Deep-buried TBM rock machine parameter fusion prediction method and system based on CNN-LSTM-Adaboost model

Abstract

The invention discloses a deep-buried TBM rock machine parameter fusion prediction method and system based on a CNN-LSTM-Adaboost model, and belongs to the field of intelligent data algorithms, and the method comprises the steps: S1, building an initial CNN-LSTM-Adaboost surrounding rock grade prediction model; s2, establishing an evaluation index and effect index model; s3, optimizing the model; s4, predicting tunneling parameters under different surrounding rock grades by using the optimized model; and S5, outputting a prediction result. The model provided by the invention combines the LSTM model and the CNN model, and can cope with relatively complex stratum conditions. The CNN-LSTM model overcomes the problems that the LSTM is poor in space processing capability and the CNN is insensitive to a time sequence. Compared with the single use of LSTM or CNN, the prediction accuracy of the model is improved by 10%; the prediction relative error rate of the data is within 10%, the root mean square error is within 5, the average absolute percentage error value is within 10%, and the goodness of fit value is 0.9 or above; according to the model, the recognition accuracy of the tunnel surrounding rock grade based on the tunneling parameters is high, and the comprehensive accuracy reaches 90%.

Description

Technical Field

[0001] The present invention relates to the field of intelligent data algorithm technology, and in particular to a deep buried TBM rock machine parameter fusion prediction method and system based on a CNN-LSTM-Adaboost model. Background Art

[0002] Tunnel boring machines (TBMs) are increasingly being used in the construction of ultra-long tunnels. Compared with traditional drilling and blasting methods, TBMs have the advantages of low safety risks, fast construction speed, low environmental disturbance, good construction conditions, and high engineering quality. However, TBMs have weak adaptability to different geological conditions during deep tunnel excavation, which places higher operational requirements on TBM drivers. Drivers are required to adjust mechanical parameters based on their personal experience, which often leads to a mismatch between driving operations and geological conditions, affecting TBM excavation efficiency, and even damaging the cutterhead, leading to tunnel surrounding rock collapse and other safety accidents. Therefore, accurate prediction of TBM excavation parameters and accurate prediction of tunnel surrounding rock grades are of great significance to improving TBM excavation efficiency and safety level, and also lay the foundation for realizing TBM intelligent excavation.

[0003] In the field of TBM excavation parameter prediction, many scholars have introduced machine learning and deep learning algorithms into their research to obtain better parameter prediction results. By comparing existing machine learning models, it can be found that the long short time series-adaptive boosting prediction model (LSTM-Adaboost) has the most outstanding processing effect on long-term fluctuation sequence data, but the parameters of the surrounding rock and TBM excavator used are relatively simple, which makes it difficult to apply to more complex formation environments. Summary of the invention

[0004] In order to solve the technical problem that the parameters of the surrounding rock and TBM tunneling machine used by LSTM in the above-mentioned prior art are relatively simple and difficult to be applied to more complex formation environments, the present invention provides a deep buried TBM rock machine parameter fusion prediction method and system based on the CNN-LSTM-Adaboost model.

[0005] In order to realize the above technical solution, the present invention provides a deep buried TBM rock machine parameter fusion prediction method based on CNN-LSTM-Adaboost model, which is characterized by comprising the steps of:

[0006] S1: Establish the initial CNN-LSTM-Adaboost surrounding rock grade prediction model;

[0007] S2: Establish the relative error rate, root mean square error, mean absolute percentage error and goodness of fit evaluation indicators and effect indicators of the model output excavation parameter prediction;

[0008] S3: Based on the TBM excavation data and tunnel surrounding rock geological data, and combined with the evaluation indicators and effect indicators in step S2, the CNN-LSTM-Adaboost model is optimized;

[0009] S4: Use the optimized model to predict the excavation parameters under different surrounding rock grades;

[0010] S5: Output the obtained prediction results to predict the tunnel surrounding rock grade.

[0011] Furthermore, in step S1, the process of establishing the CNN-LSTM-Adaboost surrounding rock grade identification model includes:

[0012] Create a training set:

[0013] T={(x1,y1),(x2,y2),…(x N ,y N )}

[0014] Set the weight distribution D1 of the initialization data to uniform distribution:

[0015]

[0016] Calculate the weak classifier G m (x) Classification error ε on the dataset m :

[0017]

[0018] Calculate the weak classifier G m (x) Weight α in the strong classifier m :

[0019]

[0020] Update the weight distribution of the training set:

[0021]

[0022]

[0023] Determine the number of iterations N, and if it reaches N, combine the weak classifiers:

[0024]

[0025] The strong classifier G(x) is obtained through the relationship function sign:

[0026]

[0027] Furthermore, the operation process of the CNN-LSTM-Adaboost model includes the following steps:

[0028] S11: Import TBM excavation data and data of tunnels with different grades of surrounding rock;

[0029] S12: Establishing data set;

[0030] S13: Perform data initialization processing;

[0031] S14: 90% of the TBM excavation data obtained under different surrounding rock grades are selected as the model training data set; 10% of the TBM excavation data obtained under different surrounding rock grades are selected as the model test set;

[0032] S15: According to the analysis of whether the data in the training data set conforms to CART, the conforming data is classified into a strong classifier, and the non-conforming data is classified into a weak classifier;

[0033] S16: After testing the data in the test set, the data is classified into a strong classifier;

[0034] S17: Obtain classification results.

[0035] Furthermore, in step S2, the calculation formula of the evaluation index is as follows:

[0036] Relative error rate calculation:

[0037]

[0038] Goodness of fit calculation:

[0039]

[0040] Root mean square error calculation:

[0041]

[0042] Root mean square error calculation:

[0043]

[0044] Furthermore, in step S2, the calculation formula of the effect index ACC is as follows:

[0045]

[0046]

[0047]

[0048]

[0049]

[0050] Furthermore, in step S3, the preferred steps of the CNN-LSTM-Adaboost model are as follows:

[0051] 90% of the TBM excavation data obtained under different surrounding rock grades were selected as the model training data set;

[0052] 10% of the TBM excavation data obtained under different surrounding rock grades were selected as the model test set;

[0053] The initial learning rate of the LSTM structural unit is set to 0.005, and the CNN structure is a single-layer convolution. After 20 repeated runs, the damage function value drops to 0.1.

[0054] Roulette wheel selection is used as the validation technique, and the validation frequency is once every 40 iterations, where the gating unit adopts ten-fold cross validation;

[0055] Set up 200 Adaboost single-base classifiers, with a minimum fork number of 4 and a maximum tree height of 20;

[0056] The excavation parameters F, T, N, and v under the four levels of surrounding rock are AR 、v PR and E SE Input the model training model, output one of the tunneling parameters in turn, and train it multiple times to improve the model prediction accuracy.

[0057] Furthermore, a deep buried TBM rock machine parameter fusion prediction system based on the CNN-LSTM-Adaboost model measures the cutterhead control parameters through equipment installed on the TBM cutterhead drive device and drive motor, and inputs all the imported geological exploration data and excavation daily report data into the computer to establish a parameter database.

[0058] Furthermore, the cutterhead control parameters include cutterhead thrust, cutterhead torque and cutterhead speed; the geological exploration data include TBM excavation specific energy and surrounding rock grade; and the excavation daily report data include construction speed and net excavation speed.

[0059] The beneficial effects of the present invention are:

[0060] The present invention provides a deep buried TBM rock-machine parameter fusion prediction method based on the CNN-LSTM-Adaboost model, which combines the long-short time series-adaptive boosting prediction model (LSTM-Adaboost) and the deep learning algorithm to optimize the LSTM model, thereby further improving the prediction accuracy of TBM excavation parameters and surrounding rock grade.

[0061] Compared with the prior art, the CNN-LSTM-Adaboost model provided by the present invention combines the long-short time series-adaptive boost prediction model (LSTM-Adaboost) and the deep learning algorithm to optimize the LSTM model. In other words, the model provided by the present invention can combine multiple surrounding rock parameters and / or TBM parameters, so that the CNN-LSTM-Adaboost model can cope with more complex formation conditions. In addition, compared with the use of the long-short time series-adaptive boost prediction model (LSTM-Adaboost) or the LSTM model alone, the present invention is based on F, T, N, v AR ,v PR and E SE The data set established by six TBM excavation parameters was used to train the CNN-LSTM-Adaboost model to predict the excavation stable section data. The relative error rate of data prediction was basically within 10%, and the prediction accuracy was high, which is of great significance for actual engineering construction.

[0062] In addition, the overall prediction success rate of the LSTM-Adaboost model is 90%, which meets the model training effect, and the data prediction success rate in the Grade V surrounding rock interval is 100%, indicating that the model has a high prediction accuracy for actual data samples. The high prediction accuracy for soft surrounding rock grades is of great significance to the safe excavation of TBMs.

[0063] The CNN-LSTM-Adaboost fusion prediction model has a good prediction effect on various rock machine parameters. In actual TBM excavation work, it can be used for TBM excavation project planning and treatment of high-risk tunnel sections to improve the safety level of TBM excavation. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0065] Figure 1 A flowchart of a deep buried TBM rock-machine parameter fusion prediction method based on a CNN-LSTM-Adaboost model provided by the present invention;

[0066] Figure 2 This is the operation flow chart of the CNN-LSTM-Adaboost surrounding rock grade identification model;

[0067] Figure 3 The comparison results of the predicted and actual values ​​of the TBM excavation parameters for Class II surrounding rock;

[0068] Figure 4 The comparison results of predicted and actual values ​​of TBM excavation parameters for Class III surrounding rock;

[0069] Figure 5 The comparison results of the predicted and actual values ​​of TBM excavation parameters for Grade IV surrounding rock;

[0070] Figure 6 The comparison results of predicted and actual values ​​of TBM excavation parameters for Grade V surrounding rock;

[0071] Figure 7 Relative error rate of tunneling parameter prediction for (a) Class II surrounding rock, (b) Class III surrounding rock, (c) Class IV surrounding rock, and (d) Class V surrounding rock;

[0072] Figure 8 Output the confusion matrix for the CNN-LSTM-Adaboost model;

[0073] Fig. 9 This is the gap diagram between the surrounding rock grade prediction results of the CNN-LSTM-Adaboost model and the actual surrounding rock;

[0074] Fig.10 This is a comparison chart of the prediction accuracy of the CNN-LSTM-Adaboost model and the traditional machine learning model;

[0075] Fig.11 It is a deep buried TBM rock-machine parameter fusion prediction system based on the CNN-LSTM-Adaboost model. DETAILED DESCRIPTION

[0076] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments can be combined with each other. The present invention will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0077] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0078] Unless otherwise specifically stated, the relative arrangement of the parts and steps described in these embodiments, numerical expressions and numerical values ​​do not limit the scope of the present invention. At the same time, it should be understood that, for ease of description, the sizes of the various parts shown in the accompanying drawings are not drawn according to the actual proportional relationship. The technology, methods and equipment known to ordinary technicians in the relevant field may not be discussed in detail, but in appropriate cases, the technology, methods and equipment should be regarded as a part of the authorization specification. In all examples shown and discussed here, any specific value should be interpreted as being merely exemplary, rather than as a limitation. Therefore, other examples of exemplary embodiments may have different values. It should be noted that similar reference numerals and letters represent similar items in the following drawings, so once a certain item is defined in an accompanying drawing, it does not need to be further discussed in subsequent drawings.

[0079] See also Figure 1 As shown, the present invention provides a deep buried TBM rock machine parameter fusion prediction method based on CNN-LSTM-Adaboost model, comprising the steps of:

[0080] S1: Establish the initial CNN-LSTM-Adaboost surrounding rock grade prediction model;

[0081] S2: Establish the relative error rate, root mean square error, mean absolute percentage error and goodness of fit evaluation indicators of the output tunneling parameter prediction of the CNN-LSTM-Adaboost model;

[0082] S3: Optimizing the CNN-LSTM-Adaboost model based on TBM excavation data and tunnel surrounding rock geological data;

[0083] S4: Use the optimized model to predict the excavation parameters under different surrounding rock grades;

[0084] S5: Output the obtained prediction results.

[0085] As a preferred method, in step S1, the CNN-LSTM-Adaboost (adaptive boost) algorithm is a serial integration algorithm, and its classification model implements the iteration of classification data by increasing the contribution of the sample data that was misclassified in the previous iteration. At the same time, the weighted majority voting method is used to increase the weight of the individual learner of the correct sample, and the weighted data set continues to train the next batch of classifiers. The establishment process of the CNN-LSTM-Adaboost surrounding rock grade recognition model is as follows:

[0086] 1. Create a training set:

[0087] T={(x1,y1),(x2,y2),…(x N ,y N)} (1)

[0088] in,

[0089] 2. Set the weight distribution D1 of the initialization data to uniform distribution, then the weight of each sample in the initialization data set is w1:

[0090]

[0091] Where i=1,2,…,N.

[0092] 3. Perform iterative operations to form several weak classifiers G m (x i ), calculate the classification error of the weak classifier on the data set:

[0093]

[0094] 4. Calculate the weak classifier G m (x) Weight α in the strong classifier m :

[0095]

[0096] 5. Update the weight distribution of the training set:

[0097]

[0098] Where i=1,2,…,N.

[0099]

[0100] When a data sample is classified correctly, its weight coefficient will decrease; the weight coefficient of an incorrectly classified sample will increase.

[0101] 6. Determine whether the number of iterations N has been reached. If it has been reached, combine the weak classifiers, that is:

[0102]

[0103] Then, the strong classifier G(x) is obtained through the relational function sign, that is:

[0104]

[0105] As a preferred embodiment, Figure 2 As shown, the operation process of the CNN-LSTM-Adaboost model includes the following steps:

[0106] S11: Import TBM excavation data and data of tunnels with different grades of surrounding rock;

[0107] S12: Establishing data set;

[0108] S13: Perform data initialization processing;

[0109] S14: selecting 90% of the TBM excavation data obtained under different surrounding rock grades as a model training data set;

[0110] 10% of the TBM excavation data obtained under different surrounding rock grades were selected as the model test set;

[0111] S15: According to the analysis of whether the data in the training data set conforms to CART, the conforming data is classified into a strong classifier, and the non-conforming data is classified into a weak classifier;

[0112] S16: After testing the data in the test set, the data is classified into a strong classifier;

[0113] S17: Obtain classification results.

[0114] During the operation of the CNN-LSTM-Adaboost model, TBM excavation data and tunnel surrounding rock geological data are collected and integrated into a tunnel surrounding rock grade prediction data set. 90% of the data set is used as the training parameters of the CNN-LSTM-Adaboost model. After the model training is completed, 10% of the data is used as the model test set data to test the training effect of the model and output evaluation parameters such as the training accuracy of the model.

[0115] As a preferred method, in step S2, the evaluation indicators of the relative error rate (Er), root mean square error (RMSE), mean absolute percentage error (MAPE) and goodness of fit (R2) of the tunneling parameter prediction output by the CNN-LSTM-Adaboost model are established, and the calculation formula of the evaluation indicators is as follows:

[0116]

[0117]

[0118]

[0119]

[0120] Where: n is the number of samples, y i is the sample measured value, f(x i ) is the sample prediction value, is the sample mean.

[0121] As a preferred embodiment, in step S2, the CNN-LSTM-Adaboost classification model selects accuracy (ACC) and F1 value as classification effect indicators. The calculation formula of the effect indicator ACC is as follows:

[0122]

[0123]

[0124]

[0125]

[0126]

[0127] Where: D i,j Indicates the actual level and predicted level of sample prediction, F1 i , P i and R i Represents the F1 value, prediction accuracy and the number of surrounding rock grade i. The ACC value and F1 value range from 0 to 1, with 1 representing the highest accuracy and 0 representing the lowest.

[0128] As a preferred method, in step S3, 90% of the TBM excavation data under four levels of surrounding rock, namely, level II, level III, level IV, and level V, are used as the model training data set, and 10% are used as the model test set to establish a CNN-LSTM-Adaboost excavation parameter prediction model. The maximum number of iterations of the LSTM structural unit is 1000 times, the initial learning rate is 0.005, and the CNN structure is a single-layer convolution. After 20 repetitions, the damage function drops to 0.1. Roulette selection is used as the verification technology, the verification frequency is once every 40 iterations, and the gate unit adopts ten-fold cross validation. 200 Adaboost single-base classifiers are set, the minimum number of forks is 4, and the maximum tree height is 20.

[0129] The excavation parameters F, T, N, and v under the four levels of surrounding rock are AR 、v PR and E SE The training set of the CNN-LSTM-Adaboost model is established, the model training model is input, and one of the tunneling parameters is output in turn. After multiple trainings, the model prediction accuracy is improved. The prediction results of the model are as follows Figures 3 to 6 shown.

[0130] In step S4, actual cases are used to verify the reliability of the deep buried TBM rock-machine parameter fusion prediction method based on the CNN-LSTM-Adaboost model.

[0131] The practical data for this project comes from a TBM tunnel construction project of a water diversion project in Xinjiang. The tunnel is deep and the crossing distance is long. The tunnel depth is generally between 300 and 500 meters. The data sample construction section is 43.861 km long and the excavation surface diameter is 8 meters. The strata crossed by the TBM construction section are Huali granite or granite gneiss, and some tunnel sections have high ground pressure characteristics. The TBM model of the data collection section is an open TBM model QJY078 designed and produced by Northern Heavy Industries Group.

[0132] According to the preliminary investigation data of the construction tunnel section and the water conservancy and hydropower survey and design specifications, the tunnel surrounding rock grade is divided into four grades: Grade II, Grade III, Grade IV, and Grade V. Among the four types of surrounding rock, Grade III and Grade IV surrounding rock account for a larger proportion, and the Grade V surrounding rock construction section is shorter. The research data was collected from the underground tunnel section of KS216+522.2~KS214+227.4m. One set of data is the average value of the stable section of the TBM excavation cycle section. A total of 1150 sets of data of four grades of surrounding rock excavation parameters were collected, including 250 sets of Grade II, 550 sets of Grade III, Grade IV, and Grade V surrounding rock data, 200 sets of Grade V, and 150 sets of Grade II, Grade III, Grade IV, and Grade V surrounding rock data, respectively.

[0133] CNN-LSTM-Adaboost model predicts TBM excavation parameter output E r , RMSE, MAPE and R2, the four indicators are used to analyze the accuracy and goodness of fit of the model in predicting TBM excavation parameters. The calculation results of the prediction relative error rate of the model under the four levels of surrounding rock are shown in the figure. Figure 7 The calculation results of RMSE, MAPE and R2 are shown in Table 1.

[0134] Table 1 Error analysis of prediction results of CNN-LSTM-Adaboost model

[0135]

[0136] According to Table 1, the overall error calculation results of various tunneling parameters obtained are kept at a low level, indicating that the CNN-LSTM-Adaboost model has a good prediction effect on TBM tunneling parameters and plays an important role in guiding TBM engineering practice.

[0137] from Figure 7 According to the relative error rate results, in the II-level surrounding rock, some v in the TBM excavation parameters AR Predicted E r The absolute value is greater than 0.1, and the average E r is 0.064, and the other parameters E r The absolute values ​​are all less than 0.1; in the Class III surrounding rock, some F, v AR E r The absolute value is greater than 0.1, F, v AR The average Er is 0,0678,0.0713, and the prediction E of the remaining parameters r The absolute values ​​are all less than 0.1; in the IV-level surrounding rock, except for v AR Prediction of other TBM excavation parameters except E r The absolute value is less than 0.1, part of v AR E r The absolute value is greater than 0.1, v AR The average E r is 0.074; in the Ⅴ-level surrounding rock, some N and v AR E r The absolute value is greater than 0.1, N, v AR The average E r The predicted E of other TBM excavation parameters is 0.0685, 0.0724 r The absolute values ​​are all less than 0.1. With the decrease of surrounding rock grade, the relative error rate of the model in predicting various tunneling parameters increases, indicating that the surrounding rock conditions have a certain influence on the prediction accuracy of the model.

[0138] By comparison, it can be found that among the six TBM excavation parameters, v PR The prediction relative error rate of is the largest among all parameters, and the fluctuation of parameter prediction is also large, indicating that the model has a low prediction accuracy. Considering the actual working conditions of TBM on site, the construction speed is not only related to the TBM excavation situation, but also to the construction progress of the connecting section between tunnels, the workers' handling of emergencies, TBM step change, and machine jam escape. These factors are not closely related to TBM excavation parameters, so they have a great impact on the prediction accuracy of the model. As the surrounding rock grade decreases, the R2 value predicted by the model is also smaller, indicating that the model has a high prediction accuracy for the excavation parameters of the surrounding rock with stable geological conditions. At the same time, among the four types of surrounding rock, the average MAPE of grade V surrounding rock is 4.995%, indicating that the lower the surrounding rock grade, the greater the prediction error of the CNN-LSTM model for excavation parameters. At the same time, as the stability of the surrounding rock decreases, the predicted RMSE value of the CNN-LSTM model will also increase, indicating that the influence of human factors on rock-machine interaction increases, which reduces the prediction accuracy of the machine learning model. On the other hand, the tunnel sections with lower surrounding rock grades have less data, which affects the training accuracy of the machine learning model and reduces the accuracy of the model's prediction of parameters. Among the four types of surrounding rock, the RMSE value of the excavation parameter prediction decreases as the surrounding rock grade decreases, indicating that the model's prediction error for the excavation parameters in the weak surrounding rock section will increase.

[0139] In step S5, the tunnel surrounding rock grade is predicted according to the deep buried TBM rock-machine parameter fusion prediction method based on the CNN-LSTM-Adaboost model provided by the present invention.

[0140] like Figure 8 As shown in the figure, 40 sets of TBM parameter data are input into the CNN-LSTM-Adaboost model with bifurcation tree type set, and the surrounding rock grade confusion matrix is ​​output. From the confusion matrix, it can be seen that the prediction accuracy of the model for the grade II and grade IV surrounding rock test sets is 100%, and the prediction accuracy of grade III and grade V surrounding rock is 90% and 87.5% respectively, which meets the maximum ACC data value requirement for model training.

[0141] In a tunnel with known surrounding rock grade, 10 groups of TBM excavation parameters are selected for each surrounding rock to form a test data set. The trained LSTM-Adaboost model is used to predict the surrounding rock grade. The output results are compared with the surrounding rock grade to verify the training effect of the model. The comparison results are as follows: Fig. 9 shown.

[0142] The prediction results of the validation data set show that the overall prediction success rate of the LSTM-Adaboost model is 90%, which meets the model training effect, and the data prediction success rate in the grade V surrounding rock interval is 100%, indicating that the model has a high prediction accuracy for actual data samples, and the high prediction accuracy for soft surrounding rock grades is of great significance to the safe excavation of TBMs.

[0143] In order to compare the difference between the deep buried TBM rock machine parameter fusion prediction method based on the CNN-LSTM-Adaboost model provided by the present invention and other classic machine learning models in the prediction of tunnel surrounding rock grade, the present invention compares the ratio between the model and common models such as SVC, ANN, KNN and other models. The collected tunnel TBM excavation data samples are used to train these models, and the obtained model training results are compared with the CNN-LSTM-Adaboost model. Table 2 lists the optimal setting parameters of each machine learning model. Fig.10 Represents the accuracy calculated for each machine learning validation data sample.

[0144] Table 2 Machine learning model parameter settings

[0145]

[0146] From the ACC value and F1 of the prediction results of each machine learning model, the ACC / F1 values ​​of the CNN-LSTM-Adaboost, SVC, ANN and KNN models are 0.857 / 0.764, 0.776 / 0.677, 0.779 / 0.677, and 0.783 / 0.674, respectively. The CNN-LSTM-Adaboost model has the highest prediction accuracy, so it is the best choice to select the CNN-LSTM-Adaboost model to predict the surrounding rock grade.

[0147] As a preferred method, a deep buried TBM rock machine parameter fusion prediction system based on CNN-LSTM-Adaboost model is proposed. Fig.11 As shown in the figure, during the TBM excavation process, the TBM cutterhead control parameters are measured by sensors installed on the TBM cutterhead drive device, drive motor and other equipment; the TBM excavation specific energy and surrounding rock grade data are obtained from the previous address exploration data; the construction speed and net excavation speed are obtained from the excavation daily report generated during the TBM excavation process. All the above features and label data are input into the computer for integration and the excavation parameter database is established; the database is divided into training set and test set according to the ratio of 0.9:0.1; the training set is substituted into the CNN-LSTM-Adaboost model for training, and then the model obtained is tested with the test set. This process is repeated to iteratively reduce the loss function until the optimal model is obtained; the data to be predicted is input into the model to obtain the TBM-related rock machine parameters and realize the prediction.

[0148] As a preferred embodiment, the cutterhead control parameters include cutterhead thrust, cutterhead torque and cutterhead speed; the geological exploration data include TBM excavation specific energy and surrounding rock grade; and the excavation daily report data include construction speed and net excavation speed.

[0149] For ease of description, spatially relative terms such as "above", "above", "on the upper surface of", "above", etc. may be used here to describe the spatial positional relationship between a device or feature and other devices or features as shown in the figure. It should be understood that spatially relative terms are intended to include different orientations of the device in use or operation in addition to the orientation described in the figure. For example, if the device in the accompanying drawings is inverted, the device described as "above other devices or structures" or "above other devices or structures" will be positioned as "below other devices or structures" or "below other devices or structures". Thus, the exemplary term "above" can include both "above" and "below". The device can also be positioned in other different ways (rotated 90 degrees or in other orientations), and the spatially relative descriptions used here are interpreted accordingly.

[0150] In addition, it should be noted that the use of terms such as "first" and "second" to limit components is only for the convenience of distinguishing the corresponding components. If not otherwise stated, the above terms have no special meaning and therefore cannot be understood as limiting the scope of protection of the present invention.

[0151] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A deep buried TBM rock machine parameter fusion prediction method based on CNN-LSTM-Adaboost model, characterized in that: Includes steps: S1: Establish the initial CNN-LSTM-Adaboost surrounding rock grade prediction model; S2: Establish the relative error rate, root mean square error, mean absolute percentage error and goodness of fit evaluation indicators and effect indicators of the model output excavation parameter prediction; S3: Based on the TBM excavation data and tunnel surrounding rock geological data, and combined with the evaluation indicators and effect indicators in step S2, the CNN-LSTM-Adaboost model is optimized; S4: Use the optimized model to predict the excavation parameters under different surrounding rock grades; S5: Output the obtained prediction results to predict the tunnel surrounding rock grade.

2. The deep buried TBM rock-machine parameter fusion prediction method based on the CNN-LSTM-Adaboost model according to claim 1 is characterized in that: In step S1, the process of establishing the CNN-LSTM-Adaboost surrounding rock grade identification model includes: Create a training set: T={(x1,y1),(x2,y2),…(x N ,y N )} Set the weight distribution D1 of the initialization data to uniform distribution: Calculate the weak classifier G m (x) Classification error ε on the dataset m : Calculate the weak classifier G m (x) Weight α in the strong classifier m : Update the weight distribution of the training set: Determine the number of iterations N, and if it reaches N, combine the weak classifiers: The strong classifier G(x) is obtained through the relationship function sign:

3. The deep buried TBM rock-machine parameter fusion prediction method based on the CNN-LSTM-Adaboost model according to claim 2 is characterized in that: The operation process of the CNN-LSTM-Adaboost model includes the following steps: S11: Import TBM excavation data and data of tunnels with different grades of surrounding rock; S12: Establishing data set; S13: Perform data initialization processing; S14: 90% of the TBM excavation data obtained under different surrounding rock grades are selected as the model training data set; 10% of the TBM excavation data obtained under different surrounding rock grades are selected as the model test set; S15: According to the analysis of whether the data in the training data set conforms to CART, the conforming data is classified into a strong classifier, and the non-conforming data is classified into a weak classifier; S16: After testing the data in the test set, the data is classified into a strong classifier; S17: Obtain classification results.

4. The deep buried TBM rock-machine parameter fusion prediction method based on the CNN-LSTM-Adaboost model according to claim 1 is characterized in that: In step S2, the calculation formula of the evaluation index is as follows: Relative error rate calculation: Goodness of fit calculation: Root mean square error calculation: Root mean square error calculation:

5. The deep buried TBM rock-machine parameter fusion prediction method based on the CNN-LSTM-Adaboost model according to claim 4 is characterized in that: In step S2, the calculation formula of the effect index ACC is as follows:

6. The deep buried TBM rock-machine parameter fusion prediction method based on the CNN-LSTM-Adaboost model according to claim 5 is characterized in that: In step S3, the optimization steps of the CNN-LSTM-Adaboost model are as follows: 90% of the TBM excavation data obtained under different surrounding rock grades were selected as the model training data set; 10% of the TBM excavation data obtained under different surrounding rock grades were selected as the model test set; The initial learning rate of the LSTM structural unit is set to 0.005, and the CNN structure is a single-layer convolution. After 20 repeated runs, the damage function value drops to 0.

1. Roulette wheel selection is used as the validation technique, and the validation frequency is once every 40 iterations, where the gating unit adopts ten-fold cross validation; Set up 200 Adaboost single-base classifiers, with a minimum fork number of 4 and a maximum tree height of 20; The excavation parameters under the four levels of surrounding rock, cutterhead thrust F, cutterhead torque T, cutterhead speed N, construction speed v AR , Net advance speed v PR and TBM excavation specific energy E SE Input the model training model, output one of the tunneling parameters in turn, and train it multiple times to improve the model prediction accuracy.

7. A deep buried TBM rock-machine parameter fusion prediction system based on CNN-LSTM-Adaboost model, comprising the deep buried TBM rock-machine parameter fusion prediction method based on CNN-LSTM-Adaboost model according to any one of claims 1 to 6, characterized in that: The cutterhead control parameters are measured by equipment installed on the TBM cutterhead drive unit and drive motor, and the imported geological exploration data and daily excavation report data are all entered into the computer to establish a parameter database.

8. The deep buried TBM rock machine parameter fusion prediction system based on CNN-LSTM-Adaboost model according to claim 7 is characterized in that: The cutterhead control parameters include cutterhead thrust, cutterhead torque and cutterhead speed; the geological exploration data include TBM excavation specific energy and surrounding rock grade; the excavation daily report data include construction speed and net excavation speed.

Images