Intelligent muck treatment method and system combined with tunnel advanced geological prediction

Abstract

The invention discloses a muck intelligent processing method combined with tunnel advanced geological prediction. The muck intelligent processing method comprises the following steps: acquiring a seismic wave training set; carrying out geological analysis on muck categories in the seismic wave training set, obtaining dynamic calculation parameters and physical data corresponding to each muck category, and enabling the seismic wave features to correspond to the dynamic calculation parameters and the physical data to obtain an updated seismic wave training set; training a neural network by using the seismic wave training set to obtain a muck prediction processing model; acquiring real-time geological forecast data, preprocessing the real-time geological forecast data, inputting the preprocessed real-time geological forecast data into the muck prediction processing model, and outputting a muck type through the muck prediction processing model; and in combination with the working state of the shield tunneling machine and the geological forecast, a muck treatment suggestion is generated so as to treat muck generated in the shield tunneling construction process. According to the method, the type of the muck generated during tunnel excavation can be predicted in time, and the prediction timeliness is improved.

Description

Technical Field

[0001] The invention relates to the technical field of shield tunnel excavation, and relates to an intelligent method and system for processing slag combined with advanced tunnel geological prediction. Background Art

[0002] With the gradual deepening of my country's urbanization process, the development and utilization of urban underground space has gradually become an important direction of urban development. Compared with traditional tunnel excavation methods, shield machines integrate multiple operations such as excavation, support, advancement, and stacking, and have significant construction efficiency advantages and good cost-effectiveness. During the construction process, shield machines do not require large-scale demolition on the ground, do not block traffic, are noiseless during construction, do not cause ground subsidence, and do not affect the normal lives of residents. They have significant advantages in environmental protection and construction safety. Therefore, it is considered a common technology in this field and is widely used in major projects and infrastructure such as urban rail transit, underground pipe galleries, and railway and highway tunnels.

[0003] The use of shield technology to excavate tunnels is bound to produce a large amount of shield slag. There is still a large amount of slag in my country that needs to be treated, and this slag has become one of the obstacles to urban tunnel construction. At present, in domestic tunnel projects, the treatment of shield slag is still mainly based on the traditional stacking landfill method, which occupies a large area and affects the appearance of the city. In the construction process of shield tunnels, in order to protect the cutterhead of the shield machine, reduce the cutterhead torque, and increase the plasticity of the stratum flow, surfactants, commonly known as foaming agents, will be added to the soil bin. These foaming agents will be adsorbed in the slag, making its water content high, difficult to transport, and also causing a series of physical and chemical reactions, which increases the difficulty of environmental protection treatment of the slag. If it is discharged directly without proper treatment, it will cause soil, surface water and groundwater pollution, causing great damage to the ecological environment. In addition, the difference in the nature of the excavation surface strata will cause the shield machine to produce a variety of different types of slag, which have different particle grading, penetration index, water content, flow state and other parameters, so the required treatment solutions are also different.

[0004] Intelligent tunnel construction is one of the current research hotspots in the field of railway tunnel construction. The shield method is a factory-based operation, and the construction process is organized according to the assembly line, with a high degree of mechanization, which has the foundation and advantages of realizing intelligence. However, the current intelligent technology is mainly focused on the centralized and unified management of visualization, digitization, and intelligence in tunnel survey, design, and construction. There are few intelligent technologies for the entire process of tunnel construction that combine survey and construction waste treatment.

[0005] In modern tunnel engineering, muck identification technology is of great significance for optimizing the construction process, improving operation efficiency, and ensuring construction safety. Currently, muck identification mainly relies on devices such as computer vision, cutter head sensors, and rheometers. Among them, computer vision technology can achieve a preliminary judgment of the geological conditions ahead through real-time analysis of the face image; while the cutter head sensor can monitor the changes in various parameters during the operation of the cutter head, indirectly reflecting the geological characteristics of the face. As for the application of the rheometer, it is mainly limited to the determination of the physical properties of the excavated muck, such as viscosity, density, etc. These information helps to understand the state of the muck in the soil bin, but cannot provide direct data on the unexcavated area.

[0006] However, the above methods have a common limitation, that is, they mainly focus on the analysis of the current or completed operation area, lacking the ability to prospectively predict the geological conditions ahead. This lack of timeliness limits the effectiveness of muck identification technology in guiding muck treatment plans. Especially in complex geological conditions, timely and accurate prediction of the geological conditions ahead is crucial for preventing potential risks. In recent years, with the development of artificial intelligence technology, advanced geological prediction based on machine learning has become possible. This technology can, to a certain extent, predict the geological characteristics of the undeveloped area by integrating multiple geological data sources and using deep learning algorithm models for training. However, existing research and practice pay more attention to how to apply AI technology to the excavation process itself to improve excavation efficiency and safety, and the combination of advanced geological prediction and muck treatment is still in the exploratory stage. Summary of the Invention

[0007] Aiming at the above deficiencies in the prior art, a muck intelligent processing method and system combined with advanced geological prediction of tunnels provided by the present invention solves the problem of poor timeliness in predicting the muck generated during tunnel excavation in the prior art.

[0008] To achieve the above invention purpose, the technical solution adopted by the present invention is as follows: In the first aspect, a muck intelligent processing method combined with advanced geological prediction of tunnels includes:

[0009] S1. Obtain a seismic wave training set;

[0010] The seismic wave training set includes seismic wave features and the muck category corresponding to each seismic wave feature;

[0011] S2. Conduct geological analysis on the muck categories in the seismic wave training set, obtain the dynamic calculation parameters and physical data corresponding to each muck category, and correspond the seismic wave features with the dynamic calculation parameters and physical data to obtain an updated seismic wave training set;

[0012] S3. Train the neural network using the seismic wave training set to obtain a muck prediction and processing model;

[0013] S4. Obtain real-time geological forecast data, preprocess the real-time geological forecast data, and input the preprocessed real-time geological forecast data into the muck prediction and processing model, and output the muck type through the muck prediction and processing model;

[0014] S5. Generate muck treatment suggestions based on the muck type and combined with the working state of the shield machine and the geological forecast to treat the muck generated during the shield construction process.

[0015] The beneficial effects of the above solution are:

[0016] (1) By combining the horizontal acoustic / seismic wave profiling method (HSP) with deep learning technology, the present invention realizes high-precision geological forecasting, can accurately predict the geological conditions ahead, and improves the safety and efficiency of construction.

[0017] (2) By intelligently predicting the muck type and selecting appropriate treatment methods, the damage to the environment is reduced, which conforms to the concept of sustainable development.

[0018] Further, after S1, the method further includes:

[0019] Preprocess the seismic wave training set to obtain a preprocessed seismic wave training set;

[0020] S2 specifically includes:

[0021] Conduct geological analysis on the muck categories in the preprocessed seismic wave training set, obtain the dynamic calculation parameters and physical data corresponding to each muck category, and correspond the seismic wave characteristics with the dynamic calculation parameters and physical data to obtain an updated seismic wave training set.

[0022] The beneficial effects of the above further solution are: After removing noise, the data more accurately reflects the real situation. Filling in missing values can ensure the integrity of the data set, enabling the data to be fully utilized in subsequent analysis and model training. Data standardization can map the value ranges of different features to similar intervals, enabling the gradient descent algorithm to converge to the optimal solution faster and improving the convergence speed of the model.

[0023] Further, in S2, the dynamic calculation parameters include longitudinal wave velocity, transverse wave velocity, wave velocity ratio, density, Poisson's ratio, Young's modulus, and bulk modulus;

[0024] The physical data includes porosity, water content, and particle size distribution.

[0025] Further, in S3, the muck prediction processing model includes an input layer, a convolutional layer, a flattening layer, a fully connected layer, and an output layer;

[0026] The data form obtained by the input layer is N*8*3;

[0027] The convolutional layer includes a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer, a third convolutional layer, a third pooling layer, a fourth convolutional layer, and a fourth pooling layer;

[0028] The number of convolutional kernels in the first convolutional layer is 64, the number of convolutional kernels in the second convolutional layer is 128, the number of convolutional kernels in the third convolutional layer is 256, the number of convolutional kernels in the third convolutional layer is 512. The sizes of the convolutional kernels in the first convolutional layer, the second convolutional layer, the third convolutional layer, and the fourth convolutional layer are all 3*3. The activation functions of the first convolutional layer, the second convolutional layer, the third convolutional layer, and the fourth convolutional layer are all ReLU functions;

[0029] The pooling window sizes of the first pooling layer, the second pooling layer, the third pooling layer, and the fourth pooling layer are all 2*2;

[0030] The fully connected layer includes a first fully connected layer and a second fully connected layer;

[0031] The number of nodes in the first fully connected layer is 1024, and the number of nodes in the second fully connected layer is 512;

[0032] The activation functions of the first fully connected layer and the second fully connected layer are both ReLU functions;

[0033] The Dropout rates of the first fully connected layer and the second fully connected layer are both 0.5.

[0034] The number of nodes in the output layer is the same as the number of muck categories.

[0035] The beneficial effects of the above further solution are as follows: Small convolutional kernels can capture local features, reduce the number of parameters at the same time, and improve the computational efficiency of the model. The convolutional kernels move pixel by pixel on the input data to ensure that as many details as possible are captured. Padding ensures that the size of the output feature map is the same as the input, avoiding size reduction caused by convolutional operations. The ReLU activation function can help the model learn more complex patterns. The max pooling layer is used for dimensionality reduction, reducing the size of the feature map while retaining the most important features. The pooling layer can extract higher-level features and continue with dimensionality reduction. A large number of nodes in the fully connected layer can capture more feature combinations and improve the expressive ability of the model. Dropout randomly discards a part of neurons to prevent overfitting and improve the generalization ability of the model.

[0036] Further, in S3, the loss function of the muck prediction processing model is Categorical Crossentropy;

[0037] The optimizer of the muck prediction and processing model is the Adam optimizer, and the initial learning rate of the Adam optimizer is 0.001;

[0038] L2 regularization is applied in the fully connected layer, and the weight decay coefficient is 0.001.

[0039] The beneficial effects of the above further solution are: CategoricalCrossentropy, which is used for multi-classification tasks and measures the difference between the probability distribution predicted by the model and the true labels. The Adam optimizer combines the advantages of momentum and adaptive learning rate, can converge faster, and is suitable for the training of deep learning models

[0040] In the second aspect, an intelligent muck processing system for joint tunnel advanced geological prediction is implemented based on the intelligent muck processing method for joint tunnel advanced geological prediction described in any one of claims 1-5, and includes:

[0041] A model training module, which is used to obtain a seismic wave training set; conduct geological analysis on the muck categories in the seismic wave training set, obtain the dynamic calculation parameters and physical data corresponding to each muck category, and correspond the seismic wave features with the dynamic calculation parameters and physical data to obtain an updated seismic wave training set; use the seismic wave training set to train a neural network to obtain a muck prediction and processing model;

[0042] Among them, the seismic wave training set includes seismic wave features and the muck category corresponding to each seismic wave feature;

[0043] A real-time data acquisition module, which is used to obtain real-time geological forecast data, preprocess the real-time geological forecast data, and input the preprocessed real-time geological forecast data into the muck prediction and processing model to obtain the muck type generated during the shield construction process;

[0044] A muck analysis module, which is used to obtain real-time geological forecast data, input the real-time geological forecast data into the muck prediction and processing model to obtain the muck type generated during the shield construction process.

[0045] A muck processing module, which is used to generate muck processing suggestions according to the muck type and in combination with the working state of the shield machine and the geological forecast, so as to process the muck generated during the shield construction process.

[0046] Further, the real-time data acquisition module includes a seismic wave exciter and a geophone;

[0047] The seismic wave exciter is used to excite active source seismic waves, and the geophone is used to receive full-space vibration signals.

[0048] Further, the intelligent muck processing system further includes:

[0049] A data preprocessing module for preprocessing a seismic wave training set to obtain a preprocessed seismic wave training set.

[0050] The beneficial effect of the above further solution is that by combining the horizontal acoustic / seismic wave profile method with deep learning technology, high-precision geological forecasting is achieved, and the muck type is intelligently predicted according to the forecasting result, and a suitable muck treatment method is selected. Description of the Drawings

[0051] Figure 1 It is a schematic flow chart of a muck intelligent treatment method for combined tunnel advanced geological prediction.

[0052] Figure 2 It is a schematic structural diagram of a muck intelligent treatment system for combined tunnel advanced geological prediction. Detailed Embodiments

[0053] The present invention will be further described below with reference to the drawings and specific embodiments.

[0054] Embodiment 1

[0055] As Figure 1 shown, a muck intelligent treatment method for combined tunnel advanced geological prediction includes:

[0056] S1. Obtain a seismic wave training set; the seismic wave training set includes seismic wave features and the muck category corresponding to each seismic wave feature.

[0057] Exemplarily, the seismic wave training set can be obtained in the following manner: collect historical geological exploration data (seismic wave features) and the analysis results of muck samples during the actual shield construction process (muck categories), and construct a seismic wave training set.

[0058] S2. Conduct geological analysis on the muck categories in the seismic wave training set, obtain the dynamic calculation parameters and physical data corresponding to each muck category, and correspond the seismic wave features with the dynamic calculation parameters and physical data to obtain an updated seismic wave training set.

[0059] In this embodiment, after S1, the method further includes: preprocessing the seismic wave training set to obtain a preprocessed seismic wave training set; S2 specifically includes:

[0060] Conduct geological analysis on the muck categories in the preprocessed seismic wave training set, obtain the dynamic calculation parameters and physical data corresponding to each muck category, and correspond the seismic wave features with the dynamic calculation parameters and physical data to obtain an updated seismic wave training set.

[0061] In this embodiment, in S2, the kinetic calculation parameters include longitudinal wave velocity, transverse wave velocity, wave velocity ratio, density, Poisson's ratio, Young's modulus, and bulk modulus; the physical data includes porosity, water content, and particle size distribution.

[0062] Optionally, preprocessing the seismic wave training set may include removing noise, filling missing values, data standardization, etc., to ensure the quality and consistency of the data.

[0063] For example, to remove noise from the seismic wave training set, a Python programming language can be used to call scipy.signa1 to construct a low-pass filter for denoising. Define a function for creating and applying a low-pass Butterworth filter. This function generates filter coefficients based on the given cut-off frequency, sampling rate, and filter order, and filters the signal.

[0064] To fill missing values in the seismic wave training set, it can be completed through the fillna() function in the Pandas library of the Python programming language. First, define an example variable, then iterate through each column, calculate the average value of the column, and use this average value to fill all null values to complete the filling of missing values.

[0065] To standardize the data of the seismic wave training set, the sklearn.preprocessing.StandardScaler class can be used for standardization. The fit_transform method automatically calculates the mean μ and standard deviation σ of each feature and transforms the data into where X represents the original data points.

[0066] Conduct a detailed geological analysis of the muck samples, record their kinetic calculation parameters (longitudinal wave velocity, transverse wave velocity, wave velocity ratio, density, Poisson's ratio, Young's modulus, bulk modulus) and physical data (porosity, water content, particle size distribution), and at the same time accurately label the data, and match the kinetic calculation parameters and physical data of different types of muck samples with the corresponding seismic wave characteristics. Different types of muck samples' kinetic calculation parameters, physical data, and corresponding seismic wave characteristics can be matched by creating a dictionary in Python.

[0067] S3. Use the seismic wave training set to train the neural network to obtain a muck prediction processing model.

[0068] In this embodiment, in S3, the muck prediction processing model includes an input layer, a convolutional layer, a flattening layer, a fully connected layer, and an output layer;

[0069] The data form obtained by the input layer is N*8*3;

[0070] The convolutional layer includes a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer, a third convolutional layer, a third pooling layer, a fourth convolutional layer, and a fourth pooling layer;

[0071] The number of convolutional kernels in the first convolutional layer is 64, the number of convolutional kernels in the second convolutional layer is 128, the number of convolutional kernels in the third convolutional layer is 256, the number of convolutional kernels in the fourth convolutional layer is 512. The size of the convolutional kernels in the first convolutional layer, the second convolutional layer, the third convolutional layer, and the fourth convolutional layer is all 3*3. The activation function of the first convolutional layer, the second convolutional layer, the third convolutional layer, and the fourth convolutional layer is the ReLU function;

[0072] The pooling window size of the first pooling layer, the second pooling layer, the third pooling layer, and the fourth pooling layer is all 2*2;

[0073] The fully connected layer includes a first fully connected layer and a second fully connected layer;

[0074] The number of nodes in the first fully connected layer is 1024, and the number of nodes in the second fully connected layer is 512;

[0075] The activation function of the first fully connected layer and the second fully connected layer is the ReLU function;

[0076] The Dropout rate of the first fully connected layer and the second fully connected layer is 0.5.

[0077] The number of nodes in the output layer is the same as the number of categories of muck.

[0078] In this embodiment, the loss function of the muck prediction processing model is Categorical Crossentropy;

[0079] The optimizer of the muck prediction processing model is the Adam optimizer, and the initial learning rate of the Adam optimizer is 0.001;

[0080] L2 regularization is applied in the fully connected layer, and the weight decay coefficient is 0.001.

[0081] Optionally, after obtaining the updated seismic wave training set, the updated seismic wave training set can be divided into a training set, a test set, and a validation set. During the subsequent model training process, the training set is used to train the model, the test set is used to test the model, and the validation set is used to evaluate the model.

[0082] Exemplarily, the framework and model design of the neural network can be in the Python language, and a Convolutional Neural Network (CNN) model is designed based on the TensorFlow framework. The data form obtained by the input layer is N*8*3, where N is the time series length, 8 represents 8 different seismic wave data, and 3 represents three components of each seismic wave data. The flattening layer can convert the multi-dimensional feature map generated by the convolutional layer into a one-dimensional vector for input to the fully connected layer.

[0083] Optionally, the strides of the first convolutional layer, the second convolutional layer, the third convolutional layer, and the fourth convolutional layer are all 1, and the padding method is'same' to keep the output size unchanged. When training the neural network using the seismic wave training set, the batch size is set to 32, and the number of iterations is set to 50 times. At the same time, an early stopping strategy is adopted, that is, when the performance on the validation set no longer improves, the training is terminated early to avoid overfitting.

[0084] For example, the input size of the input layer can be a CNN architecture including a spatio-temporal convolutional layer, which can accept an input with a shape of [number of samples, time length, spatial height, spatial width, number of channels]. The spatial height and width can respectively correspond to the spatial layout of 8 positions of the seismic wave detection device, and the number of channels can be 3.

[0085] The parameter settings of the first convolutional layer, the first pooling layer, the second convolutional layer, the second pooling layer, the third convolutional layer, the third pooling layer, the fourth convolutional layer, and the fourth pooling layer can be as follows respectively:

[0086] In the first convolutional layer, the kernel size: 3x3. A small kernel can capture local features, reduce the number of parameters at the same time, and improve the computational efficiency of the model. The number of kernels: 64, the stride: 1. A stride of 1 means that the kernel moves pixel by pixel on the input data, ensuring that as many details as possible are captured. Padding:'same'. Purpose:'same' padding ensures that the size of the output feature map is the same as the input, avoiding size reduction caused by the convolution operation. Activation function: ReLU, which helps the model learn more complex patterns. The activation function changes all negative values to 0 and retains positive values.

[0087] In the first pooling layer, the kernel size: 2x2, the stride 2. The first pooling layer is used for dimensionality reduction, reducing the size of the feature map while retaining the most important features. A stride of 2 means moving two pixels each time, further reducing the amount of data.

[0088] In the second convolutional layer, the kernel size: 3x3. The number of kernels: 128. The stride: 1. Padding:'same'. Activation function: ReLU.

[0089] In the second pooling layer, the kernel size: 2x2, the stride 2. Further extract higher-level features while continuing to reduce the dimension.

[0090] In the third convolutional layer, the kernel size is 3x3, the number of kernels is 256, the stride is 1, the padding is'same', and the activation function is ReLU.

[0091] In the third pooling layer, the kernel size is 2x2 and the stride is 2. It continues to extract more complex features and further reduces the dimensionality.

[0092] In the fourth convolutional layer, the kernel size is 3x3, the number of kernels is 512, the stride is 1, the padding is'same', and the activation function is ReLU.

[0093] In the fourth pooling layer, the kernel size is 2x2 and the stride is 2. The purpose is to further extract the deepest features and continue to reduce the dimensionality.

[0094] The parameter settings of the first fully connected layer and the second fully connected layer can be respectively:

[0095] In the first fully connected layer, the number of nodes is 1024. A large number of nodes can capture more feature combinations and improve the expression ability of the model. The activation function is ReLU. Dropout is 0.5. Dropout randomly discards a part of neurons to prevent overfitting and improve the generalization ability of the model.

[0096] In the second fully connected layer, the number of nodes is 512. The activation function is ReLU. Dropout is 0.5. It further processes the features, reduces the complexity of the model, and at the same time maintains a high expression ability.

[0097] The parameter settings of the output layer can be: the number of nodes is C (C is the number of muck types), and the activation function is Softmax.

[0098] Optionally, when the accuracy on the validation set no longer improves, stop training early. After each model training is completed, the changing trends of the training loss, training accuracy, validation loss, and validation accuracy can be analyzed to evaluate the fitting status of the model. If overfitting is found, it should return to the model training step, reset parameters such as the model iteration times, learning rate and its decay strategy, batch size, etc. based on the differences between the training and validation performances, and consider adjusting activation functions and other measures to reduce overfitting, and then train the model again until a satisfactory performance level is achieved.

[0099] Then, check whether the accuracies on the training set and the validation set have reached the expected standards. If not, it is necessary to return to the model training step to continue optimization; otherwise, once the required accuracy requirements are met, save the current best weight configuration.

[0100] Subsequently, it enters the model performance testing stage, where data independent of the training process (test set) is used to evaluate the performance of the model. By inputting the samples in the test set into the trained model, prediction results are obtained and compared with the actual values to measure the degree of consistency between the model output and the real situation. This step aims to test the generalization ability of the model for unknown data. If the accuracy obtained in the testing stage is lower than the preset target value, it is also necessary to return to the model training step for further adjustment and retraining.

[0101] S4. Obtain real-time geological prediction data, preprocess the real-time geological prediction data, and input the preprocessed real-time geological prediction data into the muck prediction and processing model, and output the muck type through the muck prediction and processing model.

[0102] Exemplarily, the real-time geological prediction data can be obtained by deploying seismic wave exciters and geophones inside the tunnel. The seismic wave exciters and geophones are deployed inside the tunnel to ensure the effective excitation of the active source seismic wave and the reception of the full-space vibration signal.

[0103] Optionally, after collecting the real-time geological prediction data, the real-time geological prediction data can be preprocessed. The data preprocessing can include noise removal, missing value filling, and data standardization.

[0104] Optionally, during the shield construction process, the prediction results and the actual construction conditions can be recorded, and the muck prediction and processing model can be continuously optimized to improve the accuracy and reliability of the prediction. The optimization method can be to fine-tune the muck prediction and processing model.

[0105] For example, 1 excitation point and 8 three-component geophones are deployed inside the tunnel. The excitation point is set on the segment to ensure the effective excitation of the active source seismic wave. The geophone spacing is 2m, the minimum offset is 2.5m, and the maximum offset is 16.5m.

[0106] S5. Generate muck treatment suggestions based on the muck type, combined with the working state of the shield machine and the geological prediction, so as to treat the muck generated during the shield construction process.

[0107] Exemplarily, select appropriate treatment methods according to the characteristics of different mucks and environmental protection requirements.

[0108] Optionally, corresponding treatment equipment can be equipped according to the actual requirements of different treatment methods, including curing agent spraying devices, screening machines, washing machines, etc.

[0109] Optionally, the muck treatment suggestions may include: For multi-soil muck, this type of muck contains more fine particles (clay, sand, etc.), has a higher water content, usually exceeds the liquid limit, and is close to the saturated state. This kind of muck generally comes from the muck generated during the excavation of fully weathered rock formations or soil layers. Treatment methods: Due to its high water content and good fluidity, natural air-drying method, adding material solidification method (such as cement, lime), filtration separation method (by setting up muck holding pits), and mechanical dehydration method (such as using filter presses) can be used to reduce the water content. The machines used include: screening system (for separating coarse particles), slurry stirring device (for adjusting fluidity), filter presses (plate and frame filter press, belt thickening and dewatering machine, centrifugal dewatering machine or spiral screw dewatering machine). For intermediate muck: The particle size distribution of this type of muck is between that of multi-soil muck and multi-rock muck, with the stone content between 40% and 70%, showing the characteristics of a mixture and forming a dense-skeleton structure. This kind of muck usually comes from moderately weathered rock formations. Treatment methods: Similarly, natural air-drying method, adding material solidification method (such as cement, lime), filtration separation method, and mechanical dehydration method can be used for treatment. The machines used are similar to those for multi-soil muck, but more attention may be needed to handle the coarse aggregate to ensure the quality of the final product. For multi-rock muck: This type of muck contains more coarse aggregate (crushed stones), has a lower water content, and stronger water permeability. This kind of muck usually comes from the muck generated during the excavation of weakly weathered rock formations. Treatment methods: Mainly separate the coarse aggregate from the fine material through a screening system, and then use natural air-drying method, adding material solidification method (such as cement, lime), or mechanical dehydration method to treat the remaining fine material. The machines used include: screening system (for separating coarse particles), slurry stirring device (for adjusting fluidity), filter presses (plate and frame filter press, belt thickening and dewatering machine, centrifugal dewatering machine or spiral screw dewatering machine).

[0110] In this embodiment, the muck prediction and treatment model of the present invention can be loaded into the computer of the shield machine control system, a stable data interface can be developed, so that the muck prediction and treatment model can obtain real-time sensor data from the shield machine central control system, and a feedback mechanism can be designed to enable the processed analysis results to be timely fed back to the control system to guide muck treatment. Make adaptive adjustments to the trained muck prediction and treatment model to make it more suitable for the data characteristics under actual working conditions, and thoroughly test the entire process in a simulation environment to ensure that the model accuracy meets the engineering requirements. Finally, establish a regular inspection mechanism, track the system operation status, collect feedback information, continuously improve the algorithm accuracy, and continuously update and improve the model according to the project progress and actual application effect, so as to achieve more efficient and accurate muck prediction.

[0111] Embodiment 2

[0112] As Figure 2As shown in the figure, an intelligent muck processing system for combined tunnel advanced geological prediction is realized based on an intelligent muck processing method for combined tunnel advanced geological prediction, including:

[0113] A model training module, which is used to obtain a seismic wave training set; conduct geological analysis on the muck categories in the seismic wave training set, obtain the dynamic calculation parameters and physical data corresponding to each muck category, and correspond the seismic wave features with the dynamic calculation parameters and physical data to obtain an updated seismic wave training set; use the seismic wave training set to train a neural network to obtain a muck prediction and processing model;

[0114] Among them, the seismic wave training set includes seismic wave features and the muck categories corresponding to each seismic wave feature;

[0115] A real-time data acquisition module, which is used to obtain real-time geological forecast data, preprocess the real-time geological forecast data, and input the preprocessed real-time geological forecast data into the muck prediction and processing model to obtain the muck types generated during the shield tunneling construction process;

[0116] A muck analysis module, which is used to obtain real-time geological forecast data and input the real-time geological forecast data into the muck prediction and processing model to obtain the muck types generated during the shield tunneling construction process.

[0117] A muck processing module, which is used to generate muck processing suggestions according to the muck types and in combination with the working state of the shield machine and the geological forecast, so as to process the muck generated during the shield tunneling construction process.

[0118] Furthermore, the real-time data acquisition module includes a seismic wave exciter and a geophone;

[0119] The seismic wave exciter is used to excite active-source seismic waves, and the geophone is used to receive full-space vibration signals.

[0120] Furthermore, the intelligent muck processing system further includes:

[0121] A data preprocessing module, which is used to preprocess the seismic wave training set to obtain a preprocessed seismic wave training set.

[0122] Those of ordinary skill in the art will realize that the embodiments described herein are for helping readers understand the principles of the present invention, and it should be understood that the protection scope of the present invention is not limited to such specific statements and embodiments. Those of ordinary skill in the art can make various other specific deformations and combinations that do not depart from the essence of the present invention according to these technical revelations disclosed by the present invention, and these deformations and combinations are still within the protection scope of the invention.

Claims

1. An intelligent treatment method for muck combined with advanced geological prediction of tunnels, characterized in that, The method includes: S1. Obtain a seismic wave training set; The seismic wave training set includes seismic wave features and the corresponding muck categories for each of the seismic wave features; S2. Conduct geological analysis on the muck categories in the seismic wave training set, obtain the corresponding dynamic calculation parameters and physical data for each of the muck categories, and correspond the seismic wave features with the dynamic calculation parameters and the physical data to obtain an updated seismic wave training set; S3. Use the seismic wave training set to train a neural network to obtain a muck prediction processing model; S4. Obtain real-time geological forecast data, preprocess the real-time geological forecast data, and input the preprocessed real-time geological forecast data into the muck prediction processing model, and output the muck type through the muck prediction processing model; S5. Generate muck treatment suggestions according to the muck type, in combination with the working state of the shield machine and the geological forecast, so as to treat the muck generated during the shield construction process.

2. The method according to claim 1, wherein After S1, the method further includes: Preprocess the seismic wave training set to obtain a preprocessed seismic wave training set; S2 specifically includes: Conduct geological analysis on the muck categories in the preprocessed seismic wave training set, obtain the corresponding dynamic calculation parameters and physical data for each of the muck categories, and correspond the seismic wave features with the dynamic calculation parameters and the physical data to obtain an updated seismic wave training set.

3. The method according to claim 2, wherein In S2, the dynamic calculation parameters include longitudinal wave velocity, transverse wave velocity, wave velocity ratio, density, Poisson's ratio, Young's modulus, and bulk modulus; The physical data includes porosity, water content, and particle size distribution.

4. The method according to claim 3, wherein In S3, the muck prediction processing model includes an input layer, a convolutional layer, a flattening layer, a fully connected layer, and an output layer; The data form obtained by the input layer is N*8*3; The convolutional layer includes a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer, a third convolutional layer, a third pooling layer, a fourth convolutional layer, and a fourth pooling layer; The number of convolutional kernels in the first convolutional layer is 64, the number of convolutional kernels in the second convolutional layer is 128, the number of convolutional kernels in the third convolutional layer is 256, the number of convolutional kernels in the third convolutional layer is 512, the sizes of the convolutional kernels in the first convolutional layer, the second convolutional layer, the third convolutional layer, and the fourth convolutional layer are all 3*3, and the activation functions of the first convolutional layer, the second convolutional layer, the third convolutional layer, and the fourth convolutional layer are all ReLU functions; The pooling window sizes of the first pooling layer, the second pooling layer, the third pooling layer, and the fourth pooling layer are all 2*2; The fully connected layer includes a first fully connected layer and a second fully connected layer; The number of nodes in the first fully connected layer is 1024, and the number of nodes in the second fully connected layer is 512; The activation functions of the first fully connected layer and the second fully connected layer are both ReLU functions; The Dropout rates of the first fully connected layer and the second fully connected layer are both 0.

5. The number of nodes in the output layer is the same as the number of muck categories.

5. The method according to claim 4, characterized in that, In S3, the loss function of the muck prediction and processing model is Categorical Crossentropy; The optimizer of the muck prediction and processing model is the Adam optimizer, and the initial learning rate of the Adam optimizer is 0.001; L2 regularization is applied in the fully connected layer, and the weight decay coefficient is 0.

001.

6. An intelligent muck processing system for advanced geological prediction of a combined tunnel, which is implemented based on the intelligent muck processing method for advanced geological prediction of a combined tunnel according to any one of claims 1-5, characterized in that, It includes: A model training module for obtaining a seismic wave training set; Performing geological analysis on the muck categories in the seismic wave training set, obtaining the dynamic calculation parameters and physical data corresponding to each muck category, and corresponding the seismic wave features with the dynamic calculation parameters and the physical data to obtain an updated seismic wave training set; using the seismic wave training set to train a neural network to obtain a muck prediction and processing model; Among them, the seismic wave training set includes seismic wave features and the muck category corresponding to each seismic wave feature; A real-time data acquisition module for acquiring real-time geological forecast data, preprocessing the real-time geological forecast data, and inputting the preprocessed real-time geological forecast data into the muck prediction and processing model to obtain the muck type generated during the shield construction process; A muck analysis module for acquiring real-time geological forecast data and inputting the real-time geological forecast data into the muck prediction and processing model to obtain the muck type generated during the shield construction process. A muck processing module for generating muck processing suggestions according to the muck type and combining the working state of the shield machine and the geological forecast to process the muck generated during the shield construction process.

7. The intelligent muck processing system according to claim 6, wherein The real-time data acquisition module includes a seismic wave exciter and a geophone; The seismic wave exciter is used to excite active source seismic waves, and the geophone is used to receive full-space vibration signals.

8. The intelligent muck processing system according to claim 7, wherein The muck intelligent processing system further includes: A data preprocessing module for preprocessing the seismic wave training set to obtain a preprocessed seismic wave training set.

Images