Tunnel construction geological disaster prevention and control intelligent decision-making method and auxiliary platform

Abstract

The invention relates to a tunnel construction geological disaster prevention and control intelligent decision-making method fusing a large model and a knowledge graph technology and an auxiliary platform, and the method comprises the steps: data collection and preprocessing: collecting tunnel general situation data, geological survey data, advanced forecast data and construction monitoring data, and carrying out the preprocessing of the data, and obtaining the processed data; a knowledge graph construction step of constructing a tunnel construction geological disaster prevention and control knowledge graph based on the preprocessed data, and performing structured representation on knowledge related to disaster prevention and control, including a disaster type entity, a geological condition entity, a prevention and control measure entity and a relationship among the entities; large model training: training the preprocessed data by adopting a large model architecture, and learning the characteristics and rules of tunnel construction geological disasters; and disaster prevention and control decision making: performing disaster prevention and control decision making in combination with output results of the knowledge graph and the large model.

Description

Technical Field

[0001] The present invention relates to the technical field of geological disaster prevention and control in tunnel construction, and in particular to an intelligent decision-making method and an auxiliary platform for geological disaster prevention and control in tunnel construction. Background Art

[0002] The occurrence of geological disasters during tunnel construction can have a serious impact on the project and even threaten the lives of personnel. Traditional geological disaster prevention and control decisions mainly rely on manual experience, which is uncertain and blind. With the development of artificial intelligence technology, large models and knowledge graph technology provide new solutions for intelligent decision-making in geological disaster prevention and control during tunnel construction, but there is currently no effective method to integrate these two technologies and apply them to this field. Summary of the invention

[0003] The technical problem to be solved by the present invention is to provide an intelligent decision-making method and auxiliary platform for geological disaster prevention and control in tunnel construction that integrates large models and knowledge graph technology.

[0004] To solve the above problems, the technical solution adopted by the present invention is:

[0005] In a first aspect, a method for intelligent decision-making for geological disaster prevention and control in tunnel construction is provided, the method comprising:

[0006] Data collection and preprocessing: Collect tunnel profile data, geological survey data, advance forecast data and construction monitoring data, and preprocess the above data to obtain processed data;

[0007] Knowledge graph construction: Based on the preprocessed data, a knowledge graph of geological disaster prevention and control in tunnel construction is constructed to structurally represent the knowledge related to disaster prevention and control, including disaster type entities, geological condition entities, prevention and control measures entities and the relationships between them;

[0008] Large model training: Using a large model architecture, the pre-processed data is trained to learn the characteristics and laws of geological hazards in tunnel construction;

[0009] Disaster prevention and control decision-making: Combine the knowledge graph and the output results of the big model to make disaster prevention and control decisions.

[0010] As an implementation mode of the present invention, in data collection and preprocessing, various data in the tunnel construction process are collected in real time through sensors and monitoring equipment, and historical data are collected from relevant databases and documents.

[0011] As an implementation of the present invention, knowledge graph construction includes the following steps:

[0012] According to the type of geological disasters, search for relevant literature on geological disasters;

[0013] According to the literature knowledge corpus, the maximum forward matching algorithm is adopted, and the named entity seeds are used to mark the literature sentences, and a geological disaster named entity recognition template is constructed, wherein a geological disaster named entity recognition corpus is constructed according to the construction of the geological disaster named entity recognition template, and the maximum forward matching algorithm is used to match the entities with the maximum length contained in the sentence forward as the named entity seeds, and these named entity seeds are used to mark the literature sentences, and the geological disaster named entity recognition corpus is constructed;

[0014] A geological disaster named entity recognition model is constructed, wherein the entity recognition model is a multi-branch bidirectional gated unit conditional random field model based on deep learning. The multi-branch bidirectional gated unit conditional random field model includes an embedding layer, a multi-branch bidirectional gated unit layer, and a conditional random field output layer. The model takes sentences as input, represents the sentences as matrices through the embedding layer, and then extracts features through a bidirectional gated unit network containing three branches. Finally, the features are re-optimized through a conditional random field model to obtain the final output, i.e., the named entity recognition result;

[0015] The multi-branch bidirectional gated unit conditional random field model is trained using the geological disaster named entity recognition corpus. The geological disaster named entity recognition corpus is divided into a training set and a test set. The training set is used to train the multi-branch bidirectional gated unit conditional random field model. The test set is used to evaluate the training results. The model that achieves the best effect in the test set is used as the final model.

[0016] Based on the geological disaster named entity recognition model, the geological disaster named entities are predicted and the geological disaster knowledge graph is constructed.

[0017] As an embodiment of the present invention, large model training includes the following steps:

[0018] Step S301: Establish a large-scale text corpus u={u1,u2,…,u n}, the corpus trains the training parameters θ of the target language model by maximizing the log-likelihood function, and these parameters are trained using random parameter descent;

[0019] Step S302: Create a data set M based on the preprocessed data, the data set including a plurality of input words m1, m2, ..., m z , and N of the labels carried by each word index, after the data set is input into the trained language model, the output parameters are input into the linear output layer to obtain the predicted label;

[0020] Step S303: Use a back propagation algorithm to train the linear output layer so that the loss function of the linear output layer is minimized.

[0021] In step S301, before training, the ui in the corpus is preprocessed with expected data; it includes:

[0022] (1) Word segmentation: divide each ui into a sequence of words or characters;

[0023] (2) Building a vocabulary: Building a vocabulary from the entire corpus to convert text into digital IDs;

[0024] (3) Convert data: Convert text data into a word index format that the model can process.

[0025] As an implementation mode of the present invention, the language model in step S301 adopts a multi-layer Transformer decoder; the multi-layer Transformer decoder respectively uses the following formula to obtain the output score [f θ (x)] i :

[0026] h0=UM e +M p ;

[0027]

[0028] [f θ (x)] i =softmax(h n M e T );

[0029] Among them, U={u i-k ,…,u i-1} represents the context vector, h0 is the initial hidden layer state, M e is the character embedding matrix, M p is the position embedding matrix.

[0030] As an implementation of the present invention, for any training example (x, y), the maximum word-level log-likelihood function is adopted as follows:

[0031] or,

[0032] The following sentence-level log-likelihood function is maximized as follows:

[0033]

[0034] As an implementation of the present invention, the loss function in step S303 can be expressed by the following formula:

[0035]

[0036] Among them, y i is the ith element in the true label encoding, It is the i-th element of the model prediction value after being processed by the softmax function.

[0037] As an implementation mode of the present invention, in the disaster prevention and control decision-making step, relevant prevention and control knowledge is mined from the knowledge graph according to the characteristics of the disaster and the on-site conditions, and the intelligent analysis capability of the large model is used to recommend the optimal prevention and control decision plan.

[0038] As an embodiment of the present invention, the method also includes: displaying relevant information and prevention and control decision-making plans of tunnel construction geological hazards to users through a visual interface, and users can provide feedback based on actual conditions to further optimize decisions based on the feedback information.

[0039] In the second aspect, an intelligent decision-making support platform for geological disaster prevention and control in tunnel construction is provided, which includes:

[0040] A data acquisition and preprocessing module is used to collect tunnel profile data, geological survey data, advance forecast data and construction monitoring data, and preprocess the above data to obtain processed data;

[0041] The knowledge graph construction module is used to construct a knowledge graph for geological disaster prevention and control in tunnel construction based on preprocessed data, and to structure the knowledge related to disaster prevention and control, including disaster type entities, geological condition entities, prevention and control measures entities, and the relationships between them;

[0042] The large model training model is used to train the preprocessed data using a large model architecture to learn the characteristics and laws of geological hazards in tunnel construction;

[0043] The disaster prevention and control decision-making module is used to combine the output results of the knowledge graph and the large model to make disaster prevention and control decisions.

[0044] The beneficial effects of adopting the above technical solution are:

[0045] The intelligent decision-making method and auxiliary platform for geological disaster prevention and control in tunnel construction provided by the present invention have the following advantages:

[0046] 1. The present invention integrates the big model and knowledge graph technology, combining the powerful data analysis ability of the big model with the knowledge structured representation ability of the knowledge graph, achieving a deeper understanding and more accurate application of geological disaster prevention and control knowledge in tunnel construction, and improving the level of intelligent decision-making.

[0047] 2. The present invention improves the accuracy and reliability of decision-making: through the knowledge guidance of the knowledge graph and the data analysis of the large model, it can more accurately identify disaster characteristics and predict disaster development trends, thereby recommending more effective prevention and control decision-making plans, improving the accuracy and reliability of decision-making.

[0048] 3. The present invention enhances the explainability of decision-making: the knowledge graph can intuitively display the relationship and logic of disaster prevention and control knowledge, making the decision-making process more explainable and facilitating users to understand and accept the decision results.

[0049] 4. The present invention can be updated and optimized in real time: the auxiliary platform can continuously update the knowledge graph and optimize the large model according to new data and cases, which improves the adaptability and real-time performance of the auxiliary platform. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 It is a flow chart of an intelligent decision-making method for geological disaster prevention and control in tunnel construction provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0051] In order to make the objectives, technical solutions and advantages of the present invention more clear, the present invention is clearly and completely described below in conjunction with specific embodiments.

[0052] Example 1

[0053] An embodiment of the present invention provides an intelligent decision-making method for geological disaster prevention and control in tunnel construction, the method comprising:

[0054] Step 1: Data collection and preprocessing: Collect tunnel profile data, geological survey data, advance forecast data and construction monitoring data, and preprocess the above data to obtain processed data;

[0055] In this step, various data including stress, displacement, water pressure, ground temperature, etc. in the tunnel construction process are collected in real time through sensors and monitoring equipment, and historical data are collected from relevant databases and documents; preprocessing includes: cleaning the collected data, removing noise and abnormal data, converting and integrating the data to meet the requirements of subsequent processing, for example: and converting this information into time series data.

[0056] In addition, before this step, the method also includes: constructing a refined three-dimensional geological model based on geological survey and advance forecast data to reveal information such as the type, scale, and spatial location of geological hazards in tunnel construction.

[0057] Step 2: Knowledge graph construction: Based on the preprocessed data, a knowledge graph of geological disaster prevention and control in tunnel construction is constructed to structurally represent the knowledge related to disaster prevention and control, including disaster type entities, geological condition entities, prevention and control measures entities and the relationships between them;

[0058] In this step, the entities, relationships, and attributes of the knowledge graph are defined, relevant information is extracted from the preprocessed data, the knowledge graph is constructed, and it is continuously improved and updated.

[0059] Specifically, it may include:

[0060] Step S201: Based on the literature knowledge corpus, the maximum forward matching algorithm is used to back-mark the literature sentences with named entity seeds, and a geological disaster named entity recognition template is constructed, wherein a geological disaster named entity recognition corpus is constructed according to the construction of the geological disaster named entity recognition template, and the maximum forward matching algorithm is used to match the entities with the maximum length positively contained in the sentence as named entity seeds, and these named entity seeds are back-marked on the literature sentences to construct a geological disaster named entity recognition corpus;

[0061] Step S202, constructing a geological disaster named entity recognition model, wherein the entity recognition model is a multi-branch bidirectional gated unit conditional random field model based on deep learning, and the multi-branch bidirectional gated unit conditional random field model includes an embedding layer, a multi-branch bidirectional gated unit layer, and a conditional random field output layer. The model takes sentences as input, represents the sentences as matrices through the embedding layer, and then extracts features through a bidirectional gated unit network containing three branches, and finally optimizes the features through a conditional random field model to obtain the final output, that is, the named entity recognition result;

[0062] Step S203, training the multi-branch bidirectional gated unit conditional random field model with the geological disaster named entity recognition corpus, dividing the geological disaster named entity recognition corpus into a training set and a test set, using the training set to train the multi-branch bidirectional gated unit conditional random field model, using the test set to evaluate the training results, and the model that achieves the best effect in the test set is used as the final model;

[0063] Step S204: predict geological disaster named entities based on the geological disaster named entity recognition model and construct a geological disaster knowledge graph.

[0064] Specifically, this step may adopt the method disclosed in the Chinese invention patent with publication number CN111125371B.

[0065] Step 3: Large model training: Use a large model architecture to train the preprocessed data to learn the characteristics and laws of geological hazards in tunnel construction;

[0066] In this step, large model training includes the following steps:

[0067] Step S301: Establish a large-scale text corpus u={u1,u2,…,u n}, the corpus trains the training parameters θ of the target language model by maximizing the log-likelihood function, and these parameters are trained using random parameter descent;

[0068] Among them, before training, the u in the corpus i Perform expected data preprocessing; which includes:

[0069] (1) Word segmentation: Each u i into sequences of words or characters;

[0070] (2) Building a vocabulary: Building a vocabulary from the entire corpus to convert text into digital IDs;

[0071] (3) Convert data: Convert text data into a word index format that the model can process.

[0072] The language model in this step uses a multi-layer Transformer decoder; the multi-layer Transformer decoder uses the following formulas to obtain the output score [f θ (x)] i :

[0073] h0=UM e +M p ;

[0074]

[0075] [f θ (x)] i =softmax(h n M e T );

[0076] Among them, U={u i-k ,…,u i-1} represents the context vector, h0 is the initial hidden layer state, M e is the character embedding matrix, M p is the position embedding matrix.

[0077] For any training example (x, y), y is the output score obtained above [f θ (x)] i , the word-level log-likelihood function is maximized as follows:

[0078] or,

[0079] The following sentence-level log-likelihood function is maximized as follows:

[0080]

[0081] Step S302: Create a data set M based on the preprocessed data, the data set including a plurality of input words m1, m2, ..., m z , and N of the labels carried by each word index, after the data set is input into the trained language model, the output parameters are input into the linear output layer to obtain the predicted label;

[0082] Step S303: Use a back propagation algorithm to train the linear output layer so that the loss function of the linear output layer is minimized.

[0083] The loss function in this step can be expressed as follows:

[0084]

[0085] Among them, y i is the i-th element in the true label encoding, It is the i-th element of the model prediction value after being processed by the softmax function.

[0086] Step 4: Disaster prevention and control decision-making: Combine the knowledge graph and the output results of the big model to make disaster prevention and control decisions.

[0087] This step includes: mining relevant prevention and control knowledge from the knowledge graph based on the characteristics of the disaster and the on-site conditions, and using the intelligent analysis capabilities of the big model to recommend the optimal prevention and control decision-making plan.

[0088] The method further comprises:

[0089] Step 5: Display relevant information and prevention and control decision-making plans of tunnel construction geological hazards to users through a visual interface. Users can provide feedback based on actual conditions to further optimize decisions based on the feedback information.

[0090] The specific steps of the present invention are as follows:

[0091] 1. Data collection: Various data during the tunnel construction process are collected in real time through sensors and monitoring equipment, and historical data are collected from relevant databases and documents.

[0092] 2. Data preprocessing: Clean the collected data, remove noise and abnormal data, and convert and integrate the data to meet the requirements of subsequent processing.

[0093] 3. Knowledge graph construction: define the entities, relationships, and attributes of the knowledge graph, extract relevant information from the preprocessed data, build the knowledge graph, and continuously improve and update it.

[0094] 4. Large model training: Select the appropriate large model architecture and algorithm, use preprocessed data for training, adjust model parameters, and improve model performance.

[0095] 5. Intelligent decision-making: In the process of disaster prevention and control decision-making, the knowledge in the knowledge graph and the analysis results of the big model are combined to comprehensively consider various factors and recommend the optimal prevention and control decision-making plan.

[0096] 6. Display and feedback: The relevant information and prevention and control decision-making plans of tunnel construction geological hazards are displayed to users through a visual interface. Users can provide feedback based on actual conditions, and the system further optimizes decisions based on the feedback information.

[0097] The intelligent decision-making method for preventing and controlling geological disasters in tunnel construction provided by the present invention has the following advantages:

[0098] 1. The present invention integrates the big model and knowledge graph technology, combining the powerful data analysis ability of the big model with the knowledge structured representation ability of the knowledge graph, achieving a deeper understanding and more accurate application of geological disaster prevention and control knowledge in tunnel construction, and improving the level of intelligent decision-making.

[0099] 2. The present invention improves the accuracy and reliability of decision-making: through the knowledge guidance of the knowledge graph and the data analysis of the large model, it can more accurately identify disaster characteristics and predict disaster development trends, thereby recommending more effective prevention and control decision-making plans, improving the accuracy and reliability of decision-making.

[0100] 3. The present invention enhances the explainability of decision-making: the knowledge graph can intuitively display the relationship and logic of disaster prevention and control knowledge, making the decision-making process more explainable and facilitating users to understand and accept the decision results.

[0101] 4. The present invention can be updated and optimized in real time: The present invention can continuously update the knowledge graph and optimize the large model according to new data and cases, thereby improving adaptability and real-time performance.

[0102] Example 2

[0103] The embodiment of the present invention provides an intelligent decision-making auxiliary platform for geological disaster prevention and control in tunnel construction, which includes:

[0104] A data acquisition and preprocessing module is used to collect tunnel profile data, geological survey data, advance forecast data and construction monitoring data, and preprocess the above data to obtain processed data;

[0105] The knowledge graph construction module is used to construct a knowledge graph for geological disaster prevention and control in tunnel construction based on preprocessed data, and to structure the knowledge related to disaster prevention and control, including disaster type entities, geological condition entities, prevention and control measures entities, and the relationships between them;

[0106] The large model training model is used to train the preprocessed data using a large model architecture to learn the characteristics and laws of geological hazards in tunnel construction;

[0107] The disaster prevention and control decision-making module is used to combine the output results of the knowledge graph and the large model to make disaster prevention and control decisions.

[0108] The intelligent decision-making auxiliary platform for geological disaster prevention and control in tunnel construction provided by the present invention has the following advantages:

[0109] 1. The present invention integrates the big model and knowledge graph technology, combining the powerful data analysis ability of the big model with the knowledge structured representation ability of the knowledge graph, achieving a deeper understanding and more accurate application of geological disaster prevention and control knowledge in tunnel construction, and improving the level of intelligent decision-making.

[0110] 2. The present invention improves the accuracy and reliability of decision-making: through the knowledge guidance of the knowledge graph and the data analysis of the large model, it can more accurately identify disaster characteristics and predict disaster development trends, thereby recommending more effective prevention and control decision-making plans, improving the accuracy and reliability of decision-making.

[0111] 3. The present invention enhances the explainability of decision-making: the knowledge graph can intuitively display the relationship and logic of disaster prevention and control knowledge, making the decision-making process more explainable and facilitating users to understand and accept the decision results.

[0112] 4. The present invention can be updated and optimized in real time: the auxiliary platform can continuously update the knowledge graph and optimize the large model according to new data and cases, improving its adaptability and real-time performance.

[0113] Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art may still modify the technical solutions described in the aforementioned embodiments, or make equivalent substitutions for some of the technical features therein; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An intelligent decision-making method for geological disaster prevention and control in tunnel construction, characterized in that: The method comprises: Data collection and preprocessing: Collect tunnel profile data, geological survey data, advance forecast data and construction monitoring data, and preprocess the above data to obtain processed data; Knowledge graph construction: Based on the preprocessed data, a knowledge graph of geological disaster prevention and control in tunnel construction is constructed to structurally represent the knowledge related to disaster prevention and control, including disaster type entities, geological condition entities, prevention and control measures entities and the relationships between them; Large model training: Using a large model architecture, the pre-processed data is trained to learn the characteristics and laws of geological hazards in tunnel construction; Disaster prevention and control decision-making: Combine the output results of the knowledge graph and the big model to make disaster prevention and control decisions.

2. According to claim 1, a tunnel construction geological disaster prevention and control intelligent decision-making method is characterized in that: In data collection and preprocessing, various data during the tunnel construction process are collected in real time through sensors and monitoring equipment, and historical data are collected from relevant databases and documents.

3. The intelligent decision-making method for geological disaster prevention and control in tunnel construction according to claim 1 is characterized in that: The knowledge graph construction includes the following steps: According to the type of geological disasters, search for relevant literature on geological disasters; According to the literature knowledge corpus, the maximum forward matching algorithm is adopted, and the named entity seeds are used to mark the literature sentences, and a geological disaster named entity recognition template is constructed, wherein a geological disaster named entity recognition corpus is constructed according to the construction of the geological disaster named entity recognition template, and the maximum forward matching algorithm is used to match the entities with the maximum length contained in the sentence forward as the named entity seeds, and these named entity seeds are used to mark the literature sentences, and the geological disaster named entity recognition corpus is constructed; A geological disaster named entity recognition model is constructed, wherein the entity recognition model is a multi-branch bidirectional gated unit conditional random field model based on deep learning. The multi-branch bidirectional gated unit conditional random field model includes an embedding layer, a multi-branch bidirectional gated unit layer, and a conditional random field output layer. The model takes sentences as input, represents the sentences as matrices through the embedding layer, and then extracts features through a bidirectional gated unit network containing three branches. Finally, the features are re-optimized through a conditional random field model to obtain the final output, i.e., the named entity recognition result; The multi-branch bidirectional gated unit conditional random field model is trained using the geological disaster named entity recognition corpus. The geological disaster named entity recognition corpus is divided into a training set and a test set. The training set is used to train the multi-branch bidirectional gated unit conditional random field model. The test set is used to evaluate the training results. The model that achieves the best effect in the test set is used as the final model. Based on the geological disaster named entity recognition model, the geological disaster named entities are predicted and the geological disaster knowledge graph is constructed.

4. The intelligent decision-making method for geological disaster prevention and control in tunnel construction according to claim 1 is characterized in that: Large model training includes the following steps: Step S301: Establish a large-scale text corpus u={u1,u2,…,u n }, the corpus trains the training parameters θ of the target language model by maximizing the log-likelihood function, and these parameters are trained using random parameter descent; Step S302: Create a data set M based on the preprocessed data, the data set including a plurality of input words m1, m2, ..., m z , and N of the labels carried by each word index, after the data set is input into the trained language model, the output parameters are input into the linear output layer to obtain the predicted label; Step S303: Use a back propagation algorithm to train the linear output layer so that the loss function of the linear output layer is minimized.

5. The intelligent decision-making method for geological disaster prevention and control in tunnel construction according to claim 4 is characterized in that: In step S301, before training, the u in the corpus is i Perform expected data preprocessing; It includes: (1) Word segmentation: Each u i into sequences of words or characters; (2) Building a vocabulary: Building a vocabulary from the entire corpus to convert text into digital IDs; (3) Convert data: Convert text data into a word index format that the model can process.

6. The intelligent decision-making method for geological disaster prevention and control in tunnel construction according to claim 4 is characterized in that: The language model in step S301 uses a multi-layer Transformer decoder; the multi-layer Transformer decoder uses the following formulas to obtain the output score [f θ (x)] i : h0=UM e +M p ; [f θ (x)] i =softmax(h n M e T ); Among them, U={u i-k ,…,u i-1 } represents the context vector, h0 is the initial hidden layer state, M e is the character embedding matrix, M p is the position embedding matrix.

7. The intelligent decision-making method for preventing and controlling geological disasters during tunnel construction according to claim 6 is characterized in that: For any training example (x, y), the word-level log-likelihood function is maximized as follows: or, The following sentence-level log-likelihood function is maximized as follows:

8. The intelligent decision-making method for geological disaster prevention and control in tunnel construction according to claim 6 is characterized in that: The loss function in step S303 can be expressed as follows: Among them, y i is the ith element in the true label encoding, It is the i-th element of the model prediction value after being processed by the softmax function.

9. The intelligent decision-making method for geological disaster prevention and control in tunnel construction according to claim 1 is characterized in that: In the disaster prevention and control decision-making step, relevant prevention and control knowledge is mined from the knowledge graph based on the characteristics of the disaster and the on-site conditions, and the intelligent analysis capabilities of the big model are used to recommend the optimal prevention and control decision plan.

10. An intelligent decision-making support platform for geological disaster prevention and control in tunnel construction, characterized in that: It includes: A data acquisition and preprocessing module is used to collect tunnel profile data, geological survey data, advance forecast data and construction monitoring data, and preprocess the above data to obtain processed data; The knowledge graph construction module is used to construct a knowledge graph for geological disaster prevention and control in tunnel construction based on preprocessed data, and to structure the knowledge related to disaster prevention and control, including disaster type entities, geological condition entities, prevention and control measures entities, and the relationships between them; The large model training model is used to train the preprocessed data using a large model architecture to learn the characteristics and laws of geological hazards in tunnel construction; The disaster prevention and control decision-making module is used to combine the output results of the knowledge graph and the large model to make disaster prevention and control decisions.

Images