Small distance parallel shield low disturbance collaborative control method based on previous tunnel protection

Abstract

This invention discloses a low-disturbance collaborative control method for parallel shield tunneling with small clearance based on prior tunnel protection. The method includes the construction of a comprehensive shield tunneling collaborative control system, feedforward control, real-time deviation monitoring, and feedback control. By constructing an integrated collaborative control system and employing a neural network model combining pre-training and online learning, control parameters can be dynamically optimized based on geological parameters and real-time monitoring data. This reduces reliance on manual experience, lowers construction risks caused by geological uncertainties, and improves the tunneling reliability of ultra-large diameter shield tunnels under harsh geological conditions and parallel shield tunneling with small clearance. It effectively prevents structural damage such as tunnel segment cracks, leakage, or axial misalignment, providing key technical support for the safe operation of urban infrastructure.

Description

Technical Field

[0001] This invention belongs to the field of underground engineering construction technology, and relates to a low-disturbance collaborative control method for parallel shield tunnels with small clearance based on prior tunnel protection. Background Technology

[0002] In urban underground space development and cross-river / sea tunnel construction, ultra-large diameter slurry balance shield tunneling machines with an excavation diameter of no less than 12 meters have become key equipment for achieving one-time formation of large-section tunnels. However, when this type of equipment is excavating in complex strata such as soft soil, gravel, or soft upper layers with hard lower layers, the processes of cutterhead cutting, slurry pressure fluctuations, shield tail gap formation, and synchronous grouting can easily disrupt the original stress balance of the strata in front of the excavation face and around the tunnel, inducing significant stratum displacement and surface subsidence or uplift, causing considerable disturbance to the environment along the construction route. These problems are particularly prominent when large-section tunnels face parallel shield tunneling construction with small clearances. Factors such as stratum stress redistribution, slurry pressure fluctuations, shield propulsion reaction force, and synchronous grouting pressure caused by the excavation of the subsequently constructed large-section tunnel will be transmitted to the preceding tunnel structure through the soil. If the stratum disturbance is not properly controlled, causing deformation to exceed the safety threshold, it may not only threaten the long-term safety of the tunnel itself, but also directly endanger the normal use and safety of existing parallel shield tunnels, resulting in significant engineering risks and social impacts.

[0003] Currently, control technologies for parallel shield tunneling with small clearances mostly focus on local adjustments of single construction parameters or passive responses based on experience, lacking coordinated control and intelligent decision-making among multiple systems such as the cutterhead system, propulsion system, slurry system, and grouting system. Existing methods often fail to establish a feedforward prediction mechanism for the impact of the tunneling process on the preceding tunnel, and also lack feedback and adaptive capabilities based on real-time monitoring data. This results in insufficient control accuracy and delayed response when tunneling in close parallel or crossing situations, making it difficult to achieve minimal or no disturbance construction.

[0004] In summary, existing technologies are insufficient to meet the high-precision, low-disturbance control requirements of parallel shield tunneling projects with small clearances. There is an urgent need for a collaborative control method that can systematically integrate multi-source monitoring information and possess feedforward prediction and real-time feedback optimization capabilities. This method would allow for refined collaborative control of the subsequent shield tunneling process, minimizing disturbances to the preceding tunnel and ensuring the structural safety and long-term stable operation of the parallel tunnels. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the prior art and propose a low-disturbance collaborative control method for parallel shield tunneling with small clearance based on prior tunnel protection. This method constructs an integrated collaborative control system and combines feedforward control, feedback control and model adaptive optimization to achieve precise management of the shield tunneling process, thereby minimizing disturbance to the strata.

[0006] The low-disturbance collaborative control method for parallel shield tunneling with small clearance based on prior tunnel protection includes the following steps:

[0007] S1. Construction of the overall shield tunneling collaborative control system:

[0008] A shield tunneling collaborative control system is constructed, including a data monitoring system, an intelligent decision-making system, a grouting system, and a cutterhead system. The systems interact with each other in real time and transmit commands through a high-speed network.

[0009] The data monitoring system is responsible for data acquisition throughout the entire construction process. It includes ground-penetrating radar, pressure gauges, flow meters integrated into the grouting system, and settlement meters installed on the ground surface, which are used to obtain the density of the grout body behind the wall. C 1 Grouting pressure C 2 and grouting volume C 3 and the subsidence of surface monitoring sites S i ;

[0010] The intelligent decision-making system incorporates a first neural network model and a second neural network model with pre-training and online learning capabilities.

[0011] The grouting system is used to control the bentonite slurry mix ratio. F 1 Grouting pressure F 2 Grouting volume F 3 And perform backfill grouting;

[0012] The cutterhead system is equipped with a rotating, variable-diameter edge cutter mechanism for performing tunneling operations and adjusting the excavation diameter in real time. D 1 ;

[0013] S2, Feedforward Control:

[0014] Before the tunnel boring machine begins each tunneling cycle into the sensitive area, the first neural network model is based on the feedforward control input set. O 1 =( D 2 , D 3 , D 4 , E 1 , E 2 , E 3 , E4 , E 5 , E 6 Predictive analysis is performed to generate feedforward control parameters. P 1 =( D 1 , F 1 , F 2 , F 3 The command is then sent to the grouting system and the cutterhead system and driven to execute; wherein... E 1 , E 2 , E 3 , E 4 , E 5 and E 6 These are the cohesion, internal friction angle, elastic modulus, density, water content, and porosity of the strata to be traversed, which were obtained in advance through exploration. D 2 For the tunnel boring machine (TBM) construction mileage, D 3 Design diameter for the tunnel, D 4 For tunnel burial depth;

[0015] S3. Real-time Deviation Monitoring:

[0016] During the tunnel boring machine's excavation process, the density of the grout behind the tunnel wall is obtained through the data monitoring system. C 1 Grouting pressure C 2 Grouting volume C 3 Subsidence at surface monitoring sites S i To form a monitoring dataset R =( C 1 , C 2 , C 3 , S i ), and then with O 1 and P 1 Together they constitute the feedback control input set O 2 =( O1 , P 1 , R );

[0017] S4, Feedback Control:

[0018] The second neural network model controls the input set based on feedback. O 2 Perform real-time calculations to generate feedback control parameters. P 2 =( D 1 +Δ D 1 , F 1 +Δ F 1 , F 2 +Δ F 2 , F 3 +Δ F 3 ), and issue the command to the grouting system and the cutterhead system for execution; wherein, Δ D 1 Δ F 1 Δ F 2 and Δ F 3 They are respectively D 1 , F 1 , F 2 and F 3 The adjustment amount.

[0019] Preferably, the ground-penetrating radar is arranged on the track of the tunnel boring machine's excavation section and is driven by a servo motor to move circumferentially along the track for scanning and detecting the entire cross-section of the grouting body behind the wall. The radar system integrates dual transmitting devices for ultrasonic and high-frequency electromagnetic waves, which alternately transmit beams and receive reflected signals. The ground-penetrating radar obtains the density of the grouting body behind the wall by transmitting, receiving, and analyzing the propagation speed and amplitude of sound waves and electromagnetic waves. C 1 The mapping relationship between propagation speed, amplitude and density was calibrated through indoor tests; the track is made of corrosion-resistant alloy material and is installed on the inner side of the shield tail of the tunnel boring machine, operating synchronously with the grouting system.

[0020] Preferably, the edge hobbing mechanism adopts a hydraulically driven telescopic hobbing arm, which is evenly distributed on the edge of the cutter head, and the single arm telescopic stroke is 0 to 200 mm.

[0021] Preferably, the bentonite grout used in the grouting system is composed of the following components in parts by weight, wherein the sum of the parts by weight of each component may not be 100 parts: 90 to 95 parts bentonite, 2 to 3 parts expanded perlite, 1 to 2 parts sodium fatty alcohol polyoxyethylene ether sulfate, 5 to 7 parts polytetrafluoroethylene, 4 to 6 parts bakelite powder, 1 to 2 parts silicone oil, 4.5 to 6 parts itaconic acid monobutyl ester, 2 to 3 parts molybdenum disulfide, 1.5 to 2.5 parts zinc naphthenate, 1 to 1.5 parts beeswax, and 3 to 4 parts modified diatomaceous earth, wherein the modified diatomaceous earth is 3-aminopropyltriethoxysilane. Alkane is used as a modifier, and the amount of modifier is 5% of the mass of diatomaceous earth. The preparation method of the bentonite slurry includes the following steps: first, bentonite, expanded perlite and molybdenum disulfide are ground together and sieved; then, sodium fatty alcohol polyoxyethylene ether sulfate, polytetrafluoroethylene and bakelite powder are initially mixed at 80°C to 90°C, then the remaining components are added and the temperature is raised to 130°C to 160°C for secondary mixing; finally, the materials obtained from the two treatments are mixed, water is added and ground to form a suspension, then modified diatomaceous earth is added and stirred, and finally dried and pulverized to obtain a finished powder of 200 to 400 mesh.

[0022] Preferably, the first neural network model is a pre-trained deep recurrent neural network model with online learning capabilities; the architecture of the first neural network model includes a first input layer, a first hidden layer, and a first output layer; the first input layer is used to receive the feedforward control input set. O 1 The first hidden layer uses a recurrent structure composed of 3 to 5 layers of long short-term memory units, with 128 to 256 neurons per layer. Residual connections are introduced between layers to alleviate the vanishing gradient problem and improve training stability. The first output layer is mapped to feedforward control parameters through a fully connected layer. P 1 A linear activation function is used to ensure that the parameters are continuously adjustable. The training of the first neural network model includes a pre-training phase and an online learning phase. In the pre-training phase, at least 1,000 sets of historical construction data from ultra-large diameter slurry balance shield tunneling are collected to construct a training sample set, covering historical data under three geological conditions: sand, silty clay, and silty clay. O 1 , P 1 and RThe network weights are optimized using the backpropagation algorithm. The loss function during training is defined as the mean squared error between the output and label values. The initial learning rate is 0.001, which decays to 0.9 times the previous rate every 1000 iterations. The iteration terminates when the loss function value is less than 0.0001. The online learning phase occurs during the actual tunneling process, using the monitoring dataset acquired in real time through the data monitoring system. R When any parameter deviates from the preset threshold, the model parameters of the first neural network model are updated through an incremental learning algorithm based on gradient accumulation. The learning rate is 1 / 10 of that in the pre-training stage, ensuring the adaptive optimization capability of the formation.

[0023] Preferably, the second neural network model is a pre-trained deep neural network with online learning capabilities; the architecture of the second neural network model includes a second input layer, a second hidden layer, and a second output layer; the second input layer is used to receive the feedforward control input set. O 1 The feedforward control parameters P 1 and the monitoring dataset R The data is then standardized and fused to form the feedback control input set. O 2 =( O 1 , P 1 , R The second hidden layer employs a hybrid structure of convolutional neural network branches and gated recurrent unit branches. The convolutional neural network branch has two convolutional layers with kernel sizes of 3×3 and 5×5, both with a stride of 1, and uses max pooling to extract the feedback control input set. O 2 The local spatial features in the convolutional neural network are analyzed. The gated recurrent unit branch has a single gated recurrent layer with 128 neurons and a stride of 2, used to extract dynamic sequence features that change over time. The convolutional neural network branch and the gated recurrent unit branch are weighted at a 1:1 ratio and then input into a fully connected layer. The second output layer outputs the feedback control parameters through a linear fully connected layer. P 2 Furthermore, a residual connection structure is introduced to accelerate convergence; in the training of the second neural network model, at least 1000 sets of historical construction data from ultra-large diameter slurry balance shield tunneling are first collected to construct a training sample set, covering three geological conditions: sandy soil, silty clay, and silty clay. O 2 and P 2The network weights are optimized using the backpropagation algorithm. The loss function during training is defined as the mean square error between the output value and the label value. The initial learning rate is 0.001, which decays to 0.9 times the previous rate every 1000 iterations. The iteration terminates when the loss function value is less than 0.0001.

[0024] In summary, compared with existing technologies, the low-disturbance collaborative control method for parallel shield tunneling with small clearance based on prior tunnel protection provided by this invention, including the construction of the overall shield tunneling collaborative control system, feedforward control, real-time deviation monitoring, and feedback control, has achieved the following significant beneficial effects:

[0025] 1) Achieved intelligent collaborative control of multiple systems, improving control accuracy and response speed: By constructing an integrated collaborative control system, data monitoring, intelligent decision-making, grouting and cutterhead systems are interconnected through a high-speed network, realizing real-time data interaction and command coordination among multiple subsystems; This method breaks through the limitations of traditional single-system local adjustment, using a first neural network model for feedforward prediction and a second neural network model for real-time feedback control, forming a feedforward-feedback closed-loop optimization mechanism, thereby predicting changes in tunneling parameters in advance and dynamically adjusting grouting parameters and cutterhead excavation diameter, effectively avoiding control response lag problems, improving control accuracy to the millimeter level, significantly reducing ground disturbance during tunneling and protecting the preceding tunnel;

[0026] 2) Significantly reduces the risk of ground deformation: The high-precision data monitoring system acquires real-time data on the density, pressure, volume, and surface settlement of the grout behind the tunnel wall, and combines this with the adaptive optimization capabilities of the intelligent decision-making system to achieve proactive control of the stability of the excavation face. Especially in parallel shield tunneling with small clearance, this method can control ground deformation within the range of micro-disturbance, effectively preventing structural damage such as tunnel segment cracks, leakage, or axial misalignment, providing key technical support for the safe operation of urban infrastructure.

[0027] 3) Enhance adaptability to complex geological conditions: The neural network model, which combines pre-training and online learning, can dynamically optimize control parameters based on geological parameters and real-time monitoring data. Through adaptive learning of the model, it can quickly adapt to changes in strata, reduce reliance on human experience, reduce construction risks caused by geological uncertainties, and improve the tunneling reliability of ultra-large diameter shield tunnels under harsh geological conditions and parallel shield tunneling with small clearance.

[0028] 4) Optimize grouting material performance: The grouting system uses a specially formulated bentonite grout. By adding modified diatomaceous earth, polytetrafluoroethylene and other components, the fluidity, stability and filling density of the grout are significantly improved. Combined with the dual-beam scanning technology of ground penetrating radar, the density of the grout body can be monitored in real time to ensure that the grout behind the wall is uniform and full. This synergistic optimization of materials and monitoring technology effectively reduces the changes in formation stress caused by grouting pressure fluctuations, reduces grout waste, and improves the efficiency and economy of grouting operations. Attached Figure Description

[0029] Figure 1 This is a flowchart illustrating the low-disturbance collaborative control method for parallel shield tunneling with small clearance based on prior tunnel protection, as shown in an embodiment of the present invention.

[0030] Figure 2 This is a schematic diagram of the connection of the shield tunneling collaborative control system according to an embodiment of the present invention;

[0031] Figure 3 This is a schematic diagram of a ground-penetrating radar arranged on the track of a tunnel boring machine excavation section, as shown in an embodiment of the present invention.

[0032] Figure 4 This is a schematic diagram of the cutterhead system during normal tunneling, as shown in an embodiment of the present invention.

[0033] Figure 5 This is a schematic diagram of the cutter head system during retraction, as shown in an embodiment of the present invention.

[0034] Figure 6 This is a schematic diagram showing the extension (left figure) and retraction (right figure) of the reset mechanism of the cutter head system according to an embodiment of the present invention;

[0035] Figure 7 This is a schematic diagram of the preliminary tunnel data monitoring system shown in an embodiment of the present invention;

[0036] Among them, 1-data monitoring system, 11-ground penetrating radar, 12-pressure gauge, 13-flow meter, 14-settlement meter, 15-strain gauge, 16-displacement sensor, 2-intelligent decision-making system, 21-first neural network model, 22-second neural network model, 3-grouting system, 4-cutter head system, 41-ball head cutter box, 42-cutter head, 43-connecting rod, 44-telescopic rod, 45-locking pin, 46-reset spring, 47-reset ring, 48-seal, 49-hydraulic cylinder, 50-cutter beam, 51-cylinder seat. Detailed Implementation

[0037] The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments given herein are for illustration and explanation only and are not intended to limit the present invention.

[0038] This application discloses, as follows: Figure 1-7 The low-disturbance collaborative control method for parallel shield tunneling with small clearance based on prior tunnel protection, as shown, includes the following steps:

[0039] The low-disturbance collaborative control method for parallel shield tunneling with small clearance based on prior tunnel protection includes the following steps:

[0040] S1. Construction of the shield tunneling collaborative control system: Construct a shield tunneling collaborative control system, including a data monitoring system 1, an intelligent decision-making system 2, a grouting system 3, and a cutterhead system 4. The systems interact with each other in real time and transmit commands via an industrial Ethernet with a transmission rate of not less than 100Mbps to ensure the real-time performance and reliability of data interaction and command transmission.

[0041] The data monitoring system 1 is responsible for data acquisition throughout the construction process, including a ground-penetrating radar 11, a pressure gauge 12, a flow meter 13 integrated into the grouting system 3, and a settlement meter 14 installed on the ground surface, which are used to obtain the density of the grout body behind the wall. C 1 Grouting pressure C 2 and grouting volume C 3 and the subsidence of surface monitoring sites S i In specific implementation, a corrosion-resistant alloy track is installed circumferentially on the inner wall of the shield tail of the tunnel boring machine. The ground-penetrating radar 11 is arranged on the corrosion-resistant alloy track and is driven by a servo motor to move circumferentially along the track for scanning and detecting the entire cross-section of the grouting body behind the wall. The radar system integrates ultrasonic and high-frequency electromagnetic wave dual-transmitting devices, wherein the ultrasonic wave transmission frequency is 50 to 100 kHz and the high-frequency electromagnetic wave frequency is 1 to 2 GHz, by alternately transmitting beams and receiving reflected signals. The ground-penetrating radar obtains the density of the grouting body behind the wall by transmitting, receiving and analyzing the propagation speed and amplitude of the sound waves and electromagnetic waves. C 1 The mapping relationship between propagation speed, amplitude, and density was calibrated through indoor tests. The track, made of corrosion-resistant alloy, is installed inside the tail of the tunnel boring machine and operates synchronously with the grouting system. A surface monitoring network is established every 10 meters along the tunnel axis to continuously measure settlement. S i .

[0042] In specific implementation, in the parallel section with small clearance, the data monitoring system 1 also includes strain gauges and displacement sensors installed at key sections inside the pilot tunnel, which are used to monitor the circumferential stress of the pilot tunnel in real time. C 4 and radial deformationC 5 The strain gauges and displacement sensors are installed at a spacing of 10 meters, and the data sampling frequency is 1 Hz to ensure high-resolution monitoring.

[0043] The intelligent decision-making system 2 relies on a high-performance industrial server and has a built-in first neural network model 21 and a second neural network model 22 with pre-training and online learning functions.

[0044] In specific implementation, the first neural network model 21 is a pre-trained deep recurrent neural network model with online learning capabilities; the architecture of the first neural network model 21 includes a first input layer, a first hidden layer, and a first output layer; the first input layer is used to receive the feedforward control input set. O 1 The first hidden layer uses a recurrent structure composed of 3 to 5 layers of long short-term memory units, with 128 to 256 neurons per layer. Residual connections are introduced between layers to alleviate the vanishing gradient problem and improve training stability. The first output layer is mapped to feedforward control parameters through a fully connected layer. P 1 A linear activation function is used to ensure that the parameters are continuously adjustable. The training of the first neural network model 21 includes a pre-training phase and an online learning phase. In the pre-training phase, at least 1,000 sets of historical construction data for ultra-large diameter slurry balance shield tunneling are first collected to construct a training sample set, covering historical data under three geological conditions: sand, silty clay, and silty clay. O 1 , P 1 and R The network weights are optimized using the backpropagation algorithm. The loss function during training is defined as the mean squared error between the output and label values. The initial learning rate is 0.001, which decays to 0.9 times the previous rate every 1000 iterations. The iteration terminates when the loss function value is less than 0.0001. The online learning phase occurs during the actual tunneling process, when the monitoring dataset is acquired in real time through the data monitoring system 1. R When any parameter deviates from a preset threshold, the model parameters of the first neural network model 21 are updated using an incremental learning algorithm based on gradient accumulation. The learning rate is 1 / 10 of that in the pre-training phase, ensuring adaptive optimization capability for the formation. The monitoring dataset... R The preset threshold for any parameter is determined based on the deformation limit of the sensitive structure and the specification requirements of the grouting body behind the wall.

[0045] In specific implementation, the second neural network model 22 is a pre-trained deep neural network with online learning capabilities; the architecture of the second neural network model includes a second input layer, a second hidden layer, and a second output layer; the second input layer is used to receive the feedforward control input set. O 1 The feedforward control parameters P 1 and the monitoring dataset R The data is then standardized and fused to form the feedback control input set. O 2 =( O 1 , P 1 , R The second hidden layer employs a hybrid structure of convolutional neural network branches and gated recurrent unit branches. The convolutional neural network branch has two convolutional layers with kernel sizes of 3×3 and 5×5, both with a stride of 1, and uses max pooling to extract the feedback control input set. O 2 The local spatial features in the convolutional neural network are analyzed. The gated recurrent unit branch has a single gated recurrent layer with 128 neurons and a stride of 2, used to extract dynamic sequence features that change over time. The convolutional neural network branch and the gated recurrent unit branch are weighted at a 1:1 ratio and then input into a fully connected layer. The second output layer outputs the feedback control parameters through a linear fully connected layer. P 2 Furthermore, a residual connection structure is introduced to accelerate convergence; in the training of the second neural network model, at least 1000 sets of historical construction data from ultra-large diameter slurry balance shield tunneling are first collected to construct a training sample set, covering three geological conditions: sandy soil, silty clay, and silty clay. O 2 and P 2 The network weights are optimized using the backpropagation algorithm. The loss function during training is defined as the mean square error between the output value and the label value. The initial learning rate is 0.001, which decays to 0.9 times the previous rate every 1000 iterations. The iteration terminates when the loss function value is less than 0.0001.

[0046] The grouting system 3 is used to control the bentonite slurry mix ratio. F 1 Grouting pressure F 2 Grouting volume F 3The wall-mounted grouting operation is then performed. Specifically, the bentonite grout used in the grouting system 3 is composed of the following components in parts by weight, wherein the sum of the parts by weight of each component may not be 100 parts: 90-95 parts bentonite, 2-3 parts expanded perlite, 1-2 parts sodium fatty alcohol polyoxyethylene ether sulfate, 5-7 parts polytetrafluoroethylene, 4-6 parts bakelite powder, 1-2 parts silicone oil, 4.5-6 parts itaconic acid monobutyl ester, 2-3 parts molybdenum disulfide, 1.5-2.5 parts zinc naphthenate, 1-1.5 parts beeswax, and 3-4 parts modified diatomaceous earth, wherein the modified diatomaceous earth is 3-ammonia-reduced diatomaceous earth. Propyltriethoxysilane is used as a modifier, and the amount of modifier is 5% of the mass of diatomaceous earth. The preparation method of the bentonite slurry includes the following steps: first, bentonite, expanded perlite and molybdenum disulfide are ground together and sieved; then, sodium fatty alcohol polyoxyethylene ether sulfate, polytetrafluoroethylene and bakelite powder are initially mixed at 80°C to 90°C, then the remaining components are added and the temperature is raised to 130°C to 160°C for secondary mixing; finally, the materials obtained from the two treatments are mixed, water is added and ground to form a suspension, then modified diatomaceous earth is added and stirred, and finally dried and pulverized to obtain a finished powder of 200 to 400 mesh.

[0047] The cutterhead system is equipped with a rotating, variable-diameter edge cutter mechanism for performing tunneling operations and adjusting the excavation diameter in real time. D 1 In specific implementation, the edge hobbing mechanism adopts 6 to 8 sets of hydraulically driven telescopic hobbing arms, which are evenly distributed on the edge of the cutter head, with a single arm telescopic stroke of 0 to 200 mm. In specific implementation, the cutter head system 4 includes a ball-end cutter box 41, a cutter head 42, a connecting rod 43, a telescopic rod 44, a locking pin 45, a return spring 46, a return ring 47, a seal 48, a hydraulic cylinder 49, a cutter beam 50, and a cylinder seat 51.

[0048] S2, Feedforward Control: Before the tunnel boring machine begins each tunneling cycle into the sensitive area, the first neural network model, based on the feedforward control input set... O 1 =( D 2 , D 3 , D 4 , E 1 , E 2 , E 3 , E 4 , E 5 , E 6 Predictive analysis is performed to generate feedforward control parameters.P 1 =( D 1 , F 1 , F 2 , F 3 The command is then sent to the grouting system and the cutterhead system and driven to execute; wherein... E 1 , E 2 , E 3 , E 4 , E 5 and E 6 These are the cohesion, internal friction angle, elastic modulus, density, water content, and porosity of the strata to be traversed, which were obtained in advance through exploration. D 2 For the tunnel boring machine (TBM) construction mileage, D 3 Design diameter for the tunnel, D 4 For tunnel burial depth;

[0049] S3. Real-time deviation monitoring: During the tunnel boring machine's excavation process, the density of the grout behind the wall is obtained through the data monitoring system. C 1 Grouting pressure C 2 Grouting volume C 3 Subsidence at surface monitoring sites S i To form a monitoring dataset R =( C 1 , C 2 , C 3 , S i ), and then with O 1 and P 1 Together they constitute the feedback control input set O 2 =( O 1 , P 1 , R );

[0050] In practice, in the parallel sections with small clearances, the circumferential stress of the preceding tunnel is also obtained through the data monitoring system. C 4 and radial deformation C 5 This forms an expanded monitoring dataset. R =( C 1 , C 2 , C 3 , S i , C 4 , C 5 ).

[0051] S4. Feedback Control: The second neural network model controls the input set based on feedback. O 2 Perform real-time calculations to generate feedback control parameters. P 2 =( D 1 +Δ D 1 , F 1 +Δ F 1 , F 2 +Δ F 2 , F 3 +Δ F 3 ), and issue the command to the grouting system and the cutterhead system for execution; wherein, Δ D 1 Δ F 1 Δ F 2 and Δ F 3 They are respectively D 1 , F 1 , F 2 and F 3 The adjustment amount.

[0052] The above describes one or more embodiments of the present invention in a relatively specific and detailed manner, but it should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this patent should be determined by the appended claims.

Claims

1. A low-disturbance collaborative control method for parallel shield tunneling with small clearance based on prior tunnel protection, characterized in that, Includes the following steps: S1. Construction of the Shield Tunneling Cooperative Control System: A shield tunneling cooperative control system is constructed, including a data monitoring system, an intelligent decision-making system, a grouting system, and a cutterhead system. These systems interact and transmit commands in real time via a high-speed network. The data monitoring system is responsible for data acquisition throughout the entire construction process, including ground-penetrating radar, pressure gauges, flow meters integrated into the grouting system, and a settlement meter installed on the ground surface, used to obtain the density of the grout behind the tunnel wall. C 1 Grouting pressure C 2 and grouting volume C 3 and the subsidence of surface monitoring sites S i The intelligent decision-making system incorporates a first neural network model and a second neural network model with pre-training and online learning capabilities. The first neural network model is a pre-trained deep recurrent neural network model with online learning capabilities. The architecture of the first neural network model includes a first input layer, a first hidden layer, and a first output layer. The first input layer is used to receive the feedforward control input set. O 1 The first hidden layer uses a recurrent structure composed of 3 to 5 layers of long short-term memory units, with 128 to 256 neurons per layer. Residual connections are introduced between layers to alleviate the vanishing gradient problem and improve training stability. The first output layer is mapped to feedforward control parameters through a fully connected layer. P 1 A linear activation function is used to ensure that the parameters are continuously adjustable. The training of the first neural network model includes a pre-training phase and an online learning phase. In the pre-training phase, at least 1,000 sets of historical construction data from ultra-large diameter slurry balance shield tunneling are collected to construct a training sample set, covering historical data under three geological conditions: sand, silty clay, and silty clay. O 1 , P 1 and R The network weights are optimized using the backpropagation algorithm. The loss function during training is defined as the mean squared error between the output and label values. The initial learning rate is 0.001, which decays to 0.9 times the previous rate every 1000 iterations. The iteration terminates when the loss function value is less than 0.0001. The online learning phase occurs during the actual tunneling process, using the monitoring dataset acquired in real time through the data monitoring system. R When any parameter deviates from the preset threshold, the model parameters of the first neural network model are updated using an incremental learning algorithm based on gradient accumulation. The learning rate is 1 / 10 of that in the pre-training stage, ensuring adaptive optimization capability for the formation. The grouting system is used to control the bentonite slurry mix ratio. F 1 Grouting pressure F 2 Grouting volume F 3 The cutterhead system is equipped with a rotating, variable-diameter edge cutter mechanism for performing tunneling operations and adjusting the excavation diameter in real time. D 1 ; S2, Feedforward Control: Before the tunnel boring machine begins each tunneling cycle into the sensitive area, the first neural network model, based on the feedforward control input set... O 1 =( D 2 , D 3 , D 4 , E 1 , E 2 , E 3 , E 4 , E 5 , E 6 Predictive analysis is performed to generate feedforward control parameters. P 1 =( D 1 , F 1 , F 2 , F 3 The command is then sent to the grouting system and the cutterhead system and driven to execute; wherein... E 1 , E 2 , E 3 , E 4 , E 5 and E 6 These are the cohesion, internal friction angle, elastic modulus, density, water content, and porosity of the strata to be traversed, which were obtained in advance through exploration. D 2 For the tunnel boring machine (TBM) construction mileage, D 3 Design diameter for the tunnel, D 4 For tunnel burial depth; S3. Real-time deviation monitoring: During the tunnel boring machine's excavation process, the density of the grout behind the wall is obtained through the data monitoring system. C 1 Grouting pressure C 2 Grouting volume C 3 Subsidence at surface monitoring sites S i To form a monitoring dataset R =( C 1 , C 2 , C 3 , S i ), and then with O 1 and P 1 Together they constitute the feedback control input set O 2 =( O 1 , P 1 , R ); S4. Feedback Control: The second neural network model controls the input set based on feedback. O 2 Perform real-time calculations to generate feedback control parameters. P 2 =( D 1 +Δ D 1 , F 1 +Δ F 1 , F 2 +Δ F 2 , F 3 +Δ F 3 ), and issue the command to the grouting system and the cutterhead system for execution; wherein, Δ D 1 Δ F 1 Δ F 2 and Δ F 3 They are respectively D 1 , F 1 , F 2 and F 3 The adjustment amount.

2. The low-disturbance collaborative control method for parallel shield tunneling with small clearance based on prior tunnel protection as described in claim 1, characterized in that, The ground-penetrating radar (GPR) is positioned on the track of the tunnel boring machine (TBM) excavation section and moves circumferentially along the track via a servo motor. It is used to scan and detect the entire cross-section of the grouting body behind the tunnel wall. The GPR integrates dual transmitting devices for ultrasonic and high-frequency electromagnetic waves, alternately emitting beams and receiving reflected signals. By transmitting, receiving, and analyzing the propagation speed and amplitude of the sound and electromagnetic waves, the GPR obtains the density of the grouting body behind the tunnel wall. C 1 The mapping relationship between propagation speed, amplitude and density was calibrated through indoor tests.

3. The low-disturbance collaborative control method for parallel shield tunneling with small clearance based on prior tunnel protection as described in claim 1, characterized in that, The edge hobbing mechanism adopts a hydraulically driven telescopic hobbing arm, which is evenly distributed on the edge of the cutter head, with a single arm telescopic stroke of 0 to 200 mm.

4. The low-disturbance collaborative control method for parallel shield tunneling with small clearance based on prior tunnel protection as described in claim 1, characterized in that, The bentonite grout used in the grouting system is composed of the following components in parts by weight: 90 to 95 parts bentonite, 2 to 3 parts expanded perlite, 1 to 2 parts sodium fatty alcohol polyoxyethylene ether sulfate, 5 to 7 parts polytetrafluoroethylene, 4 to 6 parts bakelite powder, 1 to 2 parts silicone oil, 4.5 to 6 parts itaconic acid monobutyl ester, 2 to 3 parts molybdenum disulfide, 1.5 to 2.5 parts zinc naphthenate, 1 to 1.5 parts beeswax, and 3 to 4 parts modified diatomaceous earth. The modified diatomaceous earth uses 3-aminopropyltriethoxysilane as a modifier. The amount is 5% of the diatomaceous earth mass; the preparation method of the bentonite slurry includes the following steps: first, the bentonite, expanded perlite and molybdenum disulfide are ground together and sieved; then, the fatty alcohol polyoxyethylene ether sodium sulfate, polytetrafluoroethylene and bakelite powder are initially mixed at 80°C to 90°C, then the remaining components are added and the temperature is raised to 130°C to 160°C for secondary mixing; finally, the materials obtained from the two treatments are mixed, water is added and ground to form a suspension, then modified diatomaceous earth is added and stirred, and finally dried and pulverized to obtain a finished powder of 200 to 400 mesh.

5. The low-disturbance collaborative control method for parallel shield tunneling with small clearance based on prior tunnel protection as described in claim 1, characterized in that, The second neural network model is a pre-trained deep neural network with online learning capabilities; the architecture of the second neural network model includes a second input layer, a second hidden layer, and a second output layer; the second input layer is used to receive the feedforward control input set. O 1 The feedforward control parameters P 1 and the monitoring dataset R The data is then standardized and fused to form the feedback control input set. O 2 =( O 1 , P 1 , R The second hidden layer employs a hybrid structure of convolutional neural network branches and gated recurrent unit branches. The convolutional neural network branch has two convolutional layers with kernel sizes of 3×3 and 5×5, both with a stride of 1, and uses max pooling to extract the feedback control input set. O 2 The local spatial features in the convolutional neural network are analyzed. The gated recurrent unit branch has a single gated recurrent layer with 128 neurons and a stride of 2, used to extract dynamic sequence features that change over time. The convolutional neural network branch and the gated recurrent unit branch are weighted at a 1:1 ratio and then input into a fully connected layer. The second output layer outputs the feedback control parameters through a linear fully connected layer. P 2 Furthermore, a residual connection structure is introduced to accelerate convergence; in the training of the second neural network model, at least 1000 sets of historical construction data from ultra-large diameter slurry balance shield tunneling are first collected to construct a training sample set, covering three geological conditions: sandy soil, silty clay, and silty clay. O 2 and P 2 The network weights are optimized using the backpropagation algorithm. The loss function during training is defined as the mean square error between the output value and the label value. The initial learning rate is 0.001, which decays to 0.9 times the previous rate every 1000 iterations. The iteration terminates when the loss function value is less than 0.0001.

Images