Shield tunneling machine tunneling attitude parameter prediction method based on XGBoost regression model

Abstract

The invention discloses a shield tunneling machine tunneling attitude parameter prediction method based on an XGBoost regression model, and the method comprises the steps: obtaining shield tunneling machine tunneling attitude parameters, including A-F group thrust oil cylinder pressure, shield tail and shield head horizontal and vertical deviations, and calculating a forward tunneling distance under the condition of maintaining the current subarea oil cylinder pressure setting; tunneling attitude parameters of the shield tunneling machine are read to construct a training data set; performing standardization processing on the training data set, performing feature analysis on the standardized data, and then dividing the data set; an XGBoost prediction model is constructed, and hyper-parameter tuning is carried out; the divided data sets are input into the optimized XGBoost prediction model to be trained, and evaluation is carried out; and predicting and evaluating the tunneling attitude parameters of the shield tunneling machine, which are obtained in real time, by using the trained model. According to the method, the advanced machine learning technology is utilized, the relevant tunneling attitude parameter values of the shield tunneling machine can be accurately predicted, and the reliability and effectiveness of engineering data prediction are powerfully guaranteed.

Description

Technical Field

[0001] The present invention relates to the technical field of engineering prediction, and particularly to a method for predicting the tunneling attitude parameters of a shield machine based on an XGBoost regression model. Background Art

[0002] In the fields of modern engineering technology and industrial production, the demand for predicting the operating parameters and optimizing the performance of various engineering equipment is increasing day by day. For example, during the tunneling construction of a shield machine, the accurate prediction of its attitude parameters plays a crucial role in ensuring construction safety, improving construction efficiency, and controlling engineering quality.

[0003] Traditional prediction methods often rely on simple empirical formulas or relatively basic linear models. In the face of complex engineering environments and the mutual influence of multiple variables, these methods are difficult to accurately capture the non-linear relationships between various factors and the attitude parameters of the shield machine. With the continuous progress of data acquisition technology, a large amount of engineering data has been recorded and stored, which provides a basis for data-driven prediction methods.

[0004] Machine learning technology has emerged and demonstrated powerful prediction capabilities in many fields. Among them, various machine learning models such as neural network models (including graph convolutional networks, recurrent neural networks, etc.) and support vector regression have been widely studied and applied. These models theoretically have the potential to handle complex non-linear relationships. However, when actually applied to the prediction of shield machine attitude parameters, they face many challenges. They usually require a large amount of historical data for effective training, and have extremely high requirements for the precise tuning of model parameters. Once the parameters are set improperly, the model is extremely prone to overfitting or underfitting, resulting in a significant reduction in the prediction accuracy of new data and being unable to stably output reliable prediction results. Summary of the Invention

[0005] Object of the Invention: To propose a method for predicting the tunneling attitude parameters of a shield machine based on an XGBoost regression model to solve the above problems existing in the prior art.

[0006] Technical Solution: A method for predicting the tunneling attitude parameters of a shield machine based on an XGBoost regression model is proposed, and the method specifically includes the following steps:

[0007] S1. Obtain the tunneling attitude parameters of the shield machine, including: the pressure of the propulsion cylinders in groups A - F, the horizontal and vertical deviations of the shield tail and the shield head, and calculate the distance of forward tunneling under the current partition cylinder pressure setting;

[0008] S2. Read the tunneling attitude parameters of the shield machine in step S1 to construct a training data set;

[0009] S3. Standardize the training dataset constructed in S2, perform feature analysis on the standardized data, and then divide the dataset;

[0010] S4. Construct an XGBoost prediction model and perform hyperparameter tuning;

[0011] S5. Input the dataset divided in step S3 into the tuned XGBoost prediction model in step S4 for training and evaluation;

[0012] S6. Use the model trained in step S5 to predict and evaluate the real-time obtained tunneling attitude parameters of the shield machine.

[0013] According to a further improvement of the present invention, the construction of the training dataset in step S2 includes:

[0014] Step S21. Data arrangement: Arrange the tunneling attitude parameters of the shield machine obtained in step S1 into a structured dataset, where each row represents the sampling data at a time point and each column represents a feature;

[0015] Step S22. Data storage: Store the arranged dataset in CSV or database format for subsequent processing and analysis.

[0016] According to a further improvement of the present invention, step S3 is specifically:

[0017] Step S31. Data standardization and outlier detection, including:

[0018] Data standardization: Use a standardizer to standardize each feature in the dataset. The standardization method is:

[0019]

[0020] where x i is the original input feature data, μ is the mean of this feature column, and σ is the standard deviation of this feature column;

[0021] Outlier detection: Based on the distribution of the standardized data, determine whether there are outliers. The judgment criterion for outliers is: If the standardized value z of a certain data point exceeds the preset range, then this point is considered an outlier. For the detected outliers, handle them according to business requirements, including deletion, replacement with the mean or median.

[0022] Step S32. Data distribution statistics and feature analysis, including:

[0023] Step S321: Conduct distribution statistics on the standardized data, calculate statistics such as the mean, standard deviation, minimum value, maximum value, and quartiles of each feature, and draw a distribution histogram or box plot to visually display the data distribution characteristics;

[0024] Step S322: Calculate the correlation matrix between features, and use the Pearson correlation coefficient to measure the linear correlation between features;

[0025] Step S323: According to the results of the correlation analysis, eliminate highly correlated redundant features to reduce the model complexity;

[0026] Step S33: Dataset division, including:

[0027] Division ratio: Divide the dataset into a training set and a test set according to a ratio of 8:2, where:

[0028] Training set: 80% of the data, used for model training and hyperparameter tuning;

[0029] Test set: 20% of the data, used for model performance evaluation.

[0030] Division method: Use the random stratified sampling method to divide the dataset to ensure that the proportion of sample categories in the training set and the test set is the same as that in the original dataset, and avoid data distribution bias.

[0031] According to a further improvement of the present invention, it is characterized in that the step S4 is specifically:

[0032] Step S41: Model construction: Use the XGBoost regression model as the prediction model, and its objective function is:

[0033] Step S42: Hyperparameter tuning: Use grid search for hyperparameter tuning, and the search range is as follows:

[0034] n_estimators: [100, 200, 300], max_depth: [3, 5, 7], learning_rate: [0.01, 0.1, 0.2], subsample: [0.8, 1.0], colsample_bytree: [0.8, 1.0];

[0035] Step S43: Evaluate the performance of different hyperparameter combinations through cross-validation, and select the hyperparameter combination that minimizes the test set error.

[0036] According to a further improvement of the present invention, the step S5 is specifically:

[0037] Step S51: Use the training set to train the tuned XGBoost model to minimize the loss function;

[0038] Step S52: Evaluate the model performance using the test set, and calculate the mean squared error (MSE), root mean squared error (RMSE), and mean absolute error (MAE) metrics as follows:

[0039]

[0040] where n represents the total number of samples, y i represents the true value, and y' i represents the predicted value;

[0041] Step S53: According to the evaluation results, further adjust the model parameters and feature engineering to improve the model performance.

[0042] According to a further improvement of the present invention, the specific step S6 is as follows:

[0043] Step S61: Input the real-time shield tunneling attitude parameters into the trained XGBoost model for prediction to obtain the prediction results;

[0044] Step S62: Evaluate the stability of the shield tunneling attitude according to the prediction results and provide adjustment suggestions.

[0045] Beneficial effects: By applying the grid search hyperparameter tuning technology, the present invention uses the trained model to predict the test set data and calculates evaluation metrics such as MSE, RMSE, and MAE to measure the accuracy and reliability of the model prediction, thereby optimizing the overall prediction effect and achieving accurate prediction of the shield tunneling attitude parameters;

[0046] The present invention uses the XGBoost model based on the optimal parameter combination as the core prediction layer to further process the feature space and optimize the received data features;

[0047] The data set collection method used in the present invention is relatively simple, and only the shield tunneling construction monitoring data needs to be obtained to meet the requirements of model training;

[0048] The present invention has low requirements for computer configuration and can be run on most personal computers, which is beneficial to further reducing the cost of shield tunneling attitude prediction. Brief Description of the Drawings

[0049] Figure 1 is the overall step flow chart of the present invention.

[0050] Figure 2 is the comparison chart of the predicted value and the true value of the shield head horizontal deviation parameter of the present invention.

[0051] Figure 3 is the comparison chart of the predicted value and the true value of the shield head vertical deviation parameter of the present invention.

[0052] Figure 4 This is a comparison chart of the predicted value and the true value of the shield tail horizontal deviation of the parameters of the present invention's solution.

[0053] Figure 5 This is a comparison chart of the predicted value and the true value of the shield tail vertical deviation of the parameters of the present invention's solution. Specific implementation manners

[0054] In the following description, numerous specific details are given to provide a more thorough understanding of the present invention. However, it is obvious to those skilled in the art that the present invention can be implemented without one or more of these details. In other examples, in order to avoid confusion with the present invention, some technical features known to the public are not described.

[0055] The applicant believes that the XGBoost regression model based on the decision tree integration idea has gradually received attention. The XGBoost model has high computational performance, good scalability, and powerful feature processing capabilities. When processing shield machine-related data, it is first necessary to read data from engineering data records, which often contain multiple data points related to the cylinder pressure and historical values and distance information at specific positions, etc. These data constitute the input features and target outputs of the prediction model. In the data preprocessing stage, by standardizing the input features, data of different magnitudes can have a unified scale, avoiding adverse effects on model training due to excessive differences in data magnitudes. Further, in order to improve the prediction performance of the XGBoost model, the use of hyperparameter tuning technology becomes a key step. By setting a series of candidate values of hyperparameters, such as the number of decision trees, the maximum depth of decision trees, the learning rate, etc., and using the GridSearchCV tool for cross-validation and hyperparameter search, the best combination of hyperparameters can be found within the given parameter space, thereby optimizing the fitting ability and generalization performance of the model. During the model training process, detailed recording of training information helps to analyze the learning process and convergence situation of the model and timely discover potential problems. At the same time, considering the possible overfitting phenomenon of the model, the early stopping method mechanism is introduced. By setting a validation set during the training process and monitoring the performance metrics of the model on the validation set, when the model performance no longer improves after several consecutive rounds of training, the training is terminated in advance to avoid overfitting the training data and improve the adaptability of the model to new data.

[0056] In summary, the prediction method based on the XGBoost regression model shows unique advantages and application potential in the field of shield machine attitude parameter prediction, and is expected to provide more accurate and reliable technical support for engineering construction.

[0057] Therefore, the present invention proposes a method for predicting the tunneling attitude parameters of a shield machine based on the XGBoost regression model, asFigure 1 As shown in the figure, the method specifically includes the following steps:

[0058] S1. Obtain the tunneling attitude parameters of the shield machine, including: the pressures of the propulsion cylinders in groups A - F, the tail of the shield, and the horizontal and vertical deviations of the shield head, and calculate the distance of forward tunneling under the current partition cylinder pressure setting;

[0059] S2. Read the tunneling attitude parameters of the shield machine in step S1 to construct a training data set;

[0060] S3. Perform standardization processing on the training data set constructed in S2, conduct feature analysis on the standardized data, and then divide the data set;

[0061] S4. Construct an XGBoost prediction model and perform hyperparameter tuning;

[0062] S5. Input the data set divided in step S3 into the optimized XGBoost prediction model in step S4 for training and conduct evaluation;

[0063] S6. Use the model trained in step S5 to predict and evaluate the real - time obtained tunneling attitude parameters of the shield machine.

[0064] In this embodiment, the data used in step S1 comes from project monitoring data. The data set includes the pressures of the propulsion cylinders in groups A - F, the tail of the shield, the horizontal and vertical deviations of the shield head, and the calculated distance of forward tunneling under the current partition cylinder pressure setting; the number is 66,571 groups.

[0065] According to a further improvement of the present invention, the construction of the training data set in step S2 includes:

[0066] Step S21. Data arrangement: Arrange the tunneling attitude parameters of the shield machine obtained in step S1 into a structured data set. Each row represents the sampling data at a time point, and each column represents a feature;

[0067] Step S22. Data storage: Store the arranged data set in CSV or database format for subsequent processing and analysis.

[0068] According to a further improvement of the present invention, step S3 is specifically as follows:

[0069] Step S31. Data standardization and outlier detection, including:

[0070] Data standardization: Use a standardizer to perform standardization processing on each feature in the data set. The standardization method is:

[0071]

[0072] where xi Let the original input feature data be \(X\), \(\mu\) be the mean of this feature column, and \(\sigma\) be the standard deviation of this feature column;

[0073] Outlier detection: Based on the distribution of the standardized data, determine whether there are outliers; the criterion for outlier judgment is: if the standardized value \(z\) of a certain data point exceeds the preset range, then this point is considered an outlier. For the detected outliers, handle them according to business requirements, including deletion, replacement with the mean or median.

[0074] Step S32, Data distribution statistics and feature analysis, including:

[0075] Step S321, Conduct distribution statistics on the standardized data, calculate statistics such as the mean, standard deviation, minimum value, maximum value, and quartiles of each feature, and draw a distribution histogram or box plot to visually display the data distribution characteristics;

[0076] Step S322, Calculate the correlation matrix between features, and use the Pearson correlation coefficient to measure the linear correlation between features;

[0077] Step S323, According to the results of the correlation analysis, eliminate highly correlated redundant features to reduce the model complexity;

[0078] Step S33, Dataset division, including:

[0079] Division ratio: Divide the dataset into a training set and a test set in a ratio of 8:2, where:

[0080] Training set: 80% of the data, used for model training and hyperparameter tuning;

[0081] Test set: 20% of the data, used for model performance evaluation.

[0082] Division method: Use the random stratified sampling method to divide the dataset to ensure that the proportion of various category samples in the training set and the test set is the same as that in the original dataset, and avoid data distribution deviation.

[0083] According to a further improvement of the present invention, it is characterized in that the step S4 is specifically:

[0084] Step S41, Model construction: Use the XGBoost regression model as the prediction model, and its objective function is:

[0085] Step S42, Hyperparameter tuning: Use grid search for hyperparameter tuning, and the search range is as follows:

[0086] n_estimators: [100, 200, 300], max_depth: [3, 5, 7], learning_rate: [0.01, 0.1, 0.2], subsample: [0.8, 1.0], colsample_bytree: [0.8, 1.0];

[0087] Step S43: Evaluate the performance of different hyperparameter combinations through cross-validation, and select the hyperparameter combination that minimizes the test set error.

[0088] Among them, step S42 specifically includes: using GridSearchCV to perform hyperparameter tuning on the XGBoost regression model. During the hyperparameter tuning process, set a parameter grid that covers parameters with different value ranges such as "n_estimators", "max_depth", "learning_rate", "subsample", "colsample_bytree", "min_child_weight", etc. Through cross-validation and a specific evaluation metric "neg_mean_squared_error", screen out the best parameter combination from numerous hyperparameter combinations to optimize the performance of the XGBoost regression model, so as to better predict the tunneling attitude parameters of the shield machine.

[0089] In this embodiment, construct a relevant architecture based on the XGBoost regression model for predicting the tunneling attitude parameters of the shield machine, and at the same time clarify the corresponding loss function and hyperparameter settings:

[0090] This embodiment creates a method based on the XGBoost regression model to achieve accurate prediction of the tunneling attitude parameters of the shield machine. Among them, starting from the data processing link, read the data file containing various relevant data, and then define the input columns and output columns to distinguish features and targets.

[0091] Subsequently, use StandardScaler to standardize the input features to make the data within a suitable scale range, which is more conducive to subsequent model training. Then, divide the training set and test set according to a certain ratio (here the test set accounts for 0.2 and the random seed is set to 42) to prepare the data basis for model training and evaluation.

[0092] In terms of hyperparameter tuning, a parameter grid param_grid is set, which covers candidate values of multiple key hyperparameters such as "n_estimators" (with different value ranges set as [100, 200, 300], etc.), "max_depth" (the value range includes [5, 6, 7], etc.), "learning_rate", "subsample", "colsample_bytree", "min_child_weight", etc. By using GridSearchCV in combination with cross-validation (cv = 5) and taking "neg_mean_squared_error" as the evaluation metric, the best combination of hyperparameters is searched for. Among them, neg_mean_squared_error acts as a role similar to a loss function, which is used to measure the difference between the model prediction results and the true values under different hyperparameter combinations, and guides the search for the parameter configuration that optimizes the model performance.

[0093] After obtaining the best parameters, the model is trained using them, and then predictions are made on the test set. The accuracy of the model predictions is evaluated by calculating metrics such as MSE, RMSE, and MAE, which measure the quality of the model's predictions of the target parameters from different perspectives. In addition, an early stopping mechanism is introduced. During training, some functions of the test set are used to simulate the validation set to monitor the model performance, avoiding the model overfitting the training data, and ensuring that the model can accurately predict the tunneling attitude parameters and prompt anomalies stably and accurately in practical applications.

[0094] According to a further improvement of the present invention, step S5 is specifically as follows:

[0095] Step S51: Use the training set to train the tuned XGBoost model to minimize the loss function;

[0096] Step S52: Use the test set to evaluate the model performance and calculate the mean squared error MSE, root mean squared error RMSE, and mean absolute error MAE metrics as follows:

[0097]

[0098] Where n represents the total number of samples, y i represents the true value, and y' i represents the predicted value;

[0099] Step S53: According to the evaluation results, further adjust the model parameters and feature engineering to improve the model performance.

[0100] According to a further improvement of the present invention, step S6 is specifically as follows:

[0101] Step S61: Input the real-time obtained tunneling attitude parameters of the shield machine into the trained XGBoost model for prediction to obtain the prediction result;

[0102] Step S62: Evaluate the stability of the tunneling attitude of the shield machine according to the prediction result and provide adjustment suggestions.

[0103] As described above, although the present invention has been shown and described with reference to specific preferred embodiments, it should not be construed as a limitation of the present invention itself. Various changes in form and detail may be made without departing from the spirit and scope of the present invention defined by the appended claims.

Claims

1. A prediction method for the tunneling attitude parameters of a shield machine based on the XGBoost regression model, characterized in that, It includes the following steps: S1. Obtain the tunneling attitude parameters of the shield machine, including: the pressures of the propulsion cylinders in groups A - F, the horizontal and vertical deviations of the shield tail and the shield head, and calculate the distance of forward tunneling under the current partition cylinder pressure settings; S2. Read the tunneling attitude parameters of the shield machine in step S1 to construct a training data set; S3. Perform standardization processing on the training data set constructed in S2, conduct feature analysis on the standardized data, and then divide the data set; S4. Construct an XGBoost prediction model and perform hyperparameter tuning; S5. Input the data set divided in step S3 into the optimized XGBoost prediction model in step S4 for training and evaluation; S6. Use the model trained in step S5 to predict and evaluate the real - time obtained tunneling attitude parameters of the shield machine.

2. The shield tunneling attitude parameter prediction method based on the XGBoost regression model according to claim 1, characterized in that The construction of the training data set described in step S2 includes: Step S21. Data arrangement: Arrange the tunneling attitude parameters of the shield machine obtained in step S1 into a structured data set. Each row represents the sampling data at a time point, and each column represents a feature; Step S22. Data storage: Store the arranged data set in the CSV or database format for subsequent processing and analysis.

3. The shield tunneling attitude parameter prediction method based on the XGBoost regression model according to claim 1, wherein The specific steps of S3 are as follows: Step S31. Data standardization and outlier detection, including: Data standardization: Use a standardizer to perform standardization processing on each feature in the data set. The standardization method is: where x i is the original input feature data, μ is the mean of this feature column, and σ is the standard deviation of this feature column; Outlier detection: Based on the distribution of the standardized data, determine whether there are outliers. The criterion for outlier judgment is: If the standardized value z of a certain data point exceeds the preset range, then this point is considered an outlier. For the detected outliers, handle them according to business requirements, including deletion, replacement with the mean or median. Step S32. Data distribution statistics and feature analysis, including: Step S321. Conduct distribution statistics on the standardized data, calculate statistics such as the mean, standard deviation, minimum value, maximum value, and quartiles of each feature, and draw a distribution histogram or box plot to visually display the data distribution characteristics; Step S322. Calculate the correlation matrix between features and use the Pearson correlation coefficient to measure the linear correlation between features; Step S323. According to the results of the correlation analysis, eliminate highly correlated redundant features to reduce the model complexity; Step S33. Data set division, including: Division ratio: Divide the data set into a training set and a test set according to a ratio of 8:2, where: Training set: 80% of the data, used for model training and hyperparameter tuning; Test set: 20% of the data, used for model performance evaluation. Division method: Use the random stratified sampling method to divide the data set to ensure that the proportion of various category samples in the training set and the test set is the same as that in the original data set, avoiding data distribution deviation.

4. The method for predicting the tunneling attitude parameters of a shield machine based on the XGBoost regression model according to claim 1, wherein The specific steps of S4 are as follows: Step S41. Model construction: Use the XGBoost regression model as the prediction model, and its objective function is: Step S42. Hyperparameter tuning: Use grid search for hyperparameter tuning. The search range is as follows: n_estimators: [100, 200, 300], max_depth: [3, 5, 7], learning_rate: [0.01, 0.1, 0.2], subsample: [0.8, 1.0], colsample_bytree: [0.8, 1.0]; Step S43: Evaluate the performance of different hyperparameter combinations through cross-validation, and select the hyperparameter combination that minimizes the error of the test set.

5. The shield tunneling attitude parameter prediction method based on the XGBoost regression model according to claim 1, characterized in that The specific content of step S5 is as follows: Step S51: Use the training set to train the tuned XGBoost model to minimize the loss function; Step S52: Use the test set to evaluate the model performance, and calculate the mean squared error MSE, root mean squared error RMSE, and mean absolute error MAE metrics. The method is as follows: Among them, n represents the total number of samples, y i represents the true value, and y' i represents the predicted value; Step S53: According to the evaluation results, further adjust the model parameters and feature engineering to improve the model performance.

6. The shield tunneling attitude parameter prediction method based on the XGBoost regression model according to claim 1, characterized in that The specific content of step S6 is as follows: Step S61: Input the real-time obtained shield tunneling attitude parameters into the trained XGBoost model for prediction to obtain the prediction results; Step S62: Evaluate the stability of the shield tunneling attitude according to the prediction results and provide adjustment suggestions.

Images