Real-time prediction method for muck improvement parameters of shield underneath passing existing tunnel

Abstract

The invention discloses a real-time prediction method for muck improvement parameters of a shield under-crossing existing tunnel, and the method comprises the steps: carrying out the refined preprocessing of non-under-crossing section shield tunneling data, enabling an obtained data set to be used for the training of a machine learning integrated prediction model, and obtaining a muck improver dosage integrated prediction model; meanwhile, slump experiments are carried out for different types of stratums, the optimal water content and foam injection ratio parameter range is determined, and a reliable decision basis is provided for a shield driver. The quality of a data set and the prediction precision of a machine learning model are remarkably improved through a finer data preprocessing means; the machine learning integrated prediction model fuses learning capabilities of different machine learning models, so that the prediction precision of the model is further improved; and machine learning and experimental means are combined to realize accurate and rapid prediction of the use amount of the modifier, so that safe and smooth tunneling of the shield under the existing tunnel is ensured, and the method has certain timeliness and economical efficiency.

Description

Technical Field

[0001] The invention relates to the technical field of tunnel construction, and in particular to a method for real-time prediction of parameters of slag improvement for a shield machine passing through an existing tunnel. Background Art

[0002] In urban subway construction, the shield method is widely used because of its high efficiency and speed. Shield tunneling faces many challenges in the process of crossing different strata: when the shield machine passes through silty clay strata, the clay easily adheres to the cutterhead and the cutter, resulting in the "mud cake" problem. Once the shield slag is caked, it will cause the thrust and torque of the shield to increase and the excavation efficiency to decrease. In severe cases, it will cause the cutterhead opening to be blocked and the cutter to wear unevenly. In high water pressure and highly permeable coarse-grained soil strata, the screw conveyor is prone to gushing, making it difficult to effectively control the slag discharge, and the pressure in the soil bin drops sharply, which in turn causes excessive deformation of the stratum and even instability of the excavation surface. Therefore, adjusting the soil excavated from the excavation surface into a plastic flow state in the pressure chamber is the key to the success of shield construction.

[0003] On the other hand, with the increasing density of subway lines, new shield tunnels will inevitably be close to existing tunnels. When the shield machine excavates through the existing tunnel, if the slag is not improved properly, it is easy to endanger the stability of the excavation surface, causing deformation and settlement of the existing tunnel. Therefore, for the underpass section of the shield tunnel, accurate control of slag improvement parameters is crucial. At present, the amount of improver mainly depends on the shield driver's experience based on the construction stratum conditions and shield excavation parameters such as cutter torque and penetration, which makes it difficult to ensure the accuracy of the amount of slag improver. Summary of the invention

[0004] Purpose of the invention: The purpose of the present invention is to provide a real-time prediction method for parameters of slag improvement when a shield machine is passing through an existing tunnel in order to address the shortcomings of manual experience, so as to predict the precise amount of slag improvement agent used when a shield machine is passing through an existing tunnel.

[0005] Technical solution: The method for real-time prediction of parameters of soil improvement when a shield machine passes through an existing tunnel according to the present invention comprises:

[0006] (1) Collect the shield tunneling parameters of the non-underpassing section and perform data preprocessing to construct the tunneling parameter data set D;

[0007] (2) constructing a machine learning integrated prediction model including multiple base models and a meta-model; the base model takes shield tunneling parameters as input and takes the amount of soil conditioner as output, and uses the tunneling parameter data set D constructed in step (1) to train multiple base models; the prediction results of multiple base models are integrated into a meta-data set D′ for training the meta-model, and the output of the meta-model is the final prediction result of the amount of soil conditioner;

[0008] The base model uses multiple models such as random forest RF, support vector regression SVR, XGBoost, neural network, and K nearest neighbor model, and the meta-model uses linear regression, logistic regression, or decision tree model;

[0009] (3) When the shield machine is excavating under an existing tunnel, the shield machine excavation parameters are input into the trained machine learning integrated prediction model to predict the amount of soil conditioner in real time.

[0010] Furthermore, in step (1), data preprocessing includes excavation cycle segment extraction, feature screening, outlier identification, data noise reduction, and data normalization.

[0011] Furthermore, the extraction of the excavation cycle segment includes: deleting the data of the shield machine shutdown stage by setting a time threshold, and the time threshold of the excavation cycle segment is determined according to the time length of one ring of shield excavation.

[0012] Furthermore, feature screening includes: using a gradient boosting decision tree method to sort feature importance, performing correlation analysis between features, and for a pair of highly correlated features, deleting features with lower importance.

[0013] Furthermore, outlier identification includes: using an IQR method based on a box plot to identify and delete outliers.

[0014] Furthermore, the IQR method based on the box plot is used to identify and delete outliers, including:

[0015] Sort the data by size, calculate the upper quartile Q1 and lower quartile Q3 of the data, and outliers are defined as observations that fall outside the upper and lower limits of the data set. The upper and lower limits of the data are:

[0016] IQR=Q3-Q1

[0017] Upper limit: Q3+1.5×IQR

[0018] Lower limit: Q1-1.5×IQR

[0019] Here, IQR is the interquartile range.

[0020] Furthermore, data noise reduction includes: noise reduction through a Butterworth filter.

[0021] The present invention uses more refined data preprocessing methods for a large number of shield tunneling parameters, which can significantly improve the quality of the data set, make it easier for the machine learning model to learn and understand the relationship between features, and improve the prediction accuracy of the modifier dosage.

[0022] Further, step (2) comprises:

[0023] There are K samples in the excavation parameter data set D, where x is the input variable, i.e., the excavation parameter; y is the variable to be predicted, i.e., the amount of the improver; given N base models, each base model is trained using the excavation parameter data set D, and the optimal hyperparameters of each base model are obtained through the Bayesian optimization algorithm;

[0024] D={(x1,y1),(x2,y2),(x3,y3),...,(x K ,y K )}

[0025] The prediction results of a single base model are:

[0026]

[0027] For each sample in the data set D, there are corresponding prediction results of N base models, which are integrated into a meta-data set D′. At this time, the input variable is the prediction result of the base model, and the variable to be predicted is the amount of improver:

[0028]

[0029] Use D′ to train the meta-model and obtain the final prediction result of the amount of soil conditioner:

[0030]

[0031] Furthermore, the method for real-time prediction of parameters of slag improvement when a shield machine passes through an existing tunnel also includes:

[0032] (4) Collect soil from different strata and conduct slump tests under different moisture contents and foam injection ratios to determine the optimal soil improvement parameter range suitable for the current stratum type. When the amount of soil conditioner predicted in step (3) is within the optimal soil improvement parameter range, excavate according to the predicted value.

[0033] Furthermore, in step (4), the slump test under different strata and the foam injection single-tube flow conversion are:

[0034] Carry out slump tests under different water content w and foam injection ratio FIR for different types of strata, determine the w and FIR parameter ranges, and convert the FIR parameter range into the shield machine foam injection single pipe flow range q f :

[0035]

[0036] Among them, κ is the loose coefficient of excavated soil; R is the outer diameter of the shield, m; v is the shield excavation speed, mm / min.

[0037] The present invention further conducts slump tests under different strata to determine the optimal foam injection single-tube flow range, providing auxiliary decision-making basis for shield drivers. Combining the predicted value and range value, the shield driver can quickly and accurately make decisions on the amount of modifier to ensure the safety of the tunnel underpass construction, and has certain timeliness and economy.

[0038] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages:

[0039] The present invention collects shield tunneling parameters to construct a data set. The shield tunneling parameters of the same project ensure the consistency of data distribution, thereby ensuring the applicability of the prediction model in the underpass section. The learning capabilities of different machine learning models are further integrated to construct a machine learning integrated prediction model. The model is trained using the constructed data set to ensure the prediction accuracy of the amount of slag conditioner. By accurately controlling the slag improvement parameters, the safe and smooth excavation of the shield under the existing tunnel is ensured. The slag improvement parameter prediction method is not only applicable to this working condition, but also can provide a reference for other dangerous shield tunneling conditions, such as tunnel overlaps, adjacent buildings (structures), etc. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 It is a flowchart of a method for real-time prediction of parameters of slag improvement when a shield machine passes through an existing tunnel provided by an embodiment of the present invention;

[0041] Figure 2 Schematic diagram of predicting the dosage of improver using the machine learning integrated prediction model in an embodiment of the present invention. DETAILED DESCRIPTION

[0042] The present invention will be further described below in conjunction with the accompanying drawings.

[0043] like Figure 1 As shown, an embodiment of the present invention provides a method for real-time prediction of parameters of slag improvement when a shield machine passes through an existing tunnel, comprising the following steps:

[0044] (1) The shield tunneling parameters of the non-underpassing section (such as cutter head torque, total thrust, tunneling speed, etc.) are collected. In order to improve the prediction effect of the subsequent model, the collected data are preprocessed in terms of tunneling cycle segment extraction, feature screening, outlier identification, data noise reduction, data normalization, etc. to obtain the tunneling parameter data set D;

[0045] The data of the shield machine shutdown stage is deleted by setting a time threshold. The time taken for a shield machine to excavate one ring is about 25 to 30 minutes. Therefore, 25 minutes is used as the threshold of the excavation cycle segment to find the excavation cycle segment with a non-zero total thrust for more than 25 minutes.

[0046] The gradient boosted decision tree method (GBDT) is used to rank the feature importance and perform correlation analysis between features. For a pair of highly correlated features, the features with lower importance are deleted.

[0047] For outliers, the IQR method based on box plot is used to identify and delete outliers. The data is sorted by size, and the upper quartile Q1 and lower quartile Q3 of the data are calculated. Outliers are defined as observations that fall outside the upper and lower limits of the data set. The upper and lower limits of the data are:

[0048] IQR=Q3-Q1

[0049] Upper limit: Q3+1.5×IQR

[0050] Lower limit: Q1-1.5×IQR

[0051] Here, IQR is the interquartile range.

[0052] Butterworth filter was used to reduce noise and normalize the data.

[0053] (2) Construct a machine learning integrated prediction model including multiple base models and a meta-model; the base model takes shield tunneling parameters as input and takes the amount of slag improver as output, and uses the tunneling parameter data set D constructed in step (1) to train multiple base models; the prediction results of multiple base models are integrated into a meta-data set D′ for training the meta-model, and the output of the meta-model is the final prediction result of the amount of slag improver.

[0054] The base model can be selected from a variety of models such as random forest (RF), support vector regression (SVR), XGBoost, neural network, K nearest neighbor (KNN), etc. The meta-model can be a simpler linear model, such as linear regression, logistic regression, decision tree and other models.

[0055] like Figure 2 As shown in the figure, there are K samples in the excavation parameter data set D, x is the input variable, i.e., the excavation parameter; y is the variable to be predicted, i.e., the amount of the modifier. Given N base models, each base model is trained using the excavation parameter data set D, and the optimal hyperparameters of each base model are obtained through the Bayesian optimization algorithm;

[0056] D={(x1,y1),(x2,y2),(x3,y3),...,(x K ,y K )}

[0057] The prediction results of a single base model are:

[0058]

[0059] For each sample in the data set D, there are corresponding prediction results of N base models, which are integrated into a meta-data set D′. At this time, the input variable is the prediction result of the base model, and the variable to be predicted is the amount of improver:

[0060]

[0061] Use D′ to train the meta-model and obtain the final prediction result of the amount of soil conditioner:

[0062]

[0063] (3) When the shield tunneling machine passes through the existing tunnel, the shield tunneling parameters are input into the trained machine learning integrated prediction model to predict the amount of soil conditioner in real time;

[0064] (4) Collect slag from different strata, conduct slump tests under different moisture contents and foam injection ratios, and determine the optimal slag improvement parameter range suitable for the current stratum type. This numerical range can be used as a basis for shield drivers to assist in decision-making. When the amount of slag improver predicted in step (3) is within the optimal slag improvement parameter range, the shield driver excavates according to the predicted value. When the predicted amount of slag improver is not within the optimal slag improvement parameter range, the specific reasons should be identified and the excavation parameters and slag improvement parameters should be optimized to ensure safe and smooth construction.

[0065] Slump test in different strata and single-tube flow conversion of foam injection:

[0066] Slump tests were conducted for different types of strata under different water contents w and foam injection ratios FIR to determine the parameter ranges of w and FIR. To facilitate the actual operation of the shield machine driver, the FIR parameter range was further converted into the single-tube flow range q of the shield machine foam injection. f :

[0067]

[0068] Among them, κ is the loose coefficient of excavated soil; R is the outer diameter of the shield, m; v is the shield excavation speed, mm / min; the shield machine is equipped with 6 foam pipelines.

Claims

1. A real-time prediction method for parameters of soil improvement when a shield machine is passing through an existing tunnel, characterized in that: include: (1) Collect the shield tunneling parameters of the non-underpassing section and perform data preprocessing to construct the tunneling parameter data set D; (2) constructing a machine learning integrated prediction model including multiple base models and a meta-model; the base model uses shield tunneling parameters as input and soil conditioner dosage as output, and uses the tunneling parameter data set D constructed in step (1) to train multiple base models; The prediction results of multiple base models are integrated into a meta-data set D′ for training the meta-model. The output of the meta-model is the final prediction result of the amount of soil conditioner. The base model uses multiple models such as random forest RF, support vector regression SVR, XGBoost, neural network, and K nearest neighbor model, and the meta-model uses linear regression, logistic regression, or decision tree model; (3) When the shield machine is excavating under an existing tunnel, the shield machine excavation parameters are input into the trained machine learning integrated prediction model to predict the amount of soil conditioner in real time.

2. The method for real-time prediction of parameters of soil improvement when a shield machine passes through an existing tunnel according to claim 1 is characterized in that: In step (1), data preprocessing includes excavation cycle segment extraction, feature screening, outlier identification, data noise reduction, and data normalization.

3. The method for real-time prediction of parameters of soil improvement when a shield machine passes through an existing tunnel according to claim 2 is characterized in that: The extraction of the excavation cycle segment includes: deleting the data of the shield machine shutdown stage by setting a time threshold. The time threshold of the excavation cycle segment is determined according to the time length of one ring of shield excavation.

4. The method for real-time prediction of parameters of soil improvement when a shield machine passes through an existing tunnel according to claim 2 is characterized in that: Feature screening includes: using the gradient boosting decision tree method to sort the feature importance, performing correlation analysis between features, and for a pair of highly correlated features, deleting the features with lower importance.

5. The method for real-time prediction of parameters of soil improvement when a shield machine passes through an existing tunnel according to claim 2 is characterized in that: Outlier identification includes: using the IQR method based on the box plot to identify and delete outliers.

6. The method for real-time prediction of parameters of soil improvement when a shield machine passes through an existing tunnel according to claim 5 is characterized in that: The IQR method based on the box plot is used to identify and remove outliers, including: Sort the data by size, calculate the upper quartile Q1 and lower quartile Q3 of the data, and outliers are defined as observations that fall outside the upper and lower limits of the data set. The upper and lower limits of the data are: IQR=Q3-Q1 Upper limit: Q3+1.5×IQR Lower limit: Q1-1.5×IQR Here, IQR is the interquartile range.

7. The method for real-time prediction of parameters of soil improvement when a shield machine passes through an existing tunnel according to claim 2 is characterized in that: Data denoising includes: denoising through Butterworth filter.

8. The method for real-time prediction of parameters of soil improvement when a shield machine passes through an existing tunnel according to claim 1 is characterized in that: Step (2) comprises: There are K samples in the excavation parameter data set D, where x is the input variable, i.e., the excavation parameter; y is the variable to be predicted, i.e., the amount of the improver; given N base models, each base model is trained using the excavation parameter data set D, and the optimal hyperparameters of each base model are obtained through the Bayesian optimization algorithm; D={(x1,y1),(x2,y2),(x3,y3),...,(x K ,y K )} The prediction results of a single base model are: For each sample in the data set D, there are corresponding prediction results of N base models, which are integrated into a meta-data set D′. At this time, the input variable is the prediction result of the base model, and the variable to be predicted is the amount of improver: Use D′ to train the meta-model and obtain the final prediction result of the amount of soil conditioner:

9. The method for real-time prediction of parameters for improving slag soil when a shield machine passes through an existing tunnel according to any one of claims 1 to 8, characterized in that: Also includes: (4) Collect soil from different strata and conduct slump tests under different moisture contents and foam injection ratios to determine the optimal soil improvement parameter range suitable for the current stratum type. When the amount of soil conditioner predicted in step (3) is within the optimal soil improvement parameter range, excavate according to the predicted value.

10. The method for real-time prediction of parameters of soil improvement when a shield machine passes through an existing tunnel according to claim 9, characterized in that: In step (4), slump test under different formations and foam injection single pipe flow conversion: Carry out slump tests under different water content w and foam injection ratio FIR for different types of strata, determine the w and FIR parameter ranges, and convert the FIR parameter range into the shield machine foam injection single pipe flow range q f : Among them, κ is the loose coefficient of excavated soil; R is the outer diameter of the shield, m; v is the shield excavation speed, mm / min.

Images