Multi-stage intelligent split launching method for limited space slurry balance shield

Abstract

The application discloses a water conservancy limited space vertical shaft slurry balance shield multi-stage intelligent split launching method and belongs to the technical field of water conservancy tunnel shield construction. In view of the problems of extensive division of shield split launching stages, unstable slurry circulation operation, great difficulty in attitude control, lagging safety early warning and weak geological adaptability in the water conservancy limited vertical shaft scene, the application generates an optimal launching scheme through a multi-objective genetic algorithm, realizes accurate regulation and control of slurry pressure by adopting long short-term memory neural network prediction combined model prediction control, completes multi-factor coupled attitude correction by constructing a Transform-CNN model, realizes safety risk grading early warning by relying on digital twinning and edge computing, realizes adaptive adjustment of tunneling parameters by combining with advanced geological exploration, and iteratively optimizes the model through construction data closed loop. The application significantly improves the efficiency, safety and intelligent level of limited space shield launching, and is suitable for long-distance water diversion tunnel projects in mountainous areas.

Description

Technical Field

[0001] This invention relates to the technical field of tunnel boring machine (TBM) construction, specifically a multi-stage intelligent split-type launching method for slurry balance TBMs in confined spaces, and more particularly a multi-stage intelligent split-type launching method for slurry balance TBMs in confined space shafts in hydraulic engineering. Background Technology

[0002] With the continuous advancement of water conservancy projects in my country and the gradual improvement of the national water network backbone, major water conservancy projects such as long-distance water diversion and inter-basin water transfer are being implemented continuously, leading to a rapid increase in the scale and number of water conservancy tunnel projects. As the core channel of the water network, the construction efficiency and safety stability of water conservancy tunnels directly affect the construction progress and quality of the entire water conservancy project. Among water conservancy tunnel construction technologies, the slurry balance shield tunneling method, with its significant advantages such as good excavation face stability, minimal disturbance to the surrounding environment, strong adaptability to geological formations, and high construction efficiency, has become one of the preferred technologies for long-distance, large-diameter water conservancy tunnel construction, especially suitable for tunnel projects traversing complex strata such as soft soil, sand layers, and fractured rock layers.

[0003] Unlike urban subways and municipal tunnels, water conservancy tunnel projects are often located in remote mountainous areas. Constrained by multiple factors such as topography, land acquisition and demolition, project investment, and ecological protection, the construction conditions for the tunnel boring machine (TBM) launching shaft are extremely demanding. Firstly, the complex mountainous terrain results in limited space for the launching shaft, restricting its internal clearance and failing to meet the spatial requirements for the overall assembly and launching of the TBM and its supporting trolleys. Secondly, the shafts are typically deep (most exceeding 15 meters), and limitations in construction site and transportation conditions make it difficult to bring in large hoisting equipment, further exacerbating the difficulty of launching the TBM as a whole. Furthermore, water conservancy projects have strict cost control requirements; blindly expanding the shaft size would not only significantly increase project investment but also potentially damage the surrounding ecological environment, violating the principles of green construction in water conservancy projects. Therefore, under the condition of a confined vertical shaft in water conservancy, the separate launch of the tunnel boring machine (TBM) becomes an inevitable choice. That is, by disassembling the TBM main unit and the supporting trolley into multiple stages, hoisting them into the shaft in batches, assembling them in place, and gradually completing the launch operation of the TBM.

[0004] Currently, the slurry balance shield tunneling split-type launching technology has been initially applied in some water conservancy tunnel projects. However, for the special construction scenario of vertical shafts in confined spaces in water conservancy projects, the existing technology still faces many technical challenges that urgently need to be solved, making it difficult to meet the requirements of safety, efficiency, and intelligence in engineering construction. The specific problems are as follows: I. The initial phase was divided in a crude manner and lacked a scientific planning basis. Traditional shield tunneling (TBM) segmentation relies heavily on the construction experience of engineers, typically dividing the assembly of the auxiliary equipment into two or three stages without establishing a systematic evaluation framework. In the confined space of hydraulic shaft construction, factors such as shaft clearance dimensions, the size and weight of the auxiliary equipment, pipeline layout requirements for the slurry circulation system, and tunnel cross-sectional dimensions all significantly impact stage segmentation. Current technologies fail to systematically analyze and quantitatively calculate these factors, leading to insufficient rationality in stage segmentation. Specifically, this manifests in several ways: in some stages, the number of equipment deployed exceeds the shaft's limited space and operational capacity, causing conflicts during equipment hoisting and disorganized equipment placement; in other stages, the number of equipment deployed is insufficient, resulting in excessively long transition periods between stages and low construction efficiency. Furthermore, unreasonable stage segmentation can cause pipeline extensions and difficulties in equipment docking, increasing rework rates and further delaying construction progress.

[0005] II. Poor operational stability of the slurry balance system leads to high safety risks at the excavation face. The slurry balance system is the core of slurry balance shield tunneling, and its operational stability directly determines the stability of the excavation face. The slurry circulation system of a slurry balance shield tunnel in a hydraulic tunnel consists of multiple parts, including an inlet pipe, a outlet pipe, a relay pump station, a slurry mixing station, and slurry separation equipment. The pipeline layout is complex and the lines are long. During the separate launch process, as the tunnel boring machine's excavation distance increases, the slurry pipeline needs to be gradually extended. The friction resistance and local resistance along the pipeline will dynamically change, thus affecting the matching of the inlet and outlet slurry flow rates and the stability of the slurry chamber pressure. In existing technologies, the adjustment of slurry pump parameters mainly relies on manual experience. Operators manually adjust the pump speed and flow rate based on on-site observations of slurry chamber pressure, slurry concentration, and other parameters, making real-time and precise parameter control difficult. When pipeline resistance changes, the inlet and outlet slurry flow rates cannot be matched in time, easily leading to significant fluctuations in slurry chamber pressure. This can range from affecting the slurry's slag-carrying capacity to causing excavation face instability, water inrush, and sand inrush, posing a serious threat to the safety of hydraulic tunnel construction.

[0006] Third, shield tunneling attitude control is difficult and lacks multi-factor coupling control methods. In water conservancy projects involving confined spaces and deep shafts, the initial launch phase of a tunnel boring machine (TBM) occurs in the transition zone between the shaft and the tunnel. This area presents complex geological conditions and unique stress environments, making TBM attitude control far more challenging than in conventional TBM launch projects. During the segmented launch process, factors influencing the TBM's attitude exhibit multi-dimensional and coupled characteristics: First, the extended pipelines connected to the rear of the TBM generate additional tension, the magnitude of which dynamically changes with the pipeline length and arrangement. Second, the phased lowering and assembly of the auxiliary trolleys at each stage results in uneven weight distribution, leading to stress imbalances on the TBM and affecting its attitude. Third, changes in resistance during the extension of the slurry pipelines, as well as differences in tunneling resistance due to uneven soil hardness, all adversely affect the TBM's pitch and yaw attitude. Existing technologies lack quantitative analysis models for the coupled effects of the aforementioned multiple factors, making it impossible to accurately predict the degree of influence of each factor on the shield attitude. Furthermore, the lack of intelligent control equipment means that manual operation is required to adjust the shield machine's propulsion cylinders and articulation cylinders. This results in low attitude adjustment accuracy and slow response speed, which can easily lead to loss of control of the shield machine's attitude, causing problems such as tunnel axis deviation and difficulties in segment assembly. In severe cases, it can even cause the shield machine to jam, affecting project quality and construction progress.

[0007] IV. The safety risk management system is incomplete and lacks real-time perception and early warning capabilities. The construction environment within a confined space shaft in water conservancy projects is harsh, characterized by limited space, poor ventilation, and insufficient lighting. Furthermore, the segmented initial operation involves multiple high-risk stages, resulting in numerous safety risks and significant management challenges. Key risks include: improperly suspended pipelines within the shaft, increasing the risk of pipeline detachment and collisions; collisions and track misalignment of multiple auxiliary trolleys during positioning within the tunnel; complex stress on the negative ring segments during semi-ring assembly, leading to cracking, deformation, and even detachment; and additional risks such as oxygen deficiency, accumulation of toxic gases, and falls from heights. Current technologies for managing these safety risks rely primarily on manual inspections and experience-based judgment, resulting in limited inspection frequency and incomplete coverage, hindering real-time perception and dynamic monitoring of risk factors. Furthermore, the lack of an intelligent risk warning system prevents timely alerts when risks arise, leading to delays in accident prevention and response, and compromising the safety of construction personnel and equipment.

[0008] 5. Poor geological adaptability and delayed adjustment of tunneling parameters. The geological conditions traversed by water conservancy tunnels are complex and varied. The initial section often contains a variety of complex strata, including completely weathered rock, strongly weathered rock, moderately weathered rock, fault fracture zones, and interfaces between soft soil and hard rock. The physical and mechanical properties of different strata vary significantly, and the requirements for tunnel boring machine (TBM) excavation parameters also differ. For example, when excavating in soft soil strata, it is necessary to control the advance speed and slurry chamber pressure to avoid surface subsidence; when excavating in hard rock or fractured rock strata, it is necessary to increase the advance speed and cutterhead rotation speed, while adjusting the mud concentration to enhance the muck-carrying capacity. In existing split-launch technology, the setting of excavation parameters is mainly based on preliminary geological survey data and empirical values. During construction, it is difficult to adjust the excavation parameters in a timely manner according to real-time changes in the strata, and parameter adjustments lag behind strata changes. This lag can easily lead to excessive cutter wear, cutterhead damage, and even problems such as loss of shield attitude control and excavation face instability. This not only increases equipment maintenance costs but also delays construction progress and reduces the economic benefits of the project.

[0009] VI. Fixed phase quantities, lacking adaptive adjustment capabilities. Existing shield tunneling launch technologies typically pre-determine a fixed number of launch stages, such as two-stage or three-stage launches. This approach fails to adequately consider the unique requirements of hydraulic engineering projects. Different hydraulic tunnel projects exhibit significant variations in shaft clearance dimensions, shield machine equipment parameters (main machine dimensions, number and dimensions of auxiliary trolleys), geological conditions, and construction schedule requirements. A fixed number of stages cannot match the actual project conditions. For example, for shafts with extremely small clearance dimensions, a fixed two- or three-stage launch may not meet the space operation requirements; for projects with complex geological conditions and high construction schedule requirements, a fixed number of stages may lead to low construction efficiency. Current technologies lack methods and systems for dynamically optimizing and adjusting the number of launch stages based on actual project conditions, failing to achieve personalized and precise stage division, thus affecting the construction efficiency and safety of split-stage launches.

[0010] In summary, existing slurry balance shield tunneling technology for confined space vertical shafts in hydraulic engineering suffers from significant shortcomings in areas such as stage division, slurry balance system control, shield attitude regulation, safety risk management, geological adaptability, and stage number optimization. These limitations make it difficult to adapt to the complex and demanding construction conditions of hydraulic tunnel projects, hindering the widespread application of the slurry balance shield tunneling method in the construction of confined space vertical shafts. Therefore, there is an urgent need to develop a construction method that can adapt to the conditions of confined space vertical shafts in hydraulic engineering, enabling full-process perception, dynamic optimization, and autonomous decision-making for the multi-stage intelligent split-type launching of slurry balance shield tunnels. Summary of the Invention

[0011] The purpose of this invention is to address the problems existing in the prior art by providing a multi-stage intelligent split-type launching method for slurry balance shield tunneling in confined spaces. This method achieves efficient, safe, and intelligent construction of split-type launching of slurry balance shield tunneling under confined space vertical shaft conditions through systematic confined space adaptation planning, intelligent control of slurry circulation, multi-factor coupled attitude prediction and dynamic correction, full life cycle safety risk management and geological adaptive parameter optimization.

[0012] This invention is achieved using the following technical solution: A multi-stage intelligent split-type launching method for slurry balance shield tunneling in confined space includes the following steps: S1. Multi-stage split-type intelligent planning stage based on finite space shaft constraints and equipment parameters (1) Collect the limited space structural parameters of the starting shaft - including the net size, depth, shaft opening elevation, bottom plate elevation, waist beam position and elevation, ventilation and access passage size; collect the size and weight parameters of each component of the slurry balance shield machine - including cutterhead, front shield, middle shield, tail shield, segment assembly machine, equipment bridge, and rear supporting trolley assembly; collect the equipment parameters of the slurry circulation system - including the power of the slurry pump, the diameter and length of the slurry pipeline, and the configuration of the relay pump station.

[0013] (2) Based on the matching relationship between the net size of the shaft and the size of each component in the confined space, and combined with the position and elevation of the shaft waist beam, determine the maximum number and combination scheme of the trolleys that can be lowered into the shaft at each stage; with the constraints that the total weight of the trolleys at each stage does not exceed the rated lifting capacity of the shaft hoisting equipment, the total length of the trolleys at each stage does not exceed the available traction and positioning space in the tunnel, and the number of workers at each stage does not exceed the number of people allowed in the confined space, and with the optimization objectives of equalizing the tunneling distance at each stage, minimizing the number of stage transitions, and minimizing the risk of confined space operations, a multi-objective genetic algorithm is used to generate the optimal multi-stage split launch scheme.

[0014] (3) The multi-stage split launch scheme adaptively determines the number of stages N according to the limited space constraints of the shaft, where N is an integer not less than 3; when the ratio of the net size of the shaft to the size of the trolley is within the first preset range, N is 3; when the ratio is within the second preset range, N is 4; when the ratio is greater than the upper limit of the second preset range, N is 5.

[0015] (4) In the multi-stage split launch scheme, the order of access of the supporting trolleys in each stage is determined according to the functional priority of the trolleys: the slurry circulation system trolley is lowered into the well first, the power supply system trolley is lowered second, the grouting system trolley is lowered third, and the remaining auxiliary trolleys are lowered into the well last.

[0016] S2. Multi-stage intelligent control of slurry circulation based on hydraulic characteristic parameters and machine learning. (1) Install pressure sensors, flow sensors and density sensors at preset intervals on the slurry inlet pipe and slurry outlet pipe of the slurry circulation system to collect slurry inlet and outlet pressure, flow rate and slurry density data in real time; install pressure sensors on the partition of the slurry tank to collect slurry pressure data at the top and bottom of the slurry tank in real time.

[0017] (2) Based on the tunneling distance, pipeline extension length and number of relay pump stations in each stage of the multi-stage split-start, the real-time hydraulic resistance characteristics of the slurry inlet and outlet pipelines are calculated based on Bernoulli's equation and Darcy-Weisbach's formula, and a slurry circulation resistance model is established for each stage of the split-start.

[0018] (3) Construct a slurry chamber pressure prediction model based on long short-term memory neural network (LSTM), and use historical slurry pressure data, slurry flow difference, tunneling speed and geological parameters as inputs to predict the change trend of slurry chamber pressure within a preset time window in the future.

[0019] (4) When the predicted value of the slurry tank pressure deviates from the set value by more than the preset threshold, the optimal speed adjustment of the feed pump and discharge pump is calculated based on the model predictive control (MPC) algorithm with the goal of minimizing the slurry tank pressure deviation and optimizing the matching of the feed and discharge flow rates. The result is automatically sent to the frequency converter for execution, thereby realizing the intelligent matching of the feed and discharge flow rates and the precise control of the slurry tank pressure.

[0020] (5) During the phase transition of the multi-stage split-start, the parameters of the mud-water circulation resistance model are automatically updated according to the changes in mud pipeline length and hydraulic resistance brought about by the new connected trolley, so as to realize the smooth transition of the mud-water circulation system during the phase transition process.

[0021] S3. Intelligent prediction and dynamic correction stage of shield tunnel attitude based on multi-factor coupling analysis (1) Install tilt sensors, stroke sensors and pressure sensors on the shield machine shell, propulsion cylinders and articulation devices to collect the shield machine's pitch angle, horizontal deflection angle, roll angle, thrust and stroke data of each zone's propulsion cylinder in real time.

[0022] (2) Identify and quantify the multi-factor coupling effect of each stage of the multi-stage split launch that affects the shield attitude—including the additional tension of the extended pipeline on the rear of the shield machine, the unbalanced force caused by the change in the weight of the trolley in each stage, the resistance change caused by the extension of the mud pipeline, the influence of the change in the tunneling strata on the force on the cutterhead, and the constraint force caused by the suspension arrangement of pipelines in the confined space of the shaft.

[0023] (3) Construct a multi-factor coupled attitude prediction model based on Transformer-CNN hybrid neural network: use convolutional neural network (CNN) to extract local spatiotemporal features of multi-source sensor data, use Transformer encoder to capture long-range dependencies between various influencing factors, and output the predicted values ​​of shield attitude deviation for future multi-ring.

[0024] (4) When the predicted attitude deviation exceeds the preset threshold, the optimal thrust adjustment scheme of each zone propulsion cylinder is calculated based on the multi-objective particle swarm optimization algorithm (MOPSO), with the optimization objectives of minimizing horizontal deviation, minimizing vertical deviation, and smoothing thrust change. The correction command is generated and pushed to the shield machine control system for execution.

[0025] S4. Intelligent management and control of security risks throughout the multi-stage, split-process origination based on digital twins and edge computing. (1) Construct a digital twin model of the entire process of multi-stage split launch, and map the multi-stage scheme of S1, the mud-water circulation parameters of S2, and the attitude data of S3 to the digital twin space in real time to realize the synchronous mapping and interaction between physical entities and virtual models in the limited space shaft.

[0026] (2) Fiber optic strain sensors and tilt sensors are installed on the surface of the negative ring segment and key nodes of the reaction frame to collect strain and deformation data of the negative ring segment in real time; machine vision monitoring nodes are installed along the extended pipeline to collect spatial pose images and sag data of the pipeline in real time; multi-parameter gas concentration sensors are installed in the confined space of the shaft and inside the tunnel to monitor the concentration of gases such as oxygen, methane, hydrogen sulfide, and carbon monoxide in real time; personnel positioning and counting devices are installed at the entrance and exit of the shaft to count the number of workers in the confined space in real time.

[0027] (3) Real-time preprocessing and feature extraction of sensor data based on edge computing gateway - including data cleaning, outlier removal, feature value calculation, personnel counting statistics, and uploading the processed feature data to the cloud digital twin platform, while storing the original data locally.

[0028] (4) The cloud-based digital twin platform compares and analyzes real-time monitoring data with theoretical design values ​​and historical data, and comprehensively evaluates the stress state of the negative ring segment, the safety state of the extended pipeline, the environment inside the tunnel, and the state of personnel in the confined space in combination with the preset confined space operation safety threshold system.

[0029] (5) When the comprehensive assessment results show that any risk indicator exceeds the limit, the graded early warning will be automatically triggered, including pushing early warning information to the site management personnel, automatically adjusting the construction parameters to reduce the risk, automatically executing the shutdown and pressure maintenance and confined space personnel evacuation procedures in emergency situations, and pushing the early warning information and disposal suggestions to the emergency command center at the same time.

[0030] S5. Adaptive optimization stage of tunneling parameters based on real-time geological parameter sensing. (1) Install advanced geological detection sensors in front of the shield machine cutterhead, including seismic wave reflection detectors and resistivity detection devices, to collect geological parameters such as surrounding rock grade, integrity coefficient, and water content within a preset distance in front of the tunnel face in real time.

[0031] (2) Based on the real-time sensed stratum parameters, combined with the current tunneling speed, cutterhead torque, total thrust, penetration depth and other operating parameters of the tunnel boring machine, construct a mapping relationship model between tunneling parameters and stratum parameters based on support vector regression (SVR).

[0032] (3) When the formation parameters are sensed to change, the consequences of maintaining the current tunneling parameters are predicted based on the mapping relationship model, including accelerated tool wear, increased attitude deviation, and difficulty in slag removal. Suggestions for adjusting tunneling parameters are automatically generated, including the adjustment amount of tunneling speed, the adjustment amount of cutterhead speed, and the adjustment amount of slurry pressure.

[0033] (4) Push parameter adjustment suggestions to the operator for confirmation or directly send them to the shield machine control system for execution, so as to realize the real-time adaptive adjustment of tunneling parameters to geological changes.

[0034] S6. Multi-stage, modular, fully intelligent execution and knowledge accumulation phase of the entire process. (1) According to the optimal multi-stage split-type starting scheme generated by S1, the split-type starting excavation construction from the first stage to the Nth stage is carried out in sequence.

[0035] (2) During the construction process, the intelligent control of slurry circulation in S2, the intelligent prediction and dynamic correction of shield attitude in S3, the intelligent control of safety risks throughout the entire process in S4, and the adaptive optimization of tunneling parameters in S5 are operated simultaneously to achieve coordinated operation and information sharing among the subsystems.

[0036] (3) After each stage of construction is completed, the actual construction data of this stage—including tunneling parameters, attitude data, slurry circulation data, safety monitoring data, and stratum perception data—will be uploaded to the cloud database for incremental training and iterative optimization of each intelligent model.

[0037] (4) After all N stages of construction are completed, a multi-stage split-type starting process intelligent construction report is generated, which includes construction parameter statistics for each stage, attitude deviation analysis, abnormal event records, confined space operation efficiency analysis, model optimization suggestions, and parameter recommendations for subsequent normal tunneling.

[0038] More preferably, the fitness function of the multi-objective genetic algorithm in step S1 is: F = ω 1 Dbalance + ω 2 N transition + ω 3 R hoist + ω 4 S risk ,in D balance This serves as an index for the uniformity of tunneling distance at each stage. N transition For the number of stage transitions, R hoist To improve the utilization rate of hoisting equipment, S risk This is a risk index for operations in confined spaces. ω 1. ω 2. ω 3. ω 4 represents the corresponding weighting coefficient.

[0039] More preferably, in step S1, the first preset range is the ratio of the net clearance size inside the shaft to the size of the trolley is less than 2.5, the second preset range is the ratio from 2.5 to 4.0, and when the ratio is greater than 4.0, N is 5.

[0040] In a further preferred embodiment, the input sequence length of the LSTM slurry silo pressure prediction model in step S2 is no less than 30 time steps, the prediction time window is 5-10 ring tunneling distances, the model uses mean squared error (MSE) as the loss function, and the Adam optimizer is used for training.

[0041] More preferably, the prediction time domain of the MPC algorithm in step S2 is 10-20 control steps, the control time domain is 3-5 control steps, and the constraints include the upper and lower limits of the feed pump speed, the upper and lower limits of the discharge pump speed, and the limit of the pressure change rate of the mud-water silo.

[0042] In a further preferred embodiment, in the Transformer-CNN hybrid neural network model described in step S3, the CNN module contains no less than 3 convolutional layers and 2 pooling layers, and the Transformer encoder contains no less than 4 encoding layers, with each encoding layer containing an 8-head self-attention mechanism.

[0043] Further preferably, the graded early warning in step S4 includes three levels: Level 3 early warning is a confined space attention reminder, which pushes information to the on-site technical supervisor and simultaneously marks it on the digital twin platform; Level 2 early warning is an abnormal alarm, which automatically adjusts construction parameters and notifies the safety director, while simultaneously activating enhanced ventilation in the confined space; Level 1 early warning is an emergency risk, which automatically executes the shutdown and pressure maintenance and confined space personnel evacuation procedures, and notifies the project manager to activate the emergency plan.

[0044] Preferably, the advanced geological detection sensor mentioned in step S5 has a detection distance of not less than 30m and a detection accuracy that can identify changes in the surrounding rock grade and the location of water-bearing structures.

[0045] In a further preferred embodiment, the number of workers in the confined space mentioned in step S4 is counted in real time using UWB ultra-wideband positioning technology, with a positioning accuracy of no more than 30cm and a personnel counting error of no more than 1 person.

[0046] This multi-stage intelligent split-type launching method for slurry balance shield tunneling in confined space has the following advantages: 1. By using the constraints of finite space shafts and a multi-objective genetic algorithm, the scientific planning and adaptive determination of the number of stages for multi-stage split launch are realized, which solves the problem of unreasonable stages caused by traditional experience division, and improves the utilization efficiency of finite space shafts and the safety of equipment hoisting and positioning.

[0047] 2. Through LSTM-MPC intelligent regulation of slurry circulation and automatic updating of stage transition resistance model, precise matching of slurry inlet and outlet flow rates and accurate control of slurry chamber pressure at each stage of split-type initialization are achieved, effectively ensuring the stability of the excavation face.

[0048] 3. By combining Transformer-CNN multi-factor coupled attitude prediction with MOPSO dynamic correction, the influence of multiple factors such as the additional tension of the extended pipeline, the change in trolley weight, and the vertical shaft suspension constraint force on attitude is comprehensively considered, which significantly improves the axis control accuracy in the initial stage of the vertical shaft split in a limited space.

[0049] 4. Through the collaborative architecture of digital twin and edge computing, and UWB personnel precise positioning, real-time perception, comprehensive assessment and graded early warning of safety risks throughout the entire process of starting the vertical shaft in a confined space, as well as precise control of personnel working in confined spaces, have been achieved, greatly improving construction safety.

[0050] 5. Through advanced geological exploration and SVR adaptive optimization, the tunneling parameters can be adjusted in real time to adapt to complex and variable geological conditions, effectively reducing the risk of tool wear and equipment failure.

[0051] 6. Through intelligent execution and data closed-loop optimization throughout the entire process, continuous iteration and knowledge accumulation of the split-start construction technology have been achieved, providing a reusable intelligent solution for similar water conservancy confined space projects. Attached Figure Description

[0052] Figure 1 This is a flowchart of the multi-stage intelligent split-type initiation of a slurry balance shield tunnel in a confined space; Figure 2 This is a schematic diagram of the multi-stage split-start division and trolley grouping (taking four stages as an example); Figure 3 It is a multi-stage, modular intelligent planning flowchart based on the constraints of a limited space shaft. Figure 4 This is a diagram of the intelligent control system architecture for mud-water circulation based on LSTM-MPC. Figure 5 This is a flowchart of multi-factor coupled pose prediction and dynamic correction based on Transformer-CNN; Figure 6 This is a diagram of a multi-stage, end-to-end intelligent security risk management architecture based on digital twins and edge computing. Figure 7 This is a flowchart of geological adaptive tunneling parameter optimization based on SVR. Detailed Implementation

[0053] To provide a clearer description of this technology, the following detailed explanation is provided in conjunction with the accompanying drawings and embodiments.

[0054] Example

[0055] The following steps were used to complete the multi-stage intelligent split-type launch construction of a slurry balance shield tunnel in a confined space: S1. Multi-stage Split Launch Intelligent Planning: Collect structural parameters of the launching shaft, size / weight parameters of each component of the tunnel boring machine, and equipment parameters of the slurry circulation system. Combined with the constraints of the limited space in the shaft, determine the number and combination scheme of the trolleys that can be lowered into the shaft at each stage. Using hoisting load, positioning space, and personnel limits in the limited space as constraints, and with the optimization objectives of balanced tunneling distance, minimum number of stage transitions, and lowest operational risk, a multi-objective genetic algorithm is used to generate the optimal multi-stage split launch scheme. The scheme adaptively determines the number of stages N according to the shaft constraints, where N is an integer not less than 3, and determines the order of lowering into the shaft according to the functional priority of the trolleys. When the ratio of the net size in the shaft to the size of the trolley is less than 2.5, the number of stages N is 3; when the ratio is between 2.5 and 4.0, N is 4; when the ratio is greater than 4.0, N is 5.

[0056] S2. Multi-stage intelligent control of slurry circulation: Real-time acquisition of slurry inlet and outlet pressure, flow rate, slurry density, and slurry tank pressure data; calculation of real-time hydraulic resistance characteristics of inlet and outlet pipelines; establishment of slurry circulation resistance models for each stage; construction of a slurry tank pressure prediction model based on a long short-term memory neural network to predict the slurry tank pressure change trend; when the pressure deviation exceeds a preset threshold, automatic adjustment of the inlet and outlet pump speed based on the model predictive control algorithm to achieve flow matching and precise pressure control; automatic updating of resistance model parameters during stage transitions to achieve smooth system transition.

[0057] S3. Intelligent prediction and dynamic correction of shield attitude: Real-time acquisition of shield attitude and propulsion cylinder operation data, identification and quantification of the multi-factor coupling effect on attitude at each stage of split launch; construction of a multi-factor coupled attitude prediction model based on Transformer-CNN hybrid neural network, outputting the predicted value of future multi-ring attitude deviation; when the deviation exceeds the preset threshold, the optimal thrust adjustment scheme of each zone propulsion cylinder is calculated based on the multi-objective particle swarm optimization algorithm, and correction commands are generated and executed.

[0058] S4. Intelligent Management and Control of Safety Risks Throughout the Process: Construct a multi-stage, split-type digital twin model for launch, mapping the launch plan, circulation parameters, and attitude data to the twin space in real time; collect data on the stress on the negative ring segments, the status of the extended pipeline, the environmental parameters inside the tunnel, and personnel data in the confined space in real time; upload the data to the cloud twin platform after edge computing preprocessing, and conduct a comprehensive assessment of multiple safety conditions; automatically trigger graded early warnings and corresponding handling procedures when any risk indicator exceeds the limit.

[0059] S5. Adaptive optimization of tunneling parameters: Real-time acquisition of strata parameters in front of the tunnel face through advanced geological exploration, and construction of a mapping relationship model between tunneling parameters and strata parameters; When strata parameters change, predict the construction risks under the current parameters and generate tunneling parameter adjustment suggestions, push confirmation or direct execution, so as to realize the adaptive adjustment of tunneling parameters with geological changes.

[0060] S6. Intelligent Execution and Knowledge Accumulation Throughout the Process: Each stage of tunneling construction is executed sequentially according to the optimal plan, with simultaneous operation of subsystems for slurry circulation control, attitude correction, safety management, and parameter optimization, achieving collaborative operation of multiple systems; construction data from each stage is uploaded to a cloud database for incremental training and iterative optimization of the intelligent model; and a full-process intelligent construction report is generated upon completion of all stages.

[0061] Application Examples

[0062] Taking a water conservancy tunnel in the Guangxi Water Resources Allocation Project in the Beibu Gulf as an example, the starting shaft is a confined space operation environment with an internal clearance of 17.8m long × 7.4m wide and 14.5m deep. It adopts the CTS3070H-0250 slurry balance shield machine of China Railway Equipment, with a total length of about 198m and a total weight of about 309t. It is equipped with a total of 28 trolleys.

[0063] S1, Multi-stage Split-start Intelligent Planning The structural parameters of the limited space of the launching shaft, the dimensions and weight parameters of each component of the tunnel boring machine (TBM), and the parameters of the slurry circulation system were collected. Through multi-objective genetic algorithm optimization calculations, the ratio of the net space dimension within the shaft to the trolley dimension was determined to be 3.2, falling within the second preset range (2.5-4.0). The number of adaptive determination stages was N=4. The optimization process employed the fitness function F=ω1D. balance +ω2N transition+ω3R hoist +ω4S risk Set weight coefficients ω1 and ω 2、 ω3, ω4, solve to obtain the quantitative values ​​of each target: tunneling distance balance D balance Number of stage transitions N transition =3, hoisting equipment utilization rate R hoist Risk index S for confined space operations risk The target value is solved by substituting the fitness function, and finally the four-stage split-starting optimal solution is generated. Its function is to synchronously display the weights, values ​​of each sub-item index, and the operation logic of the objective function at the algorithm output result, and can directly establish the causal relationship of "algorithm → parameters → four-stage solution".

[0064] Phase 1: The main unit, equipment bridge, and No. 1-2 trolleys are lowered into the well for the start-up, with a tunneling distance of 50m, thrust controlled within 3000kN, and tunneling speed of 5-10mm / min; The main operations are as follows: In the first stage, a full three-ring system (0 ring, -1 ring, and -8 ring) and a semi-negative six-ring system are used for launching. The tunnel boring machine (TBM) main unit, trailer No. 2, and equipment bridge (trailer No. 1) are lowered into the shaft for assembly. Trailers No. 4 and No. 5 (slurry pumps) are lowered into the shaft for placement. Due to the space required for launching the TBM at the bottom of the shaft, trailers No. 4 and No. 5 cannot be placed at the bottom of the shaft. A second lateral beam is used to expand the storage platform to temporarily store trailers No. 4 and No. 5. The storage platform is expanded using 12.5 I-beams on the side of the lateral beam. The tunnel segments are directly hoisted into the equipment bridge from the 6.4×4.8m opening of the rear pilot tunnel lateral beam for assembly. The first excavation advances 25m, and the battery truck is hoisted into the shaft. Mortar is transported to the working face via DN80 pipelines through extended pipelines.

[0065] Phase 2: Connect trolleys 3-10, with a cumulative tunneling distance of 100m, thrust not exceeding 5000kN, and speed of 10-25mm / min; The main operations are as follows: After the first phase of excavation has reached 50m, the extension pipeline is removed, trailers 3-10 are reassembled, and transported sequentially into the tunnel to connect with the main unit. The remaining trailers are then placed on the ground and connected to the underground trailers via extension pipelines. Trailers 3-6 are pushed in by a battery-powered vehicle and connected to trailer 2. The battery-powered vehicle then travels into the tunnel and onto trailers 5-6, where it stops. Subsequently, trailers 7-10 are hoisted in and pushed into the pilot tunnel. Finally, trailers 7-10 are sequentially removed from the pilot tunnel using equipment and connected to trailer 6. The remaining space in the pilot tunnel is used to store three 6-meter inclined rail sections.

[0066] Phase 3: Connecting trolleys 11-19, with a cumulative tunneling distance of 152m and a speed of 15-30mm / min; The main operations are as follows: After a cumulative excavation of 73m in the second phase, the inclined rails are installed, and the cumulative excavation reaches 100m. The extension pipeline is then removed, and trailers 11-19 are sequentially lowered into the tunnel and connected to trailer 10. The remaining trailers are then repositioned on the ground and connected to the underground trailers using extension pipelines. Mortar is transported to the tunnel boring machine's grout storage tank using a battery-powered locomotive for grouting.

[0067] Phase 4: Connect trolleys 20-28, complete the initial configuration of the entire machine, and proceed normally to tunnel breakthrough at 3126m.

[0068] The main operations are as follows: The third stage involves excavating to 152m, dismantling the extension pipeline, and then excavating a total of 176m. The inclined rails are then reinstalled, and trailers 20-28 are sequentially lowered into the tunnel and transported to connect with trailer 19. After commissioning, the entire machine begins excavation until the tunnel is completed. The fourth stage involves transporting the tunnel segments and mortar to the tunnel boring machine via a battery-powered locomotive for normal construction procedures.

[0069] The order in which the trolleys are connected at each stage follows functional priority: the trolley for the mud-water circulation system is lowered into the well first, followed by the trolley for the power supply system, then the trolley for the grouting system, and the remaining auxiliary trolleys are lowered into the well last.

[0070] The optimization process of this scheme fully recorded the values ​​of various parameters of the fitness function. Through iterative optimization using a multi-objective genetic algorithm, the optimal four-stage split construction arrangement was obtained by comparing the corresponding indicators and fitness F calculation results of different stage division schemes.

[0071] S2, Intelligent Control of Sludge Circulation Pressure, flow, and density sensors are installed on the inlet and outlet slurry pipes, and a pressure sensor is installed on the slurry silo baffle. Based on an LSTM neural network, the slurry silo pressure trend is predicted. When the pressure deviation exceeds the set value of ±0.02 MPa, the MPC algorithm calculates the optimal speed adjustment of the inlet and outlet slurry pumps and executes it automatically, thus controlling the slurry silo pressure fluctuation within ±0.01 MPa. During transitions between stages, the slurry circulation resistance model is automatically updated to achieve a smooth transition.

[0072] S3. Multi-factor coupled attitude prediction and dynamic correction The study identifies the coupled effects of multiple factors, including the additional tension of the extended pipeline, changes in the weight of the trolley, changes in the resistance of the mud pipeline, changes in the geological formation, and the constraint force of the shaft suspension. A Transformer-CNN hybrid neural network is constructed to predict the shield tunneling attitude. When the prediction deviation exceeds the threshold of 15mm horizontally or 10mm vertically, the MOPSO algorithm calculates the optimal thrust adjustment scheme for 11 sets of propulsion cylinders and automatically corrects the deviation, with the correction amount controlled within 2mm / m.

[0073] S4 Intelligent Management and Control of Safety Risks Throughout the Process A digital twin model is constructed to synchronously map the construction status at each stage. Fiber optic strain sensors (accuracy ±1με) are deployed on the surface of the negative ring tunnel segments. Machine vision nodes are installed at 50m intervals along the extended pipeline. Multi-parameter gas concentration sensors are deployed in the confined space of the shaft and inside the tunnel. UWB personnel positioning and counting devices (positioning accuracy ≤30cm, personnel counting error ≤1 person) are deployed at the shaft entrances and exits. Edge computing gateways preprocess data in real time, and the cloud-based digital twin platform comprehensively assesses risks. A safety factor <0.7 triggers a Level 3 warning, <0.5 triggers a Level 2 warning and initiates enhanced ventilation, and <0.3 triggers a Level 1 warning and automatically executes shutdown, pressure maintenance, and personnel evacuation procedures.

[0074] S5, Geological Adaptive Tunneling Parameter Optimization A seismic wave reflection detector and a resistivity detection device are installed in front of the tunnel boring machine's cutterhead to monitor the geological parameters within a 30m range in front of the tunnel face in real time. When the surrounding rock grade is detected to change from Class IV to Class V, the SVR mapping model automatically generates adjustment suggestions to reduce the tunneling speed from 25mm / min to 15mm / min and the cutterhead rotation speed from 1.5rpm to 1.0rpm. These suggestions are then automatically implemented after confirmation by the operator.

[0075] S6. Full-process execution and closed-loop optimization The construction process proceeds sequentially through stages one through four, with data from each stage being synchronously uploaded to the cloud for incremental training of various intelligent models. Upon completion, a multi-stage, modular, full-process intelligent construction report is generated, including analysis of confined space operation efficiency.

[0076] The above description is not intended to limit this application, nor is this application limited to the above examples. Any changes, modifications, additions, or substitutions made by those skilled in the art within the scope of this application should fall within the protection scope of this application.

Claims

1. A multi-stage intelligent split-type launching method for a slurry balance shield tunneling machine in a confined space, characterized in that, Includes the following steps: S1. Multi-stage split-type launch intelligent planning: Collect structural parameters of the launch shaft, size / weight parameters of each component of the tunnel boring machine, and equipment parameters of the slurry circulation system. Combine the constraints of the limited space of the shaft to determine the number and combination scheme of the trolleys that can be lowered into the shaft at each stage. With constraints such as hoisting load, positioning space, and limited personnel in the limited space, and with the optimization objectives of balanced tunneling distance, minimum number of stage transitions, and lowest operational risk, a multi-objective genetic algorithm is used to generate the optimal multi-stage split-type launch scheme. The proposed scheme adaptively determines the number of stages N based on the vertical shaft constraints, where N is an integer not less than 3, and determines the order of access to the shaft according to the functional priority of the trolley. S2. Multi-stage intelligent control of slurry circulation: Real-time acquisition of slurry inlet and outlet pressure, flow rate, slurry density, and slurry tank pressure data; calculation of real-time hydraulic resistance characteristics of inlet and outlet pipelines; establishment of slurry circulation resistance models for each stage; construction of a slurry tank pressure prediction model based on a long short-time memory neural network to predict the slurry tank pressure change trend; automatic adjustment of inlet and outlet pump speed based on model predictive control algorithm when pressure deviation exceeds a preset threshold to achieve flow matching and precise pressure control; automatic updating of resistance model parameters during stage transitions to achieve smooth system transition; S3. Intelligent prediction and dynamic correction of shield attitude: Real-time acquisition of shield attitude and propulsion cylinder operation data, identification and quantification of the multi-factor coupling effect on attitude at each stage of split launch; construction of a multi-factor coupled attitude prediction model based on Transformer-CNN hybrid neural network, outputting future multi-ring attitude deviation prediction values. When the deviation exceeds the preset threshold, the optimal thrust adjustment scheme of each partition propulsion cylinder is calculated based on the multi-objective particle swarm optimization algorithm, and a correction command is generated and executed. S4. Intelligent management and control of safety risks throughout the entire process: Construct a multi-stage split digital twin model for launch, and map the launch scheme, circulation parameters, and attitude data to the twin space in real time; Real-time data collection includes stress on negative ring tunnel segments, status of extended pipelines, environmental parameters inside the tunnel, and personnel data in confined spaces. After edge computing preprocessing, the data is uploaded to a cloud-based twin platform for comprehensive assessment of various safety conditions. When any risk indicator exceeds the limit, a graded early warning and corresponding handling procedures are automatically triggered. S5. Adaptive optimization of tunneling parameters: Real-time acquisition of strata parameters in front of the tunnel face through advanced geological exploration, and construction of a mapping relationship model between tunneling parameters and strata parameters; When strata parameters change, predict the construction risk under the current parameters and generate tunneling parameter adjustment suggestions, push confirmation or direct execution, so as to realize the adaptive adjustment of tunneling parameters with geological changes; S6. Intelligent Execution and Knowledge Accumulation Throughout the Process: Each stage of tunneling construction is executed sequentially according to the optimal plan, with simultaneous operation of subsystems for slurry circulation control, attitude correction, safety management, and parameter optimization, achieving collaborative operation of multiple systems; construction data from each stage is uploaded to a cloud database for incremental training and iterative optimization of the intelligent model; and a full-process intelligent construction report is generated upon completion of all stages.

2. The multi-stage intelligent split-type launching method for slurry balance shield tunneling in confined space according to claim 1, characterized in that, The fitness function of the multi-objective genetic algorithm in step S1 is: F = ω 1 D balance + ω 2 N transition + ω 3 R hoist + ω 4 S risk ,in D balance This serves as an index for the uniformity of tunneling distance at each stage. N transition For the number of stage transitions, R hoist To improve the utilization rate of hoisting equipment, S risk This is a risk index for operations in confined spaces. ω 1. ω 2. ω 3. ω 4 represents the corresponding weighting coefficient.

3. The multi-stage intelligent split-type launching method for slurry balance shield tunneling in confined space according to claim 1, characterized in that, In step S1, when the ratio of the net size of the shaft to the size of the trolley is less than 2.5, the number of stages N is 3; when the ratio is between 2.5 and 4.0, N is 4; when the ratio is greater than 4.0, N is 5.

4. The multi-stage intelligent split-type launching method for slurry balance shield tunneling in confined space according to claim 1, characterized in that, The long short-term memory neural network slurry chamber pressure prediction model described in step S2 takes historical slurry pressure data, slurry flow difference, tunneling speed, and geological parameters as inputs. The length of the input sequence is no less than 30 time steps, and the prediction time window is 5-10 ring tunneling distances. The model uses mean square error as the loss function and is trained using the Adam optimizer.

5. The multi-stage intelligent split-type launching method for slurry balance shield tunneling in confined space according to claim 1, characterized in that, The prediction time domain of the model predictive control algorithm described in step S2 is 10-20 control steps, and the control time domain is 3-5 control steps. The constraints include the upper and lower limits of the feed pump speed, the upper and lower limits of the discharge pump speed, and the limit of the pressure change rate of the mud-water silo.

6. The multi-stage intelligent split-type launching method for slurry balance shield tunneling in confined space according to claim 1, characterized in that, The Transformer-CNN hybrid neural network model described in step S3 uses a convolutional neural network to extract local spatiotemporal features of multi-source sensor data and a Transformer encoder to capture long-range dependencies between various influencing factors. The convolutional neural network module contains no less than 3 convolutional layers and 2 pooling layers, and the Transformer encoder contains no less than 4 encoding layers, with each encoding layer containing an 8-head self-attention mechanism.

7. The multi-stage intelligent split-type launching method for slurry balance shield tunneling in confined space according to claim 1, characterized in that, The graded early warning mentioned in step S4 includes three levels: Level 3 early warning is a limited space attention reminder, which pushes information to the on-site technical person in charge and is simultaneously marked on the digital twin platform; Level 2 warning is an abnormal alarm, which automatically adjusts construction parameters and notifies the safety director, while simultaneously activating enhanced ventilation in confined spaces; Level 1 warning is an emergency risk, which automatically executes shutdown and pressure maintenance procedures and confined space personnel evacuation procedures, and notifies the project manager to activate the emergency plan.

8. The multi-stage intelligent split-type launching method for slurry balance shield tunneling in confined space according to claim 1, characterized in that, The advanced geological exploration described in step S5 has a detection distance of not less than 30m and a detection accuracy that can identify changes in the surrounding rock grade and the location of water-bearing structures.

9. The multi-stage intelligent split-type launching method for slurry balance shield tunneling in confined space according to claim 1, characterized in that, In step S4, the number of workers in the confined space is counted in real time using UWB ultra-wideband positioning technology, with a positioning accuracy of no more than 30cm and a personnel counting error of no more than 1 person.

Images