Machine learning based prediction and optimization method for shield muck modifier

Abstract

The application provides a shield muck modifier prediction and optimization method based on machine learning, comprising: the i-ring data comprises i-ring tunneling data, i-ring geological data and i-ring modifier data, the i-ring cutter penetration, the i-ring torque cut depth index and the i-ring field cut depth index are calculated according to the i-ring tunneling data, and the data set is divided into a good data set and a poor data set through the i-ring torque cut depth index and the i-ring field cut depth index; an XGBoost prediction model is constructed, the good data set is used to train the XGBoost prediction model to obtain a trained XGBoost prediction model; the poor data set is input into the trained XGBoost prediction model to obtain corresponding optimized modifier parameters and an optimized cutter torque. Through considering the influence of the shield tunneling data and the geological data on the muck modifier dosage, the established modifier prediction model can realize real-time dynamic decision-making with the change of stratum conditions and working conditions.

Description

Technical Field

[0001] This invention relates to the field of soil improvement technology in earth pressure balance shield tunneling, and more particularly to a method for predicting and optimizing soil improvement agents for shield tunneling based on machine learning. Background Technology

[0002] Earth pressure balance (EPB) tunnel boring machines (TBMs) are widely used in urban tunnel construction due to their high degree of mechanization and minimal impact on the surrounding environment. During tunneling, the EPB's cutterhead rotates and cuts the soil in front. The excavated material passes through the cutterhead opening into the soil chamber, where it transmits pressure from the jacks to the tunnel face to balance the water and soil pressure ahead. A screw conveyor transports the excavated material to the surface, simultaneously adjusting the balance between the amount of excavated and discharged soil. However, due to the variability of engineering geology, safety hazards exist during TBM tunneling, such as cutterhead cutting tool "mud cake" formation, tool wear, and screw conveyor outlet gushing. To facilitate soil discharge and control soil chamber pressure, and improve tunneling efficiency, it is often necessary to modify the excavated material by injecting a modifier into the area in front of the cutterhead and into the soil chamber, giving the excavated material good fluidity and plasticity. Commonly used soil conditioners for sandy soil include foaming agents and bentonite. Foaming agents can improve the flowability of excavated soil and are applicable to various soil types; bentonite can improve the cohesion of excavated soil and is suitable for soil types lacking fine particles. Determining the dosage of soil conditioner is a key aspect of tunnel boring machine (TBM) excavated soil improvement.

[0003] Currently, the experimental method is the primary means of determining soil amendment schemes, providing references for soil amendment schemes in various strata such as water-rich sand, gravel, and clay. Its drawbacks include the need for multiple sets of soil amendment tests for different geological conditions in areas with complex and varied geological conditions, resulting in a large workload and long testing time; furthermore, the stress on shield tunnel excavated soil is complex, and existing test conditions are difficult to simulate real working conditions. Therefore, it is crucial to develop a method for predicting the dosage of soil amendments that matches the geological strata. In recent years, machine learning has been widely applied in the field of soil amendment.

[0004] Some scholars have extracted information from multiple sets of slag soil improvement experiments. Zhan Chao, Zhang Yan, and others used Relevance Vector Machine (RVM) and Backpropagation Neural Network to predict the improved internal friction angle and permeability coefficient using the original slag soil's permeability coefficient, internal friction angle, resistivity, and the amounts of foaming agent and bentonite. Zhang Wentao established a random forest model, using slag soil moisture content, foam injection ratio, bentonite slurry injection ratio, effective soil particle size, and hydraulic gradient as input parameters to predict the permeability coefficient of shield tunneling slag soil. Based on the predicted permeability coefficient, the presence of a blowout risk on site can be determined. These methods only validate the effectiveness of the experimental methods and do not consider the mapping relationship between shield tunneling data and slag soil mechanical parameters, thus failing to provide a reference for determining the dosage of amendments under variable geological conditions.

[0005] Chinese patent application number 201910445028.9 proposes an automatic improvement method for excavated soil in earth pressure balance shield tunneling. A camera is installed at the tail of the screw conveyor to determine the fluidity and plasticity of the excavated soil through images, and the results are fed back to the shield machine's operating platform in real time, allowing the system to automatically adjust the improvement parameters. Problems with this method include: a large number of images consume significant memory, leading to increased equipment consumption; and in cases of abnormal improvement results, the obtained image categories may be unbalanced, potentially causing decision-making errors.

[0006] Chinese patent application number 202110297054.9 proposes a method for improving muck quality in earth pressure balance shield tunneling machines by combining GBDT and random forest algorithms. The method employs GBDT to select features from geological data; obtains key muck quality parameters based on tunneling data; establishes N prediction models; predicts the amount of amendment using geological data and key tunneling data; and compares and selects the optimal prediction model to predict the amount of foaming agent to be used. The method has the following problems: it does not use data with "better" improvement effects to train the model; if the training set contains data with amendment effects that are poor, it will affect the accuracy of the obtained predictions.

[0007] In view of this, in order to predict the amount of soil amendment with good improvement effect, the model should be trained using data with "good effect". To this end, it is necessary to first determine an evaluation method for the soil amendment effect. For the amendment data with "poor effect", a machine learning model should be used for optimization to obtain a relatively good amount of amendment applicable to different strata, thereby further improving tunneling efficiency. Summary of the Invention

[0008] To address the aforementioned technical problem of ineffective soil amendment in existing technologies, this invention provides a machine learning-based method for predicting and optimizing soil amendments used in tunnel boring machines (TBMs). This invention fully explores the inherent relationships between tunneling data, geological data, and amendment data, proposes a method for evaluating amendment effectiveness, establishes a machine learning prediction model, and enables dynamic decision-making regarding amendment dosage.

[0009] The technical means employed in this invention are as follows:

[0010] This invention provides a machine learning-based method for predicting and optimizing soil conditioner for tunnel boring machines, comprising:

[0011] Obtain the database, preprocess the database to obtain the dataset, the dataset includes data from the first ring to the Nth ring, the data of the i-th ring includes the tunneling data of the i-th ring, the geological data of the i-th ring and the amendment data of the i-th ring, the tunneling data of the i-th ring includes the cutterhead torque of the i-th ring, the total propulsion force of the i-th ring, the cutterhead rotation speed of the i-th ring, and the tunneling speed of the i-th ring, N>1 and N is an integer, 1≤i≤N and i is an integer;

[0012] The cutterhead penetration of the i-th ring is calculated based on the cutterhead rotation speed and the tunneling speed of the i-th ring. The torque cutting depth index and the field cutting depth index of the i-th ring are calculated based on the cutterhead penetration of the i-th ring.

[0013] Based on the standard fitting curve of torque depth of cutter penetration, it is determined whether the torque depth of cutter penetration of the i-th ring is greater than the standard torque depth of cutter penetration of the i-th ring under the condition of cutter penetration of the i-th ring. Based on the standard fitting curve of field depth of cutter penetration, it is determined whether the field depth of cutter penetration of the i-th ring is greater than the standard field depth of cutter penetration of the i-th ring under the condition of cutter penetration of the i-th ring.

[0014] If the torque shear depth index of the i-th ring is less than or equal to the standard torque shear depth index of the i-th ring, and the field shear depth index of the i-th ring is less than or equal to the standard field shear depth index of the i-th ring, the i-th ring data is superior data; if the torque shear depth index of the i-th ring is greater than the standard torque shear depth index of the i-th ring, and / or, the field shear depth index of the i-th ring is greater than the standard field shear depth index of the i-th ring, the i-th ring data is inferior data, the superior data constitutes a superior dataset, and the inferior data constitutes a inferior dataset;

[0015] Construct an XGBoost prediction model and train the XGBoost prediction model using the superior dataset to obtain the trained XGBoost prediction model.

[0016] The differential dataset is input into the trained XGBoost prediction model to obtain the corresponding optimized modifier parameters and optimized cutter head torque.

[0017] Further, the database is preprocessed to obtain the dataset, including:

[0018] The database includes subsets 1 to N. The i-th subset includes i-th tunneling data, i-th geological data, and i-th amendment data. The i-th tunneling data includes i-th cutterhead rotation speed dataset, i-th tunneling speed dataset, i-th total thrust dataset, i-th cutterhead torque dataset, i-th soil chamber pressure dataset, i-th auger rotation speed dataset, and i-th auger torque dataset.

[0019] Data with a cutterhead rotation speed of 0, data with a tunneling speed of 0, and data with a total thrust of 0 in the i-th cutterhead rotation speed dataset, the i-th tunneling speed dataset, and the i-th total thrust dataset are removed. Outliers in these datasets are then removed using a box plotter algorithm to obtain subsets of the i-th cutterhead rotation speed, i-th tunneling speed, i-th total thrust, i-th cutterhead torque, i-th soil chamber pressure, i-th auger speed, and i-th auger torque.

[0020] The average value of the subset of i-th cutterhead rotational speeds is used to obtain the i-th ring cutterhead rotational speed; the average value of the subset of i-th tunneling speeds is used to obtain the i-th ring tunneling speed; the average value of the subset of i-th total propulsion force is used to obtain the i-th ring total propulsion force; the average value of the subset of i-th cutterhead torques is used to obtain the i-th ring cutterhead torque; the average value of the subset of i-th soil chamber pressures is used to obtain the i-th ring soil chamber pressure; the average value of the subset of i-th auger speeds is used to obtain the i-th ring auger speed; and the average value of the subset of i-th auger torques is used to obtain the i-th ring auger torque. The i-th ring cutterhead rotational speed, the i-th ring tunneling speed, the i-th ring total propulsion force, the i-th ring cutterhead torque, the i-th ring soil chamber pressure, the i-th ring auger speed, and the i-th ring auger torque constitute the i-th ring tunneling data.

[0021] The i-th geological data includes the i-th burial depth, i-th natural density, i-th deformation modulus, i-th permeability coefficient, i-th cohesion, i-th internal friction angle, i-th standard penetration test blow count, i-th dynamic elastic modulus, and i-th dynamic shear modulus. The weighted average of the i-th geological parameters is calculated based on the i-th natural density, i-th deformation modulus, i-th permeability coefficient, i-th cohesion, i-th internal friction angle, i-th standard penetration test blow count, i-th dynamic elastic modulus, and i-th dynamic shear modulus. The i-th burial depth and the weighted average of the i-th geological parameters constitute the i-th ring geological data.

[0022] The i-th modifier data includes the i-th ring foam stock solution volume dataset, the i-th ring foaming agent solution concentration dataset, the i-th ring bentonite slurry volume dataset, and the i-th ring bentonite slurry concentration dataset. The i-th ring foam stock solution volume is obtained by calculating the cumulative value of the i-th ring foam stock solution volume dataset, the i-th ring bentonite slurry volume is obtained by calculating the cumulative value of the i-th ring bentonite slurry volume dataset, the i-th ring foaming agent solution concentration is obtained by calculating the average value of the i-th ring foaming agent solution concentration dataset, and the i-th ring bentonite slurry concentration is obtained by calculating the average value of the i-th ring bentonite slurry concentration dataset. The i-th ring foam stock solution volume, the i-th ring bentonite slurry volume, the i-th ring foaming agent solution concentration, and the i-th ring bentonite slurry concentration constitute the i-th ring modifier data.

[0023] Furthermore, the superior dataset is divided into a training set and a test set. When training the XGBoost prediction model, the training set is input, and the hyperparameters of the XGBoost prediction model are optimized using the Optuna algorithm to obtain the trained XGBoost prediction model. The hyperparameters include the maximum depth of the tree, the minimum leaf weight, the parameters of the L1 regularization term, the parameters of the L2 regularization term, and the learning rate.

[0024] Further, the test set is input into the trained XGBoost prediction model to obtain predicted values, a prediction accuracy evaluation index is calculated based on the predicted values, and the prediction accuracy of the trained XGBoost prediction model is evaluated based on the prediction accuracy evaluation index.

[0025] Further, inputting the differential dataset into the trained XGBoost prediction model includes:

[0026] The XGBoost prediction model is trained by inputting the cutterhead rotation speed, tunneling speed, total propulsion force, soil chamber pressure, auger rotation speed, auger torque, and geological data of the i-th ring from the differential dataset. The trained XGBoost prediction model outputs optimized values, which include optimized amendment data and optimized cutterhead torque of the i-th ring.

[0027] Furthermore, after outputting the optimized value, the method also includes calculating the tunneling specific energy and the optimized tunneling specific energy. If the optimized tunneling specific energy is less than the tunneling specific energy, the trained XGBoost prediction model performs well; if the optimized tunneling specific energy is greater than or equal to the tunneling specific energy, the trained XGBoost prediction model performs poorly.

[0028] Furthermore, the tunneling specific energy and the optimized tunneling specific energy are calculated in the following manner:

[0029]

[0030] Among them, E S Let E′ be the tunneling specific energy. S For the optimized tunneling specific energy, N i T is the rotational speed of the i-th ring cutter head. i Let V be the torque of the i-th ring cutter head, D be the diameter of the cutter head, and V be the torque of the ring cutter head. i Let T′ be the tunneling speed of the i-th ring. i To optimize the torque of the i-th ring cutter head.

[0031] Further, the penetration depth of the i-th ring cutterhead is calculated based on the rotational speed of the i-th ring cutterhead and the tunneling speed of the i-th ring, in the following manner:

[0032]

[0033] Wherein, the S i Let N be the penetration depth of the i-th ring cutterhead. i V is the rotational speed of the i-th ring cutter head. i Let be the tunneling speed of the i-th ring.

[0034] Further, the torque depth index and the field depth index of the i-th ring are calculated based on the penetration depth of the i-th ring cutter head, and are calculated in the following manner:

[0035]

[0036] Among them, TPI i FPI is the torque depth index of the i-th ring. i Let T be the cutting depth index of the i-th ring field. i Let F be the torque of the i-th ring cutter head. i Let be the total thrust of the i-th ring.

[0037] Compared with the prior art, the present invention has the following advantages:

[0038] 1. The shield tunneling soil amendment prediction and optimization method based on machine learning provided by this invention calculates the torque depth index (TPI), field depth index (FPI), and penetration depth (S) based on the cutterhead torque, total thrust, cutterhead rotation speed, and tunneling speed. According to the TPI-S and FPI-S curves of different strata, the method divides the amendment effect into "good" and "bad" datasets. Compared with the traditional on-site manual recording method, this method saves manpower and avoids the influence of human subjectivity on the results, making the results more reliable.

[0039] 2. The machine learning-based method for predicting and optimizing soil conditioner for tunnel boring machines (TBMs) provided by this invention comprehensively considers the influence of TBM tunneling data and geological data on the amount of soil conditioner used. The established conditioner prediction model can make real-time dynamic decisions as the geological conditions and working conditions change, providing a reference for TBM construction.

[0040] 3. The machine learning-based method for predicting and optimizing the soil conditioner for tunnel boring machines provided by this invention uses the completed soil conditioner prediction model on a dataset with "poor" results to optimize the output values ​​(soil conditioner data, cutterhead torque) and output the amount of soil conditioner to adapt to the strata and tunnel boring machine conditions, thereby reducing the energy consumption of the tunnel boring machine. Attached Figure Description

[0041] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0042] Figure 1 This is a schematic flowchart of a method for predicting and optimizing tunnel boring machine excavation soil conditioner based on machine learning, provided by the present invention.

[0043] Figure 2 This is a TPI-S curve.

[0044] Figure 3 This is an FPI-S curve.

[0045] Figure 4 This is a box-shaped diagram.

[0046] Figure 5 This is a schematic diagram of the stratigraphic profile at the working face.

[0047] Figure 6 This is a diagram illustrating the input parameters and their corresponding output parameters. Detailed Implementation

[0048] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0049] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0050] Reference Figure 1 , Figure 2 and Figure 3 , Figure 1 This is a schematic flowchart of a machine learning-based method for predicting and optimizing tunnel boring machine (TBM) slag amendments provided by the present invention. Figure 2 This is a TPI-S curve. Figure 3 Using an FPI-S curve to illustrate a specific embodiment of the machine learning-based method for predicting and optimizing tunnel boring machine (TBM) slag amendments, this embodiment includes:

[0051] Obtain the database, preprocess the database to obtain the dataset, the dataset includes data from ring 1 to ring N, the data of ring i includes the tunneling data of ring i, the geological data of ring i and the amendment data of ring i, the tunneling data of ring i includes the cutterhead torque of ring i, the total propulsion force of ring i, the cutterhead rotation speed of ring i, and the tunneling speed of ring i, N>1 and N is an integer, 1≤i≤N and i is an integer;

[0052] The cutterhead penetration of the i-th ring is calculated based on the cutterhead rotation speed and the tunneling speed of the i-th ring. The torque cutting depth index and the field cutting depth index of the i-th ring are also calculated based on the cutterhead penetration of the i-th ring.

[0053] Specifically, the penetration depth of the i-th ring cutterhead is calculated based on the rotational speed and tunneling speed of the i-th ring cutterhead, as follows:

[0054]

[0055] Among them, S i Let N be the penetration depth of the i-th ring cutterhead. i V is the rotational speed of the i-th ring cutter head. i Let be the tunneling speed of the i-th ring.

[0056] The torque depth index and the field depth index of the i-th ring are calculated based on the penetration depth of the i-th ring cutter head, as follows:

[0057]

[0058] Among them, TPI i FPI is the torque depth index of the i-th ring. i Let T be the cutting depth index of the i-th ring field. i Let F be the torque of the i-th ring cutter head. i This represents the total propulsion force of the i-th ring.

[0059] Based on the standard fitting curve of torque depth of cutter penetration, determine whether the torque depth of cutter penetration of ring i is greater than the standard torque depth of cutter penetration of ring i. Based on the standard fitting curve of field depth of cutter penetration, determine whether the field depth of cutter penetration of ring i is greater than the standard field depth of cutter penetration of ring i.

[0060] If the torque shear depth index of the i-th ring is less than or equal to the standard torque shear depth index of the i-th ring, and the field shear depth index of the i-th ring is less than or equal to the standard field shear depth index of the i-th ring, the data of the i-th ring is excellent data; if the torque shear depth index of the i-th ring is greater than the standard torque shear depth index of the i-th ring, and / or, the field shear depth index of the i-th ring is greater than or equal to the standard field shear depth index of the i-th ring, the data of the i-th ring is poor data. The excellent data constitutes the excellent dataset, and the poor data constitutes the poor dataset.

[0061] Understandably, penetration depth reflects the ease or difficulty of tunneling through strata; a large penetration depth indicates better excavability, while a small penetration depth indicates poorer excavability. Literature indicates that torque depth index and field depth index can indirectly reflect the effectiveness of soil improvement. The torque depth index is the ratio of cutterhead torque to penetration depth, while the field depth index is the ratio of total thrust to penetration depth; both reflect the ground resistance during tunneling.

[0062] Figure 2 and Figure 3 Taking the 0-442 loops of the same route as an example, refer to Figure 2 The x-axis of the standard fitting curve of torque depth of cutter penetration is penetration, and the y-axis is the standard torque depth of cutter penetration. When the coordinates of the i-th ring cutter penetration and the i-th ring torque depth of cutter penetration are above the standard fitting curve of torque depth of cutter penetration, the soil improvement effect is poor and it cannot be used for subsequent training. Similarly, referring to... Figure 3 The horizontal axis of the field cutting depth index-cutter head penetration standard fitting curve is penetration, and the vertical axis is the standard torque cutting depth index. When the coordinates of the i-th ring cutter head penetration and the i-th ring standard torque cutting depth index are above the field cutting depth index-cutter head penetration standard fitting curve, it means that the soil improvement effect is poor and cannot be used for subsequent training. In this way, the dataset can be divided into excellent datasets and poor datasets, avoiding the use of data with poor improvement effect for model training, which would affect the accuracy of the model.

[0063] Construct an XGBoost prediction model and train it using a superior dataset to obtain the trained XGBoost prediction model.

[0064] It is understandable that the XGBoost prediction model is an open-source gradient boosting framework, belonging to the ensemble decision tree model. Its advantages include: it can quickly process large-scale data, it can learn autonomously, it can automatically optimize split nodes, and it can effectively handle data with a large number of outliers and missing values.

[0065] The XGBoost prediction model is calculated using the following method:

[0066] The linear correlation between any two input parameters is analyzed using the Pearson correlation coefficient, and the input parameters are dimensionality reduced using principal component analysis.

[0067] It is understandable that the correlation coefficient is a measure of the degree of correlation between variables and can be used as a basis for feature selection. This invention uses the Pearson correlation coefficient method to perform linear correlation analysis on the initially selected input features. The correlation coefficient ranges from -1 to 1, with positive values ​​indicating positive correlation and negative values ​​indicating negative correlation. The closer the absolute value is to 1, the higher the correlation between the two variables; the closer it is to 0, the weaker the correlation between the two variables. The correlation strengths corresponding to the correlation coefficients are as follows: 0.8-1.0 extremely strong correlation, 0.6-0.8 strong correlation, 0.4-0.6 moderate correlation, 0.2-0.4 correlation, 0.0-0.2 extremely weak correlation or no correlation. To avoid redundant features affecting the model's computational efficiency, principal component analysis (PCA) is used to reduce the dimensionality of the initially selected features. The PCA algorithm uses linear projection to map data from high-dimensional to low-dimensional space, maximizing variance and constructing a new coordinate space. Orthogonal features in this new space are called principal components. PCA replaces old features with fewer new ones, achieving dimensionality reduction and effectively improving model computational efficiency. Using a cumulative variance contribution rate of 0.95 as the standard, the top n principal components with a cumulative variance contribution rate of 0.95 are selected as the final input to the model, discarding data with low variance contribution rates. The input layer parameters after dimensionality reduction are updated to P = {principal component 1, principal component 2, ..., principal component n}; the output layer parameters remain: foam concentrate volume, foaming agent solution concentration, bentonite slurry volume, bentonite slurry concentration, and cutterhead torque.

[0068] The prediction expression for the XGBoost prediction model is constructed and calculated as follows:

[0069]

[0070] Where, x i For the input data of the i-th ring, f kFor the basic learner model of each regression tree, This is the prediction result for the i-th ring;

[0071] The objective function is constructed based on the prediction expression. The objective function consists of two parts: a loss function and a regularization term, and is calculated as follows:

[0072]

[0073] Among them, y i This is the actual value of the i-th ring. Let Ω(f) be the loss function, representing the difference between the predicted score and the true score. k ) represents the regularization term, used to reduce the complexity of the objective function, and K represents the number of model iterations.

[0074] Perform a second-order Taylor expansion on the objective function and find its minimum value;

[0075] Update leaf node values: Obtain the corresponding leaf node values ​​based on the minimum value of the objective function;

[0076] Update the sample predictions based on the new leaf node values, average the predictions across all trees, and output the prediction results.

[0077] The inferior dataset is input into the trained XGBoost prediction model to obtain the corresponding optimized modifier parameters and optimized tool head torque. When the test set is input into the trained XGBoost prediction model, the prediction result output by the trained XGBoost prediction model is the predicted value; when the inferior dataset is input into the trained XGBoost prediction model, the prediction result output by the trained XGBoost prediction model is the optimized value.

[0078] In some alternative embodiments, refer to Figure 4 and Figure 5 , Figure 4 This is a box-shaped diagram. Figure 5 This is a schematic diagram of the stratigraphic profile at the tunnel face. The dataset is obtained by preprocessing the database, including:

[0079] The database includes subsets 1 to N. The i-th subset includes the i-th tunneling data, the i-th geological data, and the i-th amendment data. The i-th tunneling data includes the i-th cutterhead speed dataset, the i-th tunneling speed dataset, the i-th total thrust dataset, the i-th cutterhead torque dataset, the i-th soil chamber pressure dataset, the i-th screw conveyor speed dataset, and the i-th screw conveyor torque dataset.

[0080] Remove data where the cutterhead speed is 0 from the i-th cutterhead speed dataset, data where the tunneling speed is 0 from the i-th tunneling speed dataset, and data where the total thrust is 0 from the i-th total thrust dataset. Use a box plot to remove outliers from the i-th cutterhead speed dataset, i-th tunneling speed dataset, i-th total thrust dataset, i-th cutterhead torque dataset, i-th soil chamber pressure dataset, i-th auger speed dataset, and i-th auger torque dataset to obtain subsets of i-th cutterhead speed, i-th tunneling speed, i-th total thrust, i-th cutterhead torque, i-th soil chamber pressure, i-th auger speed, and i-th auger torque.

[0081] Understandably, the data acquisition system of the tunnel boring machine (TBM) continuously reads and stores tunneling data, with a data acquisition frequency of approximately once every 10 seconds. The data volume is extremely large, with high repetition within the same segment and the presence of invalid information. Each segment requires standby for segment installation. During standby, parameters such as cutterhead torque, tunneling speed, and thrust are all zero, offering no predictive value. Therefore, this data needs to be removed, specifically the data where cutterhead torque, tunneling speed, and thrust are zero. Secondly, mechanical failures or human error are inevitable during TBM tunneling, leading to outliers. Inputting these outliers into the model will affect the model's fitting accuracy; therefore, outliers need to be screened and removed. Common outlier removal methods include box plots and the 3σ principle. Compared to the 3σ principle, box plots do not require adherence to a normal distribution and have a wider range of applications. Therefore, box plots are used to handle outliers. Figure 4 As shown.

[0082] The average value of the tunneling data for each ring is taken. The average value of the subset of cutterhead speeds of the i-th ring is obtained by calculating the average value of the subset of tunneling speeds of the i-th ring, the average value of the subset of tunneling speeds of the i-th ring is obtained by calculating the average value of the subset of total thrust of the i-th ring, the average value of the subset of cutterhead torques of the i-th ring is obtained by calculating the average value of the subset of soil chamber pressures of the i-th ring, the average value of the subset of auger speeds of the i-th ring is obtained by calculating the average value of the subset of auger torques of the i-th ring, and the average value of the subset of auger torques of the i-th ring is obtained by calculating the average value of the subset of auger torques of the i-th ring. The cutterhead speed of the i-th ring, the tunneling speed of the i-th ring, the total thrust of the i-th ring, the cutterhead torque of the i-th ring, the soil chamber pressure of the i-th ring, the auger speed of the i-th ring, and the auger torque of the i-th ring constitute the tunneling data for the i-th ring.

[0083] The i-th geological data includes the i-th burial depth, i-th natural density, i-th deformation modulus, i-th permeability coefficient, i-th cohesion, i-th internal friction angle, i-th standard penetration test blow count, i-th dynamic elastic modulus, and i-th dynamic shear modulus. The weighted average of the i-th geological parameters is calculated based on the i-th natural density, i-th deformation modulus, i-th permeability coefficient, i-th cohesion, i-th internal friction angle, i-th standard penetration test blow count, i-th dynamic elastic modulus, and i-th dynamic shear modulus. The i-th burial depth and the weighted average of the i-th geological parameters constitute the i-th ring geological data.

[0084] Understandably, the burial depth is taken from the actual value in the geological survey report; other geological parameters, namely natural density, deformation modulus, permeability coefficient, cohesion, internal friction angle, standard penetration test blow count, dynamic elastic modulus, and dynamic shear modulus, are based on the geological exploration stratigraphic profile diagram, with reference to... Figure 5 The proportion of each stratum at the tunnel face of the shield tunneling at the borehole is calculated to obtain the weighted average of the geological parameters. The data between two boreholes is obtained by linear interpolation. Specifically, the weighted average of the i-th geological parameter is calculated as follows:

[0085]

[0086] in, α is the weighted average of the i-th geological parameter. r H represents the actual geological parameters of the r-th stratum. r Let G be the actual thickness of the r-th stratum, and D be the diameter of the cutterhead. That is, for any ring, the geological data G = {burial depth, α natural density, α deformation modulus, ..., α dynamic shear modulus}.

[0087] The i-th modifier data includes the i-th ring foam stock volume dataset, the i-th ring foaming agent solution concentration dataset, the i-th ring bentonite slurry volume dataset, and the i-th ring bentonite slurry concentration dataset. The i-th ring foam stock volume is obtained by calculating the cumulative value of the i-th ring foam stock volume dataset, the i-th ring bentonite slurry volume is obtained by calculating the cumulative value of the i-th ring bentonite slurry volume dataset, the i-th ring foaming agent solution concentration is obtained by calculating the average value of the i-th ring foaming agent solution concentration dataset, and the i-th ring bentonite slurry concentration is obtained by calculating the average value of the i-th ring bentonite slurry concentration dataset. The i-th ring foam stock volume, i-th ring bentonite slurry volume, i-th ring foaming agent solution concentration, and i-th ring bentonite slurry concentration constitute the i-th ring modifier data.

[0088] In some alternative embodiments, reference continues to be made to... Figure 1The superior dataset includes a training set and a test set. When training the XGBoost prediction model, the training set is input, and the Optuna algorithm is used to optimize the hyperparameters of the XGBoost prediction model to obtain the trained XGBoost prediction model. The hyperparameters include the maximum depth of the tree, the minimum leaf weight, the parameters of the L1 regularization term, the parameters of the L2 regularization term, and the learning rate.

[0089] Understandably, a crucial step in model training is the continuous adjustment of hyperparameters. By comparing the model accuracy under different combinations of hyperparameters, the optimal hyperparameters are determined. As model framework parameters, hyperparameters significantly impact model accuracy; a reasonable combination of hyperparameters can greatly improve the model's generalization performance. Compared to manual hyperparameter tuning, automatic hyperparameter tuning algorithms offer significant advantages in computational efficiency and accuracy. The Optuna algorithm is used to fine-tune the model's hyperparameters, and the specific steps are as follows:

[0090] Determine the objective function for optimization: This invention sets the optimization objective of the Optuna algorithm as the minimum root mean square error (RMSE) between the predicted result and the actual value.

[0091] Determine the search scope: The search method of this invention is set as follows, and the search scope is shown in Table 1.

[0092] Set the number of iterations: When the maximum number of iterations is reached, output the optimal hyperparameter ratio.

[0093] The Optuna algorithm was used to optimize the hyperparameters of the XGBoost model: max depth, min child weight, learning rate, alpha, and lambda. The meanings and search ranges of the hyperparameters are shown in Table 1.

[0094] Table 1. Meaning and Search Range of Hyperparameters

[0095] max depth Maximum depth of tree [5,20] min child weight Minimum Leaf Weight (1,15)

[0096] Table 1. Meaning and Search Range of Hyperparameters

[0097] alpha Parameters of L1 regularization (0.01,10.0) lambda Parameters of L2 regularization (0.01,10.0) learning rate Learning rate (0.01,0.2)

[0098] In some alternative embodiments, refer to Figure 1 and Figure 6 , Figure 6 This diagram illustrates the input parameters and corresponding output parameters. The test set is input into the trained XGBoost prediction model to obtain predicted values. The prediction accuracy evaluation index is calculated based on the predicted values, and the prediction accuracy of the trained XGBoost prediction model is evaluated based on the prediction accuracy evaluation index.

[0099] Understandably, 80% of the data in the superior dataset is divided into a training set, and the remaining 20% ​​is divided into a test set. The training set is used to build the model, and the test set is used to verify the model's accuracy. Specifically, the following data from the test set are input into the trained XGBoost prediction model: the cutterhead rotation speed, tunneling speed, total thrust, soil chamber pressure, auger rotation speed, auger torque, and geological data of the i-th ring. The trained XGBoost prediction model outputs predicted values, including predictions of the amendment data and the cutterhead torque of the i-th ring. The coefficient of determination (R²) is then used to determine the predicted values. 2 Root mean square error (RMSE) and mean absolute error (MAE) are used as indicators to evaluate prediction accuracy. 2 The closer the value is to 1, the better the model's prediction accuracy; the smaller the RMSE and MAE, the better the model's prediction accuracy. 2 RMSE and MAE are calculated as follows:

[0100]

[0101]

[0102]

[0103] Where Q is the total number of samples, y i This is the actual value of the i-th ring. For the prediction result of the i-th ring, This is the average of the actual values.

[0104] In some alternative embodiments, reference continues to be made to... Figure 1 and Figure 6 The inferior dataset is input into the trained XGBoost prediction model, including:

[0105] Input the i-th ring cutterhead rotation speed, i-th ring tunneling speed, i-th ring total thrust, i-th ring soil chamber pressure, i-th ring auger rotation speed, i-th ring auger torque, and i-th ring geological data from the differential dataset into the trained XGBoost prediction model. The trained XGBoost prediction model outputs optimized values, which include optimized i-th ring amendment data and optimized i-th ring cutterhead torque.

[0106] After outputting the optimized value, the calculation of tunneling specific energy and optimized tunneling specific energy is also included. If the optimized tunneling specific energy is less than the tunneling specific energy, the trained XGBoost prediction model performs well; if the optimized tunneling specific energy is greater than or equal to the tunneling specific energy, the trained XGBoost prediction model performs poorly.

[0107] Tunneling specific energy and optimized tunneling specific energy are calculated as follows:

[0108]

[0109] Among them, E S For tunneling specific energy, E S To optimize the tunneling specific energy, N i Let T be the rotational speed of the i-th ring cutter head. i Let V be the torque of the i-th ring cutter head, D be the diameter of the cutter head, and V be the torque of the ring cutter head. i Let T be the tunneling speed of the i-th ring. i To optimize the torque of the i-th ring cutter head.

[0110] Understandably, if the optimized tunneling energy is less than the tunneling energy, it means that the optimization effect of the slag amendment is good and it can also save the energy consumption of the tunnel boring machine. If the optimized tunneling energy is greater than or equal to the tunneling energy, then the model needs to be further adjusted.

[0111] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0112] In the above embodiments of the present invention, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0113] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for predicting and optimizing soil amendments for tunnel boring machines based on machine learning, characterized in that, include: Obtain the database, preprocess the database to obtain the dataset, the dataset includes data from the first ring to the Nth ring, the data of the i-th ring includes the tunneling data of the i-th ring, the geological data of the i-th ring and the amendment data of the i-th ring, the tunneling data of the i-th ring includes the cutterhead torque of the i-th ring, the total propulsion force of the i-th ring, the cutterhead rotation speed of the i-th ring, and the tunneling speed of the i-th ring, N>1 and N is an integer, 1≤i≤N and i is an integer; The cutterhead penetration of the i-th ring is calculated based on the cutterhead rotation speed and the tunneling speed of the i-th ring. The torque cutting depth index and the field cutting depth index of the i-th ring are calculated based on the cutterhead penetration of the i-th ring. Based on the standard fitting curve of torque depth of cutter penetration, it is determined whether the torque depth of cutter penetration of the i-th ring is greater than the standard torque depth of cutter penetration of the i-th ring under the condition of cutter penetration of the i-th ring. Based on the standard fitting curve of field depth of cutter penetration, it is determined whether the field depth of cutter penetration of the i-th ring is greater than the standard field depth of cutter penetration of the i-th ring under the condition of cutter penetration of the i-th ring. If the torque shear depth index of the i-th ring is less than or equal to the standard torque shear depth index of the i-th ring, and the field shear depth index of the i-th ring is less than or equal to the standard field shear depth index of the i-th ring, the i-th ring data is superior data; if the torque shear depth index of the i-th ring is greater than the standard torque shear depth index of the i-th ring, and / or, the field shear depth index of the i-th ring is greater than the standard field shear depth index of the i-th ring, the i-th ring data is inferior data, the superior data constitutes a superior dataset, and the inferior data constitutes a inferior dataset; Construct an XGBoost prediction model and train the XGBoost prediction model using the superior dataset to obtain the trained XGBoost prediction model. The differential dataset is input into the trained XGBoost prediction model to obtain the corresponding optimized modifier parameters and optimized cutter head torque.

2. The method for predicting and optimizing shield tunneling soil amendments based on machine learning according to claim 1, characterized in that, The dataset is obtained by preprocessing the database, including: The database includes subsets 1 to N. The i-th subset includes i-th tunneling data, i-th geological data, and i-th amendment data. The i-th tunneling data includes i-th cutterhead rotation speed dataset, i-th tunneling speed dataset, i-th total thrust dataset, i-th cutterhead torque dataset, i-th soil chamber pressure dataset, i-th auger rotation speed dataset, and i-th auger torque dataset. Data with a cutterhead rotation speed of 0, data with a tunneling speed of 0, and data with a total thrust of 0 in the i-th cutterhead rotation speed dataset, the i-th tunneling speed dataset, and the i-th total thrust dataset are removed. Outliers in these datasets are then removed using a box plotter algorithm to obtain subsets of the i-th cutterhead rotation speed, i-th tunneling speed, i-th total thrust, i-th cutterhead torque, i-th soil chamber pressure, i-th auger speed, and i-th auger torque. The average value of the subset of i-th cutterhead rotational speeds is used to obtain the i-th ring cutterhead rotational speed; the average value of the subset of i-th tunneling speeds is used to obtain the i-th ring tunneling speed; the average value of the subset of i-th total propulsion force is used to obtain the i-th ring total propulsion force; the average value of the subset of i-th cutterhead torques is used to obtain the i-th ring cutterhead torque; the average value of the subset of i-th soil chamber pressures is used to obtain the i-th ring soil chamber pressure; the average value of the subset of i-th auger speeds is used to obtain the i-th ring auger speed; and the average value of the subset of i-th auger torques is used to obtain the i-th ring auger torque. The i-th ring cutterhead rotational speed, the i-th ring tunneling speed, the i-th ring total propulsion force, the i-th ring cutterhead torque, the i-th ring soil chamber pressure, the i-th ring auger speed, and the i-th ring auger torque constitute the i-th ring tunneling data. The i-th geological data includes the i-th burial depth, i-th natural density, i-th deformation modulus, i-th permeability coefficient, i-th cohesion, i-th internal friction angle, i-th standard penetration test blow count, i-th dynamic elastic modulus, and i-th dynamic shear modulus. The weighted average of the i-th geological parameters is calculated based on the i-th natural density, i-th deformation modulus, i-th permeability coefficient, i-th cohesion, i-th internal friction angle, i-th standard penetration test blow count, i-th dynamic elastic modulus, and i-th dynamic shear modulus. The i-th burial depth and the weighted average of the i-th geological parameters constitute the i-th ring geological data. The i-th modifier data includes the i-th ring foam stock solution volume dataset, the i-th ring foaming agent solution concentration dataset, the i-th ring bentonite slurry volume dataset, and the i-th ring bentonite slurry concentration dataset. The i-th ring foam stock solution volume is obtained by calculating the cumulative value of the i-th ring foam stock solution volume dataset, the i-th ring bentonite slurry volume is obtained by calculating the cumulative value of the i-th ring bentonite slurry volume dataset, the i-th ring foaming agent solution concentration is obtained by calculating the average value of the i-th ring foaming agent solution concentration dataset, and the i-th ring bentonite slurry concentration is obtained by calculating the average value of the i-th ring bentonite slurry concentration dataset. The i-th ring foam stock solution volume, the i-th ring bentonite slurry volume, the i-th ring foaming agent solution concentration, and the i-th ring bentonite slurry concentration constitute the i-th ring modifier data.

3. The method for predicting and optimizing shield tunneling soil amendments based on machine learning according to claim 1, characterized in that, The superior dataset is divided into a training set and a test set. When training the XGBoost prediction model, the training set is input, and the hyperparameters of the XGBoost prediction model are optimized using the Optuna algorithm to obtain the trained XGBoost prediction model. The hyperparameters include the maximum depth of the tree, the minimum leaf weight, the parameters of the L1 regularization term, the parameters of the L2 regularization term, and the learning rate.

4. The method for predicting and optimizing shield tunneling soil amendments based on machine learning according to claim 3, characterized in that, The test set is input into the trained XGBoost prediction model to obtain predicted values. A prediction accuracy evaluation index is calculated based on the predicted values, and the prediction accuracy of the trained XGBoost prediction model is evaluated based on the prediction accuracy evaluation index.

5. The method for predicting and optimizing shield tunneling spoil amendments based on machine learning according to claim 1, characterized in that, Inputting the differential dataset into the trained XGBoost prediction model includes: The XGBoost prediction model is trained by inputting the cutterhead rotation speed, tunneling speed, total propulsion force, soil chamber pressure, auger rotation speed, auger torque, and geological data of the i-th ring from the differential dataset. The trained XGBoost prediction model outputs optimized values, which include optimized amendment data and optimized cutterhead torque of the i-th ring.

6. The method for predicting and optimizing shield tunneling soil amendments based on machine learning according to claim 5, characterized in that, After outputting the optimized value, the method further includes calculating the tunneling specific energy and the optimized tunneling specific energy. If the optimized tunneling specific energy is less than the tunneling specific energy, the trained XGBoost prediction model performs well; if the optimized tunneling specific energy is greater than or equal to the tunneling specific energy, the trained XGBoost prediction model performs poorly.

7. The method for predicting and optimizing shield tunneling soil amendments based on machine learning according to claim 6, characterized in that, The tunneling specific energy and the optimized tunneling specific energy are calculated in the following manner: Among them, E S Let E′ be the tunneling specific energy. S For the optimized tunneling specific energy, N i T is the rotational speed of the i-th ring cutter head. i Let V be the torque of the i-th ring cutter head, D be the diameter of the cutter head, and V be the torque of the ring cutter head. i Let T′ be the tunneling speed of the i-th ring. i To optimize the torque of the i-th ring cutter head.

8. The method for predicting and optimizing shield tunneling soil amendments based on machine learning according to claim 1, characterized in that, The penetration depth of the i-th ring cutterhead is calculated based on the rotational speed of the i-th ring cutterhead and the tunneling speed of the i-th ring, in the following manner: Wherein, the S i Let N be the penetration depth of the i-th ring cutterhead. i V is the rotational speed of the i-th ring cutter head. i Let be the tunneling speed of the i-th ring.

9. The method for predicting and optimizing shield tunneling soil amendments based on machine learning according to claim 1, characterized in that, The torque depth index and the field depth index of the i-th ring are calculated based on the penetration depth of the i-th ring cutter head, and are calculated in the following manner: Among them, TPI i FPI is the torque depth index of the i-th ring. i Let T be the cutting depth index of the i-th ring field. i Let F be the torque of the i-th ring cutter head. i Let be the total thrust of the i-th ring.

Images