A method for intelligent preparation and control of grouting fluid for backfilling of tunnel boring machines.

Abstract

This invention belongs to the field of shield tunnel construction technology, specifically relating to an intelligent preparation and control method for grouting fluid behind the shield wall, aiming to solve the problems of poor adaptability and unstable quality in existing grouting preparation technologies. This invention includes: constructing a grout performance prediction model and combining it with real-time acquired and pre-processed geological parameters; using a control algorithm to dynamically generate adjustment signals, thereby precisely adjusting the particle size distribution, moisture content, and admixture dosage of the semi-finished grout. Simultaneously, grout performance is continuously monitored during preparation and grouting, and the data is fed back to the model for parameter self-correction, forming a closed-loop intelligent control system. This invention achieves adaptive matching of grout performance with complex geological conditions, significantly improving grouting quality and construction efficiency.

Description

Technical Field

[0001] This invention belongs to the field of shield tunnel construction technology, and specifically relates to an intelligent preparation and control method for grouting fluid behind the shield wall. Background Technology

[0002] In the construction of shield tunnels, backfilling is one of the key procedures. Its purpose is to promptly fill the structural voids formed between the outer wall of the tunnel segments and the ground during tunnel boring machine (TBM) excavation, so as to effectively control ground settlement, stabilize the surrounding rock, and ensure the long-term safety of the tunnel structure.

[0003] Existing methods for preparing grout for shield tunneling mostly rely on pre-set empirical ratios or laboratory test results. This approach has significant drawbacks in actual construction: First, the geological conditions along the tunnel, such as stratum lithology, groundwater pressure and distribution, are dynamically changing, and fixed grout ratios are difficult to adapt to these changes, potentially leading to poor grouting results, incomplete filling, and insufficient or excessive strength of the grout aggregate. Second, traditional grout preparation processes are mostly open-loop controlled, lacking real-time monitoring and feedback adjustment mechanisms for the performance of the finished grout, making it difficult to guarantee the stability of grout quality, resulting in excessive manual intervention and low levels of automation and intelligence.

[0004] Therefore, how to adjust the slurry ratio and preparation process in real time and dynamically according to the complex and ever-changing geological environment of the construction site, and achieve precise control of slurry performance, is a technical problem that urgently needs to be solved in the field of tunnel boring machine construction. Summary of the Invention

[0005] To address the aforementioned problems in the prior art, namely poor adaptability and unstable quality of grout preparation, this invention provides an intelligent preparation and control method for shield tunnel wall backfill grout, comprising:

[0006] A grout performance model is established. The grout performance model predicts the finished product performance parameters of the grout in grouting construction by analyzing the physicochemical properties of the grout raw materials and the operational variables in the preparation process. The finished product performance parameters include the grout's fluidity, setting time, and strength development characteristics.

[0007] Real-time geological parameters of the construction site are collected, including stratum lithology, groundwater distribution and surrounding rock pressure status, and the geological parameters are preprocessed to eliminate noise and outliers.

[0008] Based on the preprocessed geological parameters and the predicted output of the slurry performance model, a control signal is generated based on the control algorithm.

[0009] Based on the control signal, the particle size distribution, moisture content and admixture dosage of the semi-finished slurry are dynamically adjusted.

[0010] During the grout preparation and grouting process, the real-time performance indicators of the grout are continuously monitored, and the monitoring data is fed back to the grout performance model to correct the parameters of the grout performance model.

[0011] Furthermore, the establishment of the slurry performance model includes:

[0012] Collect historical slurry preparation data, including physical property data of raw materials, chemical composition data, ratio of raw materials, operating parameters of stirring equipment, environmental temperature and humidity data, and corresponding performance test results of finished slurry;

[0013] The collected data is preprocessed, including data cleaning to remove outliers, data normalization to eliminate the influence of units, and data standardization.

[0014] A data-driven relational network is constructed based on the preprocessed data. The relational network is a neural network, which learns the nonlinear relationship between slurry properties and performance parameters by training historical data.

[0015] The neural network is trained using the backpropagation algorithm, and the network weights are optimized using the gradient descent method.

[0016] During the model deployment and application phase, new slurry preparation process data and performance test data are continuously integrated. The neural network parameters are dynamically adjusted through incremental learning, and the model prediction accuracy is verified periodically. When the prediction deviation exceeds the first preset threshold, model reconstruction is triggered.

[0017] Furthermore, the real-time geological parameters of the construction site collected include:

[0018] Acquire stratigraphic lithology data, groundwater distribution data, and surrounding rock pressure status data;

[0019] The geological parameters are preprocessed, including data cleaning to remove outliers, data normalization to eliminate the influence of dimensions, and data fusion to integrate multi-source sensor information into a unified set of geological parameters.

[0020] The preprocessed geological parameters are used as inputs to the slurry performance model and control algorithm.

[0021] Furthermore, generating control signals through the control algorithm includes:

[0022] Based on the pretreated geological parameters, determine the target values ​​of slurry performance that are adapted to the current geological conditions. The target performance values ​​include target fluidity, target setting time, and target strength.

[0023] A multivariate objective function is constructed with the goal of optimizing slurry performance. The objective function is based on a weighted combination of the deviations between the predicted and target values ​​of slurry fluidity, setting time, and strength development. The predicted values ​​are output by the slurry performance model.

[0024] Set constraints on process variables, including particle size distribution range, moisture content range, and admixture dosage range;

[0025] Within each control cycle, starting from the current process variable setpoints, an optimization problem with the objective of minimizing the multivariate objective function and subject to the constraints is solved to calculate a sequence of optimal future process variable setpoints.

[0026] The instruction for the current control cycle is selected from the optimal setpoint sequence, and a real-time adjustment signal is generated as the control signal.

[0027] Furthermore, the dynamic adjustment of particle size distribution, moisture content, and admixture dosage of the semi-finished slurry based on control signals includes:

[0028] The control signals are analyzed into particle size distribution adjustment signals, moisture content adjustment signals, and admixture dosage adjustment signals.

[0029] The particle size distribution of the slurry is adjusted based on the particle size distribution adjustment signal, the moisture content of the slurry is adjusted based on the moisture content adjustment signal, and the admixture dosage is adjusted based on the admixture dosage adjustment signal.

[0030] The actual values ​​of particle size distribution, moisture content and admixture dosage are monitored in real time and compared with the target values. When the deviation exceeds the allowable range, the control signal is recalculated and the adjustment operation is adjusted.

[0031] Furthermore, continuous monitoring of the real-time performance indicators of the slurry and feeding the monitoring data back to the slurry performance model includes:

[0032] Real-time data on the fluidity, setting time, and strength of the slurry are obtained through online monitoring.

[0033] The monitored real-time data is compared with the predicted values ​​of the slurry performance model, and the deviation is calculated.

[0034] Adjust the internal parameters of the slurry performance model according to the magnitude of the deviation;

[0035] Feedback data is used to calibrate geological parameters and control algorithms, forming an adaptive loop.

[0036] Furthermore, the method for adjusting the internal parameters of the slurry performance model is as follows:

[0037] Based on the aforementioned deviation, the weights and bias parameters of the neural network are dynamically adjusted through incremental learning.

[0038] The model prediction accuracy is verified periodically, and when the deviation exceeds a set threshold, the reconstruction process of the slurry performance model is triggered.

[0039] Furthermore, the control signals are analyzed into particle size distribution adjustment signals, moisture content adjustment signals, and admixture dosage adjustment signals, using the following method:

[0040] Read the target values ​​contained in the control signal that correspond to particle size distribution, moisture content and admixture dosage, respectively;

[0041] The target value of particle size distribution is compared with the currently detected actual value of particle size distribution, and a particle size distribution adjustment signal is generated based on the difference.

[0042] The target moisture content value is compared with the currently detected actual moisture content value, and a moisture content adjustment signal is generated based on the difference.

[0043] The target value of the admixture dosage is compared with the actual value of the currently detected admixture dosage, and an admixture dosage adjustment signal is generated based on the difference.

[0044] Furthermore, the particle size distribution of the slurry is adjusted based on the particle size distribution adjustment signal, the moisture content of the slurry is adjusted based on the moisture content adjustment signal, and the admixture dosage is adjusted based on the admixture dosage adjustment signal. The method is as follows:

[0045] The particle size distribution adjustment signal is converted into a first control command to control the feeding actuators of the coarse aggregate bin and the fine aggregate bin. By adjusting the feeding ratio and rate of coarse and fine aggregates, the particle size distribution of the slurry is controlled.

[0046] The moisture content adjustment signal is converted into a second control command to control the regulating valve of the water supply pipeline, thereby controlling the moisture content of the slurry by precisely adjusting the water injection volume.

[0047] The admixture dosage adjustment signal is converted into a third control command to control the admixture metering pump. By precisely controlling the injection rate and total amount of the admixture, the admixture dosage can be controlled.

[0048] Furthermore, it also includes performance and safety boundary monitoring steps:

[0049] Set a safe threshold range for the real-time performance indicators of the slurry;

[0050] During continuous monitoring, when any real-time performance indicator exceeds its corresponding safety threshold range, an alarm is triggered and a preset safety policy is executed.

[0051] The safety strategy includes: pausing grouting operations, locking the current control signal and switching to manual control mode, and generating a fault diagnosis report.

[0052] The beneficial effects of this invention are:

[0053] This invention achieves dynamic and adaptive adjustment of grout mix proportions and preparation process by real-time acquisition of on-site geological parameters such as stratum lithology and groundwater, and incorporating these parameters as key input variables into the grout performance model and control algorithm. This allows the prepared grout performance to be precisely matched with the current geological environment, effectively avoiding the problem of poor grouting effect caused by unsuitable grout, and fundamentally improving the grouting quality.

[0054] The intelligent control algorithm of this invention employs a rolling optimization strategy, which can predict the trend of slurry performance changes over several future cycles based on the current state and solve the constrained optimization problem. This forward-looking control method avoids lag and overshoot in the control process, making the adjustment of process parameters more stable and efficient. It ensures that the slurry performance is always maintained at the optimal state while satisfying all process constraints, thereby improving resource utilization.

[0055] This invention precisely analyzes the control signals generated by the control algorithm into specific operational instructions for underlying execution equipment such as crushers, regulating valves, and metering pumps, and establishes localized equipment-level closed-loop feedback. This achieves a high degree of automation in the slurry preparation process, reducing reliance on operator experience and human error. Simultaneously, by setting performance and safety boundary monitoring, it can provide real-time early warnings for abnormal operating conditions and activate safety strategies, effectively ensuring the safety of the construction process. Attached Figure Description

[0056] Other features, objects, and advantages of this application will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0057] Figure 1 This is a flowchart illustrating an intelligent preparation and control method for grouting slurry behind the shield tunnel wall according to the present invention.

[0058] Figure 2 This is a flowchart of the control algorithm generating control signals in the intelligent preparation and control method for grouting slurry behind the shield wall of the present invention;

[0059] Figure 3 This is a schematic diagram of the process of dynamically adjusting the grout through control signals in the intelligent preparation and control method of grout behind the shield wall of the present invention.

[0060] Figure 4 This is a schematic diagram of the structure of a computer system used to implement the methods, systems, and electronic devices of this application. Detailed Implementation

[0061] The present application will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the invention. Furthermore, it should be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings.

[0062] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0063] See Figure 1 The first embodiment of the present invention provides a method for intelligent preparation and control of grouting fluid for backfilling of tunnel boring machines, comprising:

[0064] A grout performance model is established. The grout performance model predicts the finished product performance parameters of the grout in grouting construction by analyzing the physicochemical properties of the grout raw materials and the operational variables in the preparation process. The finished product performance parameters include the grout's fluidity, setting time, and strength development characteristics.

[0065] Real-time geological parameters of the construction site are collected, including stratum lithology, groundwater distribution and surrounding rock pressure status, and the geological parameters are preprocessed to eliminate noise and outliers.

[0066] Based on the preprocessed geological parameters and the predicted output of the slurry performance model, a control signal is generated based on the control algorithm.

[0067] Based on the control signal, the particle size distribution, moisture content and admixture dosage of the semi-finished slurry are dynamically adjusted.

[0068] During the grout preparation and grouting process, the real-time performance indicators of the grout are continuously monitored, and the monitoring data is fed back to the grout performance model to correct the parameters of the grout performance model.

[0069] Specifically, this embodiment first uses a predictive slurry performance model to predict the final slurry performance under specific raw materials and processes. The system uses sensors to monitor changes in the geological environment at the construction site in real time. Then, a control algorithm combines the model's prediction results with the current geological conditions' requirements for slurry performance to calculate the optimal slurry preparation process parameters and generate corresponding control signals. Based on these signals, the execution system precisely fine-tunes the components of the slurry online during preparation. By monitoring the actual performance of the finished slurry online and feeding the monitoring results back to the model for self-learning and correction, a complete closed loop including prediction, perception, decision-making, execution, and feedback is formed, thereby achieving dynamic matching between slurry performance and geological conditions.

[0070] By constructing such a complete closed-loop control logic, this invention fundamentally changes the traditional open-loop, passive grouting method that relies on fixed experience-based proportions. It enables the grout preparation process to actively adapt to complex and changing geological environments, ensuring the accuracy, continuity, and stability of grout performance, and significantly improving grouting quality and construction efficiency.

[0071] For example, when the tunnel boring machine is about to advance from a stable clay layer to a water-rich sandy gravel layer, the geological parameter acquisition step will first detect this geological abrupt change through changes in tunneling parameters and water pressure sensors. Based on this, the control algorithm will set new slurry performance targets, such as shortening the initial setting time from 8 hours to 4 hours and increasing the 8-hour strength from 0.1 MPa to 0.3 MPa. The slurry performance model will predict based on the current mix ratio and determine that it cannot meet the new performance targets. Therefore, the control algorithm generates new control signals, and the dynamic adjustment step will use these signals to instruct the actuators to increase the accelerator dosage and precisely reduce the water-cement ratio. At the same time, the performance monitoring step will continuously verify whether the newly prepared slurry has reached the preset performance indicators and feed the results back to the model to ensure that the decisions of the entire system are always accurate.

[0072] In this embodiment, establishing the slurry performance model includes:

[0073] Collect historical slurry preparation data, including physical property data of raw materials, chemical composition data, ratio of raw materials, operating parameters of stirring equipment, environmental temperature and humidity data, and corresponding performance test results of finished slurry;

[0074] The collected data is preprocessed, including data cleaning to remove outliers, data normalization to eliminate the influence of units, and data standardization.

[0075] A data-driven relational network is constructed based on the preprocessed data. The relational network is a neural network, which learns the nonlinear relationship between slurry properties and performance parameters by training historical data.

[0076] The neural network is trained using the backpropagation algorithm, and the network weights are optimized using the gradient descent method.

[0077] During the model deployment and application phase, new slurry preparation process data and performance test data are continuously integrated. The neural network parameters are dynamically adjusted through incremental learning, and the model prediction accuracy is verified periodically. When the prediction deviation exceeds the first preset threshold, model reconstruction is triggered.

[0078] In practice, the first step is to systematically collect massive amounts of data from multiple historical engineering projects. This data covers the physicochemical properties of raw materials, such as the specific surface area of ​​cement being 350 square meters per kilogram and the loss on ignition of fly ash being 3.5%; the mass ratio of each component, such as a water-cement ratio of 1.0; the operating parameters of the mixer, such as a stirring current of 50 amperes and a rotation speed of 30 revolutions per minute; and the ambient temperature and humidity during preparation, such as a temperature of 28 degrees Celsius. Simultaneously, corresponding test results of the finished slurry performance are collected, for example:

[0079] Flowability test results: The outflow time measured by a cement slurry flowability tester (such as the funnel flow rate method), for example, 18 seconds; or the slurry diffusion diameter measured by the diffusion diameter method, for example, 220 mm.

[0080] Setting time test results: Initial setting time measured using a Vicat apparatus, for example, 480 minutes; final setting time, for example, 600 minutes.

[0081] Strength development characteristics test results: Unconfined compressive strength of grout stone body at different ages, for example, 8-hour strength is 0.1 MPa, 24-hour strength is 0.8 MPa, and 28-day strength is 2.5 MPa.

[0082] The data were statistically analyzed using box plots to remove outliers located more than 1.5 times the interquartile range above and below the quartiles.

[0083] A max-min normalization method is used to map all feature values ​​of different dimensions to the interval between 0 and 1. Based on the preprocessed data, a fully connected feedforward neural network is constructed, which includes one input layer, three hidden layers, and one output layer. The hidden layers use ReLU (Modified Linear Unit) as the activation function, and the output layer predicts performance parameters such as mobility and condensation time.

[0084] During training, the gradient descent algorithm, specifically the Adam optimizer, is employed in conjunction with backpropagation to iteratively optimize network weights on a large dataset, minimizing the mean squared error between predicted and true values. After model deployment, the system continuously collects new data, fine-tunes the model with a minimal learning rate through incremental learning, and sets a first preset threshold. For example, if the mean absolute percentage error of prediction exceeds 5% for three consecutive days, the system automatically triggers a comprehensive offline model reconstruction process.

[0085] This modeling method, based on big data and neural networks, can accurately capture the highly complex nonlinear relationships between various influencing factors of slurry and the performance of the finished product, with prediction accuracy far exceeding that of traditional empirical formulas. Simultaneously, the combination of online updates and offline reconstruction as dual optimization mechanisms ensures that the model can adapt to changes such as raw material batch fluctuations and equipment aging over the long term, maintaining high reliability and providing a solid and reliable decision-making foundation for the entire intelligent control system.

[0086] The computational process for model training is illustrated below: The preprocessed dataset is divided into training, validation, and test sets at a ratio of 70%, 20%, and 10%, respectively. A batch of data is taken from the training set, with the input vector being a set of normalized values ​​whose original physical meanings include a cement specific surface area of ​​350 square meters per kilogram, a fly ash content of 20%, a water-cement ratio of 1.0, and an ambient temperature of 28 degrees Celsius. The corresponding real-world performance labels for this input are a flowability of 220 mm and an initial setting time of 480 minutes.

[0087] The neural network, based on its current weights, predicts a flowability of 195 mm and an initial setting time of 510 minutes. Next, the mean square error E between the predicted and actual values ​​is calculated using the formula E = 0.5 × [(220 - 195)]. 2 +(480-510) 2 ] = 0.5 × [25 2 +(-30) 2 =762.5. Using the backpropagation algorithm, the partial derivatives of the error E with respect to each weight w in the network are calculated. The Adam optimizer automatically adjusts the learning rate and updates the weights based on the direction and magnitude of the gradient represented by the partial derivative. The update rule can be expressed as follows: This process is iterated thousands of times until the model's error on the validation set converges to a minimum. Where w old The weights before the update, w new For the updated weights, learning rate This is the learning rate.

[0088] In this embodiment, the real-time geological parameters of the construction site include:

[0089] Acquire stratigraphic lithology data, groundwater distribution data, and surrounding rock pressure status data;

[0090] The geological parameters are preprocessed, including data cleaning to remove outliers, data normalization to eliminate the influence of dimensions, and data fusion to integrate multi-source sensor information into a unified set of geological parameters.

[0091] The preprocessed geological parameters are used as inputs to the slurry performance model and control algorithm.

[0092] In practice, the current lithology of the strata is comprehensively assessed by analyzing the vibration spectrum, torque, and thrust of the tunnel boring machine (TBM) cutterhead in real time, combined with existing advanced geological forecasting information. Multiple pore water pressure sensor arrays, evenly distributed circumferentially along the TBM shell, monitor the pressure and distribution of groundwater in real time. The pressure state of the surrounding rock is calculated and evaluated using pressure sensor data from the soil chamber and thrust data from the TBM propulsion cylinders. The collected raw data first undergoes median filtering to remove outliers such as jumps caused by transient interference from the sensors, and then normalizes the data. The key to this step is data fusion.

[0093] For example, multiple discrete earth pressure sensor readings are input together with the total thrust data of the tunnel boring machine into a Kalman filter model. Through the state estimation function of this model, a more stable and representative equivalent surrounding rock pressure parameter is output. Finally, the processed core parameters such as stratum lithology index, average water pressure, and equivalent surrounding rock pressure are integrated into a unified set of geological parameters and input into the control system.

[0094] This invention, through multi-source information fusion and data preprocessing, extracts stable, reliable, and standardized geological characteristic parameters from raw, noisy sensor data collected on-site. This significantly improves the accuracy and robustness of the system's perception of the geological environment, providing high-quality input information for subsequent control algorithms to make precise decisions, and is a prerequisite for realizing on-demand customization of slurry performance.

[0095] An example of the calculation process for geological parameter preprocessing is as follows: At a certain moment, the readings of the six water pressure sensors installed on the shield shell are 0.31, 0.32, 0.29, 1.85, 0.33, and 0.30, respectively, in megapascals (MPa). First, by setting a reasonable physical threshold, such as 1.0 MPa, 1.85 MPa is identified as a significant outlier and removed. Then, the arithmetic mean of the remaining five valid readings (0.31, 0.32, 0.29, 0.33, and 0.30) is calculated, yielding the average water pressure at this section as (0.31 + 0.32 + 0.29 + 0.33 + 0.30) / 5 = 0.31 MPa.

[0096] Meanwhile, the average reading of the soil chamber pressure sensor was 0.35 MPa, and the pressure converted from the total thrust of the tunnel boring machine was 0.34 MPa. The data fusion module takes these three values, namely 0.31 MPa, 0.35 MPa, and 0.34 MPa, as input and performs calculations through a preset weighted fusion model.

[0097] For example, the equivalent surrounding rock pressure = 0.3 × average water pressure + 0.5 × earth chamber pressure + 0.2 × total thrust pressure = 0.3 × 0.31 + 0.5 × 0.35 + 0.2 × 0.34 = 0.336 MPa. This value will be normalized and used as a unified index of surrounding rock pressure state for subsequent calculations.

[0098] See Figure 2 In this embodiment, generating control signals through a control algorithm includes:

[0099] Based on the pretreated geological parameters, determine the target values ​​of slurry performance that are adapted to the current geological conditions. The target performance values ​​include target fluidity, target setting time, and target strength.

[0100] A multivariate objective function is constructed with the goal of optimizing slurry performance. The objective function is based on a weighted combination of the deviations between the predicted and target values ​​of slurry fluidity, setting time, and strength development. The predicted values ​​are output by the slurry performance model.

[0101] Set constraints on process variables, including particle size distribution range, moisture content range, and admixture dosage range;

[0102] Within each control cycle, starting from the current process variable setpoints, an optimization problem with the objective of minimizing the multivariate objective function and subject to the constraints is solved to calculate a sequence of optimal future process variable setpoints.

[0103] The instruction for the current control cycle is selected from the optimal setpoint sequence, and a real-time adjustment signal is generated as the control signal.

[0104] In this embodiment, the target value of slurry performance adapted to the current geological conditions is determined based on the pre-processed geological parameters. The method is as follows:

[0105] Based on the real-time geological parameters, the target value ranges for various performance indicators of the slurry are determined, including the value ranges for fluidity, setting time, and strength development characteristics.

[0106] Within the target value range, multiple candidate combinations of slurry performance target values ​​are generated by systematically combining different values ​​of fluidity, setting time, and strength development characteristics within their respective ranges.

[0107] Obtain a set of baseline process variable settings, wherein the baseline process variable settings are the process variable settings used to stably prepare qualified slurry in current or historical production.

[0108] For each of the aforementioned candidate target performance value combinations, perform the following steps:

[0109] The geological conditions corresponding to the combination of the baseline process variable set value and the current candidate target performance value are input into the slurry performance model to predict a set of slurry performance prediction values.

[0110] The overall deviation between the current candidate target performance value combination and the predicted slurry performance value is calculated based on preset weighting coefficients.

[0111] From all the candidate target performance value combinations, the combination with the smallest overall deviation is selected as the final determined slurry performance target value.

[0112] In the specific implementation process, when determining the target values ​​of slurry performance that are adapted to the current geological conditions based on the pre-processed geological parameters, the target value range of various slurry performance indicators is first determined based on the real-time collected geological parameters, including the value range of fluidity, setting time, and strength development characteristics.

[0113] Within the target value range, multiple candidate combinations of slurry performance target values ​​are generated by systematically combining different values ​​of fluidity, setting time, and strength development characteristics within their respective ranges. Simultaneously, a set of baseline process variable settings is obtained, which are the process variable settings used to stably prepare qualified slurries in current or historical production.

[0114] For each candidate target performance value combination, the following steps are performed: input the baseline process variable setting value and the geological conditions corresponding to the current candidate target performance value combination into the slurry performance model to predict a set of slurry performance prediction values; calculate the overall deviation between the current candidate target performance value combination and the slurry performance prediction values; select the combination with the smallest overall deviation from all candidate target performance value combinations as the final determined slurry performance target value.

[0115] The calculation of the overall deviation requires integrating the deviations of three indicators with different dimensions and importance—fluidity, setting time, and strength—into a unified scalar. To this end, weighting coefficients a, b, and c are introduced, corresponding to the relative importance of the deviations in the development characteristics of fluidity, setting time, and strength, respectively. The formula for calculating the overall deviation is:

[0116] Overall deviation = a × (predicted mobility - candidate mobility) 2 +b×(Predicted setting time - Candidate setting time) 2 +c×(predicted intensity - candidate intensity) 2 ;

[0117] The specific values ​​of the weighting coefficients a, b, and c are set by the engineer based on the characteristics of the current geological conditions and the focus of the project requirements. The sum of their values ​​is usually 1 to reflect the priority of each performance index under specific working conditions.

[0118] Assuming the current geological parameters indicate high-pressure water and weak surrounding rock conditions, the system sets the target value range for the slurry performance indicators as follows: fluidity of 180-220 mm, initial setting time of 300-420 min, and 8-hour strength of 0.15-0.25 MPa. The system generates candidate combination A (candidate fluidity = 190 mm, candidate setting time = 360 min, candidate strength = 0.2 MPa). The baseline process variables are set as fineness modulus 2.5, water-cement ratio 0.95, and admixture dosage 0.8%. Inputting the baseline process variables and the current geological conditions into the slurry performance model, the predicted performance values ​​are: predicted fluidity = 192 mm, predicted setting time = 365 min, and predicted strength = 0.19 MPa.

[0119] Engineers determined that in this soft, water-rich stratum, setting time was the most critical factor, followed by strength, and then fluidity. Therefore, weighting coefficients were set as follows: a = 0.2, b = 0.5, c = 0.3.

[0120] Calculate the overall deviation of candidate combination A:

[0121] Overall deviation = 0.2 × (192 - 190) 2 +0.5×(365-360) 2 +0.3×(0.19-0.2) 2 =13.30003;

[0122] By comparing the overall deviation of all candidate combinations, the performance parameters of combination A with the smallest deviation were finally determined as the target values: target flowability 190 mm, target initial setting time 360 ​​minutes, and target 8-hour strength 0.2 MPa.

[0123] After calculating the target value, this embodiment establishes a mathematical optimization model. The objective function J can take the form of:

[0124] In this function, F, T, and S represent the model-predicted fluidity, setting time, and intensity, respectively. , , These are the slurry fluidity, setting time, and strength development characteristics predicted by the slurry performance model. , , These are performance target values ​​determined based on real-time geological parameters, namely, the desired fluidity, setting time, and strength. , , These are the weighting coefficients for each performance characteristic, reflecting their relative importance. Constraints are set based on material and equipment limitations; for example, the fineness modulus of the sand must be between 2.3 and 2.8, and the water-cement ratio must be between 0.9 and 1.1. In each pulping cycle, the control algorithm uses a slurry performance model to quickly simulate the slurry performance under different combinations of process variables within the constraints. An optimization algorithm is then used to find the optimal set of process variable settings that minimizes the objective function J, and this optimal set is used as the control command for the current cycle. Specifically:

[0125] The step of calculating the optimal setpoint sequence of process variables by solving a constrained optimization problem includes:

[0126] Step A: Starting with a set of initial process variable settings that satisfy the process variable constraints, perform iterative solutions.

[0127] Step B, in each iteration, performs the following steps:

[0128] Based on the current iteration's process variable setpoints, a quadratic programming subproblem is constructed by making a second approximation of the multivariate objective function at the current setpoint, and the nonlinear process variable constraints are linearized at the current setpoint.

[0129] Solving the quadratic programming subproblem yields a search direction vector that defines the direction and proportion of process variable adjustment;

[0130] A one-dimensional line search is performed along the search direction vector, and an optimal step size factor is determined by evaluating a value function that simultaneously contains the original objective function and a penalty term for violating the original nonlinear constraints.

[0131] Based on the search direction vector and the optimal step size factor, update the process variable settings to generate new settings for the next iteration;

[0132] Step C: Repeat the iterative process of step B until the change in the process variable setpoint is less than the first preset convergence threshold, or the norm of the search direction vector is less than the second preset convergence threshold; and

[0133] Step D: Output the final process variable setpoints obtained at the end of the iteration as the optimal setpoint sequence.

[0134] The step of calculating the optimal setpoint sequence of process variables by solving a constrained optimization problem is specifically manifested as an iterative optimization solution process based on local quadratic model approximation, which includes:

[0135] An initial process variable setting value that satisfies the basic process constraints is used as the starting state point for iteration.

[0136] Within each iteration cycle, the solution process first constructs a local quadratic programming subproblem based on the state point with the current process variable setpoints. This subproblem involves two aspects: firstly, calculating the first derivative (gradient) and second derivative (Hessian matrix) of the multivariate objective function at the current state point to form a quadratic objective function that accurately describes the local curvature and trend of the original objective function in the neighborhood of the current point; secondly, linearizing the nonlinear equations and inequalities describing the process variable constraints to form a set of linear constraint boundaries. Mathematically, this quadratic programming subproblem uses a simpler, more easily solvable quadratic function to simulate the shape of the complex, nonlinear original objective function near the current solution, and uses a set of linear planes to approximate the curved boundaries of the original feasible region.

[0137] The algorithm solves the constructed quadratic programming subproblem, and its solution is essentially a search direction vector pointing to better performance. This vector not only indicates the direction and relative proportion of each process variable that should be adjusted to reduce the objective function value, but also takes into account linearization constraints to ensure that the adjustment direction does not immediately conflict with the boundary of the feasible region.

[0138] Instead of directly adopting the endpoint of the vector as the next state, the algorithm initiates a one-dimensional line search process. This process, along the search direction, determines an optimal step size factor by evaluating a value function that incorporates the original objective function and constraint penalties. This step size factor ensures that, while moving towards a better solution, the value of the original objective function is reduced to the greatest extent possible without exceeding the feasible region defined by the original nonlinear constraints. The precision of this step lies in its balance between the true model of global nonlinearity and the approximate model of local linearization, preventing iterative divergence caused by inaccurate local approximations.

[0139] After determining the optimal step size factor, the algorithm moves the current process variable setpoint along the previously calculated search direction by a distance determined by the step size factor, thereby obtaining a new process variable setpoint with better performance, and uses this as the starting point for the next iteration.

[0140] Finally, the system continues to repeat the iterative loop of "constructing quadratic subproblems, solving for search directions, performing line searches, and updating state points" until the difference between the process variable setpoints obtained from two consecutive iterations is less than a preset minimum convergence threshold, or the length of the calculated search direction vector approaches zero. This indicates that the algorithm has found a local optimum. At this point, the iterative process terminates, and the final set of process variable setpoints is taken as the optimal solution that minimizes the objective function while satisfying all constraints, and output as the control command for the current cycle.

[0141] This control strategy, based on model prediction and rolling optimization, has significant foresight. Instead of passively correcting deviations after they occur, it pre-calculates the optimal control scheme before preparation, thus proactively avoiding lag and overshoot in the control process. This ensures that the slurry performance, while meeting all process constraints, always approaches the engineering requirements in an efficient and economical manner, achieving optimal resource utilization.

[0142] This invention transforms the grout preparation process from an open-loop operation relying on fixed proportions and experience into an intelligent closed-loop control system based on model prediction and real-time optimization. Traditional methods cannot adapt to dynamically changing geological conditions, leading to poor grouting effects and quality fluctuations. This invention fundamentally solves this problem by constructing a precise mathematical optimization framework. In each control cycle, this framework compares the key indicators such as fluidity and setting time predicted by the grout performance model with the target values ​​determined based on real-time geological conditions, and constructs a multivariate objective function. Subsequently, using the advanced optimization algorithm of sequential quadratic programming, under the strict constraint of process constraints such as particle size distribution and moisture content, a set of optimal setpoints for future process variables is solved in real time.

[0143] The innovation of this optimization process lies in its forward-looking and adaptive nature. The algorithm does not passively correct deviations only after they occur, but actively calculates the optimal control strategy that continuously brings the slurry performance close to the target through an iterative cycle of prediction, optimization, and correction. Specifically, it constructs a local quadratic programming subproblem at each iteration point, capturing the local variation trend of the system by calculating the gradient of the objective function and the Hessian matrix, and linearizing nonlinear constraints, thereby efficiently determining the optimal adjustment direction and step size of the process variables. This rolling optimization mechanism effectively overcomes the inherent lag of traditional control methods, ensuring that the slurry performance maintains a high degree of stability and consistency even in the face of complex and changing geological environments.

[0144] Furthermore, this invention seamlessly integrates the high-level decisions generated by the optimization algorithm with the underlying physical execution system, achieving a seamless transition from intelligent decision-making to precise execution. Control signals are precisely parsed into specific instructions for the coarse / fine aggregate bins, water supply regulating valves, and admixture metering pumps, and a local closed-loop feedback is formed through real-time monitoring to ensure execution accuracy. Simultaneously, the system continuously feeds back the actual performance data of the prepared grout to the grout performance model for online self-correction and parameter calibration. This forms a complete adaptive loop encompassing geological perception, performance prediction, intelligent decision-making, precise execution, and feedback learning, enabling the system to continuously learn from practice and optimize its own decisions. Ultimately, this achieves a high degree of adaptive matching between grout performance and formation conditions, significantly improving the reliability of grouting quality and construction efficiency.

[0145] The calculation process of the control algorithm is illustrated below: Given the current geological parameters of high-pressure water and weak surrounding rock, the system sets the performance target as: target mobility F. target =190 mm, target initial setting time T target =360 minutes, target 8-hour intensity S target =0.2 MPa. The weighting coefficient is set to w according to project requirements. f =0.4, w t =0.3, w s =0.3.

[0146] At this point, the control algorithm initiates optimization calculations. In one loop, the algorithm tries the first set of process variables, such as fineness modulus = 2.5, water-cement ratio = 0.95, and admixture dosage = 0.8%, inputting them into the slurry performance model to obtain the predicted performance F. pred1 =210 mm, T pred1 =400 minutes, S pred1 =0.15 MPa. Substituting into the objective function, the cost is calculated as follows:

[0147] J1 = 0.4 × (210 - 190) 2 +0.3×(400-360) 2 +0.3×(0.15-0.2) 2 =160+480+0.00075=640.00075.

[0148] Trying the second set of variables, such as fineness modulus = 2.6, water-cement ratio = 0.92, and admixture dosage = 1.0%, the predicted performance is F. pred2 =195 mm, T pred2 =370 minutes, S pred2 =0.21 MPa, calculate the cost:

[0149] J2 = 0.4 × (195 - 190)2 +0.3×(370-360) 2 +0.3×(0.21-0.2) 2 =10 + 30 + 0.00003 = 40.00003. After multiple iterative searches, the sequential quadratic programming algorithm finally finds a set of optimal variables, such that the calculated cost J... opt The optimal value is minimized, and this set of optimal variables is output as control commands.

[0150] The iterative search of the Sequential Quadratic Programming (SQP) algorithm is not a simple trial of discrete points. Its core principle is to determine an efficient direction that causes the objective function value to decrease most rapidly by solving a simplified subproblem based on the information of the current point. The following are the detailed steps of its first iterative calculation starting from point J1:

[0151] First, the algorithm operates at the initial process variable point x0 = [2.5, 0.95, 0.8]. T (Its cost J(x0) = 640.00075) is used for local characteristic analysis. The core of this step is to quantify the changing trend of the objective function at the current point. This is achieved by using the partial derivatives of the output with respect to the input provided by the slurry performance model (e.g., ...). , (etc.), and use the chain rule to calculate the gradient vector of the objective function J with respect to each process variable. For example, the calculation result is This vector precisely indicates the direction in which the cost value increases the fastest. Simultaneously, the algorithm employs quasi-Newton methods (such as BFGS) to construct or update an approximate H0 of the Hessian matrix describing the local curvature of the function, typically initialized to the identity matrix during the first iteration.

[0152] Based on this gradient vector and Hessian matrix, the algorithm constructs and solves a quadratic programming (QP) subproblem, the goal of which is to determine a search direction vector d=[d1, d2, d3] that maximizes the reduction of cost. T The objective function for this subproblem is: It approximates the complex original objective function near the x0 point using a simple quadratic function. Simultaneously, the original upper and lower bound constraints of the process variables are linearized into constraints on the search direction d; for example, 2.2 ≤ 2.5 + d1 ≤ 3.0 is transformed into -0.3 ≤ d1 ≤ 0.5. Using a standard QP solver, the algorithm obtains an optimal search direction, for example, d0 = [0.11, -0.03, 0.20]. T It provides precise proportions and directions for the next step of variable adjustment.

[0153] Since the QP subproblem is only a local approximation, directly moving a full step size d0 may not be optimal. Therefore, to determine the optimal distance to move in this direction, the algorithm performs a one-dimensional line search to find the optimal step size α > 0. This process involves testing a series of candidate points defined by x(α) = x0 + α × d0. For example, the algorithm might try α = 1.0 and obtain the point [2.61, 0.92, 1.0]. T And by trying α=0.9, we get the point [2.599, 0.923, 0.98]. T These points are then substituted into the original, accurate slurry performance model to calculate their true cost. By comparing the true cost corresponding to different α values, the algorithm finally determines an optimal step size that maximizes the decrease in the J value, for example, α0 = 0.95.

[0154] Finally, the algorithm updates the process variables by combining the search direction and the optimal step size, and calculates the new iteration point: x1 = x0 + α0 × d0 ≈ [2.60, 0.92, 0.99] T The cost J(x1) of the new point (x1) will be very close to J2=40.00003 in the example, marking a successful iteration from a high cost point to a low cost point. At this point, a complete computational loop ends, but the optimization does not stop. The algorithm will use x1 as the new starting point, update the Hessian matrix approximation H1 using the information from this iteration, and repeat the above steps of analysis, solution, line search, and update, iterating continuously until the magnitude of the search direction vector or the decrease in cost value is sufficiently small, satisfying the preset convergence condition. The algorithm will then output the process variables at this point as the optimal control command.

[0155] See Figure 3 In this embodiment, the dynamic adjustment of the particle size distribution, moisture content, and admixture dosage of the semi-finished slurry according to the control signal includes:

[0156] The control signals are analyzed into particle size distribution adjustment signals, moisture content adjustment signals, and admixture dosage adjustment signals.

[0157] The particle size distribution of the slurry is adjusted based on the particle size distribution adjustment signal, the moisture content of the slurry is adjusted based on the moisture content adjustment signal, and the admixture dosage is adjusted based on the admixture dosage adjustment signal.

[0158] The actual values ​​of particle size distribution, moisture content and admixture dosage are monitored in real time and compared with the target values. When the deviation exceeds the allowable range, the control signal is recalculated and the adjustment operation is adjusted.

[0159] Specifically, after receiving the optimal setpoint generated in the previous step, the system first parses it into three parallel adjustment tasks. Then, the system drives the actuators controlling the feeding of coarse and fine aggregates, controls the regulating valve of the water supply pipeline, and controls the admixture metering pump to adjust the corresponding components of the semi-finished slurry still in the mixing chamber.

[0160] During this process, equipment such as online laser particle size analyzers, microwave moisture meters, and high-precision flow meters are used to monitor the actual values ​​of particle size distribution, moisture content, and admixtures in real time. The controller continuously compares these actual values ​​with the target values, forming a local closed-loop feedback. Once the deviation exceeds the preset allowable range, for example, if the moisture content deviation continues to exceed 2%, the system will immediately fine-tune the output of the actuator to ensure that all components are precisely controlled before the slurry preparation is completed.

[0161] This dynamic adjustment mechanism with local closed-loop feedback enables high-precision online control of slurry components. It not only ensures that upper-level decision-making commands are executed flawlessly, but also compensates in real time for lower-level execution errors such as uneven material feeding and sensor drift. This significantly improves the first-pass yield of the slurry and reduces material waste, making it a key execution link for achieving full-process automation and precision.

[0162] An example of the dynamic adjustment calculation process is as follows: The control signal contains a target moisture content of 55.0%. At the start of adjustment, the water supply valve is opened. The initial reading of the online moisture meter is 52.5%. The controller calculates the deviation as 52.5% - 55.0% = -2.5%. According to the PID control logic, the control algorithm calculates that the initial valve opening should be 60%. As water is added, the real-time feedback reading of the moisture meter becomes 53.5%, and the deviation decreases to 53.5% - 55.0% = -1.5%. The PID controller recalculates based on the new deviation and adjusts the valve opening to 55%. This cycle of measurement, comparison, and adjustment continues until the moisture meter reading stabilizes within the allowable range of 55.0% plus or minus 0.2%. At this point, the local closed-loop adjustment is complete, and the valve closes or remains at a small opening to maintain stability.

[0163] In this embodiment, continuously monitoring the real-time performance indicators of the slurry and feeding the monitoring data back to the slurry performance model includes:

[0164] Real-time data on the fluidity, setting time, and strength of the slurry are obtained through online monitoring; the monitored real-time data are compared with the predicted values ​​of the slurry performance model, and the deviation is calculated.

[0165] Based on the magnitude of the deviation, the internal parameters of the slurry performance model are adjusted; feedback data is used to calibrate geological parameters and control algorithms to form an adaptive loop.

[0166] In implementation, online rheometers and ultrasonic analyzers are installed at the discharge port of the mixing plant or on the grouting pipeline to acquire physical quantities reflecting the slurry's fluidity, setting process, and early strength development in real time. The system compares these online monitored actual performance data with the predicted values ​​given by the model for the same batch of slurry before preparation and calculates the deviation between the two. This deviation information is used not only to adjust the internal parameters of the slurry performance model, such as the weights of the neural network, but also for higher-level calibration, such as correcting the interpretation rules of geological parameters or the weight coefficients in intelligent control algorithms.

[0167] This step establishes a feedback loop for the entire system, enabling true self-learning and self-adaptation. By continuously learning from actual construction results, the system can continuously optimize its internal model and control strategies, constantly evolving its adaptability to environmental changes and its level of intelligent decision-making, thus ensuring the continuous stability and reliability of grouting quality during long-term construction.

[0168] An example of the feedback optimization calculation process is as follows: For a certain batch of slurry, the model predicts its initial setting time to be 420 minutes before preparation. After the slurry is prepared, an online ultrasonic monitor continuously tracks the change in its sound wave propagation speed and calculates the actual initial setting time of the slurry to be 405 minutes based on a preset curve showing the relationship between wave velocity and setting time. The system records the deviation between the predicted and actual values ​​as 420 - 405 = +15 minutes. This data point, containing the input variable, the predicted value of 420, and the actual value of 405, is stored in the database. When enough such data is accumulated, these deviations are used to fine-tune the model's parameters. Furthermore, if the system finds that the predicted setting time is consistently and systematically longer under certain geological conditions, the system will also use this feedback data to fine-tune the target value setting of the setting time under those geological conditions in the control algorithm, forming a higher level of adaptation.

[0169] In this embodiment, the method for adjusting the internal parameters of the slurry performance model is as follows:

[0170] Based on the aforementioned deviation, the weights and bias parameters of the neural network are dynamically adjusted through incremental learning.

[0171] The model prediction accuracy is verified periodically, and when the deviation exceeds a set threshold, the reconstruction process of the slurry performance model is triggered.

[0172] In practice, a two-stage update strategy is adopted. The first stage is high-frequency online fine-tuning: whenever a new set of input and actual output data pairs is obtained, online gradient descent is used to update the weights and biases of the neural network with a very small learning rate. This allows the model to adapt to small changes in real time and gradually.

[0173] The second level is low-frequency offline reconstruction: the system automatically calculates the root mean square error of all predictions made in the past week each week. When this error exceeds a preset reconstruction threshold, such as 8%, the reconstruction process is automatically triggered. This process retrieves all historical data, including all new data, trains a new neural network model from scratch, and replaces the old model running online after successful validation.

[0174] This update strategy, which combines fine-tuning and reconstruction, balances the model's real-time adaptability with its long-term stability. Incremental learning ensures the model's rapid response to new situations, while periodic reconstruction fundamentally corrects potential performance degradation caused by significant changes in operating conditions, ensuring the model remains in optimal predictive condition throughout the entire project lifecycle.

[0175] The calculation process for model updates is illustrated below: Suppose that during a certain online fine-tuning, the prediction error of a new data point leads to a backpropagation, and the error with respect to a certain weight w is calculated. ij The gradient is g. The incremental learning algorithm updates the weights with a very small learning rate, for example, β equals 0.0001. The update formula is: updated weight = original weight - β × g. This update is small, but continuous updates allow the model to gradually approach the new data distribution. During model reconstruction, the system finds that the root mean square error of prediction over the past week has reached 8.5%, exceeding the 8% threshold. It collects all past data, re-divides it into 70%, 20%, and 10% portions, and then initiates a complete training process that may take several hours. Through tens of thousands of iterations, it finds a completely new set of optimal weights, thoroughly replacing the parameters of the old model.

[0176] In this embodiment, the control signal is parsed into a particle size distribution adjustment signal, a moisture content adjustment signal, and an admixture dosage adjustment signal. The method is as follows:

[0177] Read the target values ​​contained in the control signal that correspond to particle size distribution, moisture content and admixture dosage, respectively;

[0178] The target value of particle size distribution is compared with the currently detected actual value of particle size distribution, and a particle size distribution adjustment signal is generated based on the difference.

[0179] The target moisture content value is compared with the currently detected actual moisture content value, and a moisture content adjustment signal is generated based on the difference.

[0180] The target value of the admixture dosage is compared with the actual value of the currently detected admixture dosage, and an admixture dosage adjustment signal is generated based on the difference.

[0181] In software implementation, the control signal is typically a structured data object. The parsing process involves extracting specific values ​​by accessing specific fields of this object, such as a field named "target fineness modulus" or a field named "target water-cement ratio." Subsequently, the control system reads the current actual values ​​from relevant online sensors, such as particle size analyzers and moisture meters, and calculates the difference between the target value and the actual value, i.e., the deviation. Finally, based on this deviation, a specific adjustment signal with direction and amplitude is generated through a PID controller or fuzzy logic rules, such as "increase the coarse sand feeding rate by 10%" or "increase the metering pump frequency by 5 Hz."

[0182] This method of generating adjustment signals based on the deviation between target and actual values ​​is the core of achieving precise closed-loop control. It decomposes abstract performance targets layer by layer, ultimately transforming them into specific, quantifiable operational instructions for physical equipment. This ensures that control intentions are accurately transmitted to the execution layer, serving as a crucial bridge connecting decision-making and action.

[0183] The calculation process for signal generation is illustrated below: The control signal reads a target admixture dosage of 1.2%. The flow meter of the admixture metering pump reports the current cumulative dosage, converted to 0.9%. The controller calculates the error as 1.2% - 0.9% = 0.3%. Based on the preset PID parameters—proportional coefficient P = 0.5, integral coefficient I = 0.1, and derivative coefficient D = 0.01—the PID controller calculates the control quantity to be output, for example, a voltage signal representing the metering pump frequency. The calculation formula is as follows: The calculated output value is a specific, quantified adjustment signal that directly drives the metering pump to operate at a higher frequency to quickly eliminate this 0.3% deviation.

[0184] In this embodiment, the particle size distribution of the slurry is adjusted based on the particle size distribution adjustment signal, the moisture content of the slurry is adjusted based on the moisture content adjustment signal, and the admixture dosage is adjusted based on the admixture dosage adjustment signal. The method is as follows:

[0185] The particle size distribution adjustment signal is converted into a first control command to control the feeding actuators of the coarse aggregate bin and the fine aggregate bin. By adjusting the feeding ratio and rate of coarse and fine aggregates, the particle size distribution of the slurry is controlled.

[0186] The moisture content adjustment signal is converted into a second control command to control the regulating valve of the water supply pipeline, thereby controlling the moisture content of the slurry by precisely adjusting the water injection volume.

[0187] The admixture dosage adjustment signal is converted into a third control command to control the admixture metering pump. By precisely controlling the injection rate and total amount of the admixture, the admixture dosage can be controlled.

[0188] Specifically, the system converts the adjustment signals generated in the previous step into physical signals that can directly drive the hardware via a programmable logic controller (PLC) or a distributed control system (DCS). For example, the particle size distribution adjustment signal is converted into a 4-20 mA current signal for the variable frequency belt scales below the coarse and fine aggregate bins, allowing for precise proportioning by independently controlling the feeding speed of both. The moisture content adjustment signal is converted into a 0-10 volt voltage signal for the electric regulating valve positioner to precisely control the valve opening. The admixture dosage adjustment signal is converted into a pulse frequency signal for the metering pump controller to precisely control its stroke count.

[0189] This invention materializes digital control commands into precise physical operations on field equipment. By decomposing comprehensive adjustment tasks into individual, specialized actuators and driving them with standardized industrial control signals, a high degree of automation and reliable execution of the slurry preparation process is achieved, ensuring the quality of the final slurry product.

[0190] An example of the calculation process for control command conversion is as follows: A moisture content adjustment signal requires increasing the water supply valve opening from the current 40% to 45%. The total valve stroke corresponds to a control voltage of 0 to 10 volts. Therefore, the system calculates that the current required output voltage is 4.0 volts. The target state corresponds to a voltage of 4.5 volts. The PLC's analog output module will immediately adjust the output voltage from 4.0 volts to 4.5 volts. After receiving the 4.5 volt signal, the valve positioner will drive the valve to move until its internal position sensor reports an opening of 45%, thus completing the precise execution of the control command.

[0191] This embodiment also includes a performance safety boundary monitoring step: setting a safety threshold range for the real-time performance indicators of the slurry; during continuous monitoring, when any real-time performance indicator exceeds its corresponding safety threshold range, an alarm is triggered and a preset safety strategy is executed; the safety strategy includes:

[0192] Pause grouting operations, lock the current control signal and switch to manual control mode, and generate a fault diagnosis report.

[0193] During the system configuration phase, technicians will set an inviolable safety threshold for key performance indicators based on engineering specifications and geological conditions. For example, the fluidity must be within the range of 180 mm to 250 mm, and the initial setting time must be greater than 4 hours. During system operation, the online monitoring module will continuously compare real-time performance data with this safety range. Once any indicator exceeds the limit, such as a sudden drop in fluidity to 170 mm, the system will immediately trigger the highest priority safety interlock procedure. This procedure will first send an emergency stop command to the grouting pump and mixing system; simultaneously, it will freeze the output of all automatic control signals, force the human-machine interface to switch to manual control mode, and issue an audible and visual alarm; finally, the system will automatically capture all relevant data within five minutes before and after the anomaly occurs, including raw material ratios, equipment status, sensor readings, etc., and generate a detailed fault diagnosis report to provide a basis for subsequent manual troubleshooting.

[0194] This performance safety boundary monitoring mechanism provides a crucial safety barrier for the entire highly automated system. It can promptly and decisively halt unsafe operations in the event of extreme abnormal conditions, preventing substandard grout from being injected into the tunnel and causing quality accidents. It also provides valuable data support for rapid fault recovery, greatly ensuring construction safety and project quality.

[0195] An example of the calculation process triggered by the safety policy is as follows: The system sets the 8-hour intensity safety lower limit to 0.08 MPa. Based on real-time wave velocity data, the online ultrasonic instrument calculates the predicted 8-hour intensity using its internal correlation model to be only 0.07 MPa. The system immediately performs a comparison: since 0.07 MPa is less than 0.08 MPa, it is determined that the safety boundary condition has been triggered. At this point, the system immediately executes a preset logical sequence: first, it sends a "stop" command to the grouting pump controller; second, it sets the system's operating mode to "manual"; finally, it triggers a report generation program. After receiving the instruction, the PLC immediately disconnects the grouting pump's operating relay, illuminates the "manual mode" indicator on the control panel, and begins writing relevant variables from the memory buffer to the log file, generating a fault report.

[0196] The second embodiment of the present invention proposes an intelligent preparation and control system for grouting fluid behind the shield tunnel wall, comprising:

[0197] The performance prediction module is used to establish and run a slurry performance model. The slurry performance model predicts the finished product performance parameters of the slurry in grouting construction by analyzing the physicochemical properties of the slurry raw materials and the operational variables in the preparation process. The finished product performance parameters include the slurry's fluidity, setting time, and strength development characteristics.

[0198] The geological sensing module is used to collect real-time geological parameters of the construction site, including stratum lithology, groundwater distribution and surrounding rock pressure status, and to preprocess the geological parameters to eliminate noise and outliers.

[0199] The intelligent control module is communicatively connected to the performance prediction module and the geological sensing module, and is used to generate control signals through a control algorithm based on the preprocessed geological parameters and the predicted output of the slurry performance model.

[0200] The dynamic execution module is communicatively connected to the intelligent control module and is used to dynamically adjust the particle size distribution, moisture content and admixture dosage of the semi-finished slurry according to the control signal.

[0201] The online monitoring and feedback module is used to continuously monitor the real-time performance indicators of the slurry during the slurry preparation and grouting process, and to feed the monitoring data back to the performance prediction module to correct the parameters of the slurry performance model.

[0202] This embodiment provides an intelligent preparation and control system for grouting fluid behind the shield tunnel wall, used to implement the method described in the first embodiment. This system forms a complete intelligent control closed loop through the coordinated operation of multiple functional modules.

[0203] The system's performance prediction module is centered around a mathematical model of slurry performance deployed on an industrial server. This module leverages the server's large-capacity storage resources to manage massive amounts of historical slurry preparation and performance data, and interacts with other modules via industrial Ethernet. Its function is to receive real-time and historical data from the geological sensing and online monitoring modules, and output accurate predictions of key slurry performance parameters.

[0204] The geological sensing module consists of a network of various sensors and data acquisition units deployed at the tunnel boring machine (TBM) and tunnel site. This module integrates TBM tunneling parameter acquisition units, an advanced geological prediction system interface, a circumferentially deployed array of vibrating wire pore water pressure sensors, a soil chamber pressure sensor group, and TBM propulsion cylinder pressure sensors to comprehensively acquire raw information such as stratum lithology, groundwater distribution, and surrounding rock pressure status. The acquired raw data undergoes preliminary processing via a programmable logic controller (PLC) or remote input / output module, performing data cleaning, filtering, and normalization operations. Finally, a unified, standardized set of geological parameters is uploaded to the system network via a fieldbus protocol.

[0205] The intelligent control module uses a high-performance embedded industrial computer or industrial control computer as its hardware platform and runs a real-time operating system. This module incorporates advanced control algorithms and establishes connections with the performance prediction module and geological sensing module via standard industrial communication protocols to acquire slurry performance prediction results and on-site geological parameters in real time. After calculation using built-in optimization algorithms, the module generates control instructions containing target values ​​for particle size distribution, moisture content, and admixture dosage, and sends these instructions to the execution layer via real-time Ethernet.

[0206] The dynamic execution module consists of a series of controlled field actuators and their local control system. This module includes a particle size distribution control unit, a moisture content control unit, and an admixture dosage control unit. The particle size distribution control unit adjusts the feeding ratio and rate of coarse and fine aggregates via a variable frequency speed-regulating belt scale and its local proportional-integral-derivative controller. The moisture content control unit precisely controls the pipeline water injection volume via an electric regulating valve and its positioner. The admixture dosage control unit achieves precise control of the admixture injection rate and total amount via a high-precision metering pump and its variable frequency drive.

[0207] The online monitoring and feedback module consists of dedicated online monitoring instruments installed at the discharge port of the mixing plant and on the grouting pipeline. This module integrates equipment such as an online fluidity monitor, an online setting time analyzer, and an online strength development predictor. It transmits continuously collected real-time performance indicators of the slurry to the system data platform via standard analog signals or an Ethernet interface. This real-time data provides crucial feedback for parameter correction of the performance prediction model and for system adaptive optimization.

[0208] In addition, the system's intelligent control module integrates a performance and safety boundary monitoring submodule. This submodule is linked with the audible and visual alarm device in the central control room, the emergency stop circuit of the grouting system, and the human-machine interface terminal for coordinated control. When the system detects that any real-time performance indicator exceeds the preset safety threshold, it will automatically trigger a multi-level interlocking response, including activating the audible and visual alarm, executing the grouting pause command, locking the control signal, and switching to manual control mode. At the same time, it will automatically generate a detailed fault diagnosis report, thereby comprehensively ensuring the safety and quality of the construction process.

[0209] Those skilled in the art will understand that there are multiple options for the specific hardware implementation of the above modules, and different models of controllers, sensors and actuators can be configured according to the actual engineering project requirements. All hardware combinations that can achieve the functions described in this invention are within the protection scope of this invention.

[0210] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process and related descriptions of the system described above can be found in the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0211] It should be noted that the intelligent preparation and control system for grouting behind the shield tunnel wall provided in the above embodiments is only an example of the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the modules or steps in the embodiments of the present invention can be further decomposed or combined. For example, the modules in the above embodiments can be merged into one module, or further divided into multiple sub-modules to complete all or part of the functions described above. The names of the modules and steps involved in the embodiments of the present invention are only for distinguishing the various modules or steps and are not considered as an improper limitation of the present invention.

[0212] A device according to a third embodiment of the present invention includes:

[0213] At least one processor;

[0214] and a memory communicatively connected to at least one of the processors;

[0215] The memory stores instructions that can be executed by the processor to implement the above-mentioned intelligent preparation and control method for grouting fluid behind the shield wall.

[0216] A computer-readable storage medium according to a fourth embodiment of the present invention stores computer instructions, which are executed by the computer to implement the above-described intelligent preparation and control method for grouting fluid behind the shield wall.

[0217] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process and related descriptions of the storage device and processing device described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0218] The following is for reference. Figure 4 It shows a schematic diagram of the structure of a computer system for implementing embodiments of the systems, methods, and electronic devices of this application. Figure 4 The server shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0219] like Figure 4As shown, the computer system includes a Central Processing Unit (CPU) 401, which can perform various appropriate actions and processes based on programs stored in Read Only Memory (ROM) 402 or programs loaded from storage section 408 into Random Access Memory (RAM) 403. RAM 403 also stores various programs and data required for system operation. The CPU 401, ROM 402, and RAM 403 are interconnected via bus 404. Input / output (I / O) interface 405 is also connected to bus 404.

[0220] The following components are connected to I / O interface 405: an input section 406 including a keyboard, mouse, etc.; an output section 407 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and speakers, etc.; a storage section 408 including a hard disk, etc.; and a communication section 409 including a network interface card such as a LAN (Local Area Network) card, modem, etc. The communication section 409 performs communication processing via a network such as the Internet. A drive 410 is also connected to I / O interface 405 as needed. A removable medium 411, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on drive 410 as needed so that computer programs read from it can be installed into storage section 408 as needed.

[0221] Specifically, according to embodiments of this disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this disclosure include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 409, and / or installed from removable medium 411. When the computer program is executed by central processing unit (CPU) 401, it performs the functions defined in the methods of this application. It should be noted that the computer-readable medium described above in this application can be a computer-readable signal medium or a computer-readable storage medium, or any combination of the two. A computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this application, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in connection with an instruction execution system, apparatus, or device. In this application, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can also be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on a computer-readable medium can be transmitted using any suitable medium, including but not limited to: wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0222] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0223] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0224] The terms “first”, “second”, etc., are used to distinguish similar objects, not to describe or indicate a specific order or sequence.

[0225] The term "comprising" or any other similar term is intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus / device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent in such process, method, article, or apparatus / device.

[0226] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

Claims

1. A shield wall back grouting slurry intelligent preparation control method, characterized in that, The method comprises: establishing a slurry performance model for predicting the finished product performance parameters of the slurry in the grouting construction by analyzing the physical and chemical properties of the slurry raw materials and the operating variables in the preparation process, the finished product performance parameters including the fluidity, setting time and strength development characteristics of the slurry; collecting real-time geological parameters of the construction site, the geological parameters including the stratum lithology, groundwater distribution and surrounding rock pressure state, and preprocessing the geological parameters to eliminate noise and outliers; generating a control signal based on the preprocessed geological parameters and the predicted output of the slurry performance model based on a control algorithm; adjusting the particle size distribution, moisture content and admixture content of the semi-finished slurry dynamically according to the control signal; continuously monitoring the real-time performance indicators of the slurry during the preparation and grouting process of the slurry, and feeding the monitoring data back to the slurry performance model for correcting the parameters of the slurry performance model; wherein the control signal is generated by a control algorithm, which comprises: determining the slurry performance target values suitable for the current geological conditions according to the preprocessed geological parameters, the performance target values including the target fluidity, target setting time and target strength; constructing a multivariable objective function with the optimization of slurry performance as the goal, the objective function being based on the weighted combination of the deviations of the predicted values of slurry fluidity, setting time and strength development from the target values; wherein the predicted values are output by the slurry performance model; setting the constraint conditions of the process variables, including the particle size distribution range, moisture content range and admixture content range; in each control cycle, taking the current process variable set value as the starting point, calculating a sequence of future optimal process variable set values by solving an optimization problem with the minimization of the multivariable objective function as the goal and subject to the constraint conditions; selecting the instruction of the current control cycle from the optimal set value sequence to generate a real-time adjustment signal as the control signal; the step of calculating the optimal set value sequence of the process variables by solving the constrained optimization problem, comprising: Step A, taking a set of initial process variable set values satisfying the constraint conditions of the process variables as the starting point, and performing iterative solving; Step B, in each iteration, performing the following steps: based on the process variable set value of the current iteration, constructing a quadratic programming sub-problem by performing quadratic approximation of the multivariable objective function at the current set value point, and linearizing the nonlinear process variable constraint conditions at the current set value point to achieve; solving the quadratic programming sub-problem to obtain a search direction vector defining the adjustment direction and proportion of the process variables; performing one-dimensional line search along the search direction vector to determine an optimal step size factor by evaluating a value function containing the original objective function and a penalty term for violating the original nonlinear constraint conditions; updating the process variable set value according to the search direction vector and the optimal step size factor to generate a new set value for the next iteration; Step C, repeating the iteration process of step B until the variation of the process variable setting value is less than a first preset convergence threshold, or the norm of the search direction vector is less than a second preset convergence threshold; Step D, outputting the final process variable setting value obtained at the end of iteration as the optimal setting value sequence.

2. The method of claim 1, wherein, The establishment of the slurry performance model comprises: Collecting historical slurry preparation data, including physical property data, chemical composition data, ratio relationship between raw materials, stirring equipment operation parameters, environmental temperature and humidity data, and corresponding finished slurry performance test results; Pretreating the collected data, including data cleaning to remove outliers, data normalization to eliminate dimension influence, and data standardization; Based on the pretreated data, a data-driven relationship network is constructed, which is a neural network, and the historical data is trained to learn the nonlinear relationship between slurry attributes and performance parameters; The neural network is trained by a back propagation algorithm, and the network weight is optimized by a gradient descent method; In the model deployment application stage, new slurry preparation process data and performance test data are continuously integrated, the neural network parameters are dynamically adjusted by incremental learning, and the model prediction accuracy is periodically verified. When the prediction deviation exceeds a first preset threshold, the model reconstruction is triggered.

3. The method of claim 1, wherein, The collection of real-time geological parameters of the construction site comprises: Obtaining stratum lithology data, underground water distribution data and surrounding rock pressure state data; Pretreating the geological parameters, including data cleaning to remove outliers, data normalization to eliminate dimension influence, and data fusion to integrate multi-source sensor information into a unified geological parameter set; The pretreated geological parameters are used as inputs of the slurry performance model and the control algorithm.

4. The method of claim 1, wherein, Dynamic adjustment of the particle size distribution, moisture content and admixture dosage of the semi-finished slurry according to the control signal comprises: Resolving the control signal into a particle size distribution adjustment signal, a moisture content adjustment signal and an admixture dosage adjustment signal; Adjusting the particle size distribution of the slurry based on the particle size distribution adjustment signal, adjusting the moisture content of the slurry based on the moisture content adjustment signal, and adjusting the admixture dosage based on the admixture dosage adjustment signal; Real-time monitoring of the actual values of the particle size distribution, moisture content and admixture dosage, and comparison with the target values of the particle size distribution, moisture content and admixture dosage. When the deviation exceeds the allowable range, the control signal is recalculated and the adjustment operation is adjusted.

5. The method of claim 2, wherein, Continuous monitoring of real-time performance indicators of the slurry and feedback of the monitoring data to the slurry performance model comprises: Obtaining real-time data of the flowability, setting time and strength indicators of the slurry through online monitoring; Comparing the monitored real-time data with the predicted values of the slurry performance model to calculate the deviation; Adjusting the internal parameters of the slurry performance model according to the deviation; Calibrating the geological parameters and the control algorithm using the feedback data to form an adaptive cycle.

6. The method of claim 5, wherein, The method for adjusting the internal parameters of the slurry performance model comprises: Based on the deviation, the weight and bias parameters of the neural network are dynamically adjusted by incremental learning; Periodically verifying the model prediction accuracy, and when the deviation exceeds a set threshold, triggering the reconstruction process of the slurry performance model.

7. The method of claim 4, wherein, The control signal is parsed into a particle size distribution adjustment signal, a moisture content adjustment signal and an additive dosage adjustment signal, and the method is as follows: The target values corresponding to the particle size distribution, moisture content and additive dosage respectively are read from the control signal; The particle size distribution target value is compared with the actually detected particle size distribution actual value, and the particle size distribution adjustment signal is generated according to the difference between them; The moisture content target value is compared with the actually detected moisture content actual value, and the moisture content adjustment signal is generated according to the difference between them; The additive dosage target value is compared with the actually detected additive dosage actual value, and the additive dosage adjustment signal is generated according to the difference between them.

8. The method of claim 4, wherein, The slurry particle size distribution is adjusted based on the particle size distribution adjustment signal, the slurry moisture content is adjusted based on the moisture content adjustment signal, and the additive dosage is adjusted based on the additive dosage adjustment signal, and the method is as follows: The particle size distribution adjustment signal is converted into a first control instruction for controlling the discharging actuators of the coarse aggregate bin and the fine aggregate bin, and the particle size distribution of the slurry is controlled by adjusting the discharging proportion and rate of the coarse aggregate and the fine aggregate; The moisture content adjustment signal is converted into a second control instruction for controlling the water supply pipeline regulating valve, and the moisture content of the slurry is controlled by accurately adjusting the water injection amount; The additive dosage adjustment signal is converted into a third control instruction for controlling the additive metering pump, and the additive dosage is controlled by accurately controlling the injection rate and total amount of the additive.

9. The method of claim 1, wherein, It also includes a performance safety boundary monitoring step: A safety threshold range is set for the real-time performance indicators of the slurry; During continuous monitoring, when any real-time performance indicator exceeds its corresponding safety threshold range, an alarm is triggered and a preset safety strategy is executed; The safety strategy includes suspending the grouting construction, locking the current control signal and switching to a manual control mode, and generating a fault diagnosis report.

Images