Unfavorable geological type identification method and system based on joint deep learning network

Abstract

The invention provides an unfavorable geological type identification method and system based on a joint deep learning network, and relates to the technical field of geological disaster forecasting. Comprising the following steps: acquiring training information including geological data of an excavation section, geophysical prospecting inversion imaging information and drilling information; inputting the training information into a pre-established joint deep learning network model, performing data preprocessing, feature extraction and feature fusion in sequence to obtain joint feature representation, and based on the joint feature representation, obtaining a discrimination type and a morphological description of the unfavorable geologic body; wherein the joint deep learning network model comprises a DNN model, a CNN model and an RNN model, and the DNN model, the CNN model and the RNN model are respectively utilized to extract the features of the digital data, the image data and the text description data in the training information. The method can quickly and accurately carry out type discrimination and form description on the unfavorable geologic body in front of the tunnel, and reduces the manual forecasting pressure.

Description

Technical Field

[0001] The present invention belongs to the technical field of geological disaster prediction, and in particular to a method and system for identifying adverse geological types based on a joint deep learning network. Background Art

[0002] In the field of tunnel engineering, the existence of unfavorable geological bodies poses many challenges and hidden dangers to the construction process. If it is not possible to accurately predict unfavorable geological bodies and take targeted measures, it will cause unpredictable harm to the safety of tunnel construction personnel, greatly prolong the construction period, and lead to more cost investment. For example, tunnel engineering carried out in broken strata often suffers from collapse due to the relatively broken rock mass; and if the broken zone is rich in water, there will be hidden dangers of sudden water and mud during excavation. Therefore, it is very important and necessary to predict the unfavorable geology ahead of the tunnel, define the types of unfavorable geology, characterize the unfavorable geological morphology, and take targeted measures in tunnel construction.

[0003] Due to the development of detection methods, the current location prediction technology for unfavorable geological bodies has become very mature. However, the technology for determining the type and characteristics of unfavorable geological bodies and guiding excavation and risk management based on the type and characteristics is still immature and still highly dependent on the judgment of the expert team.

[0004] The expert team has a high degree of credibility in distinguishing the types and characterizing the features of bad geological bodies, but the overall distinction still relies on manual work, and the results of type judgment and characterization vary from person to person, and there is no unified standard basis for distinction. In addition, when distinguishing the types and characterizing the features of bad geological bodies, the expert team cannot integrate multiple information due to the limitations of human brain computing power, and can only rely on memory and experience to link several types of information for identification.

[0005] In addition, there are many tunnel projects on a large scale, but there are few expert teams capable of identifying the types and characterizing the characteristics of unfavorable geological bodies. Without an expert team, construction workers are unable to accurately identify the types and characterize the unfavorable geological bodies in front of the tunnel. Summary of the invention

[0006] In order to overcome the shortcomings of the above-mentioned prior art, the present invention provides a method and system for identifying the type of adverse geological bodies based on a joint deep learning network, which can quickly and accurately identify the type and morphology of the adverse geological bodies in front of the tunnel, make geological forecasting intelligent, and greatly reduce the pressure of manual forecasting.

[0007] To achieve the above objectives, one or more embodiments of the present invention provide the following technical solutions:

[0008] The first aspect of the present invention provides a method for identifying adverse geological types based on a joint deep learning network.

[0009] The method for identifying adverse geological types based on a joint deep learning network includes the following steps:

[0010] Obtain training information, including geological data of the excavation section, geophysical inversion imaging information and drilling information;

[0011] The training information is input into the pre-established joint deep learning network model, and data preprocessing, feature extraction and feature fusion are performed in sequence to obtain a joint feature representation, and based on the joint feature representation, the discrimination type and morphological characterization of the adverse geological body are obtained;

[0012] Among them, the joint deep learning network model includes a DNN model, a CNN model and a RNN model arranged in parallel. When performing feature extraction, the DNN model, the CNN model and the RNN model are used to extract the features of digital data, image data and text description data in the training information respectively.

[0013] The second aspect of the present invention provides a system for identifying adverse geological types based on a joint deep learning network.

[0014] Adverse geological type identification system based on joint deep learning network, including:

[0015] The data acquisition module is configured to: acquire training information, including geological data of the excavation section, geophysical inversion imaging information and drilling information;

[0016] The model prediction module is configured to: input the training information into a pre-established joint deep learning network model, perform data preprocessing, feature extraction and feature fusion in sequence, obtain a joint feature representation, and obtain the discrimination type and morphological characterization of the adverse geological body based on the joint feature representation;

[0017] Among them, the joint deep learning network model includes a DNN model, a CNN model and a RNN model arranged in parallel. When performing feature extraction, the DNN model, the CNN model and the RNN model are used to extract the features of digital data, image data and text description data in the training information respectively.

[0018] One or more of the above technical solutions have the following beneficial effects:

[0019] 1. The present invention proposes a method and system for identifying the type of adverse geological conditions based on a joint deep learning network. The joint deep learning network is used to process multi-source information to more quickly and accurately identify the type and morphologically characterize the adverse geological body in front of the tunnel. The joint deep learning network is used for adverse geological forecasting, which greatly speeds up the speed and accuracy of processing complex multi-source information and reduces manual workload.

[0020] 2. The present invention adopts a joint deep learning network, combines multiple deep learning networks, selects a specific deep learning network for a specific data type, and uses the DNN model, CNN model and RNN model to extract the features of digital data, image data and text description data in the training information during feature extraction, thereby greatly improving the accuracy of feature extraction of multi-source information.

[0021] 3. The present invention adopts a joint deep learning network to characterize the morphological characteristics of unfavorable geological bodies, uses a natural language generation model to provide textual descriptions, and uses a generative adversarial network to draw conceptual maps. This can provide a more intuitive and clearer understanding of the unfavorable geological conditions ahead of the tunnel and enable more accurate disposal measures to be taken.

[0022] 4. When training the model, the loss functions used for training the poor geological body type discrimination and morphological characterization are different. The loss function for training the poor geological body type discrimination includes the cross entropy function, to which physical constraints can be added, and the loss function for training the poor geological body morphological characterization includes the mean square error function. The loss function for the poor geological body type discrimination and the loss function for the poor geological body morphological characterization are weighted to form a joint loss function of the joint deep learning network model, thereby realizing the overall training of the poor geological body type discrimination and morphological characterization.

[0023] Advantages of additional aspects of the present invention will be given in part in the following description, and in part will become obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding 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 improper limitations on the present invention.

[0025] Figure 1 This is a flow chart of the method of the first embodiment. DETAILED DESCRIPTION

[0026] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0027] It should be noted that the terms used herein are for describing specific embodiments only and are not intended to be limiting of exemplary embodiments according to the present invention.

[0028] In the absence of conflict, the embodiments of the present invention and the features of the embodiments may be combined with each other.

[0029] Embodiment 1

[0030] As mentioned in the above background technology, when judging the types and characteristics of bad geological bodies and guiding excavation and risk management according to the type characteristics, there is a high reliance on the judgment of the expert team. However, the expert team can only rely on memory and experience to link several kinds of information for judgment, and cannot integrate multiple kinds of information well, and the subjectivity is strong. In view of this, this embodiment discloses a method for judging the type of bad geological bodies based on a joint deep learning network, which uses a joint deep learning network to process multi-source information of the forecast segment to complete a fast and accurate forecast of bad geological bodies.

[0031] like Figure 1 The diagram is a framework diagram of a joint deep learning network. The method for identifying unfavorable geological types based on a joint deep learning network proposed in this embodiment may specifically include the following steps:

[0032] (1) Establish a database of adverse geological bodies.

[0033] (2) The geological data, geophysical inversion imaging information and drilling information of the excavated sections are screened and summarized and organized into training information.

[0034] (3) Establish a joint deep learning network input layer to preprocess the training information.

[0035] (4) Establish a joint deep learning network extraction layer and use different deep learning network models to extract features of different types of information in geological data, geophysical inversion imaging information, and drilling information.

[0036] Among them, the joint deep learning network model includes a DNN model, a CNN model and a RNN model arranged in parallel. When performing feature extraction, the DNN model, the CNN model and the RNN model are used to extract the features of digital data, image data and text description data in the training information respectively.

[0037] (5) A joint deep learning network fusion layer is established, and the attention mechanism is used to fuse the features of the extracted geological information, geophysical inversion imaging information, and drilling information to obtain a joint feature representation.

[0038] (6) Establish a joint deep learning network output layer to identify the type and characterize the morphology of adverse geological bodies based on the obtained joint feature representation.

[0039] (7) Train the above model in a way that minimizes the loss function.

[0040] (8) After training is completed, the geological information, geophysical inversion information, and drilling information of the forecast period are input into the joint deep learning network to obtain the prediction results;

[0041] Compare the prediction results output by the joint deep learning network with the results of the adverse geological body database to complete the identification of the adverse geological type and morphological characterization of the forecast section.

[0042] The prediction results of the joint deep learning network are compared and verified with the results of the adverse geological body database to prove the reliability of the prediction results of the joint deep learning network.

[0043] Furthermore, in step (1), a database is established based on the existing information of adverse geological bodies, and the database information covers the classification of adverse geological types and the characteristic description of adverse geological bodies.

[0044] Types of adverse geology include but are not limited to: fault fracture zones, karst, soft rock formations, water-rich faults and aquifers, coal seams, frozen soil, etc.

[0045] The characteristic description of adverse geology includes, but is not limited to, the description of adverse geology types, development conditions (scope of impact, scale) and possible geological disasters.

[0046] In the step (2), the geological data may specifically include but are not limited to descriptions of the main lithological compositions, descriptions of groundwater distribution, descriptions of the distribution of major faults, geological profiles and other related geological maps; the geophysical inversion imaging information may specifically include but are not limited to seismic inversion imaging information, resistivity inversion imaging information, induced polarization inversion imaging information, transient electromagnetic inversion imaging information, and geological radar imaging information; the drilling information may specifically include but are not limited to core images, core columnar diagrams, drilling information, etc.

[0047] The geological data, geophysical inversion imaging information and drilling information of the excavated sections are screened. The specific screening methods can be:

[0048] (1) For geological data, firstly, the representative lithological parameters such as rock type, strength, weathering degree, etc. can be screened out by analyzing the lithological data of the excavated sections. Data with special geological structures (such as faults and folds) are screened out. Secondly, for structural information related to adverse geology, such as fault fracture zones, landslide bodies, and caves, detailed descriptions of these features are screened out and distinguished from normal geological sections. Thirdly, by comparing geological information at different time periods and locations, parts that are significantly different from normal strata (such as porosity and rock layer inclination) are screened out as key research objects. In addition, by statistically analyzing a large amount of geological data, factors that have a greater impact on tunnel construction safety and stability are screened out, and duplicate or irrelevant data are eliminated.

[0049] (2) For geophysical inversion imaging information, first use inversion imaging technology (such as resistivity imaging, seismic wave reflection, etc.) to identify abnormal bodies in the strata, screen out inversion results related to unfavorable geology such as faults, fracture zones, and caves, and eliminate noise and abnormal interference. Secondly, for the results of various geophysical methods (such as electrical methods, seismic exploration, and radar), high-quality signals in specific frequency bands can be screened out through spectrum analysis, frequency band filtering, etc., and multi-source data fusion can be performed to eliminate irrelevant or duplicate information. Thirdly, based on the differences in the physical properties of different geological bodies in the imaging results (such as resistivity, reflection coefficient, etc.), areas that meet the characteristics of unfavorable geological types can be screened out, and areas consistent with normal strata can be filtered out.

[0050] (3) Regarding drilling information: First, by analyzing the core samples of the drilling section, select samples related to poor geology (such as fracture development, fragmentation, and water-rich layers), and record key indicators in the core, such as rock integrity and water content. Secondly, combined with the various geological parameters obtained during the drilling process (such as drilling depth, rock formation pressure, rock density, etc.), select parameters with a high correlation with geological disaster risks, and eliminate data information that has no obvious correlation with poor geological bodies. Thirdly, the real-time parameters during drilling (such as drill bit speed, drilling pressure, footage speed, drilling power consumption, etc.) can be monitored to select abnormally fluctuating data, which may indicate that poor geological bodies such as fracture zones and fault mud have been encountered.

[0051] In addition, you can use automated screening methods: Here are two screening ideas:

[0052] 1. Feature selection based on machine learning: Machine learning algorithms such as random forest and Lasso regression can be used to automatically select the most representative and discriminative features. These algorithms can automatically select the optimal feature combination based on the importance of the input data.

[0053] 2. Anomaly detection algorithm: The input data is analyzed through anomaly detection algorithms (such as isolation forest, LOF, etc.) to filter out data that may belong to abnormal geological areas, focusing on the parts that are significantly different from ordinary strata.

[0054] In step (3), data preprocessing specifically includes:

[0055] 1. Data cleaning: Perform preliminary cleaning of geological, geophysical and drilling information to remove missing, duplicate or abnormal data points to ensure data quality.

[0056] 2. Normalization processing: For digital data, normalization methods can be used to convert data of different magnitudes into the same range, eliminate deviations between numerical values, and facilitate subsequent model processing.

[0057] 3. Feature selection: Use correlation analysis, principal component analysis (PCA) and other algorithms to reduce the dimension of the data, screen out features that have a high contribution to the discrimination of adverse geological types, and reduce the impact of invalid or redundant information on the model.

[0058] It also includes preprocessing of text description data, including but not limited to clarification and word segmentation.

[0059] In step (4), geological data, geophysical inversion imaging information and drilling information can be regarded as three major categories of information. Each major category of information may include multiple types of information, such as geological information may contain text type, image type and digital type information. For different information types, the feature extraction of each major category of information needs to be extracted using multiple deep learning network models.

[0060] Digital data features can be extracted using, but not limited to, DNN models, image data features can be extracted using, but not limited to, CNN models, and text description information can be extracted using, but not limited to, RNN.

[0061] As a further technical solution:

[0062] (I) Digital data features can be extracted using, but not limited to, DNN models. The specific extraction process is as follows:

[0063] 1. Data preprocessing:

[0064] Normalization processing: Normalize the input digital data (such as formation depth, rock density, rock thickness, drilling data, etc.) to normalize its value range to [0,1] or [-1,1], eliminate numerical differences of different magnitudes, and avoid deviations in model training.

[0065] Missing value processing: Use interpolation, mean filling, or front and back value filling to handle missing values ​​in the data to ensure that no anomalies caused by missing data occur during model training.

[0066] Feature selection: Through correlation analysis or feature importance ranking, features that have no contribution or low contribution to the identification of poor geological bodies are removed to reduce the complexity of the model.

[0067] 2. Input layer:

[0068] The preprocessed digital data is used as the input layer of the DNN (deep neural network), and each feature corresponds to a neuron. The features of the digital data include but are not limited to parameters that are highly correlated with the characteristics of adverse geological bodies, such as drilling depth, rock formation pressure, rock density, water content, and torque during drilling.

[0069] 3. Hidden layer design:

[0070] Number of layers and neurons: The hidden layer usually contains multiple layers, and the number of neurons in each layer is determined according to the complexity of the data and the model requirements. Typically, the number of hidden layers can be from 3 to 5 layers, and the number of neurons in each layer decreases (for example, 128 in the first layer, 64 in the second layer, and 32 in the third layer).

[0071] Activation function: Each neuron in the hidden layer passes through the ReLU (linear rectification) activation function to introduce nonlinear features and speed up the convergence of the model.

[0072] 4. Feature extraction:

[0073] The hidden layer gradually extracts meaningful high-dimensional features by calculating the nonlinear relationship between features layer by layer. Finally, high-level abstract features related to adverse geological features are extracted from hidden layers at different depths, such as the density change trend of a geological body, abnormal stress distribution, etc.

[0074] 5. Output layer:

[0075] The final output layer can be a softmax layer for classification of bad geological body types, or a linear output layer for predicting the numerical values ​​of geological parameters. In this way, DNN can effectively process complex digital features, automatically learn the relationship between digital features, and extract features that are useful for identifying bad geological bodies.

[0076] In the data processing process of the DNN model, the improvement methods adopted in this embodiment include:

[0077] 1. Data enhancement: Use simulation data based on physical models to enhance digital data and improve the generalization ability of the model.

[0078] 2. Regularization technology: L2 regularization or Dropout technology can be used to prevent DNN models from overfitting. Especially when there is less training data, regularization can significantly improve the robustness of the model.

[0079] (II) Image data features can be extracted using, but not limited to, CNN models. The specific extraction process is as follows:

[0080] 1. Data preprocessing:

[0081] Image standardization: Standardize the input geological inversion images, drilling profiles, radar images, etc., so that their pixel values ​​are normalized to the range of [0,1] to reduce calculation errors.

[0082] Resizing: resize images of different sizes to the same size, usually using interpolation or cropping methods, so that they can be input into CNN in batches.

[0083] Data enhancement: Increase data diversity and avoid model overfitting by rotating, flipping, scaling, and other operations on images.

[0084] 2. Input layer:

[0085] The preprocessed image is input to the input layer of the CNN model, and each pixel of the image is used as an input neuron. Typically, the input data is a three-dimensional tensor (height, width, number of channels), such as a two-dimensional profile of geological radar imaging or geophysical imaging results of drilling.

[0086] 3. Convolutional layer:

[0087] Convolution operation: In the convolution layer, multiple convolution kernels (such as 3×3 and 5×5 filters) are used to extract local features of the image, such as the boundaries of rock formations, the outlines of fractures, the texture of cracks, etc. The convolution kernel scans the entire image and identifies local features by weighted summation of local information.

[0088] Activation function: The ReLU activation function is usually used for the output after the convolution operation to increase nonlinearity and ensure that the network can handle complex features.

[0089] 4. Pooling layer:

[0090] Max pooling or average pooling: Through pooling operations (such as 2×2 max pooling), the size of the feature map is reduced, important local information is retained, the amount of calculation is reduced, and overfitting is avoided. The pooling layer helps to retain large-scale information such as rock formation characteristics and fault orientation.

[0091] 5. Fully connected layer:

[0092] The outputs of the convolutional layer and the pooling layer are expanded into a one-dimensional vector and input into the fully connected layer to further extract features related to the adverse geological body. The fully connected layer learns high-level global features such as the fault direction of the entire stratum and the distribution of water bodies through the weights between neurons.

[0093] 6. Output layer:

[0094] Finally, the bad geological types in the image are classified through the softmax or sigmoid function. For example, the output can be the geological body types such as fracture, fracture zone, karst cave, etc., and the location and size of the bad geological body can also be predicted through the regression model.

[0095] In the data processing process of the CNN model, the improvement methods adopted in this embodiment include:

[0096] 1. Multi-scale feature extraction: By adding a pyramid pooling layer, geological features at different scales, such as small cracks and large-scale fault structures, are extracted to improve the ability to recognize complex geological environments.

[0097] 2. Transfer learning: Use pre-trained models (such as ResNet, VGG, etc.) to accelerate the convergence of CNN and improve the feature extraction capabilities on small data sets.

[0098] (III) Text description information can be extracted using, but not limited to, RNN. The specific process of extraction is as follows:

[0099] 1. Data preprocessing:

[0100] Text cleaning: pre-process the text descriptions in geological exploration reports, remove irrelevant characters and punctuation marks, unify terminology, eliminate redundancy and typos, etc.

[0101] Word segmentation: Break down the long text data in the geological report into words or phrases and build a vocabulary. You can use the Jieba word segmentation tool to perform Chinese word segmentation.

[0102] Text encoding: The segmented text is encoded into word vectors using Word2Vec, Glove, or the pre-trained BERT model, converting each word into a vector representation of a fixed dimension.

[0103] 2. Input layer:

[0104] The encoded text data is input into the input layer of the RNN, and each word or phrase corresponds to a time step of the RNN. The RNN will gradually process and learn the contextual relationship between words based on the serialized text information.

[0105] 3. Hidden layer (circular layer):

[0106] Recurrent Neural Network (RNN) or Long Short-Term Memory Network (LSTM): In order to better capture long-distance dependencies, it is recommended to use LSTM or GRU instead of traditional RNN. Through the memory unit of LSTM, it is possible to capture dependencies that are far apart in geological descriptions, such as the connection between the "fracture zone" and "karst cave" mentioned before and after when describing a bad geological body.

[0107] Bidirectional RNN: Through the bidirectional RNN model, information is extracted from the forward and backward directions of the text at the same time to obtain a more complete representation of geological features.

[0108] 4. Attention Mechanism:

[0109] By adding an attention mechanism to the output of the RNN, the model can focus on important words in the text that are related to the characteristics of adverse geological bodies, such as key terms such as "fault", "crack", and "water-rich layer", thereby improving the ability to distinguish geological bodies.

[0110] 5. Output layer:

[0111] The final output can be the classification result of the bad geological body or a text description of the characteristics of the bad geological body. By connecting the natural language generation model, the summary of the geological report can be automatically generated, or the morphology and location of the bad geological body can be described in text.

[0112] In the data processing process of the RNN model, the improvement methods adopted in this embodiment include:

[0113] 1. Pre-trained language model: Use pre-trained language models such as BERT and GPT as the basis for feature extraction, combined with specially trained geological data sets to enhance the model's understanding and discrimination capabilities of geological language.

[0114] 2. Multi-task learning: Combine the bad geological body classification task with the geological description generation task, and improve the model's ability to understand text descriptions through multi-task learning.

[0115] This embodiment improves the above-mentioned model structures. The specific structural improvements of each model will be described in detail below.

[0116] Improvements in DNN structure

[0117] 1.1 Hierarchical design optimization:

[0118] Adaptive layer adjustment: Different geological data may have different complexity. By introducing an adaptive network structure, the number of network layers can be dynamically adjusted according to the complexity of the input data, rather than a fixed number of layers. Adaptive depth can be achieved by monitoring the changes in the loss function during training, so that the network has different depths at different stages.

[0119] 1.2 Residual Connections:

[0120] In deep networks, increased depth may lead to the vanishing or exploding gradient problem.

[0121] By adding residual connections (ResNet), some layers are allowed to skip connections directly to deeper layers, ensuring that gradients can be propagated smoothly. This can help DNN avoid performance degradation when processing complex geological data.

[0122] 1.3Dropout and Batch Normalization:

[0123] Dropout: Add Dropout operation to each layer to randomly close neurons with a certain probability to prevent network overfitting. Especially in geological prediction tasks with less training data, Dropout can effectively improve the generalization ability of the model.

[0124] Batch Normalization: Perform batch normalization on the output of each layer to reduce the internal covariate shift problem, help the model converge faster during training, and avoid local optimal solutions.

[0125] 1.4 Mixed activation function:

[0126] Traditional DNNs usually use a single activation function (such as ReLU). A mixed activation function can be introduced to use different activation functions in different layers, such as Leaky ReLU, ELU, etc., to better capture different types of nonlinear relationships. This has better adaptability to the complex nonlinear features in geological data.

[0127] To summarize the above, in this embodiment, in the DNN model, the number of layers is adaptively adjusted, residual connections are added to ensure that the gradient can be smoothly propagated, the Dropout operation is added in each layer to effectively improve the generalization ability of the model, batch normalization operations are performed on the output of each layer to help the model converge faster during training, and a mixed activation function is introduced to better adapt to the complex nonlinear characteristics of geological data. The traditional DNN model is improved to be more suitable for the processing of digital data in the process of geological type identification.

[0128] 2. Improvement of CNN structure

[0129] 2.1 Multi-scale convolution kernel:

[0130] Multi-scale convolution: Use convolution kernels of different sizes (such as 3×3, 5×5, 7×7) in the convolution layer to capture geological features of different scales simultaneously. For example, a 3×3 convolution kernel can extract detailed information, while a 5×5 or 7×7 convolution kernel helps capture geological structures in a larger range, such as the overall morphology of a fault or fracture zone.

[0131] 2.2 Combination of deep convolution and shallow convolution:

[0132] Feature Pyramid Network (FPN): It combines the high-dimensional features extracted by the deep convolution layer with the low-dimensional features extracted by the shallow convolution layer to extract information from different scales. It is especially suitable for the small cracks and large fault structures that exist simultaneously in geological images. The FPN network can effectively improve the expression ability of features at different scales.

[0133] 2.3 Dilated Convolution:

[0134] The dilated convolution can expand the receptive field without increasing the number of parameters, which is particularly applicable to scenes where features in geological images are sparse or unevenly distributed. By increasing the receptive field, the dilated convolution can better capture distant geological features at the same level and improve the model's ability to understand complex structures.

[0135] 2.4 Convolutional Attention Mechanism:

[0136] Integrate the attention mechanism into the convolutional network, so that the model can focus on the key parts of the geological image, such as the boundary of rock layers, fault zones, etc., and ignore irrelevant information. By introducing channel attention mechanism (such as Squeeze-and-Excitation Networks, SENet) or spatial attention mechanism, the discriminative power of the model can be improved.

[0137] 2.5 Convolutional variants:

[0138] Depthwise Separable Convolution: Split the standard convolution into depthwise convolution and pointwise convolution, significantly reducing the number of parameters and calculations, suitable for resource-limited application scenarios while maintaining high feature extraction capabilities. This method is suitable for real-time or lightweight geological prediction systems.

[0139] In summary, in this embodiment, in the CNN model, multi-scale convolution is used, the high-dimensional features extracted by the deep convolution layer and the low-dimensional features extracted by the shallow convolution layer are fused, the receptive field is expanded without increasing the number of parameters through the hole convolution, the attention mechanism is integrated into the convolutional network, and the deep separable convolution is used to improve the traditional CNN model, so that it is more suitable for the processing of image data in the process of geological type identification.

[0140] 3. Improvement of RNN structure

[0141] 3.1 Bidirectional LSTM / GRU:

[0142] Bidirectional RNN: Ordinary RNN or LSTM can only capture unidirectional sequence information, while bidirectional LSTM / GRU can simultaneously capture the front-end and back-end dependency information in geological texts. This has obvious advantages for complex language relationships in geological descriptions, such as the front-end and back-end related words can be better understood when describing fault structures.

[0143] 3.2 Self-Attention Mechanism:

[0144] The self-attention mechanism can help RNN find important features in long sequence inputs without being limited to the relationship between adjacent time steps. Through the attention mechanism, RNN can identify key geological terms and descriptions such as "fault zone" and "collapse layer" when processing geological reports, ignore minor information, and improve the modeling ability of long sequence data.

[0145] 3.3 Transformer network replaces RNN:

[0146] Although RNN is good at processing sequence data, it has limitations in processing long sequences. You can consider using the Transformer architecture to replace the traditional RNN / GRU / LSTM. Based on the global attention mechanism, Transformer can more efficiently handle the long dependency problem in geological texts, and has a higher degree of computational parallelism, which is suitable for processing large-scale geological data.

[0147] 3.4 Layered RNN:

[0148] Introducing the hierarchical RNN architecture to process geological texts in layers. For example, the first layer can process word relationships within a sentence, and the second layer can process logical relationships between different sentences. This can better capture the multi-level information in complex geological reports.

[0149] 3.5 Memory enhancement mechanism:

[0150] Improve the LSTM or GRU model and introduce a memory augmentation mechanism (Memory Augmented Networks). Use external memory units to record and store important geological information, improve the model's ability to remember long-term information, and avoid information forgetting. This is very helpful for dealing with geological reports with long-term dependencies (such as the connection between multiple fault descriptions).

[0151] To sum up the above, in this embodiment, in the RNN model, a bidirectional RNN is used to simultaneously capture the forward and backward dependency information in the geological text, a self-attention mechanism is introduced to help the RNN find important features in a long sequence input, the Transformer architecture is used to replace the traditional RNN / GRU / LSTM, a layered RNN architecture is introduced to process the geological text in layers, and a memory enhancement mechanism is introduced to improve the traditional RNN model, so that it is more suitable for the processing of image data in the process of geological type identification.

[0152] Furthermore, in step (6), the type of unfavorable geological body can be determined by using methods including but not limited to the softmax function, and the morphology of the unfavorable geological body can be described in words by using methods including but not limited to accessing a natural language generation model, and can be drawn by using methods including but not limited to accessing a generative adversarial network.

[0153] Furthermore, in step (7), the loss functions used for training the bad geological body type discrimination and the morphological characterization are different.

[0154] The loss function for training the bad geological body type discrimination may include but is not limited to the cross entropy function, in which physical constraints may be added:

[0155] L total =L task +L constraint

[0156] Among them, L total is the final loss function used; L task The loss function chosen at the beginning of training; L constraint is a physical constraint.

[0157] The loss function for training the bad geological body morphology characterization may include but is not limited to the mean square error function.

[0158] The above two loss functions are weighted to form a joint loss function to achieve overall training of poor geological body type discrimination and morphological characterization.

[0159] Combination Figure 1 As shown, the specific training process of the joint deep learning network in this embodiment includes the following steps:

[0160] (1) The geological data, geophysical inversion imaging information and drilling information of the excavated sections are screened and summarized to form training information. The geophysical inversion imaging information is Figure 1 The geophysical data in the drilling information is Figure 1 Drilling data in .

[0161] (2) The training information is input into the input layer of the joint deep learning network for data preprocessing to obtain geological information, geophysical information, and drilling information for the next step.

[0162] (3) Feature extraction of three major categories of information is performed in the extraction layer.

[0163] The geological information, geophysical information and drilling information after data preprocessing are input into the extraction layer, and the text in the three categories of information is extracted using including but not limited to the RNN deep learning network; the image in the three categories of information is extracted using including but not limited to the CNN deep learning network; the number in the three categories of information is extracted using including but not limited to the DNN deep learning network; so as to obtain the geological information features, geophysical information features and drilling information features.

[0164] (4) In the fusion layer, the attention mechanism is used to fuse the features extracted in the extraction layer into a joint feature representation.

[0165] The geological information features, geophysical information features and drilling information features extracted by the extraction layer are input into the fusion layer, and feature fusion is performed based on the attention mechanism to obtain a joint feature representation.

[0166] (5) In the output layer, based on the joint feature representation, different methods are used to complete the type discrimination and morphological characterization of the adverse geological body.

[0167] Here, the type discrimination of the bad geological body includes using the softmax function for discrimination;

[0168] Morphological characterization of adverse geological bodies includes the use of textual descriptions and conceptual diagrams.

[0169] (6) Train the above model in a way that minimizes the loss function.

[0170] For type discrimination, the loss function may include but is not limited to the cross entropy function; for morphological characterization, the loss function may include but is not limited to the mean square error function; the above two loss functions are weighted to establish a joint loss function for overall training to obtain a trained joint deep learning network.

[0171] This embodiment processes multi-source information through a joint deep learning network, which can achieve faster and more accurate type identification and morphological characterization of the unfavorable geological body in front of the tunnel.

[0172] Embodiment 2

[0173] This embodiment discloses a system for identifying adverse geological types based on a joint deep learning network.

[0174] Adverse geological type identification system based on joint deep learning network, including:

[0175] The data acquisition module is configured to: acquire training information, including geological data of the excavation section, geophysical inversion imaging information and drilling information;

[0176] The model prediction module is configured to: input the training information into a pre-established joint deep learning network model, perform data preprocessing, feature extraction and feature fusion in sequence, obtain a joint feature representation, and obtain the discrimination type and morphological characterization of the adverse geological body based on the joint feature representation;

[0177] Among them, the joint deep learning network model includes a DNN model, a CNN model and a RNN model arranged in parallel. When performing feature extraction, the DNN model, the CNN model and the RNN model are used to extract the features of digital data, image data and text description data in the training information respectively.

[0178] Furthermore, the specific steps performed by the above-mentioned data acquisition module and model prediction module of this embodiment are:

[0179] (1) Establish a database of adverse geological bodies.

[0180] (2) The geological data, geophysical inversion imaging information and drilling information of the excavated sections are screened and summarized and organized into training information.

[0181] (3) Establish a joint deep learning network input layer to preprocess the training information.

[0182] (4) Establish a joint deep learning network extraction layer and use different deep learning network models to extract features of different types of information in geological data, geophysical inversion imaging information, and drilling information.

[0183] Among them, the joint deep learning network model includes a DNN model, a CNN model and a RNN model arranged in parallel. When performing feature extraction, the DNN model, the CNN model and the RNN model are used to extract the features of digital data, image data and text description data in the training information respectively.

[0184] (5) A joint deep learning network fusion layer is established, and the attention mechanism is used to fuse the features of the extracted geological information, geophysical inversion imaging information, and drilling information to obtain a joint feature representation.

[0185] (6) Establish a joint deep learning network output layer to identify the type and characterize the morphology of adverse geological bodies based on the obtained joint feature representation.

[0186] (7) Train the above model in a way that minimizes the loss function.

[0187] (8) After training is completed, the geological information, geophysical inversion information, and drilling information of the forecast period are input into the joint deep learning network to obtain the prediction results;

[0188] Compare the prediction results output by the joint deep learning network with the results of the adverse geological body database to complete the identification of the adverse geological type and morphological characterization of the forecast section.

[0189] The prediction results of the joint deep learning network are compared and verified with the results of the adverse geological body database to prove the reliability of the prediction results of the joint deep learning network.

[0190] Those skilled in the art should understand that the modules or steps of the present invention described above can be implemented by a general-purpose computer device, or alternatively, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. The present invention is not limited to any specific combination of hardware and software.

[0191] Although the above describes the specific implementation mode of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without creative work are still within the scope of protection of the present invention.

Claims

1. A method for identifying adverse geological types based on a joint deep learning network, characterized in that: The following steps are involved: Obtain training information, including geological data of the excavation section, geophysical inversion imaging information and drilling information; The training information is input into the pre-established joint deep learning network model, and data preprocessing, feature extraction and feature fusion are performed in sequence to obtain a joint feature representation, and based on the joint feature representation, the discrimination type and morphological characterization of the adverse geological body are obtained; Among them, the joint deep learning network model includes a DNN model, a CNN model and a RNN model arranged in parallel. When performing feature extraction, the DNN model, the CNN model and the RNN model are used to extract the features of digital data, image data and text description data in the training information respectively.

2. The method for identifying adverse geological types based on a joint deep learning network according to claim 1, characterized in that: The joint deep learning network model includes an input layer, a feature extraction layer, a feature fusion layer and an output layer, wherein: The input layer is used to perform data preprocessing on the training information; The feature extraction layer is used to extract features of digital data, image data and text description data in the preprocessed training information based on the DNN model, the CNN model and the RNN model; The feature fusion layer is used to fuse the features extracted by the feature extraction layer based on the attention mechanism to obtain a joint feature representation; The output layer is used to distinguish the type and characterize the morphology of the bad geological body based on the joint feature representation.

3. The method for identifying adverse geological types based on a joint deep learning network as claimed in claim 2, characterized in that: The types of unfavorable geological bodies are identified using functions including but not limited to softmax functions; The morphology of unfavorable geological bodies is characterized by methods including but not limited to accessing a natural language generation model for textual description, and / or by methods including but not limited to accessing a generative adversarial network for drawing a conceptual map.

4. The method for identifying adverse geological types based on a joint deep learning network according to claim 1, characterized in that: The method also includes training the joint deep learning network model to obtain a trained joint deep learning network model, wherein: The loss function for training the discrimination of bad geological body types includes the cross entropy function, in which physical constraints can be added; The loss functions for training the morphological characterization of bad geological bodies include the mean square error function; The loss function for identifying the type of bad geological body and the loss function for describing the morphology of bad geological body are weighted to form a joint loss function of the joint deep learning network model, and the identification of the type of bad geological body and the morphology description are trained as a whole.

5. The method for identifying adverse geological types based on a joint deep learning network according to claim 4, characterized in that: The training of the joint deep learning network model is completed by minimizing the joint loss function.

6. The method for identifying adverse geological types based on a joint deep learning network according to claim 4, characterized in that: Also includes: Establish a database of adverse geological bodies; Input the geological information, geophysical inversion information and drilling information of the forecast section into the trained joint deep learning network model to obtain the prediction results; The prediction results are compared with the data in the bad geological body database to complete the bad geological type identification and morphological characterization of the forecast section.

7. The method for identifying adverse geological types based on a joint deep learning network according to claim 6, characterized in that: The bad geological body database is established based on the existing bad geological body information, and the bad geological body database data covers the type classification of bad geology and the characteristic description of bad geology.

8. The method for identifying adverse geological types based on a joint deep learning network according to claim 1, characterized in that: The training information includes at least one of digital data, image data and text description data.

9. The method for identifying adverse geological types based on a joint deep learning network according to claim 2, characterized in that: The data preprocessing of the training information includes normalizing the digital data and clarifying and segmenting the text description data.

10. The unfavorable geological type identification system based on the joint deep learning network is characterized by: include: The data acquisition module is configured to: acquire training information, including geological data of the excavation section, geophysical inversion imaging information and drilling information; The model prediction module is configured to: input the training information into a pre-established joint deep learning network model, perform data preprocessing, feature extraction and feature fusion in sequence, obtain a joint feature representation, and obtain the discrimination type and morphological characterization of the adverse geological body based on the joint feature representation; Among them, the joint deep learning network model includes a DNN model, a CNN model and a RNN model arranged in parallel. When performing feature extraction, the DNN model, the CNN model and the RNN model are used to extract the features of digital data, image data and text description data in the training information respectively.

Images