Loess stratum large-diameter pipe jacking tunneling auxiliary decision-making method based on data driving and deep learning

Abstract

The invention discloses a loess stratum large-diameter pipe jacking tunneling auxiliary decision-making method based on data driving and deep learning, and the method comprises the steps: obtaining related data of a pipe jacking machine in a construction process, and carrying out the preprocessing and normalization of the data; performing signal decomposition by using a VMD algorithm; setting a light GBM hyper-parameter, and carrying out parameter adjustment on the hyper-parameter; the data set is used for training a VMD-Light GBM model; the determined parameters are adjusted according to actual engineering conditions, limit values are set, and variable constraint conditions are formed; taking a nonlinear mapping function, which is fitted by a VMD-Light GBM algorithm and is related to the relationship between the pipe jacking construction parameters and the pipe jacking postures, as an optimized fitness function; values of parameters of the NSGA-III algorithm are determined, minimization of absolute values of multiple pipe jacking attitude targets is taken as a target, global optimization is carried out by adopting the NSGA-III algorithm, and an optimal solution of pipe jacking construction parameters is determined and is taken as a guidance scheme of field construction; according to the method, the pipe jacking implementation posture can be accurately predicted, and an optimal tunneling parameter control scheme is provided.

Description

Technical Field

[0001] The present invention relates to the technical field of pipe jacking attitude control, and particularly to an auxiliary decision-making method for large-diameter pipe jacking tunneling in loess stratum based on data driving and deep learning. Background Technique

[0002] With the rapid growth of urban population, the problem of traffic congestion has become increasingly serious. In order to make more efficient use of underground space, a large number of pipe jacking tunnel projects have been widely applied to underground development. However, during the pipe jacking construction process, due to factors such as uneven formation conditions and non-standard construction operations, the pipe jacking machine often experiences uneven stress, resulting in the deflection of its attitude and deviation from the designed axis. This makes the attitude control of large-section rectangular pipe jacking tunnels extremely difficult.

[0003] At present, the pipe jacking construction technology faces problems such as low degree of intelligence and dependence on manual operation for guiding measurement accuracy, and it is urgently necessary to be improved through technological innovation and intelligent means. Factors such as complex and changeable geological environment, uneven cutter head force, uneven grouting, and uneven thrust often lead to the deviation of the actual pipe jacking trajectory from the preset trajectory, increasing the construction difficulty and risk. During the pipe jacking construction process, due to the failure to timely master the tunneling attitude information of the pipe jacking machine, accidents such as ground settlement, stuck cutter head, deformation and even rupture of pipes often occur, which not only pose a threat to the construction quality but also may cause serious economic losses. Therefore, timely mastering the attitude information of the pipe jacking machine and ensuring the guiding accuracy are the keys to ensuring the construction quality of pipe jacking.

[0004] Pipe jacking attitude control plays a key role in the pipe jacking tunneling system. With the continuous development of pipe jacking construction technology, this technology is also constantly innovating and improving. In view of the complexity of the pipe jacking tunneling control system, intelligence will become an inevitable trend in the development of pipe jacking attitude control technology. Summary of the Invention

[0005] In order to solve the problems existing in the prior art, the purpose of the present invention is to provide an auxiliary decision-making method for large-diameter pipe jacking tunneling in loess stratum based on data driving and deep learning, and the present invention can accurately predict the pipe jacking implementation attitude and provide an optimal tunneling parameter control scheme.

[0006] To achieve the above purpose, the technical solution adopted by the present invention is: an auxiliary decision-making method for large-diameter pipe jacking tunneling in loess stratum based on data driving and deep learning, including the following steps:

[0007] Step 1: Obtain the relevant data of the pipe jacking machine during the construction process, and perform preprocessing and normalization on the data;

[0008] Step 2: Apply the VMD algorithm for signal decomposition: Input the acquired time series data, decompose it into multiple modal IMFs through VMD, and obtain the decomposed dataset for training the VMD-LightGBM model;

[0009] Step 3: Set the LightGBM hyperparameters and tune the hyperparameters by gradually adjusting within a certain parameter range to find the parameter combination with the best accuracy on the validation set;

[0010] Step 4: Use the dataset to train the VMD-LightGBM model, and combine 5-fold cross-validation to avoid overfitting; the output result is the predicted value of the pipe jacking attitude deviation, compare it with the actual value, and use RMSE, MAE, R 2 metrics to evaluate the performance of the VMD-LightGBM model, and rank the importance of the influence of different parameters on the pipe jacking attitude deviation;

[0011] Step 5: Adjust the parameters determined in Step 4 according to the actual engineering situation and set the limit values to form variable constraint conditions; use the non-linear mapping function of the relationship between pipe jacking construction parameters and pipe jacking attitude fitted by the VMD-LightGBM algorithm as the optimized fitness function; determine the value range of the NSGA-Ⅲ algorithm parameters, take the minimization of the absolute values of multiple pipe jacking attitude targets as the goal, and use the NSGA-Ⅲ algorithm for global optimization to determine the optimal solution of the pipe jacking construction parameters as the guiding scheme for on-site construction.

[0012] As a further improvement of the present invention, Step 1 specifically includes the following steps:

[0013] Step 1.1: Collect data under normal working conditions through sensors and data loggers on the tunneling system;

[0014] Step 1.2: Normalize the data samples of different targets and scale them proportionally to [0, 1];

[0015] Step 1.3: Use the median to fill in the missing values, use the box plot to detect outliers, and use the mean to replace the outliers.

[0016] As a further improvement of the present invention, Step 2 is specifically as follows:

[0017] The time series data is optimized by adaptively selecting the frequency, and the VMD algorithm uses the following variational method and constraint conditions:

[0018]

[0019] where: u k (t) is the k-th intrinsic mode function, w kis the central frequency of each modal function, α is a regularization parameter that controls the smoothness of signal decomposition, K is the number of decomposed modes;

[0020] A set of intrinsic mode functions is obtained, thereby decomposing the original data into several IMF components.

[0021] As a further improvement of the present invention, in step 3, the hyperparameters include: L2 regularization parameter, learning rate, and maximum depth of the tree; the Bayesian optimization method is used to tune the hyperparameters of LightGBM, and the L2 regularization parameter, learning rate, and maximum depth of the tree are selected as optimization parameters, the mean squared error loss function of the training model is used as the objective function, and the initial ranges of the hyperparameters are respectively set in [3, 9], [0, 0.5], and [4, 8], and adjusted step by step to find the parameter combination with the best accuracy on the validation set.

[0022] As a further improvement of the present invention, step 4 specifically includes the following steps:

[0023] Step 4.1, construct a VMD-LightGBM model;

[0024] Step 4.2, use the 5-fold cross-validation method to divide the entire data set into 5 parts, where 4 parts are used for model training and 1 part is used to evaluate the model performance; and this process is repeated multiple times to ensure that the entire data set participates in the evaluation of the model performance;

[0025] Step 4.3, use the data set constructed in step 4.2 to train the VMD-LightGBM model;

[0026] Step 4.4, use RMSE, MAE, R 2 indicators to evaluate the performance of the model;

[0027] Step 4.5, through the LightGBM built-in gain algorithm, measure the contribution of each feature to the improvement of the model performance based on the feature importance of information gain, and rank the importance of each feature.

[0028] As a further improvement of the present invention, step 5 specifically includes the following steps:

[0029] Step 5.1, introduce the non-linear mapping function about the relationship between the pipe jacking construction parameters and the pipe jacking attitude fitted by the VMD-LightGBM algorithm as the optimized fitness function;

[0030] Step 5.2, form different construction scenarios according to the actual engineering requirements and project adjustments of each parameter, and use the existing data to set the limit range for the value of each parameter to form variable constraint conditions.

[0031] Step 5.3: Determine the values of the NSGA-Ⅲ algorithm parameters. Taking the minimization of the absolute values of the four pipe jacking attitude targets as the goal, use the NSGA-Ⅲ method for global optimization to determine the optimal solution of the pipe jacking construction parameters.

[0032] As a further improvement of the present invention, in Step 5.1, the objective function of the pipe jacking attitude based on the VMD-LightGBM algorithm is:

[0033] min f(x) = [f1(x), f2(x), …, f4(x)]

[0034] f i (x) = LightGBM(x1, x2, …, x 14 ).

[0035] Where: f i (x) is a non-linear mapping function about the relationship between the pipe jacking construction parameters and the pipe jacking attitude fitted by the VMD-LightGBM algorithm; x1, x2, …, x 14 are the pipe jacking construction parameters;

[0036] As a further improvement of the present invention, in Step 5.3, use the Technique for Order Preference by Similarity to an Ideal Solution (TOPSIS) method to select an optimal solution from the optimal solution set after multi-objective optimization by numerous NSGA-Ⅲ algorithms; specifically as follows:

[0037] For each solution x j = [f1(x j ), f2(x j ), …, f4(x j )] on the front of the optimal solution set, calculate the Euclidean distance between this solution and the ideal solution min f(x); the distance d j from the solution x j to the ideal solution min f(x) is:

[0038]

[0039] Select the solution closest to the ideal solution, that is, minimize the distance; for each solution x j , calculate its distance d j to the ideal solution, and then select the solution with the minimum distance as the optimal solution:

[0040]

[0041] That is, select the solution x j that minimizes the distance d j :

[0042]

[0043] Select a set of solutions with the smallest distance as the guiding scheme for on-site construction.

[0044] The beneficial effects of the present invention are as follows:

[0045] The pipe jacking tunneling attitude prediction and control method based on the VMD-LightGBM-NSGA-Ⅲ algorithm of the present invention can accurately predict the pipe jacking implementation attitude and provide an optimal tunneling parameter control scheme, avoiding the disadvantages of traditional control mainly relying on manual experience, and greatly improving aspects such as the completion period, construction cost, quality and safety of engineering projects. Description of the Drawings

[0046] Figure 1 is the flowchart of the embodiment of the present invention;

[0047] Figure 2 is the VMD-LightGBM-NSGA-Ⅲ model structure diagram in the embodiment of the present invention. Detailed Embodiment

[0048] The embodiments of the present invention will be described in detail below with reference to the drawings.

[0049] Embodiment

[0050] As Figure 1 and Figure 2 shown, a pipe jacking tunneling attitude prediction and control method based on the VMD-LightGBM-NSGA-Ⅲ algorithm includes the following steps:

[0051] S1: Obtain relevant data during the construction of the pipe jacking machine, such as propulsion speed, rotation angle, earth pressure, geological conditions, etc., preprocess the data, clean the data, process missing values and outliers, and perform data normalization.

[0052] The specific process of step S1 further includes the following steps:

[0053] S1.1: Collect data in the normal working state through sensors and data collectors on the tunneling system;

[0054] Specifically, the data is obtained by a real-time measurement system, and the data of the pipe jacking machine tunneling 1 ring to 200 rings is collected as samples. Each ring of data includes 14 parameters, namely buried depth CD, jack thrust JF, propulsion speed AR, cutter head torque CT, screw conveyor speed SRS, cutter head earth pressure CEP, cutter head speed CR, grouting volume GV, grouting pressure GP, foam injection volume FJV, soil output volume EV, tunneling length EL, articulated horizontal deviation AHD, and articulated vertical deviation AVD.

[0055] S1.2: Normalize the data samples of different targets and scale them proportionally to [0, 1];

[0056] Specifically, the MinMaxScaler algorithm is used to scale the data proportionally to the range [0, 1]. Compressing the values of data features into a unified scale range can prevent certain features from affecting the performance of the model during training due to their large or small ranges.

[0057] S1.3: Use the median to fill in missing values, use box plots to detect outliers, and use the mean to replace outliers.

[0058] Specifically, according to the order of the data, find the median after sorting, and fill the missing values in this column of data with the median to ensure that the missing values are reasonably filled without introducing too much bias. Use the box method for outlier detection. According to the interquartile range (IQR, the difference between the upper quartile Q1 and the lower quartile Q3), define the range of outliers. Usually, values less than Q1 - 1.5IQR or greater than Q3 + 1.5IQR are marked as outliers, and the mean of the data at both ends of the outliers is used to replace the outliers.

[0059] S2: Apply the VMD algorithm for signal decomposition. Input the time series data (such as propulsion speed and rotation angle) collected in S1 and decompose it into multiple modes (IMFs) through VMD to obtain the decomposed data set. The training set is used to train the VMD-LightGBM model.

[0060] Specifically, input the time series data (such as propulsion speed and rotation angle) collected in S1, and ensure that there is no overlap between the frequency bands corresponding to each IMF by adaptively selecting the frequency. Use the following variational method and constraint conditions for optimization to obtain a set of intrinsic mode functions. Decompose the original data into several IMF components. Specifically, VMD achieves signal decomposition by minimizing the following variational problem:

[0061]

[0062] where: u k (t) is the k-th intrinsic mode function; w k is the central frequency of each mode function; α is a regularization parameter that controls the smoothness of signal decomposition; K is the number of decomposed modes.

[0063] S3: Set the LightGBM hyperparameters. There are mainly 3 parameters to be adjusted: the L2 regularization parameter, the learning rate, and the maximum depth of the tree. Use the Bayesian optimization method to tune the hyperparameters, and gradually adjust within a certain parameter range to find the parameter combination with the best accuracy on the validation set.

[0064] Specifically, the Bayesian optimization method is used to tune the hyperparameters of LightGBM. The L2 regularization parameter, learning rate, and maximum depth of the tree are selected as the optimization parameters. The mean squared error loss function of the training model is used as the objective function, and the initial ranges of the parameters are set at [3, 9], [0, 0.5], and [4, 8], and adjusted step by step to find the parameter combination with the best accuracy on the validation set.

[0065] S4: Use the dataset constructed in S2 to train the VMD-LightGBM model, and combine 5-fold cross-validation to avoid overfitting; the output result is the predicted value of the pipe jacking attitude deviation, compare it with the actual value, and use RMSE, MAE, R 2 indicators to evaluate the performance of the model. Through the built-in "gain" algorithm of the model, rank the importance of the influence of different parameters on the pipe jacking attitude deviation.

[0066] S4.1: Construct the VMD-LightGBM model;

[0067] Specifically, LightGBM is an efficient algorithm based on gradient boosting decision trees, with the advantages of fast training and low memory consumption, and supports parallel processing of massive data. It accelerates training and improves accuracy by discretizing continuous data into histograms, and uses the gain method to extract features. The LightGBM objective function can be expressed by the following formula:

[0068]

[0069] where: y i is the true value of the sample, is the value predicted by the model, and n is the total number of samples.

[0070] S4.2: Use the 5-fold cross-validation method to divide the entire dataset into 5 parts, where 4 parts are used for model training and 1 part is used to evaluate the model performance. This process is repeated 5 times to ensure that the entire dataset participates in the evaluation of the model performance;

[0071] Specifically, to fully utilize the dataset to evaluate the model performance, the 5-fold cross-validation method is adopted to reduce the fluctuations caused by unreasonable data division. Specifically, the entire dataset is divided into 5 parts, where 4 parts are used for model training and 1 part is used to evaluate the model performance. Through multiple validations, cross-validation can reduce the variance of the model evaluation results, making the evaluation more stable. It improves the generalization ability of the model on different data subsets and avoids relying on a single training / test set. Each data point is used as the validation set at least once, which enables the model to make the most of the data.

[0072] S4.3: Use the dataset constructed in S4.2 to train the VMD-LightGBM model;

[0073] S4.4: Use RMSE, MAE, and R 2 metrics to evaluate the performance of this model.

[0074] Specifically, the model output data are the shield head horizontal deviation, shield tail horizontal deviation, shield head vertical deviation, and shield tail vertical deviation. RMSE is one of the common metrics for evaluating the prediction accuracy of a regression model, representing the square root of the mean of the squares of the differences between the predicted values and the actual values. Its principle is to calculate the squares of all prediction errors (i.e., the gaps between the predicted values and the true values), emphasizing larger errors, so as to give an error measure of the same order of magnitude as the original data. MAE is another commonly used metric for evaluating the prediction accuracy of a model. It calculates the average of the absolute values of all prediction errors and can provide the average magnitude of the model prediction errors. R 2 is a metric used to evaluate the goodness of fit in a regression model, representing the ability of the model to explain the variation of the data. The value ranges between 0 and 1. It measures the correlation between the model prediction results and the actual results. The closer the value is to 1, the more accurately the model can fit the data. The results of these metrics prove that the performance of the pipe jacking attitude prediction model established is accurate, can effectively map the relationship between the input and output, and there is no overfitting or underfitting situation.

[0075] S4.5: Through the built-in "gain" algorithm of LightGBM, measure the contribution of each feature to the improvement of the model performance based on the feature importance of information gain, and rank the importance of each feature.

[0076] Specifically, analyze the importance of 14 pipe jacking parameters based on the gain algorithm, select 5 pipe jacking construction parameters that have a significant impact on the 4 pipe jacking attitude targets as the main optimization parameters, and optimize the 5 pipe jacking construction parameters with a higher importance ranking in order to propose an executable construction plan.

[0077] S5: Adjust the parameters determined in S4 according to the actual engineering situation, and set limits to form variable constraint conditions. Use the non-linear mapping function of the relationship between pipe jacking construction parameters and pipe jacking attitude fitted by the VMD-LightGBM algorithm as the optimized fitness function. Determine the values of the NSGA-Ⅲ algorithm parameters, take the minimization of the absolute values of the 4 pipe jacking attitude targets as the goal, and use the NSGA-Ⅲ algorithm for global optimization to determine the optimal solution of the pipe jacking construction parameters as the guiding plan for on-site construction.

[0078] S5.1: Introduce the non-linear mapping function of the relationship between pipe jacking construction parameters and pipe jacking attitude fitted by the VMD-LightGBM algorithm as the optimized fitness function;

[0079] Specifically, the objective function for the pipe jacking attitude based on the LightGBM algorithm is as follows:

[0080] min f(x) = [f1(x), f2(x), …, f4(x)]

[0081] f i (x) = LightGBM(x1, x2, …, x 14 ).

[0082] Where: f i (x) is a non - linear mapping function that fits the relationship between the pipe jacking construction parameters and the pipe jacking attitude by the VMD - LightGBM algorithm; x1, x2, …, x 14 are the pipe jacking construction parameters;

[0083] S5.2: According to the actual engineering requirements and project adjustments, 14 parameters are formed into different construction scenarios, and the value ranges of each parameter are set with restrictions using the existing data to form variable constraint conditions;

[0084] S5.3: Determine the values of the NSGA - Ⅲ algorithm parameters. With the goal of minimizing the absolute values of the 4 pipe jacking attitude objectives, the NSGA - Ⅲ algorithm is used for global optimization to determine the optimal solutions of the pipe jacking construction parameters.

[0085] Specifically, the number of objectives of the genetic algorithm is set to 4, the population size is set to 100, the crossover operator is 0.9, the mutation operator is 0.01, and the maximum number of generations and the stopping generation are 50. Multi - objective optimization is carried out according to the steps of the NSGA - Ⅲ algorithm to obtain the Pareto optimal solution set. This method can obtain multiple pipe jacking parameter optimization schemes. However, in actual engineering, only one optimization scheme is needed to guide the on - site construction. To achieve the best optimization effect, the method of approaching the ideal solution distance is used to select an optimal scheme from the numerous Pareto solution sets.

[0086] For each solution x j = [f1(x j ), f2(x j ), …, f4(x j )] on the Pareto front, calculate the Euclidean distance between this solution and the ideal solution min f(x). According to the Euclidean distance formula, the distance d j from the solution x j to the ideal solution min f(x) is:

[0087]

[0088] Select the solution closest to the ideal solution, that is, minimize this distance. For each solution x j calculate its distance d to the ideal solutionj , and then select the solution with the minimum distance as the optimal solution:

[0089]

[0090] That is, select the solution that makes the distance d j the minimum solution x j :

[0091]

[0092] Select a set of solutions with the minimum distance as the guiding scheme for on-site construction.

[0093] It should be noted that the optimal solution of the pipe jacking construction parameters needs to be combined with the on-site labor for guiding the construction, playing an auxiliary decision-making role.

[0094] The above-described embodiments only represent the specific implementation manners of the present invention, and their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of the present invention patent. It should be pointed out that for those of ordinary skill in the art, without departing from the concept of the present invention, several modifications and improvements can still be made, and these all belong to the protection scope of the present invention.

Claims

1. A large-diameter pipe jacking tunneling auxiliary decision-making method for loess strata based on data-driven and deep learning, characterized in that, It includes the following steps: Step 1: Obtain the relevant data of the pipe jacking machine during construction, and preprocess and normalize the data; Step 2: Apply the VMD algorithm for signal decomposition: Input the obtained time series data, decompose it into multiple modal IMFs through VMD, and obtain the decomposed data set for training the VMD-LightGBM model; Step 3: Set the LightGBM hyperparameters, and tune the hyperparameters. Gradually adjust within a certain parameter range to find the parameter combination with the best accuracy on the validation set; Step 4: Use the dataset to train the VMD-LightGBM model, and combine 5-fold cross-validation to avoid overfitting; the output result is the predicted value of the pipe jacking attitude deviation. Compare it with the actual value, and use RMSE, MAE, and R 2 metrics to evaluate the performance of the VMD-LightGBM model, and rank the importance of the influence of different parameters on the pipe jacking attitude deviation; Step 5: Adjust the parameters determined in Step 4 according to the actual engineering situation, and set limits to form variable constraint conditions; Use the non-linear mapping function about the relationship between pipe jacking construction parameters and pipe jacking attitude fitted by the VMD-LightGBM algorithm as the optimized fitness function; Determine the value range of the NSGA-Ⅲ algorithm parameters. With the goal of minimizing the absolute value of multiple pipe jacking attitude targets, use the NSGA-Ⅲ algorithm for global optimization to determine the optimal solution of the pipe jacking construction parameters as the guiding plan for on-site construction.

2. The large-diameter pipe jacking tunneling auxiliary decision-making method for loess stratum based on data-driven and deep learning according to claim 1, characterized in that The specific steps of Step 1 include the following steps: Step 1.1: Collect data in the normal working state through the sensors and data collectors on the tunneling system; Step 1.2: Normalize the data samples of different targets and scale them proportionally to [0, 1]; Step 1.3: Use the median to fill in the missing values, use the box plot to detect outliers, and use the mean to replace the outliers.

3. The large-diameter pipe jacking tunneling auxiliary decision-making method for loess stratum based on data-driven and deep learning according to claim 1, characterized in that The specific steps of Step 2 are as follows: For the time series data, through adaptive selection of frequency, the VMD algorithm uses the following variational method and constraint conditions for optimization: where: u k (t) is the k-th intrinsic mode function, w k is the central frequency of each mode function, α is a regularization parameter that controls the smoothness of the signal decomposition, and K is the number of decomposed modes; Obtain a set of intrinsic mode functions, thereby decomposing the original data into several IMF components.

4. The large-diameter pipe jacking tunneling auxiliary decision-making method for loess stratum based on data-driven and deep learning according to claim 1, characterized in that In Step 3, the hyperparameters include: L2 regularization parameter, learning rate, and maximum depth of the tree; Use the Bayesian optimization method to tune the LightGBM hyperparameters. Select the L2 regularization parameter, learning rate, and maximum depth of the tree as the optimization parameters, use the mean squared error loss function of the training model as the objective function, and set the initial ranges of the hyperparameters in [3, 9], [0, 0.5], and [4, 8] respectively, and gradually adjust to find the parameter combination with the best accuracy on the validation set.

5. The large-diameter pipe jacking tunneling auxiliary decision-making method for loess stratum based on data-driven and deep learning according to claim 1, characterized in that The specific steps of Step 4 include the following steps: Step 4.1: Construct a VMD-LightGBM model; Step 4.2: Use the 5-fold cross-validation method to divide the entire data set into 5 parts. Among them, 4 parts are used for model training, and 1 part is used to evaluate the model performance; And repeat this process multiple times to ensure that the entire data set participates in the evaluation of the model performance; Step 4.3: Use the data set constructed in Step 4.2 to train the VMD-LightGBM model; Step 4.4: Use RMSE, MAE, and R 2 metrics to evaluate the performance of this model; Step 4.5: Through the LightGBM built-in gain algorithm, measure the contribution of each feature to the improvement of the model performance based on the feature importance of information gain, and rank the importance of each feature.

6. The large-diameter pipe jacking tunneling auxiliary decision-making method for loess strata based on data-driven and deep learning according to claim 1, characterized in that The specific steps of Step 5 include the following steps: Step 5.1: Introduce the non-linear mapping function between the pipe jacking construction parameters and the pipe jacking attitude fitted by the VMD-LightGBM algorithm as the optimized fitness function. Step 5.2: Form different construction scenarios according to the actual engineering requirements and project adjustments of each parameter, and set the limit range for the value of each parameter using the existing data to form variable constraint conditions. Step 5.3: Determine the value of the NSGA-Ⅲ algorithm parameters, aiming at minimizing the absolute values of the 4 pipe jacking attitude targets, and use the NSGA-Ⅲ method for global optimization to determine the optimal solution of the pipe jacking construction parameters.

7. The large-diameter pipe jacking tunneling auxiliary decision-making method for loess stratum based on data-driven and deep learning according to claim 6, characterized in that, In Step 5.1, the objective function of the pipe jacking attitude based on the VMD-LightGBM algorithm is: minf(x)=f1(x),f2(x),…,f4(x) f i (x) = LightGBM x1, x2, …, x 14 。 where: f i (x) is a non-linear mapping function of the relationship between the pipe jacking construction parameters and the pipe jacking attitude fitted by the VMD-LightGBM algorithm; x1, x2, …, x 14 are the pipe jacking construction parameters.

8. The large-diameter pipe jacking tunneling auxiliary decision-making method for loess strata based on data-driven and deep learning according to claim 6, characterized in that, In Step 5.3, the method of approaching the ideal solution distance is used to select an optimal solution from the optimal solution set after multi-objective optimization of many NSGA-Ⅲ algorithms; specifically as follows: For each solution x on the frontier of the optimal solution set j = f1(x j ), f2(x j ), …, f4(x j ), calculate the Euclidean distance between this solution and the ideal solution minf(x); from the Euclidean distance formula, the distance d j from the solution x j to the ideal solution minf(x) is: Select the solution that is closest to the ideal solution, i.e., minimize the distance; for each solution x j calculate its distance d j to the ideal solution, and then select the solution with the minimum distance as the optimal solution: That is, select the solution x j that minimizes the distance d j : Select a set of solutions with the smallest distance as the guiding plan for on-site construction.

Images