Geological anomaly identification method and system based on knowledge and data dual drive

Abstract

The present invention relates to the technical field of bad geological identification, and specifically discloses a method and system for identifying geological anomalies based on dual drive of knowledge and data, the method comprising: constructing a database of bad geological features; constructing a knowledge base of geological anomaly spatial relations; screening out a set of geophysical abnormal indicators reflecting the existence of bad geological features in front of the tunnel; embedding the abnormal indicator set into words to obtain an abnormal indicator matrix; acquiring bad geological spatial information revealed in stages during tunnel excavation to form a spatial feature set; embedding the spatial feature set into words to obtain a semantic vector matrix; extracting features from the abnormal indicator matrix and the semantic vector matrix respectively to obtain a data feature sequence and a knowledge feature sequence; using a sparse autoencoder to fuse the data feature sequence and the knowledge feature sequence, and obtaining a bad geological identification result using a classifier based on the fused features. The present invention improves the interpretability of bad geological prediction for tunnels.

Description

Technical Field

[0001] The present invention relates to the technical field of adverse geological identification, and in particular to a geological anomaly identification method and system based on dual drive of knowledge and data. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] Geological anomaly identification plays a very important and fundamental role in the fields of tunnel and underground engineering construction. During the tunnel excavation process, the construction party will inevitably face the threat of various adverse geological disasters. Among them, typical adverse geology such as fault zones, sudden water and mud has a serious impact on the actual engineering construction and directly threatens the life and property safety of the workers.

[0004] As the number of long and deep tunnels under construction gradually increases, the geological environment in which the tunnels are located becomes increasingly extremely complex, and the difficulty of identifying geological abnormalities and disasters increases dramatically. At the same time, unidentified adverse geology poses a very serious safety hazard to tunnel construction. How to accurately predict and effectively identify the adverse geological phenomena that may exist ahead of tunnel excavation has become a major problem that needs to be solved urgently.

[0005] There are many methods for identifying adverse geology, but they can be roughly divided into two categories:

[0006] The first category is to directly identify bad geology and judge the scale and shape of bad geology. This type of method mainly includes bad geology identification based on geophysical detection, and the use of electromagnetic waves, seismic waves and other methods to judge bad geology. In the process of identifying bad geology, this type of method generally focuses on the research of information extraction process, that is, using data to drive the identification of bad geology, and using mathematical methods and models as much as possible to continuously mine geological, geophysical, remote sensing and other information. This method does not take into account the formation mechanism and process of bad geology, as well as its correlation with the surrounding environment, and the interpretability of the mining results is poor.

[0007] The second type is based on the formation mechanism of bad geology, identifying and analyzing bad geology from a geological perspective. This type of method mainly includes identification of bad geology based on geochemical geological anomalies, analysis of bad geology based on traditional geological perspectives, etc. This type of method is based on geological mechanisms for identification and prediction, that is, knowledge is used to drive the identification of bad geology, considering the formation and evolution of bad geology, and then combining geochemical and other data to identify and predict bad geology. This method is highly dependent on the rich experience of geological experts and less considers the direct impact on the project.

[0008] It can be seen from this that the currently available knowledge-driven identification and prediction of adverse geology emphasizes the causes, formation processes, types and impact levels of adverse geology. Without the coupling of these knowledge factors in time and space, adverse geology will not be able to be preserved and further studied. Moreover, the above-mentioned knowledge and laws are necessary but not sufficient conditions for adverse geology, and it is impossible to use this knowledge to drive the prediction of adverse geology.

[0009] At present, in terms of available data-driven identification and prediction of adverse geology, emphasis is placed on the laws and data of geological anomalies caused by adverse geology, but the two are poorly coupled. Therefore, neither single data-driven nor knowledge-driven methods can effectively complete the identification and prediction of adverse geology. Summary of the invention

[0010] In order to solve the above problems, the present invention proposes a geological anomaly identification method and system based on dual drive of knowledge and data, which integrates the theoretical knowledge of adverse geological formation and geological anomaly detection data to accurately predict tunnel adverse geology in an all-round, multi-angle and accurate manner.

[0011] In some embodiments, the following technical solutions are adopted:

[0012] A geological anomaly identification method based on knowledge and data dual drive, including:

[0013] Acquire geophysical detection data generated during tunnel adverse geological detection and build a database of adverse geological features;

[0014] Obtain relevant terms in the field of geology and the relationship between various terms, and at the same time obtain the formation factors of geological anomalies, consider the adverse geological types, formation factors and the relationship between the two, and establish a geological anomaly knowledge ontology model;

[0015] Obtain a large amount of spatial correlation information about the formation and development process of adverse geology, and build a geological anomaly spatial relationship knowledge base based on the geological anomaly knowledge ontology model;

[0016] Based on the actual engineering construction data, the geophysical abnormal indicator set reflecting the existence of bad geology in front of the tunnel is screened out from the bad geological feature database; the abnormal indicator set is word embedded to obtain the abnormal indicator matrix;

[0017] Based on actual engineering construction data, the adverse geological spatial information revealed in stages during tunnel excavation is obtained, and the key geological texts at the spatial level are obtained; the key geological texts are segmented to obtain an entity set; a concept search is performed on each entity through a geological anomaly spatial relationship knowledge base to form a spatial feature set; the spatial feature set is word embedded to obtain a semantic vector matrix;

[0018] Feature extraction is performed on the abnormal indicator matrix and the semantic vector matrix respectively to obtain the data feature sequence and the knowledge feature sequence; the sparse autoencoder is used to fuse the data feature sequence and the knowledge feature sequence, and based on the fused features, the classifier is used to obtain the adverse geological identification results.

[0019] Among them, for the abnormal indicator matrix, a fully connected network model is used for feature extraction. The fully connected network model includes two hidden layers. First, the input layer data is weighted summed, and after the weighted summation, each node of the hidden layer 1 is output; the output data of each node of the hidden layer 1 is used as the input data of the hidden layer 2, and the weighted summation is performed, and each node of the hidden layer 2 is output, and finally the output of each node of the output layer is obtained as the data feature sequence.

[0020] For the semantic vector matrix, a convolutional neural network model is used for feature extraction.

[0021] Use sparse autoencoder to fuse data feature sequence and knowledge feature sequence, specifically:

[0022] Perform the “⊕” operation on the data feature sequence and the knowledge feature sequence to obtain the merged feature ;

[0023] Merge features As the input of the sparse autoencoder, it is nonlinearly mapped to the hidden layer for sparse representation output as T, and then the hidden layer output result T is mapped to the output layer to obtain the binary fusion feature .

[0024] In other embodiments, the following technical solutions are adopted:

[0025] A geological anomaly identification system based on knowledge and data dual drive, including:

[0026] A data acquisition module is used to acquire geophysical detection data generated during the tunnel adverse geological detection process and to construct an adverse geological feature database;

[0027] Obtain relevant terms in the field of geology and the relationships between various terms, and at the same time obtain the formation factors of geological anomalies. Consider the adverse geological types, formation factors and the relationship between the two, and combine the geological anomaly knowledge ontology model to establish a geological anomaly knowledge ontology model;

[0028] Obtain a large amount of spatial correlation information about the formation and development process of adverse geological conditions and build a knowledge base of spatial relations of geological anomalies;

[0029] The abnormal indicator matrix construction module is used to screen out a set of abnormal indicators of geophysical exploration that reflect the existence of bad geology ahead of the tunnel from the database of bad geology features based on actual engineering construction data; embed the abnormal indicator set into words to obtain the abnormal indicator matrix;

[0030] The semantic vector matrix construction module is used to obtain the adverse geological spatial information revealed in stages during tunnel excavation based on actual engineering construction data, and obtain the key geological texts at the spatial level; segment the key geological texts to obtain an entity set; perform concept retrieval on each entity through a geological anomaly spatial relationship knowledge base to form a spatial feature set; perform word embedding on the spatial feature set to obtain a semantic vector matrix;

[0031] The bad geological identification module is used to extract features from the abnormal indicator matrix and the semantic vector matrix respectively to obtain data feature sequences and knowledge feature sequences; a sparse autoencoder is used to fuse the data feature sequences and the knowledge feature sequences, and based on the fused features, a classifier is used to obtain the bad geological identification results.

[0032] In other embodiments, the following technical solutions are adopted:

[0033] A terminal device includes a processor and a memory, wherein the processor is used to implement instructions; the memory is used to store multiple instructions, and the instructions are suitable for being loaded by the processor and executing the above-mentioned geological anomaly identification method based on dual drive of knowledge and data.

[0034] In other embodiments, the following technical solutions are adopted:

[0035] A computer-readable storage medium stores a plurality of instructions, wherein the instructions are suitable for being loaded by a processor of a terminal device and executing the above-mentioned geological anomaly identification method based on dual drive of knowledge and data.

[0036] Compared with the prior art, the present invention has the following beneficial effects:

[0037] (1) The present invention constructs a typical bad geological body feature database and a geological anomaly spatial relationship knowledge base, combines actual engineering data, integrates geophysical data indicators with geological anomaly knowledge, and makes full use of knowledge and data resources, thereby overcoming the shortcomings of single geophysical detection method and geological analysis method in the identification and prediction of tunnel bad geology; it provides a new dual-driven accurate prediction method for the identification and prediction of tunnel bad geology, and improves the interpretability of tunnel bad geology forecast.

[0038] (2) In the process of knowledge and data dual drive, the present invention embeds the abnormal indicators and the spatial feature set into two types of matrices through vector embedding, and uses the fully connected neural network and convolutional neural network to preliminarily extract the features, splice and fuse the features, and uses the sparse autoencoder to deeply fuse the merged features. The sparse autoencoder introduces the sparsity penalty term on the basis of the autoencoder, and learns the sparse representation of the data by changing the weight of the penalty term and adjusting the sparse activation degree of the hidden layer, so as to realize the data reconstruction of the merged features, and can better capture the important information of the input data, thereby mining the potential correlation between different feature data. It proposes a new idea for the fusion of multi-source features in the identification and prediction of tunnel adverse geology, and enhances the accuracy of adverse geology type prediction.

[0039] The present invention can achieve dimensionality reduction processing of high-dimensional vector matrix data by performing feature extraction, while retaining key feature information in the original data, thereby facilitating the sparse autoencoder to fully integrate knowledge and data features.

[0040] Other features and 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 the practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 It is a flow chart of a geological anomaly identification method based on knowledge and data dual drive in an embodiment of the present invention;

[0042] Figure 2 This is a schematic diagram of a fully connected network structure in an embodiment of the present invention;

[0043] Figure 3 Schematic diagram of the network architecture of the sparse autoencoder in an embodiment of the present invention. DETAILED DESCRIPTION

[0044] It should be noted that the following detailed descriptions are illustrative and are intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used in the present invention have the same meanings as those commonly understood by those skilled in the art to which the present application belongs.

[0045] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0046] Embodiment 1

[0047] In one or more embodiments, a knowledge- and data-driven geological anomaly identification method is disclosed, combining Figure 1 , specifically including the following process:

[0048] S101: Acquire geophysical detection data generated during the tunnel adverse geological detection process and construct an adverse geological feature database.

[0049] In this embodiment, based on actual engineering data, information extraction is performed through existing and collected case data of geophysical detection of poor geology in tunnels to obtain geophysical and geological information data.

[0050] From the extracted geophysical and geological information data, through the selection of different geophysical methods, the geophysical detection data generated in the process of detecting typical adverse geology such as fault zones and alteration zones are obtained. For example, according to the different geophysical methods adopted, the seismic wave reflection method (a type of seismic wave method) obtains abnormal wave velocity signals about the front detection section; the transient electromagnetic method (a type of electrical and electromagnetic method) obtains low resistivity data about the water-bearing structural adverse geological body in the front detection section. Comprehensively considering the properties of the data of the seismic, electrical and electromagnetic detection methods, such as data structure, complexity, size, etc., as well as considering the scale issues such as data read and write operation frequency, response time and concurrency, a matching database type is selected to build a database of adverse geological features with standardized data format.

[0051] S102: Obtain relevant terms in the field of geology and the relationships between various terms, and at the same time obtain the formation factors of geological anomalies, consider the types of adverse geology, formation factors and the relationship between the two, and establish a geological anomaly knowledge ontology model; obtain a large amount of spatial correlation information about the formation and development process of adverse geology, combine it with the geological anomaly knowledge ontology model, and construct a geological anomaly spatial relationship knowledge base.

[0052] In this embodiment, according to the large-scale text information mining module, a large number of Internet encyclopedia geology terms are crawled and the hierarchical relationship of the geology thesaurus is extracted. For example, there is a large amount of semi-structured data on geology terms on the Internet encyclopedia, such as the "fault" encyclopedia page, which indicates that the main cause of the fault is tectonic movement, and the basic types of faults are normal faults, reverse faults, and thrust faults; the hierarchical relationship of the geology thesaurus is to obtain the relationship between various terms through the compilation system in the professional terminology table, such as the geological action terminology is divided into internal geological action and external geological action, etc. Based on the above two methods, the formation factors of the geological anomaly body in the relevant literature on bad geological identification are obtained at the same time, and the expert assists in establishing the geological anomaly knowledge ontology model.

[0053] The geological anomaly knowledge ontology model of this embodiment stores a large amount of conceptual knowledge related to the formation of unfavorable geological bodies, which is substantially expressed as a "entity-relationship-entity" triple, and the connection between the specific unfavorable geological body itself and the related factors is constructed, forming a large amount of structured information such as unfavorable geological types, formation factors, and the relationship between the two; for example, the unfavorable geology in the "fault" tunnel, one of the main reasons affecting its formation and evolution is "tectonic movement", that is, the basic unit of the "tectonic movement-cause-fault" form is formed.

[0054] This embodiment is based on the strict constraints of the concepts in the ontology model to achieve accurate extraction of entities corresponding to various causes in the formation process of adverse geological conditions in the knowledge base; at the same time, through a semantic construction module that takes spatial relationships into account, a large amount of spatial correlation information about the formation and development process of adverse geological conditions is obtained to construct a knowledge base of spatial relationships of geological anomalies.

[0055] Because the entities used to represent spatial information in general ontology, although their semantics represent the concept of space, do not construct specific spatial positions and shapes in actual operations, making them indistinguishable from other types of entities and unable to accurately express information in the dimension of space. The semantic building module that takes into account spatial relationships in this embodiment is used to effectively express entities with spatial dimensional characteristics.

[0056] Spatial correlation information is the relevant information that can effectively express the size, shape, and orientation of specific adverse geological bodies. For example, when typical adverse geological bodies such as faults are identified using geological analysis methods, they are mainly judged based on their spatial distribution characteristics such as strike, dip, and inclination.

[0057] The spatial correlation information about the formation and development process of the unfavorable geological bodies in this embodiment refers to the spatial distribution characteristic information that can identify the unfavorable geological bodies at the geological knowledge level.

[0058] S103: Based on actual engineering construction data, a set of geophysical abnormal indicators reflecting the existence of adverse geology in front of the tunnel is screened out from the adverse geology feature database; and word embedding is performed on the abnormal indicator set to obtain an abnormal indicator matrix.

[0059] This embodiment is based on the actual data in the engineering construction, extracts the relevant information about the geophysical method test, and selects the geophysical method abnormal indicators reflecting the existence of the bad geology in front of the tunnel from the established bad geology feature database according to the actual type of geophysical method used, such as low impedance rate, abnormal wave velocity, etc., and collects the relevant abnormal indicators to form an abnormal indicator set A={a1, a2, a3, ... a n}.

[0060] The abnormal index of this embodiment is determined according to the index value; for example, one of the geophysical exploration methods described above, the transient electromagnetic method, is used to detect the resistivity data of the unfavorable geological body of the water-bearing structure ahead. If there is an unfavorable geological body ahead, the resistivity measured there is significantly lower than the normal index value. The abnormal index is to screen and summarize the indicator types that are different from the normal index value, and finally obtain the abnormal indicator set A.

[0061] The abnormal indicators of different adverse geological types are not necessarily different. The abnormal indicators are mainly determined by the applicable scope of the geophysical prospecting method used; for example:

[0062] ① The seismic wave reflection method is mainly used to obtain elastic parameters such as wave velocity and density of the rock mass within a certain range in front of the tunnel face. Based on the abnormal wave velocity or wave impedance, it is inferred whether there are faults, fracture zones, karst, weak interlayers and other unfavorable geological bodies. The abnormal indicators are abnormal wave velocity or abnormal wave impedance.

[0063] ② The transient electromagnetic method is to infer whether there is a water-bearing poorly structured geological body based on the low resistivity data of the front detection section, and its abnormal indicator is low resistivity;

[0064] ③ The ground penetrating radar method receives the reflected signals of electromagnetic waves of different frequencies according to the different electromagnetic characteristics of different terrain structural layers, and then infers the unfavorable geological bodies such as karst, faults, fractured zones, weak interlayers and groundwater-rich areas that may exist in front of the tunnel.

[0065] Therefore, for different types of adverse geology, the abnormal indicators may be the same or different.

[0066] The obtained anomaly indicator set A is input into the vector embedding model for word embedding to form an anomaly indicator matrix V. The vector embedding model of this embodiment may use a skip-gram model in Word2vec.

[0067] S104: Based on actual engineering construction data, obtain the adverse geological spatial information revealed in stages during tunnel excavation, and obtain the key geological text at the spatial level; segment the key geological text to obtain an entity set; perform concept retrieval on each entity through a geological anomaly spatial relationship knowledge base to form a spatial feature set; perform word embedding on the spatial feature set to obtain a semantic vector matrix.

[0068] In this embodiment, based on the process data of the actual excavation of the tunnel project, the geological characteristics of the tunnel site area at the project site, the geological sketch of the face and the geological exploration data are analyzed to obtain the adverse geological spatial information revealed in the tunnel excavation periodically, and obtain the key geological text at the spatial level. For example, when using the comprehensive geological analysis method to identify the fault zone, the basic geological data such as lithology, joints and fissures near the excavation face are mainly obtained through geological cataloging. At the same time, based on the spatial distribution precursor characteristics such as the strike, dip, and inclination of certain faults, qualitative speculation is made on the possible existence of faults in front of the face. The spatial distribution precursor characteristic information is the required key geological text.

[0069] The key geological texts representing spatial information are processed by entity processing. First, the maximum likelihood algorithm in the word segmentation tool is used to segment the text. At the same time, the semantic compatibility of each segment with the context is considered to eliminate the local entity ambiguity of the text segmentation fragments. The best text segmentation fragments are obtained and used as the entity set E = {e1, e2, e3, …e n}.

[0070] The segmented fragments in this embodiment are actually short text entities, each of which contains spatial feature information and related information that may reveal the unfavorable geology ahead, such as {"excavation surface", "influence zone", "joint", "inclination", "inclination angle"}, which include three specific objects: tunnel excavation surface, fault influence zone and joints, as well as two specific spatial features of inclination and inclination angle related to joints.

[0071] Each entity e in the obtained entity set E i , the concept retrieval is performed through the geological anomaly spatial relationship knowledge base. During the retrieval process, the semantic relevance calculation method is used to compare each entity e i The semantic relevance between concepts in the knowledge base is obtained by i , thus forming a spatial feature set B = {b1, b2, b3, …b n}.

[0072] The spatial feature set B is also input into the vector embedding model to form the semantic vector matrix X.

[0073] S105: Extract features from the abnormal indicator matrix and the semantic vector matrix respectively to obtain a data feature sequence and a knowledge feature sequence.

[0074] In this embodiment, for the obtained anomaly indicator matrix V, a fully connected network (hereinafter referred to as FC) model is used for feature extraction. FC consists of an input layer, a hidden layer and an output layer. The number of neuron nodes in the input layer is determined according to the variable dimension in the matrix V; in order to ensure the fitting accuracy of the smooth mapping and improve the performance of high-order feature sequence extraction, the hidden layer is composed of two layers, the number of neuron nodes in the first layer is initially set to 80, and the number of neuron nodes in the second layer is set to 40. The tanh function is used as the activation function, and the number of neuron nodes is adjusted according to the fitting results of the training data, that is, if it is underfitting, more neurons are used, and if it is overfitting, the number of neurons is reduced, until the number of neurons in the first and second layers is taken as h and k, so that the fitting result is more accurate. Assuming that the input layer has m nodes and the output layer has n nodes, the FC structure is as follows Figure 2 When calculating the FC network output, the input layer data is first weighted and summed, and the formula is:

[0075] ;

[0076] in, represents the weight matrix from the ath node in the input layer to the bth node in the hidden layer 1, is the ath input data, is the bias term of the bth node in hidden layer 1.

[0077] After weighted summation, each node in hidden layer 1 is output, and the formula is as follows:

[0078] ;

[0079] Similarly, the output data of each node in hidden layer 1 is used as the input data of hidden layer 2, and weighted summation is performed. Each node in hidden layer 2 is output, and finally the output of each node in the output layer is obtained. The formula is:

[0080] ;

[0081] ;

[0082] ;

[0083] ;

[0084] in, and They represent the weight matrices from the bth node in hidden layer 1 to the cth node in hidden layer 2, and from the cth node in hidden layer 2 to the dth node in the output layer, respectively. and Respectively represent the output of each node in the hidden layer 2 and the output layer, and They are respectively represented as the bias terms of the cth node in the hidden layer 2 and the dth node in the output layer.

[0085] Finally, we get the data feature sequence ={ .

[0086] For the semantic vector matrix X, a convolutional neural network (CNN) model is used to extract features from the matrix X. The CNN consists of four parts: input layer, convolution layer, pooling layer, and output layer. The input layer is the obtained semantic vector matrix X; the convolution layer is a one-dimensional convolution, and uses convolution kernels of size 3 and 4 to extract features to form a richer feature map. The activation function uses the ReLU function; the pooling layer uses the maximum pooling (max pooling) to reduce the number of model parameters and the overfitting problem of the model, and extract strong features; the output layer includes a fully connected layer and a softmax output layer. The fully connected layer splices the maximum pooling results of the previous steps together, and then inputs them into the softmax layer to obtain the probability of each category, calculate the loss function based on the prediction results, and update the parameters to complete a round of training.

[0087] After multiple rounds of training, we finally get a more accurate training result and extract the knowledge feature sequence. .

[0088] S106: Using a sparse autoencoder to fuse the data feature sequence and the knowledge feature sequence; based on the fused features, using a classifier to obtain a poor geological identification result.

[0089] In this embodiment, the data feature sequence Knowledge Feature Sequence Perform the “⊕” operation, which is a direct connection in the feature fusion method. It shares the features of the data side and the knowledge side to form a feature space, and concatenates the two sequence feature vectors together to obtain the merged features. The specific formula is expressed as:

[0090] ;

[0091] in, is the merged feature output in the feature space, is the data feature sequence Input, is a sequence of knowledge features Input, is the weight matrix.

[0092] For the obtained merged features, simple concatenation and fusion cannot fully reflect the potential correlation information between knowledge text and abnormal data. This embodiment uses sparse autoencoder (SAE) to fully explore the correlation between the two. Figure 3 , the number of nodes in the input layer and output layer of SAE is the same, the number of nodes in the hidden layer is greater than the number of nodes in the input and output layers, and the activation function uses the sigmoid function to merge the features As the input of SAE, it is nonlinearly mapped to the hidden layer for sparse representation output as T, and then the hidden layer output result T is mapped to the output layer to obtain the binary fusion feature , the above process is encoding and decoding. The formula is:

[0093] ;

[0094] ;

[0095] in, and distribution is the weight matrix of the encoder and decoder, and are the bias terms of the encoder and decoder respectively.

[0096] The binary fusion features obtained The input is input into the softmax classifier, the output values ​​of different neurons are calculated, and the adverse geological type corresponding to the neuron with the maximum output value is selected as the final recognition result of this identification, thereby realizing adverse geological prediction based on both knowledge and data drive.

[0097] It should be noted that the data acquisition process involved in this embodiment adopts reasonable and legal means to acquire data, and does not involve problems such as illegal acquisition of data.

[0098] Embodiment 2

[0099] In one or more embodiments, a geological anomaly identification system based on knowledge and data dual drive is disclosed, specifically including:

[0100] A data acquisition module is used to acquire geophysical detection data generated during the tunnel adverse geological detection process and to construct an adverse geological feature database;

[0101] Obtain relevant terms in the field of geology and the relationships between various terms, and at the same time obtain the formation factors of geological anomalies. Consider the adverse geological types, formation factors and the relationship between the two, and combine the geological anomaly knowledge ontology model to establish a geological anomaly knowledge ontology model;

[0102] Obtain a large amount of spatial correlation information about the formation and development process of adverse geological conditions and build a knowledge base of spatial relations of geological anomalies;

[0103] The abnormal indicator matrix construction module is used to screen out the abnormal indicator set of geophysical exploration method reflecting the existence of bad geology in front of the tunnel from the bad geology feature database based on actual engineering construction data; embed the abnormal indicator set into words to obtain the abnormal indicator matrix;

[0104] The semantic vector matrix construction module is used to obtain the adverse geological spatial information revealed in stages during tunnel excavation based on actual engineering construction data, and obtain the key geological texts at the spatial level; segment the key geological texts to obtain an entity set; perform concept retrieval on each entity through a geological anomaly spatial relationship knowledge base to form a spatial feature set; perform word embedding on the spatial feature set to obtain a semantic vector matrix;

[0105] The bad geological identification module is used to extract features from the abnormal indicator matrix and the semantic vector matrix respectively to obtain data feature sequences and knowledge feature sequences; a sparse autoencoder is used to fuse the data feature sequences and the knowledge feature sequences, and based on the fused features, a classifier is used to obtain the bad geological identification results.

[0106] The specific implementation of the above modules is the same as that in Example 1 and will not be described in detail.

[0107] Embodiment 3

[0108] In one or more embodiments, a terminal device is disclosed, including a server, the server including a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the knowledge- and data-driven geological anomaly identification method in Embodiment 1 when executing the program. For the sake of brevity, it will not be described in detail here.

[0109] It should be understood that in this embodiment, the processor may be a central processing unit CPU, and the processor may also be other general-purpose processors, digital signal processors DSP, application-specific integrated circuits ASIC, off-the-shelf programmable gate arrays FPGA or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0110] The memory may include a read-only memory and a random access memory, and provide instructions and data to the processor. A portion of the memory may also include a non-volatile random access memory. For example, the memory may also store information about the device type.

[0111] In the implementation process, each step of the above method can be completed by an integrated logic circuit of hardware in a processor or an instruction in the form of software.

[0112] Embodiment 4

[0113] In one or more embodiments, a computer-readable storage medium is disclosed, in which a plurality of instructions are stored, wherein the instructions are suitable for being loaded by a processor of a terminal device and executing the geological anomaly identification method based on dual drive of knowledge and data described in Example 1.

[0114] 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 geological anomaly identification method based on knowledge and data dual drive, characterized in that: include: Acquire geophysical detection data generated during tunnel adverse geological detection and build a database of adverse geological features; Obtain relevant terms in the field of geology and the relationship between various terms, and at the same time obtain the formation factors of geological anomalies, consider the adverse geological types, formation factors and the relationship between the two, and establish a geological anomaly knowledge ontology model; Obtain a large amount of spatial correlation information about the formation and development process of adverse geology, and build a geological anomaly spatial relationship knowledge base based on the geological anomaly knowledge ontology model; Based on the actual engineering construction data, the geophysical abnormal indicator set reflecting the existence of bad geology in front of the tunnel is screened out from the bad geological feature database; the abnormal indicator set is word embedded to obtain the abnormal indicator matrix; Based on actual engineering construction data, the adverse geological spatial information revealed in stages during tunnel excavation is obtained, and the key geological texts at the spatial level are obtained; the key geological texts are segmented to obtain an entity set; a concept search is performed on each entity through a geological anomaly spatial relationship knowledge base to form a spatial feature set; the spatial feature set is word embedded to obtain a semantic vector matrix; Feature extraction is performed on the abnormal indicator matrix and the semantic vector matrix respectively to obtain the data feature sequence and the knowledge feature sequence; the sparse autoencoder is used to fuse the data feature sequence and the knowledge feature sequence, and based on the fused features, the classifier is used to obtain the adverse geological identification results.

2. A method for identifying geological anomalies based on knowledge and data dual drive according to claim 1, characterized in that: The key geological text is segmented to obtain an entity set, specifically: The maximum likelihood algorithm in the word segmentation tool is used to segment the key geological texts, considering the semantic compatibility of each segment with the context, eliminating the local entity ambiguity of the text segmentation fragments, and obtaining the best text segmentation fragments as the entity set.

3. The method for identifying geological anomalies based on knowledge and data dual drive according to claim 1, characterized in that: The concept retrieval is performed on each entity through the geological anomaly spatial relationship knowledge base. The specific process is as follows: Based on the semantic association calculation method, each entity e i The semantic correlation between concepts in the knowledge base of geological anomaly spatial relationship is obtained to obtain the concept b corresponding to the maximum correlation. i .

4. The method for identifying geological anomalies based on knowledge and data dual drive according to claim 1, characterized in that: For the abnormal indicator matrix, a fully connected network model is used for feature extraction. The fully connected network model includes two hidden layers. First, the input layer data is weighted summed. After the weighted summation, each node of the hidden layer 1 is output; The output data of each node in hidden layer 1 is used as the input data of hidden layer 2, and weighted summation is performed. Each node in hidden layer 2 is output, and finally the output of each node in the output layer is obtained as the data feature sequence.

5. The method for identifying geological anomalies based on knowledge and data dual drive according to claim 1, characterized in that: For the semantic vector matrix, a convolutional neural network model is used for feature extraction.

6. The method for identifying geological anomalies based on knowledge and data dual drive according to claim 1, characterized in that: Use sparse autoencoder to fuse data feature sequence and knowledge feature sequence, specifically: Perform the "⊕" operation on the data feature sequence and the knowledge feature sequence to obtain the merged feature ; Merge features As the input of the sparse autoencoder, it is nonlinearly mapped to the hidden layer for sparse representation output as T, and then the hidden layer output result T is mapped to the output layer to obtain the binary fusion feature .

7. The method for identifying geological anomalies based on knowledge and data dual drive according to claim 6, characterized in that: The binary fusion feature Specifically: ； ； in, and distribution is the weight matrix of the encoder and decoder, and are the bias terms of the encoder and decoder respectively.

8. A geological anomaly identification system based on knowledge and data dual drive, characterized in that: include: A data acquisition module is used to acquire geophysical detection data generated during the tunnel adverse geological detection process and to construct an adverse geological feature database; Obtain relevant terms in the field of geology and the relationships between various terms, and at the same time obtain the formation factors of geological anomalies. Consider the adverse geological types, formation factors and the relationship between the two, and combine the geological anomaly knowledge ontology model to establish a geological anomaly knowledge ontology model; Obtain a large amount of spatial correlation information about the formation and development process of adverse geological conditions and build a knowledge base of spatial relations of geological anomalies; The abnormal indicator matrix construction module is used to screen out the abnormal indicator set of geophysical exploration method reflecting the existence of bad geology in front of the tunnel from the bad geology feature database based on actual engineering construction data; embed the abnormal indicator set into words to obtain the abnormal indicator matrix; The semantic vector matrix construction module is used to obtain the adverse geological spatial information revealed in stages during tunnel excavation based on actual engineering construction data, and obtain the key geological texts at the spatial level; segment the key geological texts to obtain an entity set; perform concept retrieval on each entity through a geological anomaly spatial relationship knowledge base to form a spatial feature set; perform word embedding on the spatial feature set to obtain a semantic vector matrix; The bad geological identification module is used to extract features from the abnormal indicator matrix and the semantic vector matrix respectively to obtain data feature sequences and knowledge feature sequences; a sparse autoencoder is used to fuse the data feature sequences and the knowledge feature sequences, and based on the fused features, a classifier is used to obtain the bad geological identification results.

9. A terminal device, comprising a processor and a memory, wherein the processor is used to implement instructions; and the memory is used to store multiple instructions, characterized in that: The instructions are suitable for being loaded by a processor and executing the geological anomaly identification method based on dual drive of knowledge and data as described in any one of claims 1-7.

10. A computer-readable storage medium storing a plurality of instructions, characterized in that: The instructions are suitable for being loaded by a processor of a terminal device and executing the geological anomaly identification method based on dual drive of knowledge and data as described in any one of claims 1-7.

Images