Intelligent shield muck improvement system based on machine learning

Abstract

The application discloses an intelligent shield muck improvement system based on machine learning, and relates to the technical field of shield tunnel construction.The system comprises a muck area division module, a reverse stability analysis module and a strategy generation module.The muck area division module collects muck macroscopic distribution data through three-dimensional laser scanning, obtains microscopic characteristics such as average particle size and clay mineral composition by combining technologies such as microscopic imaging and XRD, and divides the muck into improvement units with similar physical properties by using K-means clustering.Meanwhile, a machine learning regression model is combined to construct the feasibility index of each improvement unit, and the improvement units are marked as three levels of green, yellow and red to intuitively present the improvement difficulty.The reverse stability analysis module collects transportation and stacking data, accurately identifies disturbance risks caused by water migration and path overlap between improvement units, and the strategy generation module generates a priority improvement sequence according to the improvement level and the disturbance level, and matches a differentiated strategy.

Description

Technical Field

[0001] This invention relates to the field of shield tunnel construction technology, specifically to an intelligent shield tunnel spoil improvement system based on machine learning. Background Technology

[0002] During shield tunnel construction, the tunnel boring machine (TBM) excavates and transports underground soil out of the tunnel using mechanical cutting and conveying systems. The resulting excavated soil typically consists of a mixture of fine-grained soil, gravel, cement slurry, and small amounts of chemical additives. It is characterized by complex particle composition, uneven moisture content, high fluidity, and poor short-term stability. Traditionally, TBM excavated soil is often directly discharged or stockpiled in its original state. However, this method presents several problems: firstly, stockpiled excavated soil is prone to settlement, slippage, or localized collapse, affecting site safety and transportation access; secondly, the heterogeneity and high moisture content of the excavated soil lead to unstable performance during reuse, making it difficult to meet the requirements for backfilling, foundation improvement, or other engineering applications, resulting in resource waste and environmental pollution risks.

[0003] Traditional methods typically divide the waste soil into simple sections based on construction phases or soil volume, neglecting the microstructural characteristics, particle composition, and moisture content distribution. This leads to uneven treatment of waste soil with different physical properties, resulting in localized under-treatment or overuse of amendments. Secondly, moisture migration and disturbance occur during transportation, stockpiling, and zoning of the waste soil. Traditional methods often ignore the mutual influence and reverse disturbance between different types of waste soil, causing sudden changes in fluidity or instability of the amendment during stockpiling or transportation, affecting the overall improvement effect and construction efficiency. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides an intelligent shield tunneling muck improvement system based on machine learning to solve the problems mentioned in the background section.

[0005] To achieve the above objectives, the present invention provides the following technical solution: an intelligent shield tunneling spoil remediation system based on machine learning, comprising: The waste soil area division module is used to collect waste soil samples and related construction parameters generated during shield tunnel construction, establish a three-dimensional spatial model of the waste soil, and identify the microstructural characteristics, particle composition, and moisture content distribution of the waste soil. Based on machine learning K-means clustering, the waste soil is divided into several improvement units, and the feasibility index of the c-th improvement unit is constructed through a trained machine learning model. Based on the prediction results, several improvement units are classified to obtain the corresponding improvement levels and marked in the three-dimensional slag model. The reverse stability analysis module is used to collect historical stockpiling data, moisture change data during transportation, and influence path information of adjacent improvement units for each improvement unit, and to establish an influence dataset to construct the reverse influence coefficient between the c-th improvement unit and the d-th adjacent improvement unit. The disturbances are then categorized to obtain their corresponding levels. The strategy generation module generates a priority improvement sequence based on the corresponding improvement level and disturbance level of the c-th improvement unit, and implements the corresponding strategy on the c-th improvement unit according to the priority improvement sequence.

[0006] Preferably, the waste soil area division module includes a waste soil data acquisition unit and a microstructure identification unit; The excavated soil data acquisition unit is used to collect excavated soil samples generated at different times, under different cutterhead torques and propulsion pressures during shield tunneling construction, and to record the construction ring number and relevant construction parameters for each construction process, including the cutterhead torque of the corresponding ring segment. Cutting area of ​​the cutter head and propulsion speed ; Furthermore, a 3D laser scanner was used to scan the formation state of the construction waste during the construction process, obtain information on the morphology and volume distribution of the waste accumulation, and establish a preliminary spatial distribution dataset of the waste waste S0(x,y,z). The microstructure identification unit is used to identify the microstructure of slag and soil based on microscopic imaging and particle image analysis technology, and to obtain the average particle size. Fine particle content Specific surface area σ, porosity n e And the natural moisture content w0, to establish a micro-dataset; Combining XRD and FTIR spectroscopy results, the main clay mineral components were identified, including the proportions of illite, montmorillonite, and kaolinite; a mineral composition vector M=(m1,m2,...,m) was constructed for each slag soil sample point. k );m j This represents the mass fraction of the j-th clay mineral; And calculate the mineral sensitivity coefficient of the i-th slag soil sample point. :

[0007] in, The expansion sensitivity factor corresponding to the j-th clay mineral is represented, including: montmorillonite r=1.0, illite r=0.5, and kaolinite r=0.2. The larger the mineral sensitivity coefficient Ms is, the stronger the swelling of the slag soil when it comes into contact with water, and the higher the dosage of the amendment is required for stabilization treatment. J represents the number of clay mineral types identified at the i-th slag soil sample point.

[0008] Preferably, the waste soil area division module also includes an area segmentation unit and a physical characteristic calculation unit; The region segmentation unit is used to project the preliminary waste soil spatial distribution dataset S0(x,y,z) onto the shield tunnel coordinate system, establish a three-dimensional waste soil spatial model, and perform spatial partitioning according to the construction ring number or soil silo volume improvement unit. Extract each soil sample point generated during the tunnel boring machine (TBM) excavation process, obtain its spatial coordinates (x, y, z), and the average particle size obtained from the microscopic dataset. Fine particle content Specific surface area σ, porosity n e and natural moisture content And the mineral composition vector M, which forms the physical property feature vector of each slag soil sample point:

[0009] The physical property feature vectors of each soil sample point were normalized to a range of 0 and 1. Based on the shield tunneling section length and the number of samples, K-means clustering was used, with K being the number of clusters. Each cluster was numbered c=1, 2, ..., K, and each cluster corresponded to an "improvement unit". An initial cluster center was set for every 5 meters of tunneling length, and K soil sample points were selected as the initial cluster centers. : Calculate the weighted Euclidean distance between the i-th slag sample and each cluster center. : Calculate the set of distances from samples within each cluster c to the center, and set a spatial radius threshold. The spatial range of the improvement unit is defined, and the spatial range of the c-th improvement unit is defined as follows:

[0010] in, This represents the weighted Euclidean distance between a sample point and the cluster center. To improve the threshold of the unit space radius.

[0011] Preferably, the physical characteristic calculation unit is used to calculate the plastic flow pressure coefficient of the c-th improved unit. Permeability coefficient and cohesive strength index This allows for the quantification of the physical properties of each improved unit, including: The plastic pressure coefficient of the c-th improved unit is calculated based on the principles of energy conservation and cutting specific energy. ; Based on the Kozeny–Carman equation seepage formula, the permeability coefficient of the c-th improved unit is calculated. ; Based on the Coulomb failure criterion and natural moisture content To calculate the cohesive strength index of the c-th improved unit. .

[0012] Preferably, the waste soil area division module also includes a feasibility calculation unit and a classification marking unit; The feasibility calculation unit is first used to calculate the number of clustered samples in the c-th improvement unit. mineral sensitivity coefficient The average value is then used to obtain the comprehensive mineral sensitivity coefficient of the c-th improved unit. ; And extract the plastic flow pressure coefficient of the c-th improved unit. Permeability coefficient Cohesive strength index and comprehensive mineral sensitivity coefficient After performing Min-Max normalization on the inverse indicators based on historical data, the feasibility index of the c-th improvement unit is obtained by weighted summation. .

[0013] Preferably, the classification marker unit is used for the feasibility index based on the c-th improvement unit. To obtain the corresponding improvement level, including: like The first improvement level is generated, and the improvement unit is marked in green in the three-dimensional slag model. like A second improvement level is generated, and the improved unit is marked in yellow in the three-dimensional slag model. like The third improvement level is generated, and the improvement unit is marked in red in the three-dimensional slag model.

[0014] Preferably, the reverse stability analysis module includes an impact data acquisition unit and a path coupling unit; The impact data acquisition unit is used to collect environmental and status data of each improvement unit during shield tunneling and spoil transportation, including stockpiling time, transportation distance, transportation batches, change in moisture content Δa before and after transportation, as well as the spatial location and slope relationship of adjacent improvement units, to establish an improvement impact dataset. The expression for the change in moisture content Δa before and after transportation is: ;in, and These represent the pore water volumes before and after transportation. For solid volume; The path coupling unit is used to calculate the path coupling coefficient between each improved unit based on the transportation route and stacking order. Specifically: Calculate the overlap ratio and stacking sequence time difference between the c-th improved unit and its adjacent d-th improved unit on the transportation path to obtain the path coupling coefficient. :

[0015] in, Let be the overlap distance between the c-th improved unit and its d-th adjacent improved unit in the transportation route. The total transportation distance. The time difference between stacking the two improved units. is the time decay coefficient, and e is the natural exponential decay function.

[0016] Preferably, the reverse stability analysis module also includes an influence coefficient calculation unit and a disturbance classification unit; The influence coefficient calculation unit is used to combine the path coupling coefficient. Calculate the reverse influence coefficient between the c-th improvement unit and the d-th adjacent improvement unit, based on the change in moisture content Δa before and after transportation. ; Disturbance-level unit, used to determine the reverse influence coefficient between the c-th improved unit and its d-th adjacent improved unit. Generate corresponding disturbance levels, including: like This generates the first level of disturbance. like This generates a second level of disturbance. like This generates the third level of disturbance.

[0017] Preferably, the strategy generation module includes a priority improvement sequence generation unit and a strategy generation unit. The priority improvement sequence generation unit is used to generate a priority improvement sequence based on the soil improvement level and disturbance level of the c-th improvement unit, including: For an improved unit with an improvement level of the first improvement level and a perturbation level of the third perturbation level, generate the front part of the sequence; For improved units with an improvement level of first improvement level and a disturbance level of second disturbance level, and improved units with an improvement level of second improvement level and a disturbance level of third disturbance level, the middle part of the merged sequence is included. For improved units with an improvement level of the first improvement level and a disturbance level of the first disturbance level, and improved units with an improvement level of the second improvement level and a disturbance level of the first or second disturbance level, the middle and later parts of the generation sequence are generated. For improved units with an improvement level of three, regardless of the perturbation level, the end part of the generated sequence is generated.

[0018] Preferably, the strategy generation unit is used to generate corresponding strategies for the priority improvement unit sequence, including: For the front part of the sequence, a first strategy is generated, including: using an enhanced stirring intensity of +30% of the standard stirring intensity, increasing the modifier dosage by 20%, and adding a water-retaining agent by 10% during the mixing process; For the middle part of the sequence, a second strategy is generated, including: increasing the stirring intensity by 15%, increasing the modifier dosage by 10%, and increasing the water-retaining agent by 5% in moisture adjustment; For the middle and later parts of the generated sequence, a third strategy is generated, which includes: maintaining standard stirring intensity and modifier dosage, and maintaining the original moisture conditions; For the end portion of the sequence, a fourth strategy is generated, including: increasing the stirring intensity by 20%, increasing the modifier dosage by 30%, and increasing the water-retaining agent by 10% to 15% in moisture adjustment.

[0019] This invention provides an intelligent shield tunneling muck improvement system based on machine learning. It has the following beneficial effects: (1) By dividing the improvement units with similar physical properties through K-means clustering, and combining physical characteristic calculation and feasibility index evaluation, the problem of insufficient improvement or excessive improvement agent caused by traditional extensive division is avoided. At the same time, the improvement level is marked by the color mark of the three-dimensional model, providing clear guidance for construction.

[0020] (2) By quantifying the reverse influence coefficient and disturbance level between units through the reverse stability analysis module, the disturbance risks such as moisture migration and path coupling during transportation and stacking can be identified in advance, avoiding problems such as fluidity mutation and improver failure caused by ignoring reverse disturbances in the traditional way.

[0021] (3) The generation module combines the improvement level and the disturbance level to generate a priority sequence, match the differentiated improvement strategy, prioritize the processing of high-value and high-risk units, avoid resource waste and decision-making blindness, and achieve optimization of improvement efficiency, cost and stability. Attached Figure Description

[0022] Figure 1 This is a schematic diagram of the system flow of the present invention. Detailed Implementation

[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] Example 1 Please see Figure 1This invention provides an intelligent shield tunneling spoil remediation system based on machine learning, comprising: The waste soil area division module is used to collect waste soil samples and related construction parameters generated during shield tunnel construction, establish a three-dimensional spatial model of the waste soil, and identify the microstructural characteristics, particle composition, and moisture content distribution of the waste soil. Based on machine learning K-means clustering, the waste soil is divided into several improvement units, and the feasibility index of the c-th improvement unit is constructed through a trained machine learning model. Based on the prediction results, several improvement units are classified to obtain the corresponding improvement levels and marked in the three-dimensional slag model. The reverse stability analysis module is used to collect historical stockpiling data, moisture change data during transportation, and influence path information of adjacent improvement units for each improvement unit, and to establish an influence dataset to construct the reverse influence coefficient between the c-th improvement unit and the d-th adjacent improvement unit. The disturbances are then categorized to obtain their corresponding levels. The strategy generation module generates a priority improvement sequence based on the corresponding improvement level and disturbance level of the c-th improvement unit, and implements the corresponding strategy on the c-th improvement unit according to the priority improvement sequence.

[0025] In this embodiment, addressing the problems of high safety risks, significant resource waste, uneven improvement effects, and susceptibility to disturbance and instability inherent in traditional shield tunneling excavation waste disposal, this system utilizes a waste area division module. Combined with machine learning K-means clustering and a feasibility index model, it accurately captures differences in the microstructure, particle composition, and moisture content distribution of the waste soil. This divides the waste soil into improvement units with similar physical properties and labels them with improvement levels, effectively avoiding the problems of insufficient improvement or excessive amendments caused by simple segment or volume division in traditional methods. Simultaneously, the reverse stability analysis module constructs reverse influence coefficients between improvement units to fully consider... The study measured the effects of moisture migration during transportation and stockpiling of construction waste and the mutual disturbance between adjacent improvement units, thus overcoming the shortcomings of traditional methods that neglect reverse disturbances leading to sudden changes in fluidity or instability due to dilution of the improver. Finally, the strategy generation module generates a priority improvement sequence based on the feasibility index and the reverse influence coefficient, enabling differentiated and precise implementation of construction waste improvement. This not only improves the overall improvement effect and construction efficiency, but also reduces safety risks such as settlement and slippage at the construction site. Furthermore, by optimizing the use of improvers and improving the stability of construction waste performance, it creates conditions for reuse such as construction waste backfilling and foundation improvement, thereby reducing resource waste and environmental pollution risks.

[0026] Example 2 This embodiment is an explanation based on Embodiment 1. Please refer to it. Figure 1 Specifically, the waste soil area division module includes a waste soil data acquisition unit and a microstructure identification unit; The excavated soil data acquisition unit is used to collect excavated soil samples generated at different times, under different cutterhead torques and propulsion pressures during shield tunneling construction, and to record relevant construction parameters for each construction ring, including the cutterhead torque of the corresponding ring segment. Cutting area of ​​the cutter head and propulsion speed ; Furthermore, a 3D laser scanner was used to scan the formation state of the construction waste during the construction process, obtain information on the morphology and volume distribution of the waste accumulation, and establish a preliminary spatial distribution dataset of the waste waste S0(x,y,z). The microstructure identification unit is used to identify the microstructure of slag and soil based on microscopic imaging and particle image analysis technology, and to obtain the average particle size. Fine particle content Specific surface area σ, porosity n e And the natural moisture content w0, to establish a micro-dataset; Combining XRD and FTIR spectroscopy results, the main clay mineral components were identified, including the proportions of illite, montmorillonite, and kaolinite; a mineral composition vector M=(m1,m2,...,m) was constructed for each slag soil sample point. k );m j This represents the mass fraction of the j-th clay mineral; And calculate the mineral sensitivity coefficient of the i-th slag soil sample point. :

[0027] in, The expansion sensitivity factor corresponding to the j-th clay mineral is represented, including: montmorillonite r=1.0, illite r=0.5, and kaolinite r=0.2. The larger the mineral sensitivity coefficient Ms is, the stronger the swelling of the slag soil when it comes into contact with water, and the higher the dosage of the amendment is required for stabilization treatment. J represents the number of clay mineral types identified at the i-th slag soil sample point.

[0028] The waste soil area division module also includes area segmentation units and physical characteristic calculation units; The region segmentation unit is used to project the preliminary waste soil spatial distribution dataset S0(x,y,z) onto the shield tunnel coordinate system, establish a three-dimensional waste soil spatial model, and perform spatial partitioning according to the construction ring number or soil silo volume improvement unit. Extract each soil sample point generated during the tunnel boring machine (TBM) excavation process, obtain its spatial coordinates (x, y, z), and extract the average particle size from the microscopic dataset. Fine particle content Specific surface area σ, porosity n e and natural moisture content And the mineral composition vector M, which forms the physical property feature vector of each slag soil sample point:

[0029] The physical property feature vectors of each soil sample point were normalized to a range of 0 and 1. Based on the shield tunneling section length and the number of samples, K-means clustering was used, with K being the number of clusters. Each cluster was numbered c=1, 2, ..., K, and each cluster corresponded to an "improvement unit". An initial cluster center was set for every 5 meters of tunneling length, and K soil sample points were selected as the initial cluster centers. : Calculate the weighted Euclidean distance between the i-th slag sample and each cluster center. ; Calculate the set of distances from samples within each cluster c to the center, and set a spatial radius threshold. The spatial range of the improvement unit is defined, and the spatial range of the c-th improvement unit is defined as follows:

[0030] in, This represents the weighted Euclidean distance between a sample point and the cluster center. To improve the threshold for the spatial radius of the unit cell; After performing the above steps, several non-overlapping improvement units will be obtained. Each improvement unit represents a type of slag soil area with similar physical properties, which is used to optimize the ratio of improvement agents, determine the feasibility of improvement, and evaluate the stability of improvement effects.

[0031] In this embodiment, the slag area division module brings several targeted benefits through refined data acquisition, accurate microscopic characteristic identification, and scientific spatial partitioning and clustering design: First, the slag data acquisition unit, combining construction parameters (cutterhead torque, propulsion speed, etc.) with slag accumulation morphology and volume information obtained from 3D laser scanning, constructs a preliminary slag spatial distribution dataset S0(x,y,z), which can comprehensively capture the macroscopic distribution characteristics of slag, avoiding the spatial information loss caused by the traditional coarse recording relying solely on construction ring numbers, thus laying the foundation for subsequent accurate partitioning; Second, the microstructure identification unit, through microscopic imaging, particle image analysis, and XRD and FTIR tests, not only obtains key parameters such as average particle size and moisture content, but also quantifies the clay mineral expansion characteristics through the mineral composition vector M and the mineral sensitivity coefficient Ms, accurately identifying the microstructure of the slag. By observing differences, this approach solves the problem of poor compatibility of soil amendments caused by neglecting microstructure in traditional methods. In particular, the identification of highly expansive minerals such as montmorillonite allows for the early prediction of high-dose amendment requirements, avoiding construction risks caused by insufficient amendment. Furthermore, the regional segmentation unit combines macroscopic spatial data with microscopic physical property characteristics. Through normalization processing and K-means clustering (setting initial centers based on 5-meter tunneling lengths and using weighted Euclidean distance), it divides the soil into non-overlapping amendment units with similar physical properties. This overcomes the shortcomings of traditional segmentation based on ring / soil volume that ignores the differences in the inherent properties of the excavated soil. It also clarifies the unit range through spatial radius thresholds, ensuring that the amendment requirements of each unit are consistent. This provides a precise basis for subsequent optimization of amendment ratios, while avoiding waste and risks from excessive or insufficient amendment in certain areas. Ultimately, it achieves refined management and control of the entire process from excavated soil data collection to zoning.

[0032] Example 3 This embodiment is an explanation based on Embodiment 1. Please refer to it. Figure 1 , specifically Based on the regression prediction model, a machine learning model is established. The machine learning model is then trained by first calculating the plastic flow pressure coefficient of the c-th improved unit. Permeability coefficient and cohesive strength index Then, combining the four feature indicators, the final prediction formula is obtained through least squares fitting, and the feasibility index of the c-th improvement unit is obtained. The predicted results are output and validated using a validation set after the machine learning model is trained. The "improvement level" of each new improvement unit (e.g., green for Level I, blue for Level II, orange for Level III, and red for Level IV) is associated with a "3D spoil space model" (built from regional segmentation units), visually displaying the distribution of spoil improvement difficulty at different locations within the tunnel, providing spatial guidance for construction personnel. As new construction projects progress, the "predicted index of new improvement unit - actual improvement effect" data is added to the historical database, and the model is retrained periodically (e.g., every 100 rings of construction completed). This includes updating the maximum / minimum values ​​of historical data, optimizing normalization accuracy, and recalculating feature weights to adapt to changes in feature importance under new geological conditions.

[0033] The physical characteristic calculation unit is used to calculate the plastic flow pressure coefficient of the c-th improved unit. Permeability coefficient and cohesive strength index This allows for the quantification of the physical properties of each improved unit, including: The plastic pressure coefficient of the c-th improved unit is calculated based on the principles of energy conservation and cutting specific energy. ;

[0034] in, This indicates the torque of the corresponding ring-shaped cutter head. Indicates the cutting area of ​​the cutter head. Indicates the propulsion speed; commonly used in tunnel boring machines to determine the "plasticity and flow resistance of the excavated soil"; (Reference: Maidl, Shield Tunneling Machines, 2012). Plastic flow pressure coefficient. Larger cutter head → higher torque, harder to push the excavated soil → stickier and thicker excavated soil → poorer fluidity; Based on the Kozeny–Carman equation seepage formula, the permeability coefficient of the c-th improved unit is calculated. ;

[0035] in, This indicates the particle size of 10% of the particles in the slag. Indicates the void ratio. For void volume, For solid volume; Larger particles → coarser pores → higher permeability; Smaller pores mean fewer pores and therefore lower permeability; The large amount indicates that the slag soil has rapid water seepage, poor water retention, and the soil conditioner is easily lost. Based on the Coulomb failure criterion and natural moisture content To calculate the cohesive strength index of the c-th improved unit. ;

[0036] in, The intrinsic cohesion of the undisturbed soil is obtained through laboratory triaxial or direct shear tests. This represents the proportion of the cohesive component, with a specific value between 0 and 1, reflecting the proportion of fine particles in silt and clay. This represents the natural moisture content; when the lateral stress is constant, the shear strength of the slag soil is mainly determined by... Control; increased moisture content softens the soil, but can increase surface adhesion within a small range, therefore an empirical correction term is introduced. ; Larger sizes mean higher viscosity components and higher moisture content, making the slag stickier and more difficult to break and mix.

[0037] The waste soil area delineation module also includes a feasibility calculation unit and a classification labeling unit; The feasibility calculation unit is first used to calculate the number of clustered samples in the c-th improvement unit. mineral sensitivity coefficient The average value is then used to obtain the comprehensive mineral sensitivity coefficient of the c-th improved unit. :

[0038] And extract the plastic flow pressure coefficient of the c-th improved unit. Permeability coefficient Cohesive strength index and comprehensive mineral sensitivity coefficient After performing Min-Max normalization of the inverse indicator based on historical data, the feasibility index of the c-th improvement unit is calculated using the following formula. :

[0039] In the formula, , , and Let represent the normalized plastic flow pressure coefficients of the c-th improved unit. Permeability coefficient Cohesive strength index and comprehensive mineral sensitivity coefficient Weighting coefficients; In clay soil strata, the cohesiveness of the slag and the expansibility of minerals have a greater impact on the improvement effect, with weights set as a3=0.3, a4=0.3, a1=0.2, and a2=0.2. In gravel strata, permeability (due to easy loss of the amendment) and fluidity (due to easy collapse of the slag) are key factors, with weights adjusted to a1=0.3, a2=0.3, a3=0.2, and a4=0.2. In practice, the least squares method can be used to fit historical data, minimizing the error between the model's predicted feasibility index and the actual improvement effect (such as expert ratings and construction efficiency). This automatically determines the optimal weight combination, ensuring that the weights both conform to engineering experience and accurately reflect the correlation between characteristics and improvement feasibility.

[0040] The normalization steps are as follows:

[0041]

[0042] in, and These represent the maximum and minimum values ​​of the plastic flow pressure coefficient in historical data, respectively. and These represent the maximum and minimum values ​​of the permeability coefficient in historical data, respectively. and These represent the maximum and minimum values ​​of the cohesive strength index in historical data, respectively; and These represent the maximum and minimum values ​​of the comprehensive mineral sensitivity coefficient in historical data, respectively. The significance of inverse normalization is that, in shield tunneling slag improvement analysis, the magnitude of different physical indicators is inversely correlated with the feasibility of improvement. High plasticity pressure coefficient → sticky slag, difficult to push the cutterhead → difficult to improve; high permeability coefficient → rapid water seepage in slag, loss of improver → difficult to improve; high cohesive strength → slag is too sticky, difficult to mix → difficult to improve; high mineral sensitivity → prone to expansion and instability → difficult to improve. Therefore, to ensure that all indicators reflect the feasibility of improvement in the same direction, the larger the value, the better.

[0043] Classification label unit, used for the feasibility index based on the c-th improvement unit. To obtain the corresponding improvement level, including: like This indicates that the improvement unit has the highest feasibility, the slag has good fluidity, is easy to mix, has moderate water retention, and the improvement agent has high efficiency. It can be prioritized for improvement and generated as the first improvement level. In the three-dimensional slag model, the improvement unit is marked in green. like This indicates that the improvement unit has moderate feasibility, the fluidity and water retention of the slag are average, the use of the improver needs to be adjusted appropriately, and a second improvement level is generated. In the three-dimensional slag model, the improvement unit is marked in yellow. like This indicates that the improvement unit has the lowest feasibility, poor fluidity, strong cohesion, and high difficulty in improvement. It is necessary to increase the dosage of the improver or adopt a special stirring / moisture adjustment strategy to generate a third improvement level. In the three-dimensional slag model, this improvement unit is marked in red.

[0044] In this embodiment, the physical characteristic calculation unit, based on energy conservation, the Kozeny-Carman equation, and the Coulomb criterion, accurately quantifies the plasticity pressure coefficient, permeability coefficient, and cohesive strength index, reflecting the fluidity, water retention, and cohesiveness of the slag, respectively. This solves the problem of blind improvement schemes caused by the difficulty in quantifying the physical properties of slag in traditional methods. The feasibility calculation unit integrates microscopic expansion characteristics by comprehensively considering mineral sensitivity coefficients, eliminates dimensional differences in indicators by combining inverse Min-Max normalization, and obtains the feasibility index by fitting weights using the least squares method. This transforms multi-dimensional indicators negatively correlated with improvement difficulty into a unified positive evaluation standard, ensuring the comparability and judgment of feasibility among different improvement units. The classification and labeling unit... Based on the feasibility index, three levels of improvement are divided and marked with color codes in the 3D model, which intuitively presents the spatial distribution of improvement difficulty within the tunnel, providing clear spatial guidance for construction personnel and avoiding the confusion of improvement priorities caused by the lack of clear level division in traditional methods. At the same time, the model is continuously retrained by incorporating new data and dynamically updating feature weights and normalization parameters to ensure high evaluation accuracy under different geological conditions. Ultimately, this achieves a scientific process from quantifying physical properties to marking improvement levels, ensuring precise matching of improvement agents (such as targeted dosage increases for red-level units) and improving construction efficiency through intuitive visualization, effectively reducing safety risks and cost waste caused by insufficient or excessive improvement.

[0045] Example 4 This embodiment is an explanation based on Embodiment 1. Please refer to it. Figure 1 Specifically, the reverse stability analysis module includes an impact data acquisition unit and a path coupling unit; The impact data acquisition unit is used to collect environmental and status data of each improvement unit during shield tunneling and spoil transportation, including stockpiling time, transportation distance, transportation batches, change in moisture content Δa before and after transportation, as well as the spatial location and slope relationship of adjacent improvement units, to establish an improvement impact dataset. The expression for the change in moisture content Δa before and after transportation is: ;in, and These represent the pore water volumes before and after transportation. The solid volume represents the change in moisture content Δa before and after transportation, which reflects the rate of change in pore water in the slag and the volumetric moisture content gradient per unit volume of soil. The path coupling unit is used to calculate the path coupling coefficient between each improved unit based on the transportation route and stacking order. Specifically: Calculate the overlap ratio and stacking sequence time difference between the c-th improved unit and its adjacent d-th improved unit on the transportation path to obtain the path coupling coefficient. :

[0046] in, Let be the overlap distance between the c-th improved unit and its d-th adjacent improved unit in the transportation route. The total transportation distance. The time difference between the stacking of the two improved units, in hours. is the time decay coefficient; representing the exponential decay rate of transport / stacking disturbance over time; e is the natural exponential decay function, with a value of 2.71828; The time decay factor unifies the time unit to hours: For obviously coarse-grained materials such as gravelly sand and crushed stone soil, the clay content is <5%. For high-viscosity / high-fineness materials (such as cohesive soils and high-fineness silts with montmorillonite content >20%), take... .

[0047] Classification and particle size standards: For obviously coarse-grained materials... Includes the following categories; Gravelly soil: Particles larger than 2mm account for more than 50% of the total mass, including boulders (more than 50% of particles larger than 200mm, rounded), boulders (more than 50% of particles larger than 200mm, angular), pebbles (more than 50% of particles larger than 2mm, rounded), and crushed stone (more than 50% of particles larger than 2mm, angular). Sandy soil (coarse-grained dominant): Particles larger than 0.075mm account for more than 50% of the total mass, and fine particles (<0.075mm) account for a very low percentage (usually <5%), such as gravelly sand (25%-50% of particles larger than 2mm) and coarse sand (more than 50% of particles larger than 0.5mm).

[0048] For high viscosity / high fineness materials include: The core characteristics of high-viscosity / high-fineness materials are a high proportion of fine particles, high clay mineral content, and strong plasticity. A comprehensive assessment is needed, considering particle size distribution, plasticity indices, and clay mineral type. Clayey soil (high viscosity dominant): Plasticity index IP > 10, high proportion of clay particles with a particle size < 0.002 mm, and clay mineral (montmorillonite, illite, kaolinite, etc.) content usually > 20%.

[0049] High fineness silt (fine particles dominant, weak cohesion): the content of particles with a diameter > 0.075mm is ≤ 50%, the plasticity index IP is ≤ 10, but the proportion of clay particles (< 0.002mm) is 10%-30%, the clay mineral content is 5%-20%, the properties are between sandy soil and clayey soil, the water retention is better than sandy soil, but the cohesion is weaker than silty clay.

[0050] The layered structure and hydrophilicity of clay minerals are the core reasons for high viscosity. The enhancing effect of different minerals on viscosity is ranked as follows: montmorillonite > illite > kaolinite.

[0051] In this embodiment, the impact data acquisition unit focuses on the entire process of waste soil transportation and stockpiling. It not only collects basic information such as stockpiling time and transportation distance, but also accurately captures the impact of moisture migration on the state of the waste soil by measuring the change in moisture content Δa (quantifying the rate of change in pore water volume) before and after transportation. Simultaneously, it incorporates the spatial location and slope aspect relationship of adjacent units, filling the gap in traditional methods that only focus on the excavation stage and lack data on subsequent stages. This provides comprehensive and accurate data support for subsequent analysis of reverse disturbances, avoiding problems such as amendment failure and sudden changes in waste soil fluidity due to moisture loss or accumulation. Secondly, the path coupling unit innovatively introduces path coupling coefficient calculation. It uses the overlap ratio of transportation routes (reflecting the degree of physical disturbance superposition) and the stockpiling time difference (combined with the time decay coefficient a) to adapt to the disturbance decay law of waste soils with different particle characteristics, such as coarse-grained materials. Higher values, faster disturbance decay, high viscosity materials With lower values ​​and longer-lasting disturbances, this module quantifies the mutual influence intensity between the c-th and d-th adjacent units, breaking the limitations of traditional methods that struggle to quantify the coupling effects between units. It can accurately identify "high-coupling-risk unit combinations" (such as units with high path coupling coefficients, which are prone to strong mutual disturbances due to overlapping transportation and close stacking times). Overall, this module can predict the risk of reverse disturbances during transportation and stacking (such as dilution of the modifier caused by moisture migration from adjacent units, and fluidity imbalance caused by mixing coarse and viscous units). This provides a scientific basis for the subsequent strategy generation module to formulate differentiated protection measures (such as adjusting transportation batches and optimizing stacking layouts for high-coupling-risk units), ensuring the stability of the soil improvement effect throughout the entire process from excavation to stacking, avoiding construction interruptions and safety hazards caused by reverse disturbances, and further improving the systematicness and reliability of shield tunneling soil improvement.

[0052] Example 5 This embodiment is an explanation based on Embodiment 1. Please refer to it. Figure 1Specifically, the reverse stability analysis module also includes an influence coefficient calculation unit and a disturbance classification unit; The influence coefficient calculation unit is used to combine the path coupling coefficient. Calculate the reverse influence coefficient between the c-th improvement unit and the d-th adjacent improvement unit, based on the change in moisture content Δa before and after transportation. :

[0053] in, and These represent the changes in moisture content before and after transportation for the c-th improved unit and the d-th adjacent improved unit, respectively. This represents the maximum absolute value of the change in moisture content of all improved units during the transportation phase in this analysis period; and Indicates the weighting coefficient; if →The two improved units are closely coupled in space and in transportation path, if The large difference in water content between the two improved units indicates that water redistribution or seepage may be intense. Larger units may cause significant disturbance during transportation or storage, increasing the risk of soil fluidity or dilution of the amendment. When transportation routes overlap significantly (e.g., the same vehicle transports excavated soil from adjacent units back and forth) and the stockpiling times are similar (e.g., the interval between stockpiling excavated soil from two units is less than 2 hours), the physical disturbances caused by path coupling (e.g., excavated soil mixing, cross-contamination of amendments) become more prominent. In this case, it is necessary to improve... If the transportation routes are independent and the storage areas are isolated, the impact of path coupling is weak, which can reduce... If the project is characterized by "mixed soil (silt + sand), semi-overlapping transportation routes, and semi-centralized stockpiling" (the most common scenario in the industry, with equilibrium under two-factor disturbances), then the path coupling coefficient weight should be set. =0.5, weight of the difference in moisture content =0.5, at which point the two types of disturbance factors contribute equally to the reverse influence coefficient, which can balance the risk of physical mixing caused by path overlap and the risk of sudden state change caused by moisture difference.

[0054] If the project targets "high-cohesive soil (montmorillonite content > 20%), with independent transportation routes but concentrated stockpiling" (moisture migration being the main disturbance), then set the following: =0.35, =0.65, focusing on amplifying the contribution of moisture content variation differences to the reverse effect, accurately identifying the risks of amendment dilution and soil clumping caused by water seepage.

[0055] If the project targets "coarse-grained soil (gravel + crushed stone, clay content <5%), with highly overlapping transport routes but dispersed stockpiling" (path coupling is the main disturbance), then set... =0.65, =0.35, which prioritizes the risk of soil mixing and pipeline blockage caused by overlapping transportation, and is suitable for the characteristics of coarse soil with small impact from moisture changes.

[0056] Disturbance-level unit, used to determine the reverse influence coefficient between the c-th improved unit and its d-th adjacent improved unit. Generate corresponding disturbance levels, including: like This indicates that the improved unit is less disturbed by adjacent improved units during transportation and stacking, has a low path coupling coefficient, and weak moisture transfer or disturbance potential, thus generating the first disturbance level. like This indicates that the modified unit may be affected by the disturbance of adjacent modified units to a certain extent during the transportation or stacking stage. The path coupling coefficient is moderate, and there is potential for moisture migration or disturbance, generating a second disturbance level. The fluidity of the slag may fluctuate locally, and the mixing of the modifier may be uneven. It is necessary to pay appropriate attention to the amount of modifier used or the mixing strategy. like This indicates that the modified unit is significantly disturbed by adjacent modified units during transportation or storage, with a high path coupling coefficient and strong potential for moisture migration or disturbance, generating a third disturbance level. The third disturbance level carries the risk of sudden changes in the fluidity of the slag and soil, and the modifier may be diluted or locally unstable, making the modification difficult. It is necessary to focus on adopting strategies such as increasing the amount of modifier, optimizing mixing, or adjusting moisture.

[0057] In this embodiment, the reverse influence coefficient calculation unit innovatively integrates the path coupling coefficient and the moisture content change Δa to construct the reverse influence coefficient. The path coupling coefficient reflects the physical correlation between the overlapping transport paths and stacking sequence of units, while the difference in moisture content change (combined with the normalization of the maximum moisture content change over the entire cycle) reflects the chemical and physical state disturbances caused by moisture migration. Furthermore, a weighting coefficient balances the influence of these two types of factors, breaking the limitations of traditional methods that rely solely on experience to judge interference between units. This design can accurately capture high-risk unit combinations with "tight physical coupling + large moisture differences" (such as units with large reverse influence coefficients), identifying potential problems such as uneven mixing of amendments due to path overlap and sudden changes in soil fluidity caused by moisture migration in advance. This provides a quantitative basis for subsequent risk prevention and control, avoiding amendment failure due to hidden disturbances.

[0058] On the other hand, the disturbance classification unit divides disturbance levels into three levels based on the reverse influence coefficient, clearly defining the risk characteristics and response directions for different levels: the first disturbance level corresponds to low interference, requiring no additional strategy adjustments and improving construction efficiency; the second disturbance level indicates the need to pay attention to the dosage of modifiers and mixing strategies to prevent local fluctuations from developing into construction problems; the third disturbance level directly points to high risk, requiring enhanced measures such as increasing the dosage of modifiers and optimizing mixing to accurately match the control needs of different risk levels. This classification method allows construction personnel to quickly locate high-disturbance-risk areas within the tunnel, avoiding the cost waste or safety hazards caused by "overlooking high risks" resulting from traditional "indiscriminate protection".

[0059] Furthermore, the synergistic effect of the two types of units forms a closed loop of "quantified impact - hierarchical control": the calculation of impact coefficients provides scientific data support for disturbance classification, ensuring that the classification results do not rely on subjective experience; and the disturbance classification transforms the abstract coefficient values ​​into intuitive risk levels.

[0060] Example 6 This embodiment is an explanation based on Embodiment 1. Please refer to it. Figure 1 Specifically, the strategy generation module includes a priority improvement sequence generation unit and a strategy generation unit. The priority improvement sequence generation unit is used to generate a priority improvement sequence based on the soil improvement level and disturbance level of the c-th improvement unit, in order to improve improvement efficiency and reduce transportation and storage disturbance risks, including: For improvement units with an improvement level of first improvement level and a disturbance level of third disturbance level, the front part of the generation sequence is prioritized for improvement to ensure that improvement units with high soil improvementability but high risk of disturbance from neighboring improvement units can be treated in a timely and sufficient manner. For improved units with an improvement level of first improvement level and a disturbance level of second disturbance level, and improved units with an improvement level of second improvement level and a disturbance level of third disturbance level, improvements are arranged in priority order according to the middle part of the similar improvement strategy merging sequence to ensure that improved units with high improveability or large disturbance impact can be processed appropriately in advance. For improved units with a first improvement level and a first disturbance level, and improved units with a second improvement level and a first or second disturbance level, the middle and later parts of the generation sequence are arranged in a conventional order to handle improved units with moderate soil improvementability and less impact from disturbance. For improvement units with an improvement level of three, regardless of the disturbance level, the end part of the generation sequence is improved according to the on-site resources, and special mixing or moisture adjustment strategies are adopted as needed to handle improvement units with high difficulty in improving slag and soil. By combining them in the above order, a priority improvement unit sequence is obtained, which is used to guide the implementation sequence of shield tunneling spoil improvement.

[0061] A policy generation unit, used to generate corresponding policies for a priority improvement unit sequence, including: For the front part of the sequence, a first strategy is generated, including: increasing the mixing intensity to +30% of the standard mixing intensity, increasing the dosage of the amendment by 20%, and adding 10% of the water-retaining agent during the mixing process, so as to ensure that the amendment units with high soil amendmentability but high risk of disturbance from adjacent amendment units can be treated in a timely and sufficient manner. For the middle part of the sequence, a second strategy is generated, including: increasing the mixing intensity by 15%, increasing the dosage of the amendment by 10%, and increasing the water-retaining agent by 5% in the moisture adjustment, so as to ensure that the amendment units with high soil amendmentability and moderate disturbance impact can be uniformly mixed and reduce the risk of local fluidity. For the middle and later parts of the generated sequence, a third strategy is generated, including: maintaining standard mixing intensity and modifier dosage, and maintaining the original moisture conditions, in order to handle the modification units with high soil modification potential and low disturbance from adjacent modification units, and to arrange modification in the conventional order. For the final part of the sequence, a fourth strategy is generated, including: increasing the mixing intensity by 20%, increasing the dosage of the amendment by 30%, and increasing the water-retaining agent by 10% to 15% in moisture adjustment, to handle the amendment units with high difficulty in soil amendment, and to adopt enhanced mixing or moisture adjustment strategies according to site requirements. In this embodiment, the strategy generation module addresses the pain points of "disordered sequence and blind strategy" in traditional improvement methods through a dual design of "precise priority sorting + targeted strategy matching." On one hand, the priority improvement sequence generation unit breaks away from the limitations of traditional "single sorting by segment or difficulty," innovatively combining improvement level (reflecting its own improvement difficulty) and disturbance level (reflecting external interference risk) to construct a four-level priority sequence: units with "high improveability but high disturbance risk" are placed at the front for priority processing to prevent them from becoming ineffective due to interference from adjacent units caused by delayed improvement; units with "high improveability / high disturbance risk" are placed in the middle for secondary priority, balancing efficiency and risk; units with "medium improveability and low disturbance" are sorted according to conventional methods to avoid resource waste; and units with "high improvement difficulty" are placed at the end to meet their specific resource requirements. This sorting method ensures that high-value (easy-to-improve) units are not destroyed by disturbances and reserves sufficient processing resources for high-difficulty units, completely solving the problem of "easy-to-improve units becoming ineffective due to disturbances and high-difficulty units being processed hastily" caused by traditional sorting, significantly improving overall improvement efficiency.

[0062] The strategy generation unit customizes differentiated strategies for different priority sequences to achieve precise matching of "demand and measures": the front-end unit adopts an enhanced scheme of "+30% mixing intensity + 20% modifier + 10% water-retaining agent" to specifically resist the risk of high disturbance; the middle-stage unit balances the improvement effect and cost through "15% enhanced mixing + 10% increased modifier"; the mid-to-late-stage units follow the standard scheme to avoid resource redundancy; and the final high-difficulty units tackle improvement challenges with "20% improved mixing + 30% increased modifier + 10-15% water-retaining agent". This "on-demand" design not only solves the problem of "excessive waste in easy-to-improve units and insufficient failure in difficult-to-improve units" caused by the traditional "one-size-fits-all" strategy, but also further ensures the stability of the improved slag and reduces the risk of sudden changes in fluidity during transportation and stockpiling through the fine adjustment of water-retaining agent and mixing intensity. The two main units work together to form a "sequencing-strategy" closed loop, which not only provides construction personnel with clear operational guidance on "what to modify first and how to modify it," reducing the difficulty of on-site decision-making, but also optimizes the cost structure and reduces ineffective investment through targeted allocation of resources (such as prioritizing the use of high-dose modifiers and strong mixing equipment in the front and rear units). At the same time, this design is deeply linked with the aforementioned reverse stability analysis and feasibility assessment (such as referencing the reverse influence coefficient and the physical properties corresponding to the modification level for strategy parameters), ensuring that the modification scheme is both adapted to the characteristics of the excavated soil and resistant to external disturbances, ultimately achieving a triple optimization of "efficiency, cost, and stability" in the modification of shield tunnel excavated soil.

[0063] The threshold is set to facilitate comparison. The size of the threshold depends on the amount of sample data and the number of bases set by those skilled in the art for each set of sample data; as long as it does not affect the ratio between the parameter and the quantized value, it is acceptable.

[0064] The above formulas are all derived from software simulation using a large amount of data and are selected to be close to the actual values. The coefficients in the formulas are set by those skilled in the art according to the actual situation. The above description is only a preferred embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any equivalent substitutions or changes made by those skilled in the art within the technical scope disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the protection scope of the present invention.

Claims

1. A machine learning-based intelligent shield tunneling spoil remediation system, characterized in that, include: The waste soil area division module is used to collect waste soil samples and related construction parameters generated during shield tunnel construction, establish a three-dimensional spatial model of the waste soil, and identify the microstructural characteristics, particle composition, and moisture content distribution of the waste soil. Based on machine learning K-means clustering, the waste soil is divided into several improvement units, and the feasibility index of the c-th improvement unit is constructed through a trained machine learning model. Based on the prediction results, several improvement units are classified to obtain the corresponding improvement levels and marked in the three-dimensional slag model. The waste soil area division module also includes a feasibility calculation unit and a classification marking unit; The feasibility calculation unit is first used to calculate the number of clustered samples in the c-th improvement unit. mineral sensitivity coefficient The average value is then used to obtain the comprehensive mineral sensitivity coefficient of the c-th improved unit. ; Mineral sensitivity coefficient The data was obtained by combining XRD and FTIR spectroscopy results to identify the main clay mineral components, including the proportions of illite, montmorillonite, and kaolinite; and constructing a mineral composition vector M=(m1,m2,...,m...) for each slag soil sample point. k );m j This represents the mass fraction of the j-th clay mineral; And calculate the mineral sensitivity coefficient of the i-th slag soil sample point. : in, The expansion sensitivity factor corresponding to the j-th clay mineral is represented, including: montmorillonite r=1.0, illite r=0.5, kaolinite r=0.2; the larger the mineral sensitivity coefficient Ms is, the stronger the swelling of the slag soil when it comes into contact with water, and the higher the dosage of the amendment is required for stabilization treatment. J represents the number of clay mineral types identified at the i-th slag soil sample point. And extract the plastic flow pressure coefficient of the c-th improved unit. Permeability coefficient Cohesive strength index and comprehensive mineral sensitivity coefficient After performing Min-Max normalization on the inverse indicators based on historical data, the feasibility index of the c-th improvement unit is obtained by weighted summation. ; The reverse stability analysis module is used to collect historical stockpiling data, moisture change data during transportation, and influence path information of adjacent improvement units for each improvement unit, and to establish an influence dataset to construct the reverse influence coefficient between the c-th improvement unit and the d-th adjacent improvement unit. The disturbances are then categorized to obtain their corresponding levels. The reverse stability analysis module includes an impact data acquisition unit and a path coupling unit; The impact data acquisition unit is used to collect environmental and state data of each improvement unit during shield tunneling and spoil transportation, including stockpiling time, transportation distance, transportation batches, change in moisture content Δa before and after transportation, as well as the spatial location and slope relationship of adjacent improvement units, to establish an improvement impact dataset. The expression for the change in moisture content Δa before and after transportation is: ;in, and These represent the pore water volumes before and after transportation. For solid volume; The path coupling unit is used to calculate the path coupling coefficient between each improvement unit based on the transportation route and stacking order, specifically: Calculate the overlap ratio and stacking sequence time difference between the c-th improved unit and its adjacent d-th improved unit on the transportation path to obtain the path coupling coefficient. : in, Let be the overlap distance between the c-th improved unit and its d-th adjacent improved unit in the transportation route. The total transportation distance. The time difference between stacking the two improved units. is the time decay coefficient; e is the natural exponential decay function; The reverse stability analysis module also includes an influence coefficient calculation unit and a disturbance classification unit; The influence coefficient calculation unit is used to combine the path coupling coefficient. Calculate the reverse influence coefficient between the c-th improvement unit and the d-th adjacent improvement unit, based on the change in moisture content Δa before and after transportation. ; The perturbation classification unit is used to determine the reverse influence coefficient between the c-th improvement unit and its d-th adjacent improvement unit. Generate corresponding disturbance levels, including: like This generates the first level of disturbance. like This generates a second level of disturbance. like This generates a third level of disturbance. The strategy generation module generates a priority improvement sequence based on the corresponding improvement level and disturbance level of the c-th improvement unit, and implements the corresponding strategy on the c-th improvement unit according to the priority improvement sequence.

2. The intelligent shield tunneling muck improvement system based on machine learning according to claim 1, characterized in that, The waste soil area division module includes a waste soil data acquisition unit and a microstructure identification unit. The excavated soil data acquisition unit is used to collect excavated soil samples generated at different times, under different cutterhead torques and propulsion pressures during shield tunneling construction, and to record the ring number and relevant construction parameters for each construction process, including the cutterhead torque of the corresponding ring segment. Cutting area of ​​the cutter head and propulsion speed ; Furthermore, a 3D laser scanner was used to scan the formation state of the construction waste during the construction process, obtain information on the morphology and volume distribution of the waste accumulation, and establish a preliminary spatial distribution dataset of the waste waste S0(x,y,z). The microstructure identification unit is used to identify the microstructure of the slag soil based on microscopic imaging and particle image analysis technology, and to obtain the average particle size. Fine particle content Specific surface area σ, porosity n e In addition to the natural moisture content w0, a micro-dataset was established.

3. The intelligent shield tunneling spoil remediation system based on machine learning according to claim 2, characterized in that, The waste soil area division module also includes an area segmentation unit and a physical feature calculation unit; The region segmentation unit is used to project the preliminary slag spatial distribution dataset S0(x,y,z) onto the shield tunnel coordinate system, establish a three-dimensional slag spatial model, and spatially partition the improvement unit according to the construction ring number or soil silo volume. Extract each soil sample point generated during the tunnel boring machine (TBM) excavation process, obtain its spatial coordinates (x, y, z), and the average particle size obtained from the microscopic dataset. Fine particle content Specific surface area σ, porosity n e and natural moisture content And the mineral composition vector M, which forms the physical property feature vector of each slag soil sample point: The physical property feature vectors of each soil sample point were normalized to a range of 0 and 1. Based on the shield tunneling section length and the number of samples, the K-means clustering method was used, setting the number of clusters K. Each cluster was numbered c=1, 2, ..., K, and each cluster corresponded to an "improvement unit". An initial cluster center was set for every 5 meters of tunneling length, and K soil sample points were selected as the initial cluster centers. : Calculate the weighted Euclidean distance between the i-th slag sample and each cluster center. ; Calculate the set of distances from samples within each cluster c to the center, and set a spatial radius threshold. The spatial range of the improvement unit is defined, and the spatial range of the c-th improvement unit is defined as follows: in, This represents the weighted Euclidean distance between a sample point and the cluster center. To improve the threshold of the unit space radius.

4. The intelligent shield tunneling muck improvement system based on machine learning according to claim 3, characterized in that, The physical characteristic calculation unit is used to calculate the plastic pressure coefficient of the c-th improved unit. Permeability coefficient and cohesive strength index This allows for the quantification of the physical properties of each improved unit, including: The plastic pressure coefficient of the c-th improved unit is calculated based on the principles of energy conservation and cutting specific energy. ; Based on the Kozeny–Carman equation seepage formula, the permeability coefficient of the c-th improved unit is calculated. ; Based on the Coulomb failure criterion and natural moisture content To calculate the cohesive strength index of the c-th improved unit. .

5. The intelligent shield tunneling muck improvement system based on machine learning according to claim 1, characterized in that, The classification labeling unit is used to determine the feasibility index based on the c-th improvement unit. To obtain the corresponding improvement level, including: like The first improvement level is generated, and the improvement unit is marked in green in the three-dimensional slag model. like A second improvement level is generated, and the improved unit is marked in yellow in the three-dimensional slag model. like The third improvement level is generated, and the improvement unit is marked in red in the three-dimensional slag model.

6. The intelligent shield tunneling muck improvement system based on machine learning according to claim 1, characterized in that, The strategy generation module includes a priority improvement sequence generation unit and a strategy generation unit. The priority improvement sequence generation unit is used to generate a priority improvement sequence based on the soil improvement level and disturbance level of the c-th improvement unit, including: For an improved unit with an improvement level of the first improvement level and a perturbation level of the third perturbation level, generate the front part of the sequence; For improved units with an improvement level of first improvement level and a disturbance level of second disturbance level, and improved units with an improvement level of second improvement level and a disturbance level of third disturbance level, the middle part of the merged sequence is included. For improved units with an improvement level of the first improvement level and a disturbance level of the first disturbance level, and improved units with an improvement level of the second improvement level and a disturbance level of the first or second disturbance level, the middle and later parts of the generation sequence are generated. For improved units with an improvement level of three, regardless of the perturbation level, the end part of the generated sequence is generated.

7. The intelligent shield tunneling muck improvement system based on machine learning according to claim 6, characterized in that, The strategy generation unit is used to generate corresponding strategies for the priority improvement unit sequence, including: For the front part of the sequence, a first strategy is generated, including: using an enhanced stirring intensity of +30% of the standard stirring intensity, increasing the modifier dosage by 20%, and adding a water-retaining agent by 10% during the mixing process; For the middle part of the sequence, a second strategy is generated, including: increasing the stirring intensity by 15%, increasing the modifier dosage by 10%, and increasing the water-retaining agent by 5% in moisture adjustment; For the middle and later parts of the generated sequence, a third strategy is generated, which includes: maintaining standard stirring intensity and modifier dosage, and maintaining the original moisture conditions; For the end portion of the sequence, a fourth strategy is generated, including: increasing the stirring intensity by 20%, increasing the modifier dosage by 30%, and increasing the water-retaining agent by 10% to 15% in moisture adjustment.

Images