways, underground utility tunnels, river-crossing tunnels, and high-speed rail tunnels. In the future, it can be extended to a fully automatic intelligent shield construction system, promoting the development of tunnel construction towards the direction of intelligence, unmanned operation, and high precision;

[0136] Shield construction with extremely limited radii is commonly used in narrow sections of subways and tunnels with small turning radii. This method can improve tunneling accuracy, reduce settlement, tail leakage, and tunnel axis deviation, ensure construction safety, and improve the construction efficiency of urban rail transit;

[0137] Furthermore, intelligent tunneling trajectory optimization for shield tunnels with extreme radii:

[0138] 1. Implementation background:

[0139] In a certain urban subway tunnel project, a shield machine with a 6m diameter needs to tunnel in a space with an extremely small curve radius of 250m, and the geological conditions along the line are complex (including silty clay and sand and gravel layers). The traditional manual adjustment method is difficult to ensure high-precision tunneling and is prone to problems such as attitude deviation and segment misalignment. Based on deep reinforcement learning (DRL) combined with an adaptive fuzzy PID control algorithm, this embodiment optimizes the attitude adjustment of the shield machine and improves the tunneling accuracy.

[0140] 2. Data acquisition method:

[0141] (1) Shield machine attitude data acquisition:

[0142] Equipment: IMU inertial navigation, laser rangefinder, attitude sensor;

[0143] The sampling frequency is shown in Table 1;

[0144] Table 1

[0145]

[0146] (2) Multi-sensor data fusion (based on Bayesian estimation)

[0147] Bayesian estimation is used to calculate the optimal estimated value of the shield machine attitude, and the equation is:

[0148] ;

[0149] P(H∣D) represents the posterior probability of the construction environment information H under the sensor data D;

[0150] P(D∣H) represents the likelihood of observing the sensor data D under a specific construction environment information H, the likelihood function;

[0151] P(H) represents the prior probability of the construction environment information, which is modeled based on historical data;

[0152] P(D) represents the total probability of the sensor data and is used for normalization.

[0153] This algorithm is used to integrate shield machine sensor data (such as thrust, torque, settlement, etc.), provide a more stable and noise-resistant construction environment assessment, can effectively improve data accuracy, and reduce the impact of single-sensor errors on construction parameter adjustment.

[0154] 3. Trajectory optimization calculation process:

[0155] (1) Deep reinforcement learning (DRL)

[0156] ;

[0157] Training process:

[0158] State space St:

[0159] St = (θt, φt, ψt, vt, Tt)

[0160] Action space at:

[0161] at = (N, F, P)

[0162] where: N: Cutter head rotation speed adjustment (1 - 3 rpm)

[0163] F: Shield machine propulsion force (5 - 15 MPa)

[0164] P: Steering cylinder pressure adjustment (20 - 100 kPa)

[0165] Reward function:

[0166] ;

[0167] where α, β, γ ensure the minimum attitude error.

[0168] After 20,000 rounds of training:

[0169] The attitude deviation is reduced by 40% (from 0.5° to 0.3°)

[0170] The tunneling trajectory error is reduced from 30 mm to 10 mm (improving the accuracy by 66%)

[0171] (2) Fuzzy PID control strategy

[0172] Use fuzzy PID for real-time attitude correction:

[0173] ;

[0174] Error adjustment rule:

[0175] If |e(t)| > 2°, increase Kp for rapid adjustment;

[0176] If |e(t)| < 0.5°, decrease Kd to avoid oscillation.

[0177] As can be seen from the above, the implementation effects include

[0178] The construction time is reduced by 15%, saving a large amount of construction costs.

[0179] The number of shield machine attitude adjustments is reduced by 50%, reducing the risk of segment misalignment.

[0180] The tunnel inner contour error is reduced by 66% (from 30 mm to 10 mm).

[0181] Example 3

[0182] Intelligent optimization of synchronous grouting for shield tunnels with extreme radii:

[0183] 1. Implementation background:

[0184] During the shield construction of a certain high-speed rail underground passage, it passed through soft soil layers with an extreme radius of only 300 m. With traditional manual adjustment of synchronous grouting parameters, it was difficult to control the penetration range of the grout, resulting in ground instability and shield tail leakage. In this example, particle swarm optimization (PSO) + genetic algorithm (GA-MOO) is used for synchronous grouting optimization to reduce ground disturbance.

[0185] 2. Data acquisition method:

[0186] (1) Layout of monitoring points:

[0187] Fifty groups of pore water pressure sensors are arranged around the tunnel:

[0188] The sampling parameters are shown in Table 2:

[0189] Table 2

[0190]

[0191] (2) Calculation of soil layer deformation (finite element simulation FEM)

[0192] Calculate the stress distribution of the soil around the tunnel:

[0193] ;

[0194] Among them:

[0195] E: Elastic modulus of the soil layer;

[0196] ν: Poisson's ratio;

[0197] u: Soil layer displacement;

[0198] 3. Construction parameter optimization calculation process:

[0199] (1) Construction parameter optimization (PSO + GA-MOO)

[0200] Optimize the grouting pressure P and flow rate Q as the objectives:

[0201] ;

[0202] f1(x): Minimize ground deformation;

[0203] f2(x): Maximize the uniformity of the grout;

[0204] Particle swarm optimization update:

[0205] ;

[0206] Iterate 500 rounds to obtain the optimum:

[0207] Grouting pressure: 200 - 250 kPa;

[0208] Grouting flow rate: 12 - 15 L / min;

[0209] (2) Intelligent grouting control of shield machine:

[0210] ;

[0211] If the seepage pressure > 250 kPa, reduce the grouting pressure by 10%;

[0212] If there is leakage at the shield tail, adjust the water - cement ratio of the slurry (W / C = 0.6);

[0213] As can be seen from the above, the implementation effects include:

[0214] The ground settlement is reduced by 60% (from 10 mm to 4 mm).

[0215] The leakage at the shield tail is reduced by 90%, improving the construction safety.

[0216] The utilization rate of the slurry is increased by 20%, saving a large amount of material costs.

[0217] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An intelligent control method for a shield tunnel with a limited radius in a narrow space, characterized in that: include: Acquire multi-source data during shield construction, and use multi-modal data fusion algorithm to fuse the data to obtain shield machine status information; Collect formation pressure, water content, and soil deformation data, and use a formation dynamic evolution neural network model combined with shield machine status information to predict formation response during shield tunneling and obtain formation characteristic information; The formula of the formation dynamic evolution neural network model is: ； ； Where: X t represents the input stratum monitoring data at construction time t; h t represents the hidden state at time t, h t-1 represents the hidden state at time t-1, indicating the formation evolution information; W h , W x , W y Represents the network weight parameters; b h , b y represents the bias term; Y t represents the predicted formation response at time t; σ represents the activation function; Based on the formation characteristic information, reinforcement learning is used to optimize the excavation parameters, and the optimal construction adjustment strategy is obtained by combining the digital twin simulation analysis; based on the formation characteristic information, the reinforcement learning algorithm is used to optimize the excavation trajectory to obtain the optimal excavation path; the thrust, torque, cutter head speed, and synchronous grouting pressure parameters are comprehensively considered, and a multi-objective optimization method based on a genetic algorithm is used to optimize the construction strategy to obtain the optimal construction adjustment strategy; Based on the optimal construction adjustment strategy, deep reinforcement learning combined with fuzzy PID control is used to calculate the optimal excavation trajectory of the shield machine, and an adaptive construction plan with a limit radius in a narrow space is obtained; based on the optimal construction adjustment strategy, special environment adaptability optimization is performed to obtain an adaptive construction plan with a limit radius in a narrow space, including the following steps: according to the tunnel curvature radius and the minimum construction space constraint, a dynamic adjustment model of the shield posture is constructed; a nonlinear trajectory optimization algorithm is used to calculate the optimal adjustment strategy of the shield posture in the limit radius area; combined with stratum deformation feedback, the construction parameters are adjusted in real time to improve the adaptability of the shield construction with a limit radius in a narrow space; The expression of deep reinforcement learning is: ； Where Q(st,at) represents the value of executing action at in state st; α represents the learning rate, which controls the update speed; rt represents the reward of the current step; γ represents the discount factor, which weighs the impact of future rewards on current decisions; Represents the value of the optimal action in the next state st+1.

2. The intelligent control method for a narrow space extreme radius shield tunnel according to claim 1 is characterized in that: The multimodal data fusion algorithm is Bayesian estimation, and the formula of Bayesian estimation is: ； P(H|D) represents the posterior probability of the construction environment information H under the sensor data D; P(D|H) represents the possibility of sensor data D being observed under specific construction environment information H, the likelihood function; P(H) represents the prior probability of construction environment information, which is modeled based on historical data; P(D) represents the total probability of sensor data and is used for normalization.

3. The intelligent control method for a narrow space extreme radius shield tunnel according to claim 1 is characterized in that: The expression of the multi-objective optimization method based on genetic algorithm includes: Fitness calculation: ； Where: fi(x) represents the i-th optimization objective; wi represents the weight of the target; Crossover operation: ； Among them: xnew is the new construction parameter combination after the intersection; x1, x2 parent parameter set; Mutation operation: ； Among them, xmut is the new construction parameter combination after the mutation operation; N(0,σ 2 ) represents the random variation term that follows Gaussian distribution.

4. The intelligent control method for a narrow space extreme radius shield tunnel according to claim 1 is characterized in that: The expression combined with fuzzy PID control is: ； Where: e(t) represents the attitude error of the shield machine; Kp, Ki, and Kd represent PID control parameters, which are dynamically adjusted by fuzzy logic rules; The fuzzy logic adjustment rules are as follows: If |e(t)| is large, increase Kp to speed up the response; If e(t) remains unchanged for a long time, increase Ki to reduce the steady-state error; If e(t) changes too quickly, increase Kd to suppress oscillation.

5. The intelligent control method for a narrow space extreme radius shield tunnel according to claim 1 is characterized in that: The multi-source data include shield machine posture, thrust, torque, tunneling speed, cutter head rotation speed, formation resistance and shield tail clearance.

Images