A mud dispersant injection decision system and platform based on multi-source parameter fusion

Abstract

This invention relates to the field of intelligent control technology for tunnel boring machine (TBM) construction, and discloses a mud dispersant injection decision-making system and platform based on multi-source parameter fusion. The system includes a multi-source parameter acquisition module, a state quantification module, a conflict evaluation module, a collaborative control module, and an injection execution control module. By mapping mud performance maintenance efficiency values ​​and mud cake prevention efficiency values ​​to the phase plane, phase trajectory points are constructed, and the type and degree of efficiency imbalance are identified. Then, a machine learning model is used to collaboratively control the dispersant injection amount and propulsion speed. This invention fundamentally reconstructs the dispersant injection decision-making from the traditional prediction bias correction paradigm to a structure-performance conflict collaborative control paradigm, resolving the fundamental contradiction that mud performance maintenance and mud cake prevention cannot be simultaneously optimized, and realizing intelligent decision-making for dispersant injection during TBM tunneling.

Description

Technical Field

[0001] This invention relates to the field of intelligent control technology for tunnel boring machine (TBM) construction, and in particular to a mud dispersant injection decision system and platform based on multi-source parameter fusion. Background Technology

[0002] Currently, in the field of slurry shield tunneling, especially in the excavation of highly viscous strata, cutterhead cake formation is a major technical bottleneck causing construction stagnation and safety risks. Clay particles continuously accumulate on the cutterhead surface to form cakes, leading to a sharp increase in cutterhead torque and a significant decrease in tunneling speed. In severe cases, it requires shutdown and cleaning, causing project delays and economic losses. Dispersants, as key materials for inhibiting cake formation, reduce the risk of cake formation by improving the dispersion state of clay particles. They have been widely used in slurry shield tunneling projects. For example, PLC systems are used to achieve linked control of foam volume, slurry pressure, and propulsion speed, or fuzzy comprehensive evaluation models are used to evaluate slurry delivery performance through multiple indicators to guide dispersant injection.

[0003] However, existing dispersant injection decision-making methods generally suffer from a fundamental technical flaw: they fail to recognize the inherent structure-property conflict between maintaining mud performance and preventing cake formation, and even more so, they fail to treat this conflict as a core management object of the decision-making system. Specifically, increasing the amount of dispersant injected is beneficial for inhibiting cake formation, but it weakens the mud's slag-carrying capacity and excavation face support capacity; reducing the amount injected is beneficial for maintaining mud performance, but it increases the risk of cake formation. Existing technologies either employ fixed threshold-triggered injection or passively correct based on multi-source data prediction of cake risk. Essentially, both are static trade-offs between two conflicting performance objectives, rather than dynamic and coordinated regulation. When facing complex and variable high-viscosity strata, this decision-making method, lacking conflict perception and proactive coordination capabilities, cannot achieve a dynamic balance between maintaining mud performance and preventing cake formation, fundamentally restricting the synergistic improvement of efficiency and safety in slurry shield tunneling. Furthermore, existing technologies do not rely on industrial internet platforms to achieve cloud-based aggregation of multi-source data from shield tunneling, cross-equipment collaboration, and real-time analysis across the entire domain. The problem of data silos is prominent, making it difficult to adapt to the dynamic decision-making needs of dispersant injection in complex strata.

[0004] Therefore, this invention proposes a mud dispersant injection decision system and platform based on multi-source parameter fusion. Summary of the Invention

[0005] This invention provides a mud dispersant injection decision system and platform based on multi-source parameter fusion. By upgrading the natural contradiction between maintaining mud performance and preventing mud cake from a static trade-off to dynamic synergistic regulation, it has for the first time achieved a paradigm shift in dispersant injection decision-making from "passive response and single-objective optimization" to "active management of structure-performance conflict and synergistic balance of dual effectiveness".

[0006] This invention provides a mud dispersant injection decision system based on multi-source parameter fusion, comprising: The multi-source parameter acquisition module is used to acquire mud rheological parameters, shield tunneling parameters, and formation parameters during the shield tunneling process, and to perform industrial data processing on the acquired mud rheological parameters, shield tunneling parameters, and formation parameters. The state quantification module is used to calculate the mud performance maintenance efficiency value based on the mud rheological parameters and formation parameters after industrial data processing, and to calculate the mud cake prevention efficiency value based on the shield tunneling parameters and formation parameters after industrial data processing. The mud performance maintenance efficiency value reflects the degree of maintenance of the mud's slag carrying capacity and support capacity, and the mud cake prevention efficiency value reflects the degree of suppression of the risk of mud cake formation. The conflict evaluation module is used to map the mud performance maintenance efficiency value and mud cake prevention efficiency value to a phase plane with the mud performance maintenance efficiency value as the first coordinate axis and the mud cake prevention efficiency value as the second coordinate axis. It constructs phase trajectory points that characterize the current shield tunneling state and determines the type and degree of efficiency imbalance in the current state based on the deviation direction and deviation distance between the phase trajectory points and the optimal working point in the phase plane. The collaborative control module is used to calculate the dispersant injection amount control parameters and the propulsion speed control parameters based on the type and degree of performance imbalance using a machine learning model. The dispersant injection amount control parameters are used to adjust the amount of dispersant injected into the mud and water chamber injection point and the slurry discharge pipe injection point, and the propulsion speed control parameters are used to adjust the tunnel boring machine's propulsion speed. The injection execution control module is used to control the dispersant injection execution unit to add the corresponding amount of dispersant to the injection points of the mud and water chamber and the injection points of the slurry discharge pipe according to the dispersant injection amount control parameters, and to control the shield machine propulsion mechanism to adjust the propulsion speed according to the propulsion speed control parameters.

[0007] Furthermore, the conflict assessment module also includes a built-in self-correction unit for the performance evaluation framework. This self-correction unit performs the following functions: The instability monitoring subunit is used to continuously collect the historical sequence of mud cake control efficacy value and the historical sequence of mud performance maintenance efficacy value within a specified historical window, and calculate the global instability index of the historical sequence of mud cake control efficacy value. The global instability index is the ratio of the variance of the historical sequence of mud cake control efficacy value to the mean of the historical sequence of mud cake control efficacy value. The self-calibration triggering subunit is used to determine that the current formation is in a heterogeneous abrupt state when the global instability index exceeds the preset instability threshold, and to trigger the self-calibration of the performance evaluation framework. The local model training subunit is used to collect mud parameters, shield parameters and formation parameters of multiple consecutive tunneling rings as a self-calibration sample set after triggering the self-calibration of the performance evaluation framework. The clay content, water content and mineral composition data of the formation parameters of each sample in the self-calibration sample set are used as input features, and the actual mud cake prevention efficiency value and actual mud performance maintenance efficiency value of the corresponding sample are used as output labels to train and obtain a local performance prediction model. The effectiveness prediction sub-unit is used to input the latest formation parameters into the local effectiveness prediction model to obtain the predicted mud cake control effectiveness value and the predicted mud performance maintenance effectiveness value that reflect the current formation characteristics. The migration sub-unit is used to recalculate the coordinates of the optimal working point in the conflict evaluation module based on the estimated mud cake prevention efficiency value and the estimated mud performance maintenance efficiency value, and to migrate the original optimal working point coordinates on the phase plane to the new coordinates in a gradual manner, thereby completing the self-correction of the efficiency evaluation framework.

[0008] Furthermore, the state quantization module also incorporates a built-in apparent viscosity-true performance asymmetric response compensation unit, which includes: The sequence acquisition subunit is used to continuously acquire the mud viscosity sequence and the corresponding mud cake control efficacy value sequence within a preset observation window after the dispersant is injected; The apparent normalization subunit is used to normalize the mud viscosity sequence to obtain the apparent dispersion sequence. The true normalization subunit is used to normalize the mud cake control efficacy value sequence to obtain the true efficacy response sequence. The cross-correlation calculation subunit is used to calculate the dynamic cross-correlation function between the apparent dispersion sequence and the true performance response sequence, and to determine the lag time of the true performance response sequence relative to the apparent dispersion sequence. The gain coefficient calculation subunit is used to calculate the ratio of the instantaneous rate of change of the apparent dispersion sequence to the instantaneous rate of change of the actual performance response sequence, and to obtain the asymmetric response gain coefficient. The asymmetric compensation subunit is used to determine that the current response mode is asymmetric when the asymmetric response gain coefficient is lower than the preset lower threshold or exceeds the preset upper threshold. It multiplies the asymmetric response gain coefficient with the dispersant injection amount control parameter calculated by the collaborative control module to generate the compensated dispersant injection amount control parameter, and feeds the compensated dispersant injection amount control parameter back to the collaborative control module as the final value of the dispersant injection amount control parameter.

[0009] Furthermore, the coordinated control module also incorporates a built-in hydraulic transport lag effect prediction and compensation unit, which includes: The pipeline hysteresis calculation subunit is used to obtain the real-time mud flow velocity, length and cross-sectional area of ​​the slurry discharge pipeline in the shield tunnel circulation system, and to calculate the pipeline transport hysteresis time of mud flowing from the mud-water chamber to the outlet of the slurry discharge pipe. The mud-water tank replacement calculation subunit is used to divide the mud-water tank volume by the current mud flow rate to calculate the time for complete mud replacement in the mud-water tank. The total hysteresis calculation subunit is used to add the pipeline transport hysteresis time to the time of complete replacement of the slurry tank to obtain the total hydraulic hysteresis time of the circulation system. The time alignment subunit is used to obtain the mud cake control efficiency value currently calculated by the state quantization module, and to align the current mud cake control efficiency value with the historical mud cake control efficiency value before the total hydraulic delay time to obtain the mud cake control efficiency value before correction and the mud cake control efficiency value after correction. The trend prediction subunit is used to predict the trend of mud cake control effectiveness change reflected by the mud that is about to reach the outlet of the discharge pipe within the hysteresis time window, based on the mud cake control effectiveness value before correction and the mud cake control effectiveness value after correction. The hysteresis compensation solution subunit is used to input the trend of mud cake control efficacy changes into the control parameter solution process of the collaborative control module, and solve the dispersant injection amount control parameter and propulsion speed control parameter after hysteresis effect prediction compensation.

[0010] Furthermore, when the collaborative control module determines that the type of performance imbalance is insufficient mud cake control performance, it calculates the dispersant injection rate control parameters and the propulsion speed control parameters according to the following rules: The dispersant injection amount control parameter is taken as a positive increment, which is the product of the degree of imbalance and the baseline dispersant injection amount; The propulsion speed control parameter is set to a negative reduction, which is the product of the degree of imbalance and the reference propulsion speed. When the imbalance is determined to be due to insufficient mud performance maintenance efficiency, the dispersant injection amount control parameter is set to a negative decrease, which is the product of the degree of imbalance and the baseline dispersant injection amount. The propulsion speed control parameter is set to a positive increase, which is the product of the degree of imbalance and the baseline propulsion speed.

[0011] Furthermore, the optimal operating point in the conflict assessment module is determined as follows: From the historical tunneling database, tunneling rings that simultaneously meet the preset excellent standards for mud performance maintenance efficiency and mud cake prevention efficiency are selected, and the coordinates of the phase trajectory points corresponding to the tunneling rings are used as candidate optimal points. For each candidate optimal point, calculate the number of candidate optimal points contained within a preset neighborhood radius centered on each candidate optimal point, and use this as the local density of each candidate optimal point; Sort all candidate optima in descending order of local density, and select the candidate optima with the largest local density as the optimal operating point.

[0012] Furthermore, it also includes a model self-calibration function module, which includes: The deviation calculation subunit is used to collect the actual mud performance maintenance efficiency value and the actual mud cake prevention efficiency value after each adjustment, and to calculate the change in the actual efficiency value. The deviation comparison subunit is used to compare the actual change in efficiency value with the theoretical change in efficiency value and calculate the efficiency prediction deviation. The theoretical change in efficiency value is the expected change in efficiency value for this regulation determined by the model self-correction function module based on the historical change in efficiency value corresponding to samples with similar features to the current input in the training data of the state quantization module. The calibration trigger subunit is used to trigger parameter calibration of the state quantization module and the conflict evaluation module when the performance prediction deviation exceeds the preset deviation threshold in a series of preset adjustments.

[0013] Furthermore, the parameter calibration process of the model self-calibration function module includes: Obtain the pre-control state point corresponding to the performance prediction deviation exceeding the preset deviation threshold, and use the mud rheological parameters, shield tunneling parameters and formation parameters of the corresponding pre-control state point as input features, and the actual performance value change as output label to construct calibration samples; Add the calibration samples to the training dataset of the state quantization module; With the goal of minimizing the prediction error of the efficiency value of the calibration sample by the state quantization module, the gradient descent method is used to incrementally update the efficiency calculation coefficients in the state quantization module.

[0014] Furthermore, the conflict assessment module also has experience-learning capabilities, including: The experience storage subunit is used to obtain the state parameters at each time the control is executed and the actual performance compensation amount after the control is executed. The state parameters include the coordinates, deviation direction and deviation distance of the phase trajectory point before the control. The state parameters and the actual performance compensation amount are stored as a historical control experience record and associated with it in the control experience database. The experience retrieval subunit is used to retrieve historical control experience records in the control experience base after the conflict evaluation module determines the new type and degree of performance imbalance, using the coordinates, deviation direction and deviation distance of the current phase trajectory point as retrieval conditions. The similarity between the records and the retrieval conditions exceeds a preset similarity threshold. The regulation effect prediction subunit is used to predict the regulation effect of the current regulation based on the actual efficiency value compensation amount in the retrieved historical regulation experience records. The regulation correction subunit is used to correct the dispersant injection amount regulation parameters and propulsion speed regulation parameters calculated by the collaborative regulation module when the estimated regulation effect is lower than the preset minimum expected compensation amount.

[0015] This invention provides a mud dispersant injection decision platform based on multi-source parameter fusion, including a processor and a machine-readable storage medium. The machine-readable storage medium is connected to the processor and stores a computer program. The processor executes the computer program to achieve the functions of any of the above-mentioned mud dispersant injection decision systems based on multi-source parameter fusion.

[0016] The beneficial effects of this invention compared to existing technologies are as follows: This invention fundamentally reconstructs the dispersant injection decision-making from the traditional "prediction bias correction" paradigm to a "structure-performance conflict synergistic control" paradigm. For the first time, it quantitatively manages the mud performance maintenance efficiency and mud cake prevention efficiency as a pair of efficiency pairs with a non-simultaneously optimal structure-performance conflict relationship. By identifying the type and degree of efficiency imbalance through phase plane mapping and deviation vector analysis, and then using a machine learning model to synergistically control the dispersant injection amount and propulsion speed, it maintains an optimal balance between the two efficiencies throughout the tunneling process in dynamically changing high-viscosity strata. This fundamentally solves the "one-sided" dilemma caused by the inability of existing technologies to handle structure-performance conflicts, achieving a synergistic improvement in shield tunneling efficiency and safety. This invention can also connect to an industrial internet platform to achieve cloud-based fusion of multi-source parameters, real-time issuance of decision commands, and remote monitoring of construction status, further improving the intelligent management and control system for shield tunneling.

[0017] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in this application.

[0018] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0019] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a schematic diagram of the mud dispersant injection decision system based on multi-source parameter fusion in an embodiment of the present invention; Figure 2 This is a schematic diagram of the structure-performance conflict synergistic regulation closed-loop architecture in an embodiment of the present invention. Detailed Implementation

[0020] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0021] refer to Figure 1 and Figure 2 This invention provides an embodiment of a mud dispersant injection decision system based on multi-source parameter fusion, comprising: The multi-source parameter acquisition module is used to acquire mud rheological parameters, shield tunneling parameters, and formation parameters during the shield tunneling process, and to perform industrial data processing on the acquired mud rheological parameters, shield tunneling parameters, and formation parameters. The state quantification module is used to calculate the mud performance maintenance efficiency value based on the mud rheological parameters and formation parameters after industrial data processing, and to calculate the mud cake prevention efficiency value based on the shield tunneling parameters and formation parameters after industrial data processing. The mud performance maintenance efficiency value reflects the degree of maintenance of the mud's slag carrying capacity and support capacity, and the mud cake prevention efficiency value reflects the degree of suppression of the risk of mud cake formation. The conflict evaluation module is used to map the mud performance maintenance efficiency value and mud cake prevention efficiency value to a phase plane with the mud performance maintenance efficiency value as the first coordinate axis and the mud cake prevention efficiency value as the second coordinate axis. It constructs phase trajectory points that characterize the current shield tunneling state and determines the type and degree of efficiency imbalance in the current state based on the deviation direction and deviation distance between the phase trajectory points and the optimal working point in the phase plane. The collaborative control module is used to calculate the dispersant injection amount control parameters and the propulsion speed control parameters based on the type and degree of performance imbalance using a machine learning model. The dispersant injection amount control parameters are used to adjust the amount of dispersant injected into the mud and water chamber injection point and the slurry discharge pipe injection point, and the propulsion speed control parameters are used to adjust the tunnel boring machine's propulsion speed. The injection execution control module is used to control the dispersant injection execution unit to add the corresponding amount of dispersant to the injection points of the mud and water chamber and the injection points of the slurry discharge pipe according to the dispersant injection amount control parameters, and to control the shield machine propulsion mechanism to adjust the propulsion speed according to the propulsion speed control parameters.

[0022] In this embodiment, the shield tunneling process refers to the process where a slurry shield machine uses its cutterhead to cut through the rock and soil at the tunnel face, while simultaneously transporting the cut material to a surface slurry treatment station via a slurry circulation system. After tunneling one ring width of tunnel segments, cutting stops for segment assembly, and then the cycle of tunneling the next ring begins. During shield tunneling, the slurry rheological parameters, shield tunneling parameters, and geological parameters continuously change as the tunneling ring advances.

[0023] In this embodiment, mud rheological parameters refer to physical parameters reflecting the flow and deformation characteristics of mud in the real-time monitoring of the circulating mud system in a slurry shield tunnel. These parameters include mud viscosity, mud density, and dynamic shear force. Mud viscosity reflects the internal frictional resistance of mud flow, mud density reflects the mass per unit volume of mud, and dynamic shear force reflects the minimum shear stress required for mud to begin flowing. These parameters are collected in real-time by viscometers and densitometers installed on the inlet and outlet pipes.

[0024] In this embodiment, the tunneling parameters refer to the mechanical parameters that the slurry shield tunneling machine monitors in real time during the tunneling process, reflecting the interaction between the cutterhead and the tunnel face. These parameters include cutterhead thrust, cutterhead torque, propulsion speed, and cutterhead rotation speed. Cutterhead thrust reflects the total pressure exerted by the cutterhead on the tunnel face; cutterhead torque reflects the resistance torque experienced by the cutterhead as it rotates and cuts through the rock and soil; propulsion speed reflects the linear velocity of the tunneling machine's forward tunneling; and cutterhead rotation speed reflects the number of revolutions the cutterhead makes per minute. These tunneling parameters are collected in real time by sensors installed on the main drive system and propulsion cylinders of the tunneling machine.

[0025] In this embodiment, the formation parameters refer to parameters reflecting the physical and mechanical properties of the strata currently being traversed by the tunnel boring machine (TBM). These parameters include clay content, moisture content, and mineral composition. Clay content reflects the percentage by mass of clay particles smaller than 0.005 mm in diameter; moisture content reflects the ratio of water mass to dry soil particle mass; and mineral composition reflects the types and proportions of clay minerals in the strata, particularly the content of highly expansive minerals such as montmorillonite. These formation parameters are obtained from the geological survey report prior to TBM excavation and are dynamically updated during the excavation process based on the analysis results of soil and slag samples taken from the slurry outlet.

[0026] In this embodiment, industrial data processing is performed on the collected mud rheological parameters, tunnel boring parameters, and formation parameters. Industrial data processing refers to a series of preprocessing operations on the raw sensor data, including removing outliers caused by sensor malfunctions, filling missing data with the average of normal values ​​from the previous and next time points, filtering out high-frequency random noise using a moving average method, and aligning parameters from different sampling frequencies to the same timestamp. Industrial data processing eliminates noise and errors in the raw data, providing stable and reliable input for subsequent quantitative calculations of performance. The multi-source parameter acquisition module is the main entity performing industrial data processing.

[0027] In this embodiment, the mud performance maintenance effectiveness value is calculated based on the mud rheological parameters and formation parameters after industrial data processing. Specifically, the mud viscosity, mud density, and dynamic shear stress are compared with their respective preset optimal ranges and normalized to obtain viscosity maintenance scores, density maintenance scores, and dynamic shear stress maintenance scores. Based on the clay content and water content in the formation parameters, the demand weights for mud carrying capacity and support capacity of the current formation are determined. These demand weights reflect the higher requirements for mud carrying capacity in high-clay formations and the higher requirements for mud support capacity in high-permeability formations. The viscosity maintenance score, density maintenance score, and dynamic shear stress maintenance score are weighted and summed with the demand weights to obtain the mud performance maintenance effectiveness value. The mud performance maintenance effectiveness value ranges from 0 to 1, with a value closer to 1 indicating better mud performance maintenance.

[0028] In this embodiment, the mud cake prevention effectiveness value is calculated based on the shield tunneling parameters and formation parameters after industrial data processing. Specifically, the ratio of cutterhead torque to propulsion speed is used as the penetration resistance index, and the ratio of cutterhead thrust to propulsion speed is used as the propulsion resistance index. The penetration resistance index and propulsion resistance index are combined with the clay content and mineral composition data in the formation parameters to construct a mud cake formation feature vector. This feature vector is then input into a pre-trained random forest regression model, which outputs the probability of mud cake formation. The random forest regression model is constructed as follows: the mud cake formation degree labels confirmed by cutterhead opening inspections in historical tunneling loops are used as training targets, and the penetration resistance index, propulsion resistance index, clay content, and montmorillonite ratio of the corresponding historical tunneling loops are used as training features. The model is trained using a random forest algorithm with 100 decision trees and a maximum depth of 10 layers. The result of subtracting the probability of mud cake formation from 1 is taken as the mud cake control efficacy value. The mud cake control efficacy value ranges from 0 to 1. The closer the value is to 1, the better the mud cake control efficacy.

[0029] In this embodiment, the mud carrying capacity refers to the mud's ability to carry clay particles and rock cuttings from the cutterhead from the mud tank to the surface mud treatment station. The mud carrying capacity depends on the matching degree between the mud viscosity and density and the mud flow velocity in the discharge pipeline. The support maintenance capacity refers to the ability of the mud film formed at the excavation face to prevent groundwater infiltration and maintain the stability of the working face. The support maintenance capacity depends on the matching degree between the mud density and dynamic shear force and the formation permeability coefficient. The mud performance maintenance efficiency value comprehensively reflects the maintenance degree of both the mud carrying capacity and the support capacity.

[0030] In this embodiment, the degree of suppression of cake formation risk refers to the reduction in the probability of cake formation on the cutterhead surface and in the slurry chamber after improving the dispersion state of clay particles through dispersant injection. Cake formation is a process in which clay particles continuously aggregate and adhere under the high temperature and pressure of the cutterhead. The higher the degree of suppression of cake formation risk, the greater the probability that the cutterhead torque and cutterhead thrust will remain within the normal range.

[0031] In this embodiment, mapping the mud performance maintenance effectiveness value and the mud cake prevention effectiveness value onto a phase plane with the mud performance maintenance effectiveness value as the first coordinate axis and the mud cake prevention effectiveness value as the second coordinate axis means that a point is marked in a two-dimensional rectangular coordinate system with the mud performance maintenance effectiveness value as the abscissa and the mud cake prevention effectiveness value as the ordinate, as the phase trajectory point of the current shield tunneling state. The phase trajectory point moves continuously in the phase plane as the shield tunneling progresses, forming a phase trajectory. Phase plane analysis provides a visualized and quantifiable state characterization basis for subsequent conflict assessment.

[0032] In this embodiment, the optimal working point in the phase plane refers to the coordinate point in the phase plane where the mud performance maintenance efficiency value and the mud cake prevention efficiency value simultaneously reach an ideal balance. The optimal working point is determined as follows: From the historical tunneling database, tunneling loops where both mud performance maintenance efficiency and mud cake prevention efficiency values ​​simultaneously meet a preset excellent standard are selected. The coordinates of the phase trajectory points corresponding to these tunneling loops are used as candidate optimal points. The number of candidate optimal points within a preset neighborhood radius for each candidate optimal point is calculated as the local density. All candidate optimal points are sorted in descending order of local density, and the candidate optimal point with the highest local density is selected as the optimal working point. The preset excellent standard means that both the mud performance maintenance efficiency value and the mud cake prevention efficiency value are not lower than 0.7.

[0033] In this embodiment, the deviation direction of the phase trajectory point from the optimal working point in the phase plane refers to the vector direction of the current phase trajectory point relative to the optimal working point with the optimal working point as the origin; the deviation distance refers to the Euclidean distance between the current phase trajectory point and the optimal working point. The larger the deviation distance, the more the current shield tunneling state deviates from the ideal equilibrium state.

[0034] In this embodiment, the type and degree of performance imbalance in the current state are determined based on the deviation direction and distance between the phase trajectory point and the optimal operating point in the phase plane. Specifically, the phase plane is divided into four quadrants, with the optimal operating point as the origin. When the phase trajectory point is located in the first quadrant and its ordinate is less than the ordinate of the optimal operating point while its abscissa is greater, the performance imbalance type is determined to be an imbalance of excessive mud performance and high cake risk. When the phase trajectory point is located in the third quadrant and its abscissa is less than the abscissa of the optimal operating point while its ordinate is greater, the performance imbalance type is determined to be an imbalance of excessive cake prevention and low mud performance. When the phase trajectory point is located in the fourth quadrant and both its abscissa and ordinate are less than the corresponding coordinates of the optimal operating point, the performance imbalance type is determined to be a double performance degradation imbalance. The degree of imbalance is quantified by the deviation distance.

[0035] In this embodiment, the machine learning model refers to a pre-trained gradient boosting tree model. The gradient boosting tree model is trained using the type and degree of performance imbalance, current mud viscosity, current cutterhead torque, and current clay content as input features, and the dispersant injection rate control parameters and propulsion speed control parameters as output labels. The gradient boosting tree model is constructed as follows: using the actual dispersant injection rate adjustment values ​​and actual propulsion speed adjustment values ​​recorded before and after each dispersant injection control in the historical tunneling database as training targets, and using the performance imbalance type, degree of imbalance, mud viscosity, cutterhead torque, and clay content at the corresponding historical moment before control as training features, it is trained using a gradient boosting tree algorithm with 200 trees, a learning rate of 0.05, and a maximum depth of 6 layers. The inputs to the gradient boosting tree model are the performance imbalance type, degree of imbalance, current mud viscosity, current cutterhead torque, and current clay content, and the outputs are the dispersant injection rate control parameters and propulsion speed control parameters.

[0036] In this embodiment, during the initial deployment of the system, historical tunneling data from existing shield tunneling projects in the same region or similar geological formations are used to pre-train the gradient boosting tree model and the random forest regression model. After the system goes online, a control rule based on a fixed threshold is used as a transition scheme within an initial preset number of tunneling loops, with the preset number being 100 loops, while training data is accumulated. Once the accumulated training data reaches a preset amount, the system automatically switches to a machine learning model-driven control mode.

[0037] In this embodiment, based on the type and degree of performance imbalance, a machine learning model calculates the dispersant injection rate control parameters and the propulsion speed control parameters. This involves inputting the performance imbalance type and degree output from the conflict assessment module, along with the current mud viscosity, current cutterhead torque, and current clay content provided by the multi-source parameter acquisition module, into a pre-trained gradient boosting tree model. The gradient boosting tree model simultaneously outputs the dispersant injection rate control parameters and the propulsion speed control parameters. The dispersant injection rate control parameter represents the absolute value by which the current dispersant injection rate needs to be increased or decreased, in liters per minute (L / min). The propulsion speed control parameter represents the absolute value by which the current propulsion speed needs to be increased or decreased, in millimeters per minute (mm / min).

[0038] In this embodiment, the amount of dispersant injected into the slurry tank injection point and the slurry discharge pipe injection point refers to the volume of dispersant added to the slurry tank and the slurry discharge pipe per unit time through the injection pipeline. The slurry tank injection point and the slurry discharge pipe injection point have different functional roles: the dispersant injected into the slurry tank acts directly on the cutterhead cutting area and the clay particles in the slurry tank, inhibiting the aggregation of clay particles on the cutterhead surface to form a mud cake; the dispersant injected into the slurry discharge pipe acts on the slurry in the slurry discharge pipe, preventing clay particles from accumulating on the inner wall of the slurry discharge pipe and clogging the pipe. The dispersant injection execution unit achieves precise control of the dispersant injection amount by adjusting the opening of the electric proportional valve on the injection pipeline.

[0039] In this embodiment, the tunnel boring machine (TBM) propulsion speed refers to the distance the TBM's propulsion cylinder advances per unit time, measured in millimeters per minute. The propulsion speed is adjusted by controlling the flow of hydraulic oil into the cylinder via a hydraulic proportional valve. Increasing the propulsion speed reduces the residence time of clay particles on the cutterhead surface, thus helping to suppress mud cake formation; however, excessive propulsion speed may lead to instability of the excavation face. Conversely, decreasing the propulsion speed helps maintain excavation face stability but increases the residence time of clay particles on the cutterhead surface, thereby increasing the risk of mud cake formation.

[0040] In this embodiment, controlling the dispersant injection execution unit to add the corresponding amount of dispersant to the mud-water tank injection point and the slurry discharge pipe injection point according to the dispersant injection amount control parameter means that the injection execution control module converts the dispersant injection amount control parameter output by the coordination control module into a control signal for the electric proportional valve. The electric proportional valve adjusts the valve core opening according to the control signal, so that the volume of dispersant added to the mud-water tank injection point and the slurry discharge pipe injection point per unit time is equal to the injection amount value specified by the dispersant injection amount control parameter. The dispersant injection execution unit is equipped with an independent electric proportional valve and a flow meter at each of the mud-water tank injection point and the slurry discharge pipe injection point. The flow meter provides real-time feedback on the actual injection amount, and the electric proportional valve automatically adjusts its opening according to the deviation between the feedback value and the target value, realizing closed-loop control of the injection amount.

[0041] In this embodiment, the dispersant injection execution unit refers to a collection of equipment used for storing, transporting, and injecting dispersant, including a dispersant storage tank, an injection pump, an electric proportional valve, a flow meter, and an injection pipeline. The dispersant storage tank is used to store liquid dispersant; the injection pump is used to extract the dispersant from the storage tank and pressurize it for delivery to the injection pipeline; the electric proportional valve is used to adjust the flow cross-sectional area of ​​the injection pipeline according to a control signal to control the dispersant flow rate; the flow meter is used to measure the volume of dispersant passing through the injection pipeline per unit time in real time and feed it back to the injection execution control module; the injection pipeline is used to transport the dispersant from the storage tank to the mud-water tank injection point and the slurry discharge pipe injection point.

[0042] In this embodiment, the slurry injection point refers to the interface location of the dispersant injection pipeline on the slurry chamber wall of the slurry shield machine. The slurry chamber is a sealed compartment located behind the cutterhead and in front of the air pressure regulation system, used to contain slurry and balance the water and soil pressure at the working face. The clay particles cut by the cutterhead are fully mixed with the slurry in the slurry chamber. After the dispersant enters the slurry chamber through the slurry injection point, it is fully mixed with the slurry containing clay particles under the stirring action of the slurry chamber agitator, directly acting on the surface of the clay particles, improving the dispersion state of the clay particles, and inhibiting the aggregation of clay particles on the cutterhead surface and in the slurry chamber.

[0043] In this embodiment, the slurry discharge pipe injection point refers to the interface location of the dispersant injection pipeline on the slurry discharge pipeline of the slurry shield tunnel. The slurry discharge pipeline is a slurry transport pipeline connecting the slurry tank and the ground slurry treatment station, responsible for transporting slurry carrying excavated soil from the slurry tank to the ground. After the dispersant enters the slurry discharge pipeline through the injection point, it mixes with the slurry during the slurry transport along the pipeline, inhibiting the adhesion and aggregation of clay particles on the inner wall of the slurry discharge pipeline, and preventing slurry circulation system failure due to slurry cake blockage.

[0044] In this embodiment, controlling the tunnel boring machine's propulsion mechanism to adjust the propulsion speed based on the propulsion speed control parameters means that the injection execution control module converts the propulsion speed control parameters output by the collaborative control module into control signals for the hydraulic proportional valves of the tunnel boring machine's propulsion cylinders. The hydraulic proportional valves adjust the flow rate of hydraulic oil entering the propulsion cylinders according to the control signals, changing the piston rod extension speed of the propulsion cylinders, thereby adjusting the forward propulsion speed of the tunnel boring machine. The propulsion mechanism includes propulsion cylinders, a hydraulic pump station, hydraulic proportional valves, and displacement sensors. The displacement sensors measure the piston rod extension speed of the propulsion cylinders in real time and feed it back to the injection execution control module. The injection execution control module compares the actual propulsion speed with the target propulsion speed and adjusts the control signal of the hydraulic proportional valves according to the deviation, achieving closed-loop control of the propulsion speed.

[0045] In this embodiment, the tunnel boring machine (TBM) propulsion mechanism refers to the mechanical and hydraulic equipment assembly that drives the TBM forward. It includes multiple sets of propulsion cylinders evenly distributed along the circumference of the TBM, a hydraulic pump station supplying high-pressure hydraulic oil to the propulsion cylinders, hydraulic proportional valves controlling the flow rate and direction of hydraulic oil entering each set of propulsion cylinders, and displacement sensors that measure the extension length and speed of the piston rod of each set of propulsion cylinders in real time. Through the coordinated extension and retraction of multiple sets of propulsion cylinders, the propulsion mechanism not only provides forward propulsion force to the TBM but also adjusts the tunneling direction by controlling the stroke difference between the sets of propulsion cylinders.

[0046] Furthermore, the conflict assessment module also includes a built-in self-correction unit for the performance evaluation framework, which comprises: The instability monitoring subunit is used to continuously collect the historical sequence of mud cake control efficacy value and the historical sequence of mud performance maintenance efficacy value within a specified historical window, and calculate the global instability index of the historical sequence of mud cake control efficacy value. The global instability index is the ratio of the variance of the historical sequence of mud cake control efficacy value to the mean of the historical sequence of mud cake control efficacy value. The self-calibration triggering subunit is used to determine that the current formation is in a heterogeneous abrupt state when the global instability index exceeds the preset instability threshold, and to trigger the self-calibration of the performance evaluation framework. The local model training subunit is used to collect mud parameters, shield parameters and formation parameters of multiple consecutive tunneling rings as a self-calibration sample set after triggering the self-calibration of the performance evaluation framework. The clay content, water content and mineral composition data of the formation parameters of each sample in the self-calibration sample set are used as input features, and the actual mud cake prevention efficiency value and actual mud performance maintenance efficiency value of the corresponding sample are used as output labels to train and obtain a local performance prediction model. The effectiveness prediction sub-unit is used to input the latest formation parameters into the local effectiveness prediction model to obtain the predicted mud cake control effectiveness value and the predicted mud performance maintenance effectiveness value that reflect the current formation characteristics. The migration sub-unit is used to recalculate the coordinates of the optimal working point in the conflict evaluation module based on the estimated mud cake prevention efficiency value and the estimated mud performance maintenance efficiency value, and to migrate the original optimal working point coordinates on the phase plane to the new coordinates in a gradual manner, thereby completing the self-correction of the efficiency evaluation framework.

[0047] In this embodiment, the designated historical window refers to the time interval consisting of a preset number of tunneling rings, ending at the current tunneling ring. The preset number is set to 50 rings, and the window length is adjusted according to the homogeneity of the strata traversed by the tunnel boring machine. The more frequent the changes in the strata, the shorter the window length is set.

[0048] In this embodiment, continuously collecting the historical sequence of mud cake prevention effectiveness value and the historical sequence of mud performance maintenance effectiveness value within a specified historical window means that the self-correction unit of the effectiveness evaluation framework extracts the mud cake prevention effectiveness value and the mud performance maintenance effectiveness value corresponding to each tunneling ring from the effectiveness value log of the state quantification module according to the time range of the specified historical window, and arranges them according to the time order of the tunneling rings to form the historical sequence of mud cake prevention effectiveness value and the historical sequence of mud performance maintenance effectiveness value.

[0049] In this embodiment, the preset instability threshold is a threshold value used to determine whether there are global abnormal fluctuations in the historical sequence of mud cake control effectiveness. The preset instability threshold is set to 0.3. When the global instability index exceeds 0.3, it indicates that the mud cake control effectiveness has experienced drastic changes beyond the normal random fluctuation range within the specified historical window. Such drastic changes usually mean that the formation physical properties have undergone fundamental changes, rather than being caused by normal adjustments to the tunneling parameters.

[0050] In this embodiment, the current formation being in a heterogeneous abrupt change state means that the formation currently being traversed by the tunnel boring machine has undergone significant changes in terms of clay content, water content, or mineral composition, causing the original performance evaluation logic to no longer be applicable to the new formation conditions. The heterogeneous abrupt change state is determined by the global instability index exceeding a preset instability threshold. Once the heterogeneous abrupt change state is confirmed, the performance evaluation framework self-correction unit automatically initiates the self-correction process.

[0051] In this embodiment, a tunneling ring refers to the work unit corresponding to the tunneling machine completing the excavation of one ring of tunnel segments. A tunneling ring includes the entire process from the start of cutting by the cutterhead to the cessation of cutting and the completion of the assembly of the ring of tunnel segments. Shield tunneling construction uses the tunneling ring as the basic unit for data recording and management. Each tunneling ring corresponds to a complete set of data records for mud rheological parameters, shield tunneling parameters, and formation parameters.

[0052] In this embodiment, collecting mud parameters, shield parameters, and formation parameters from multiple consecutive tunneling rings as a self-calibration sample set means that after triggering the performance evaluation framework self-calibration, the performance evaluation framework self-calibration unit obtains the mud parameters, shield parameters, and formation parameters of the current tunneling ring and a preset number of tunneling rings preceding it from the multi-source parameter acquisition module. The preset number is 20 rings, and the 20 sets of data collected constitute the self-calibration sample set. Each set of data includes the mud rheological parameters, shield tunneling parameters, and formation parameters of the corresponding tunneling ring.

[0053] In this embodiment, the sample formation parameters refer to the formation parameters corresponding to each sample in the self-calibrated sample set, including clay content, water content, and mineral composition data. The sample formation parameters are obtained from the geological survey report of each tunneling ring or the analysis results of the slag sample taken from the outlet of the slurry pipe.

[0054] In this embodiment, clay content refers to the percentage of clay particles with a diameter less than 0.005 mm in the total mass of the soil. Moisture content refers to the ratio of the mass of water in the formation to the dry mass of soil particles, expressed as a percentage. Mineral composition data refers to the types of clay minerals in the formation and the percentage of each type by mass, including the percentages of montmorillonite, illite, and kaolinite. Montmorillonite is a clay mineral with high expansibility and a high specific surface area; a higher montmorillonite percentage increases the risk of mud cake formation.

[0055] In this embodiment, the clay content, water content, and mineral composition data of each sample formation parameter in the self-calibrated sample set are used as input features, and the actual mud cake control efficacy value and actual mud performance maintenance efficacy value of the corresponding sample are used as output labels to train a local efficacy prediction model. The specific process is as follows: the clay content, water content, montmorillonite ratio, illite ratio, and kaolinite ratio of each sample in the self-calibrated sample set are used as input features, and the mud cake control efficacy value and mud performance maintenance efficacy value of the corresponding sample are used as two independent output labels; the support vector regression algorithm is adopted, with the radial basis function as the kernel function and the penalty coefficient set to 1.0, to train a first support vector regression model with mud cake control efficacy value as the output label and a second support vector regression model with mud performance maintenance efficacy value as the output label. The first support vector regression model and the second support vector regression model together constitute the local efficacy prediction model in this embodiment.

[0056] In this embodiment, the temporary local performance prediction model refers to a performance prediction model trained using a self-calibration sample set during the performance evaluation framework self-calibration process, which is only applicable to the current formation conditions. "Temporary" means that this model is only trained and used under conditions of formation heterogeneity and abrupt changes. When formation heterogeneity and abrupt changes are detected again, a new self-calibration sample set will be collected and a new local performance prediction model will be trained, with the old model being overwritten by the new model.

[0057] In this embodiment, the latest formation parameters are input into the local performance prediction model to obtain estimated mud cake control effectiveness and estimated mud performance maintenance effectiveness, reflecting the current formation characteristics. This involves using the clay content, water content, montmorillonite percentage, illite percentage, and kaolinite percentage of the current tunneling ring provided by the multi-source parameter acquisition module as input, and feeding them into the first and second support vector regression models, respectively. The first support vector regression model outputs the estimated mud cake control effectiveness, and the second support vector regression model outputs the estimated mud performance maintenance effectiveness. These estimated effectiveness values ​​comprehensively reflect the potential impact of the current formation's clay mineral type and content on both mud cake control and mud performance maintenance effectiveness.

[0058] In this embodiment, recalculating the coordinates of the optimal working point in the conflict assessment module based on the estimated mud cake prevention effectiveness and the estimated mud performance maintenance effectiveness means using the estimated mud cake prevention effectiveness as the second coordinate axis and the estimated mud performance maintenance effectiveness as the first coordinate axis to mark a new optimal working point in the phase plane. The specific process of gradually migrating the original optimal working point coordinates to the new coordinates on the phase plane is as follows: setting the migration step size to 0.05, the optimal working point's horizontal coordinate value is moved one step towards the new horizontal coordinate value and its vertical coordinate value is moved one step towards the new vertical coordinate value every one tunneling ring. This process is repeated over multiple tunneling rings to gradually migrate the optimal working point from its original coordinate position to the new coordinate position. During the migration, the conflict assessment module uses the gradually changing optimal working point coordinates to determine the type and degree of performance imbalance. This gradual migration avoids the parameter shocks caused by sudden changes in the optimal working point to the collaborative control module, ensuring a smooth and stable control process. After completing the optimal working point coordinate migration, the self-correction process of the performance assessment framework ends.

[0059] During the optimal operating point migration, the state quantization module continues to use the pre-trained random forest regression model to calculate the mud cake control efficacy value, ensuring that the phase trajectory point data used by the conflict assessment module and the efficacy value data used by the collaborative control module come from the same efficacy calculation model. The estimated efficacy value output by the local efficacy prediction model is only used to calculate the target coordinates of the new optimal operating point and does not participate in the real-time efficacy value calculation during the migration. After the optimal operating point migration is completed, the local efficacy prediction model is incorporated into the state quantization module as a new auxiliary prediction model and runs in parallel with the original random forest regression model.

[0060] Furthermore, the state quantization module also incorporates a built-in apparent viscosity-true performance asymmetric response compensation unit, which includes: The sequence acquisition subunit is used to continuously acquire the mud viscosity sequence and the corresponding mud cake control efficacy value sequence within a preset observation window after the dispersant is injected; The apparent normalization subunit is used to normalize the mud viscosity sequence to obtain the apparent dispersion sequence. The true normalization subunit is used to normalize the mud cake control efficacy value sequence to obtain the true efficacy response sequence. The cross-correlation calculation subunit is used to calculate the dynamic cross-correlation function between the apparent dispersion sequence and the true performance response sequence, and to determine the lag time of the true performance response sequence relative to the apparent dispersion sequence. The gain coefficient calculation subunit is used to calculate the ratio of the instantaneous rate of change of the apparent dispersion sequence to the instantaneous rate of change of the actual performance response sequence, and to obtain the asymmetric response gain coefficient. The asymmetric compensation subunit is used to determine that the current response mode is asymmetric when the asymmetric response gain coefficient is lower than the preset lower threshold or exceeds the preset upper threshold. It multiplies the asymmetric response gain coefficient with the dispersant injection amount control parameter calculated by the collaborative control module to generate the compensated dispersant injection amount control parameter, and feeds the compensated dispersant injection amount control parameter back to the collaborative control module as the final value of the dispersant injection amount control parameter.

[0061] In this embodiment, the preset observation window refers to a monitoring period of a preset duration, starting from the moment the dispersant injection execution unit completes one dispersant injection action. The duration of the preset observation window is set to the tunneling time of three tunneling loops, that is, spanning three complete tunneling loops from the moment the injection is completed. The basis for setting three tunneling loops is that the sufficient mixing and effect manifestation of the dispersant in the slurry circulation system requires approximately two tunneling loop cycles, and reserving a margin of one tunneling loop ensures that the complete performance response process is captured.

[0062] In this embodiment, within a preset observation window after the dispersant is injected, a mud viscosity sequence and a corresponding mud cake control efficacy value sequence are continuously collected. This means that the apparent viscosity-true efficacy asymmetric response compensation unit obtains the mud viscosity value at each sampling time within the preset observation window from the multi-source parameter acquisition module, and arranges them in chronological order to form a mud viscosity sequence; simultaneously, it obtains the mud cake control efficacy value corresponding to each sampling time within the preset observation window from the state quantization module, and arranges them in the same chronological order to form a mud cake control efficacy value sequence. The sampling time interval is set to ten times per tunneling ring, the two sequences are of the same length, and the mud viscosity value and mud cake control efficacy value at each position in the sequence correspond to the same sampling time.

[0063] In this embodiment, normalizing the mud viscosity sequence to obtain the apparent dispersion sequence involves subtracting the minimum mud viscosity value from each mud viscosity value in the sequence and then dividing by the difference between the maximum and minimum mud viscosity values. This converts the mud viscosity sequence into an apparent dispersion sequence with values ​​ranging from zero to one. The closer the value in the apparent dispersion sequence is to one, the greater the decrease in mud viscosity, indicating a more significant apparent dispersion effect observed from the perspective of mud rheological changes.

[0064] In this embodiment, normalizing the mud cake control efficacy value sequence to obtain the true efficacy response sequence involves subtracting the minimum mud cake control efficacy value from each mud cake control efficacy value in the sequence, and then dividing by the difference between the maximum and minimum mud cake control efficacy values. This transforms the mud cake control efficacy value sequence into a true efficacy response sequence with values ​​ranging from zero to one. The closer the values ​​in the true efficacy response sequence are to one, the greater the improvement in mud cake control efficacy, meaning a more significant true dispersion effect assessed from the perspective of mud cake formation risk suppression.

[0065] In this embodiment, calculating the dynamic cross-correlation function between the apparent dispersion sequence and the true performance response sequence involves fixing the apparent dispersion sequence and gradually shifting the true performance response sequence along the time axis. Each time the sequence is shifted by a sampling interval, the Pearson correlation coefficient between the apparent dispersion sequence and the shifted true performance response sequence is calculated, resulting in a set of correlation coefficients that change with the number of shift steps. A positive shift step indicates that the true performance response sequence has shifted in the direction of time lag, while a negative shift step indicates that the true performance response sequence has shifted in the direction of time lead.

[0066] In this embodiment, determining the lag time of the true effectiveness response sequence relative to the apparent dispersion sequence refers to finding the number of shift steps corresponding to the maximum correlation coefficient in the dynamic cross-correlation function calculation results. When the maximum correlation coefficient appears at a position where the true effectiveness response sequence is shifted several steps in the time lag direction, the number of shift steps multiplied by the sampling time interval is the lag time. The lag time reflects the time difference between the apparent decrease in mud viscosity after dispersant injection and the actual improvement in mud cake control effectiveness. For example, if the maximum correlation coefficient appears at a position three sampling intervals behind the true effectiveness response sequence, and the sampling time interval is one-tenth of a tunneling loop time, then the lag time is three-tenths of a tunneling loop time.

[0067] In this embodiment, the asymmetric response gain coefficient is obtained by calculating the ratio of the instantaneous rate of change of the apparent dispersion sequence to the instantaneous rate of change of the actual effectiveness response sequence. This involves taking the difference between two adjacent values ​​in the apparent dispersion sequence divided by the sampling time interval as the apparent instantaneous rate of change, and taking the difference between two corresponding adjacent values ​​in the actual effectiveness response sequence divided by the sampling time interval as the actual instantaneous rate of change. Dividing the apparent instantaneous rate of change by the actual instantaneous rate of change yields the asymmetric response ratio for each sampling interval. The average of all asymmetric response ratios is taken as the asymmetric response gain coefficient. An asymmetric response gain coefficient greater than one indicates that the rate of change of the apparent dispersion is faster than the rate of change of the actual effectiveness response, and the dispersant's effect on reducing mud viscosity is numerically more significant than its actual contribution to mud cake control. An asymmetric response gain coefficient less than one indicates that the rate of change of the actual effectiveness response is faster than the rate of change of the apparent dispersion.

[0068] In this embodiment, the preset lower threshold is a boundary value used to determine whether the asymmetric response gain coefficient is too low, resulting in insufficient compensation for the actual efficacy of mud cake control. The preset lower threshold is set to 0.8. When the asymmetric response gain coefficient is lower than 0.8, it indicates that the rate of change of the actual efficacy response is significantly faster than the rate of change of the apparent dispersion degree. Indirectly judging the effect of the dispersant based on the change of mud viscosity will underestimate the actual improvement in mud cake control efficacy, and it is necessary to amplify and compensate the dispersant injection amount control parameter calculated by the collaborative control module.

[0069] In this embodiment, the preset upper limit threshold is a boundary value used to determine whether the asymmetric response gain coefficient is too high, leading to overcompensation of the actual efficacy of mud cake control. The preset upper limit threshold is set to 1.2. When the asymmetric response gain coefficient exceeds 1.2, it indicates that the rate of change of the apparent dispersion degree is significantly faster than the rate of change of the actual efficacy response. Indirectly judging the effect of the dispersant based on the change of mud viscosity will overestimate the actual improvement in mud cake control efficacy, and it is necessary to suppress and compensate the dispersant injection amount control parameter calculated by the collaborative control module.

[0070] In this embodiment, the current response mode is asymmetric, meaning the asymmetric response gain coefficient is below a preset lower threshold of 0.8 or above a preset upper threshold of 1.2. The asymmetric mode indicates a significant mismatch between the changing trend of mud viscosity and the actual changing trend of mud cake control efficacy; the change in mud viscosity cannot be directly used to equivalently represent the actual contribution of the dispersant to mud cake control efficacy. When the asymmetric response gain coefficient is between 0.8 and 1.2, the response mode is symmetric, and the apparent dispersion degree is basically synchronized with the actual efficacy response, requiring no asymmetric compensation.

[0071] In this embodiment, the final value of the dispersant injection amount control parameter refers to the dispersant injection amount control parameter value that, after being compensated and corrected by the apparent viscosity-true performance asymmetric response compensation unit, is finally sent to the injection execution control module to control the action of the dispersant injection execution unit. The dispersant injection amount control parameter calculated by the collaborative control module is used as the initial value. This initial value is multiplied by the asymmetric response gain coefficient determined according to the current response mode to obtain the compensated dispersant injection amount control parameter. This compensated dispersant injection amount control parameter is used as the final value, updated by the collaborative control module and output to the injection execution control module. The injection execution control module controls the opening of the electric proportional valve according to the final value of the dispersant injection amount control parameter to complete the dispersant injection.

[0072] Furthermore, the coordinated control module also incorporates a built-in hydraulic transport lag effect prediction and compensation unit, which includes: The pipeline hysteresis calculation subunit is used to obtain the real-time mud flow velocity, length and cross-sectional area of ​​the slurry discharge pipeline in the shield tunnel circulation system, and to calculate the pipeline transport hysteresis time of mud flowing from the mud-water chamber to the outlet of the slurry discharge pipe. The mud-water tank replacement calculation subunit is used to divide the mud-water tank volume by the current mud flow rate to calculate the time for complete mud replacement in the mud-water tank. The total hysteresis calculation subunit is used to add the pipeline transport hysteresis time to the time of complete replacement of the slurry tank to obtain the total hydraulic hysteresis time of the circulation system. The time alignment subunit is used to obtain the mud cake control efficiency value currently calculated by the state quantization module, and to align the current mud cake control efficiency value with the historical mud cake control efficiency value before the total hydraulic delay time to obtain the mud cake control efficiency value before correction and the mud cake control efficiency value after correction. The trend prediction subunit is used to predict the trend of mud cake control effectiveness change reflected by the mud that is about to reach the outlet of the discharge pipe within the hysteresis time window, based on the mud cake control effectiveness value before correction and the mud cake control effectiveness value after correction. The hysteresis compensation solution subunit is used to input the trend of mud cake control efficacy changes into the control parameter solution process of the collaborative control module, and solve the dispersant injection amount control parameter and propulsion speed control parameter after hysteresis effect prediction compensation.

[0073] In this embodiment, the shield circulation system refers to the closed pipeline system used by the slurry shield machine to maintain the circulating flow of slurry. It consists of an inlet pipeline, a slurry chamber, a discharge pipeline, and a surface slurry treatment station connected end to end. The inlet pipeline transports fresh slurry treated at the surface slurry treatment station to the slurry chamber. In the slurry chamber, the slurry mixes with the excavated soil cut by the cutterhead and is then transported back to the surface slurry treatment station through the discharge pipeline. The surface slurry treatment station separates the excavated soil and sends the slurry back into the inlet pipeline, forming a continuous circulation of slurry. The main functions of the shield circulation system are to carry and transport excavated soil, balance the water and soil pressure at the tunnel face, and cool the cutterhead.

[0074] In this embodiment, the slurry discharge pipeline refers to the slurry transport pipeline in the shield tunneling circulation system that connects the outlet of the slurry chamber to the inlet of the surface slurry treatment station. The diameter and length of the slurry discharge pipeline depend on the specifications of the shield machine and the tunnel excavation distance. The inner wall of the slurry discharge pipeline is equipped with a wear-resistant lining to resist long-term erosion by slag-containing slurry.

[0075] In this embodiment, the real-time mud flow velocity of the slurry discharge pipeline in the shield tunnel circulation system is obtained from the multi-source parameter acquisition module. This refers to reading the current flow velocity of the mud in the slurry discharge pipeline from the electromagnetic flowmeter installed on the pipeline, expressed in meters per second. The length of the slurry discharge pipeline refers to the total length of the pipeline from the outlet of the mud-water chamber to the inlet of the surface mud-water treatment station. The cross-sectional area of ​​the slurry discharge pipeline refers to the circular cross-sectional area of ​​the pipeline. The length and cross-sectional area of ​​the slurry discharge pipeline are stored as preset system parameters in the parameter configuration table of the collaborative control module, and are obtained and entered from the design drawings and construction records of the shield tunnel circulation system during system initialization.

[0076] In this embodiment, the slurry chamber refers to a sealed pressure chamber located behind the cutterhead and in front of the air pressure regulation system of the slurry shield machine. It is filled with slurry and connected to the slurry inlet and outlet pipes. The slurry chamber balances the water and soil pressure at the tunnel face through slurry pressure, and simultaneously accommodates the mixture of excavated soil and slurry cut by the cutterhead. It is the initial location where the dispersant interacts with the clay-containing slurry, and also the location of the slurry chamber injection point.

[0077] In this embodiment, calculating the pipeline transport delay time from the mud tank outlet to the discharge pipe outlet refers to dividing the length of the discharge pipe by the real-time mud flow velocity in the discharge pipe to obtain the time elapsed from when the mud enters the discharge pipe from the mud tank outlet to when it reaches the inlet of the surface mud treatment station. The pipeline transport delay time depends on the length of the discharge pipe and the current mud flow velocity; the longer the discharge pipe and the lower the mud flow velocity, the longer the pipeline transport delay time.

[0078] In this embodiment, the slurry tank volume refers to the total volume of slurry that the slurry tank can hold, measured in cubic meters. The slurry tank volume is stored as a preset parameter in the parameter configuration table of the collaborative control module and is entered during system initialization based on the design parameters of the slurry tunnel boring machine.

[0079] In this embodiment, the current mud flow rate refers to the volume of mud passing through the discharge pipe per unit time, expressed in cubic meters per second. The current mud flow rate is equal to the cross-sectional area of ​​the discharge pipe multiplied by the real-time mud flow velocity of the discharge pipe.

[0080] In this embodiment, the complete mud replacement time in the mud tank refers to the value obtained by dividing the volume of the mud tank by the current mud flow rate. Physically, the complete mud replacement time is the time required for all the existing mud in the mud tank to be completely replaced by fresh mud entering from the inlet pipe. The larger the volume of the mud tank or the smaller the mud flow rate, the longer the complete mud replacement time.

[0081] In this embodiment, the total hydraulic lag time of the circulation system refers to the sum of the pipeline transport lag time and the time for complete displacement of the slurry tank. The physical meaning of the total hydraulic lag time is the total time required for the dispersant-treated slurry to be transported along the discharge pipe outlet after being mixed and displaced in the slurry tank following the injection of the dispersant from the slurry tank injection point. The total hydraulic lag time reflects the time delay between the dispersant injection action and the perception of its effect at the discharge pipe outlet.

[0082] In this embodiment, the pre-correction mud cake control efficiency value refers to the mud cake control efficiency value calculated and output by the state quantification module at the current moment, while the corrected mud cake control efficiency value refers to the mud cake control efficiency value calculated and output by the state quantification module at a historical moment before the total hydraulic lag time. Time-aligning the current mud cake control efficiency value with the historical mud cake control efficiency value before the total hydraulic lag time means establishing a correspondence between the pre-correction efficiency value and the corrected efficiency value. The corrected efficiency value reflects the actual efficiency level detectable at the slurry discharge pipe outlet after the total hydraulic lag time, while the pre-correction efficiency value reflects the latest efficiency level in the mud-water chamber at the current moment. The difference between the two reflects the information deviation caused by hydraulic lag.

[0083] In this embodiment, the hysteresis time window refers to the time interval extending forward from the current moment to the total hydraulic hysteresis time. Within the hysteresis time window, the mud in the mud tank will gradually reach the outlet of the discharge pipe.

[0084] In this embodiment, the prediction of the trend of mud cake control effectiveness change reflected by the mud reaching the discharge pipe outlet within the lag time window is based on the mud cake control effectiveness value before and after correction. This means dividing the difference between the mud cake control effectiveness value before and after correction by the total hydraulic lag time to obtain the average rate of change of the mud cake control effectiveness value from the current time to the time before the total hydraulic lag time. The predicted mud cake control effectiveness value is obtained by adding the product of the average rate of change and the total hydraulic lag time to the mud cake control effectiveness value before correction. The predicted mud cake control effectiveness value reflects the effectiveness level that will be detected when the mud in the mud-water tank reaches the discharge pipe outlet after the total hydraulic lag time. The trend of mud cake control effectiveness change is jointly characterized by the mud cake control effectiveness value before correction and the predicted mud cake control effectiveness value.

[0085] In this embodiment, the process of inputting the trend of mud cake control efficiency changes into the control parameter calculation process of the collaborative control module refers to using the uncorrected mud cake control efficiency value and the predicted mud cake control efficiency value as additional features, along with the efficiency imbalance type, imbalance degree, current mud viscosity, current cutterhead torque, and current clay content, as input into the gradient boosting tree model to solve for the dispersant injection amount control parameters and propulsion speed control parameters after hysteresis effect prediction compensation. The control parameters after hysteresis effect prediction compensation can respond in advance to impending efficiency changes when the hydraulic hysteresis effect causes a lag in the detection information at the slurry discharge pipe outlet, avoiding delays in control timing due to information lag.

[0086] Furthermore, when the collaborative control module determines that the type of performance imbalance is insufficient mud cake control performance, it calculates the dispersant injection rate control parameters and the propulsion speed control parameters according to the following rules: The dispersant injection amount control parameter is taken as a positive increment, which is the product of the degree of imbalance and the baseline dispersant injection amount; The propulsion speed control parameter is set to a negative reduction, which is the product of the degree of imbalance and the reference propulsion speed. When the imbalance is determined to be due to insufficient mud performance maintenance efficiency, the dispersant injection amount control parameter is set to a negative decrease, which is the product of the degree of imbalance and the baseline dispersant injection amount. The propulsion speed control parameter is set to a positive increase, which is the product of the degree of imbalance and the baseline propulsion speed.

[0087] In this embodiment, the imbalance type is determined to be insufficient mud cake prevention effectiveness. This means that the conflict evaluation module determines the current state as an imbalance where the mud cake prevention effectiveness is inferior to the mud performance maintenance effectiveness based on the quadrant and deviation direction of the phase trajectory point in the phase plane. When the phase trajectory point is located in the first quadrant and its ordinate value is less than the ordinate value of the optimal working point, but its abscissa value is greater than the abscissa value of the optimal working point, it indicates that the mud performance maintenance effectiveness is too high while the mud cake prevention effectiveness is too low. The main risk currently faced by the tunnel boring machine is mud cake formation rather than mud performance deterioration, therefore it is determined to be an imbalance of insufficient mud cake prevention effectiveness.

[0088] In this embodiment, the dispersant injection rate control parameter is set to a positive increment. This means that in the case of an imbalance due to insufficient mud cake control efficacy, the collaborative control module increases the dispersant injection rate by a positive value based on the current value. The positive increment is equal to the imbalance level multiplied by the baseline dispersant injection rate. The physical meaning of the positive increment is to enhance the dispersion effect on clay particles by increasing the dispersant injection rate, reduce the risk of mud cake formation, and bring the mud cake control efficacy value back towards the optimal operating point.

[0089] In this embodiment, the dispersant baseline injection rate refers to a reference value for the amount of dispersant injected per unit time, pre-set according to the tunnel boring machine specifications and the flow rate of the slurry circulation system. The dispersant baseline injection rate is set to 0.05% of the total flow rate of the tunnel boring machine's slurry circulation system. When the total flow rate of the slurry circulation system is 1,200 cubic meters per hour, the dispersant baseline injection rate is 0.6 cubic meters per hour. The dispersant baseline injection rate is stored as a system preset parameter of the collaborative control module.

[0090] In this embodiment, the propulsion speed control parameter is set to a negative reduction. This means that in the case of an imbalance caused by insufficient mud cake control effectiveness, the collaborative control module reduces the propulsion speed by a certain value from the current value. The negative reduction value is equal to the imbalance degree value multiplied by the baseline propulsion speed. The physical meaning of the negative reduction is to reduce the volume of clay cut by the cutterhead per unit time by reducing the tunnel boring machine's propulsion speed, thereby prolonging the residence time of clay particles in the slurry chamber. This allows the limited amount of dispersant to have more time to act on the surface of the clay particles, assisting in the recovery of mud cake control effectiveness.

[0091] In this embodiment, the reference advance speed refers to a pre-set reference value for the advance speed based on the tunnel boring machine's design parameters and geological conditions. The reference advance speed is set to 60% of the tunnel boring machine's maximum design advance speed. When the maximum design advance speed of the tunnel boring machine is 50 millimeters per minute, the reference advance speed is 30 millimeters per minute. The reference advance speed is stored as a system preset parameter of the collaborative control module.

[0092] In this embodiment, the imbalance is determined to be of insufficient mud performance maintenance efficiency. This means that the conflict evaluation module determines the current state as an imbalance where the mud performance maintenance efficiency is inferior to the mud cake prevention efficiency based on the quadrant and deviation direction of the phase trajectory point in the phase plane. When the phase trajectory point is located in the third quadrant and its abscissa value is less than that of the optimal working point, while its ordinate value is greater than that of the optimal working point, it indicates that the mud cake prevention efficiency is too high and the mud performance maintenance efficiency is too low. The main risk currently faced by the tunnel boring machine is insufficient mud slag carrying capacity or support capacity rather than mud cake formation. Therefore, it is determined to be an imbalance of insufficient mud performance maintenance efficiency.

[0093] In this embodiment, the dispersant injection rate control parameter is set to a negative reduction. This means that in the case of an imbalance due to insufficient mud performance maintenance efficiency, the collaborative control module reduces the dispersant injection rate by a certain value from the current value. The negative reduction value is equal to the imbalance degree value multiplied by the baseline dispersant injection rate. The physical meaning of the negative reduction is to reduce the continuous weakening effect of the dispersant on the mud rheology by reducing the dispersant injection rate, so that the mud viscosity and dynamic shear force can be restored to a range that is conducive to cuttings carrying and support, and the mud performance maintenance efficiency value can be restored towards the optimal operating point.

[0094] In this embodiment, the propulsion speed control parameter is set to a positive increment. This means that in the event of an imbalance due to insufficient mud performance maintenance efficiency, the collaborative control module increases the propulsion speed by a positive value based on the current value. The positive increment is equal to the imbalance level multiplied by the baseline propulsion speed. The physical meaning of the positive increment is to increase the flow rate of mud through the discharge pipe per unit time by increasing the tunnel boring machine's propulsion speed, thereby accelerating the mud circulation speed and indirectly assisting in the recovery of the mud performance maintenance efficiency. The extent of the increase in propulsion speed is constrained by the formation stability and cannot exceed the critical propulsion speed corresponding to the current formation clay content.

[0095] Furthermore, the optimal operating point in the conflict assessment module is determined as follows: From the historical tunneling database, tunneling rings that simultaneously meet the preset excellent standards for mud performance maintenance efficiency and mud cake prevention efficiency are selected, and the coordinates of the phase trajectory points corresponding to the tunneling rings are used as candidate optimal points. For each candidate optimal point, calculate the number of candidate optimal points contained within a preset neighborhood radius centered on each candidate optimal point, and use this as the local density of each candidate optimal point; Sort all candidate optima in descending order of local density, and select the candidate optima with the largest local density as the optimal operating point.

[0096] In this embodiment, the historical tunneling database refers to a database system used to store and manage construction data for all tunneling rings completed by the tunnel boring machine (TBM). Each record in the historical tunneling database corresponds to a tunneling ring, and the record content includes the tunneling ring number, the start and end times of the tunneling ring, the average value of the mud rheological parameters of the tunneling ring, the average value of the TBM tunneling parameters, the stratum parameters, the mud performance maintenance efficiency value and mud cake prevention efficiency value calculated by the state quantification module, the phase trajectory point coordinates recorded by the conflict evaluation module, the presence of mud cake tags, and the presence of excavation face instability tags. The mud cake tag is determined by the results of mud cake fragment monitoring at the cutterhead opening inspection or the slurry pipe outlet, and the excavation face instability tag is determined by the pressure fluctuation of the slurry chamber and the surface settlement monitoring data. The historical tunneling database is continuously updated during the TBM tunneling process, with a new record added after each tunneling ring is completed.

[0097] In this embodiment, the preset excellent standard refers to the judgment condition used to determine whether the mud performance maintenance efficiency value and mud cake prevention efficiency value of a tunneling ring are simultaneously within the ideal range. The preset excellent standard is set as a mud performance maintenance efficiency value not lower than 0.7 and a mud cake prevention efficiency value not lower than 0.7. The threshold of 0.7 is determined based on shield tunneling construction experience: when the mud performance maintenance efficiency value is lower than 0.7, the mud's slag carrying capacity or support capacity begins to show an observable downward trend; when the mud cake prevention efficiency value is lower than 0.7, the cutterhead torque begins to show an abnormal upward trend caused by mud cake. A tunneling ring with both efficiency values ​​not lower than 0.7 indicates that the tunneling ring is in a healthy state in terms of both mud performance maintenance and mud cake prevention.

[0098] In this embodiment, selecting tunneling loops from the historical tunneling database that simultaneously meet preset excellent standards for both mud performance maintenance efficiency and mud cake prevention efficiency refers to checking each tunneling loop record in the historical tunneling database one by one, selecting tunneling loops with a mud performance maintenance efficiency value of not less than 0.7 and a mud cake prevention efficiency value of not less than 0.7, extracting the phase trajectory point coordinates corresponding to these tunneling loops on the phase plane, and using all the extracted phase trajectory point coordinates as a set of candidate optimal points. Each point in the candidate optimal point set represents an excellent balance state that has actually been achieved in historical tunneling.

[0099] In this embodiment, the preset neighborhood radius refers to the radius of a circular region centered on each candidate optimum for calculating local density. The preset neighborhood radius is set to 0.15. It is calculated by sorting the Euclidean distances between the candidate point and all other candidate optima from smallest to largest, and taking the distance at the 10th percentile of the sorted distances as the neighborhood radius. A neighborhood radius of 0.15 ensures that a reasonable range is delineated on the phase plane that can distinguish local density. The more candidate optima contained within the neighborhood radius, the denser the phase plane region where the candidate point is located.

[0100] In this embodiment, for each candidate optimal point, the number of candidate optimal points contained within a preset neighborhood radius centered on each candidate optimal point is calculated. This means drawing a circular region on the phase plane with the coordinates of the currently calculated candidate point as the center and a radius of 0.15, and counting the number of candidate points falling within this circular region. This number is the local density of the currently calculated candidate point. The physical meaning of local density is the density of excellent state points clustered in the neighborhood centered on the candidate point on the phase plane. The higher the local density, the more historical excellent state points exist around the candidate point. The optimal working area represented by the candidate point appears repeatedly in multiple historically excellent tunneling loops, exhibiting good robustness and reproducibility.

[0101] In this embodiment, all candidate optimal points are sorted in descending order of local density, and the candidate optimal point with the highest local density is selected as the optimal operating point. This means that each candidate optimal point is sorted from largest to smallest local density, and the candidate optimal point with the highest local density at the top of the sorted list is determined as the optimal operating point currently used by the conflict evaluation module. The reason for selecting the candidate point with the highest local density as the optimal operating point is that the point with the highest local density is located in the center of the phase plane region where historical excellent state points are most densely concentrated. It represents not an excellent state under individual extreme conditions, but a typical excellent state that has been stably appearing in multiple constructions. Using it as the control target can make the convergence process of collaborative control more stable and reliable.

[0102] Furthermore, it also includes a model self-calibration function module, which includes: The deviation calculation subunit is used to collect the actual mud performance maintenance efficiency value and the actual mud cake prevention efficiency value after each adjustment, and to calculate the change in the actual efficiency value. The deviation comparison subunit is used to compare the actual change in efficiency value with the theoretical change in efficiency value and calculate the efficiency prediction deviation. The theoretical change in efficiency value is the expected change in efficiency value for this regulation determined by the model self-correction function module based on the historical change in efficiency value corresponding to samples with similar features to the current input in the training data of the state quantization module. The calibration trigger subunit is used to trigger parameter calibration of the state quantization module and the conflict evaluation module when the performance prediction deviation exceeds the preset deviation threshold in a series of preset adjustments.

[0103] In this embodiment, the actual mud performance maintenance efficiency value and the actual mud cake prevention efficiency value are collected after each adjustment. The change in actual efficiency value is calculated as follows: After the injection execution control module completes one dispersant injection and propulsion speed adjustment based on the final value of the dispersant injection amount adjustment parameter calculated by the collaborative control module and the propulsion speed adjustment parameter, the model self-correction function module obtains the adjusted mud performance maintenance efficiency value and mud cake prevention efficiency value from the state quantification module. The actual change in mud performance maintenance efficiency value is obtained by subtracting the mud performance maintenance efficiency value before adjustment from the adjusted mud performance maintenance efficiency value, and the actual change in mud cake prevention efficiency value is obtained by subtracting the mud cake prevention efficiency value before adjustment from the adjusted mud cake prevention efficiency value. The actual change in mud performance maintenance efficiency value and the actual change in mud cake prevention efficiency value together constitute the change in actual efficiency value.

[0104] In this embodiment, the estimated change in theoretical effectiveness value when the collaborative regulation module calculates the regulation parameters refers to the estimated change in the expected change of mud performance maintenance effectiveness and mud cake control effectiveness value after this regulation, output by the gradient boosting tree model in the collaborative regulation module, based on the historical effectiveness value changes corresponding to samples with similar characteristics to the current input in the training data, while outputting the dispersant injection amount regulation parameters and propulsion speed regulation parameters. The theoretical effectiveness value change includes the theoretical mud performance maintenance effectiveness change and the theoretical mud cake control effectiveness change, reflecting the gradient boosting tree model's expectation of the regulation effect based on historical experience.

[0105] In this embodiment, the efficiency prediction deviation is calculated by comparing the actual change in efficiency value with the theoretical change in efficiency value estimated by the collaborative control module when calculating the control parameters. Specifically, the absolute value of the difference between the actual change in mud performance maintenance efficiency and the theoretical change in mud performance maintenance efficiency is taken as the mud performance maintenance efficiency prediction deviation, and the absolute value of the difference between the actual change in mud cake control efficiency and the theoretical change in mud cake control efficiency is taken as the mud cake control efficiency prediction deviation. The larger of the two deviations is taken as the efficiency prediction deviation. The efficiency prediction deviation reflects whether there is a significant decrease in the gradient boosting tree model's ability to predict the control effect.

[0106] In this embodiment, the preset number of times refers to the threshold for the number of consecutive adjustments used to trigger parameter calibration. The preset number of times is set to five. The basis for setting five times is that the performance prediction deviation of a single or two adjustments may be caused by the special working conditions of individual tunneling loops. If the performance prediction deviation of five consecutive adjustments exceeds the preset deviation threshold, it indicates that the decline in the model's prediction capability is systematic, rather than caused by accidental factors.

[0107] In this embodiment, the preset deviation threshold is a boundary value used to determine whether the performance prediction deviation is significant enough to trigger parameter calibration. The preset deviation threshold is set to 0.15. When the performance prediction deviation exceeds 0.15, it indicates a significant difference between the performance change estimated by the gradient boosting tree model and the actual performance change measured by the state quantization module. This difference means that the model's prediction of the control effect has deviated from the true response pattern under the current formation and tunneling conditions.

[0108] In this embodiment, triggering parameter calibration for the state quantification module and the conflict assessment module means that when the performance prediction deviation exceeds 0.15 in five consecutive adjustments, the model self-calibration function module sends a parameter calibration command for the random forest regression model to the state quantification module and a recalculation command for the optimal operating point to the conflict assessment module. Triggering parameter calibration indicates that the system has detected that the coordinates of the random forest regression model in the state quantification module and the optimal operating point in the conflict assessment module are no longer suitable for the current construction environment due to changes in geological conditions or tunneling conditions, and parameter calibration is needed to restore the decision-making accuracy of the control system.

[0109] Furthermore, the parameter calibration process of the model self-calibration function module includes: Obtain the pre-control state point corresponding to the performance prediction deviation exceeding the preset deviation threshold, and use the mud rheological parameters, shield tunneling parameters and formation parameters of the corresponding pre-control state point as input features, and the actual performance value change as output label to construct calibration samples; Add the calibration samples to the training dataset of the state quantization module; With the goal of minimizing the prediction error of the efficiency value of the calibration sample by the state quantization module, the gradient descent method is used to incrementally update the efficiency calculation coefficients in the state quantization module.

[0110] In this embodiment, obtaining the pre-control state point corresponding to the performance prediction deviation exceeding a preset deviation threshold means that the model self-calibration function module retrieves the five pre-control execution times corresponding to the five consecutive controls with performance prediction deviations exceeding 0.15 from the control record log, and extracts the mud rheological parameters, shield tunneling parameters, and formation parameters of the state point recorded by the multi-source parameter acquisition module at each pre-control execution time. The parameter data of these five state points will be used to construct calibration samples to correct the performance calculation model in the state quantification module that is no longer suitable for the current formation conditions.

[0111] In this embodiment, the mud rheological parameters, shield tunneling parameters, and formation parameters of the corresponding pre-control state point are used as input features, and the actual change in performance value is used as the output label to construct a calibration sample. This means using the mud viscosity, mud density, and dynamic shear force from the mud rheological parameters of a pre-control state point, and the clay content, water content, and the proportions of montmorillonite, illite, and kaolinite from the formation parameters of the formation, as input features. The actual measured changes in mud performance maintenance efficiency and mud cake prevention efficiency after the control are used as output labels to construct a complete training sample that simultaneously contains input features and corresponding output labels. One calibration sample is constructed for each of the five pre-control state points, resulting in five calibration samples. The input feature dimensions of the calibration samples are completely consistent with the training feature dimensions used by the random forest regression model in the state quantification module during initial training. The output label of the calibration samples is the actual performance response observed under the latest formation conditions.

[0112] In this embodiment, the training dataset for the state quantization module refers to the set of samples used to train the random forest regression model in the state quantization module. During system initialization, the training dataset is constructed from all tunneling loop data with confirmed mud cake formation labels obtained from the historical tunneling database. The input features of each training sample include the penetration resistance index, propulsion resistance index, clay content, and montmorillonite percentage; the output label is the probability of mud cake formation. The training dataset is dynamically expanded during parameter calibration, with new calibration samples continuously added, allowing the random forest regression model to gradually adapt to changes in formation conditions.

[0113] In this embodiment, aiming to minimize the prediction error of the efficiency value of the state quantization module for the calibration samples, the gradient descent method is used to incrementally update the efficiency calculation coefficients in the state quantization module. This involves sequentially inputting the input features of the five calibration samples into the currently trained random forest regression model in the state quantization module to obtain the predicted probability of mud cake formation for each calibration sample. The mean squared error (MSE) between the predicted probability and the actual probability of mud cake formation in the calibration samples is then calculated, and this MSE is used as the prediction error. The output value of each leaf node in the random forest regression model is used as the efficiency calculation coefficient. The partial derivative of the MSE with respect to each efficiency calculation coefficient is calculated, and the efficiency calculation coefficient is slightly adjusted along the negative direction of the partial derivative, with an adjustment step size set to 0.01. This process of calculating the partial derivative and adjusting the coefficients is repeated multiple times until the MSE no longer decreases or the decrease is less than 0.001. Incremental update means only fine-tuning the existing coefficients of the model rather than retraining the model from scratch, ensuring that the model can quickly adapt to new geological conditions while retaining historical learning knowledge. After the gradient descent method update is completed, the random forest regression model in the state quantization module completes parameter calibration.

[0114] Furthermore, the conflict assessment module also has experience-learning capabilities, including: The experience storage subunit is used to obtain the state parameters at each time the control is executed and the actual performance compensation amount after the control is executed. The state parameters include the coordinates, deviation direction and deviation distance of the phase trajectory point before the control. The state parameters and the actual performance compensation amount are stored as a historical control experience record and associated with it in the control experience database. The experience retrieval subunit is used to retrieve historical control experience records in the control experience base after the conflict evaluation module determines the new type and degree of performance imbalance, using the coordinates, deviation direction and deviation distance of the current phase trajectory point as retrieval conditions. The similarity between the records and the retrieval conditions exceeds a preset similarity threshold. The regulation effect prediction subunit is used to predict the regulation effect of the current regulation based on the actual efficiency value compensation amount in the retrieved historical regulation experience records. The regulation correction subunit is used to correct the dispersant injection amount regulation parameters and propulsion speed regulation parameters calculated by the collaborative regulation module when the estimated regulation effect is lower than the preset minimum expected compensation amount.

[0115] In this embodiment, the coordinates of the control preceding trajectory point refer to the abscissa and ordinate values ​​of the current shield tunneling state point marked on the phase plane by the conflict evaluation module when determining the type and degree of performance imbalance. The abscissa value is the current mud performance maintenance efficiency value, and the ordinate value is the current mud cake prevention efficiency value. The deviation direction refers to the vector direction of the control preceding trajectory point relative to the optimal working point, with the optimal working point as the origin. The deviation distance refers to the Euclidean distance between the control preceding trajectory point and the optimal working point. The coordinates, deviation direction, and deviation distance of the control preceding trajectory point together constitute the three-dimensional state parameters describing the state at which the control triggering timing occurs.

[0116] In this embodiment, the actual efficiency value compensation after the adjustment refers to the change in mud performance maintenance efficiency value relative to before the adjustment, and the change in mud cake control efficiency value relative to before the adjustment, measured by the state quantification module at the end of the preset observation window after the injection execution control module completes one dispersant injection and propulsion speed adjustment based on the final value of the dispersant injection amount adjustment parameter calculated by the collaborative adjustment module and the propulsion speed adjustment parameter. The actual efficiency value compensation reflects the real improvement effect of this adjustment action on mud performance maintenance efficiency and mud cake control efficiency, and its unit is a dimensionless numerical value with the same efficiency value.

[0117] In this embodiment, the conflict assessment module determines the new type and degree of performance imbalance. This means that during the tunnel boring machine (TBM) excavation process, as the conflict assessment module continuously operates, after the multi-source parameter acquisition module updates the mud rheological parameters, TBM excavation parameters, and formation parameters after industrial data processing, the state quantification module recalculates the current mud performance maintenance efficiency value and mud cake prevention efficiency value. The conflict assessment module maps the new efficiency values ​​to the phase plane, constructs new phase trajectory points, calculates the deviation direction and distance between the new phase trajectory points and the current optimal working point, and determines the new type and degree of performance imbalance based on the deviation direction and distance. Determining the new type and degree of performance imbalance is a prerequisite for initiating the step of retrieving similar historical control experiences in the experience learning function.

[0118] In this embodiment, the similarity with the search criteria refers to the weighted average of three factors: the Euclidean distance between the coordinates of the pre-control phase trajectory point and the coordinates of the current phase trajectory point in each historical control experience record in the control experience database; the cosine value of the angle between the deviation directions; and the absolute value of the difference between the deviation distances. The weighted average is assigned as follows: Euclidean distance weight 0.4, cosine value weight 0.3, and deviation distance difference weight 0.3. The similarity value ranges from zero to one; the closer the value is to one, the more similar the historical control experience record's scenario is to the current scenario to be controlled.

[0119] In this embodiment, the preset similarity threshold refers to the lowest similarity limit value used to filter historical control experience records that have sufficient similarity to the current control scenario. The preset similarity threshold is set to 0.7. Historical control experience records with a similarity lower than 0.7 are considered to be significantly different from the current scenario, and their historical control experience has limited reference value for the current control, and is not included in the prediction calculation of subsequent control effects.

[0120] In this embodiment, retrieving historical control experience records in the control experience base whose similarity to the search criteria exceeds a preset similarity threshold means using the coordinates, deviation direction, and deviation distance of the current phase trajectory point as search criteria. This involves traversing all historical control experience records stored in the control experience base, calculating the similarity between each historical control experience record and the search criteria, filtering out all historical control experience records with a similarity greater than 0.7, sorting the filtered records from highest to lowest similarity, and selecting the top ten records as the current control's historical reference record set. If there are fewer than ten records in the control experience base that meet the criteria, all records that meet the criteria are used. If there are no historical control experience records in the control experience base whose similarity exceeds the preset similarity threshold, the correction step of the experience learning function is skipped, and the dispersant injection amount control parameters and propulsion speed control parameters calculated by the collaborative control module are directly used as the final values.

[0121] In this embodiment, when the number of retrieved historical control experience records is not zero, the control effect of the current control is estimated based on the actual efficiency compensation amount in the retrieved historical control experience records. This involves multiplying the actual mud performance maintenance efficiency compensation amount and the actual mud cake prevention efficiency compensation amount of each record in the historical reference record set by the similarity of that record, summing the results, and then dividing by the sum of the similarities of all records in the historical reference record set. The estimated mud performance maintenance efficiency compensation amount and the estimated mud cake prevention efficiency compensation amount are obtained from these two estimated compensation amounts. These two estimated compensation amounts together constitute the estimated control effect. The estimated control effect reflects the average efficiency improvement achieved by control actions similar to the current scenario in historical experience.

[0122] In this embodiment, the preset minimum expected compensation amount refers to the minimum efficiency compensation threshold value used to determine whether the estimated control effect is sufficient to compensate for the control cost. The preset minimum expected compensation amount is set to 0.05. When the estimated mud performance maintenance efficiency compensation amount or the estimated mud cake prevention efficiency compensation amount is both lower than 0.05, it indicates that even in the most similar historical scenario, the efficiency improvement in both aspects after implementing standard control actions is very limited. The expected benefits of this control may not be sufficient to cover the dispersant consumption cost and the impact of tunneling speed adjustment on the project schedule.

[0123] In this embodiment, the correction of the dispersant injection rate control parameters and propulsion speed control parameters calculated by the collaborative control module refers to the process where, when the estimated control effect is lower than the preset minimum expected compensation amount, the experience learning function multiplies the dispersant injection rate control parameters and propulsion speed control parameters output by the gradient boosting tree model in the collaborative control module by a correction coefficient. The correction coefficient is determined by the ratio of the actual performance value compensation to the theoretical performance value change of the historical record closest to the deviation direction of the current phase trajectory point in the historical reference record set. The corrected dispersant injection rate control parameters and propulsion speed control parameters replace the uncorrected parameters as the final values ​​of the dispersant injection rate control parameters. These final values ​​are further processed by the subsequent apparent viscosity-true performance asymmetric response compensation unit and hydraulic transport hysteresis effect prediction compensation unit before being executed by the injection execution control module.

[0124] This invention provides an embodiment of a mud dispersant injection decision platform based on multi-source parameter fusion, including a processor and a machine-readable storage medium. The machine-readable storage medium is connected to the processor and stores a computer program. The processor executes the computer program to achieve the functions of any of the above-mentioned mud dispersant injection decision systems based on multi-source parameter fusion.

[0125] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of this invention and its equivalents, this invention also intends to include these modifications and variations.

Claims

1. A mud dispersant injection decision system based on multi-source parameter fusion, characterized in that, include: The multi-source parameter acquisition module is used to acquire mud rheological parameters, shield tunneling parameters, and formation parameters during the shield tunneling process, and to perform industrial data processing on the acquired mud rheological parameters, shield tunneling parameters, and formation parameters. The state quantification module is used to calculate the mud performance maintenance efficiency value based on the mud rheological parameters and formation parameters after industrial data processing, and to calculate the mud cake prevention efficiency value based on the shield tunneling parameters and formation parameters after industrial data processing. The mud performance maintenance efficiency value reflects the degree of maintenance of the mud's slag carrying capacity and support capacity, and the mud cake prevention efficiency value reflects the degree of suppression of the risk of mud cake formation. The conflict evaluation module is used to map the mud performance maintenance efficiency value and mud cake prevention efficiency value to a phase plane with the mud performance maintenance efficiency value as the first coordinate axis and the mud cake prevention efficiency value as the second coordinate axis. It constructs phase trajectory points that characterize the current shield tunneling state and determines the type and degree of efficiency imbalance in the current state based on the deviation direction and deviation distance between the phase trajectory points and the optimal working point in the phase plane. The collaborative control module is used to calculate the dispersant injection amount control parameters and the propulsion speed control parameters based on the type and degree of performance imbalance using a machine learning model. The dispersant injection amount control parameters are used to adjust the amount of dispersant injected into the mud and water chamber injection point and the slurry discharge pipe injection point, and the propulsion speed control parameters are used to adjust the tunnel boring machine's propulsion speed. The injection execution control module is used to control the dispersant injection execution unit to add the corresponding amount of dispersant to the injection points of the mud and water chamber and the injection points of the slurry discharge pipe according to the dispersant injection amount control parameters, and to control the shield machine propulsion mechanism to adjust the propulsion speed according to the propulsion speed control parameters.

2. The mud dispersant injection decision system based on multi-source parameter fusion according to claim 1, characterized in that, The conflict assessment module also includes a built-in self-correcting unit for the performance evaluation framework, which comprises: The instability monitoring subunit is used to continuously collect the historical sequence of mud cake control efficacy value and the historical sequence of mud performance maintenance efficacy value within a specified historical window, and calculate the global instability index of the historical sequence of mud cake control efficacy value. The global instability index is the ratio of the variance of the historical sequence of mud cake control efficacy value to the mean of the historical sequence of mud cake control efficacy value. The self-calibration triggering subunit is used to determine that the current formation is in a heterogeneous abrupt state when the global instability index exceeds the preset instability threshold, and to trigger the self-calibration of the performance evaluation framework. The local model training subunit is used to collect mud parameters, shield parameters and formation parameters of multiple consecutive tunneling rings as a self-calibration sample set after triggering the self-calibration of the performance evaluation framework. The clay content, water content and mineral composition data of the formation parameters of each sample in the self-calibration sample set are used as input features, and the actual mud cake prevention efficiency value and actual mud performance maintenance efficiency value of the corresponding sample are used as output labels to train and obtain a local performance prediction model. The effectiveness prediction sub-unit is used to input the latest formation parameters into the local effectiveness prediction model to obtain the predicted mud cake control effectiveness value and the predicted mud performance maintenance effectiveness value that reflect the current formation characteristics. The migration sub-unit is used to recalculate the coordinates of the optimal working point in the conflict evaluation module based on the estimated mud cake prevention efficiency value and the estimated mud performance maintenance efficiency value, and to migrate the original optimal working point coordinates on the phase plane to the new coordinates in a gradual manner, thereby completing the self-correction of the efficiency evaluation framework.

3. The mud dispersant injection decision system based on multi-source parameter fusion according to claim 1, characterized in that, The state quantization module also incorporates a built-in apparent viscosity-true performance asymmetric response compensation unit, which includes: The sequence acquisition subunit is used to continuously acquire the mud viscosity sequence and the corresponding mud cake control efficacy value sequence within a preset observation window after the dispersant is injected; The apparent normalization subunit is used to normalize the mud viscosity sequence to obtain the apparent dispersion sequence. The true normalization subunit is used to normalize the mud cake control efficacy value sequence to obtain the true efficacy response sequence. The cross-correlation calculation subunit is used to calculate the dynamic cross-correlation function between the apparent dispersion sequence and the true performance response sequence, and to determine the lag time of the true performance response sequence relative to the apparent dispersion sequence. The gain coefficient calculation subunit is used to calculate the ratio of the instantaneous rate of change of the apparent dispersion sequence to the instantaneous rate of change of the actual performance response sequence, and to obtain the asymmetric response gain coefficient. The asymmetric compensation subunit is used to determine that the current response mode is asymmetric when the asymmetric response gain coefficient is lower than the preset lower threshold or exceeds the preset upper threshold. It multiplies the asymmetric response gain coefficient with the dispersant injection amount control parameter calculated by the collaborative control module to generate the compensated dispersant injection amount control parameter, and feeds the compensated dispersant injection amount control parameter back to the collaborative control module as the final value of the dispersant injection amount control parameter.

4. The mud dispersant injection decision system based on multi-source parameter fusion according to claim 1, characterized in that, The coordinated control module also has a built-in hydraulic transport lag effect prediction and compensation unit, which includes: The pipeline hysteresis calculation subunit is used to obtain the real-time mud flow velocity, length and cross-sectional area of ​​the slurry discharge pipeline in the shield tunnel circulation system, and to calculate the pipeline transport hysteresis time of mud flowing from the mud-water chamber to the outlet of the slurry discharge pipe. The mud-water tank replacement calculation subunit is used to divide the mud-water tank volume by the current mud flow rate to calculate the time for complete mud replacement in the mud-water tank. The total hysteresis calculation subunit is used to add the pipeline transport hysteresis time to the time of complete replacement of the slurry tank to obtain the total hydraulic hysteresis time of the circulation system. The time alignment subunit is used to obtain the mud cake control efficiency value currently calculated by the state quantization module, and to align the current mud cake control efficiency value with the historical mud cake control efficiency value before the total hydraulic delay time to obtain the mud cake control efficiency value before correction and the mud cake control efficiency value after correction. The trend prediction subunit is used to predict the trend of mud cake control effectiveness change reflected by the mud that is about to reach the outlet of the discharge pipe within the hysteresis time window, based on the mud cake control effectiveness value before correction and the mud cake control effectiveness value after correction. The hysteresis compensation solution subunit is used to input the trend of mud cake control efficacy changes into the control parameter solution process of the collaborative control module, and solve the dispersant injection amount control parameter and propulsion speed control parameter after hysteresis effect prediction compensation.

5. The mud dispersant injection decision system based on multi-source parameter fusion according to claim 1, characterized in that, When the collaborative control module determines that the type of performance imbalance is insufficient mud cake control performance, it calculates the dispersant injection rate control parameters and the propulsion speed control parameters according to the following rules: The dispersant injection amount control parameter is taken as a positive increment, which is the product of the degree of imbalance and the baseline dispersant injection amount; The propulsion speed control parameter is set to a negative reduction, which is the product of the degree of imbalance and the reference propulsion speed. When the imbalance is determined to be due to insufficient mud performance maintenance efficiency, the dispersant injection amount control parameter is set to a negative decrease, which is the product of the degree of imbalance and the baseline dispersant injection amount. The propulsion speed control parameter is set to a positive increase, which is the product of the degree of imbalance and the baseline propulsion speed.

6. The mud dispersant injection decision system based on multi-source parameter fusion according to claim 1, characterized in that, The optimal operating point is determined in the conflict evaluation module as follows: From the historical tunneling database, tunneling rings that simultaneously meet the preset excellent standards for mud performance maintenance efficiency and mud cake prevention efficiency are selected, and the coordinates of the phase trajectory points corresponding to the tunneling rings are used as candidate optimal points. For each candidate optimal point, calculate the number of candidate optimal points contained within a preset neighborhood radius centered on each candidate optimal point, and use this as the local density of each candidate optimal point; Sort all candidate optima in descending order of local density, and select the candidate optima with the largest local density as the optimal operating point.

7. The mud dispersant injection decision system based on multi-source parameter fusion according to claim 1, characterized in that, It also includes a model self-calibration function module, which includes: The deviation calculation subunit is used to collect the actual mud performance maintenance efficiency value and the actual mud cake prevention efficiency value after each adjustment, and to calculate the change in the actual efficiency value. The deviation comparison subunit is used to compare the actual change in efficiency value with the theoretical change in efficiency value and calculate the efficiency prediction deviation. The theoretical change in efficiency value is the expected change in efficiency value for this regulation determined by the model self-correction function module based on the historical change in efficiency value corresponding to samples with similar features to the current input in the training data of the state quantization module. The calibration trigger subunit is used to trigger parameter calibration of the state quantization module and the conflict evaluation module when the performance prediction deviation exceeds the preset deviation threshold in a series of preset adjustments.

8. The mud dispersant injection decision system based on multi-source parameter fusion according to claim 7, characterized in that, The parameter calibration process of the model self-calibration function module includes: Obtain the pre-control state point corresponding to the performance prediction deviation exceeding the preset deviation threshold, and use the mud rheological parameters, shield tunneling parameters and formation parameters of the corresponding pre-control state point as input features, and the actual performance value change as output label to construct calibration samples; Add the calibration samples to the training dataset of the state quantization module; With the goal of minimizing the prediction error of the efficiency value of the calibration sample by the state quantization module, the gradient descent method is used to incrementally update the efficiency calculation coefficients in the state quantization module.

9. The mud dispersant injection decision system based on multi-source parameter fusion according to claim 1, characterized in that, The conflict assessment module also has experience-learning capabilities, including: The experience storage subunit is used to obtain the state parameters at each time the control is executed and the actual performance compensation amount after the control is executed. The state parameters include the coordinates, deviation direction and deviation distance of the phase trajectory point before the control. The state parameters and the actual performance compensation amount are stored as a historical control experience record and associated with it in the control experience database. The experience retrieval subunit is used to retrieve historical control experience records in the control experience base after the conflict evaluation module determines the new type and degree of performance imbalance, using the coordinates, deviation direction and deviation distance of the current phase trajectory point as retrieval conditions. The similarity between the records and the retrieval conditions exceeds a preset similarity threshold. The regulation effect prediction subunit is used to predict the regulation effect of the current regulation based on the actual efficiency value compensation amount in the retrieved historical regulation experience records. The regulation correction subunit is used to correct the dispersant injection amount regulation parameters and propulsion speed regulation parameters calculated by the collaborative regulation module when the estimated regulation effect is lower than the preset minimum expected compensation amount.

10. A mud dispersant injection decision platform based on multi-source parameter fusion, characterized in that, It includes a processor and a machine-readable storage medium connected to the processor and storing a computer program. The processor executes the computer program to achieve the function of the mud dispersant injection decision system based on multi-source parameter fusion as described in any one of claims 1 to 9.

Images